xref: /openbmc/linux/net/ipv4/icmp.c (revision 9d749629)
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 #include <net/ip_fib.h>
99 
100 /*
101  *	Build xmit assembly blocks
102  */
103 
104 struct icmp_bxm {
105 	struct sk_buff *skb;
106 	int offset;
107 	int data_len;
108 
109 	struct {
110 		struct icmphdr icmph;
111 		__be32	       times[3];
112 	} data;
113 	int head_len;
114 	struct ip_options_data replyopts;
115 };
116 
117 /* An array of errno for error messages from dest unreach. */
118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
119 
120 const struct icmp_err icmp_err_convert[] = {
121 	{
122 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
123 		.fatal = 0,
124 	},
125 	{
126 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
127 		.fatal = 0,
128 	},
129 	{
130 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
131 		.fatal = 1,
132 	},
133 	{
134 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
135 		.fatal = 1,
136 	},
137 	{
138 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
139 		.fatal = 0,
140 	},
141 	{
142 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
143 		.fatal = 0,
144 	},
145 	{
146 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
147 		.fatal = 1,
148 	},
149 	{
150 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
163 		.fatal = 1,
164 	},
165 	{
166 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
167 		.fatal = 0,
168 	},
169 	{
170 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
171 		.fatal = 0,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
175 		.fatal = 1,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
179 		.fatal = 1,
180 	},
181 	{
182 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
183 		.fatal = 1,
184 	},
185 };
186 EXPORT_SYMBOL(icmp_err_convert);
187 
188 /*
189  *	ICMP control array. This specifies what to do with each ICMP.
190  */
191 
192 struct icmp_control {
193 	void (*handler)(struct sk_buff *skb);
194 	short   error;		/* This ICMP is classed as an error message */
195 };
196 
197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
198 
199 /*
200  *	The ICMP socket(s). This is the most convenient way to flow control
201  *	our ICMP output as well as maintain a clean interface throughout
202  *	all layers. All Socketless IP sends will soon be gone.
203  *
204  *	On SMP we have one ICMP socket per-cpu.
205  */
206 static struct sock *icmp_sk(struct net *net)
207 {
208 	return net->ipv4.icmp_sk[smp_processor_id()];
209 }
210 
211 static inline struct sock *icmp_xmit_lock(struct net *net)
212 {
213 	struct sock *sk;
214 
215 	local_bh_disable();
216 
217 	sk = icmp_sk(net);
218 
219 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
220 		/* This can happen if the output path signals a
221 		 * dst_link_failure() for an outgoing ICMP packet.
222 		 */
223 		local_bh_enable();
224 		return NULL;
225 	}
226 	return sk;
227 }
228 
229 static inline void icmp_xmit_unlock(struct sock *sk)
230 {
231 	spin_unlock_bh(&sk->sk_lock.slock);
232 }
233 
234 /*
235  *	Send an ICMP frame.
236  */
237 
238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 				      struct flowi4 *fl4, int type, int code)
240 {
241 	struct dst_entry *dst = &rt->dst;
242 	bool rc = true;
243 
244 	if (type > NR_ICMP_TYPES)
245 		goto out;
246 
247 	/* Don't limit PMTU discovery. */
248 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
249 		goto out;
250 
251 	/* No rate limit on loopback */
252 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
253 		goto out;
254 
255 	/* Limit if icmp type is enabled in ratemask. */
256 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 		struct inet_peer *peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1);
258 		rc = inet_peer_xrlim_allow(peer,
259 					   net->ipv4.sysctl_icmp_ratelimit);
260 		if (peer)
261 			inet_putpeer(peer);
262 	}
263 out:
264 	return rc;
265 }
266 
267 /*
268  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
269  */
270 void icmp_out_count(struct net *net, unsigned char type)
271 {
272 	ICMPMSGOUT_INC_STATS(net, type);
273 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
274 }
275 
276 /*
277  *	Checksum each fragment, and on the first include the headers and final
278  *	checksum.
