xref: /linux/net/ipv4/icmp.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 #include <net/vrf.h>
100 
101 /*
102  *	Build xmit assembly blocks
103  */
104 
105 struct icmp_bxm {
106 	struct sk_buff *skb;
107 	int offset;
108 	int data_len;
109 
110 	struct {
111 		struct icmphdr icmph;
112 		__be32	       times[3];
113 	} data;
114 	int head_len;
115 	struct ip_options_data replyopts;
116 };
117 
118 /* An array of errno for error messages from dest unreach. */
119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
120 
121 const struct icmp_err icmp_err_convert[] = {
122 	{
123 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
124 		.fatal = 0,
125 	},
126 	{
127 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
128 		.fatal = 0,
129 	},
130 	{
131 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
132 		.fatal = 1,
133 	},
134 	{
135 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
136 		.fatal = 1,
137 	},
138 	{
139 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
140 		.fatal = 0,
141 	},
142 	{
143 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
144 		.fatal = 0,
145 	},
146 	{
147 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
148 		.fatal = 1,
149 	},
150 	{
151 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
152 		.fatal = 1,
153 	},
154 	{
155 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
156 		.fatal = 1,
157 	},
158 	{
159 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
160 		.fatal = 1,
161 	},
162 	{
163 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
164 		.fatal = 1,
165 	},
166 	{
167 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
168 		.fatal = 0,
169 	},
170 	{
171 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
172 		.fatal = 0,
173 	},
174 	{
175 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
176 		.fatal = 1,
177 	},
178 	{
179 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
180 		.fatal = 1,
181 	},
182 	{
183 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
184 		.fatal = 1,
185 	},
186 };
187 EXPORT_SYMBOL(icmp_err_convert);
188 
189 /*
190  *	ICMP control array. This specifies what to do with each ICMP.
191  */
192 
193 struct icmp_control {
194 	bool (*handler)(struct sk_buff *skb);
195 	short   error;		/* This ICMP is classed as an error message */
196 };
197 
198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
199 
200 /*
201  *	The ICMP socket(s). This is the most convenient way to flow control
202  *	our ICMP output as well as maintain a clean interface throughout
203  *	all layers. All Socketless IP sends will soon be gone.
204  *
205  *	On SMP we have one ICMP socket per-cpu.
206  */
207 static struct sock *icmp_sk(struct net *net)
208 {
209 	return *this_cpu_ptr(net->ipv4.icmp_sk);
210 }
211 
212 static inline struct sock *icmp_xmit_lock(struct net *net)
213 {
214 	struct sock *sk;
215 
216 	local_bh_disable();
217 
218 	sk = icmp_sk(net);
219 
220 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
221 		/* This can happen if the output path signals a
222 		 * dst_link_failure() for an outgoing ICMP packet.
223 		 */
224 		local_bh_enable();
225 		return NULL;
226 	}
227 	return sk;
228 }
229 
230 static inline void icmp_xmit_unlock(struct sock *sk)
231 {
232 	spin_unlock_bh(&sk->sk_lock.slock);
233 }
234 
235 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
236 int sysctl_icmp_msgs_burst __read_mostly = 50;
237 
238 static struct {
239 	spinlock_t	lock;
240 	u32		credit;
241 	u32		stamp;
242 } icmp_global = {
243 	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
244 };
245 
246 /**
247  * icmp_global_allow - Are we allowed to send one more ICMP message ?
248  *
249  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
250  * Returns false if we reached the limit and can not send another packet.
251  * Note: called with BH disabled
252  */
253 bool icmp_global_allow(void)
254 {
255 	u32 credit, delta, incr = 0, now = (u32)jiffies;
256 	bool rc = false;
257 
258 	/* Check if token bucket is empty and cannot be refilled
259 	 * without taking the spinlock.
