1 /*
2  *	NET3:	Implementation of the ICMP protocol layer.
3  *
4  *		Alan Cox, <alan@redhat.com>
5  *
6  *	Version: $Id: icmp.c,v 1.82.2.1 2001/12/13 08:59:27 davem Exp $
7  *
8  *	This program is free software; you can redistribute it and/or
9  *	modify it under the terms of the GNU General Public License
10  *	as published by the Free Software Foundation; either version
11  *	2 of the License, or (at your option) any later version.
12  *
13  *	Some of the function names and the icmp unreach table for this
14  *	module were derived from [icmp.c 1.0.11 06/02/93] by
15  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
16  *	Other than that this module is a complete rewrite.
17  *
18  *	Fixes:
19  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
20  *					with icmp type masking ability instead
21  *					of broken per type icmp timeouts.
22  *		Mike Shaver	:	RFC1122 checks.
23  *		Alan Cox	:	Multicast ping reply as self.
24  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
25  *					call.
26  *		Alan Cox	:	Added 216,128 byte paths to the MTU
27  *					code.
28  *		Martin Mares	:	RFC1812 checks.
29  *		Martin Mares	:	Can be configured to follow redirects
30  *					if acting as a router _without_ a
31  *					routing protocol (RFC 1812).
32  *		Martin Mares	:	Echo requests may be configured to
33  *					be ignored (RFC 1812).
34  *		Martin Mares	:	Limitation of ICMP error message
35  *					transmit rate (RFC 1812).
36  *		Martin Mares	:	TOS and Precedence set correctly
37  *					(RFC 1812).
38  *		Martin Mares	:	Now copying as much data from the
39  *					original packet as we can without
40  *					exceeding 576 bytes (RFC 1812).
41  *	Willy Konynenberg	:	Transparent proxying support.
42  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
43  *					path MTU bug.
44  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
45  *					valid (RFC 1812).
46  *		Andi Kleen	:	Check all packet lengths properly
47  *					and moved all kfree_skb() up to
48  *					icmp_rcv.
49  *		Andi Kleen	:	Move the rate limit bookkeeping
50  *					into the dest entry and use a token
51  *					bucket filter (thanks to ANK). Make
52  *					the rates sysctl configurable.
53  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
54  *					- IP option length was accounted wrongly
55  *					- ICMP header length was not accounted at all.
56  *              Tristan Greaves :       Added sysctl option to ignore bogus broadcast
57  *                                      responses from broken routers.
58  *
59  * To Fix:
60  *
61  *	- Should use skb_pull() instead of all the manual checking.
62  *	  This would also greatly simply some upper layer error handlers. --AK
63  *
64  */
65 
66 #include <linux/config.h>
67 #include <linux/types.h>
68 #include <linux/sched.h>
69 #include <linux/kernel.h>
70 #include <linux/fcntl.h>
71 #include <linux/socket.h>
72 #include <linux/in.h>
73 #include <linux/inet.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 
94 /*
95  *	Build xmit assembly blocks
96  */
97 
98 struct icmp_bxm
99 {
100 	struct sk_buff *skb;
101 	int offset;
102 	int data_len;
103 
104 	unsigned int csum;
105 	struct {
106 		struct icmphdr icmph;
107 		__u32	       times[3];
108 	} data;
109 	int head_len;
110 	struct ip_options replyopts;
111 	unsigned char  optbuf[40];
112 };
113 
114 /*
115  *	Statistics
116  */
117 
118 struct icmp_mib icmp_statistics[NR_CPUS*2];
119 
120 /* An array of errno for error messages from dest unreach. */
121 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOS_UNREACH and SR_FAIELD MUST be considered 'transient errs'. */
122 
123 struct icmp_err icmp_err_convert[] = {
124   { ENETUNREACH,	0 },	/*	ICMP_NET_UNREACH	*/
125   { EHOSTUNREACH,	0 },	/*	ICMP_HOST_UNREACH	*/
126   { ENOPROTOOPT,	1 },	/*	ICMP_PROT_UNREACH	*/
127   { ECONNREFUSED,	1 },	/*	ICMP_PORT_UNREACH	*/
128   { EMSGSIZE,		0 },	/*	ICMP_FRAG_NEEDED	*/
129   { EOPNOTSUPP,		0 },	/*	ICMP_SR_FAILED		*/
130   { ENETUNREACH,	1 },	/* 	ICMP_NET_UNKNOWN	*/
131   { EHOSTDOWN,		1 },	/*	ICMP_HOST_UNKNOWN	*/
132   { ENONET,		1 },	/*	ICMP_HOST_ISOLATED	*/
133   { ENETUNREACH,	1 },	/*	ICMP_NET_ANO		*/
134   { EHOSTUNREACH,	1 },	/*	ICMP_HOST_ANO		*/
135   { ENETUNREACH,	0 },	/*	ICMP_NET_UNR_TOS	*/
136   { EHOSTUNREACH,	0 },	/*	ICMP_HOST_UNR_TOS	*/
137   { EHOSTUNREACH,	1 },	/*	ICMP_PKT_FILTERED	*/
138   { EHOSTUNREACH,	1 },	/*	ICMP_PREC_VIOLATION	*/
139   { EHOSTUNREACH,	1 }	/*	ICMP_PREC_CUTOFF	*/
140 };
141 
142 extern int sysctl_ip_default_ttl;
143 
144 /* Control parameters for ECHO replies. */
145 int sysctl_icmp_echo_ignore_all;
146 int sysctl_icmp_echo_ignore_broadcasts;
147 
148 /* Control parameter - ignore bogus broadcast responses? */
149 int sysctl_icmp_ignore_bogus_error_responses;
150 
151 /*
152  * 	Configurable global rate limit.
