xref: /openbmc/linux/net/ipv4/icmp.c (revision 5086ea4b)
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. The READ_ONCE() are paired
253 	 * with the following WRITE_ONCE() in this same function.
254 	 */
255 	if (!READ_ONCE(icmp_global.credit)) {
256 		delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
257 		if (delta < HZ / 50)
258 			return false;
259 	}
260 
261 	spin_lock(&icmp_global.lock);
262 	delta = min_t(u32, now - icmp_global.stamp, HZ);
263 	if (delta >= HZ / 50) {
264 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
265 		if (incr)
266 			WRITE_ONCE(icmp_global.stamp, now);
267 	}
268 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
269 	if (credit) {
270 		/* We want to use a credit of one in average, but need to randomize
271 		 * it for security reasons.
272 		 */
273 		credit = max_t(int, credit - prandom_u32_max(3), 0);
274 		rc = true;
275 	}
276 	WRITE_ONCE(icmp_global.credit, credit);
277 	spin_unlock(&icmp_global.lock);
278 	return rc;
279 }
280 EXPORT_SYMBOL(icmp_global_allow);
281 
282 static bool icmpv4_mask_allow(struct net *net, int type, int code)
283 {
284 	if (type > NR_ICMP_TYPES)
285 		return true;
286 
287 	/* Don't limit PMTU discovery. */
288 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
289 		return true;
290 
291 	/* Limit if icmp type is enabled in ratemask. */
292 	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
293 		return true;
294 
295 	return false;
296 }
297 
298 static bool icmpv4_global_allow(struct net *net, int type, int code)
299 {
300 	if (icmpv4_mask_allow(net, type, code))
301 		return true;
302 
303 	if (icmp_global_allow())
304 		return true;
305 
306 	return false;
307 }
308 
309 /*
310  *	Send an ICMP frame.
311  */
312 
313 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
314 			       struct flowi4 *fl4, int type, int code)
315 {
316 	struct dst_entry *dst = &rt->dst;
317 	struct inet_peer *peer;
318 	bool rc = true;
319 	int vif;
320 
321 	if (icmpv4_mask_allow(net, type, code))
322 		goto out;
323 
324 	/* No rate limit on loopback */
325 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
326 		goto out;
327 
328 	vif = l3mdev_master_ifindex(dst->dev);
329 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
330 	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
331 	if (peer)
332 		inet_putpeer(peer);
333 out:
334 	return rc;
335 }
336 
337 /*
338  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
339  */
340 void icmp_out_count(struct net *net, unsigned char type)
341 {
342 	ICMPMSGOUT_INC_STATS(net, type);
343 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
344 }
345 
346 /*
347  *	Checksum each fragment, and on the first include the headers and final
348  *	checksum.
349  */
350 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
351 			  struct sk_buff *skb)
352 {
353 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
354 	__wsum csum;
355 
356 	csum = skb_copy_and_csum_bits(icmp_param->skb,
357 				      icmp_param->offset + offset,
358 				      to, len);
359 
360 	skb->csum = csum_block_add(skb->csum, csum, odd);
361 	if (icmp_pointers[icmp_param->data.icmph.type].error)
362 		nf_ct_attach(skb, icmp_param->skb);
363 	return 0;
364 }
365 
366 static void icmp_push_reply(struct icmp_bxm *icmp_param,
367 			    struct flowi4 *fl4,
368 			    struct ipcm_cookie *ipc, struct rtable **rt)
369 {
370 	struct sock *sk;
371 	struct sk_buff *skb;
372 
373 	sk = icmp_sk(dev_net((*rt)->dst.dev));
374 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
375 			   icmp_param->data_len+icmp_param->head_len,
376 			   icmp_param->head_len,
377 			   ipc, rt, MSG_DONTWAIT) < 0) {
378 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
379 		ip_flush_pending_frames(sk);
380 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
381 		struct icmphdr *icmph = icmp_hdr(skb);
382 		__wsum csum;
383 		struct sk_buff *skb1;
384 
385 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
386 						 (char *)icmph,
387 						 icmp_param->head_len);
388 		skb_queue_walk(&sk->sk_write_queue, skb1) {
389 			csum = csum_add(csum, skb1->csum);
390 		}
391 		icmph->checksum = csum_fold(csum);
392 		skb->ip_summed = CHECKSUM_NONE;
393 		ip_push_pending_frames(sk, fl4);
394 	}
395 }
396 
397 /*
398  *	Driving logic for building and sending ICMP messages.
