xref: /linux/net/ipv4/icmp.c (revision 42466b9f29b415c254dc4c2f4618e2a96951a406)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	NET3:	Implementation of the ICMP protocol layer.
4  *
5  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
6  *
7  *	Some of the function names and the icmp unreach table for this
8  *	module were derived from [icmp.c 1.0.11 06/02/93] by
9  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
10  *	Other than that this module is a complete rewrite.
11  *
12  *	Fixes:
13  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
14  *					with icmp type masking ability instead
15  *					of broken per type icmp timeouts.
16  *		Mike Shaver	:	RFC1122 checks.
17  *		Alan Cox	:	Multicast ping reply as self.
18  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
19  *					call.
20  *		Alan Cox	:	Added 216,128 byte paths to the MTU
21  *					code.
22  *		Martin Mares	:	RFC1812 checks.
23  *		Martin Mares	:	Can be configured to follow redirects
24  *					if acting as a router _without_ a
25  *					routing protocol (RFC 1812).
26  *		Martin Mares	:	Echo requests may be configured to
27  *					be ignored (RFC 1812).
28  *		Martin Mares	:	Limitation of ICMP error message
29  *					transmit rate (RFC 1812).
30  *		Martin Mares	:	TOS and Precedence set correctly
31  *					(RFC 1812).
32  *		Martin Mares	:	Now copying as much data from the
33  *					original packet as we can without
34  *					exceeding 576 bytes (RFC 1812).
35  *	Willy Konynenberg	:	Transparent proxying support.
36  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
37  *					path MTU bug.
38  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
39  *					valid (RFC 1812).
40  *		Andi Kleen	:	Check all packet lengths properly
41  *					and moved all kfree_skb() up to
42  *					icmp_rcv.
43  *		Andi Kleen	:	Move the rate limit bookkeeping
44  *					into the dest entry and use a token
45  *					bucket filter (thanks to ANK). Make
46  *					the rates sysctl configurable.
47  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
48  *					- IP option length was accounted wrongly
49  *					- ICMP header length was not accounted
50  *					  at all.
51  *              Tristan Greaves :       Added sysctl option to ignore bogus
52  *              			broadcast responses from broken routers.
53  *
54  * To Fix:
55  *
56  *	- Should use skb_pull() instead of all the manual checking.
57  *	  This would also greatly simply some upper layer error handlers. --AK
58  */
59 
60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
61 
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/jiffies.h>
65 #include <linux/kernel.h>
66 #include <linux/fcntl.h>
67 #include <linux/socket.h>
68 #include <linux/in.h>
69 #include <linux/inet.h>
70 #include <linux/inetdevice.h>
71 #include <linux/netdevice.h>
72 #include <linux/string.h>
73 #include <linux/netfilter_ipv4.h>
74 #include <linux/slab.h>
75 #include <net/snmp.h>
76 #include <net/ip.h>
77 #include <net/route.h>
78 #include <net/protocol.h>
79 #include <net/icmp.h>
80 #include <net/tcp.h>
81 #include <net/udp.h>
82 #include <net/raw.h>
83 #include <net/ping.h>
84 #include <linux/skbuff.h>
85 #include <net/sock.h>
86 #include <linux/errno.h>
87 #include <linux/timer.h>
88 #include <linux/init.h>
89 #include <linux/uaccess.h>
90 #include <net/checksum.h>
91 #include <net/xfrm.h>
92 #include <net/inet_common.h>
93 #include <net/ip_fib.h>
94 #include <net/l3mdev.h>
95 
96 /*
97  *	Build xmit assembly blocks
98  */
99 
100 struct icmp_bxm {
101 	struct sk_buff *skb;
102 	int offset;
103 	int data_len;
104 
105 	struct {
106 		struct icmphdr icmph;
107 		__be32	       times[3];
108 	} data;
109 	int head_len;
110 	struct ip_options_data replyopts;
111 };
112 
113 /* An array of errno for error messages from dest unreach. */
114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
115 
116 const struct icmp_err icmp_err_convert[] = {
117 	{
118 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
119 		.fatal = 0,
120 	},
121 	{
122 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
123 		.fatal = 0,
124 	},
125 	{
126 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
127 		.fatal = 1,
128 	},
129 	{
130 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
131 		.fatal = 1,
132 	},
133 	{
134 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
135 		.fatal = 0,
136 	},
137 	{
138 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
139 		.fatal = 0,
140 	},
141 	{
142 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
143 		.fatal = 1,
144 	},
145 	{
146 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
147 		.fatal = 1,
148 	},
149 	{
150 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
163 		.fatal = 0,
164 	},
165 	{
166 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
167 		.fatal = 0,
168 	},
169 	{
170 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
171 		.fatal = 1,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
175 		.fatal = 1,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
179 		.fatal = 1,
180 	},
181 };
182 EXPORT_SYMBOL(icmp_err_convert);
183 
184 /*
185  *	ICMP control array. This specifies what to do with each ICMP.