153  *
154  *	ratelimit defines tokens/packet consumed for dst->rate_token bucket
155  *	ratemask defines which icmp types are ratelimited by setting
156  * 	it's bit position.
157  *
158  *	default:
159  *	dest unreachable (3), source quench (4),
160  *	time exceeded (11), parameter problem (12)
161  */
162 
163 int sysctl_icmp_ratelimit = 1*HZ;
164 int sysctl_icmp_ratemask = 0x1818;
165 
166 /*
167  *	ICMP control array. This specifies what to do with each ICMP.
168  */
169 
170 struct icmp_control
171 {
172 	unsigned long *output;		/* Address to increment on output */
173 	unsigned long *input;		/* Address to increment on input */
174 	void (*handler)(struct sk_buff *skb);
175 	short	error;		/* This ICMP is classed as an error message */
176 };
177 
178 static struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
179 
180 /*
181  *	The ICMP socket(s). This is the most convenient way to flow control
182  *	our ICMP output as well as maintain a clean interface throughout
183  *	all layers. All Socketless IP sends will soon be gone.
184  */
185 
186 static struct inode __icmp_inode[NR_CPUS];
187 #define icmp_socket (&__icmp_inode[smp_processor_id()].u.socket_i)
188 #define icmp_socket_cpu(X) (&__icmp_inode[(X)].u.socket_i)
189 
icmp_xmit_lock(void)190 static int icmp_xmit_lock(void)
191 {
192 	local_bh_disable();
193 	if (unlikely(!spin_trylock(&icmp_socket->sk->lock.slock))) {
194 		/* This can happen if the output path signals a
195 		 * dst_link_failure() for an outgoing ICMP packet.
196 		 */
197 		local_bh_enable();
198 		return 1;
199 	}
200 	return 0;
201 }
202 
icmp_xmit_unlock(void)203 static void icmp_xmit_unlock(void)
204 {
205 	spin_unlock_bh(&icmp_socket->sk->lock.slock);
206 }
207 
208 /*
209  *	Send an ICMP frame.
210  */
211 
212 /*
213  *	Check transmit rate limitation for given message.
214  *	The rate information is held in the destination cache now.
215  *	This function is generic and could be used for other purposes
216  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
217  *
218  *	Note that the same dst_entry fields are modified by functions in
219  *	route.c too, but these work for packet destinations while xrlim_allow
220  *	works for icmp destinations. This means the rate limiting information
221  *	for one "ip object" is shared - and these ICMPs are twice limited:
222  *	by source and by destination.
223  *
224  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
225  *			  SHOULD allow setting of rate limits
226  *
227  * 	Shared between ICMPv4 and ICMPv6.
228  */
229 #define XRLIM_BURST_FACTOR 6
xrlim_allow(struct dst_entry * dst,int timeout)230 int xrlim_allow(struct dst_entry *dst, int timeout)
231 {
232 	unsigned long now;
233 
234 	now = jiffies;
235 	dst->rate_tokens += now - dst->rate_last;
236 	dst->rate_last = now;
237 	if (dst->rate_tokens > XRLIM_BURST_FACTOR*timeout)
238         	dst->rate_tokens = XRLIM_BURST_FACTOR*timeout;
239 	if (dst->rate_tokens >= timeout) {
240 		dst->rate_tokens -= timeout;
241 		return 1;
242 	}
243 	return 0;
244 }
245 
icmpv4_xrlim_allow(struct rtable * rt,int type,int code)246 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
247 {
248 	struct dst_entry *dst = &rt->u.dst;
249 
250 	if (type > NR_ICMP_TYPES)
251 		return 1;
252 
253 	/* Don't limit PMTU discovery. */
254 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
255 		return 1;
256 
257 	/* No rate limit on loopback */
258 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
259  		return 1;
260 
261 	/* Limit if icmp type is enabled in ratemask. */
262 	if((1 << type) & sysctl_icmp_ratemask)
263 		return xrlim_allow(dst, sysctl_icmp_ratelimit);
264 	else
265 		return 1;
266 }
267 
268 /*
269  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
270  */
271 
icmp_out_count(int type)272 static void icmp_out_count(int type)
273 {
274 	if (type>NR_ICMP_TYPES)
275 		return;
276 	(icmp_pointers[type].output)[(smp_processor_id()*2+!in_softirq())*sizeof(struct icmp_mib)/sizeof(unsigned long)]++;
277 	ICMP_INC_STATS(IcmpOutMsgs);
278 }
279 
280 /*
281  *	Checksum each fragment, and on the first include the headers and final checksum.
