xref: /linux/net/ipv4/icmp.c (revision 02000b55850deeadffe433e4b4930a8831f477de)
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 <linux/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/l3mdev.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 /* Called with BH disabled */
213 static inline struct sock *icmp_xmit_lock(struct net *net)
214 {
215 	struct sock *sk;
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 		return NULL;
224 	}
225 	return sk;
226 }
227 
228 static inline void icmp_xmit_unlock(struct sock *sk)
229 {
230 	spin_unlock(&sk->sk_lock.slock);
231 }
232 
233 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
234 int sysctl_icmp_msgs_burst __read_mostly = 50;
235 
236 static struct {
237 	spinlock_t	lock;
238 	u32		credit;
239 	u32		stamp;
240 } icmp_global = {
241 	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
242 };
243 
244 /**
245  * icmp_global_allow - Are we allowed to send one more ICMP message ?
246  *
247  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
248  * Returns false if we reached the limit and can not send another packet.
249  * Note: called with BH disabled
250  */
251 bool icmp_global_allow(void)
252 {
253 	u32 credit, delta, incr = 0, now = (u32)jiffies;
254 	bool rc = false;
255 
256 	/* Check if token bucket is empty and cannot be refilled
257 	 * without taking the spinlock.
258 	 */
259 	if (!icmp_global.credit) {
260 		delta = min_t(u32, now - icmp_global.stamp, HZ);
261 		if (delta < HZ / 50)
262 			return false;
263 	}
264 
265 	spin_lock(&icmp_global.lock);
266 	delta = min_t(u32, now - icmp_global.stamp, HZ);
267 	if (delta >= HZ / 50) {
268 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
269 		if (incr)
270 			icmp_global.stamp = now;
271 	}
272 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
273 	if (credit) {
274 		credit--;
275 		rc = true;
276 	}
277 	icmp_global.credit = credit;
278 	spin_unlock(&icmp_global.lock);
279 	return rc;
280 }
281 EXPORT_SYMBOL(icmp_global_allow);
282 
283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
284 {
285 	if (type > NR_ICMP_TYPES)
286 		return true;
287 
288 	/* Don't limit PMTU discovery. */
289 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
290 		return true;
291 
292 	/* Limit if icmp type is enabled in ratemask. */
293 	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
294 		return true;
295 
296 	return false;
297 }
298 
299 static bool icmpv4_global_allow(struct net *net, int type, int code)
300 {
301 	if (icmpv4_mask_allow(net, type, code))
302 		return true;
303 
304 	if (icmp_global_allow())
305 		return true;
306 
307 	return false;
308 }
309 
310 /*
311  *	Send an ICMP frame.
312  */
313 
314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
315 			       struct flowi4 *fl4, int type, int code)
316 {
317 	struct dst_entry *dst = &rt->dst;
318 	struct inet_peer *peer;
319 	bool rc = true;
320 	int vif;
321 
322 	if (icmpv4_mask_allow(net, type, code))
323 		goto out;
324 
325 	/* No rate limit on loopback */
326 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
327 		goto out;
328 
329 	vif = l3mdev_master_ifindex(dst->dev);
330 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
331 	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
332 	if (peer)
333 		inet_putpeer(peer);
334 out:
335 	return rc;
336 }
337 
338 /*
339  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
340  */
341 void icmp_out_count(struct net *net, unsigned char type)
342 {
343 	ICMPMSGOUT_INC_STATS(net, type);
344 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
345 }
346 
347 /*
348  *	Checksum each fragment, and on the first include the headers and final
349  *	checksum.