186  */
187 
188 struct icmp_control {
189 	bool (*handler)(struct sk_buff *skb);
190 	short   error;		/* This ICMP is classed as an error message */
191 };
192 
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194 
195 /*
196  *	The ICMP socket(s). This is the most convenient way to flow control
197  *	our ICMP output as well as maintain a clean interface throughout
198  *	all layers. All Socketless IP sends will soon be gone.
199  *
200  *	On SMP we have one ICMP socket per-cpu.
201  */
202 static struct sock *icmp_sk(struct net *net)
203 {
204 	return this_cpu_read(*net->ipv4.icmp_sk);
205 }
206 
207 /* Called with BH disabled */
208 static inline struct sock *icmp_xmit_lock(struct net *net)
209 {
210 	struct sock *sk;
211 
212 	sk = icmp_sk(net);
213 
214 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
215 		/* This can happen if the output path signals a
216 		 * dst_link_failure() for an outgoing ICMP packet.
217 		 */
218 		return NULL;
219 	}
220 	return sk;
221 }
222 
223 static inline void icmp_xmit_unlock(struct sock *sk)
224 {
225 	spin_unlock(&sk->sk_lock.slock);
226 }
227 
228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229 int sysctl_icmp_msgs_burst __read_mostly = 50;
230 
231 static struct {
232 	spinlock_t	lock;
233 	u32		credit;
234 	u32		stamp;
235 } icmp_global = {
236 	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
237 };
238 
239 /**
240  * icmp_global_allow - Are we allowed to send one more ICMP message ?
241  *
242  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
243  * Returns false if we reached the limit and can not send another packet.
244  * Note: called with BH disabled
245  */
246 bool icmp_global_allow(void)
247 {
248 	u32 credit, delta, incr = 0, now = (u32)jiffies;
249 	bool rc = false;
250 
251 	/* Check if token bucket is empty and cannot be refilled
252 	 * without taking the spinlock.
253 	 */
254 	if (!icmp_global.credit) {
255 		delta = min_t(u32, now - icmp_global.stamp, HZ);
256 		if (delta < HZ / 50)
257 			return false;
258 	}
259 
260 	spin_lock(&icmp_global.lock);
261 	delta = min_t(u32, now - icmp_global.stamp, HZ);
262 	if (delta >= HZ / 50) {
263 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
264 		if (incr)
265 			icmp_global.stamp = now;
266 	}
267 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
268 	if (credit) {
269 		credit--;
270 		rc = true;
271 	}
272 	icmp_global.credit = credit;
273 	spin_unlock(&icmp_global.lock);
274 	return rc;
275 }
276 EXPORT_SYMBOL(icmp_global_allow);
277 
278 static bool icmpv4_mask_allow(struct net *net, int type, int code)
279 {
280 	if (type > NR_ICMP_TYPES)
281 		return true;
282 
283 	/* Don't limit PMTU discovery. */
284 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 		return true;
286 
287 	/* Limit if icmp type is enabled in ratemask. */
288 	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
289 		return true;
290 
291 	return false;
292 }
293 
294 static bool icmpv4_global_allow(struct net *net, int type, int code)
295 {
296 	if (icmpv4_mask_allow(net, type, code))
297 		return true;
298 
299 	if (icmp_global_allow())
300 		return true;
301 
302 	return false;
303 }
304 
305 /*
306  *	Send an ICMP frame.
307  */
308 
309 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
310 			       struct flowi4 *fl4, int type, int code)
311 {
312 	struct dst_entry *dst = &rt->dst;
313 	struct inet_peer *peer;
314 	bool rc = true;
315 	int vif;
316 
317 	if (icmpv4_mask_allow(net, type, code))
318 		goto out;
319 
320 	/* No rate limit on loopback */
321 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
322 		goto out;
323 
324 	vif = l3mdev_master_ifindex(dst->dev);
325 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
326 	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
327 	if (peer)
328 		inet_putpeer(peer);
329 out:
330 	return rc;
331 }
332 
333 /*
334  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
335  */
336 void icmp_out_count(struct net *net, unsigned char type)
337 {
338 	ICMPMSGOUT_INC_STATS(net, type);
339 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
340 }
341 
342 /*
343  *	Checksum each fragment, and on the first include the headers and final
344  *	checksum.