279  */
280 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
281 			  struct sk_buff *skb)
282 {
283 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
284 	__wsum csum;
285 
286 	csum = skb_copy_and_csum_bits(icmp_param->skb,
287 				      icmp_param->offset + offset,
288 				      to, len, 0);
289 
290 	skb->csum = csum_block_add(skb->csum, csum, odd);
291 	if (icmp_pointers[icmp_param->data.icmph.type].error)
292 		nf_ct_attach(skb, icmp_param->skb);
293 	return 0;
294 }
295 
296 static void icmp_push_reply(struct icmp_bxm *icmp_param,
297 			    struct flowi4 *fl4,
298 			    struct ipcm_cookie *ipc, struct rtable **rt)
299 {
300 	struct sock *sk;
301 	struct sk_buff *skb;
302 
303 	sk = icmp_sk(dev_net((*rt)->dst.dev));
304 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
305 			   icmp_param->data_len+icmp_param->head_len,
306 			   icmp_param->head_len,
307 			   ipc, rt, MSG_DONTWAIT) < 0) {
308 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
309 		ip_flush_pending_frames(sk);
310 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
311 		struct icmphdr *icmph = icmp_hdr(skb);
312 		__wsum csum = 0;
313 		struct sk_buff *skb1;
314 
315 		skb_queue_walk(&sk->sk_write_queue, skb1) {
316 			csum = csum_add(csum, skb1->csum);
317 		}
318 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
319 						 (char *)icmph,
320 						 icmp_param->head_len, csum);
321 		icmph->checksum = csum_fold(csum);
322 		skb->ip_summed = CHECKSUM_NONE;
323 		ip_push_pending_frames(sk, fl4);
324 	}
325 }
326 
327 /*
328  *	Driving logic for building and sending ICMP messages.
329  */
330 
331 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
332 {
333 	struct ipcm_cookie ipc;
334 	struct rtable *rt = skb_rtable(skb);
335 	struct net *net = dev_net(rt->dst.dev);
336 	struct flowi4 fl4;
337 	struct sock *sk;
338 	struct inet_sock *inet;
339 	__be32 daddr, saddr;
340 
341 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
342 		return;
343 
344 	sk = icmp_xmit_lock(net);
345 	if (sk == NULL)
346 		return;
347 	inet = inet_sk(sk);
348 
349 	icmp_param->data.icmph.checksum = 0;
350 
351 	inet->tos = ip_hdr(skb)->tos;
352 	daddr = ipc.addr = ip_hdr(skb)->saddr;
353 	saddr = fib_compute_spec_dst(skb);
354 	ipc.opt = NULL;
355 	ipc.tx_flags = 0;
356 	if (icmp_param->replyopts.opt.opt.optlen) {
357 		ipc.opt = &icmp_param->replyopts.opt;
358 		if (ipc.opt->opt.srr)
359 			daddr = icmp_param->replyopts.opt.opt.faddr;
360 	}
361 	memset(&fl4, 0, sizeof(fl4));
362 	fl4.daddr = daddr;
363 	fl4.saddr = saddr;
364 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
365 	fl4.flowi4_proto = IPPROTO_ICMP;
366 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
367 	rt = ip_route_output_key(net, &fl4);
368 	if (IS_ERR(rt))
369 		goto out_unlock;
370 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
371 			       icmp_param->data.icmph.code))
372 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
373 	ip_rt_put(rt);
374 out_unlock:
375 	icmp_xmit_unlock(sk);
376 }
377 
378 static struct rtable *icmp_route_lookup(struct net *net,
379 					struct flowi4 *fl4,
380 					struct sk_buff *skb_in,
381 					const struct iphdr *iph,
382 					__be32 saddr, u8 tos,
383 					int type, int code,
384 					struct icmp_bxm *param)
385 {
386 	struct rtable *rt, *rt2;
387 	struct flowi4 fl4_dec;
388 	int err;
389 
390 	memset(fl4, 0, sizeof(*fl4));
391 	fl4->daddr = (param->replyopts.opt.opt.srr ?