282  */
283 
icmp_glue_bits(const void * p,char * to,unsigned int offset,unsigned int fraglen,struct sk_buff * skb)284 static int icmp_glue_bits(const void *p, char *to, unsigned int offset,
285                           unsigned int fraglen, struct sk_buff *skb)
286 {
287 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)p;
288 	struct icmphdr *icmph;
289 	unsigned int csum;
290 
291 	if (icmp_pointers[icmp_param->data.icmph.type].error)
292 		nf_ct_attach(skb, icmp_param->skb);
293 
294 	if (offset) {
295 		icmp_param->csum=skb_copy_and_csum_bits(icmp_param->skb,
296 							icmp_param->offset+(offset-icmp_param->head_len),
297 							to, fraglen,icmp_param->csum);
298 		return 0;
299 	}
300 
301 	/*
302 	 *	First fragment includes header. Note that we've done
303 	 *	the other fragments first, so that we get the checksum
304 	 *	for the whole packet here.
305 	 */
306 	csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
307 		to, icmp_param->head_len,
308 		icmp_param->csum);
309 	csum=skb_copy_and_csum_bits(icmp_param->skb,
310 				    icmp_param->offset,
311 				    to+icmp_param->head_len,
312 				    fraglen-icmp_param->head_len,
313 				    csum);
314 	icmph=(struct icmphdr *)to;
315 	icmph->checksum = csum_fold(csum);
316 	return 0;
317 }
318 
319 /*
320  *	Driving logic for building and sending ICMP messages.
321  */
322 
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)323 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
324 {
325 	struct sock *sk=icmp_socket->sk;
326 	struct ipcm_cookie ipc;
327 	struct rtable *rt = (struct rtable*)skb->dst;
328 	u32 daddr;
329 
330 	if (ip_options_echo(&icmp_param->replyopts, skb))
331 		return;
332 
333 	if (icmp_xmit_lock())
334 		return;
335 
336 	icmp_param->data.icmph.checksum=0;
337 	icmp_param->csum=0;
338 	icmp_out_count(icmp_param->data.icmph.type);
339 
340 	sk->protinfo.af_inet.tos = skb->nh.iph->tos;
341 	sk->protinfo.af_inet.ttl = sysctl_ip_default_ttl;
342 	daddr = ipc.addr = rt->rt_src;
343 	ipc.opt = NULL;
344 	if (icmp_param->replyopts.optlen) {
345 		ipc.opt = &icmp_param->replyopts;
346 		if (ipc.opt->srr)
347 			daddr = icmp_param->replyopts.faddr;
348 	}
349 	if (ip_route_output(&rt, daddr, rt->rt_spec_dst, RT_TOS(skb->nh.iph->tos), 0))
350 		goto out;
351 	if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
352 			       icmp_param->data.icmph.code)) {
353 		ip_build_xmit(sk, icmp_glue_bits, icmp_param,
354 			      icmp_param->data_len+icmp_param->head_len,
355 			      &ipc, rt, MSG_DONTWAIT);
356 	}
357 	ip_rt_put(rt);
358 out:
359 	icmp_xmit_unlock();
360 }
361 
362 
363 /*
364  *	Send an ICMP message in response to a situation
365  *
366  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header. MAY send more (we do).
367  *			MUST NOT change this header information.
368  *			MUST NOT reply to a multicast/broadcast IP address.
369  *			MUST NOT reply to a multicast/broadcast MAC address.
370  *			MUST reply to only the first fragment.
371  */
372 
icmp_send(struct sk_buff * skb_in,int type,int code,u32 info)373 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info)
374 {
375 	struct iphdr *iph;
376 	int room;
377 	struct icmp_bxm icmp_param;
378 	struct rtable *rt = (struct rtable*)skb_in->dst;
379 	struct ipcm_cookie ipc;
380 	u32 saddr;
381 	u8  tos;
382 
383 	if (!rt)
384 		return;
385 
386 	/*
387 	 *	Find the original header. It is expected to be valid, of course.
388 	 *	Check this, icmp_send is called from the most obscure devices
389 	 *	sometimes.