345  */
346 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
347 			  struct sk_buff *skb)
348 {
349 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
350 	__wsum csum;
351 
352 	csum = skb_copy_and_csum_bits(icmp_param->skb,
353 				      icmp_param->offset + offset,
354 				      to, len, 0);
355 
356 	skb->csum = csum_block_add(skb->csum, csum, odd);
357 	if (icmp_pointers[icmp_param->data.icmph.type].error)
358 		nf_ct_attach(skb, icmp_param->skb);
359 	return 0;
360 }
361 
362 static void icmp_push_reply(struct icmp_bxm *icmp_param,
363 			    struct flowi4 *fl4,
364 			    struct ipcm_cookie *ipc, struct rtable **rt)
365 {
366 	struct sock *sk;
367 	struct sk_buff *skb;
368 
369 	sk = icmp_sk(dev_net((*rt)->dst.dev));
370 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
371 			   icmp_param->data_len+icmp_param->head_len,
372 			   icmp_param->head_len,
373 			   ipc, rt, MSG_DONTWAIT) < 0) {
374 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
375 		ip_flush_pending_frames(sk);
376 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
377 		struct icmphdr *icmph = icmp_hdr(skb);
378 		__wsum csum = 0;
379 		struct sk_buff *skb1;
380 
381 		skb_queue_walk(&sk->sk_write_queue, skb1) {
382 			csum = csum_add(csum, skb1->csum);
383 		}
384 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
385 						 (char *)icmph,
386 						 icmp_param->head_len, csum);
387 		icmph->checksum = csum_fold(csum);
388 		skb->ip_summed = CHECKSUM_NONE;
389 		ip_push_pending_frames(sk, fl4);
390 	}
391 }
392 
393 /*
394  *	Driving logic for building and sending ICMP messages.
395  */
396 
397 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
398 {
399 	struct ipcm_cookie ipc;
400 	struct rtable *rt = skb_rtable(skb);
401 	struct net *net = dev_net(rt->dst.dev);
402 	struct flowi4 fl4;
403 	struct sock *sk;
404 	struct inet_sock *inet;
405 	__be32 daddr, saddr;
406 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
407 	int type = icmp_param->data.icmph.type;
408 	int code = icmp_param->data.icmph.code;
409 
410 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
411 		return;
412 
413 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
414 	local_bh_disable();
415 
416 	/* global icmp_msgs_per_sec */
417 	if (!icmpv4_global_allow(net, type, code))
418 		goto out_bh_enable;
419 
420 	sk = icmp_xmit_lock(net);
421 	if (!sk)
422 		goto out_bh_enable;
423 	inet = inet_sk(sk);
424 
425 	icmp_param->data.icmph.checksum = 0;
426 
427 	ipcm_init(&ipc);
428 	inet->tos = ip_hdr(skb)->tos;
429 	sk->sk_mark = mark;
430 	daddr = ipc.addr = ip_hdr(skb)->saddr;
431 	saddr = fib_compute_spec_dst(skb);
432 
433 	if (icmp_param->replyopts.opt.opt.optlen) {
434 		ipc.opt = &icmp_param->replyopts.opt;
435 		if (ipc.opt->opt.srr)
436 			daddr = icmp_param->replyopts.opt.opt.faddr;
437 	}
438 	memset(&fl4, 0, sizeof(fl4));
439 	fl4.daddr = daddr;
440 	fl4.saddr = saddr;
441 	fl4.flowi4_mark = mark;
442 	fl4.flowi4_uid = sock_net_uid(net, NULL);
443 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
444 	fl4.flowi4_proto = IPPROTO_ICMP;
445 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
446 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
447 	rt = ip_route_output_key(net, &fl4);
448 	if (IS_ERR(rt))
449 		goto out_unlock;
450 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
451 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
452 	ip_rt_put(rt);
453 out_unlock:
454 	icmp_xmit_unlock(sk);
455 out_bh_enable:
456 	local_bh_enable();
457 }
458 
459 static struct rtable *icmp_route_lookup(struct net *net,
460 					struct flowi4 *fl4,
461 					struct sk_buff *skb_in,
462 					const struct iphdr *iph,
463 					__be32 saddr, u8 tos, u32 mark,
464 					int type, int code,
465 					struct icmp_bxm *param)
466 {
467 	struct rtable *rt, *rt2;
468 	struct flowi4 fl4_dec;
469 	int err;
470 
471 	memset(fl4, 0, sizeof(*fl4));
472 	fl4->daddr = (param->replyopts.opt.opt.srr ?