350  */
351 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
352 			  struct sk_buff *skb)
353 {
354 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
355 	__wsum csum;
356 
357 	csum = skb_copy_and_csum_bits(icmp_param->skb,
358 				      icmp_param->offset + offset,
359 				      to, len, 0);
360 
361 	skb->csum = csum_block_add(skb->csum, csum, odd);
362 	if (icmp_pointers[icmp_param->data.icmph.type].error)
363 		nf_ct_attach(skb, icmp_param->skb);
364 	return 0;
365 }
366 
367 static void icmp_push_reply(struct icmp_bxm *icmp_param,
368 			    struct flowi4 *fl4,
369 			    struct ipcm_cookie *ipc, struct rtable **rt)
370 {
371 	struct sock *sk;
372 	struct sk_buff *skb;
373 
374 	sk = icmp_sk(dev_net((*rt)->dst.dev));
375 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
376 			   icmp_param->data_len+icmp_param->head_len,
377 			   icmp_param->head_len,
378 			   ipc, rt, MSG_DONTWAIT) < 0) {
379 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
380 		ip_flush_pending_frames(sk);
381 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
382 		struct icmphdr *icmph = icmp_hdr(skb);
383 		__wsum csum = 0;
384 		struct sk_buff *skb1;
385 
386 		skb_queue_walk(&sk->sk_write_queue, skb1) {
387 			csum = csum_add(csum, skb1->csum);
388 		}
389 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
390 						 (char *)icmph,
391 						 icmp_param->head_len, csum);
392 		icmph->checksum = csum_fold(csum);
393 		skb->ip_summed = CHECKSUM_NONE;
394 		ip_push_pending_frames(sk, fl4);
395 	}
396 }
397 
398 /*
399  *	Driving logic for building and sending ICMP messages.
400  */
401 
402 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
403 {
404 	struct ipcm_cookie ipc;
405 	struct rtable *rt = skb_rtable(skb);
406 	struct net *net = dev_net(rt->dst.dev);
407 	struct flowi4 fl4;
408 	struct sock *sk;
409 	struct inet_sock *inet;
410 	__be32 daddr, saddr;
411 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
412 	int type = icmp_param->data.icmph.type;
413 	int code = icmp_param->data.icmph.code;
414 
415 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
416 		return;
417 
418 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
419 	local_bh_disable();
420 
421 	/* global icmp_msgs_per_sec */
422 	if (!icmpv4_global_allow(net, type, code))
423 		goto out_bh_enable;
424 
425 	sk = icmp_xmit_lock(net);
426 	if (!sk)
427 		goto out_bh_enable;
428 	inet = inet_sk(sk);
429 
430 	icmp_param->data.icmph.checksum = 0;
431 
432 	inet->tos = ip_hdr(skb)->tos;
433 	sk->sk_mark = mark;
434 	daddr = ipc.addr = ip_hdr(skb)->saddr;
435 	saddr = fib_compute_spec_dst(skb);
436 	ipc.opt = NULL;
437 	ipc.tx_flags = 0;
438 	ipc.ttl = 0;
439 	ipc.tos = -1;
440 	ipc.sockc.transmit_time = 0;
441 
442 	if (icmp_param->replyopts.opt.opt.optlen) {
443 		ipc.opt = &icmp_param->replyopts.opt;
444 		if (ipc.opt->opt.srr)
445 			daddr = icmp_param->replyopts.opt.opt.faddr;
446 	}
447 	memset(&fl4, 0, sizeof(fl4));
448 	fl4.daddr = daddr;
449 	fl4.saddr = saddr;
450 	fl4.flowi4_mark = mark;
451 	fl4.flowi4_uid = sock_net_uid(net, NULL);
452 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
453 	fl4.flowi4_proto = IPPROTO_ICMP;
454 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
455 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
456 	rt = ip_route_output_key(net, &fl4);
457 	if (IS_ERR(rt))
458 		goto out_unlock;
459 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
460 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
461 	ip_rt_put(rt);
462 out_unlock:
463 	icmp_xmit_unlock(sk);
464 out_bh_enable:
465 	local_bh_enable();
466 }
467 
468 static struct rtable *icmp_route_lookup(struct net *net,
469 					struct flowi4 *fl4,
470 					struct sk_buff *skb_in,
471 					const struct iphdr *iph,
472 					__be32 saddr, u8 tos, u32 mark,
473 					int type, int code,
474 					struct icmp_bxm *param)
475 {
476 	struct rtable *rt, *rt2;
477 	struct flowi4 fl4_dec;
478 	int err;
479 
480 	memset(fl4, 0, sizeof(*fl4));
481 	fl4->daddr = (param->replyopts.opt.opt.srr ?
