xref: /openbmc/linux/net/ipv4/icmp.c (revision 63dc02bd)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
66 
67 #include <linux/module.h>
68 #include <linux/types.h>
69 #include <linux/jiffies.h>
70 #include <linux/kernel.h>
71 #include <linux/fcntl.h>
72 #include <linux/socket.h>
73 #include <linux/in.h>
74 #include <linux/inet.h>
75 #include <linux/inetdevice.h>
76 #include <linux/netdevice.h>
77 #include <linux/string.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <linux/slab.h>
80 #include <net/snmp.h>
81 #include <net/ip.h>
82 #include <net/route.h>
83 #include <net/protocol.h>
84 #include <net/icmp.h>
85 #include <net/tcp.h>
86 #include <net/udp.h>
87 #include <net/raw.h>
88 #include <net/ping.h>
89 #include <linux/skbuff.h>
90 #include <net/sock.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/init.h>
94 #include <asm/uaccess.h>
95 #include <net/checksum.h>
96 #include <net/xfrm.h>
97 #include <net/inet_common.h>
98 
99 /*
100  *	Build xmit assembly blocks
101  */
102 
103 struct icmp_bxm {
104 	struct sk_buff *skb;
105 	int offset;
106 	int data_len;
107 
108 	struct {
109 		struct icmphdr icmph;
110 		__be32	       times[3];
111 	} data;
112 	int head_len;
113 	struct ip_options_data replyopts;
114 };
115 
116 /* An array of errno for error messages from dest unreach. */
117 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
118 
119 const struct icmp_err icmp_err_convert[] = {
120 	{
121 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
122 		.fatal = 0,
123 	},
124 	{
125 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
126 		.fatal = 0,
127 	},
128 	{
129 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
130 		.fatal = 1,
131 	},
132 	{
133 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
134 		.fatal = 1,
135 	},
136 	{
137 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
138 		.fatal = 0,
139 	},
140 	{
141 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
142 		.fatal = 0,
143 	},
144 	{
145 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
146 		.fatal = 1,
147 	},
148 	{
149 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
150 		.fatal = 1,
151 	},
152 	{
153 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
154 		.fatal = 1,
155 	},
156 	{
157 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
158 		.fatal = 1,
159 	},
160 	{
161 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
162 		.fatal = 1,
163 	},
164 	{
165 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
166 		.fatal = 0,
167 	},
168 	{
169 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
170 		.fatal = 0,
171 	},
172 	{
173 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
174 		.fatal = 1,
175 	},
176 	{
177 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
178 		.fatal = 1,
179 	},
180 	{
181 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
182 		.fatal = 1,
183 	},
184 };
185 EXPORT_SYMBOL(icmp_err_convert);
186 
187 /*
188  *	ICMP control array. This specifies what to do with each ICMP.
189  */
190 
191 struct icmp_control {
192 	void (*handler)(struct sk_buff *skb);
193 	short   error;		/* This ICMP is classed as an error message */
194 };
195 
196 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
197 
198 /*
199  *	The ICMP socket(s). This is the most convenient way to flow control
200  *	our ICMP output as well as maintain a clean interface throughout
201  *	all layers. All Socketless IP sends will soon be gone.
202  *
203  *	On SMP we have one ICMP socket per-cpu.
204  */
205 static struct sock *icmp_sk(struct net *net)
206 {
207 	return net->ipv4.icmp_sk[smp_processor_id()];
208 }
209 
210 static inline struct sock *icmp_xmit_lock(struct net *net)
211 {
212 	struct sock *sk;
213 
214 	local_bh_disable();
215 
216 	sk = icmp_sk(net);
217 
218 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
219 		/* This can happen if the output path signals a
220 		 * dst_link_failure() for an outgoing ICMP packet.
221 		 */
222 		local_bh_enable();
223 		return NULL;
224 	}
225 	return sk;
226 }
227 
228 static inline void icmp_xmit_unlock(struct sock *sk)
229 {
230 	spin_unlock_bh(&sk->sk_lock.slock);
231 }
232 
233 /*
234  *	Send an ICMP frame.