473 		      param->replyopts.opt.opt.faddr : iph->saddr);
474 	fl4->saddr = saddr;
475 	fl4->flowi4_mark = mark;
476 	fl4->flowi4_uid = sock_net_uid(net, NULL);
477 	fl4->flowi4_tos = RT_TOS(tos);
478 	fl4->flowi4_proto = IPPROTO_ICMP;
479 	fl4->fl4_icmp_type = type;
480 	fl4->fl4_icmp_code = code;
481 	fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
482 
483 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
484 	rt = ip_route_output_key_hash(net, fl4, skb_in);
485 	if (IS_ERR(rt))
486 		return rt;
487 
488 	/* No need to clone since we're just using its address. */
489 	rt2 = rt;
490 
491 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
492 					   flowi4_to_flowi(fl4), NULL, 0);
493 	if (!IS_ERR(rt)) {
494 		if (rt != rt2)
495 			return rt;
496 	} else if (PTR_ERR(rt) == -EPERM) {
497 		rt = NULL;
498 	} else
499 		return rt;
500 
501 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
502 	if (err)
503 		goto relookup_failed;
504 
505 	if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
506 				     fl4_dec.saddr) == RTN_LOCAL) {
507 		rt2 = __ip_route_output_key(net, &fl4_dec);
508 		if (IS_ERR(rt2))
509 			err = PTR_ERR(rt2);
510 	} else {
511 		struct flowi4 fl4_2 = {};
512 		unsigned long orefdst;
513 
514 		fl4_2.daddr = fl4_dec.saddr;
515 		rt2 = ip_route_output_key(net, &fl4_2);
516 		if (IS_ERR(rt2)) {
517 			err = PTR_ERR(rt2);
518 			goto relookup_failed;
519 		}
520 		/* Ugh! */
521 		orefdst = skb_in->_skb_refdst; /* save old refdst */
522 		skb_dst_set(skb_in, NULL);
523 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
524 				     RT_TOS(tos), rt2->dst.dev);
525 
526 		dst_release(&rt2->dst);
527 		rt2 = skb_rtable(skb_in);
528 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
529 	}
530 
531 	if (err)
532 		goto relookup_failed;
533 
534 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
535 					    flowi4_to_flowi(&fl4_dec), NULL,
536 					    XFRM_LOOKUP_ICMP);
537 	if (!IS_ERR(rt2)) {
538 		dst_release(&rt->dst);
539 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
540 		rt = rt2;
541 	} else if (PTR_ERR(rt2) == -EPERM) {
542 		if (rt)
543 			dst_release(&rt->dst);
544 		return rt2;
545 	} else {
546 		err = PTR_ERR(rt2);
547 		goto relookup_failed;
548 	}
549 	return rt;
550 
551 relookup_failed:
552 	if (rt)
553 		return rt;
554 	return ERR_PTR(err);
555 }
556 
557 /*
558  *	Send an ICMP message in response to a situation
559  *
560  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
561  *		  MAY send more (we do).
562  *			MUST NOT change this header information.
563  *			MUST NOT reply to a multicast/broadcast IP address.
564  *			MUST NOT reply to a multicast/broadcast MAC address.
565  *			MUST reply to only the first fragment.
566  */
567 
568 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
569 		 const struct ip_options *opt)
570 {
571 	struct iphdr *iph;
572 	int room;
573 	struct icmp_bxm icmp_param;
574 	struct rtable *rt = skb_rtable(skb_in);
575 	struct ipcm_cookie ipc;
576 	struct flowi4 fl4;
577 	__be32 saddr;
578 	u8  tos;
579 	u32 mark;
580 	struct net *net;
581 	struct sock *sk;
582 
583 	if (!rt)
584 		goto out;
585 
586 	if (rt->dst.dev)
587 		net = dev_net(rt->dst.dev);
588 	else if (skb_in->dev)
589 		net = dev_net(skb_in->dev);
590 	else
591 		goto out;
592 
593 	/*
594 	 *	Find the original header. It is expected to be valid, of course.
595 	 *	Check this, icmp_send is called from the most obscure devices
596 	 *	sometimes.
597 	 */
598 	iph = ip_hdr(skb_in);
599 
600 	if ((u8 *)iph < skb_in->head ||
601 	    (skb_network_header(skb_in) + sizeof(*iph)) >
602 	    skb_tail_pointer(skb_in))
603 		goto out;
604 
605 	/*
606 	 *	No replies to physical multicast/broadcast
607 	 */
608 	if (skb_in->pkt_type != PACKET_HOST)
609 		goto out;
610 
611 	/*
612 	 *	Now check at the protocol level
613 	 */
614 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
615 		goto out;
616 
617 	/*
618 	 *	Only reply to fragment 0. We byte re-order the constant
619 	 *	mask for efficiency.