399  */
400 
401 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
402 {
403 	struct ipcm_cookie ipc;
404 	struct rtable *rt = skb_rtable(skb);
405 	struct net *net = dev_net(rt->dst.dev);
406 	struct flowi4 fl4;
407 	struct sock *sk;
408 	struct inet_sock *inet;
409 	__be32 daddr, saddr;
410 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
411 	int type = icmp_param->data.icmph.type;
412 	int code = icmp_param->data.icmph.code;
413 
414 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
415 		return;
416 
417 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
418 	local_bh_disable();
419 
420 	/* global icmp_msgs_per_sec */
421 	if (!icmpv4_global_allow(net, type, code))
422 		goto out_bh_enable;
423 
424 	sk = icmp_xmit_lock(net);
425 	if (!sk)
426 		goto out_bh_enable;
427 	inet = inet_sk(sk);
428 
429 	icmp_param->data.icmph.checksum = 0;
430 
431 	ipcm_init(&ipc);
432 	inet->tos = ip_hdr(skb)->tos;
433 	ipc.sockc.mark = mark;
434 	daddr = ipc.addr = ip_hdr(skb)->saddr;
435 	saddr = fib_compute_spec_dst(skb);
436 
437 	if (icmp_param->replyopts.opt.opt.optlen) {
438 		ipc.opt = &icmp_param->replyopts.opt;
439 		if (ipc.opt->opt.srr)
440 			daddr = icmp_param->replyopts.opt.opt.faddr;
441 	}
442 	memset(&fl4, 0, sizeof(fl4));
443 	fl4.daddr = daddr;
444 	fl4.saddr = saddr;
445 	fl4.flowi4_mark = mark;
446 	fl4.flowi4_uid = sock_net_uid(net, NULL);
447 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
448 	fl4.flowi4_proto = IPPROTO_ICMP;
449 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
450 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
451 	rt = ip_route_output_key(net, &fl4);
452 	if (IS_ERR(rt))
453 		goto out_unlock;
454 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
455 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
456 	ip_rt_put(rt);
457 out_unlock:
458 	icmp_xmit_unlock(sk);
459 out_bh_enable:
460 	local_bh_enable();
461 }
462 
463 /*
464  * The device used for looking up which routing table to use for sending an ICMP
465  * error is preferably the source whenever it is set, which should ensure the
466  * icmp error can be sent to the source host, else lookup using the routing
467  * table of the destination device, else use the main routing table (index 0).
468  */
469 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
470 {
471 	struct net_device *route_lookup_dev = NULL;
472 
473 	if (skb->dev)
474 		route_lookup_dev = skb->dev;
475 	else if (skb_dst(skb))
476 		route_lookup_dev = skb_dst(skb)->dev;
477 	return route_lookup_dev;
478 }
479 
480 static struct rtable *icmp_route_lookup(struct net *net,
481 					struct flowi4 *fl4,
482 					struct sk_buff *skb_in,
483 					const struct iphdr *iph,
484 					__be32 saddr, u8 tos, u32 mark,
485 					int type, int code,
486 					struct icmp_bxm *param)
487 {
488 	struct net_device *route_lookup_dev;
489 	struct rtable *rt, *rt2;
490 	struct flowi4 fl4_dec;
491 	int err;
492 
493 	memset(fl4, 0, sizeof(*fl4));
494 	fl4->daddr = (param->replyopts.opt.opt.srr ?