235  */
236 
237 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
238 				      struct flowi4 *fl4, int type, int code)
239 {
240 	struct dst_entry *dst = &rt->dst;
241 	bool rc = true;
242 
243 	if (type > NR_ICMP_TYPES)
244 		goto out;
245 
246 	/* Don't limit PMTU discovery. */
247 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
248 		goto out;
249 
250 	/* No rate limit on loopback */
251 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
252 		goto out;
253 
254 	/* Limit if icmp type is enabled in ratemask. */
255 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
256 		if (!rt->peer)
257 			rt_bind_peer(rt, fl4->daddr, 1);
258 		rc = inet_peer_xrlim_allow(rt->peer,
259 					   net->ipv4.sysctl_icmp_ratelimit);
260 	}
261 out:
262 	return rc;
263 }
264 
265 /*
266  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
267  */
268 void icmp_out_count(struct net *net, unsigned char type)
269 {
270 	ICMPMSGOUT_INC_STATS(net, type);
271 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
272 }
273 
274 /*
275  *	Checksum each fragment, and on the first include the headers and final
276  *	checksum.
277  */
278 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
279 			  struct sk_buff *skb)
280 {
281 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
282 	__wsum csum;
283 
284 	csum = skb_copy_and_csum_bits(icmp_param->skb,
285 				      icmp_param->offset + offset,
286 				      to, len, 0);
287 
288 	skb->csum = csum_block_add(skb->csum, csum, odd);
289 	if (icmp_pointers[icmp_param->data.icmph.type].error)
290 		nf_ct_attach(skb, icmp_param->skb);
291 	return 0;
292 }
293 
294 static void icmp_push_reply(struct icmp_bxm *icmp_param,
295 			    struct flowi4 *fl4,
296 			    struct ipcm_cookie *ipc, struct rtable **rt)
297 {
298 	struct sock *sk;
299 	struct sk_buff *skb;
300 
301 	sk = icmp_sk(dev_net((*rt)->dst.dev));
302 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
303 			   icmp_param->data_len+icmp_param->head_len,
304 			   icmp_param->head_len,
305 			   ipc, rt, MSG_DONTWAIT) < 0) {
306 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
307 		ip_flush_pending_frames(sk);
308 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
309 		struct icmphdr *icmph = icmp_hdr(skb);
310 		__wsum csum = 0;
311 		struct sk_buff *skb1;
312 
313 		skb_queue_walk(&sk->sk_write_queue, skb1) {
314 			csum = csum_add(csum, skb1->csum);
315 		}
316 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
317 						 (char *)icmph,
318 						 icmp_param->head_len, csum);
319 		icmph->checksum = csum_fold(csum);
320 		skb->ip_summed = CHECKSUM_NONE;
321 		ip_push_pending_frames(sk, fl4);
322 	}
323 }
324 
325 /*
326  *	Driving logic for building and sending ICMP messages.
327  */
328 
329 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
330 {
331 	struct ipcm_cookie ipc;
332 	struct rtable *rt = skb_rtable(skb);
333 	struct net *net = dev_net(rt->dst.dev);
334 	struct flowi4 fl4;
335 	struct sock *sk;
336 	struct inet_sock *inet;
337 	__be32 daddr;
338 
339 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
340 		return;
341 
342 	sk = icmp_xmit_lock(net);
343 	if (sk == NULL)
344 		return;
345 	inet = inet_sk(sk);
346 
347 	icmp_param->data.icmph.checksum = 0;
348 
349 	inet->tos = ip_hdr(skb)->tos;
350 	daddr = ipc.addr = ip_hdr(skb)->saddr;
351 	ipc.opt = NULL;
352 	ipc.tx_flags = 0;
353 	if (icmp_param->replyopts.opt.opt.optlen) {
354 		ipc.opt = &icmp_param->replyopts.opt;
355 		if (ipc.opt->opt.srr)
356 			daddr = icmp_param->replyopts.opt.opt.faddr;
357 	}
358 	memset(&fl4, 0, sizeof(fl4));
359 	fl4.daddr = daddr;
360 	fl4.saddr = rt->rt_spec_dst;
361 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
362 	fl4.flowi4_proto = IPPROTO_ICMP;
363 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
364 	rt = ip_route_output_key(net, &fl4);
365 	if (IS_ERR(rt))
366 		goto out_unlock;
367 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
368 			       icmp_param->data.icmph.code))
369 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
370 	ip_rt_put(rt);
371 out_unlock:
372 	icmp_xmit_unlock(sk);
373 }
374 
375 static struct rtable *icmp_route_lookup(struct net *net,
376 					struct flowi4 *fl4,
377 					struct sk_buff *skb_in,
378 					const struct iphdr *iph,
379 					__be32 saddr, u8 tos,
380 					int type, int code,
381 					struct icmp_bxm *param)
382 {
383 	struct rtable *rt, *rt2;
384 	struct flowi4 fl4_dec;
385 	int err;
386 
387 	memset(fl4, 0, sizeof(*fl4));
388 	fl4->daddr = (param->replyopts.opt.opt.srr ?
