xref: /openbmc/linux/net/ipv4/icmp.c (revision 930beb5a)
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 	ipc.ttl = 0;
357 	ipc.tos = -1;
358 
359 	if (icmp_param->replyopts.opt.opt.optlen) {
360 		ipc.opt = &icmp_param->replyopts.opt;
361 		if (ipc.opt->opt.srr)
362 			daddr = icmp_param->replyopts.opt.opt.faddr;
363 	}
364 	memset(&fl4, 0, sizeof(fl4));
365 	fl4.daddr = daddr;
366 	fl4.saddr = saddr;
367 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
368 	fl4.flowi4_proto = IPPROTO_ICMP;
369 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
370 	rt = ip_route_output_key(net, &fl4);
371 	if (IS_ERR(rt))
372 		goto out_unlock;
373 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
374 			       icmp_param->data.icmph.code))
375 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
376 	ip_rt_put(rt);
377 out_unlock:
378 	icmp_xmit_unlock(sk);
379 }
380 
381 static struct rtable *icmp_route_lookup(struct net *net,
382 					struct flowi4 *fl4,
383 					struct sk_buff *skb_in,
384 					const struct iphdr *iph,
385 					__be32 saddr, u8 tos,
386 					int type, int code,
387 					struct icmp_bxm *param)
388 {
389 	struct rtable *rt, *rt2;
390 	struct flowi4 fl4_dec;
391 	int err;
392 
393 	memset(fl4, 0, sizeof(*fl4));
394 	fl4->daddr = (param->replyopts.opt.opt.srr ?
395 		      param->replyopts.opt.opt.faddr : iph->saddr);
396 	fl4->saddr = saddr;
397 	fl4->flowi4_tos = RT_TOS(tos);
398 	fl4->flowi4_proto = IPPROTO_ICMP;
399 	fl4->fl4_icmp_type = type;
400 	fl4->fl4_icmp_code = code;
401 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
402 	rt = __ip_route_output_key(net, fl4);
403 	if (IS_ERR(rt))
404 		return rt;
405 
406 	/* No need to clone since we're just using its address. */
407 	rt2 = rt;
408 
409 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
410 					   flowi4_to_flowi(fl4), NULL, 0);
411 	if (!IS_ERR(rt)) {
412 		if (rt != rt2)
413 			return rt;
414 	} else if (PTR_ERR(rt) == -EPERM) {
415 		rt = NULL;
416 	} else
417 		return rt;
418 
419 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
420 	if (err)
421 		goto relookup_failed;
422 
423 	if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
424 		rt2 = __ip_route_output_key(net, &fl4_dec);
425 		if (IS_ERR(rt2))
426 			err = PTR_ERR(rt2);
427 	} else {
428 		struct flowi4 fl4_2 = {};
429 		unsigned long orefdst;
430 
431 		fl4_2.daddr = fl4_dec.saddr;
432 		rt2 = ip_route_output_key(net, &fl4_2);
433 		if (IS_ERR(rt2)) {
434 			err = PTR_ERR(rt2);
435 			goto relookup_failed;
436 		}
437 		/* Ugh! */
438 		orefdst = skb_in->_skb_refdst; /* save old refdst */
439 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
440 				     RT_TOS(tos), rt2->dst.dev);
441 
442 		dst_release(&rt2->dst);
443 		rt2 = skb_rtable(skb_in);
444 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
445 	}
446 
447 	if (err)
448 		goto relookup_failed;
449 
450 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
451 					    flowi4_to_flowi(&fl4_dec), NULL,
452 					    XFRM_LOOKUP_ICMP);
453 	if (!IS_ERR(rt2)) {
454 		dst_release(&rt->dst);
455 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
456 		rt = rt2;
457 	} else if (PTR_ERR(rt2) == -EPERM) {
458 		if (rt)
459 			dst_release(&rt->dst);
460 		return rt2;
461 	} else {
462 		err = PTR_ERR(rt2);
463 		goto relookup_failed;
464 	}
465 	return rt;
466 
467 relookup_failed:
468 	if (rt)
469 		return rt;
470 	return ERR_PTR(err);
471 }
472 
473 /*
474  *	Send an ICMP message in response to a situation
475  *
476  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
477  *		  MAY send more (we do).
