xref: /openbmc/linux/net/ipv4/icmp.c (revision 827634ad)
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 	bool (*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 *this_cpu_ptr(net->ipv4.icmp_sk);
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 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
235 int sysctl_icmp_msgs_burst __read_mostly = 50;
236 
237 static struct {
238 	spinlock_t	lock;
239 	u32		credit;
240 	u32		stamp;
241 } icmp_global = {
242 	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
243 };
244 
245 /**
246  * icmp_global_allow - Are we allowed to send one more ICMP message ?
247  *
248  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
249  * Returns false if we reached the limit and can not send another packet.
250  * Note: called with BH disabled
251  */
252 bool icmp_global_allow(void)
253 {
254 	u32 credit, delta, incr = 0, now = (u32)jiffies;
255 	bool rc = false;
256 
257 	/* Check if token bucket is empty and cannot be refilled
258 	 * without taking the spinlock.
259 	 */
260 	if (!icmp_global.credit) {
261 		delta = min_t(u32, now - icmp_global.stamp, HZ);
262 		if (delta < HZ / 50)
263 			return false;
264 	}
265 
266 	spin_lock(&icmp_global.lock);
267 	delta = min_t(u32, now - icmp_global.stamp, HZ);
268 	if (delta >= HZ / 50) {
269 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
270 		if (incr)
271 			icmp_global.stamp = now;
272 	}
273 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
274 	if (credit) {
275 		credit--;
276 		rc = true;
277 	}
278 	icmp_global.credit = credit;
279 	spin_unlock(&icmp_global.lock);
280 	return rc;
281 }
282 EXPORT_SYMBOL(icmp_global_allow);
283 
284 /*
285  *	Send an ICMP frame.
286  */
287 
288 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
289 			       struct flowi4 *fl4, int type, int code)
290 {
291 	struct dst_entry *dst = &rt->dst;
292 	bool rc = true;
293 
294 	if (type > NR_ICMP_TYPES)
295 		goto out;
296 
297 	/* Don't limit PMTU discovery. */
298 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
299 		goto out;
300 
301 	/* No rate limit on loopback */
302 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
303 		goto out;
304 
305 	/* Limit if icmp type is enabled in ratemask. */
306 	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
307 		goto out;
308 
309 	rc = false;
310 	if (icmp_global_allow()) {
311 		struct inet_peer *peer;
312 
313 		peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1);
314 		rc = inet_peer_xrlim_allow(peer,
315 					   net->ipv4.sysctl_icmp_ratelimit);
316 		if (peer)
317 			inet_putpeer(peer);
318 	}
319 out:
320 	return rc;
321 }
322 
323 /*
324  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
325  */
326 void icmp_out_count(struct net *net, unsigned char type)
327 {
328 	ICMPMSGOUT_INC_STATS(net, type);
329 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
330 }
331 
332 /*
333  *	Checksum each fragment, and on the first include the headers and final
334  *	checksum.
335  */
336 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
337 			  struct sk_buff *skb)
338 {
339 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
340 	__wsum csum;
341 
342 	csum = skb_copy_and_csum_bits(icmp_param->skb,
343 				      icmp_param->offset + offset,
344 				      to, len, 0);
345 
346 	skb->csum = csum_block_add(skb->csum, csum, odd);
347 	if (icmp_pointers[icmp_param->data.icmph.type].error)
348 		nf_ct_attach(skb, icmp_param->skb);
349 	return 0;
350 }
351 
352 static void icmp_push_reply(struct icmp_bxm *icmp_param,
353 			    struct flowi4 *fl4,
354 			    struct ipcm_cookie *ipc, struct rtable **rt)
355 {
356 	struct sock *sk;
357 	struct sk_buff *skb;
358 
359 	sk = icmp_sk(dev_net((*rt)->dst.dev));
360 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
361 			   icmp_param->data_len+icmp_param->head_len,
362 			   icmp_param->head_len,
363 			   ipc, rt, MSG_DONTWAIT) < 0) {
364 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
365 		ip_flush_pending_frames(sk);
366 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
367 		struct icmphdr *icmph = icmp_hdr(skb);
368 		__wsum csum = 0;
369 		struct sk_buff *skb1;
370 
371 		skb_queue_walk(&sk->sk_write_queue, skb1) {
372 			csum = csum_add(csum, skb1->csum);
373 		}
374 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
375 						 (char *)icmph,
376 						 icmp_param->head_len, csum);
377 		icmph->checksum = csum_fold(csum);
378 		skb->ip_summed = CHECKSUM_NONE;
379 		ip_push_pending_frames(sk, fl4);
380 	}
381 }
382 
383 /*
384  *	Driving logic for building and sending ICMP messages.