620 	 */
621 	if (iph->frag_off & htons(IP_OFFSET))
622 		goto out;
623 
624 	/*
625 	 *	If we send an ICMP error to an ICMP error a mess would result..
626 	 */
627 	if (icmp_pointers[type].error) {
628 		/*
629 		 *	We are an error, check if we are replying to an
630 		 *	ICMP error
631 		 */
632 		if (iph->protocol == IPPROTO_ICMP) {
633 			u8 _inner_type, *itp;
634 
635 			itp = skb_header_pointer(skb_in,
636 						 skb_network_header(skb_in) +
637 						 (iph->ihl << 2) +
638 						 offsetof(struct icmphdr,
639 							  type) -
640 						 skb_in->data,
641 						 sizeof(_inner_type),
642 						 &_inner_type);
643 			if (!itp)
644 				goto out;
645 
646 			/*
647 			 *	Assume any unknown ICMP type is an error. This
648 			 *	isn't specified by the RFC, but think about it..
649 			 */
650 			if (*itp > NR_ICMP_TYPES ||
651 			    icmp_pointers[*itp].error)
652 				goto out;
653 		}
654 	}
655 
656 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
657 	local_bh_disable();
658 
659 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
660 	 * incoming dev is loopback.  If outgoing dev change to not be
661 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
662 	 */
663 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
664 	      !icmpv4_global_allow(net, type, code))
665 		goto out_bh_enable;
666 
667 	sk = icmp_xmit_lock(net);
668 	if (!sk)
669 		goto out_bh_enable;
670 
671 	/*
672 	 *	Construct source address and options.
673 	 */
674 
675 	saddr = iph->daddr;
676 	if (!(rt->rt_flags & RTCF_LOCAL)) {
677 		struct net_device *dev = NULL;
678 
679 		rcu_read_lock();
680 		if (rt_is_input_route(rt) &&
681 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
682 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
683 
684 		if (dev)
685 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
686 		else
687 			saddr = 0;
688 		rcu_read_unlock();
689 	}
690 
691 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
692 					   IPTOS_PREC_INTERNETCONTROL) :
693 					  iph->tos;
694 	mark = IP4_REPLY_MARK(net, skb_in->mark);
695 
696 	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
697 		goto out_unlock;
698 
699 
700 	/*
701 	 *	Prepare data for ICMP header.
702 	 */
703 
704 	icmp_param.data.icmph.type	 = type;
705 	icmp_param.data.icmph.code	 = code;
706 	icmp_param.data.icmph.un.gateway = info;
707 	icmp_param.data.icmph.checksum	 = 0;
708 	icmp_param.skb	  = skb_in;
709 	icmp_param.offset = skb_network_offset(skb_in);
710 	inet_sk(sk)->tos = tos;
711 	sk->sk_mark = mark;
712 	ipcm_init(&ipc);
713 	ipc.addr = iph->saddr;
714 	ipc.opt = &icmp_param.replyopts.opt;
715 
716 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
717 			       type, code, &icmp_param);
718 	if (IS_ERR(rt))
719 		goto out_unlock;
720 
721 	/* peer icmp_ratelimit */
722 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
723 		goto ende;
724 
725 	/* RFC says return as much as we can without exceeding 576 bytes. */
726 
727 	room = dst_mtu(&rt->dst);
728 	if (room > 576)
729 		room = 576;
730 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
731 	room -= sizeof(struct icmphdr);
732 
733 	icmp_param.data_len = skb_in->len - icmp_param.offset;
734 	if (icmp_param.data_len > room)
735 		icmp_param.data_len = room;
736 	icmp_param.head_len = sizeof(struct icmphdr);
737 
738 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
739 ende:
740 	ip_rt_put(rt);
741 out_unlock:
742 	icmp_xmit_unlock(sk);
743 out_bh_enable:
744 	local_bh_enable();
745 out:;
746 }
747 EXPORT_SYMBOL(__icmp_send);
748 
749 
750 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
751 {
752 	const struct iphdr *iph = (const struct iphdr *) skb->data;
753 	const struct net_protocol *ipprot;
754 	int protocol = iph->protocol;
755 
756 	/* Checkin full IP header plus 8 bytes of protocol to
757 	 * avoid additional coding at protocol handlers.
758 	 */
759 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
760 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
761 		return;
762 	}
763 
764 	raw_icmp_error(skb, protocol, info);
765 
766 	ipprot = rcu_dereference(inet_protos[protocol]);
767 	if (ipprot && ipprot->err_handler)
768 		ipprot->err_handler(skb, info);
769 }
770 
771 static bool icmp_tag_validation(int proto)
772 {
773 	bool ok;
774 
775 	rcu_read_lock();
776 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
777 	rcu_read_unlock();
778 	return ok;
779 }
780 
781 /*
782  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
783  *	ICMP_PARAMETERPROB.
