xref: /openbmc/linux/net/ipv4/icmp.c (revision 93dc544c)
1 /*
2  *	NET3:	Implementation of the ICMP protocol layer.
3  *
4  *		Alan Cox, <alan@redhat.com>
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 int icmp_xmit_lock(struct sock *sk)
208 {
209 	local_bh_disable();
210 
211 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
212 		/* This can happen if the output path signals a
213 		 * dst_link_failure() for an outgoing ICMP packet.
214 		 */
215 		local_bh_enable();
216 		return 1;
217 	}
218 	return 0;
219 }
220 
221 static inline void icmp_xmit_unlock(struct sock *sk)
222 {
223 	spin_unlock_bh(&sk->sk_lock.slock);
224 }
225 
226 /*
227  *	Send an ICMP frame.
228  */
229 
230 /*
231  *	Check transmit rate limitation for given message.
232  *	The rate information is held in the destination cache now.
233  *	This function is generic and could be used for other purposes
234  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
235  *
236  *	Note that the same dst_entry fields are modified by functions in
237  *	route.c too, but these work for packet destinations while xrlim_allow
238  *	works for icmp destinations. This means the rate limiting information
239  *	for one "ip object" is shared - and these ICMPs are twice limited:
240  *	by source and by destination.
241  *
242  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
243  *			  SHOULD allow setting of rate limits
244  *
245  * 	Shared between ICMPv4 and ICMPv6.
246  */
247 #define XRLIM_BURST_FACTOR 6
248 int xrlim_allow(struct dst_entry *dst, int timeout)
249 {
250 	unsigned long now, token = dst->rate_tokens;
251 	int rc = 0;
252 
253 	now = jiffies;
254 	token += now - dst->rate_last;
255 	dst->rate_last = now;
256 	if (token > XRLIM_BURST_FACTOR * timeout)
257 		token = XRLIM_BURST_FACTOR * timeout;
258 	if (token >= timeout) {
259 		token -= timeout;
260 		rc = 1;
261 	}
262 	dst->rate_tokens = token;
263 	return rc;
264 }
265 
266 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
267 		int type, int code)
268 {
269 	struct dst_entry *dst = &rt->u.dst;
270 	int rc = 1;
271 
272 	if (type > NR_ICMP_TYPES)
273 		goto out;
274 
275 	/* Don't limit PMTU discovery. */
276 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
277 		goto out;
278 
279 	/* No rate limit on loopback */
280 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
281 		goto out;
282 
283 	/* Limit if icmp type is enabled in ratemask. */
284 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
285 		rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
286 out:
287 	return rc;
288 }
289 
290 /*
291  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
292  */
293 void icmp_out_count(struct net *net, unsigned char type)
294 {
295 	ICMPMSGOUT_INC_STATS(net, type);
296 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
297 }
298 
299 /*
300  *	Checksum each fragment, and on the first include the headers and final
301  *	checksum.
302  */
303 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
304 			  struct sk_buff *skb)
305 {
306 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
307 	__wsum csum;
308 
309 	csum = skb_copy_and_csum_bits(icmp_param->skb,
310 				      icmp_param->offset + offset,
311 				      to, len, 0);
312 
313 	skb->csum = csum_block_add(skb->csum, csum, odd);
314 	if (icmp_pointers[icmp_param->data.icmph.type].error)
315 		nf_ct_attach(skb, icmp_param->skb);
316 	return 0;
317 }
318 
319 static void icmp_push_reply(struct icmp_bxm *icmp_param,
320 			    struct ipcm_cookie *ipc, struct rtable *rt)
321 {
322 	struct sock *sk;
323 	struct sk_buff *skb;
324 
325 	sk = icmp_sk(dev_net(rt->u.dst.dev));
326 	if (ip_append_data(sk, icmp_glue_bits, icmp_param,
327 			   icmp_param->data_len+icmp_param->head_len,
328 			   icmp_param->head_len,
329 			   ipc, rt, MSG_DONTWAIT) < 0)
330 		ip_flush_pending_frames(sk);
331 	else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
332 		struct icmphdr *icmph = icmp_hdr(skb);
333 		__wsum csum = 0;
334 		struct sk_buff *skb1;
335 
336 		skb_queue_walk(&sk->sk_write_queue, skb1) {
337 			csum = csum_add(csum, skb1->csum);
338 		}
339 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
340 						 (char *)icmph,
341 						 icmp_param->head_len, csum);
342 		icmph->checksum = csum_fold(csum);
343 		skb->ip_summed = CHECKSUM_NONE;
344 		ip_push_pending_frames(sk);
345 	}
346 }
347 
348 /*
349  *	Driving logic for building and sending ICMP messages.