482 		      param->replyopts.opt.opt.faddr : iph->saddr);
483 	fl4->saddr = saddr;
484 	fl4->flowi4_mark = mark;
485 	fl4->flowi4_uid = sock_net_uid(net, NULL);
486 	fl4->flowi4_tos = RT_TOS(tos);
487 	fl4->flowi4_proto = IPPROTO_ICMP;
488 	fl4->fl4_icmp_type = type;
489 	fl4->fl4_icmp_code = code;
490 	fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
491 
492 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
493 	rt = ip_route_output_key_hash(net, fl4, skb_in);
494 	if (IS_ERR(rt))
495 		return rt;
496 
497 	/* No need to clone since we're just using its address. */
498 	rt2 = rt;
499 
500 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
501 					   flowi4_to_flowi(fl4), NULL, 0);
502 	if (!IS_ERR(rt)) {
503 		if (rt != rt2)
504 			return rt;
505 	} else if (PTR_ERR(rt) == -EPERM) {
506 		rt = NULL;
507 	} else
508 		return rt;
509 
510 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
511 	if (err)
512 		goto relookup_failed;
513 
514 	if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
515 				     fl4_dec.saddr) == RTN_LOCAL) {
516 		rt2 = __ip_route_output_key(net, &fl4_dec);
517 		if (IS_ERR(rt2))
518 			err = PTR_ERR(rt2);
519 	} else {
520 		struct flowi4 fl4_2 = {};
521 		unsigned long orefdst;
522 
523 		fl4_2.daddr = fl4_dec.saddr;
524 		rt2 = ip_route_output_key(net, &fl4_2);
525 		if (IS_ERR(rt2)) {
526 			err = PTR_ERR(rt2);
527 			goto relookup_failed;
528 		}
529 		/* Ugh! */
530 		orefdst = skb_in->_skb_refdst; /* save old refdst */
531 		skb_dst_set(skb_in, NULL);
532 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
533 				     RT_TOS(tos), rt2->dst.dev);
534 
535 		dst_release(&rt2->dst);
536 		rt2 = skb_rtable(skb_in);
537 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
538 	}
539 
540 	if (err)
541 		goto relookup_failed;
542 
543 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
544 					    flowi4_to_flowi(&fl4_dec), NULL,
545 					    XFRM_LOOKUP_ICMP);
546 	if (!IS_ERR(rt2)) {
547 		dst_release(&rt->dst);
548 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
549 		rt = rt2;
550 	} else if (PTR_ERR(rt2) == -EPERM) {
551 		if (rt)
552 			dst_release(&rt->dst);
553 		return rt2;
554 	} else {
555 		err = PTR_ERR(rt2);
556 		goto relookup_failed;
557 	}
558 	return rt;
559 
560 relookup_failed:
561 	if (rt)
562 		return rt;
563 	return ERR_PTR(err);
564 }
565 
566 /*
567  *	Send an ICMP message in response to a situation
568  *
569  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
570  *		  MAY send more (we do).
571  *			MUST NOT change this header information.
572  *			MUST NOT reply to a multicast/broadcast IP address.
573  *			MUST NOT reply to a multicast/broadcast MAC address.
574  *			MUST reply to only the first fragment.
575  */
576 
577 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
578 {
579 	struct iphdr *iph;
580 	int room;
581 	struct icmp_bxm icmp_param;
582 	struct rtable *rt = skb_rtable(skb_in);
583 	struct ipcm_cookie ipc;
584 	struct flowi4 fl4;
585 	__be32 saddr;
586 	u8  tos;
587 	u32 mark;
588 	struct net *net;
589 	struct sock *sk;
590 
591 	if (!rt)
592 		goto out;
593 	net = dev_net(rt->dst.dev);
594 
595 	/*
596 	 *	Find the original header. It is expected to be valid, of course.