385  */
386 
387 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
388 {
389 	struct ipcm_cookie ipc;
390 	struct rtable *rt = skb_rtable(skb);
391 	struct net *net = dev_net(rt->dst.dev);
392 	struct flowi4 fl4;
393 	struct sock *sk;
394 	struct inet_sock *inet;
395 	__be32 daddr, saddr;
396 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
397 
398 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
399 		return;
400 
401 	sk = icmp_xmit_lock(net);
402 	if (!sk)
403 		return;
404 	inet = inet_sk(sk);
405 
406 	icmp_param->data.icmph.checksum = 0;
407 
408 	inet->tos = ip_hdr(skb)->tos;
409 	sk->sk_mark = mark;
410 	daddr = ipc.addr = ip_hdr(skb)->saddr;
411 	saddr = fib_compute_spec_dst(skb);
412 	ipc.opt = NULL;
413 	ipc.tx_flags = 0;
414 	ipc.ttl = 0;
415 	ipc.tos = -1;
416 
417 	if (icmp_param->replyopts.opt.opt.optlen) {
418 		ipc.opt = &icmp_param->replyopts.opt;
419 		if (ipc.opt->opt.srr)
420 			daddr = icmp_param->replyopts.opt.opt.faddr;
421 	}
422 	memset(&fl4, 0, sizeof(fl4));
423 	fl4.daddr = daddr;
424 	fl4.saddr = saddr;
425 	fl4.flowi4_mark = mark;
426 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
427 	fl4.flowi4_proto = IPPROTO_ICMP;
428 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
429 	rt = ip_route_output_key(net, &fl4);
430 	if (IS_ERR(rt))
431 		goto out_unlock;
432 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
433 			       icmp_param->data.icmph.code))
434 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
435 	ip_rt_put(rt);
436 out_unlock:
437 	icmp_xmit_unlock(sk);
438 }
439 
440 static struct rtable *icmp_route_lookup(struct net *net,
441 					struct flowi4 *fl4,
442 					struct sk_buff *skb_in,
443 					const struct iphdr *iph,
444 					__be32 saddr, u8 tos, u32 mark,
445 					int type, int code,
446 					struct icmp_bxm *param)
447 {
448 	struct rtable *rt, *rt2;
449 	struct flowi4 fl4_dec;
450 	int err;
451 
452 	memset(fl4, 0, sizeof(*fl4));
453 	fl4->daddr = (param->replyopts.opt.opt.srr ?
454 		      param->replyopts.opt.opt.faddr : iph->saddr);
455 	fl4->saddr = saddr;
456 	fl4->flowi4_mark = mark;
457 	fl4->flowi4_tos = RT_TOS(tos);
458 	fl4->flowi4_proto = IPPROTO_ICMP;
459 	fl4->fl4_icmp_type = type;
460 	fl4->fl4_icmp_code = code;
461 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
462 	rt = __ip_route_output_key(net, fl4);
463 	if (IS_ERR(rt))
464 		return rt;
465 
466 	/* No need to clone since we're just using its address. */
467 	rt2 = rt;
468 
469 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
470 					   flowi4_to_flowi(fl4), NULL, 0);
471 	if (!IS_ERR(rt)) {
472 		if (rt != rt2)
473 			return rt;
474 	} else if (PTR_ERR(rt) == -EPERM) {
475 		rt = NULL;
476 	} else
477 		return rt;
478 
479 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
480 	if (err)
481 		goto relookup_failed;
482 
483 	if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
484 		rt2 = __ip_route_output_key(net, &fl4_dec);
485 		if (IS_ERR(rt2))
486 			err = PTR_ERR(rt2);
487 	} else {
488 		struct flowi4 fl4_2 = {};
489 		unsigned long orefdst;
490 
491 		fl4_2.daddr = fl4_dec.saddr;
492 		rt2 = ip_route_output_key(net, &fl4_2);
493 		if (IS_ERR(rt2)) {
494 			err = PTR_ERR(rt2);
495 			goto relookup_failed;
496 		}
497 		/* Ugh! */
498 		orefdst = skb_in->_skb_refdst; /* save old refdst */
499 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
500 				     RT_TOS(tos), rt2->dst.dev);