350  */
351 
352 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
353 {
354 	struct ipcm_cookie ipc;
355 	struct rtable *rt = skb->rtable;
356 	struct net *net = dev_net(rt->u.dst.dev);
357 	struct sock *sk = icmp_sk(net);
358 	struct inet_sock *inet = inet_sk(sk);
359 	__be32 daddr;
360 
361 	if (ip_options_echo(&icmp_param->replyopts, skb))
362 		return;
363 
364 	if (icmp_xmit_lock(sk))
365 		return;
366 
367 	icmp_param->data.icmph.checksum = 0;
368 
369 	inet->tos = ip_hdr(skb)->tos;
370 	daddr = ipc.addr = rt->rt_src;
371 	ipc.opt = NULL;
372 	if (icmp_param->replyopts.optlen) {
373 		ipc.opt = &icmp_param->replyopts;
374 		if (ipc.opt->srr)
375 			daddr = icmp_param->replyopts.faddr;
376 	}
377 	{
378 		struct flowi fl = { .nl_u = { .ip4_u =
379 					      { .daddr = daddr,
380 						.saddr = rt->rt_spec_dst,
381 						.tos = RT_TOS(ip_hdr(skb)->tos) } },
382 				    .proto = IPPROTO_ICMP };
383 		security_skb_classify_flow(skb, &fl);
384 		if (ip_route_output_key(net, &rt, &fl))
385 			goto out_unlock;
386 	}
387 	if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
388 			       icmp_param->data.icmph.code))
389 		icmp_push_reply(icmp_param, &ipc, rt);
390 	ip_rt_put(rt);
391 out_unlock:
392 	icmp_xmit_unlock(sk);
393 }
394 
395 
396 /*
397  *	Send an ICMP message in response to a situation
398  *
399  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
400  *		  MAY send more (we do).
401  *			MUST NOT change this header information.
402  *			MUST NOT reply to a multicast/broadcast IP address.
403  *			MUST NOT reply to a multicast/broadcast MAC address.
404  *			MUST reply to only the first fragment.
405  */
406 
407 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
408 {
409 	struct iphdr *iph;
410 	int room;
411 	struct icmp_bxm icmp_param;
412 	struct rtable *rt = skb_in->rtable;
413 	struct ipcm_cookie ipc;
414 	__be32 saddr;
415 	u8  tos;
416 	struct net *net;
417 	struct sock *sk;
418 
419 	if (!rt)
420 		goto out;
421 	net = dev_net(rt->u.dst.dev);
422 	sk = icmp_sk(net);
423 
424 	/*
425 	 *	Find the original header. It is expected to be valid, of course.
426 	 *	Check this, icmp_send is called from the most obscure devices
427 	 *	sometimes.
428 	 */
429 	iph = ip_hdr(skb_in);
430 
431 	if ((u8 *)iph < skb_in->head ||
432 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
433 		goto out;
434 
435 	/*
436 	 *	No replies to physical multicast/broadcast
437 	 */
438 	if (skb_in->pkt_type != PACKET_HOST)
439 		goto out;
440 
441 	/*
442 	 *	Now check at the protocol level
443 	 */
444 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
445 		goto out;
446 
447 	/*
448 	 *	Only reply to fragment 0. We byte re-order the constant
449 	 *	mask for efficiency.