597 	 *	Check this, icmp_send is called from the most obscure devices
598 	 *	sometimes.
599 	 */
600 	iph = ip_hdr(skb_in);
601 
602 	if ((u8 *)iph < skb_in->head ||
603 	    (skb_network_header(skb_in) + sizeof(*iph)) >
604 	    skb_tail_pointer(skb_in))
605 		goto out;
606 
607 	/*
608 	 *	No replies to physical multicast/broadcast
609 	 */
610 	if (skb_in->pkt_type != PACKET_HOST)
611 		goto out;
612 
613 	/*
614 	 *	Now check at the protocol level
615 	 */
616 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
617 		goto out;
618 
619 	/*
620 	 *	Only reply to fragment 0. We byte re-order the constant
621 	 *	mask for efficiency.
622 	 */
623 	if (iph->frag_off & htons(IP_OFFSET))
624 		goto out;
625 
626 	/*
627 	 *	If we send an ICMP error to an ICMP error a mess would result..
628 	 */
629 	if (icmp_pointers[type].error) {
630 		/*
631 		 *	We are an error, check if we are replying to an
632 		 *	ICMP error
633 		 */
634 		if (iph->protocol == IPPROTO_ICMP) {
635 			u8 _inner_type, *itp;
636 
637 			itp = skb_header_pointer(skb_in,
638 						 skb_network_header(skb_in) +
639 						 (iph->ihl << 2) +
640 						 offsetof(struct icmphdr,
641 							  type) -
642 						 skb_in->data,
643 						 sizeof(_inner_type),
644 						 &_inner_type);
645 			if (!itp)
646 				goto out;
647 
648 			/*
649 			 *	Assume any unknown ICMP type is an error. This
650 			 *	isn't specified by the RFC, but think about it..
651 			 */
652 			if (*itp > NR_ICMP_TYPES ||
653 			    icmp_pointers[*itp].error)
654 				goto out;
655 		}
656 	}
657 
658 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
659 	local_bh_disable();
660 
661 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
662 	 * incoming dev is loopback.  If outgoing dev change to not be
663 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
664 	 */
665 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
666 	      !icmpv4_global_allow(net, type, code))
667 		goto out_bh_enable;
668 
669 	sk = icmp_xmit_lock(net);
670 	if (!sk)
671 		goto out_bh_enable;
672 
673 	/*
674 	 *	Construct source address and options.
675 	 */
676 
677 	saddr = iph->daddr;
678 	if (!(rt->rt_flags & RTCF_LOCAL)) {
679 		struct net_device *dev = NULL;
680 
681 		rcu_read_lock();
682 		if (rt_is_input_route(rt) &&
683 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
684 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
685 
686 		if (dev)
687 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
688 		else
689 			saddr = 0;
690 		rcu_read_unlock();
691 	}
692 
693 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
694 					   IPTOS_PREC_INTERNETCONTROL) :
695 					  iph->tos;
696 	mark = IP4_REPLY_MARK(net, skb_in->mark);
697 
698 	if (ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in))
699 		goto out_unlock;
700 
701 
702 	/*
703 	 *	Prepare data for ICMP header.