478  *			MUST NOT change this header information.
479  *			MUST NOT reply to a multicast/broadcast IP address.
480  *			MUST NOT reply to a multicast/broadcast MAC address.
481  *			MUST reply to only the first fragment.
482  */
483 
484 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
485 {
486 	struct iphdr *iph;
487 	int room;
488 	struct icmp_bxm *icmp_param;
489 	struct rtable *rt = skb_rtable(skb_in);
490 	struct ipcm_cookie ipc;
491 	struct flowi4 fl4;
492 	__be32 saddr;
493 	u8  tos;
494 	struct net *net;
495 	struct sock *sk;
496 
497 	if (!rt)
498 		goto out;
499 	net = dev_net(rt->dst.dev);
500 
501 	/*
502 	 *	Find the original header. It is expected to be valid, of course.
503 	 *	Check this, icmp_send is called from the most obscure devices
504 	 *	sometimes.
505 	 */
506 	iph = ip_hdr(skb_in);
507 
508 	if ((u8 *)iph < skb_in->head ||
509 	    (skb_network_header(skb_in) + sizeof(*iph)) >
510 	    skb_tail_pointer(skb_in))
511 		goto out;
512 
513 	/*
514 	 *	No replies to physical multicast/broadcast
515 	 */
516 	if (skb_in->pkt_type != PACKET_HOST)
517 		goto out;
518 
519 	/*
520 	 *	Now check at the protocol level
521 	 */
522 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
523 		goto out;
524 
525 	/*
526 	 *	Only reply to fragment 0. We byte re-order the constant
527 	 *	mask for efficiency.
528 	 */
529 	if (iph->frag_off & htons(IP_OFFSET))
530 		goto out;
531 
532 	/*
533 	 *	If we send an ICMP error to an ICMP error a mess would result..
534 	 */
535 	if (icmp_pointers[type].error) {
536 		/*
537 		 *	We are an error, check if we are replying to an
538 		 *	ICMP error
539 		 */
540 		if (iph->protocol == IPPROTO_ICMP) {
541 			u8 _inner_type, *itp;
542 
543 			itp = skb_header_pointer(skb_in,
544 						 skb_network_header(skb_in) +
545 						 (iph->ihl << 2) +
546 						 offsetof(struct icmphdr,
547 							  type) -
548 						 skb_in->data,
549 						 sizeof(_inner_type),
550 						 &_inner_type);
551 			if (itp == NULL)
552 				goto out;
553 
554 			/*
555 			 *	Assume any unknown ICMP type is an error. This
556 			 *	isn't specified by the RFC, but think about it..
557 			 */
558 			if (*itp > NR_ICMP_TYPES ||
559 			    icmp_pointers[*itp].error)
560 				goto out;
561 		}
562 	}
563 
564 	icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
565 	if (!icmp_param)
566 		return;
567 
568 	sk = icmp_xmit_lock(net);
569 	if (sk == NULL)
570 		goto out_free;
571 
572 	/*
573 	 *	Construct source address and options.
574 	 */
575 
576 	saddr = iph->daddr;
577 	if (!(rt->rt_flags & RTCF_LOCAL)) {
578 		struct net_device *dev = NULL;
579 
580 		rcu_read_lock();
581 		if (rt_is_input_route(rt) &&
582 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
583 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
584 
585 		if (dev)
586 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
587 		else
588 			saddr = 0;
589 		rcu_read_unlock();
590 	}
591 
592 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
593 					   IPTOS_PREC_INTERNETCONTROL) :
594 					  iph->tos;
595 
596 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
597 		goto out_unlock;
598 
599 
600 	/*
601 	 *	Prepare data for ICMP header.