784  */
785 
786 static bool icmp_unreach(struct sk_buff *skb)
787 {
788 	const struct iphdr *iph;
789 	struct icmphdr *icmph;
790 	struct net *net;
791 	u32 info = 0;
792 
793 	net = dev_net(skb_dst(skb)->dev);
794 
795 	/*
796 	 *	Incomplete header ?
797 	 * 	Only checks for the IP header, there should be an
798 	 *	additional check for longer headers in upper levels.
799 	 */
800 
801 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
802 		goto out_err;
803 
804 	icmph = icmp_hdr(skb);
805 	iph   = (const struct iphdr *)skb->data;
806 
807 	if (iph->ihl < 5) /* Mangled header, drop. */
808 		goto out_err;
809 
810 	switch (icmph->type) {
811 	case ICMP_DEST_UNREACH:
812 		switch (icmph->code & 15) {
813 		case ICMP_NET_UNREACH:
814 		case ICMP_HOST_UNREACH:
815 		case ICMP_PROT_UNREACH:
816 		case ICMP_PORT_UNREACH:
817 			break;
818 		case ICMP_FRAG_NEEDED:
819 			/* for documentation of the ip_no_pmtu_disc
820 			 * values please see
821 			 * Documentation/networking/ip-sysctl.txt
822 			 */
823 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
824 			default:
825 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
826 						    &iph->daddr);
827 				break;
828 			case 2:
829 				goto out;
830 			case 3:
831 				if (!icmp_tag_validation(iph->protocol))
832 					goto out;
833 				/* fall through */
834 			case 0:
835 				info = ntohs(icmph->un.frag.mtu);
836 			}
837 			break;
838 		case ICMP_SR_FAILED:
839 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
840 					    &iph->daddr);
841 			break;
842 		default:
843 			break;
844 		}
845 		if (icmph->code > NR_ICMP_UNREACH)
846 			goto out;
847 		break;
848 	case ICMP_PARAMETERPROB:
849 		info = ntohl(icmph->un.gateway) >> 24;
850 		break;
851 	case ICMP_TIME_EXCEEDED:
852 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
853 		if (icmph->code == ICMP_EXC_FRAGTIME)
854 			goto out;
855 		break;
856 	}
857 
858 	/*
859 	 *	Throw it at our lower layers
860 	 *
861 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
862 	 *		  header.
863 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
864 	 *		  transport layer.
865 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
866 	 *		  transport layer.
867 	 */
868 
869 	/*
870 	 *	Check the other end isn't violating RFC 1122. Some routers send
871 	 *	bogus responses to broadcast frames. If you see this message
872 	 *	first check your netmask matches at both ends, if it does then
873 	 *	get the other vendor to fix their kit.
874 	 */
875 
876 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
877 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
878 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
879 				     &ip_hdr(skb)->saddr,
880 				     icmph->type, icmph->code,
881 				     &iph->daddr, skb->dev->name);
882 		goto out;
883 	}
884 
885 	icmp_socket_deliver(skb, info);
886 
887 out:
888 	return true;
889 out_err:
890 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
891 	return false;
892 }
893 
894 
895 /*
896  *	Handle ICMP_REDIRECT.
897  */
898 
899 static bool icmp_redirect(struct sk_buff *skb)
900 {
901 	if (skb->len < sizeof(struct iphdr)) {
902 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
903 		return false;
904 	}
905 
906 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
907 		/* there aught to be a stat */
908 		return false;
909 	}
910 
911 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
912 	return true;
913 }
914 
915 /*
916  *	Handle ICMP_ECHO ("ping") requests.
917  *
918  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
919  *		  requests.
920  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
921  *		  included in the reply.
922  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
923  *		  echo requests, MUST have default=NOT.
924  *	See also WRT handling of options once they are done and working.
925  */
926 
927 static bool icmp_echo(struct sk_buff *skb)
928 {
929 	struct net *net;
930 
931 	net = dev_net(skb_dst(skb)->dev);
932 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
933 		struct icmp_bxm icmp_param;
934 
935 		icmp_param.data.icmph	   = *icmp_hdr(skb);
936 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
937 		icmp_param.skb		   = skb;
938 		icmp_param.offset	   = 0;
939 		icmp_param.data_len	   = skb->len;
940 		icmp_param.head_len	   = sizeof(struct icmphdr);
941 		icmp_reply(&icmp_param, skb);
942 	}
943 	/* should there be an ICMP stat for ignored echos? */
944 	return true;
945 }
946 
947 /*
948  *	Handle ICMP Timestamp requests.