495 		      param->replyopts.opt.opt.faddr : iph->saddr);
496 	fl4->saddr = saddr;
497 	fl4->flowi4_mark = mark;
498 	fl4->flowi4_uid = sock_net_uid(net, NULL);
499 	fl4->flowi4_tos = RT_TOS(tos);
500 	fl4->flowi4_proto = IPPROTO_ICMP;
501 	fl4->fl4_icmp_type = type;
502 	fl4->fl4_icmp_code = code;
503 	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
504 	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
505 
506 	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
507 	rt = ip_route_output_key_hash(net, fl4, skb_in);
508 	if (IS_ERR(rt))
509 		return rt;
510 
511 	/* No need to clone since we're just using its address. */
512 	rt2 = rt;
513 
514 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
515 					   flowi4_to_flowi(fl4), NULL, 0);
516 	if (!IS_ERR(rt)) {
517 		if (rt != rt2)
518 			return rt;
519 	} else if (PTR_ERR(rt) == -EPERM) {
520 		rt = NULL;
521 	} else
522 		return rt;
523 
524 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
525 	if (err)
526 		goto relookup_failed;
527 
528 	if (inet_addr_type_dev_table(net, route_lookup_dev,
529 				     fl4_dec.saddr) == RTN_LOCAL) {
530 		rt2 = __ip_route_output_key(net, &fl4_dec);
531 		if (IS_ERR(rt2))
532 			err = PTR_ERR(rt2);
533 	} else {
534 		struct flowi4 fl4_2 = {};
535 		unsigned long orefdst;
536 
537 		fl4_2.daddr = fl4_dec.saddr;
538 		rt2 = ip_route_output_key(net, &fl4_2);
539 		if (IS_ERR(rt2)) {
540 			err = PTR_ERR(rt2);
541 			goto relookup_failed;
542 		}
543 		/* Ugh! */
544 		orefdst = skb_in->_skb_refdst; /* save old refdst */
545 		skb_dst_set(skb_in, NULL);
546 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
547 				     RT_TOS(tos), rt2->dst.dev);
548 
549 		dst_release(&rt2->dst);
550 		rt2 = skb_rtable(skb_in);
551 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
552 	}
553 
554 	if (err)
555 		goto relookup_failed;
556 
557 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
558 					    flowi4_to_flowi(&fl4_dec), NULL,
559 					    XFRM_LOOKUP_ICMP);
560 	if (!IS_ERR(rt2)) {
561 		dst_release(&rt->dst);
562 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
563 		rt = rt2;
564 	} else if (PTR_ERR(rt2) == -EPERM) {
565 		if (rt)
566 			dst_release(&rt->dst);
567 		return rt2;
568 	} else {
569 		err = PTR_ERR(rt2);
570 		goto relookup_failed;
571 	}
572 	return rt;
573 
574 relookup_failed:
575 	if (rt)
576 		return rt;
577 	return ERR_PTR(err);
578 }
579 
580 /*
581  *	Send an ICMP message in response to a situation
582  *
583  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
584  *		  MAY send more (we do).
585  *			MUST NOT change this header information.
586  *			MUST NOT reply to a multicast/broadcast IP address.
587  *			MUST NOT reply to a multicast/broadcast MAC address.
588  *			MUST reply to only the first fragment.
589  */
590 
591 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
592 		 const struct ip_options *opt)
593 {
594 	struct iphdr *iph;
595 	int room;
596 	struct icmp_bxm icmp_param;
597 	struct rtable *rt = skb_rtable(skb_in);
598 	struct ipcm_cookie ipc;
599 	struct flowi4 fl4;
600 	__be32 saddr;
601 	u8  tos;
602 	u32 mark;
603 	struct net *net;
604 	struct sock *sk;
605 
606 	if (!rt)
607 		goto out;
608 
609 	if (rt->dst.dev)
610 		net = dev_net(rt->dst.dev);
611 	else if (skb_in->dev)
612 		net = dev_net(skb_in->dev);
613 	else
614 		goto out;
615 
616 	/*
617 	 *	Find the original header. It is expected to be valid, of course.
618 	 *	Check this, icmp_send is called from the most obscure devices
619 	 *	sometimes.
620 	 */
621 	iph = ip_hdr(skb_in);
622 
623 	if ((u8 *)iph < skb_in->head ||
624 	    (skb_network_header(skb_in) + sizeof(*iph)) >
625 	    skb_tail_pointer(skb_in))
626 		goto out;
627 
628 	/*
629 	 *	No replies to physical multicast/broadcast
630 	 */
631 	if (skb_in->pkt_type != PACKET_HOST)
632 		goto out;
633 
634 	/*
635 	 *	Now check at the protocol level
636 	 */
637 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
638 		goto out;
639 
640 	/*
641 	 *	Only reply to fragment 0. We byte re-order the constant
642 	 *	mask for efficiency.
643 	 */
644 	if (iph->frag_off & htons(IP_OFFSET))
645 		goto out;
646 
647 	/*
648 	 *	If we send an ICMP error to an ICMP error a mess would result..