450 	 */
451 	if (iph->frag_off & htons(IP_OFFSET))
452 		goto out;
453 
454 	/*
455 	 *	If we send an ICMP error to an ICMP error a mess would result..
456 	 */
457 	if (icmp_pointers[type].error) {
458 		/*
459 		 *	We are an error, check if we are replying to an
460 		 *	ICMP error
461 		 */
462 		if (iph->protocol == IPPROTO_ICMP) {
463 			u8 _inner_type, *itp;
464 
465 			itp = skb_header_pointer(skb_in,
466 						 skb_network_header(skb_in) +
467 						 (iph->ihl << 2) +
468 						 offsetof(struct icmphdr,
469 							  type) -
470 						 skb_in->data,
471 						 sizeof(_inner_type),
472 						 &_inner_type);
473 			if (itp == NULL)
474 				goto out;
475 
476 			/*
477 			 *	Assume any unknown ICMP type is an error. This
478 			 *	isn't specified by the RFC, but think about it..
479 			 */
480 			if (*itp > NR_ICMP_TYPES ||
481 			    icmp_pointers[*itp].error)
482 				goto out;
483 		}
484 	}
485 
486 	if (icmp_xmit_lock(sk))
487 		return;
488 
489 	/*
490 	 *	Construct source address and options.
491 	 */
492 
493 	saddr = iph->daddr;
494 	if (!(rt->rt_flags & RTCF_LOCAL)) {
495 		struct net_device *dev = NULL;
496 
497 		if (rt->fl.iif &&
498 			net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
499 			dev = dev_get_by_index(net, rt->fl.iif);
500 
501 		if (dev) {
502 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
503 			dev_put(dev);
504 		} else
505 			saddr = 0;
506 	}
507 
508 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
509 					   IPTOS_PREC_INTERNETCONTROL) :
510 					  iph->tos;
511 
512 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
513 		goto out_unlock;
514 
515 
516 	/*
517 	 *	Prepare data for ICMP header.
518 	 */
519 
520 	icmp_param.data.icmph.type	 = type;
521 	icmp_param.data.icmph.code	 = code;
522 	icmp_param.data.icmph.un.gateway = info;
523 	icmp_param.data.icmph.checksum	 = 0;
524 	icmp_param.skb	  = skb_in;
525 	icmp_param.offset = skb_network_offset(skb_in);
526 	inet_sk(sk)->tos = tos;
527 	ipc.addr = iph->saddr;
528 	ipc.opt = &icmp_param.replyopts;
529 
530 	{
531 		struct flowi fl = {
532 			.nl_u = {
533 				.ip4_u = {
534 					.daddr = icmp_param.replyopts.srr ?