390 	 */
391 	iph = skb_in->nh.iph;
392 
393 	if ((u8*)iph < skb_in->head || (u8*)(iph+1) > skb_in->tail)
394 		return;
395 
396 	/*
397 	 *	No replies to physical multicast/broadcast
398 	 */
399 	if (skb_in->pkt_type!=PACKET_HOST)
400 		return;
401 
402 	/*
403 	 *	Now check at the protocol level
404 	 */
405 	if (rt->rt_flags&(RTCF_BROADCAST|RTCF_MULTICAST))
406 		return;
407 
408 	/*
409 	 *	Only reply to fragment 0. We byte re-order the constant
410 	 *	mask for efficiency.
411 	 */
412 	if (iph->frag_off&htons(IP_OFFSET))
413 		return;
414 
415 	/*
416 	 *	If we send an ICMP error to an ICMP error a mess would result..
417 	 */
418 	if (icmp_pointers[type].error) {
419 		/*
420 		 *	We are an error, check if we are replying to an ICMP error
421 		 */
422 		if (iph->protocol==IPPROTO_ICMP) {
423 			u8 inner_type;
424 
425 			if (skb_copy_bits(skb_in,
426 					  skb_in->nh.raw + (iph->ihl<<2)
427 					  + offsetof(struct icmphdr, type)
428 					  - skb_in->data,
429 					  &inner_type, 1))
430 				return;
431 
432 			/*
433 			 *	Assume any unknown ICMP type is an error. This isn't
434 			 *	specified by the RFC, but think about it..
435 			 */
436 			if (inner_type>NR_ICMP_TYPES || icmp_pointers[inner_type].error)
437 				return;
438 		}
439 	}
440 
441 	if (icmp_xmit_lock())
442 		return;
443 
444 	/*
445 	 *	Construct source address and options.
446 	 */
447 
448 #ifdef CONFIG_IP_ROUTE_NAT
449 	/*
450 	 *	Restore original addresses if packet has been translated.
451 	 */
452 	if (rt->rt_flags&RTCF_NAT && IPCB(skb_in)->flags&IPSKB_TRANSLATED) {
453 		iph->daddr = rt->key.dst;
454 		iph->saddr = rt->key.src;
455 	}
456 #endif
457 
458 	saddr = iph->daddr;
459 	if (!(rt->rt_flags & RTCF_LOCAL))
460 		saddr = 0;
461 
462 	tos = icmp_pointers[type].error ?
463 		((iph->tos & IPTOS_TOS_MASK) | IPTOS_PREC_INTERNETCONTROL) :
464 			iph->tos;
465 
466 	if (ip_route_output(&rt, iph->saddr, saddr, RT_TOS(tos), 0))
467 		goto out;
468 
469 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
470 		goto ende;
471 
472 
473 	/*
474 	 *	Prepare data for ICMP header.
475 	 */
476 
477 	icmp_param.data.icmph.type=type;
478 	icmp_param.data.icmph.code=code;
479 	icmp_param.data.icmph.un.gateway = info;
480 	icmp_param.data.icmph.checksum=0;
481 	icmp_param.csum=0;
482 	icmp_param.skb=skb_in;
483 	icmp_param.offset=skb_in->nh.raw - skb_in->data;
484 	icmp_out_count(icmp_param.data.icmph.type);
485 	icmp_socket->sk->protinfo.af_inet.tos = tos;
486 	icmp_socket->sk->protinfo.af_inet.ttl = sysctl_ip_default_ttl;
487 	ipc.addr = iph->saddr;
488 	ipc.opt = &icmp_param.replyopts;
489 	if (icmp_param.replyopts.srr) {
490 		ip_rt_put(rt);
491 		if (ip_route_output(&rt, icmp_param.replyopts.faddr, saddr, RT_TOS(tos), 0))
492 			goto out;
493 	}
494 
495 	if (!icmpv4_xrlim_allow(rt, type, code))
496 		goto ende;
497 
498 	/* RFC says return as much as we can without exceeding 576 bytes. */
499 
500 	room = rt->u.dst.pmtu;
501 	if (room > 576)
502 		room = 576;
503 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
504 	room -= sizeof(struct icmphdr);
505 
506 	icmp_param.data_len=skb_in->len-icmp_param.offset;
507 	if (icmp_param.data_len > room)
508 		icmp_param.data_len = room;
509 	icmp_param.head_len = sizeof(struct icmphdr);
510 
511 	ip_build_xmit(icmp_socket->sk, icmp_glue_bits, &icmp_param,
512 		icmp_param.data_len+sizeof(struct icmphdr),
513 		&ipc, rt, MSG_DONTWAIT);
514 
515 ende:
516 	ip_rt_put(rt);
517 out:
518 	icmp_xmit_unlock();
519 }
520 
521 
522 /*
523  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
524  */
525 
icmp_unreach(struct sk_buff * skb)526 static void icmp_unreach(struct sk_buff *skb)
527 {
528 	struct iphdr *iph;
529 	struct icmphdr *icmph;
530 	int hash, protocol;
531 	struct inet_protocol *ipprot;
532 	struct sock *raw_sk;
533 	u32 info = 0;
534 
535 	/*
536 	 *	Incomplete header ?