392 		      param->replyopts.opt.opt.faddr : iph->saddr);
393 	fl4->saddr = saddr;
394 	fl4->flowi4_tos = RT_TOS(tos);
395 	fl4->flowi4_proto = IPPROTO_ICMP;
396 	fl4->fl4_icmp_type = type;
397 	fl4->fl4_icmp_code = code;
398 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
399 	rt = __ip_route_output_key(net, fl4);
400 	if (IS_ERR(rt))
401 		return rt;
402 
403 	/* No need to clone since we're just using its address. */
404 	rt2 = rt;
405 
406 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
407 					   flowi4_to_flowi(fl4), NULL, 0);
408 	if (!IS_ERR(rt)) {
409 		if (rt != rt2)
410 			return rt;
411 	} else if (PTR_ERR(rt) == -EPERM) {
412 		rt = NULL;
413 	} else
414 		return rt;
415 
416 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
417 	if (err)
418 		goto relookup_failed;
419 
420 	if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
421 		rt2 = __ip_route_output_key(net, &fl4_dec);
422 		if (IS_ERR(rt2))
423 			err = PTR_ERR(rt2);
424 	} else {
425 		struct flowi4 fl4_2 = {};
426 		unsigned long orefdst;
427 
428 		fl4_2.daddr = fl4_dec.saddr;
429 		rt2 = ip_route_output_key(net, &fl4_2);
430 		if (IS_ERR(rt2)) {
431 			err = PTR_ERR(rt2);
432 			goto relookup_failed;
433 		}
434 		/* Ugh! */
435 		orefdst = skb_in->_skb_refdst; /* save old refdst */
436 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
437 				     RT_TOS(tos), rt2->dst.dev);
438 
439 		dst_release(&rt2->dst);
440 		rt2 = skb_rtable(skb_in);
441 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
442 	}
443 
444 	if (err)
445 		goto relookup_failed;
446 
447 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
448 					    flowi4_to_flowi(&fl4_dec), NULL,
449 					    XFRM_LOOKUP_ICMP);
450 	if (!IS_ERR(rt2)) {
451 		dst_release(&rt->dst);
452 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
453 		rt = rt2;
454 	} else if (PTR_ERR(rt2) == -EPERM) {
455 		if (rt)
456 			dst_release(&rt->dst);
457 		return rt2;
458 	} else {
459 		err = PTR_ERR(rt2);
460 		goto relookup_failed;
461 	}
462 	return rt;
463 
464 relookup_failed:
465 	if (rt)
466 		return rt;
467 	return ERR_PTR(err);
468 }
469 
470 /*
471  *	Send an ICMP message in response to a situation
472  *
473  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
474  *		  MAY send more (we do).
475  *			MUST NOT change this header information.
476  *			MUST NOT reply to a multicast/broadcast IP address.
477  *			MUST NOT reply to a multicast/broadcast MAC address.
478  *			MUST reply to only the first fragment.
479  */
480 
481 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
482 {
483 	struct iphdr *iph;
484 	int room;
485 	struct icmp_bxm icmp_param;
486 	struct rtable *rt = skb_rtable(skb_in);
487 	struct ipcm_cookie ipc;
488 	struct flowi4 fl4;
489 	__be32 saddr;
490 	u8  tos;
491 	struct net *net;
492 	struct sock *sk;
493 
494 	if (!rt)
495 		goto out;
496 	net = dev_net(rt->dst.dev);
497 
498 	/*
499 	 *	Find the original header. It is expected to be valid, of course.
500 	 *	Check this, icmp_send is called from the most obscure devices
501 	 *	sometimes.
502 	 */
503 	iph = ip_hdr(skb_in);
504 
505 	if ((u8 *)iph < skb_in->head ||
506 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
507 		goto out;
508 
509 	/*
510 	 *	No replies to physical multicast/broadcast
511 	 */
512 	if (skb_in->pkt_type != PACKET_HOST)
513 		goto out;
514 
515 	/*
516 	 *	Now check at the protocol level
517 	 */
518 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
519 		goto out;
520 
521 	/*
522 	 *	Only reply to fragment 0. We byte re-order the constant
523 	 *	mask for efficiency.