535 						icmp_param.replyopts.faddr :
536 						iph->saddr,
537 					.saddr = saddr,
538 					.tos = RT_TOS(tos)
539 				}
540 			},
541 			.proto = IPPROTO_ICMP,
542 			.uli_u = {
543 				.icmpt = {
544 					.type = type,
545 					.code = code
546 				}
547 			}
548 		};
549 		int err;
550 		struct rtable *rt2;
551 
552 		security_skb_classify_flow(skb_in, &fl);
553 		if (__ip_route_output_key(net, &rt, &fl))
554 			goto out_unlock;
555 
556 		/* No need to clone since we're just using its address. */
557 		rt2 = rt;
558 
559 		err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
560 		switch (err) {
561 		case 0:
562 			if (rt != rt2)
563 				goto route_done;
564 			break;
565 		case -EPERM:
566 			rt = NULL;
567 			break;
568 		default:
569 			goto out_unlock;
570 		}
571 
572 		if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
573 			goto relookup_failed;
574 
575 		if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
576 			err = __ip_route_output_key(net, &rt2, &fl);
577 		else {
578 			struct flowi fl2 = {};
579 			struct dst_entry *odst;
580 
581 			fl2.fl4_dst = fl.fl4_src;
582 			if (ip_route_output_key(net, &rt2, &fl2))
583 				goto relookup_failed;
584 
585 			/* Ugh! */
586 			odst = skb_in->dst;
587 			err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
588 					     RT_TOS(tos), rt2->u.dst.dev);
589 
590 			dst_release(&rt2->u.dst);
591 			rt2 = skb_in->rtable;
592 			skb_in->dst = odst;
593 		}
594 
595 		if (err)
596 			goto relookup_failed;
597 
598 		err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
599 				  XFRM_LOOKUP_ICMP);
600 		switch (err) {
601 		case 0:
602 			dst_release(&rt->u.dst);
603 			rt = rt2;
604 			break;
605 		case -EPERM:
606 			goto ende;
607 		default:
608 relookup_failed:
609 			if (!rt)
610 				goto out_unlock;
611 			break;
612 		}
613 	}
614 
615 route_done:
616 	if (!icmpv4_xrlim_allow(net, rt, type, code))
617 		goto ende;
618 
619 	/* RFC says return as much as we can without exceeding 576 bytes. */
620 
621 	room = dst_mtu(&rt->u.dst);
622 	if (room > 576)
623 		room = 576;
624 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
625 	room -= sizeof(struct icmphdr);
626 
627 	icmp_param.data_len = skb_in->len - icmp_param.offset;
628 	if (icmp_param.data_len > room)
629 		icmp_param.data_len = room;
630 	icmp_param.head_len = sizeof(struct icmphdr);
631 
632 	icmp_push_reply(&icmp_param, &ipc, rt);
633 ende:
634 	ip_rt_put(rt);
635 out_unlock:
636 	icmp_xmit_unlock(sk);
637 out:;
638 }
639 
640 
641 /*
642  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
643  */
644 
645 static void icmp_unreach(struct sk_buff *skb)
646 {
647 	struct iphdr *iph;
648 	struct icmphdr *icmph;
649 	int hash, protocol;
650 	struct net_protocol *ipprot;
651 	u32 info = 0;
652 	struct net *net;
653 
654 	net = dev_net(skb->dst->dev);
655 
656 	/*
657 	 *	Incomplete header ?
658 	 * 	Only checks for the IP header, there should be an
659 	 *	additional check for longer headers in upper levels.
660 	 */
661 
662 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
663 		goto out_err;
664 
665 	icmph = icmp_hdr(skb);
666 	iph   = (struct iphdr *)skb->data;
667 
668 	if (iph->ihl < 5) /* Mangled header, drop. */
669 		goto out_err;
670 
671 	if (icmph->type == ICMP_DEST_UNREACH) {
672 		switch (icmph->code & 15) {
673 		case ICMP_NET_UNREACH:
674 		case ICMP_HOST_UNREACH:
675 		case ICMP_PROT_UNREACH:
676 		case ICMP_PORT_UNREACH:
677 			break;
678 		case ICMP_FRAG_NEEDED:
679 			if (ipv4_config.no_pmtu_disc) {
680 				LIMIT_NETDEBUG(KERN_INFO "ICMP: " NIPQUAD_FMT ": "
681 							 "fragmentation needed "
682 							 "and DF set.\n",
683 					       NIPQUAD(iph->daddr));
684 			} else {
685 				info = ip_rt_frag_needed(net, iph,
686 							 ntohs(icmph->un.frag.mtu),
687 							 skb->dev);
688 				if (!info)
689 					goto out;
690 			}
691 			break;
692 		case ICMP_SR_FAILED:
693 			LIMIT_NETDEBUG(KERN_INFO "ICMP: " NIPQUAD_FMT ": Source "
694 						 "Route Failed.\n",
695 				       NIPQUAD(iph->daddr));
696 			break;
697 		default:
698 			break;
699 		}
700 		if (icmph->code > NR_ICMP_UNREACH)
701 			goto out;
702 	} else if (icmph->type == ICMP_PARAMETERPROB)
703 		info = ntohl(icmph->un.gateway) >> 24;
704 
705 	/*
706 	 *	Throw it at our lower layers
707 	 *
708 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
709 	 *		  header.