649 	 */
650 	if (icmp_pointers[type].error) {
651 		/*
652 		 *	We are an error, check if we are replying to an
653 		 *	ICMP error
654 		 */
655 		if (iph->protocol == IPPROTO_ICMP) {
656 			u8 _inner_type, *itp;
657 
658 			itp = skb_header_pointer(skb_in,
659 						 skb_network_header(skb_in) +
660 						 (iph->ihl << 2) +
661 						 offsetof(struct icmphdr,
662 							  type) -
663 						 skb_in->data,
664 						 sizeof(_inner_type),
665 						 &_inner_type);
666 			if (!itp)
667 				goto out;
668 
669 			/*
670 			 *	Assume any unknown ICMP type is an error. This
671 			 *	isn't specified by the RFC, but think about it..
672 			 */
673 			if (*itp > NR_ICMP_TYPES ||
674 			    icmp_pointers[*itp].error)
675 				goto out;
676 		}
677 	}
678 
679 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
680 	local_bh_disable();
681 
682 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
683 	 * incoming dev is loopback.  If outgoing dev change to not be
684 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
685 	 */
686 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
687 	      !icmpv4_global_allow(net, type, code))
688 		goto out_bh_enable;
689 
690 	sk = icmp_xmit_lock(net);
691 	if (!sk)
692 		goto out_bh_enable;
693 
694 	/*
695 	 *	Construct source address and options.
696 	 */
697 
698 	saddr = iph->daddr;
699 	if (!(rt->rt_flags & RTCF_LOCAL)) {
700 		struct net_device *dev = NULL;
701 
702 		rcu_read_lock();
703 		if (rt_is_input_route(rt) &&
704 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
705 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
706 
707 		if (dev)
708 			saddr = inet_select_addr(dev, iph->saddr,
709 						 RT_SCOPE_LINK);
710 		else
711 			saddr = 0;
712 		rcu_read_unlock();
713 	}
714 
715 	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
716 					   IPTOS_PREC_INTERNETCONTROL) :
717 					   iph->tos;
718 	mark = IP4_REPLY_MARK(net, skb_in->mark);
719 
720 	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
721 		goto out_unlock;
722 
723 
724 	/*
725 	 *	Prepare data for ICMP header.
726 	 */
727 
728 	icmp_param.data.icmph.type	 = type;
729 	icmp_param.data.icmph.code	 = code;
730 	icmp_param.data.icmph.un.gateway = info;
731 	icmp_param.data.icmph.checksum	 = 0;
732 	icmp_param.skb	  = skb_in;
733 	icmp_param.offset = skb_network_offset(skb_in);
734 	inet_sk(sk)->tos = tos;
735 	ipcm_init(&ipc);
736 	ipc.addr = iph->saddr;
737 	ipc.opt = &icmp_param.replyopts.opt;
738 	ipc.sockc.mark = mark;
739 
740 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
741 			       type, code, &icmp_param);
742 	if (IS_ERR(rt))
743 		goto out_unlock;
744 
745 	/* peer icmp_ratelimit */
746 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
747 		goto ende;
748 
749 	/* RFC says return as much as we can without exceeding 576 bytes. */
750 
751 	room = dst_mtu(&rt->dst);
752 	if (room > 576)
753 		room = 576;
754 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
755 	room -= sizeof(struct icmphdr);
756 
757 	icmp_param.data_len = skb_in->len - icmp_param.offset;
758 	if (icmp_param.data_len > room)
759 		icmp_param.data_len = room;
760 	icmp_param.head_len = sizeof(struct icmphdr);
761 
762 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
763 ende:
764 	ip_rt_put(rt);
765 out_unlock:
766 	icmp_xmit_unlock(sk);
767 out_bh_enable:
768 	local_bh_enable();
769 out:;
770 }
771 EXPORT_SYMBOL(__icmp_send);
772 
773 #if IS_ENABLED(CONFIG_NF_NAT)
774 #include <net/netfilter/nf_conntrack.h>
775 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
776 {
777 	struct sk_buff *cloned_skb = NULL;
778 	struct ip_options opts = { 0 };
779 	enum ip_conntrack_info ctinfo;
780 	struct nf_conn *ct;
781 	__be32 orig_ip;
782 
783 	ct = nf_ct_get(skb_in, &ctinfo);
784 	if (!ct || !(ct->status & IPS_SRC_NAT)) {
785 		__icmp_send(skb_in, type, code, info, &opts);
786 		return;
787 	}
788 
789 	if (skb_shared(skb_in))
790 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
791 
792 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
793 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
794 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
795 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
796 		goto out;
797 
798 	orig_ip = ip_hdr(skb_in)->saddr;
799 	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
800 	__icmp_send(skb_in, type, code, info, &opts);
801 	ip_hdr(skb_in)->saddr = orig_ip;
802 out:
803 	consume_skb(cloned_skb);
804 }
805 EXPORT_SYMBOL(icmp_ndo_send);
806 #endif
807 
808 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
809 {
810 	const struct iphdr *iph = (const struct iphdr *)skb->data;
811 	const struct net_protocol *ipprot;
812 	int protocol = iph->protocol;
813 
814 	/* Checkin full IP header plus 8 bytes of protocol to
815 	 * avoid additional coding at protocol handlers.