260 	 */
261 	if (!icmp_global.credit) {
262 		delta = min_t(u32, now - icmp_global.stamp, HZ);
263 		if (delta < HZ / 50)
264 			return false;
265 	}
266 
267 	spin_lock(&icmp_global.lock);
268 	delta = min_t(u32, now - icmp_global.stamp, HZ);
269 	if (delta >= HZ / 50) {
270 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
271 		if (incr)
272 			icmp_global.stamp = now;
273 	}
274 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
275 	if (credit) {
276 		credit--;
277 		rc = true;
278 	}
279 	icmp_global.credit = credit;
280 	spin_unlock(&icmp_global.lock);
281 	return rc;
282 }
283 EXPORT_SYMBOL(icmp_global_allow);
284 
285 /*
286  *	Send an ICMP frame.
287  */
288 
289 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
290 			       struct flowi4 *fl4, int type, int code)
291 {
292 	struct dst_entry *dst = &rt->dst;
293 	bool rc = true;
294 
295 	if (type > NR_ICMP_TYPES)
296 		goto out;
297 
298 	/* Don't limit PMTU discovery. */
299 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
300 		goto out;
301 
302 	/* No rate limit on loopback */
303 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
304 		goto out;
305 
306 	/* Limit if icmp type is enabled in ratemask. */
307 	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
308 		goto out;
309 
310 	rc = false;
311 	if (icmp_global_allow()) {
312 		int vif = vrf_master_ifindex(dst->dev);
313 		struct inet_peer *peer;
314 
315 		peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
316 		rc = inet_peer_xrlim_allow(peer,
317 					   net->ipv4.sysctl_icmp_ratelimit);
318 		if (peer)
319 			inet_putpeer(peer);
320 	}
321 out:
322 	return rc;
323 }
324 
325 /*
326  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
327  */
328 void icmp_out_count(struct net *net, unsigned char type)
329 {
330 	ICMPMSGOUT_INC_STATS(net, type);
331 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
332 }
333 
334 /*
335  *	Checksum each fragment, and on the first include the headers and final
336  *	checksum.
337  */
338 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
339 			  struct sk_buff *skb)
340 {
341 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
342 	__wsum csum;
343 
344 	csum = skb_copy_and_csum_bits(icmp_param->skb,
345 				      icmp_param->offset + offset,
346 				      to, len, 0);
347 
348 	skb->csum = csum_block_add(skb->csum, csum, odd);
349 	if (icmp_pointers[icmp_param->data.icmph.type].error)
350 		nf_ct_attach(skb, icmp_param->skb);
351 	return 0;
352 }
353 
354 static void icmp_push_reply(struct icmp_bxm *icmp_param,
355 			    struct flowi4 *fl4,
356 			    struct ipcm_cookie *ipc, struct rtable **rt)
357 {
358 	struct sock *sk;
359 	struct sk_buff *skb;
360 
361 	sk = icmp_sk(dev_net((*rt)->dst.dev));
362 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
363 			   icmp_param->data_len+icmp_param->head_len,
364 			   icmp_param->head_len,
365 			   ipc, rt, MSG_DONTWAIT) < 0) {
366 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
367 		ip_flush_pending_frames(sk);
368 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
369 		struct icmphdr *icmph = icmp_hdr(skb);
370 		__wsum csum = 0;
371 		struct sk_buff *skb1;
372 
373 		skb_queue_walk(&sk->sk_write_queue, skb1) {
374 			csum = csum_add(csum, skb1->csum);
375 		}
376 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
377 						 (char *)icmph,
378 						 icmp_param->head_len, csum);
379 		icmph->checksum = csum_fold(csum);
380 		skb->ip_summed = CHECKSUM_NONE;
381 		ip_push_pending_frames(sk, fl4);
382 	}
383 }
384 
385 /*
386  *	Driving logic for building and sending ICMP messages.