389 		      param->replyopts.opt.opt.faddr : iph->saddr);
390 	fl4->saddr = saddr;
391 	fl4->flowi4_tos = RT_TOS(tos);
392 	fl4->flowi4_proto = IPPROTO_ICMP;
393 	fl4->fl4_icmp_type = type;
394 	fl4->fl4_icmp_code = code;
395 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
396 	rt = __ip_route_output_key(net, fl4);
397 	if (IS_ERR(rt))
398 		return rt;
399 
400 	/* No need to clone since we're just using its address. */
401 	rt2 = rt;
402 
403 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
404 					   flowi4_to_flowi(fl4), NULL, 0);
405 	if (!IS_ERR(rt)) {
406 		if (rt != rt2)
407 			return rt;
408 	} else if (PTR_ERR(rt) == -EPERM) {
409 		rt = NULL;
410 	} else
411 		return rt;
412 
413 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
414 	if (err)
415 		goto relookup_failed;
416 
417 	if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
418 		rt2 = __ip_route_output_key(net, &fl4_dec);
419 		if (IS_ERR(rt2))
420 			err = PTR_ERR(rt2);
421 	} else {
422 		struct flowi4 fl4_2 = {};
423 		unsigned long orefdst;
424 
425 		fl4_2.daddr = fl4_dec.saddr;
426 		rt2 = ip_route_output_key(net, &fl4_2);
427 		if (IS_ERR(rt2)) {
428 			err = PTR_ERR(rt2);
429 			goto relookup_failed;
430 		}
431 		/* Ugh! */
432 		orefdst = skb_in->_skb_refdst; /* save old refdst */
433 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
434 				     RT_TOS(tos), rt2->dst.dev);
435 
436 		dst_release(&rt2->dst);
437 		rt2 = skb_rtable(skb_in);
438 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
439 	}
440 
441 	if (err)
442 		goto relookup_failed;
443 
444 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
445 					    flowi4_to_flowi(&fl4_dec), NULL,
446 					    XFRM_LOOKUP_ICMP);
447 	if (!IS_ERR(rt2)) {
448 		dst_release(&rt->dst);
449 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
450 		rt = rt2;
451 	} else if (PTR_ERR(rt2) == -EPERM) {
452 		if (rt)
453 			dst_release(&rt->dst);
454 		return rt2;
455 	} else {
456 		err = PTR_ERR(rt2);
457 		goto relookup_failed;
458 	}
459 	return rt;
460 
461 relookup_failed:
462 	if (rt)
463 		return rt;
464 	return ERR_PTR(err);
465 }
466 
467 /*
468  *	Send an ICMP message in response to a situation
469  *
470  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
471  *		  MAY send more (we do).
472  *			MUST NOT change this header information.
473  *			MUST NOT reply to a multicast/broadcast IP address.
474  *			MUST NOT reply to a multicast/broadcast MAC address.
475  *			MUST reply to only the first fragment.
476  */
477 
478 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
479 {
480 	struct iphdr *iph;
481 	int room;
482 	struct icmp_bxm icmp_param;
483 	struct rtable *rt = skb_rtable(skb_in);
484 	struct ipcm_cookie ipc;
485 	struct flowi4 fl4;
486 	__be32 saddr;
487 	u8  tos;
488 	struct net *net;
489 	struct sock *sk;
490 
491 	if (!rt)
492 		goto out;
493 	net = dev_net(rt->dst.dev);
494 
495 	/*
496 	 *	Find the original header. It is expected to be valid, of course.
497 	 *	Check this, icmp_send is called from the most obscure devices
498 	 *	sometimes.