816 	 */
817 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
818 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
819 		return;
820 	}
821 
822 	raw_icmp_error(skb, protocol, info);
823 
824 	ipprot = rcu_dereference(inet_protos[protocol]);
825 	if (ipprot && ipprot->err_handler)
826 		ipprot->err_handler(skb, info);
827 }
828 
829 static bool icmp_tag_validation(int proto)
830 {
831 	bool ok;
832 
833 	rcu_read_lock();
834 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
835 	rcu_read_unlock();
836 	return ok;
837 }
838 
839 /*
840  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
841  *	ICMP_PARAMETERPROB.
842  */
843 
844 static bool icmp_unreach(struct sk_buff *skb)
845 {
846 	const struct iphdr *iph;
847 	struct icmphdr *icmph;
848 	struct net *net;
849 	u32 info = 0;
850 
851 	net = dev_net(skb_dst(skb)->dev);
852 
853 	/*
854 	 *	Incomplete header ?
855 	 * 	Only checks for the IP header, there should be an
856 	 *	additional check for longer headers in upper levels.
857 	 */
858 
859 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
860 		goto out_err;
861 
862 	icmph = icmp_hdr(skb);
863 	iph   = (const struct iphdr *)skb->data;
864 
865 	if (iph->ihl < 5) /* Mangled header, drop. */
866 		goto out_err;
867 
868 	switch (icmph->type) {
869 	case ICMP_DEST_UNREACH:
870 		switch (icmph->code & 15) {
871 		case ICMP_NET_UNREACH:
872 		case ICMP_HOST_UNREACH:
873 		case ICMP_PROT_UNREACH:
874 		case ICMP_PORT_UNREACH:
875 			break;
876 		case ICMP_FRAG_NEEDED:
877 			/* for documentation of the ip_no_pmtu_disc
878 			 * values please see
879 			 * Documentation/networking/ip-sysctl.rst
880 			 */
881 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
882 			default:
883 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
884 						    &iph->daddr);
885 				break;
886 			case 2:
887 				goto out;
888 			case 3:
889 				if (!icmp_tag_validation(iph->protocol))
890 					goto out;
891 				fallthrough;
892 			case 0:
893 				info = ntohs(icmph->un.frag.mtu);
894 			}
895 			break;
896 		case ICMP_SR_FAILED:
897 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
898 					    &iph->daddr);
899 			break;
900 		default:
901 			break;
902 		}
903 		if (icmph->code > NR_ICMP_UNREACH)
904 			goto out;
905 		break;
906 	case ICMP_PARAMETERPROB:
907 		info = ntohl(icmph->un.gateway) >> 24;
908 		break;
909 	case ICMP_TIME_EXCEEDED:
910 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
911 		if (icmph->code == ICMP_EXC_FRAGTIME)
912 			goto out;
913 		break;
914 	}
915 
916 	/*
917 	 *	Throw it at our lower layers
918 	 *
919 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
920 	 *		  header.
921 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
922 	 *		  transport layer.
923 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
924 	 *		  transport layer.
925 	 */
926 
927 	/*
928 	 *	Check the other end isn't violating RFC 1122. Some routers send
929 	 *	bogus responses to broadcast frames. If you see this message
930 	 *	first check your netmask matches at both ends, if it does then
931 	 *	get the other vendor to fix their kit.