387  */
388 
389 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
390 {
391 	struct ipcm_cookie ipc;
392 	struct rtable *rt = skb_rtable(skb);
393 	struct net *net = dev_net(rt->dst.dev);
394 	struct flowi4 fl4;
395 	struct sock *sk;
396 	struct inet_sock *inet;
397 	__be32 daddr, saddr;
398 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
399 
400 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
401 		return;
402 
403 	sk = icmp_xmit_lock(net);
404 	if (!sk)
405 		return;
406 	inet = inet_sk(sk);
407 
408 	icmp_param->data.icmph.checksum = 0;
409 
410 	inet->tos = ip_hdr(skb)->tos;
411 	sk->sk_mark = mark;
412 	daddr = ipc.addr = ip_hdr(skb)->saddr;
413 	saddr = fib_compute_spec_dst(skb);
414 	ipc.opt = NULL;
415 	ipc.tx_flags = 0;
416 	ipc.ttl = 0;
417 	ipc.tos = -1;
418 
419 	if (icmp_param->replyopts.opt.opt.optlen) {
420 		ipc.opt = &icmp_param->replyopts.opt;
421 		if (ipc.opt->opt.srr)
422 			daddr = icmp_param->replyopts.opt.opt.faddr;
423 	}
424 	memset(&fl4, 0, sizeof(fl4));
425 	fl4.daddr = daddr;
426 	fl4.saddr = saddr;
427 	fl4.flowi4_mark = mark;
428 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
429 	fl4.flowi4_proto = IPPROTO_ICMP;
430 	fl4.flowi4_oif = vrf_master_ifindex(skb->dev);
431 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
432 	rt = ip_route_output_key(net, &fl4);
433 	if (IS_ERR(rt))
434 		goto out_unlock;
435 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
436 			       icmp_param->data.icmph.code))
437 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
438 	ip_rt_put(rt);
439 out_unlock:
440 	icmp_xmit_unlock(sk);
441 }
442 
443 static struct rtable *icmp_route_lookup(struct net *net,
444 					struct flowi4 *fl4,
445 					struct sk_buff *skb_in,
446 					const struct iphdr *iph,
447 					__be32 saddr, u8 tos, u32 mark,
448 					int type, int code,
449 					struct icmp_bxm *param)
450 {
451 	struct rtable *rt, *rt2;
452 	struct flowi4 fl4_dec;
453 	int err;
454 
455 	memset(fl4, 0, sizeof(*fl4));
456 	fl4->daddr = (param->replyopts.opt.opt.srr ?
457 		      param->replyopts.opt.opt.faddr : iph->saddr);
458 	fl4->saddr = saddr;
459 	fl4->flowi4_mark = mark;
460 	fl4->flowi4_tos = RT_TOS(tos);
461 	fl4->flowi4_proto = IPPROTO_ICMP;
462 	fl4->fl4_icmp_type = type;
463 	fl4->fl4_icmp_code = code;
464 	fl4->flowi4_oif = vrf_master_ifindex(skb_in->dev);
465 
466 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
467 	rt = __ip_route_output_key(net, fl4);
468 	if (IS_ERR(rt))
469 		return rt;
470 
471 	/* No need to clone since we're just using its address. */
472 	rt2 = rt;
473 
474 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
475 					   flowi4_to_flowi(fl4), NULL, 0);
476 	if (!IS_ERR(rt)) {
477 		if (rt != rt2)
478 			return rt;
479 	} else if (PTR_ERR(rt) == -EPERM) {
480 		rt = NULL;
481 	} else
482 		return rt;
483 
484 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
485 	if (err)
486 		goto relookup_failed;
487 
488 	if (inet_addr_type_dev_table(net, skb_in->dev,
489 				     fl4_dec.saddr) == RTN_LOCAL) {
490 		rt2 = __ip_route_output_key(net, &fl4_dec);
491 		if (IS_ERR(rt2))
492 			err = PTR_ERR(rt2);
493 	} else {
494 		struct flowi4 fl4_2 = {};
495 		unsigned long orefdst;
496 
497 		fl4_2.daddr = fl4_dec.saddr;
498 		rt2 = ip_route_output_key(net, &fl4_2);
499 		if (IS_ERR(rt2)) {
500 			err = PTR_ERR(rt2);
501 			goto relookup_failed;
502 		}
503 		/* Ugh! */
504 		orefdst = skb_in->_skb_refdst; /* save old refdst */
505 		skb_dst_set(skb_in, NULL);
506 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
507 				     RT_TOS(tos), rt2->dst.dev);
508 
509 		dst_release(&rt2->dst);
510 		rt2 = skb_rtable(skb_in);
511 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
512 	}
513 
514 	if (err)
515 		goto relookup_failed;
516 
517 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
518 					    flowi4_to_flowi(&fl4_dec), NULL,
519 					    XFRM_LOOKUP_ICMP);
520 	if (!IS_ERR(rt2)) {
521 		dst_release(&rt->dst);
522 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
523 		rt = rt2;
524 	} else if (PTR_ERR(rt2) == -EPERM) {
525 		if (rt)
526 			dst_release(&rt->dst);
527 		return rt2;
528 	} else {
529 		err = PTR_ERR(rt2);
530 		goto relookup_failed;
531 	}
532 	return rt;
533 
534 relookup_failed:
535 	if (rt)
536 		return rt;
537 	return ERR_PTR(err);
538 }
539 
540 /*
541  *	Send an ICMP message in response to a situation
542  *
543  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
544  *		  MAY send more (we do).