524 	 */
525 	if (iph->frag_off & htons(IP_OFFSET))
526 		goto out;
527 
528 	/*
529 	 *	If we send an ICMP error to an ICMP error a mess would result..
530 	 */
531 	if (icmp_pointers[type].error) {
532 		/*
533 		 *	We are an error, check if we are replying to an
534 		 *	ICMP error
535 		 */
536 		if (iph->protocol == IPPROTO_ICMP) {
537 			u8 _inner_type, *itp;
538 
539 			itp = skb_header_pointer(skb_in,
540 						 skb_network_header(skb_in) +
541 						 (iph->ihl << 2) +
542 						 offsetof(struct icmphdr,
543 							  type) -
544 						 skb_in->data,
545 						 sizeof(_inner_type),
546 						 &_inner_type);
547 			if (itp == NULL)
548 				goto out;
549 
550 			/*
551 			 *	Assume any unknown ICMP type is an error. This
552 			 *	isn't specified by the RFC, but think about it..
553 			 */
554 			if (*itp > NR_ICMP_TYPES ||
555 			    icmp_pointers[*itp].error)
556 				goto out;
557 		}
558 	}
559 
560 	sk = icmp_xmit_lock(net);
561 	if (sk == NULL)
562 		return;
563 
564 	/*
565 	 *	Construct source address and options.
566 	 */
567 
568 	saddr = iph->daddr;
569 	if (!(rt->rt_flags & RTCF_LOCAL)) {
570 		struct net_device *dev = NULL;
571 
572 		rcu_read_lock();
573 		if (rt_is_input_route(rt) &&
574 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
575 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
576 
577 		if (dev)
578 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
579 		else
580 			saddr = 0;
581 		rcu_read_unlock();
582 	}
583 
584 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
585 					   IPTOS_PREC_INTERNETCONTROL) :
586 					  iph->tos;
587 
588 	if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
589 		goto out_unlock;
590 
591 
592 	/*
593 	 *	Prepare data for ICMP header.
594 	 */
595 
596 	icmp_param.data.icmph.type	 = type;
597 	icmp_param.data.icmph.code	 = code;
598 	icmp_param.data.icmph.un.gateway = info;
599 	icmp_param.data.icmph.checksum	 = 0;
600 	icmp_param.skb	  = skb_in;
601 	icmp_param.offset = skb_network_offset(skb_in);
602 	inet_sk(sk)->tos = tos;
603 	ipc.addr = iph->saddr;
604 	ipc.opt = &icmp_param.replyopts.opt;
605 	ipc.tx_flags = 0;
606 
607 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
608 			       type, code, &icmp_param);
609 	if (IS_ERR(rt))
610 		goto out_unlock;
611 
612 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
613 		goto ende;
614 
615 	/* RFC says return as much as we can without exceeding 576 bytes. */
616 
617 	room = dst_mtu(&rt->dst);
618 	if (room > 576)
619 		room = 576;
620 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
621 	room -= sizeof(struct icmphdr);
622 
623 	icmp_param.data_len = skb_in->len - icmp_param.offset;
624 	if (icmp_param.data_len > room)
625 		icmp_param.data_len = room;
626 	icmp_param.head_len = sizeof(struct icmphdr);
627 
628 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
629 ende:
630 	ip_rt_put(rt);
631 out_unlock:
632 	icmp_xmit_unlock(sk);
633 out:;
634 }
635 EXPORT_SYMBOL(icmp_send);
636 
637 
638 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
639 {
640 	const struct iphdr *iph = (const struct iphdr *) skb->data;
641 	const struct net_protocol *ipprot;
642 	int protocol = iph->protocol;
643 
644 	/* Checkin full IP header plus 8 bytes of protocol to
645 	 * avoid additional coding at protocol handlers.