501 
502 		dst_release(&rt2->dst);
503 		rt2 = skb_rtable(skb_in);
504 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
505 	}
506 
507 	if (err)
508 		goto relookup_failed;
509 
510 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
511 					    flowi4_to_flowi(&fl4_dec), NULL,
512 					    XFRM_LOOKUP_ICMP);
513 	if (!IS_ERR(rt2)) {
514 		dst_release(&rt->dst);
515 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
516 		rt = rt2;
517 	} else if (PTR_ERR(rt2) == -EPERM) {
518 		if (rt)
519 			dst_release(&rt->dst);
520 		return rt2;
521 	} else {
522 		err = PTR_ERR(rt2);
523 		goto relookup_failed;
524 	}
525 	return rt;
526 
527 relookup_failed:
528 	if (rt)
529 		return rt;
530 	return ERR_PTR(err);
531 }
532 
533 /*
534  *	Send an ICMP message in response to a situation
535  *
536  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
537  *		  MAY send more (we do).
538  *			MUST NOT change this header information.
539  *			MUST NOT reply to a multicast/broadcast IP address.
540  *			MUST NOT reply to a multicast/broadcast MAC address.
541  *			MUST reply to only the first fragment.
542  */
543 
544 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
545 {
546 	struct iphdr *iph;
547 	int room;
548 	struct icmp_bxm *icmp_param;
549 	struct rtable *rt = skb_rtable(skb_in);
550 	struct ipcm_cookie ipc;
551 	struct flowi4 fl4;
552 	__be32 saddr;
553 	u8  tos;
554 	u32 mark;
555 	struct net *net;
556 	struct sock *sk;
557 
558 	if (!rt)
559 		goto out;
560 	net = dev_net(rt->dst.dev);
561 
562 	/*
563 	 *	Find the original header. It is expected to be valid, of course.
564 	 *	Check this, icmp_send is called from the most obscure devices
565 	 *	sometimes.
566 	 */
567 	iph = ip_hdr(skb_in);
568 
569 	if ((u8 *)iph < skb_in->head ||
570 	    (skb_network_header(skb_in) + sizeof(*iph)) >
571 	    skb_tail_pointer(skb_in))
572 		goto out;
573 
574 	/*
575 	 *	No replies to physical multicast/broadcast
576 	 */
577 	if (skb_in->pkt_type != PACKET_HOST)
578 		goto out;
579 
580 	/*
581 	 *	Now check at the protocol level
582 	 */
583 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
584 		goto out;
585 
586 	/*
587 	 *	Only reply to fragment 0. We byte re-order the constant
588 	 *	mask for efficiency.
589 	 */
590 	if (iph->frag_off & htons(IP_OFFSET))
591 		goto out;
592 
593 	/*
594 	 *	If we send an ICMP error to an ICMP error a mess would result..
595 	 */
596 	if (icmp_pointers[type].error) {
597 		/*
598 		 *	We are an error, check if we are replying to an
599 		 *	ICMP error
600 		 */
601 		if (iph->protocol == IPPROTO_ICMP) {
602 			u8 _inner_type, *itp;
603 
604 			itp = skb_header_pointer(skb_in,
605 						 skb_network_header(skb_in) +
606 						 (iph->ihl << 2) +
607 						 offsetof(struct icmphdr,
608 							  type) -
609 						 skb_in->data,
610 						 sizeof(_inner_type),
611 						 &_inner_type);
612 			if (!itp)
613 				goto out;
614 
615 			/*
616 			 *	Assume any unknown ICMP type is an error. This
617 			 *	isn't specified by the RFC, but think about it..
618 			 */
619 			if (*itp > NR_ICMP_TYPES ||
620 			    icmp_pointers[*itp].error)
621 				goto out;
622 		}
623 	}
624 
625 	icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
626 	if (!icmp_param)
627 		return;
628 
629 	sk = icmp_xmit_lock(net);
630 	if (!sk)
631 		goto out_free;
632 
633 	/*
634 	 *	Construct source address and options.