949  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
950  *		  SHOULD be in the kernel for minimum random latency.
951  *		  MUST be accurate to a few minutes.
952  *		  MUST be updated at least at 15Hz.
953  */
954 static bool icmp_timestamp(struct sk_buff *skb)
955 {
956 	struct icmp_bxm icmp_param;
957 	/*
958 	 *	Too short.
959 	 */
960 	if (skb->len < 4)
961 		goto out_err;
962 
963 	/*
964 	 *	Fill in the current time as ms since midnight UT:
965 	 */
966 	icmp_param.data.times[1] = inet_current_timestamp();
967 	icmp_param.data.times[2] = icmp_param.data.times[1];
968 
969 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
970 
971 	icmp_param.data.icmph	   = *icmp_hdr(skb);
972 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
973 	icmp_param.data.icmph.code = 0;
974 	icmp_param.skb		   = skb;
975 	icmp_param.offset	   = 0;
976 	icmp_param.data_len	   = 0;
977 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
978 	icmp_reply(&icmp_param, skb);
979 	return true;
980 
981 out_err:
982 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
983 	return false;
984 }
985 
986 static bool icmp_discard(struct sk_buff *skb)
987 {
988 	/* pretend it was a success */
989 	return true;
990 }
991 
992 /*
993  *	Deal with incoming ICMP packets.
994  */
995 int icmp_rcv(struct sk_buff *skb)
996 {
997 	struct icmphdr *icmph;
998 	struct rtable *rt = skb_rtable(skb);
999 	struct net *net = dev_net(rt->dst.dev);
1000 	bool success;
1001 
1002 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1003 		struct sec_path *sp = skb_sec_path(skb);
1004 		int nh;
1005 
1006 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1007 				 XFRM_STATE_ICMP))
1008 			goto drop;
1009 
1010 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1011 			goto drop;
1012 
1013 		nh = skb_network_offset(skb);
1014 		skb_set_network_header(skb, sizeof(*icmph));
1015 
1016 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1017 			goto drop;
1018 
1019 		skb_set_network_header(skb, nh);
1020 	}
1021 
1022 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1023 
1024 	if (skb_checksum_simple_validate(skb))
1025 		goto csum_error;
1026 
1027 	if (!pskb_pull(skb, sizeof(*icmph)))
1028 		goto error;
1029 
1030 	icmph = icmp_hdr(skb);
1031 
1032 	ICMPMSGIN_INC_STATS(net, icmph->type);
1033 	/*
1034 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1035 	 *
1036 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1037 	 *		  discarded.
1038 	 */
1039 	if (icmph->type > NR_ICMP_TYPES)
1040 		goto error;
1041 
1042 
1043 	/*
1044 	 *	Parse the ICMP message
1045 	 */
1046 
1047 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1048 		/*
1049 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1050 		 *	  silently ignored (we let user decide with a sysctl).
1051 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1052 		 *	  discarded if to broadcast/multicast.
1053 		 */
1054 		if ((icmph->type == ICMP_ECHO ||
1055 		     icmph->type == ICMP_TIMESTAMP) &&
1056 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1057 			goto error;
1058 		}
1059 		if (icmph->type != ICMP_ECHO &&
1060 		    icmph->type != ICMP_TIMESTAMP &&
1061 		    icmph->type != ICMP_ADDRESS &&
1062 		    icmph->type != ICMP_ADDRESSREPLY) {
1063 			goto error;
1064 		}
1065 	}
1066 
1067 	success = icmp_pointers[icmph->type].handler(skb);
1068 
1069 	if (success)  {
1070 		consume_skb(skb);
1071 		return NET_RX_SUCCESS;
1072 	}
1073 
1074 drop:
1075 	kfree_skb(skb);
1076 	return NET_RX_DROP;
1077 csum_error:
1078 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1079 error:
1080 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1081 	goto drop;
1082 }
1083 
1084 int icmp_err(struct sk_buff *skb, u32 info)
1085 {
1086 	struct iphdr *iph = (struct iphdr *)skb->data;
1087 	int offset = iph->ihl<<2;
1088 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1089 	int type = icmp_hdr(skb)->type;
1090 	int code = icmp_hdr(skb)->code;
1091 	struct net *net = dev_net(skb->dev);
1092 
1093 	/*
1094 	 * Use ping_err to handle all icmp errors except those
1095 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1096 	 */
1097 	if (icmph->type != ICMP_ECHOREPLY) {
1098 		ping_err(skb, offset, info);
1099 		return 0;
1100 	}
1101 
1102 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1103 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1104 	else if (type == ICMP_REDIRECT)
1105 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1106 
1107 	return 0;
1108 }
1109 
1110 /*
1111  *	This table is the definition of how we handle ICMP.