646 	 */
647 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
648 		return;
649 
650 	raw_icmp_error(skb, protocol, info);
651 
652 	rcu_read_lock();
653 	ipprot = rcu_dereference(inet_protos[protocol]);
654 	if (ipprot && ipprot->err_handler)
655 		ipprot->err_handler(skb, info);
656 	rcu_read_unlock();
657 }
658 
659 /*
660  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
661  */
662 
663 static void icmp_unreach(struct sk_buff *skb)
664 {
665 	const struct iphdr *iph;
666 	struct icmphdr *icmph;
667 	struct net *net;
668 	u32 info = 0;
669 
670 	net = dev_net(skb_dst(skb)->dev);
671 
672 	/*
673 	 *	Incomplete header ?
674 	 * 	Only checks for the IP header, there should be an
675 	 *	additional check for longer headers in upper levels.
676 	 */
677 
678 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
679 		goto out_err;
680 
681 	icmph = icmp_hdr(skb);
682 	iph   = (const struct iphdr *)skb->data;
683 
684 	if (iph->ihl < 5) /* Mangled header, drop. */
685 		goto out_err;
686 
687 	if (icmph->type == ICMP_DEST_UNREACH) {
688 		switch (icmph->code & 15) {
689 		case ICMP_NET_UNREACH:
690 		case ICMP_HOST_UNREACH:
691 		case ICMP_PROT_UNREACH:
692 		case ICMP_PORT_UNREACH:
693 			break;
694 		case ICMP_FRAG_NEEDED:
695 			if (ipv4_config.no_pmtu_disc) {
696 				LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
697 					       &iph->daddr);
698 			} else {
699 				info = ntohs(icmph->un.frag.mtu);
700 				if (!info)
701 					goto out;
702 			}
703 			break;
704 		case ICMP_SR_FAILED:
705 			LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
706 				       &iph->daddr);
707 			break;
708 		default:
709 			break;
710 		}
711 		if (icmph->code > NR_ICMP_UNREACH)
712 			goto out;
713 	} else if (icmph->type == ICMP_PARAMETERPROB)
714 		info = ntohl(icmph->un.gateway) >> 24;
715 
716 	/*
717 	 *	Throw it at our lower layers
718 	 *
719 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
720 	 *		  header.
721 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
722 	 *		  transport layer.
723 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
724 	 *		  transport layer.
725 	 */
726 
727 	/*
728 	 *	Check the other end isn't violating RFC 1122. Some routers send
729 	 *	bogus responses to broadcast frames. If you see this message
730 	 *	first check your netmask matches at both ends, if it does then
731 	 *	get the other vendor to fix their kit.
732 	 */
733 
734 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
735 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
736 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
737 				     &ip_hdr(skb)->saddr,
738 				     icmph->type, icmph->code,
739 				     &iph->daddr, skb->dev->name);
740 		goto out;
741 	}
742 
743 	icmp_socket_deliver(skb, info);
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 	if (skb->len < sizeof(struct iphdr)) {
760 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
761 		return;
762 	}
763 
764 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
765 		return;
766 
767 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
768 }
769 
770 /*
771  *	Handle ICMP_ECHO ("ping") requests.
772  *
773  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
774  *		  requests.
775  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
776  *		  included in the reply.
777  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
778  *		  echo requests, MUST have default=NOT.
779  *	See also WRT handling of options once they are done and working.
780  */
781 
782 static void icmp_echo(struct sk_buff *skb)
783 {
784 	struct net *net;
785 
786 	net = dev_net(skb_dst(skb)->dev);
787 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
788 		struct icmp_bxm icmp_param;
789 
790 		icmp_param.data.icmph	   = *icmp_hdr(skb);
791 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
792 		icmp_param.skb		   = skb;
793 		icmp_param.offset	   = 0;
794 		icmp_param.data_len	   = skb->len;
795 		icmp_param.head_len	   = sizeof(struct icmphdr);
796 		icmp_reply(&icmp_param, skb);
797 	}
798 }
799 
800 /*
801  *	Handle ICMP Timestamp requests.
802  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
803  *		  SHOULD be in the kernel for minimum random latency.
804  *		  MUST be accurate to a few minutes.
805  *		  MUST be updated at least at 15Hz.