710 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
711 	 *		  transport layer.
712 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
713 	 *		  transport layer.
714 	 */
715 
716 	/*
717 	 *	Check the other end isnt violating RFC 1122. Some routers send
718 	 *	bogus responses to broadcast frames. If you see this message
719 	 *	first check your netmask matches at both ends, if it does then
720 	 *	get the other vendor to fix their kit.
721 	 */
722 
723 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
724 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
725 		if (net_ratelimit())
726 			printk(KERN_WARNING NIPQUAD_FMT " sent an invalid ICMP "
727 					    "type %u, code %u "
728 					    "error to a broadcast: " NIPQUAD_FMT " on %s\n",
729 			       NIPQUAD(ip_hdr(skb)->saddr),
730 			       icmph->type, icmph->code,
731 			       NIPQUAD(iph->daddr),
732 			       skb->dev->name);
733 		goto out;
734 	}
735 
736 	/* Checkin full IP header plus 8 bytes of protocol to
737 	 * avoid additional coding at protocol handlers.
738 	 */
739 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
740 		goto out;
741 
742 	iph = (struct iphdr *)skb->data;
743 	protocol = iph->protocol;
744 
745 	/*
746 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
747 	 */
748 	raw_icmp_error(skb, protocol, info);
749 
750 	hash = protocol & (MAX_INET_PROTOS - 1);
751 	rcu_read_lock();
752 	ipprot = rcu_dereference(inet_protos[hash]);
753 	if (ipprot && ipprot->err_handler)
754 		ipprot->err_handler(skb, info);
755 	rcu_read_unlock();
756 
757 out:
758 	return;
759 out_err:
760 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
761 	goto out;
762 }
763 
764 
765 /*
766  *	Handle ICMP_REDIRECT.
767  */
768 
769 static void icmp_redirect(struct sk_buff *skb)
770 {
771 	struct iphdr *iph;
772 
773 	if (skb->len < sizeof(struct iphdr))
774 		goto out_err;
775 
776 	/*
777 	 *	Get the copied header of the packet that caused the redirect
778 	 */
779 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
780 		goto out;
781 
782 	iph = (struct iphdr *)skb->data;
783 
784 	switch (icmp_hdr(skb)->code & 7) {
785 	case ICMP_REDIR_NET:
786 	case ICMP_REDIR_NETTOS:
787 		/*
788 		 * As per RFC recommendations now handle it as a host redirect.
789 		 */
790 	case ICMP_REDIR_HOST:
791 	case ICMP_REDIR_HOSTTOS:
792 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
793 			       icmp_hdr(skb)->un.gateway,
794 			       iph->saddr, skb->dev);
795 		break;
796 	}
797 out:
798 	return;
799 out_err:
800 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
801 	goto out;
802 }
803 
804 /*
805  *	Handle ICMP_ECHO ("ping") requests.
806  *
807  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
808  *		  requests.
809  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
810  *		  included in the reply.
811  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
812  *		  echo requests, MUST have default=NOT.
813  *	See also WRT handling of options once they are done and working.
814  */
815 
816 static void icmp_echo(struct sk_buff *skb)
817 {
818 	struct net *net;
819 
820 	net = dev_net(skb->dst->dev);
821 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
822 		struct icmp_bxm icmp_param;
823 
824 		icmp_param.data.icmph	   = *icmp_hdr(skb);
825 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
826 		icmp_param.skb		   = skb;
827 		icmp_param.offset	   = 0;
828 		icmp_param.data_len	   = skb->len;
829 		icmp_param.head_len	   = sizeof(struct icmphdr);
830 		icmp_reply(&icmp_param, skb);
831 	}
832 }
833 
834 /*
835  *	Handle ICMP Timestamp requests.