545  *			MUST NOT change this header information.
546  *			MUST NOT reply to a multicast/broadcast IP address.
547  *			MUST NOT reply to a multicast/broadcast MAC address.
548  *			MUST reply to only the first fragment.
549  */
550 
551 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
552 {
553 	struct iphdr *iph;
554 	int room;
555 	struct icmp_bxm *icmp_param;
556 	struct rtable *rt = skb_rtable(skb_in);
557 	struct ipcm_cookie ipc;
558 	struct flowi4 fl4;
559 	__be32 saddr;
560 	u8  tos;
561 	u32 mark;
562 	struct net *net;
563 	struct sock *sk;
564 
565 	if (!rt)
566 		goto out;
567 	net = dev_net(rt->dst.dev);
568 
569 	/*
570 	 *	Find the original header. It is expected to be valid, of course.
571 	 *	Check this, icmp_send is called from the most obscure devices
572 	 *	sometimes.
573 	 */
574 	iph = ip_hdr(skb_in);
575 
576 	if ((u8 *)iph < skb_in->head ||
577 	    (skb_network_header(skb_in) + sizeof(*iph)) >
578 	    skb_tail_pointer(skb_in))
579 		goto out;
580 
581 	/*
582 	 *	No replies to physical multicast/broadcast
583 	 */
584 	if (skb_in->pkt_type != PACKET_HOST)
585 		goto out;
586 
587 	/*
588 	 *	Now check at the protocol level
589 	 */
590 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
591 		goto out;
592 
593 	/*
594 	 *	Only reply to fragment 0. We byte re-order the constant
595 	 *	mask for efficiency.
596 	 */
597 	if (iph->frag_off & htons(IP_OFFSET))
598 		goto out;
599 
600 	/*
601 	 *	If we send an ICMP error to an ICMP error a mess would result..
602 	 */
603 	if (icmp_pointers[type].error) {
604 		/*
605 		 *	We are an error, check if we are replying to an
606 		 *	ICMP error
607 		 */
608 		if (iph->protocol == IPPROTO_ICMP) {
609 			u8 _inner_type, *itp;
610 
611 			itp = skb_header_pointer(skb_in,
612 						 skb_network_header(skb_in) +
613 						 (iph->ihl << 2) +
614 						 offsetof(struct icmphdr,
615 							  type) -
616 						 skb_in->data,
617 						 sizeof(_inner_type),
618 						 &_inner_type);
619 			if (!itp)
620 				goto out;
621 
622 			/*
623 			 *	Assume any unknown ICMP type is an error. This
624 			 *	isn't specified by the RFC, but think about it..
625 			 */
626 			if (*itp > NR_ICMP_TYPES ||
627 			    icmp_pointers[*itp].error)
628 				goto out;
629 		}
630 	}
631 
632 	icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
633 	if (!icmp_param)
634 		return;
635 
636 	sk = icmp_xmit_lock(net);
637 	if (!sk)
638 		goto out_free;
639 
640 	/*
641 	 *	Construct source address and options.