537 	 * 	Only checks for the IP header, there should be an
538 	 *	additional check for longer headers in upper levels.
539 	 */
540 
541 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
542 		ICMP_INC_STATS_BH(IcmpInErrors);
543 		return;
544 	}
545 
546 	icmph = skb->h.icmph;
547 	iph = (struct iphdr *) skb->data;
548 
549 	if (iph->ihl<5) {
550 		/* Mangled header, drop. */
551 		ICMP_INC_STATS_BH(IcmpInErrors);
552 		return;
553 	}
554 
555 	if(icmph->type==ICMP_DEST_UNREACH) {
556 		switch(icmph->code & 15) {
557 			case ICMP_NET_UNREACH:
558 				break;
559 			case ICMP_HOST_UNREACH:
560 				break;
561 			case ICMP_PROT_UNREACH:
562 				break;
563 			case ICMP_PORT_UNREACH:
564 				break;
565 			case ICMP_FRAG_NEEDED:
566 				if (ipv4_config.no_pmtu_disc) {
567 					if (net_ratelimit())
568 						printk(KERN_INFO "ICMP: %u.%u.%u.%u: fragmentation needed and DF set.\n",
569 						       NIPQUAD(iph->daddr));
570 				} else {
571 					info = ip_rt_frag_needed(iph, ntohs(icmph->un.frag.mtu));
572 					if (!info)
573 						goto out;
574 				}
575 				break;
576 			case ICMP_SR_FAILED:
577 				if (net_ratelimit())
578 					printk(KERN_INFO "ICMP: %u.%u.%u.%u: Source Route Failed.\n", NIPQUAD(iph->daddr));
579 				break;
580 			default:
581 				break;
582 		}
583 		if (icmph->code>NR_ICMP_UNREACH)
584 			goto out;
585 	} else if (icmph->type == ICMP_PARAMETERPROB) {
586 		info = ntohl(icmph->un.gateway)>>24;
587 	}
588 
589 	/*
590 	 *	Throw it at our lower layers
591 	 *
592 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed header.
593 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the transport layer.
594 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to transport layer.
595 	 */
596 
597 	/*
598 	 *	Check the other end isnt violating RFC 1122. Some routers send
599 	 *	bogus responses to broadcast frames. If you see this message
600 	 *	first check your netmask matches at both ends, if it does then
601 	 *	get the other vendor to fix their kit.
602 	 */
603 
604 	if (!sysctl_icmp_ignore_bogus_error_responses)
605 	{
606 
607 		if (inet_addr_type(iph->daddr) == RTN_BROADCAST)
608 		{
609 			if (net_ratelimit())
610 				printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP type %u, code %u error to a broadcast: %u.%u.%u.%u on %s\n",
611 					NIPQUAD(skb->nh.iph->saddr),
612 					icmph->type, icmph->code,
613 					NIPQUAD(iph->daddr),
614 					skb->dev->name);
615 			goto out;
616 		}
617 	}
618 
619 	/* Checkin full IP header plus 8 bytes of protocol to
620 	 * avoid additional coding at protocol handlers.
621 	 */
622 	if (!pskb_may_pull(skb, iph->ihl*4+8))
623 		goto out;
624 
625 	iph = (struct iphdr *) skb->data;
626 	protocol = iph->protocol;
627 
628 	/*
629 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
630 	 */
631 
632 	/* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
633 	hash = protocol & (MAX_INET_PROTOS - 1);
634 	read_lock(&raw_v4_lock);
635 	if ((raw_sk = raw_v4_htable[hash]) != NULL)
636 	{
637 		while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
638 						 iph->saddr, skb->dev->ifindex)) != NULL) {
639 			raw_err(raw_sk, skb, info);
640 			raw_sk = raw_sk->next;
641 			iph = (struct iphdr *)skb->data;
642 		}
643 	}
644 	read_unlock(&raw_v4_lock);
645 
646 	/*
647 	 *	This can't change while we are doing it.
648 	 *	Callers have obtained BR_NETPROTO_LOCK so
649 	 *	we are OK.
650 	 */
651 
652 	ipprot = (struct inet_protocol *) inet_protos[hash];
653 	while (ipprot) {
654 		struct inet_protocol *nextip;
655 
656 		nextip = (struct inet_protocol *) ipprot->next;
657 
658 		/*
659 		 *	Pass it off to everyone who wants it.
660 		 */
661 
662 		/* RFC1122: OK. Passes appropriate ICMP errors to the */
663 		/* appropriate protocol layer (MUST), as per 3.2.2. */
664 
665 		if (protocol == ipprot->protocol && ipprot->err_handler)
666  			ipprot->err_handler(skb, info);
667 
668 		ipprot = nextip;
669   	}
670 out:;
671 }
672 
673 
674 /*
675  *	Handle ICMP_REDIRECT.