602 	 */
603 
604 	icmp_param->data.icmph.type	 = type;
605 	icmp_param->data.icmph.code	 = code;
606 	icmp_param->data.icmph.un.gateway = info;
607 	icmp_param->data.icmph.checksum	 = 0;
608 	icmp_param->skb	  = skb_in;
609 	icmp_param->offset = skb_network_offset(skb_in);
610 	inet_sk(sk)->tos = tos;
611 	ipc.addr = iph->saddr;
612 	ipc.opt = &icmp_param->replyopts.opt;
613 	ipc.tx_flags = 0;
614 	ipc.ttl = 0;
615 	ipc.tos = -1;
616 
617 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
618 			       type, code, icmp_param);
619 	if (IS_ERR(rt))
620 		goto out_unlock;
621 
622 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
623 		goto ende;
624 
625 	/* RFC says return as much as we can without exceeding 576 bytes. */
626 
627 	room = dst_mtu(&rt->dst);
628 	if (room > 576)
629 		room = 576;
630 	room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
631 	room -= sizeof(struct icmphdr);
632 
633 	icmp_param->data_len = skb_in->len - icmp_param->offset;
634 	if (icmp_param->data_len > room)
635 		icmp_param->data_len = room;
636 	icmp_param->head_len = sizeof(struct icmphdr);
637 
638 	icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
639 ende:
640 	ip_rt_put(rt);
641 out_unlock:
642 	icmp_xmit_unlock(sk);
643 out_free:
644 	kfree(icmp_param);
645 out:;
646 }
647 EXPORT_SYMBOL(icmp_send);
648 
649 
650 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
651 {
652 	const struct iphdr *iph = (const struct iphdr *) skb->data;
653 	const struct net_protocol *ipprot;
654 	int protocol = iph->protocol;
655 
656 	/* Checkin full IP header plus 8 bytes of protocol to
657 	 * avoid additional coding at protocol handlers.
658 	 */
659 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
660 		return;
661 
662 	raw_icmp_error(skb, protocol, info);
663 
664 	rcu_read_lock();
665 	ipprot = rcu_dereference(inet_protos[protocol]);
666 	if (ipprot && ipprot->err_handler)
667 		ipprot->err_handler(skb, info);
668 	rcu_read_unlock();
669 }
670 
671 /*
672  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
673  *	ICMP_PARAMETERPROB.
674  */
675 
676 static void icmp_unreach(struct sk_buff *skb)
677 {
678 	const struct iphdr *iph;
679 	struct icmphdr *icmph;
680 	struct net *net;
681 	u32 info = 0;
682 
683 	net = dev_net(skb_dst(skb)->dev);
684 
685 	/*
686 	 *	Incomplete header ?
687 	 * 	Only checks for the IP header, there should be an
688 	 *	additional check for longer headers in upper levels.
689 	 */
690 
691 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
692 		goto out_err;
693 
694 	icmph = icmp_hdr(skb);
695 	iph   = (const struct iphdr *)skb->data;
696 
697 	if (iph->ihl < 5) /* Mangled header, drop. */
698 		goto out_err;
699 
700 	if (icmph->type == ICMP_DEST_UNREACH) {
701 		switch (icmph->code & 15) {
702 		case ICMP_NET_UNREACH:
703 		case ICMP_HOST_UNREACH:
704 		case ICMP_PROT_UNREACH:
705 		case ICMP_PORT_UNREACH:
706 			break;
707 		case ICMP_FRAG_NEEDED:
708 			if (ipv4_config.no_pmtu_disc) {
709 				LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
710 					       &iph->daddr);
711 			} else {
712 				info = ntohs(icmph->un.frag.mtu);
713 				if (!info)
714 					goto out;
715 			}
716 			break;
717 		case ICMP_SR_FAILED:
718 			LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
719 				       &iph->daddr);
720 			break;
721 		default:
722 			break;
723 		}
724 		if (icmph->code > NR_ICMP_UNREACH)
725 			goto out;
726 	} else if (icmph->type == ICMP_PARAMETERPROB)
727 		info = ntohl(icmph->un.gateway) >> 24;
728 
729 	/*
730 	 *	Throw it at our lower layers
731 	 *
732 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
733 	 *		  header.