642 	 */
643 
644 	saddr = iph->daddr;
645 	if (!(rt->rt_flags & RTCF_LOCAL)) {
646 		struct net_device *dev = NULL;
647 
648 		rcu_read_lock();
649 		if (rt_is_input_route(rt) &&
650 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
651 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
652 
653 		if (dev)
654 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
655 		else
656 			saddr = 0;
657 		rcu_read_unlock();
658 	}
659 
660 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
661 					   IPTOS_PREC_INTERNETCONTROL) :
662 					  iph->tos;
663 	mark = IP4_REPLY_MARK(net, skb_in->mark);
664 
665 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
666 		goto out_unlock;
667 
668 
669 	/*
670 	 *	Prepare data for ICMP header.
671 	 */
672 
673 	icmp_param->data.icmph.type	 = type;
674 	icmp_param->data.icmph.code	 = code;
675 	icmp_param->data.icmph.un.gateway = info;
676 	icmp_param->data.icmph.checksum	 = 0;
677 	icmp_param->skb	  = skb_in;
678 	icmp_param->offset = skb_network_offset(skb_in);
679 	inet_sk(sk)->tos = tos;
680 	sk->sk_mark = mark;
681 	ipc.addr = iph->saddr;
682 	ipc.opt = &icmp_param->replyopts.opt;
683 	ipc.tx_flags = 0;
684 	ipc.ttl = 0;
685 	ipc.tos = -1;
686 
687 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
688 			       type, code, icmp_param);
689 	if (IS_ERR(rt))
690 		goto out_unlock;
691 
692 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
693 		goto ende;
694 
695 	/* RFC says return as much as we can without exceeding 576 bytes. */
696 
697 	room = dst_mtu(&rt->dst);
698 	if (room > 576)
699 		room = 576;
700 	room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
701 	room -= sizeof(struct icmphdr);
702 
703 	icmp_param->data_len = skb_in->len - icmp_param->offset;
704 	if (icmp_param->data_len > room)
705 		icmp_param->data_len = room;
706 	icmp_param->head_len = sizeof(struct icmphdr);
707 
708 	icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
709 ende:
710 	ip_rt_put(rt);
711 out_unlock:
712 	icmp_xmit_unlock(sk);
713 out_free:
714 	kfree(icmp_param);
715 out:;
716 }
717 EXPORT_SYMBOL(icmp_send);
718 
719 
720 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
721 {
722 	const struct iphdr *iph = (const struct iphdr *) skb->data;
723 	const struct net_protocol *ipprot;
724 	int protocol = iph->protocol;
725 
726 	/* Checkin full IP header plus 8 bytes of protocol to
727 	 * avoid additional coding at protocol handlers.
728 	 */
729 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
730 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
731 		return;
732 	}
733 
734 	raw_icmp_error(skb, protocol, info);
735 
736 	ipprot = rcu_dereference(inet_protos[protocol]);
737 	if (ipprot && ipprot->err_handler)
738 		ipprot->err_handler(skb, info);
739 }
740 
741 static bool icmp_tag_validation(int proto)
742 {
743 	bool ok;
744 
745 	rcu_read_lock();
746 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
747 	rcu_read_unlock();
748 	return ok;
749 }
750 
751 /*
752  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
753  *	ICMP_PARAMETERPROB.
754  */
755 
756 static bool icmp_unreach(struct sk_buff *skb)
757 {
758 	const struct iphdr *iph;
759 	struct icmphdr *icmph;
760 	struct net *net;
761 	u32 info = 0;
762 
763 	net = dev_net(skb_dst(skb)->dev);
764 
765 	/*
766 	 *	Incomplete header ?
767 	 * 	Only checks for the IP header, there should be an
768 	 *	additional check for longer headers in upper levels.