676  */
677 
icmp_redirect(struct sk_buff * skb)678 static void icmp_redirect(struct sk_buff *skb)
679 {
680 	struct iphdr *iph;
681 	unsigned long ip;
682 
683 	if (skb->len < sizeof(struct iphdr)) {
684 		ICMP_INC_STATS_BH(IcmpInErrors);
685 		return;
686 	}
687 
688 	/*
689 	 *	Get the copied header of the packet that caused the redirect
690 	 */
691 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
692 		return;
693 
694 	iph = (struct iphdr *) skb->data;
695 	ip = iph->daddr;
696 
697 	switch (skb->h.icmph->code & 7) {
698 		case ICMP_REDIR_NET:
699 		case ICMP_REDIR_NETTOS:
700 			/*
701 			 *	As per RFC recommendations now handle it as
702 			 *	a host redirect.
703 			 */
704 
705 		case ICMP_REDIR_HOST:
706 		case ICMP_REDIR_HOSTTOS:
707 			ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway, iph->saddr, iph->tos, skb->dev);
708 			break;
709 		default:
710 			break;
711   	}
712 }
713 
714 /*
715  *	Handle ICMP_ECHO ("ping") requests.
716  *
717  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo requests.
718  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be included in the reply.
719  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring echo requests, MUST have default=NOT.
720  *	See also WRT handling of options once they are done and working.
721  */
722 
icmp_echo(struct sk_buff * skb)723 static void icmp_echo(struct sk_buff *skb)
724 {
725 	if (!sysctl_icmp_echo_ignore_all) {
726 		struct icmp_bxm icmp_param;
727 
728 		icmp_param.data.icmph=*skb->h.icmph;
729 		icmp_param.data.icmph.type=ICMP_ECHOREPLY;
730 		icmp_param.skb=skb;
731 		icmp_param.offset=0;
732 		icmp_param.data_len=skb->len;
733 		icmp_param.head_len=sizeof(struct icmphdr);
734 		icmp_reply(&icmp_param, skb);
735 	}
736 }
737 
738 /*
739  *	Handle ICMP Timestamp requests.
740  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
741  *		  SHOULD be in the kernel for minimum random latency.
742  *		  MUST be accurate to a few minutes.
743  *		  MUST be updated at least at 15Hz.
744  */
745 
icmp_timestamp(struct sk_buff * skb)746 static void icmp_timestamp(struct sk_buff *skb)
747 {
748 	struct timeval tv;
749 	struct icmp_bxm icmp_param;
750 
751 	/*
752 	 *	Too short.
753 	 */
754 
755 	if (skb->len < 4) {
756 		ICMP_INC_STATS_BH(IcmpInErrors);
757 		return;
758 	}
759 
760 	/*
761 	 *	Fill in the current time as ms since midnight UT:
762 	 */
763 	do_gettimeofday(&tv);
764 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 + tv.tv_usec / 1000);
765 	icmp_param.data.times[2] = icmp_param.data.times[1];
766 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
767 		BUG();
768 	icmp_param.data.icmph=*skb->h.icmph;
769 	icmp_param.data.icmph.type=ICMP_TIMESTAMPREPLY;
770 	icmp_param.data.icmph.code=0;
771 	icmp_param.skb=skb;
772 	icmp_param.offset=0;
773 	icmp_param.data_len=0;
774 	icmp_param.head_len=sizeof(struct icmphdr)+12;
775 	icmp_reply(&icmp_param, skb);
776 }
777 
778 
779 /*
780  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
781  *
782  * RFC1122 (3.2.2.9).  A host MUST only send replies to
783  * ADDRESS_MASK requests if it's been configured as an address mask
784  * agent.  Receiving a request doesn't constitute implicit permission to
785  * act as one. Of course, implementing this correctly requires (SHOULD)
786  * a way to turn the functionality on and off.  Another one for sysctl(),
787  * I guess. -- MS
788  *
789  * RFC1812 (4.3.3.9).	A router MUST implement it.
790  *			A router SHOULD have switch turning it on/off.
791  *		      	This switch MUST be ON by default.
792  *
793  * Gratuitous replies, zero-source replies are not implemented,
794  * that complies with RFC. DO NOT implement them!!! All the idea
795  * of broadcast addrmask replies as specified in RFC950 is broken.
796  * The problem is that it is not uncommon to have several prefixes
797  * on one physical interface. Moreover, addrmask agent can even be
798  * not aware of existing another prefixes.
799  * If source is zero, addrmask agent cannot choose correct prefix.
800  * Gratuitous mask announcements suffer from the same problem.