806  */
807 static void icmp_timestamp(struct sk_buff *skb)
808 {
809 	struct timespec tv;
810 	struct icmp_bxm icmp_param;
811 	/*
812 	 *	Too short.
813 	 */
814 	if (skb->len < 4)
815 		goto out_err;
816 
817 	/*
818 	 *	Fill in the current time as ms since midnight UT:
819 	 */
820 	getnstimeofday(&tv);
821 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
822 					 tv.tv_nsec / NSEC_PER_MSEC);
823 	icmp_param.data.times[2] = icmp_param.data.times[1];
824 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
825 		BUG();
826 	icmp_param.data.icmph	   = *icmp_hdr(skb);
827 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
828 	icmp_param.data.icmph.code = 0;
829 	icmp_param.skb		   = skb;
830 	icmp_param.offset	   = 0;
831 	icmp_param.data_len	   = 0;
832 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
833 	icmp_reply(&icmp_param, skb);
834 out:
835 	return;
836 out_err:
837 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
838 	goto out;
839 }
840 
841 static void icmp_discard(struct sk_buff *skb)
842 {
843 }
844 
845 /*
846  *	Deal with incoming ICMP packets.
847  */
848 int icmp_rcv(struct sk_buff *skb)
849 {
850 	struct icmphdr *icmph;
851 	struct rtable *rt = skb_rtable(skb);
852 	struct net *net = dev_net(rt->dst.dev);
853 
854 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
855 		struct sec_path *sp = skb_sec_path(skb);
856 		int nh;
857 
858 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
859 				 XFRM_STATE_ICMP))
860 			goto drop;
861 
862 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
863 			goto drop;
864 
865 		nh = skb_network_offset(skb);
866 		skb_set_network_header(skb, sizeof(*icmph));
867 
868 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
869 			goto drop;
870 
871 		skb_set_network_header(skb, nh);
872 	}
873 
874 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
875 
876 	switch (skb->ip_summed) {
877 	case CHECKSUM_COMPLETE:
878 		if (!csum_fold(skb->csum))
879 			break;
880 		/* fall through */
881 	case CHECKSUM_NONE:
882 		skb->csum = 0;
883 		if (__skb_checksum_complete(skb))
884 			goto error;
885 	}
886 
887 	if (!pskb_pull(skb, sizeof(*icmph)))
888 		goto error;
889 
890 	icmph = icmp_hdr(skb);
891 
892 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
893 	/*
894 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
895 	 *
896 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
897 	 *		  discarded.
898 	 */
899 	if (icmph->type > NR_ICMP_TYPES)
900 		goto error;
901 
902 
903 	/*
904 	 *	Parse the ICMP message
905 	 */
906 
907 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
908 		/*
909 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
910 		 *	  silently ignored (we let user decide with a sysctl).
911 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
912 		 *	  discarded if to broadcast/multicast.
913 		 */
914 		if ((icmph->type == ICMP_ECHO ||
915 		     icmph->type == ICMP_TIMESTAMP) &&
916 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
917 			goto error;
918 		}
919 		if (icmph->type != ICMP_ECHO &&
920 		    icmph->type != ICMP_TIMESTAMP &&
921 		    icmph->type != ICMP_ADDRESS &&
922 		    icmph->type != ICMP_ADDRESSREPLY) {
923 			goto error;
924 		}
925 	}
926 
927 	icmp_pointers[icmph->type].handler(skb);
928 
929 drop:
930 	kfree_skb(skb);
931 	return 0;
932 error:
933 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
934 	goto drop;
935 }
936 
937 /*
938  *	This table is the definition of how we handle ICMP.