499 	 */
500 	iph = ip_hdr(skb_in);
501 
502 	if ((u8 *)iph < skb_in->head ||
503 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
504 		goto out;
505 
506 	/*
507 	 *	No replies to physical multicast/broadcast
508 	 */
509 	if (skb_in->pkt_type != PACKET_HOST)
510 		goto out;
511 
512 	/*
513 	 *	Now check at the protocol level
514 	 */
515 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
516 		goto out;
517 
518 	/*
519 	 *	Only reply to fragment 0. We byte re-order the constant
520 	 *	mask for efficiency.
521 	 */
522 	if (iph->frag_off & htons(IP_OFFSET))
523 		goto out;
524 
525 	/*
526 	 *	If we send an ICMP error to an ICMP error a mess would result..
527 	 */
528 	if (icmp_pointers[type].error) {
529 		/*
530 		 *	We are an error, check if we are replying to an
531 		 *	ICMP error
532 		 */
533 		if (iph->protocol == IPPROTO_ICMP) {
534 			u8 _inner_type, *itp;
535 
536 			itp = skb_header_pointer(skb_in,
537 						 skb_network_header(skb_in) +
538 						 (iph->ihl << 2) +
539 						 offsetof(struct icmphdr,
540 							  type) -
541 						 skb_in->data,
542 						 sizeof(_inner_type),
543 						 &_inner_type);
544 			if (itp == NULL)
545 				goto out;
546 
547 			/*
548 			 *	Assume any unknown ICMP type is an error. This
549 			 *	isn't specified by the RFC, but think about it..
550 			 */
551 			if (*itp > NR_ICMP_TYPES ||
552 			    icmp_pointers[*itp].error)
553 				goto out;
554 		}
555 	}
556 
557 	sk = icmp_xmit_lock(net);
558 	if (sk == NULL)
559 		return;
560 
561 	/*
562 	 *	Construct source address and options.
563 	 */
564 
565 	saddr = iph->daddr;
566 	if (!(rt->rt_flags & RTCF_LOCAL)) {
567 		struct net_device *dev = NULL;
568 
569 		rcu_read_lock();
570 		if (rt_is_input_route(rt) &&
571 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
572 			dev = dev_get_by_index_rcu(net, rt->rt_iif);
573 
574 		if (dev)
575 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
576 		else
577 			saddr = 0;
578 		rcu_read_unlock();
579 	}
580 
581 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
582 					   IPTOS_PREC_INTERNETCONTROL) :
583 					  iph->tos;
584 
585 	if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
586 		goto out_unlock;
587 
588 
589 	/*
590 	 *	Prepare data for ICMP header.
591 	 */
592 
593 	icmp_param.data.icmph.type	 = type;
594 	icmp_param.data.icmph.code	 = code;
595 	icmp_param.data.icmph.un.gateway = info;
596 	icmp_param.data.icmph.checksum	 = 0;
597 	icmp_param.skb	  = skb_in;
598 	icmp_param.offset = skb_network_offset(skb_in);
599 	inet_sk(sk)->tos = tos;
600 	ipc.addr = iph->saddr;
601 	ipc.opt = &icmp_param.replyopts.opt;
602 	ipc.tx_flags = 0;
603 
604 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
605 			       type, code, &icmp_param);
606 	if (IS_ERR(rt))
607 		goto out_unlock;
608 
609 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
610 		goto ende;
611 
612 	/* RFC says return as much as we can without exceeding 576 bytes. */
613 
614 	room = dst_mtu(&rt->dst);
615 	if (room > 576)
616 		room = 576;
617 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
618 	room -= sizeof(struct icmphdr);
619 
620 	icmp_param.data_len = skb_in->len - icmp_param.offset;
621 	if (icmp_param.data_len > room)
622 		icmp_param.data_len = room;
623 	icmp_param.head_len = sizeof(struct icmphdr);
624 
625 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
626 ende:
627 	ip_rt_put(rt);
628 out_unlock:
629 	icmp_xmit_unlock(sk);
630 out:;
631 }
632 EXPORT_SYMBOL(icmp_send);
633 
634 
635 /*
636  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
637  */
638 
639 static void icmp_unreach(struct sk_buff *skb)
640 {
641 	const struct iphdr *iph;
642 	struct icmphdr *icmph;
643 	int hash, protocol;
644 	const struct net_protocol *ipprot;
645 	u32 info = 0;
646 	struct net *net;
647 
648 	net = dev_net(skb_dst(skb)->dev);
649 
650 	/*
651 	 *	Incomplete header ?