836  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
837  *		  SHOULD be in the kernel for minimum random latency.
838  *		  MUST be accurate to a few minutes.
839  *		  MUST be updated at least at 15Hz.
840  */
841 static void icmp_timestamp(struct sk_buff *skb)
842 {
843 	struct timespec tv;
844 	struct icmp_bxm icmp_param;
845 	/*
846 	 *	Too short.
847 	 */
848 	if (skb->len < 4)
849 		goto out_err;
850 
851 	/*
852 	 *	Fill in the current time as ms since midnight UT:
853 	 */
854 	getnstimeofday(&tv);
855 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
856 					 tv.tv_nsec / NSEC_PER_MSEC);
857 	icmp_param.data.times[2] = icmp_param.data.times[1];
858 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
859 		BUG();
860 	icmp_param.data.icmph	   = *icmp_hdr(skb);
861 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
862 	icmp_param.data.icmph.code = 0;
863 	icmp_param.skb		   = skb;
864 	icmp_param.offset	   = 0;
865 	icmp_param.data_len	   = 0;
866 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
867 	icmp_reply(&icmp_param, skb);
868 out:
869 	return;
870 out_err:
871 	ICMP_INC_STATS_BH(dev_net(skb->dst->dev), ICMP_MIB_INERRORS);
872 	goto out;
873 }
874 
875 
876 /*
877  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
878  *
879  * RFC1122 (3.2.2.9).  A host MUST only send replies to
880  * ADDRESS_MASK requests if it's been configured as an address mask
881  * agent.  Receiving a request doesn't constitute implicit permission to
882  * act as one. Of course, implementing this correctly requires (SHOULD)
883  * a way to turn the functionality on and off.  Another one for sysctl(),
884  * I guess. -- MS
885  *
886  * RFC1812 (4.3.3.9).	A router MUST implement it.
887  *			A router SHOULD have switch turning it on/off.
888  *		      	This switch MUST be ON by default.
889  *
890  * Gratuitous replies, zero-source replies are not implemented,
891  * that complies with RFC. DO NOT implement them!!! All the idea
892  * of broadcast addrmask replies as specified in RFC950 is broken.
893  * The problem is that it is not uncommon to have several prefixes
894  * on one physical interface. Moreover, addrmask agent can even be
895  * not aware of existing another prefixes.
896  * If source is zero, addrmask agent cannot choose correct prefix.
897  * Gratuitous mask announcements suffer from the same problem.
898  * RFC1812 explains it, but still allows to use ADDRMASK,
899  * that is pretty silly. --ANK
900  *
901  * All these rules are so bizarre, that I removed kernel addrmask
902  * support at all. It is wrong, it is obsolete, nobody uses it in
903  * any case. --ANK
904  *
905  * Furthermore you can do it with a usermode address agent program
906  * anyway...
907  */
908 
909 static void icmp_address(struct sk_buff *skb)
910 {
911 #if 0
912 	if (net_ratelimit())
913 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
914 #endif
915 }
916 
917 /*
918  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
919  *			loudly if an inconsistency is found.
920  */
921 
922 static void icmp_address_reply(struct sk_buff *skb)
923 {
924 	struct rtable *rt = skb->rtable;
925 	struct net_device *dev = skb->dev;
926 	struct in_device *in_dev;
927 	struct in_ifaddr *ifa;
928 
929 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
930 		goto out;
931 
932 	in_dev = in_dev_get(dev);
933 	if (!in_dev)
934 		goto out;
935 	rcu_read_lock();
936 	if (in_dev->ifa_list &&
937 	    IN_DEV_LOG_MARTIANS(in_dev) &&
938 	    IN_DEV_FORWARD(in_dev)) {
939 		__be32 _mask, *mp;
940 
941 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
942 		BUG_ON(mp == NULL);
943 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
944 			if (*mp == ifa->ifa_mask &&
945 			    inet_ifa_match(rt->rt_src, ifa))
946 				break;
947 		}
948 		if (!ifa && net_ratelimit()) {
949 			printk(KERN_INFO "Wrong address mask " NIPQUAD_FMT " from "
950 					 "%s/" NIPQUAD_FMT "\n",
951 			       NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
952 		}
953 	}
954 	rcu_read_unlock();
955 	in_dev_put(in_dev);
956 out:;
957 }
958 
959 static void icmp_discard(struct sk_buff *skb)
960 {
961 }
962 
963 /*
964  *	Deal with incoming ICMP packets.