734 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
735 	 *		  transport layer.
736 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
737 	 *		  transport layer.
738 	 */
739 
740 	/*
741 	 *	Check the other end isn't violating RFC 1122. Some routers send
742 	 *	bogus responses to broadcast frames. If you see this message
743 	 *	first check your netmask matches at both ends, if it does then
744 	 *	get the other vendor to fix their kit.
745 	 */
746 
747 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
748 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
749 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
750 				     &ip_hdr(skb)->saddr,
751 				     icmph->type, icmph->code,
752 				     &iph->daddr, skb->dev->name);
753 		goto out;
754 	}
755 
756 	icmp_socket_deliver(skb, info);
757 
758 out:
759 	return;
760 out_err:
761 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
762 	goto out;
763 }
764 
765 
766 /*
767  *	Handle ICMP_REDIRECT.
768  */
769 
770 static void icmp_redirect(struct sk_buff *skb)
771 {
772 	if (skb->len < sizeof(struct iphdr)) {
773 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
774 		return;
775 	}
776 
777 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
778 		return;
779 
780 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
781 }
782 
783 /*
784  *	Handle ICMP_ECHO ("ping") requests.
785  *
786  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
787  *		  requests.
788  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
789  *		  included in the reply.
790  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
791  *		  echo requests, MUST have default=NOT.
792  *	See also WRT handling of options once they are done and working.
793  */
794 
795 static void icmp_echo(struct sk_buff *skb)
796 {
797 	struct net *net;
798 
799 	net = dev_net(skb_dst(skb)->dev);
800 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
801 		struct icmp_bxm icmp_param;
802 
803 		icmp_param.data.icmph	   = *icmp_hdr(skb);
804 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
805 		icmp_param.skb		   = skb;
806 		icmp_param.offset	   = 0;
807 		icmp_param.data_len	   = skb->len;
808 		icmp_param.head_len	   = sizeof(struct icmphdr);
809 		icmp_reply(&icmp_param, skb);
810 	}
811 }
812 
813 /*
814  *	Handle ICMP Timestamp requests.
815  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
816  *		  SHOULD be in the kernel for minimum random latency.
817  *		  MUST be accurate to a few minutes.
818  *		  MUST be updated at least at 15Hz.
819  */
820 static void icmp_timestamp(struct sk_buff *skb)
821 {
822 	struct timespec tv;
823 	struct icmp_bxm icmp_param;
824 	/*
825 	 *	Too short.
826 	 */
827 	if (skb->len < 4)
828 		goto out_err;
829 
830 	/*
831 	 *	Fill in the current time as ms since midnight UT:
832 	 */
833 	getnstimeofday(&tv);
834 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
835 					 tv.tv_nsec / NSEC_PER_MSEC);
836 	icmp_param.data.times[2] = icmp_param.data.times[1];
837 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
838 		BUG();
839 	icmp_param.data.icmph	   = *icmp_hdr(skb);
840 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
841 	icmp_param.data.icmph.code = 0;
842 	icmp_param.skb		   = skb;
843 	icmp_param.offset	   = 0;
844 	icmp_param.data_len	   = 0;
845 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
846 	icmp_reply(&icmp_param, skb);
847 out:
848 	return;
849 out_err:
850 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
851 	goto out;
852 }
853 
854 static void icmp_discard(struct sk_buff *skb)
855 {
856 }
857 
858 /*
859  *	Deal with incoming ICMP packets.