652 	 * 	Only checks for the IP header, there should be an
653 	 *	additional check for longer headers in upper levels.
654 	 */
655 
656 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
657 		goto out_err;
658 
659 	icmph = icmp_hdr(skb);
660 	iph   = (const struct iphdr *)skb->data;
661 
662 	if (iph->ihl < 5) /* Mangled header, drop. */
663 		goto out_err;
664 
665 	if (icmph->type == ICMP_DEST_UNREACH) {
666 		switch (icmph->code & 15) {
667 		case ICMP_NET_UNREACH:
668 		case ICMP_HOST_UNREACH:
669 		case ICMP_PROT_UNREACH:
670 		case ICMP_PORT_UNREACH:
671 			break;
672 		case ICMP_FRAG_NEEDED:
673 			if (ipv4_config.no_pmtu_disc) {
674 				LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
675 					       &iph->daddr);
676 			} else {
677 				info = ip_rt_frag_needed(net, iph,
678 							 ntohs(icmph->un.frag.mtu),
679 							 skb->dev);
680 				if (!info)
681 					goto out;
682 			}
683 			break;
684 		case ICMP_SR_FAILED:
685 			LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
686 				       &iph->daddr);
687 			break;
688 		default:
689 			break;
690 		}
691 		if (icmph->code > NR_ICMP_UNREACH)
692 			goto out;
693 	} else if (icmph->type == ICMP_PARAMETERPROB)
694 		info = ntohl(icmph->un.gateway) >> 24;
695 
696 	/*
697 	 *	Throw it at our lower layers
698 	 *
699 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
700 	 *		  header.
701 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
702 	 *		  transport layer.
703 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
704 	 *		  transport layer.
705 	 */
706 
707 	/*
708 	 *	Check the other end isn't violating RFC 1122. Some routers send
709 	 *	bogus responses to broadcast frames. If you see this message
710 	 *	first check your netmask matches at both ends, if it does then
711 	 *	get the other vendor to fix their kit.
712 	 */
713 
714 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
715 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
716 		if (net_ratelimit())
717 			pr_warn("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
718 				&ip_hdr(skb)->saddr,
719 				icmph->type, icmph->code,
720 				&iph->daddr, skb->dev->name);
721 		goto out;
722 	}
723 
724 	/* Checkin full IP header plus 8 bytes of protocol to
725 	 * avoid additional coding at protocol handlers.
726 	 */
727 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
728 		goto out;
729 
730 	iph = (const struct iphdr *)skb->data;
731 	protocol = iph->protocol;
732 
733 	/*
734 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
735 	 */
736 	raw_icmp_error(skb, protocol, info);
737 
738 	hash = protocol & (MAX_INET_PROTOS - 1);
739 	rcu_read_lock();
740 	ipprot = rcu_dereference(inet_protos[hash]);
741 	if (ipprot && ipprot->err_handler)
742 		ipprot->err_handler(skb, info);
743 	rcu_read_unlock();
744 
745 out:
746 	return;
747 out_err:
748 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
749 	goto out;
750 }
751 
752 
753 /*
754  *	Handle ICMP_REDIRECT.
755  */
756 
757 static void icmp_redirect(struct sk_buff *skb)
758 {
759 	const struct iphdr *iph;
760 
761 	if (skb->len < sizeof(struct iphdr))
762 		goto out_err;
763 
764 	/*
765 	 *	Get the copied header of the packet that caused the redirect
766 	 */
767 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
768 		goto out;
769 
770 	iph = (const struct iphdr *)skb->data;
771 
772 	switch (icmp_hdr(skb)->code & 7) {
773 	case ICMP_REDIR_NET:
774 	case ICMP_REDIR_NETTOS:
775 		/*
776 		 * As per RFC recommendations now handle it as a host redirect.
777 		 */
778 	case ICMP_REDIR_HOST:
779 	case ICMP_REDIR_HOSTTOS:
780 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
781 			       icmp_hdr(skb)->un.gateway,
782 			       iph->saddr, skb->dev);
783 		break;
784 	}
785 
786 	/* Ping wants to see redirects.