965  */
966 int icmp_rcv(struct sk_buff *skb)
967 {
968 	struct icmphdr *icmph;
969 	struct rtable *rt = skb->rtable;
970 	struct net *net = dev_net(rt->u.dst.dev);
971 
972 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
973 		int nh;
974 
975 		if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
976 				 XFRM_STATE_ICMP))
977 			goto drop;
978 
979 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
980 			goto drop;
981 
982 		nh = skb_network_offset(skb);
983 		skb_set_network_header(skb, sizeof(*icmph));
984 
985 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
986 			goto drop;
987 
988 		skb_set_network_header(skb, nh);
989 	}
990 
991 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
992 
993 	switch (skb->ip_summed) {
994 	case CHECKSUM_COMPLETE:
995 		if (!csum_fold(skb->csum))
996 			break;
997 		/* fall through */
998 	case CHECKSUM_NONE:
999 		skb->csum = 0;
1000 		if (__skb_checksum_complete(skb))
1001 			goto error;
1002 	}
1003 
1004 	if (!pskb_pull(skb, sizeof(*icmph)))
1005 		goto error;
1006 
1007 	icmph = icmp_hdr(skb);
1008 
1009 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1010 	/*
1011 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1012 	 *
1013 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1014 	 *		  discarded.
1015 	 */
1016 	if (icmph->type > NR_ICMP_TYPES)
1017 		goto error;
1018 
1019 
1020 	/*
1021 	 *	Parse the ICMP message
1022 	 */
1023 
1024 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1025 		/*
1026 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1027 		 *	  silently ignored (we let user decide with a sysctl).
1028 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1029 		 *	  discarded if to broadcast/multicast.
1030 		 */
1031 		if ((icmph->type == ICMP_ECHO ||
1032 		     icmph->type == ICMP_TIMESTAMP) &&
1033 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1034 			goto error;
1035 		}
1036 		if (icmph->type != ICMP_ECHO &&
1037 		    icmph->type != ICMP_TIMESTAMP &&
1038 		    icmph->type != ICMP_ADDRESS &&
1039 		    icmph->type != ICMP_ADDRESSREPLY) {
1040 			goto error;
1041 		}
1042 	}
1043 
1044 	icmp_pointers[icmph->type].handler(skb);
1045 
1046 drop:
1047 	kfree_skb(skb);
1048 	return 0;
1049 error:
1050 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1051 	goto drop;
1052 }
1053 
1054 /*
1055  *	This table is the definition of how we handle ICMP.