1112  */
1113 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1114 	[ICMP_ECHOREPLY] = {
1115 		.handler = ping_rcv,
1116 	},
1117 	[1] = {
1118 		.handler = icmp_discard,
1119 		.error = 1,
1120 	},
1121 	[2] = {
1122 		.handler = icmp_discard,
1123 		.error = 1,
1124 	},
1125 	[ICMP_DEST_UNREACH] = {
1126 		.handler = icmp_unreach,
1127 		.error = 1,
1128 	},
1129 	[ICMP_SOURCE_QUENCH] = {
1130 		.handler = icmp_unreach,
1131 		.error = 1,
1132 	},
1133 	[ICMP_REDIRECT] = {
1134 		.handler = icmp_redirect,
1135 		.error = 1,
1136 	},
1137 	[6] = {
1138 		.handler = icmp_discard,
1139 		.error = 1,
1140 	},
1141 	[7] = {
1142 		.handler = icmp_discard,
1143 		.error = 1,
1144 	},
1145 	[ICMP_ECHO] = {
1146 		.handler = icmp_echo,
1147 	},
1148 	[9] = {
1149 		.handler = icmp_discard,
1150 		.error = 1,
1151 	},
1152 	[10] = {
1153 		.handler = icmp_discard,
1154 		.error = 1,
1155 	},
1156 	[ICMP_TIME_EXCEEDED] = {
1157 		.handler = icmp_unreach,
1158 		.error = 1,
1159 	},
1160 	[ICMP_PARAMETERPROB] = {
1161 		.handler = icmp_unreach,
1162 		.error = 1,
1163 	},
1164 	[ICMP_TIMESTAMP] = {
1165 		.handler = icmp_timestamp,
1166 	},
1167 	[ICMP_TIMESTAMPREPLY] = {
1168 		.handler = icmp_discard,
1169 	},
1170 	[ICMP_INFO_REQUEST] = {
1171 		.handler = icmp_discard,
1172 	},
1173 	[ICMP_INFO_REPLY] = {
1174 		.handler = icmp_discard,
1175 	},
1176 	[ICMP_ADDRESS] = {
1177 		.handler = icmp_discard,
1178 	},
1179 	[ICMP_ADDRESSREPLY] = {
1180 		.handler = icmp_discard,
1181 	},
1182 };
1183 
1184 static void __net_exit icmp_sk_exit(struct net *net)
1185 {
1186 	int i;
1187 
1188 	for_each_possible_cpu(i)
1189 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1190 	free_percpu(net->ipv4.icmp_sk);
1191 	net->ipv4.icmp_sk = NULL;
1192 }
1193 
1194 static int __net_init icmp_sk_init(struct net *net)
1195 {
1196 	int i, err;
1197 
1198 	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1199 	if (!net->ipv4.icmp_sk)
1200 		return -ENOMEM;
1201 
1202 	for_each_possible_cpu(i) {
1203 		struct sock *sk;
1204 
1205 		err = inet_ctl_sock_create(&sk, PF_INET,
1206 					   SOCK_RAW, IPPROTO_ICMP, net);
1207 		if (err < 0)
1208 			goto fail;
1209 
1210 		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1211 
1212 		/* Enough space for 2 64K ICMP packets, including
1213 		 * sk_buff/skb_shared_info struct overhead.
1214 		 */
1215 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1216 
1217 		/*
1218 		 * Speedup sock_wfree()
1219 		 */
1220 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1221 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1222 	}
1223 
1224 	/* Control parameters for ECHO replies. */
1225 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1226 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1227 
1228 	/* Control parameter - ignore bogus broadcast responses? */
1229 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1230 
1231 	/*
1232 	 * 	Configurable global rate limit.
1233 	 *
1234 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1235 	 *	bucket ratemask defines which icmp types are ratelimited by
1236 	 *	setting	it's bit position.
1237 	 *
1238 	 *	default:
1239 	 *	dest unreachable (3), source quench (4),
1240 	 *	time exceeded (11), parameter problem (12)
1241 	 */
1242 
1243 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1244 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1245 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1246 
1247 	return 0;
1248 
1249 fail:
1250 	icmp_sk_exit(net);
1251 	return err;
1252 }
1253 
1254 static struct pernet_operations __net_initdata icmp_sk_ops = {
1255        .init = icmp_sk_init,
1256        .exit = icmp_sk_exit,
1257 };
1258 
1259 int __init icmp_init(void)
1260 {
1261 	return register_pernet_subsys(&icmp_sk_ops);
1262 }
1263