801  * RFC1812 explains it, but still allows to use ADDRMASK,
802  * that is pretty silly. --ANK
803  *
804  * All these rules are so bizarre, that I removed kernel addrmask
805  * support at all. It is wrong, it is obsolete, nobody uses it in
806  * any case. --ANK
807  *
808  * Furthermore you can do it with a usermode address agent program
809  * anyway...
810  */
811 
icmp_address(struct sk_buff * skb)812 static void icmp_address(struct sk_buff *skb)
813 {
814 #if 0
815 	if (net_ratelimit())
816 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
817 #endif
818 }
819 
820 /*
821  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
822  *			loudly if an inconsistency is found.
823  */
824 
icmp_address_reply(struct sk_buff * skb)825 static void icmp_address_reply(struct sk_buff *skb)
826 {
827 	struct rtable *rt = (struct rtable*)skb->dst;
828 	struct net_device *dev = skb->dev;
829 	struct in_device *in_dev;
830 	struct in_ifaddr *ifa;
831 	u32 mask;
832 
833 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
834 		return;
835 
836 	in_dev = in_dev_get(dev);
837 	if (!in_dev)
838 		return;
839 	read_lock(&in_dev->lock);
840 	if (in_dev->ifa_list &&
841 	    IN_DEV_LOG_MARTIANS(in_dev) &&
842 	    IN_DEV_FORWARD(in_dev)) {
843 		if (skb_copy_bits(skb, 0, &mask, 4))
844 			BUG();
845 		for (ifa=in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
846 			if (mask == ifa->ifa_mask && inet_ifa_match(rt->rt_src, ifa))
847 				break;
848 		}
849 		if (!ifa && net_ratelimit()) {
850 			printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from %s/%u.%u.%u.%u\n",
851 			       NIPQUAD(mask), dev->name, NIPQUAD(rt->rt_src));
852 		}
853 	}
854 	read_unlock(&in_dev->lock);
855 	in_dev_put(in_dev);
856 }
857 
icmp_discard(struct sk_buff * skb)858 static void icmp_discard(struct sk_buff *skb)
859 {
860 }
861 
862 /*
863  *	Deal with incoming ICMP packets.
864  */
865 
icmp_rcv(struct sk_buff * skb)866 int icmp_rcv(struct sk_buff *skb)
867 {
868 	struct icmphdr *icmph;
869 	struct rtable *rt = (struct rtable*)skb->dst;
870 
871 	ICMP_INC_STATS_BH(IcmpInMsgs);
872 
873 	switch (skb->ip_summed) {
874 	case CHECKSUM_HW:
875 		if ((u16)csum_fold(skb->csum) == 0)
876 			break;
877 		NETDEBUG(if (net_ratelimit()) printk(KERN_DEBUG "icmp v4 hw csum failure\n"));
878 	case CHECKSUM_NONE:
879 		if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0)))
880 			goto error;
881 	default:;
882 	}
883 
884 	if (!pskb_pull(skb, sizeof(struct icmphdr)))
885 		goto error;
886 
887 	icmph = skb->h.icmph;
888 
889 	/*
890 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
891 	 *
892 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently discarded.
893 	 */
894 	if (icmph->type > NR_ICMP_TYPES)
895 		goto error;
896 
897 
898 	/*
899 	 *	Parse the ICMP message
900 	 */
901 
902  	if (rt->rt_flags&(RTCF_BROADCAST|RTCF_MULTICAST)) {
903 		/*
904 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
905 		 *	  silently ignored (we let user decide with a sysctl).
906 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
907 		 *	  discarded if to broadcast/multicast.
908 		 */
909 		if (icmph->type == ICMP_ECHO &&
910 		    sysctl_icmp_echo_ignore_broadcasts) {
911 			goto error;
912 		}
913 		if (icmph->type != ICMP_ECHO &&
914 		    icmph->type != ICMP_TIMESTAMP &&
915 		    icmph->type != ICMP_ADDRESS &&
916 		    icmph->type != ICMP_ADDRESSREPLY) {
917 			goto error;
918   		}
919 	}
920 
921 	icmp_pointers[icmph->type].input[smp_processor_id()*2*sizeof(struct icmp_mib)/sizeof(unsigned long)]++;
922 	(icmp_pointers[icmph->type].handler)(skb);
923 
924 drop:
925 	kfree_skb(skb);
926 	return 0;
927 error:
928 	ICMP_INC_STATS_BH(IcmpInErrors);
929 	goto drop;
930 }
931 
932 /*
933  *	This table is the definition of how we handle ICMP.