860  */
861 int icmp_rcv(struct sk_buff *skb)
862 {
863 	struct icmphdr *icmph;
864 	struct rtable *rt = skb_rtable(skb);
865 	struct net *net = dev_net(rt->dst.dev);
866 
867 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
868 		struct sec_path *sp = skb_sec_path(skb);
869 		int nh;
870 
871 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
872 				 XFRM_STATE_ICMP))
873 			goto drop;
874 
875 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
876 			goto drop;
877 
878 		nh = skb_network_offset(skb);
879 		skb_set_network_header(skb, sizeof(*icmph));
880 
881 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
882 			goto drop;
883 
884 		skb_set_network_header(skb, nh);
885 	}
886 
887 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
888 
889 	switch (skb->ip_summed) {
890 	case CHECKSUM_COMPLETE:
891 		if (!csum_fold(skb->csum))
892 			break;
893 		/* fall through */
894 	case CHECKSUM_NONE:
895 		skb->csum = 0;
896 		if (__skb_checksum_complete(skb))
897 			goto csum_error;
898 	}
899 
900 	if (!pskb_pull(skb, sizeof(*icmph)))
901 		goto error;
902 
903 	icmph = icmp_hdr(skb);
904 
905 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
906 	/*
907 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
908 	 *
909 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
910 	 *		  discarded.
911 	 */
912 	if (icmph->type > NR_ICMP_TYPES)
913 		goto error;
914 
915 
916 	/*
917 	 *	Parse the ICMP message
918 	 */
919 
920 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
921 		/*
922 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
923 		 *	  silently ignored (we let user decide with a sysctl).
924 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
925 		 *	  discarded if to broadcast/multicast.
926 		 */
927 		if ((icmph->type == ICMP_ECHO ||
928 		     icmph->type == ICMP_TIMESTAMP) &&
929 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
930 			goto error;
931 		}
932 		if (icmph->type != ICMP_ECHO &&
933 		    icmph->type != ICMP_TIMESTAMP &&
934 		    icmph->type != ICMP_ADDRESS &&
935 		    icmph->type != ICMP_ADDRESSREPLY) {
936 			goto error;
937 		}
938 	}
939 
940 	icmp_pointers[icmph->type].handler(skb);
941 
942 drop:
943 	kfree_skb(skb);
944 	return 0;
945 csum_error:
946 	ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
947 error:
948 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
949 	goto drop;
950 }
951 
952 void icmp_err(struct sk_buff *skb, u32 info)
953 {
954 	struct iphdr *iph = (struct iphdr *)skb->data;
955 	int offset = iph->ihl<<2;
956 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
957 	int type = icmp_hdr(skb)->type;
958 	int code = icmp_hdr(skb)->code;
959 	struct net *net = dev_net(skb->dev);
960 
961 	/*
962 	 * Use ping_err to handle all icmp errors except those
963 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
964 	 */
965 	if (icmph->type != ICMP_ECHOREPLY) {
966 		ping_err(skb, offset, info);
967 		return;
968 	}
969 
970 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
971 		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
972 	else if (type == ICMP_REDIRECT)
973 		ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
974 }
975 
976 /*
977  *	This table is the definition of how we handle ICMP.