769 	 */
770 
771 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
772 		goto out_err;
773 
774 	icmph = icmp_hdr(skb);
775 	iph   = (const struct iphdr *)skb->data;
776 
777 	if (iph->ihl < 5) /* Mangled header, drop. */
778 		goto out_err;
779 
780 	if (icmph->type == ICMP_DEST_UNREACH) {
781 		switch (icmph->code & 15) {
782 		case ICMP_NET_UNREACH:
783 		case ICMP_HOST_UNREACH:
784 		case ICMP_PROT_UNREACH:
785 		case ICMP_PORT_UNREACH:
786 			break;
787 		case ICMP_FRAG_NEEDED:
788 			/* for documentation of the ip_no_pmtu_disc
789 			 * values please see
790 			 * Documentation/networking/ip-sysctl.txt
791 			 */
792 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
793 			default:
794 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
795 						    &iph->daddr);
796 				break;
797 			case 2:
798 				goto out;
799 			case 3:
800 				if (!icmp_tag_validation(iph->protocol))
801 					goto out;
802 				/* fall through */
803 			case 0:
804 				info = ntohs(icmph->un.frag.mtu);
805 			}
806 			break;
807 		case ICMP_SR_FAILED:
808 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
809 					    &iph->daddr);
810 			break;
811 		default:
812 			break;
813 		}
814 		if (icmph->code > NR_ICMP_UNREACH)
815 			goto out;
816 	} else if (icmph->type == ICMP_PARAMETERPROB)
817 		info = ntohl(icmph->un.gateway) >> 24;
818 
819 	/*
820 	 *	Throw it at our lower layers
821 	 *
822 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
823 	 *		  header.
824 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
825 	 *		  transport layer.
826 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
827 	 *		  transport layer.
828 	 */
829 
830 	/*
831 	 *	Check the other end isn't violating RFC 1122. Some routers send
832 	 *	bogus responses to broadcast frames. If you see this message
833 	 *	first check your netmask matches at both ends, if it does then
834 	 *	get the other vendor to fix their kit.
835 	 */
836 
837 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
838 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
839 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
840 				     &ip_hdr(skb)->saddr,
841 				     icmph->type, icmph->code,
842 				     &iph->daddr, skb->dev->name);
843 		goto out;
844 	}
845 
846 	icmp_socket_deliver(skb, info);
847 
848 out:
849 	return true;
850 out_err:
851 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
852 	return false;
853 }
854 
855 
856 /*
857  *	Handle ICMP_REDIRECT.
858  */
859 
860 static bool icmp_redirect(struct sk_buff *skb)
861 {
862 	if (skb->len < sizeof(struct iphdr)) {
863 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
864 		return false;
865 	}
866 
867 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
868 		/* there aught to be a stat */
869 		return false;
870 	}
871 
872 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
873 	return true;
874 }
875 
876 /*
877  *	Handle ICMP_ECHO ("ping") requests.
878  *
879  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
880  *		  requests.
881  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
882  *		  included in the reply.
883  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
884  *		  echo requests, MUST have default=NOT.
885  *	See also WRT handling of options once they are done and working.
886  */
887 
888 static bool icmp_echo(struct sk_buff *skb)
889 {
890 	struct net *net;
891 
892 	net = dev_net(skb_dst(skb)->dev);
893 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
894 		struct icmp_bxm icmp_param;
895 
896 		icmp_param.data.icmph	   = *icmp_hdr(skb);
897 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
898 		icmp_param.skb		   = skb;
899 		icmp_param.offset	   = 0;
900 		icmp_param.data_len	   = skb->len;
901 		icmp_param.head_len	   = sizeof(struct icmphdr);
902 		icmp_reply(&icmp_param, skb);
903 	}
904 	/* should there be an ICMP stat for ignored echos? */
905 	return true;
906 }
907 
908 /*
909  *	Handle ICMP Timestamp requests.
910  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
911  *		  SHOULD be in the kernel for minimum random latency.
912  *		  MUST be accurate to a few minutes.
913  *		  MUST be updated at least at 15Hz.
914  */
915 static bool icmp_timestamp(struct sk_buff *skb)
916 {
917 	struct timespec tv;
918 	struct icmp_bxm icmp_param;
919 	/*
920 	 *	Too short.
921 	 */
922 	if (skb->len < 4)
923 		goto out_err;
924 
925 	/*
926 	 *	Fill in the current time as ms since midnight UT:
927 	 */
928 	getnstimeofday(&tv);
929 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
930 					 tv.tv_nsec / NSEC_PER_MSEC);
931 	icmp_param.data.times[2] = icmp_param.data.times[1];
932 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
933 		BUG();
934 	icmp_param.data.icmph	   = *icmp_hdr(skb);
935 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
936 	icmp_param.data.icmph.code = 0;
937 	icmp_param.skb		   = skb;
938 	icmp_param.offset	   = 0;
939 	icmp_param.data_len	   = 0;
940 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
941 	icmp_reply(&icmp_param, skb);
942 	return true;
943 
944 out_err:
945 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
946 	return false;
947 }
948 
949 static bool icmp_discard(struct sk_buff *skb)
950 {
951 	/* pretend it was a success */
952 	return true;
953 }
954 
955 /*
956  *	Deal with incoming ICMP packets.