704 	 */
705 
706 	icmp_param.data.icmph.type	 = type;
707 	icmp_param.data.icmph.code	 = code;
708 	icmp_param.data.icmph.un.gateway = info;
709 	icmp_param.data.icmph.checksum	 = 0;
710 	icmp_param.skb	  = skb_in;
711 	icmp_param.offset = skb_network_offset(skb_in);
712 	inet_sk(sk)->tos = tos;
713 	sk->sk_mark = mark;
714 	ipc.addr = iph->saddr;
715 	ipc.opt = &icmp_param.replyopts.opt;
716 	ipc.tx_flags = 0;
717 	ipc.ttl = 0;
718 	ipc.tos = -1;
719 	ipc.sockc.transmit_time = 0;
720 
721 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
722 			       type, code, &icmp_param);
723 	if (IS_ERR(rt))
724 		goto out_unlock;
725 
726 	/* peer icmp_ratelimit */
727 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
728 		goto ende;
729 
730 	/* RFC says return as much as we can without exceeding 576 bytes. */
731 
732 	room = dst_mtu(&rt->dst);
733 	if (room > 576)
734 		room = 576;
735 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
736 	room -= sizeof(struct icmphdr);
737 
738 	icmp_param.data_len = skb_in->len - icmp_param.offset;
739 	if (icmp_param.data_len > room)
740 		icmp_param.data_len = room;
741 	icmp_param.head_len = sizeof(struct icmphdr);
742 
743 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
744 ende:
745 	ip_rt_put(rt);
746 out_unlock:
747 	icmp_xmit_unlock(sk);
748 out_bh_enable:
749 	local_bh_enable();
750 out:;
751 }
752 EXPORT_SYMBOL(icmp_send);
753 
754 
755 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
756 {
757 	const struct iphdr *iph = (const struct iphdr *) skb->data;
758 	const struct net_protocol *ipprot;
759 	int protocol = iph->protocol;
760 
761 	/* Checkin full IP header plus 8 bytes of protocol to
762 	 * avoid additional coding at protocol handlers.
763 	 */
764 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
765 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
766 		return;
767 	}
768 
769 	raw_icmp_error(skb, protocol, info);
770 
771 	ipprot = rcu_dereference(inet_protos[protocol]);
772 	if (ipprot && ipprot->err_handler)
773 		ipprot->err_handler(skb, info);
774 }
775 
776 static bool icmp_tag_validation(int proto)
777 {
778 	bool ok;
779 
780 	rcu_read_lock();
781 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
782 	rcu_read_unlock();
783 	return ok;
784 }
785 
786 /*
787  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
788  *	ICMP_PARAMETERPROB.
789  */
790 
791 static bool icmp_unreach(struct sk_buff *skb)
792 {
793 	const struct iphdr *iph;
794 	struct icmphdr *icmph;
795 	struct net *net;
796 	u32 info = 0;
797 
798 	net = dev_net(skb_dst(skb)->dev);
799 
800 	/*
801 	 *	Incomplete header ?
802 	 * 	Only checks for the IP header, there should be an
803 	 *	additional check for longer headers in upper levels.
804 	 */
805 
806 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
807 		goto out_err;
808 
809 	icmph = icmp_hdr(skb);
810 	iph   = (const struct iphdr *)skb->data;
811 
812 	if (iph->ihl < 5) /* Mangled header, drop. */
813 		goto out_err;
814 
815 	switch (icmph->type) {
816 	case ICMP_DEST_UNREACH:
817 		switch (icmph->code & 15) {
818 		case ICMP_NET_UNREACH:
819 		case ICMP_HOST_UNREACH:
820 		case ICMP_PROT_UNREACH:
821 		case ICMP_PORT_UNREACH:
822 			break;
823 		case ICMP_FRAG_NEEDED:
824 			/* for documentation of the ip_no_pmtu_disc
825 			 * values please see
826 			 * Documentation/networking/ip-sysctl.txt
827 			 */
828 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
829 			default:
830 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
831 						    &iph->daddr);
832 				break;
833 			case 2:
834 				goto out;
835 			case 3:
836 				if (!icmp_tag_validation(iph->protocol))
837 					goto out;
838 				/* fall through */
839 			case 0:
840 				info = ntohs(icmph->un.frag.mtu);
841 			}
842 			break;
843 		case ICMP_SR_FAILED:
844 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
845 					    &iph->daddr);
846 			break;
847 		default:
848 			break;
849 		}
850 		if (icmph->code > NR_ICMP_UNREACH)
851 			goto out;
852 		break;
853 	case ICMP_PARAMETERPROB:
854 		info = ntohl(icmph->un.gateway) >> 24;
855 		break;
856 	case ICMP_TIME_EXCEEDED:
857 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
858 		if (icmph->code == ICMP_EXC_FRAGTIME)
859 			goto out;
860 		break;
861 	}
862 
863 	/*
864 	 *	Throw it at our lower layers
865 	 *
866 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
867 	 *		  header.