635 	 */
636 
637 	saddr = iph->daddr;
638 	if (!(rt->rt_flags & RTCF_LOCAL)) {
639 		struct net_device *dev = NULL;
640 
641 		rcu_read_lock();
642 		if (rt_is_input_route(rt) &&
643 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
644 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
645 
646 		if (dev)
647 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
648 		else
649 			saddr = 0;
650 		rcu_read_unlock();
651 	}
652 
653 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
654 					   IPTOS_PREC_INTERNETCONTROL) :
655 					  iph->tos;
656 	mark = IP4_REPLY_MARK(net, skb_in->mark);
657 
658 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
659 		goto out_unlock;
660 
661 
662 	/*
663 	 *	Prepare data for ICMP header.
664 	 */
665 
666 	icmp_param->data.icmph.type	 = type;
667 	icmp_param->data.icmph.code	 = code;
668 	icmp_param->data.icmph.un.gateway = info;
669 	icmp_param->data.icmph.checksum	 = 0;
670 	icmp_param->skb	  = skb_in;
671 	icmp_param->offset = skb_network_offset(skb_in);
672 	inet_sk(sk)->tos = tos;
673 	sk->sk_mark = mark;
674 	ipc.addr = iph->saddr;
675 	ipc.opt = &icmp_param->replyopts.opt;
676 	ipc.tx_flags = 0;
677 	ipc.ttl = 0;
678 	ipc.tos = -1;
679 
680 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
681 			       type, code, icmp_param);
682 	if (IS_ERR(rt))
683 		goto out_unlock;
684 
685 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
686 		goto ende;
687 
688 	/* RFC says return as much as we can without exceeding 576 bytes. */
689 
690 	room = dst_mtu(&rt->dst);
691 	if (room > 576)
692 		room = 576;
693 	room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
694 	room -= sizeof(struct icmphdr);
695 
696 	icmp_param->data_len = skb_in->len - icmp_param->offset;
697 	if (icmp_param->data_len > room)
698 		icmp_param->data_len = room;
699 	icmp_param->head_len = sizeof(struct icmphdr);
700 
701 	icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
702 ende:
703 	ip_rt_put(rt);
704 out_unlock:
705 	icmp_xmit_unlock(sk);
706 out_free:
707 	kfree(icmp_param);
708 out:;
709 }
710 EXPORT_SYMBOL(icmp_send);
711 
712 
713 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
714 {
715 	const struct iphdr *iph = (const struct iphdr *) skb->data;
716 	const struct net_protocol *ipprot;
717 	int protocol = iph->protocol;
718 
719 	/* Checkin full IP header plus 8 bytes of protocol to
720 	 * avoid additional coding at protocol handlers.
721 	 */
722 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
723 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
724 		return;
725 	}
726 
727 	raw_icmp_error(skb, protocol, info);
728 
729 	ipprot = rcu_dereference(inet_protos[protocol]);
730 	if (ipprot && ipprot->err_handler)
731 		ipprot->err_handler(skb, info);
732 }
733 
734 static bool icmp_tag_validation(int proto)
735 {
736 	bool ok;
737 
738 	rcu_read_lock();
739 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
740 	rcu_read_unlock();
741 	return ok;
742 }
743 
744 /*
745  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
746  *	ICMP_PARAMETERPROB.
747  */
748 
749 static bool icmp_unreach(struct sk_buff *skb)
750 {
751 	const struct iphdr *iph;
752 	struct icmphdr *icmph;
753 	struct net *net;
754 	u32 info = 0;
755 
756 	net = dev_net(skb_dst(skb)->dev);
757 
758 	/*
759 	 *	Incomplete header ?
760 	 * 	Only checks for the IP header, there should be an
761 	 *	additional check for longer headers in upper levels.