787          * Let's pretend they are errors of sorts... */
788 	if (iph->protocol == IPPROTO_ICMP &&
789 	    iph->ihl >= 5 &&
790 	    pskb_may_pull(skb, (iph->ihl<<2)+8)) {
791 		ping_err(skb, icmp_hdr(skb)->un.gateway);
792 	}
793 
794 out:
795 	return;
796 out_err:
797 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
798 	goto out;
799 }
800 
801 /*
802  *	Handle ICMP_ECHO ("ping") requests.
803  *
804  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
805  *		  requests.
806  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
807  *		  included in the reply.
808  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
809  *		  echo requests, MUST have default=NOT.
810  *	See also WRT handling of options once they are done and working.
811  */
812 
813 static void icmp_echo(struct sk_buff *skb)
814 {
815 	struct net *net;
816 
817 	net = dev_net(skb_dst(skb)->dev);
818 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
819 		struct icmp_bxm icmp_param;
820 
821 		icmp_param.data.icmph	   = *icmp_hdr(skb);
822 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
823 		icmp_param.skb		   = skb;
824 		icmp_param.offset	   = 0;
825 		icmp_param.data_len	   = skb->len;
826 		icmp_param.head_len	   = sizeof(struct icmphdr);
827 		icmp_reply(&icmp_param, skb);
828 	}
829 }
830 
831 /*
832  *	Handle ICMP Timestamp requests.
833  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
834  *		  SHOULD be in the kernel for minimum random latency.
835  *		  MUST be accurate to a few minutes.
836  *		  MUST be updated at least at 15Hz.
837  */
838 static void icmp_timestamp(struct sk_buff *skb)
839 {
840 	struct timespec tv;
841 	struct icmp_bxm icmp_param;
842 	/*
843 	 *	Too short.
844 	 */
845 	if (skb->len < 4)
846 		goto out_err;
847 
848 	/*
849 	 *	Fill in the current time as ms since midnight UT:
850 	 */
851 	getnstimeofday(&tv);
852 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
853 					 tv.tv_nsec / NSEC_PER_MSEC);
854 	icmp_param.data.times[2] = icmp_param.data.times[1];
855 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
856 		BUG();
857 	icmp_param.data.icmph	   = *icmp_hdr(skb);
858 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
859 	icmp_param.data.icmph.code = 0;
860 	icmp_param.skb		   = skb;
861 	icmp_param.offset	   = 0;
862 	icmp_param.data_len	   = 0;
863 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
864 	icmp_reply(&icmp_param, skb);
865 out:
866 	return;
867 out_err:
868 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
869 	goto out;
870 }
871 
872 
873 /*
874  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
875  *
876  * RFC1122 (3.2.2.9).  A host MUST only send replies to
877  * ADDRESS_MASK requests if it's been configured as an address mask
878  * agent.  Receiving a request doesn't constitute implicit permission to
879  * act as one. Of course, implementing this correctly requires (SHOULD)
880  * a way to turn the functionality on and off.  Another one for sysctl(),
881  * I guess. -- MS
882  *
883  * RFC1812 (4.3.3.9).	A router MUST implement it.
884  *			A router SHOULD have switch turning it on/off.
885  *		      	This switch MUST be ON by default.
886  *
887  * Gratuitous replies, zero-source replies are not implemented,
888  * that complies with RFC. DO NOT implement them!!! All the idea
889  * of broadcast addrmask replies as specified in RFC950 is broken.
890  * The problem is that it is not uncommon to have several prefixes
891  * on one physical interface. Moreover, addrmask agent can even be
892  * not aware of existing another prefixes.
893  * If source is zero, addrmask agent cannot choose correct prefix.
894  * Gratuitous mask announcements suffer from the same problem.
895  * RFC1812 explains it, but still allows to use ADDRMASK,
896  * that is pretty silly. --ANK
897  *
898  * All these rules are so bizarre, that I removed kernel addrmask
899  * support at all. It is wrong, it is obsolete, nobody uses it in
900  * any case. --ANK
901  *
902  * Furthermore you can do it with a usermode address agent program
903  * anyway...
904  */
905 
906 static void icmp_address(struct sk_buff *skb)
907 {
908 #if 0
909 	if (net_ratelimit())
910 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
911 #endif
912 }
913 
914 /*
915  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
916  *			loudly if an inconsistency is found.