957  */
958 int icmp_rcv(struct sk_buff *skb)
959 {
960 	struct icmphdr *icmph;
961 	struct rtable *rt = skb_rtable(skb);
962 	struct net *net = dev_net(rt->dst.dev);
963 	bool success;
964 
965 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
966 		struct sec_path *sp = skb_sec_path(skb);
967 		int nh;
968 
969 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
970 				 XFRM_STATE_ICMP))
971 			goto drop;
972 
973 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
974 			goto drop;
975 
976 		nh = skb_network_offset(skb);
977 		skb_set_network_header(skb, sizeof(*icmph));
978 
979 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
980 			goto drop;
981 
982 		skb_set_network_header(skb, nh);
983 	}
984 
985 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
986 
987 	if (skb_checksum_simple_validate(skb))
988 		goto csum_error;
989 
990 	if (!pskb_pull(skb, sizeof(*icmph)))
991 		goto error;
992 
993 	icmph = icmp_hdr(skb);
994 
995 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
996 	/*
997 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
998 	 *
999 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1000 	 *		  discarded.
1001 	 */
1002 	if (icmph->type > NR_ICMP_TYPES)
1003 		goto error;
1004 
1005 
1006 	/*
1007 	 *	Parse the ICMP message
1008 	 */
1009 
1010 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1011 		/*
1012 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1013 		 *	  silently ignored (we let user decide with a sysctl).
1014 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1015 		 *	  discarded if to broadcast/multicast.
1016 		 */
1017 		if ((icmph->type == ICMP_ECHO ||
1018 		     icmph->type == ICMP_TIMESTAMP) &&
1019 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1020 			goto error;
1021 		}
1022 		if (icmph->type != ICMP_ECHO &&
1023 		    icmph->type != ICMP_TIMESTAMP &&
1024 		    icmph->type != ICMP_ADDRESS &&
1025 		    icmph->type != ICMP_ADDRESSREPLY) {
1026 			goto error;
1027 		}
1028 	}
1029 
1030 	success = icmp_pointers[icmph->type].handler(skb);
1031 
1032 	if (success)  {
1033 		consume_skb(skb);
1034 		return 0;
1035 	}
1036 
1037 drop:
1038 	kfree_skb(skb);
1039 	return 0;
1040 csum_error:
1041 	ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
1042 error:
1043 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1044 	goto drop;
1045 }
1046 
1047 void icmp_err(struct sk_buff *skb, u32 info)
1048 {
1049 	struct iphdr *iph = (struct iphdr *)skb->data;
1050 	int offset = iph->ihl<<2;
1051 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1052 	int type = icmp_hdr(skb)->type;
1053 	int code = icmp_hdr(skb)->code;
1054 	struct net *net = dev_net(skb->dev);
1055 
1056 	/*
1057 	 * Use ping_err to handle all icmp errors except those
1058 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1059 	 */
1060 	if (icmph->type != ICMP_ECHOREPLY) {
1061 		ping_err(skb, offset, info);
1062 		return;
1063 	}
1064 
1065 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1066 		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1067 	else if (type == ICMP_REDIRECT)
1068 		ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1069 }
1070 
1071 /*
1072  *	This table is the definition of how we handle ICMP.