762 	 */
763 
764 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
765 		goto out_err;
766 
767 	icmph = icmp_hdr(skb);
768 	iph   = (const struct iphdr *)skb->data;
769 
770 	if (iph->ihl < 5) /* Mangled header, drop. */
771 		goto out_err;
772 
773 	if (icmph->type == ICMP_DEST_UNREACH) {
774 		switch (icmph->code & 15) {
775 		case ICMP_NET_UNREACH:
776 		case ICMP_HOST_UNREACH:
777 		case ICMP_PROT_UNREACH:
778 		case ICMP_PORT_UNREACH:
779 			break;
780 		case ICMP_FRAG_NEEDED:
781 			/* for documentation of the ip_no_pmtu_disc
782 			 * values please see
783 			 * Documentation/networking/ip-sysctl.txt
784 			 */
785 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
786 			default:
787 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
788 						    &iph->daddr);
789 				break;
790 			case 2:
791 				goto out;
792 			case 3:
793 				if (!icmp_tag_validation(iph->protocol))
794 					goto out;
795 				/* fall through */
796 			case 0:
797 				info = ntohs(icmph->un.frag.mtu);
798 			}
799 			break;
800 		case ICMP_SR_FAILED:
801 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
802 					    &iph->daddr);
803 			break;
804 		default:
805 			break;
806 		}
807 		if (icmph->code > NR_ICMP_UNREACH)
808 			goto out;
809 	} else if (icmph->type == ICMP_PARAMETERPROB)
810 		info = ntohl(icmph->un.gateway) >> 24;
811 
812 	/*
813 	 *	Throw it at our lower layers
814 	 *
815 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
816 	 *		  header.
817 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
818 	 *		  transport layer.
819 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
820 	 *		  transport layer.
821 	 */
822 
823 	/*
824 	 *	Check the other end isn't violating RFC 1122. Some routers send
825 	 *	bogus responses to broadcast frames. If you see this message
826 	 *	first check your netmask matches at both ends, if it does then
827 	 *	get the other vendor to fix their kit.
828 	 */
829 
830 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
831 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
832 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
833 				     &ip_hdr(skb)->saddr,
834 				     icmph->type, icmph->code,
835 				     &iph->daddr, skb->dev->name);
836 		goto out;
837 	}
838 
839 	icmp_socket_deliver(skb, info);
840 
841 out:
842 	return true;
843 out_err:
844 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
845 	return false;
846 }
847 
848 
849 /*
850  *	Handle ICMP_REDIRECT.
851  */
852 
853 static bool icmp_redirect(struct sk_buff *skb)
854 {
855 	if (skb->len < sizeof(struct iphdr)) {
856 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
857 		return false;
858 	}
859 
860 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
861 		/* there aught to be a stat */
862 		return false;
863 	}
864 
865 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
866 	return true;
867 }
868 
869 /*
870  *	Handle ICMP_ECHO ("ping") requests.
871  *
872  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
873  *		  requests.
874  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
875  *		  included in the reply.
876  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
877  *		  echo requests, MUST have default=NOT.
878  *	See also WRT handling of options once they are done and working.
879  */
880 
881 static bool icmp_echo(struct sk_buff *skb)
882 {
883 	struct net *net;
884 
885 	net = dev_net(skb_dst(skb)->dev);
886 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
887 		struct icmp_bxm icmp_param;
888 
889 		icmp_param.data.icmph	   = *icmp_hdr(skb);
890 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
891 		icmp_param.skb		   = skb;
892 		icmp_param.offset	   = 0;
893 		icmp_param.data_len	   = skb->len;
894 		icmp_param.head_len	   = sizeof(struct icmphdr);
895 		icmp_reply(&icmp_param, skb);
896 	}
897 	/* should there be an ICMP stat for ignored echos? */
898 	return true;
899 }
900 
901 /*
902  *	Handle ICMP Timestamp requests.
903  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
904  *		  SHOULD be in the kernel for minimum random latency.
905  *		  MUST be accurate to a few minutes.
906  *		  MUST be updated at least at 15Hz.
907  */
908 static bool icmp_timestamp(struct sk_buff *skb)
909 {
910 	struct timespec tv;
911 	struct icmp_bxm icmp_param;
912 	/*
913 	 *	Too short.