868 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
869 	 *		  transport layer.
870 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
871 	 *		  transport layer.
872 	 */
873 
874 	/*
875 	 *	Check the other end isn't violating RFC 1122. Some routers send
876 	 *	bogus responses to broadcast frames. If you see this message
877 	 *	first check your netmask matches at both ends, if it does then
878 	 *	get the other vendor to fix their kit.
879 	 */
880 
881 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
882 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
883 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
884 				     &ip_hdr(skb)->saddr,
885 				     icmph->type, icmph->code,
886 				     &iph->daddr, skb->dev->name);
887 		goto out;
888 	}
889 
890 	icmp_socket_deliver(skb, info);
891 
892 out:
893 	return true;
894 out_err:
895 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
896 	return false;
897 }
898 
899 
900 /*
901  *	Handle ICMP_REDIRECT.
902  */
903 
904 static bool icmp_redirect(struct sk_buff *skb)
905 {
906 	if (skb->len < sizeof(struct iphdr)) {
907 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
908 		return false;
909 	}
910 
911 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
912 		/* there aught to be a stat */
913 		return false;
914 	}
915 
916 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
917 	return true;
918 }
919 
920 /*
921  *	Handle ICMP_ECHO ("ping") requests.
922  *
923  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
924  *		  requests.
925  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
926  *		  included in the reply.
927  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
928  *		  echo requests, MUST have default=NOT.
929  *	See also WRT handling of options once they are done and working.
930  */
931 
932 static bool icmp_echo(struct sk_buff *skb)
933 {
934 	struct net *net;
935 
936 	net = dev_net(skb_dst(skb)->dev);
937 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
938 		struct icmp_bxm icmp_param;
939 
940 		icmp_param.data.icmph	   = *icmp_hdr(skb);
941 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
942 		icmp_param.skb		   = skb;
943 		icmp_param.offset	   = 0;
944 		icmp_param.data_len	   = skb->len;
945 		icmp_param.head_len	   = sizeof(struct icmphdr);
946 		icmp_reply(&icmp_param, skb);
947 	}
948 	/* should there be an ICMP stat for ignored echos? */
949 	return true;
950 }
951 
952 /*
953  *	Handle ICMP Timestamp requests.
954  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
955  *		  SHOULD be in the kernel for minimum random latency.
956  *		  MUST be accurate to a few minutes.
957  *		  MUST be updated at least at 15Hz.
958  */
959 static bool icmp_timestamp(struct sk_buff *skb)
960 {
961 	struct icmp_bxm icmp_param;
962 	/*
963 	 *	Too short.
964 	 */
965 	if (skb->len < 4)
966 		goto out_err;
967 
968 	/*
969 	 *	Fill in the current time as ms since midnight UT:
970 	 */
971 	icmp_param.data.times[1] = inet_current_timestamp();
972 	icmp_param.data.times[2] = icmp_param.data.times[1];
973 
974 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
975 
976 	icmp_param.data.icmph	   = *icmp_hdr(skb);
977 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
978 	icmp_param.data.icmph.code = 0;
979 	icmp_param.skb		   = skb;
980 	icmp_param.offset	   = 0;
981 	icmp_param.data_len	   = 0;
982 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
983 	icmp_reply(&icmp_param, skb);
984 	return true;
985 
986 out_err:
987 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
988 	return false;
989 }
990 
991 static bool icmp_discard(struct sk_buff *skb)
992 {
993 	/* pretend it was a success */
994 	return true;
995 }
996 
997 /*
998  *	Deal with incoming ICMP packets.