939  */
940 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
941 	[ICMP_ECHOREPLY] = {
942 		.handler = ping_rcv,
943 	},
944 	[1] = {
945 		.handler = icmp_discard,
946 		.error = 1,
947 	},
948 	[2] = {
949 		.handler = icmp_discard,
950 		.error = 1,
951 	},
952 	[ICMP_DEST_UNREACH] = {
953 		.handler = icmp_unreach,
954 		.error = 1,
955 	},
956 	[ICMP_SOURCE_QUENCH] = {
957 		.handler = icmp_unreach,
958 		.error = 1,
959 	},
960 	[ICMP_REDIRECT] = {
961 		.handler = icmp_redirect,
962 		.error = 1,
963 	},
964 	[6] = {
965 		.handler = icmp_discard,
966 		.error = 1,
967 	},
968 	[7] = {
969 		.handler = icmp_discard,
970 		.error = 1,
971 	},
972 	[ICMP_ECHO] = {
973 		.handler = icmp_echo,
974 	},
975 	[9] = {
976 		.handler = icmp_discard,
977 		.error = 1,
978 	},
979 	[10] = {
980 		.handler = icmp_discard,
981 		.error = 1,
982 	},
983 	[ICMP_TIME_EXCEEDED] = {
984 		.handler = icmp_unreach,
985 		.error = 1,
986 	},
987 	[ICMP_PARAMETERPROB] = {
988 		.handler = icmp_unreach,
989 		.error = 1,
990 	},
991 	[ICMP_TIMESTAMP] = {
992 		.handler = icmp_timestamp,
993 	},
994 	[ICMP_TIMESTAMPREPLY] = {
995 		.handler = icmp_discard,
996 	},
997 	[ICMP_INFO_REQUEST] = {
998 		.handler = icmp_discard,
999 	},
1000 	[ICMP_INFO_REPLY] = {
1001 		.handler = icmp_discard,
1002 	},
1003 	[ICMP_ADDRESS] = {
1004 		.handler = icmp_discard,
1005 	},
1006 	[ICMP_ADDRESSREPLY] = {
1007 		.handler = icmp_discard,
1008 	},
1009 };
1010 
1011 static void __net_exit icmp_sk_exit(struct net *net)
1012 {
1013 	int i;
1014 
1015 	for_each_possible_cpu(i)
1016 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1017 	kfree(net->ipv4.icmp_sk);
1018 	net->ipv4.icmp_sk = NULL;
1019 }
1020 
1021 static int __net_init icmp_sk_init(struct net *net)
1022 {
1023 	int i, err;
1024 
1025 	net->ipv4.icmp_sk =
1026 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1027 	if (net->ipv4.icmp_sk == NULL)
1028 		return -ENOMEM;
1029 
1030 	for_each_possible_cpu(i) {
1031 		struct sock *sk;
1032 
1033 		err = inet_ctl_sock_create(&sk, PF_INET,
1034 					   SOCK_RAW, IPPROTO_ICMP, net);
1035 		if (err < 0)
1036 			goto fail;
1037 
1038 		net->ipv4.icmp_sk[i] = sk;
1039 
1040 		/* Enough space for 2 64K ICMP packets, including
1041 		 * sk_buff/skb_shared_info struct overhead.
1042 		 */
1043 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1044 
1045 		/*
1046 		 * Speedup sock_wfree()
1047 		 */
1048 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1049 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1050 	}
1051 
1052 	/* Control parameters for ECHO replies. */
1053 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1054 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1055 
1056 	/* Control parameter - ignore bogus broadcast responses? */
1057 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1058 
1059 	/*
1060 	 * 	Configurable global rate limit.
1061 	 *
1062 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1063 	 *	bucket ratemask defines which icmp types are ratelimited by
1064 	 *	setting	it's bit position.
1065 	 *
1066 	 *	default:
1067 	 *	dest unreachable (3), source quench (4),
1068 	 *	time exceeded (11), parameter problem (12)
1069 	 */
1070 
1071 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1072 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1073 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1074 
1075 	return 0;
1076 
1077 fail:
1078 	for_each_possible_cpu(i)
1079 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1080 	kfree(net->ipv4.icmp_sk);
1081 	return err;
1082 }
1083 
1084 static struct pernet_operations __net_initdata icmp_sk_ops = {
1085        .init = icmp_sk_init,
1086        .exit = icmp_sk_exit,
1087 };
1088 
1089 int __init icmp_init(void)
1090 {
1091 	return register_pernet_subsys(&icmp_sk_ops);
1092 }
1093