978  */
979 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
980 	[ICMP_ECHOREPLY] = {
981 		.handler = ping_rcv,
982 	},
983 	[1] = {
984 		.handler = icmp_discard,
985 		.error = 1,
986 	},
987 	[2] = {
988 		.handler = icmp_discard,
989 		.error = 1,
990 	},
991 	[ICMP_DEST_UNREACH] = {
992 		.handler = icmp_unreach,
993 		.error = 1,
994 	},
995 	[ICMP_SOURCE_QUENCH] = {
996 		.handler = icmp_unreach,
997 		.error = 1,
998 	},
999 	[ICMP_REDIRECT] = {
1000 		.handler = icmp_redirect,
1001 		.error = 1,
1002 	},
1003 	[6] = {
1004 		.handler = icmp_discard,
1005 		.error = 1,
1006 	},
1007 	[7] = {
1008 		.handler = icmp_discard,
1009 		.error = 1,
1010 	},
1011 	[ICMP_ECHO] = {
1012 		.handler = icmp_echo,
1013 	},
1014 	[9] = {
1015 		.handler = icmp_discard,
1016 		.error = 1,
1017 	},
1018 	[10] = {
1019 		.handler = icmp_discard,
1020 		.error = 1,
1021 	},
1022 	[ICMP_TIME_EXCEEDED] = {
1023 		.handler = icmp_unreach,
1024 		.error = 1,
1025 	},
1026 	[ICMP_PARAMETERPROB] = {
1027 		.handler = icmp_unreach,
1028 		.error = 1,
1029 	},
1030 	[ICMP_TIMESTAMP] = {
1031 		.handler = icmp_timestamp,
1032 	},
1033 	[ICMP_TIMESTAMPREPLY] = {
1034 		.handler = icmp_discard,
1035 	},
1036 	[ICMP_INFO_REQUEST] = {
1037 		.handler = icmp_discard,
1038 	},
1039 	[ICMP_INFO_REPLY] = {
1040 		.handler = icmp_discard,
1041 	},
1042 	[ICMP_ADDRESS] = {
1043 		.handler = icmp_discard,
1044 	},
1045 	[ICMP_ADDRESSREPLY] = {
1046 		.handler = icmp_discard,
1047 	},
1048 };
1049 
1050 static void __net_exit icmp_sk_exit(struct net *net)
1051 {
1052 	int i;
1053 
1054 	for_each_possible_cpu(i)
1055 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1056 	kfree(net->ipv4.icmp_sk);
1057 	net->ipv4.icmp_sk = NULL;
1058 }
1059 
1060 static int __net_init icmp_sk_init(struct net *net)
1061 {
1062 	int i, err;
1063 
1064 	net->ipv4.icmp_sk =
1065 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1066 	if (net->ipv4.icmp_sk == NULL)
1067 		return -ENOMEM;
1068 
1069 	for_each_possible_cpu(i) {
1070 		struct sock *sk;
1071 
1072 		err = inet_ctl_sock_create(&sk, PF_INET,
1073 					   SOCK_RAW, IPPROTO_ICMP, net);
1074 		if (err < 0)
1075 			goto fail;
1076 
1077 		net->ipv4.icmp_sk[i] = sk;
1078 
1079 		/* Enough space for 2 64K ICMP packets, including
1080 		 * sk_buff/skb_shared_info struct overhead.
1081 		 */
1082 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1083 
1084 		/*
1085 		 * Speedup sock_wfree()
1086 		 */
1087 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1088 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1089 	}
1090 
1091 	/* Control parameters for ECHO replies. */
1092 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1093 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1094 
1095 	/* Control parameter - ignore bogus broadcast responses? */
1096 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1097 
1098 	/*
1099 	 * 	Configurable global rate limit.
1100 	 *
1101 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1102 	 *	bucket ratemask defines which icmp types are ratelimited by
1103 	 *	setting	it's bit position.
1104 	 *
1105 	 *	default:
1106 	 *	dest unreachable (3), source quench (4),
1107 	 *	time exceeded (11), parameter problem (12)
1108 	 */
1109 
1110 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1111 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1112 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1113 
1114 	return 0;
1115 
1116 fail:
1117 	for_each_possible_cpu(i)
1118 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1119 	kfree(net->ipv4.icmp_sk);
1120 	return err;
1121 }
1122 
1123 static struct pernet_operations __net_initdata icmp_sk_ops = {
1124        .init = icmp_sk_init,
1125        .exit = icmp_sk_exit,
1126 };
1127 
1128 int __init icmp_init(void)
1129 {
1130 	return register_pernet_subsys(&icmp_sk_ops);
1131 }
1132