917  * called with rcu_read_lock()
918  */
919 
920 static void icmp_address_reply(struct sk_buff *skb)
921 {
922 	struct rtable *rt = skb_rtable(skb);
923 	struct net_device *dev = skb->dev;
924 	struct in_device *in_dev;
925 	struct in_ifaddr *ifa;
926 
927 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
928 		return;
929 
930 	in_dev = __in_dev_get_rcu(dev);
931 	if (!in_dev)
932 		return;
933 
934 	if (in_dev->ifa_list &&
935 	    IN_DEV_LOG_MARTIANS(in_dev) &&
936 	    IN_DEV_FORWARD(in_dev)) {
937 		__be32 _mask, *mp;
938 
939 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
940 		BUG_ON(mp == NULL);
941 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
942 			if (*mp == ifa->ifa_mask &&
943 			    inet_ifa_match(ip_hdr(skb)->saddr, ifa))
944 				break;
945 		}
946 		if (!ifa && net_ratelimit()) {
947 			pr_info("Wrong address mask %pI4 from %s/%pI4\n",
948 				mp, dev->name, &ip_hdr(skb)->saddr);
949 		}
950 	}
951 }
952 
953 static void icmp_discard(struct sk_buff *skb)
954 {
955 }
956 
957 /*
958  *	Deal with incoming ICMP packets.
959  */
960 int icmp_rcv(struct sk_buff *skb)
961 {
962 	struct icmphdr *icmph;
963 	struct rtable *rt = skb_rtable(skb);
964 	struct net *net = dev_net(rt->dst.dev);
965 
966 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
967 		struct sec_path *sp = skb_sec_path(skb);
968 		int nh;
969 
970 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
971 				 XFRM_STATE_ICMP))
972 			goto drop;
973 
974 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
975 			goto drop;
976 
977 		nh = skb_network_offset(skb);
978 		skb_set_network_header(skb, sizeof(*icmph));
979 
980 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
981 			goto drop;
982 
983 		skb_set_network_header(skb, nh);
984 	}
985 
986 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
987 
988 	switch (skb->ip_summed) {
989 	case CHECKSUM_COMPLETE:
990 		if (!csum_fold(skb->csum))
991 			break;
992 		/* fall through */
993 	case CHECKSUM_NONE:
994 		skb->csum = 0;
995 		if (__skb_checksum_complete(skb))
996 			goto error;
997 	}
998 
999 	if (!pskb_pull(skb, sizeof(*icmph)))
1000 		goto error;
1001 
1002 	icmph = icmp_hdr(skb);
1003 
1004 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1005 	/*
1006 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1007 	 *
1008 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1009 	 *		  discarded.
1010 	 */
1011 	if (icmph->type > NR_ICMP_TYPES)
1012 		goto error;
1013 
1014 
1015 	/*
1016 	 *	Parse the ICMP message
1017 	 */
1018 
1019 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1020 		/*
1021 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1022 		 *	  silently ignored (we let user decide with a sysctl).
1023 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1024 		 *	  discarded if to broadcast/multicast.
1025 		 */
1026 		if ((icmph->type == ICMP_ECHO ||
1027 		     icmph->type == ICMP_TIMESTAMP) &&
1028 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1029 			goto error;
1030 		}
1031 		if (icmph->type != ICMP_ECHO &&
1032 		    icmph->type != ICMP_TIMESTAMP &&
1033 		    icmph->type != ICMP_ADDRESS &&
1034 		    icmph->type != ICMP_ADDRESSREPLY) {
1035 			goto error;
1036 		}
1037 	}
1038 
1039 	icmp_pointers[icmph->type].handler(skb);
1040 
1041 drop:
1042 	kfree_skb(skb);
1043 	return 0;
1044 error:
1045 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1046 	goto drop;
1047 }
1048 
1049 /*
1050  *	This table is the definition of how we handle ICMP.