914 	 */
915 	if (skb->len < 4)
916 		goto out_err;
917 
918 	/*
919 	 *	Fill in the current time as ms since midnight UT:
920 	 */
921 	getnstimeofday(&tv);
922 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
923 					 tv.tv_nsec / NSEC_PER_MSEC);
924 	icmp_param.data.times[2] = icmp_param.data.times[1];
925 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
926 		BUG();
927 	icmp_param.data.icmph	   = *icmp_hdr(skb);
928 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
929 	icmp_param.data.icmph.code = 0;
930 	icmp_param.skb		   = skb;
931 	icmp_param.offset	   = 0;
932 	icmp_param.data_len	   = 0;
933 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
934 	icmp_reply(&icmp_param, skb);
935 	return true;
936 
937 out_err:
938 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
939 	return false;
940 }
941 
942 static bool icmp_discard(struct sk_buff *skb)
943 {
944 	/* pretend it was a success */
945 	return true;
946 }
947 
948 /*
949  *	Deal with incoming ICMP packets.
950  */
951 int icmp_rcv(struct sk_buff *skb)
952 {
953 	struct icmphdr *icmph;
954 	struct rtable *rt = skb_rtable(skb);
955 	struct net *net = dev_net(rt->dst.dev);
956 	bool success;
957 
958 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
959 		struct sec_path *sp = skb_sec_path(skb);
960 		int nh;
961 
962 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
963 				 XFRM_STATE_ICMP))
964 			goto drop;
965 
966 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
967 			goto drop;
968 
969 		nh = skb_network_offset(skb);
970 		skb_set_network_header(skb, sizeof(*icmph));
971 
972 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
973 			goto drop;
974 
975 		skb_set_network_header(skb, nh);
976 	}
977 
978 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
979 
980 	if (skb_checksum_simple_validate(skb))
981 		goto csum_error;
982 
983 	if (!pskb_pull(skb, sizeof(*icmph)))
984 		goto error;
985 
986 	icmph = icmp_hdr(skb);
987 
988 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
989 	/*
990 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
991 	 *
992 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
993 	 *		  discarded.
994 	 */
995 	if (icmph->type > NR_ICMP_TYPES)
996 		goto error;
997 
998 
999 	/*
1000 	 *	Parse the ICMP message
1001 	 */
1002 
1003 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1004 		/*
1005 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1006 		 *	  silently ignored (we let user decide with a sysctl).
1007 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1008 		 *	  discarded if to broadcast/multicast.
1009 		 */
1010 		if ((icmph->type == ICMP_ECHO ||
1011 		     icmph->type == ICMP_TIMESTAMP) &&
1012 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1013 			goto error;
1014 		}
1015 		if (icmph->type != ICMP_ECHO &&
1016 		    icmph->type != ICMP_TIMESTAMP &&
1017 		    icmph->type != ICMP_ADDRESS &&
1018 		    icmph->type != ICMP_ADDRESSREPLY) {
1019 			goto error;
1020 		}
1021 	}
1022 
1023 	success = icmp_pointers[icmph->type].handler(skb);
1024 
1025 	if (success)  {
1026 		consume_skb(skb);
1027 		return 0;
1028 	}
1029 
1030 drop:
1031 	kfree_skb(skb);
1032 	return 0;
1033 csum_error:
1034 	ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
1035 error:
1036 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1037 	goto drop;
1038 }
1039 
1040 void icmp_err(struct sk_buff *skb, u32 info)
1041 {
1042 	struct iphdr *iph = (struct iphdr *)skb->data;
1043 	int offset = iph->ihl<<2;
1044 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1045 	int type = icmp_hdr(skb)->type;
1046 	int code = icmp_hdr(skb)->code;
1047 	struct net *net = dev_net(skb->dev);
1048 
1049 	/*
1050 	 * Use ping_err to handle all icmp errors except those
1051 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1052 	 */
1053 	if (icmph->type != ICMP_ECHOREPLY) {
1054 		ping_err(skb, offset, info);
1055 		return;
1056 	}
1057 
1058 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1059 		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1060 	else if (type == ICMP_REDIRECT)
1061 		ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1062 }
1063 
1064 /*
1065  *	This table is the definition of how we handle ICMP.