999  */
1000 int icmp_rcv(struct sk_buff *skb)
1001 {
1002 	struct icmphdr *icmph;
1003 	struct rtable *rt = skb_rtable(skb);
1004 	struct net *net = dev_net(rt->dst.dev);
1005 	bool success;
1006 
1007 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1008 		struct sec_path *sp = skb_sec_path(skb);
1009 		int nh;
1010 
1011 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1012 				 XFRM_STATE_ICMP))
1013 			goto drop;
1014 
1015 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1016 			goto drop;
1017 
1018 		nh = skb_network_offset(skb);
1019 		skb_set_network_header(skb, sizeof(*icmph));
1020 
1021 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1022 			goto drop;
1023 
1024 		skb_set_network_header(skb, nh);
1025 	}
1026 
1027 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1028 
1029 	if (skb_checksum_simple_validate(skb))
1030 		goto csum_error;
1031 
1032 	if (!pskb_pull(skb, sizeof(*icmph)))
1033 		goto error;
1034 
1035 	icmph = icmp_hdr(skb);
1036 
1037 	ICMPMSGIN_INC_STATS(net, icmph->type);
1038 	/*
1039 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1040 	 *
1041 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1042 	 *		  discarded.
1043 	 */
1044 	if (icmph->type > NR_ICMP_TYPES)
1045 		goto error;
1046 
1047 
1048 	/*
1049 	 *	Parse the ICMP message
1050 	 */
1051 
1052 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1053 		/*
1054 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1055 		 *	  silently ignored (we let user decide with a sysctl).
1056 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1057 		 *	  discarded if to broadcast/multicast.
1058 		 */
1059 		if ((icmph->type == ICMP_ECHO ||
1060 		     icmph->type == ICMP_TIMESTAMP) &&
1061 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1062 			goto error;
1063 		}
1064 		if (icmph->type != ICMP_ECHO &&
1065 		    icmph->type != ICMP_TIMESTAMP &&
1066 		    icmph->type != ICMP_ADDRESS &&
1067 		    icmph->type != ICMP_ADDRESSREPLY) {
1068 			goto error;
1069 		}
1070 	}
1071 
1072 	success = icmp_pointers[icmph->type].handler(skb);
1073 
1074 	if (success)  {
1075 		consume_skb(skb);
1076 		return NET_RX_SUCCESS;
1077 	}
1078 
1079 drop:
1080 	kfree_skb(skb);
1081 	return NET_RX_DROP;
1082 csum_error:
1083 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1084 error:
1085 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1086 	goto drop;
1087 }
1088 
1089 void icmp_err(struct sk_buff *skb, u32 info)
1090 {
1091 	struct iphdr *iph = (struct iphdr *)skb->data;
1092 	int offset = iph->ihl<<2;
1093 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1094 	int type = icmp_hdr(skb)->type;
1095 	int code = icmp_hdr(skb)->code;
1096 	struct net *net = dev_net(skb->dev);
1097 
1098 	/*
1099 	 * Use ping_err to handle all icmp errors except those
1100 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1101 	 */
1102 	if (icmph->type != ICMP_ECHOREPLY) {
1103 		ping_err(skb, offset, info);
1104 		return;
1105 	}
1106 
1107 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1108 		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1109 	else if (type == ICMP_REDIRECT)
1110 		ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1111 }
1112 
1113 /*
1114  *	This table is the definition of how we handle ICMP.