1051  */
1052 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1053 	[ICMP_ECHOREPLY] = {
1054 		.handler = ping_rcv,
1055 	},
1056 	[1] = {
1057 		.handler = icmp_discard,
1058 		.error = 1,
1059 	},
1060 	[2] = {
1061 		.handler = icmp_discard,
1062 		.error = 1,
1063 	},
1064 	[ICMP_DEST_UNREACH] = {
1065 		.handler = icmp_unreach,
1066 		.error = 1,
1067 	},
1068 	[ICMP_SOURCE_QUENCH] = {
1069 		.handler = icmp_unreach,
1070 		.error = 1,
1071 	},
1072 	[ICMP_REDIRECT] = {
1073 		.handler = icmp_redirect,
1074 		.error = 1,
1075 	},
1076 	[6] = {
1077 		.handler = icmp_discard,
1078 		.error = 1,
1079 	},
1080 	[7] = {
1081 		.handler = icmp_discard,
1082 		.error = 1,
1083 	},
1084 	[ICMP_ECHO] = {
1085 		.handler = icmp_echo,
1086 	},
1087 	[9] = {
1088 		.handler = icmp_discard,
1089 		.error = 1,
1090 	},
1091 	[10] = {
1092 		.handler = icmp_discard,
1093 		.error = 1,
1094 	},
1095 	[ICMP_TIME_EXCEEDED] = {
1096 		.handler = icmp_unreach,
1097 		.error = 1,
1098 	},
1099 	[ICMP_PARAMETERPROB] = {
1100 		.handler = icmp_unreach,
1101 		.error = 1,
1102 	},
1103 	[ICMP_TIMESTAMP] = {
1104 		.handler = icmp_timestamp,
1105 	},
1106 	[ICMP_TIMESTAMPREPLY] = {
1107 		.handler = icmp_discard,
1108 	},
1109 	[ICMP_INFO_REQUEST] = {
1110 		.handler = icmp_discard,
1111 	},
1112 	[ICMP_INFO_REPLY] = {
1113 		.handler = icmp_discard,
1114 	},
1115 	[ICMP_ADDRESS] = {
1116 		.handler = icmp_address,
1117 	},
1118 	[ICMP_ADDRESSREPLY] = {
1119 		.handler = icmp_address_reply,
1120 	},
1121 };
1122 
1123 static void __net_exit icmp_sk_exit(struct net *net)
1124 {
1125 	int i;
1126 
1127 	for_each_possible_cpu(i)
1128 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1129 	kfree(net->ipv4.icmp_sk);
1130 	net->ipv4.icmp_sk = NULL;
1131 }
1132 
1133 static int __net_init icmp_sk_init(struct net *net)
1134 {
1135 	int i, err;
1136 
1137 	net->ipv4.icmp_sk =
1138 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1139 	if (net->ipv4.icmp_sk == NULL)
1140 		return -ENOMEM;
1141 
1142 	for_each_possible_cpu(i) {
1143 		struct sock *sk;
1144 
1145 		err = inet_ctl_sock_create(&sk, PF_INET,
1146 					   SOCK_RAW, IPPROTO_ICMP, net);
1147 		if (err < 0)
1148 			goto fail;
1149 
1150 		net->ipv4.icmp_sk[i] = sk;
1151 
1152 		/* Enough space for 2 64K ICMP packets, including
1153 		 * sk_buff/skb_shared_info struct overhead.
1154 		 */
1155 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1156 
1157 		/*
1158 		 * Speedup sock_wfree()
1159 		 */
1160 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1161 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1162 	}
1163 
1164 	/* Control parameters for ECHO replies. */
1165 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1166 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1167 
1168 	/* Control parameter - ignore bogus broadcast responses? */
1169 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1170 
1171 	/*
1172 	 * 	Configurable global rate limit.
1173 	 *
1174 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1175 	 *	bucket ratemask defines which icmp types are ratelimited by
1176 	 *	setting	it's bit position.
1177 	 *
1178 	 *	default:
1179 	 *	dest unreachable (3), source quench (4),
1180 	 *	time exceeded (11), parameter problem (12)
1181 	 */
1182 
1183 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1184 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1185 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1186 
1187 	return 0;
1188 
1189 fail:
1190 	for_each_possible_cpu(i)
1191 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1192 	kfree(net->ipv4.icmp_sk);
1193 	return err;
1194 }
1195 
1196 static struct pernet_operations __net_initdata icmp_sk_ops = {
1197        .init = icmp_sk_init,
1198        .exit = icmp_sk_exit,
1199 };
1200 
1201 int __init icmp_init(void)
1202 {
1203 	return register_pernet_subsys(&icmp_sk_ops);
1204 }
1205