1066  */
1067 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1068 	[ICMP_ECHOREPLY] = {
1069 		.handler = ping_rcv,
1070 	},
1071 	[1] = {
1072 		.handler = icmp_discard,
1073 		.error = 1,
1074 	},
1075 	[2] = {
1076 		.handler = icmp_discard,
1077 		.error = 1,
1078 	},
1079 	[ICMP_DEST_UNREACH] = {
1080 		.handler = icmp_unreach,
1081 		.error = 1,
1082 	},
1083 	[ICMP_SOURCE_QUENCH] = {
1084 		.handler = icmp_unreach,
1085 		.error = 1,
1086 	},
1087 	[ICMP_REDIRECT] = {
1088 		.handler = icmp_redirect,
1089 		.error = 1,
1090 	},
1091 	[6] = {
1092 		.handler = icmp_discard,
1093 		.error = 1,
1094 	},
1095 	[7] = {
1096 		.handler = icmp_discard,
1097 		.error = 1,
1098 	},
1099 	[ICMP_ECHO] = {
1100 		.handler = icmp_echo,
1101 	},
1102 	[9] = {
1103 		.handler = icmp_discard,
1104 		.error = 1,
1105 	},
1106 	[10] = {
1107 		.handler = icmp_discard,
1108 		.error = 1,
1109 	},
1110 	[ICMP_TIME_EXCEEDED] = {
1111 		.handler = icmp_unreach,
1112 		.error = 1,
1113 	},
1114 	[ICMP_PARAMETERPROB] = {
1115 		.handler = icmp_unreach,
1116 		.error = 1,
1117 	},
1118 	[ICMP_TIMESTAMP] = {
1119 		.handler = icmp_timestamp,
1120 	},
1121 	[ICMP_TIMESTAMPREPLY] = {
1122 		.handler = icmp_discard,
1123 	},
1124 	[ICMP_INFO_REQUEST] = {
1125 		.handler = icmp_discard,
1126 	},
1127 	[ICMP_INFO_REPLY] = {
1128 		.handler = icmp_discard,
1129 	},
1130 	[ICMP_ADDRESS] = {
1131 		.handler = icmp_discard,
1132 	},
1133 	[ICMP_ADDRESSREPLY] = {
1134 		.handler = icmp_discard,
1135 	},
1136 };
1137 
1138 static void __net_exit icmp_sk_exit(struct net *net)
1139 {
1140 	int i;
1141 
1142 	for_each_possible_cpu(i)
1143 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1144 	free_percpu(net->ipv4.icmp_sk);
1145 	net->ipv4.icmp_sk = NULL;
1146 }
1147 
1148 static int __net_init icmp_sk_init(struct net *net)
1149 {
1150 	int i, err;
1151 
1152 	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1153 	if (!net->ipv4.icmp_sk)
1154 		return -ENOMEM;
1155 
1156 	for_each_possible_cpu(i) {
1157 		struct sock *sk;
1158 
1159 		err = inet_ctl_sock_create(&sk, PF_INET,
1160 					   SOCK_RAW, IPPROTO_ICMP, net);
1161 		if (err < 0)
1162 			goto fail;
1163 
1164 		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1165 
1166 		/* Enough space for 2 64K ICMP packets, including
1167 		 * sk_buff/skb_shared_info struct overhead.
1168 		 */
1169 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1170 
1171 		/*
1172 		 * Speedup sock_wfree()
1173 		 */
1174 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1175 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1176 	}
1177 
1178 	/* Control parameters for ECHO replies. */
1179 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1180 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1181 
1182 	/* Control parameter - ignore bogus broadcast responses? */
1183 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1184 
1185 	/*
1186 	 * 	Configurable global rate limit.
1187 	 *
1188 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1189 	 *	bucket ratemask defines which icmp types are ratelimited by
1190 	 *	setting	it's bit position.
1191 	 *
1192 	 *	default:
1193 	 *	dest unreachable (3), source quench (4),
1194 	 *	time exceeded (11), parameter problem (12)
1195 	 */
1196 
1197 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1198 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1199 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1200 
1201 	return 0;
1202 
1203 fail:
1204 	for_each_possible_cpu(i)
1205 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1206 	free_percpu(net->ipv4.icmp_sk);
1207 	return err;
1208 }
1209 
1210 static struct pernet_operations __net_initdata icmp_sk_ops = {
1211        .init = icmp_sk_init,
1212        .exit = icmp_sk_exit,
1213 };
1214 
1215 int __init icmp_init(void)
1216 {
1217 	return register_pernet_subsys(&icmp_sk_ops);
1218 }
1219