1056  */
1057 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1058 	[ICMP_ECHOREPLY] = {
1059 		.handler = icmp_discard,
1060 	},
1061 	[1] = {
1062 		.handler = icmp_discard,
1063 		.error = 1,
1064 	},
1065 	[2] = {
1066 		.handler = icmp_discard,
1067 		.error = 1,
1068 	},
1069 	[ICMP_DEST_UNREACH] = {
1070 		.handler = icmp_unreach,
1071 		.error = 1,
1072 	},
1073 	[ICMP_SOURCE_QUENCH] = {
1074 		.handler = icmp_unreach,
1075 		.error = 1,
1076 	},
1077 	[ICMP_REDIRECT] = {
1078 		.handler = icmp_redirect,
1079 		.error = 1,
1080 	},
1081 	[6] = {
1082 		.handler = icmp_discard,
1083 		.error = 1,
1084 	},
1085 	[7] = {
1086 		.handler = icmp_discard,
1087 		.error = 1,
1088 	},
1089 	[ICMP_ECHO] = {
1090 		.handler = icmp_echo,
1091 	},
1092 	[9] = {
1093 		.handler = icmp_discard,
1094 		.error = 1,
1095 	},
1096 	[10] = {
1097 		.handler = icmp_discard,
1098 		.error = 1,
1099 	},
1100 	[ICMP_TIME_EXCEEDED] = {
1101 		.handler = icmp_unreach,
1102 		.error = 1,
1103 	},
1104 	[ICMP_PARAMETERPROB] = {
1105 		.handler = icmp_unreach,
1106 		.error = 1,
1107 	},
1108 	[ICMP_TIMESTAMP] = {
1109 		.handler = icmp_timestamp,
1110 	},
1111 	[ICMP_TIMESTAMPREPLY] = {
1112 		.handler = icmp_discard,
1113 	},
1114 	[ICMP_INFO_REQUEST] = {
1115 		.handler = icmp_discard,
1116 	},
1117 	[ICMP_INFO_REPLY] = {
1118 		.handler = icmp_discard,
1119 	},
1120 	[ICMP_ADDRESS] = {
1121 		.handler = icmp_address,
1122 	},
1123 	[ICMP_ADDRESSREPLY] = {
1124 		.handler = icmp_address_reply,
1125 	},
1126 };
1127 
1128 static void __net_exit icmp_sk_exit(struct net *net)
1129 {
1130 	int i;
1131 
1132 	for_each_possible_cpu(i)
1133 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1134 	kfree(net->ipv4.icmp_sk);
1135 	net->ipv4.icmp_sk = NULL;
1136 }
1137 
1138 static int __net_init icmp_sk_init(struct net *net)
1139 {
1140 	int i, err;
1141 
1142 	net->ipv4.icmp_sk =
1143 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1144 	if (net->ipv4.icmp_sk == NULL)
1145 		return -ENOMEM;
1146 
1147 	for_each_possible_cpu(i) {
1148 		struct sock *sk;
1149 
1150 		err = inet_ctl_sock_create(&sk, PF_INET,
1151 					   SOCK_RAW, IPPROTO_ICMP, net);
1152 		if (err < 0)
1153 			goto fail;
1154 
1155 		net->ipv4.icmp_sk[i] = sk;
1156 
1157 		/* Enough space for 2 64K ICMP packets, including
1158 		 * sk_buff struct overhead.
1159 		 */
1160 		sk->sk_sndbuf =
1161 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1162 
1163 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1164 	}
1165 
1166 	/* Control parameters for ECHO replies. */
1167 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1168 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1169 
1170 	/* Control parameter - ignore bogus broadcast responses? */
1171 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1172 
1173 	/*
1174 	 * 	Configurable global rate limit.
1175 	 *
1176 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1177 	 *	bucket ratemask defines which icmp types are ratelimited by
1178 	 *	setting	it's bit position.
1179 	 *
1180 	 *	default:
1181 	 *	dest unreachable (3), source quench (4),
1182 	 *	time exceeded (11), parameter problem (12)
1183 	 */
1184 
1185 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1186 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1187 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1188 
1189 	return 0;
1190 
1191 fail:
1192 	for_each_possible_cpu(i)
1193 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1194 	kfree(net->ipv4.icmp_sk);
1195 	return err;
1196 }
1197 
1198 static struct pernet_operations __net_initdata icmp_sk_ops = {
1199        .init = icmp_sk_init,
1200        .exit = icmp_sk_exit,
1201 };
1202 
1203 int __init icmp_init(void)
1204 {
1205 	return register_pernet_device(&icmp_sk_ops);
1206 }
1207 
1208 EXPORT_SYMBOL(icmp_err_convert);
1209 EXPORT_SYMBOL(icmp_send);
1210 EXPORT_SYMBOL(xrlim_allow);
1211