1115  */
1116 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1117 	[ICMP_ECHOREPLY] = {
1118 		.handler = ping_rcv,
1119 	},
1120 	[1] = {
1121 		.handler = icmp_discard,
1122 		.error = 1,
1123 	},
1124 	[2] = {
1125 		.handler = icmp_discard,
1126 		.error = 1,
1127 	},
1128 	[ICMP_DEST_UNREACH] = {
1129 		.handler = icmp_unreach,
1130 		.error = 1,
1131 	},
1132 	[ICMP_SOURCE_QUENCH] = {
1133 		.handler = icmp_unreach,
1134 		.error = 1,
1135 	},
1136 	[ICMP_REDIRECT] = {
1137 		.handler = icmp_redirect,
1138 		.error = 1,
1139 	},
1140 	[6] = {
1141 		.handler = icmp_discard,
1142 		.error = 1,
1143 	},
1144 	[7] = {
1145 		.handler = icmp_discard,
1146 		.error = 1,
1147 	},
1148 	[ICMP_ECHO] = {
1149 		.handler = icmp_echo,
1150 	},
1151 	[9] = {
1152 		.handler = icmp_discard,
1153 		.error = 1,
1154 	},
1155 	[10] = {
1156 		.handler = icmp_discard,
1157 		.error = 1,
1158 	},
1159 	[ICMP_TIME_EXCEEDED] = {
1160 		.handler = icmp_unreach,
1161 		.error = 1,
1162 	},
1163 	[ICMP_PARAMETERPROB] = {
1164 		.handler = icmp_unreach,
1165 		.error = 1,
1166 	},
1167 	[ICMP_TIMESTAMP] = {
1168 		.handler = icmp_timestamp,
1169 	},
1170 	[ICMP_TIMESTAMPREPLY] = {
1171 		.handler = icmp_discard,
1172 	},
1173 	[ICMP_INFO_REQUEST] = {
1174 		.handler = icmp_discard,
1175 	},
1176 	[ICMP_INFO_REPLY] = {
1177 		.handler = icmp_discard,
1178 	},
1179 	[ICMP_ADDRESS] = {
1180 		.handler = icmp_discard,
1181 	},
1182 	[ICMP_ADDRESSREPLY] = {
1183 		.handler = icmp_discard,
1184 	},
1185 };
1186 
1187 static void __net_exit icmp_sk_exit(struct net *net)
1188 {
1189 	int i;
1190 
1191 	for_each_possible_cpu(i)
1192 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1193 	free_percpu(net->ipv4.icmp_sk);
1194 	net->ipv4.icmp_sk = NULL;
1195 }
1196 
1197 static int __net_init icmp_sk_init(struct net *net)
1198 {
1199 	int i, err;
1200 
1201 	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1202 	if (!net->ipv4.icmp_sk)
1203 		return -ENOMEM;
1204 
1205 	for_each_possible_cpu(i) {
1206 		struct sock *sk;
1207 
1208 		err = inet_ctl_sock_create(&sk, PF_INET,
1209 					   SOCK_RAW, IPPROTO_ICMP, net);
1210 		if (err < 0)
1211 			goto fail;
1212 
1213 		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1214 
1215 		/* Enough space for 2 64K ICMP packets, including
1216 		 * sk_buff/skb_shared_info struct overhead.
1217 		 */
1218 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1219 
1220 		/*
1221 		 * Speedup sock_wfree()
1222 		 */
1223 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1224 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1225 	}
1226 
1227 	/* Control parameters for ECHO replies. */
1228 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1229 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1230 
1231 	/* Control parameter - ignore bogus broadcast responses? */
1232 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1233 
1234 	/*
1235 	 * 	Configurable global rate limit.
1236 	 *
1237 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1238 	 *	bucket ratemask defines which icmp types are ratelimited by
1239 	 *	setting	it's bit position.
1240 	 *
1241 	 *	default:
1242 	 *	dest unreachable (3), source quench (4),
1243 	 *	time exceeded (11), parameter problem (12)
1244 	 */
1245 
1246 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1247 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1248 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1249 
1250 	return 0;
1251 
1252 fail:
1253 	for_each_possible_cpu(i)
1254 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1255 	free_percpu(net->ipv4.icmp_sk);
1256 	return err;
1257 }
1258 
1259 static struct pernet_operations __net_initdata icmp_sk_ops = {
1260        .init = icmp_sk_init,
1261        .exit = icmp_sk_exit,
1262 };
1263 
1264 int __init icmp_init(void)
1265 {
1266 	return register_pernet_subsys(&icmp_sk_ops);
1267 }
1268