1073  */
1074 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1075 	[ICMP_ECHOREPLY] = {
1076 		.handler = ping_rcv,
1077 	},
1078 	[1] = {
1079 		.handler = icmp_discard,
1080 		.error = 1,
1081 	},
1082 	[2] = {
1083 		.handler = icmp_discard,
1084 		.error = 1,
1085 	},
1086 	[ICMP_DEST_UNREACH] = {
1087 		.handler = icmp_unreach,
1088 		.error = 1,
1089 	},
1090 	[ICMP_SOURCE_QUENCH] = {
1091 		.handler = icmp_unreach,
1092 		.error = 1,
1093 	},
1094 	[ICMP_REDIRECT] = {
1095 		.handler = icmp_redirect,
1096 		.error = 1,
1097 	},
1098 	[6] = {
1099 		.handler = icmp_discard,
1100 		.error = 1,
1101 	},
1102 	[7] = {
1103 		.handler = icmp_discard,
1104 		.error = 1,
1105 	},
1106 	[ICMP_ECHO] = {
1107 		.handler = icmp_echo,
1108 	},
1109 	[9] = {
1110 		.handler = icmp_discard,
1111 		.error = 1,
1112 	},
1113 	[10] = {
1114 		.handler = icmp_discard,
1115 		.error = 1,
1116 	},
1117 	[ICMP_TIME_EXCEEDED] = {
1118 		.handler = icmp_unreach,
1119 		.error = 1,
1120 	},
1121 	[ICMP_PARAMETERPROB] = {
1122 		.handler = icmp_unreach,
1123 		.error = 1,
1124 	},
1125 	[ICMP_TIMESTAMP] = {
1126 		.handler = icmp_timestamp,
1127 	},
1128 	[ICMP_TIMESTAMPREPLY] = {
1129 		.handler = icmp_discard,
1130 	},
1131 	[ICMP_INFO_REQUEST] = {
1132 		.handler = icmp_discard,
1133 	},
1134 	[ICMP_INFO_REPLY] = {
1135 		.handler = icmp_discard,
1136 	},
1137 	[ICMP_ADDRESS] = {
1138 		.handler = icmp_discard,
1139 	},
1140 	[ICMP_ADDRESSREPLY] = {
1141 		.handler = icmp_discard,
1142 	},
1143 };
1144 
1145 static void __net_exit icmp_sk_exit(struct net *net)
1146 {
1147 	int i;
1148 
1149 	for_each_possible_cpu(i)
1150 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1151 	free_percpu(net->ipv4.icmp_sk);
1152 	net->ipv4.icmp_sk = NULL;
1153 }
1154 
1155 static int __net_init icmp_sk_init(struct net *net)
1156 {
1157 	int i, err;
1158 
1159 	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1160 	if (!net->ipv4.icmp_sk)
1161 		return -ENOMEM;
1162 
1163 	for_each_possible_cpu(i) {
1164 		struct sock *sk;
1165 
1166 		err = inet_ctl_sock_create(&sk, PF_INET,
1167 					   SOCK_RAW, IPPROTO_ICMP, net);
1168 		if (err < 0)
1169 			goto fail;
1170 
1171 		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1172 
1173 		/* Enough space for 2 64K ICMP packets, including
1174 		 * sk_buff/skb_shared_info struct overhead.
1175 		 */
1176 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1177 
1178 		/*
1179 		 * Speedup sock_wfree()
1180 		 */
1181 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1182 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1183 	}
1184 
1185 	/* Control parameters for ECHO replies. */
1186 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1187 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1188 
1189 	/* Control parameter - ignore bogus broadcast responses? */
1190 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1191 
1192 	/*
1193 	 * 	Configurable global rate limit.
1194 	 *
1195 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1196 	 *	bucket ratemask defines which icmp types are ratelimited by
1197 	 *	setting	it's bit position.
1198 	 *
1199 	 *	default:
1200 	 *	dest unreachable (3), source quench (4),
1201 	 *	time exceeded (11), parameter problem (12)
1202 	 */
1203 
1204 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1205 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1206 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1207 
1208 	return 0;
1209 
1210 fail:
1211 	for_each_possible_cpu(i)
1212 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1213 	free_percpu(net->ipv4.icmp_sk);
1214 	return err;
1215 }
1216 
1217 static struct pernet_operations __net_initdata icmp_sk_ops = {
1218        .init = icmp_sk_init,
1219        .exit = icmp_sk_exit,
1220 };
1221 
1222 int __init icmp_init(void)
1223 {
1224 	return register_pernet_subsys(&icmp_sk_ops);
1225 }
1226