934  */
935 
936 static struct icmp_control icmp_pointers[NR_ICMP_TYPES+1] = {
937 /* ECHO REPLY (0) */
938  { &icmp_statistics[0].IcmpOutEchoReps, &icmp_statistics[0].IcmpInEchoReps, icmp_discard, 0 },
939  { &icmp_statistics[0].dummy, &icmp_statistics[0].IcmpInErrors, icmp_discard, 1 },
940  { &icmp_statistics[0].dummy, &icmp_statistics[0].IcmpInErrors, icmp_discard, 1 },
941 /* DEST UNREACH (3) */
942  { &icmp_statistics[0].IcmpOutDestUnreachs, &icmp_statistics[0].IcmpInDestUnreachs, icmp_unreach, 1 },
943 /* SOURCE QUENCH (4) */
944  { &icmp_statistics[0].IcmpOutSrcQuenchs, &icmp_statistics[0].IcmpInSrcQuenchs, icmp_unreach, 1 },
945 /* REDIRECT (5) */
946  { &icmp_statistics[0].IcmpOutRedirects, &icmp_statistics[0].IcmpInRedirects, icmp_redirect, 1 },
947  { &icmp_statistics[0].dummy, &icmp_statistics[0].IcmpInErrors, icmp_discard, 1 },
948  { &icmp_statistics[0].dummy, &icmp_statistics[0].IcmpInErrors, icmp_discard, 1 },
949 /* ECHO (8) */
950  { &icmp_statistics[0].IcmpOutEchos, &icmp_statistics[0].IcmpInEchos, icmp_echo, 0 },
951  { &icmp_statistics[0].dummy, &icmp_statistics[0].IcmpInErrors, icmp_discard, 1 },
952  { &icmp_statistics[0].dummy, &icmp_statistics[0].IcmpInErrors, icmp_discard, 1 },
953 /* TIME EXCEEDED (11) */
954  { &icmp_statistics[0].IcmpOutTimeExcds, &icmp_statistics[0].IcmpInTimeExcds, icmp_unreach, 1 },
955 /* PARAMETER PROBLEM (12) */
956  { &icmp_statistics[0].IcmpOutParmProbs, &icmp_statistics[0].IcmpInParmProbs, icmp_unreach, 1 },
957 /* TIMESTAMP (13) */
958  { &icmp_statistics[0].IcmpOutTimestamps, &icmp_statistics[0].IcmpInTimestamps, icmp_timestamp, 0  },
959 /* TIMESTAMP REPLY (14) */
960  { &icmp_statistics[0].IcmpOutTimestampReps, &icmp_statistics[0].IcmpInTimestampReps, icmp_discard, 0 },
961 /* INFO (15) */
962  { &icmp_statistics[0].dummy, &icmp_statistics[0].dummy, icmp_discard, 0 },
963 /* INFO REPLY (16) */
964  { &icmp_statistics[0].dummy, &icmp_statistics[0].dummy, icmp_discard, 0 },
965 /* ADDR MASK (17) */
966  { &icmp_statistics[0].IcmpOutAddrMasks, &icmp_statistics[0].IcmpInAddrMasks, icmp_address, 0  },
967 /* ADDR MASK REPLY (18) */
968  { &icmp_statistics[0].IcmpOutAddrMaskReps, &icmp_statistics[0].IcmpInAddrMaskReps, icmp_address_reply, 0 }
969 };
970 
icmp_init(struct net_proto_family * ops)971 void __init icmp_init(struct net_proto_family *ops)
972 {
973 	int err, i;
974 
975 	for (i = 0; i < NR_CPUS; i++) {
976 		__icmp_inode[i].i_mode = S_IFSOCK;
977 		__icmp_inode[i].i_sock = 1;
978 		__icmp_inode[i].i_uid = 0;
979 		__icmp_inode[i].i_gid = 0;
980 		init_waitqueue_head(&__icmp_inode[i].i_wait);
981 		init_waitqueue_head(&__icmp_inode[i].u.socket_i.wait);
982 
983 		icmp_socket_cpu(i)->inode = &__icmp_inode[i];
984 		icmp_socket_cpu(i)->state = SS_UNCONNECTED;
985 		icmp_socket_cpu(i)->type = SOCK_RAW;
986 
987 		if ((err=ops->create(icmp_socket_cpu(i), IPPROTO_ICMP)) < 0)
988 			panic("Failed to create the ICMP control socket.\n");
989 
990 		icmp_socket_cpu(i)->sk->allocation=GFP_ATOMIC;
991 
992 		/* Enough space for 2 64K ICMP packets, including
993 		 * sk_buff struct overhead.
994 		 */
995 		icmp_socket_cpu(i)->sk->sndbuf =
996 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
997 
998 		icmp_socket_cpu(i)->sk->protinfo.af_inet.ttl = MAXTTL;
999 		icmp_socket_cpu(i)->sk->protinfo.af_inet.pmtudisc = IP_PMTUDISC_DONT;
1000 
1001 		/* Unhash it so that IP input processing does not even
1002 		 * see it, we do not wish this socket to see incoming
1003 		 * packets.
1004 		 */
1005 		icmp_socket_cpu(i)->sk->prot->unhash(icmp_socket_cpu(i)->sk);
1006 	}
1007 }
1008