932 	 */
933 
934 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
935 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
936 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
937 				     &ip_hdr(skb)->saddr,
938 				     icmph->type, icmph->code,
939 				     &iph->daddr, skb->dev->name);
940 		goto out;
941 	}
942 
943 	icmp_socket_deliver(skb, info);
944 
945 out:
946 	return true;
947 out_err:
948 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
949 	return false;
950 }
951 
952 
953 /*
954  *	Handle ICMP_REDIRECT.
955  */
956 
957 static bool icmp_redirect(struct sk_buff *skb)
958 {
959 	if (skb->len < sizeof(struct iphdr)) {
960 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
961 		return false;
962 	}
963 
964 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
965 		/* there aught to be a stat */
966 		return false;
967 	}
968 
969 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
970 	return true;
971 }
972 
973 /*
974  *	Handle ICMP_ECHO ("ping") requests.
975  *
976  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
977  *		  requests.
978  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
979  *		  included in the reply.
980  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
981  *		  echo requests, MUST have default=NOT.
982  *	See also WRT handling of options once they are done and working.
983  */
984 
985 static bool icmp_echo(struct sk_buff *skb)
986 {
987 	struct net *net;
988 
989 	net = dev_net(skb_dst(skb)->dev);
990 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
991 		struct icmp_bxm icmp_param;
992 
993 		icmp_param.data.icmph	   = *icmp_hdr(skb);
994 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
995 		icmp_param.skb		   = skb;
996 		icmp_param.offset	   = 0;
997 		icmp_param.data_len	   = skb->len;
998 		icmp_param.head_len	   = sizeof(struct icmphdr);
999 		icmp_reply(&icmp_param, skb);
1000 	}
1001 	/* should there be an ICMP stat for ignored echos? */
1002 	return true;
1003 }
1004 
1005 /*
1006  *	Handle ICMP Timestamp requests.
1007  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1008  *		  SHOULD be in the kernel for minimum random latency.
1009  *		  MUST be accurate to a few minutes.
1010  *		  MUST be updated at least at 15Hz.
1011  */
1012 static bool icmp_timestamp(struct sk_buff *skb)
1013 {
1014 	struct icmp_bxm icmp_param;
1015 	/*
1016 	 *	Too short.
1017 	 */
1018 	if (skb->len < 4)
1019 		goto out_err;
1020 
1021 	/*
1022 	 *	Fill in the current time as ms since midnight UT:
1023 	 */
1024 	icmp_param.data.times[1] = inet_current_timestamp();
1025 	icmp_param.data.times[2] = icmp_param.data.times[1];
1026 
1027 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1028 
1029 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1030 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1031 	icmp_param.data.icmph.code = 0;
1032 	icmp_param.skb		   = skb;
1033 	icmp_param.offset	   = 0;
1034 	icmp_param.data_len	   = 0;
1035 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1036 	icmp_reply(&icmp_param, skb);
1037 	return true;
1038 
1039 out_err:
1040 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1041 	return false;
1042 }
1043 
1044 static bool icmp_discard(struct sk_buff *skb)
1045 {
1046 	/* pretend it was a success */
1047 	return true;
1048 }
1049 
1050 /*
1051  *	Deal with incoming ICMP packets.
1052  */
1053 int icmp_rcv(struct sk_buff *skb)
1054 {
1055 	struct icmphdr *icmph;
1056 	struct rtable *rt = skb_rtable(skb);
1057 	struct net *net = dev_net(rt->dst.dev);
1058 	bool success;
1059 
1060 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1061 		struct sec_path *sp = skb_sec_path(skb);
1062 		int nh;
1063 
1064 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1065 				 XFRM_STATE_ICMP))
1066 			goto drop;
1067 
1068 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1069 			goto drop;
1070 
1071 		nh = skb_network_offset(skb);
1072 		skb_set_network_header(skb, sizeof(*icmph));
1073 
1074 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1075 			goto drop;
1076 
1077 		skb_set_network_header(skb, nh);
1078 	}
1079 
1080 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1081 
1082 	if (skb_checksum_simple_validate(skb))
1083 		goto csum_error;
1084 
1085 	if (!pskb_pull(skb, sizeof(*icmph)))
1086 		goto error;
1087 
1088 	icmph = icmp_hdr(skb);
1089 
1090 	ICMPMSGIN_INC_STATS(net, icmph->type);
1091 	/*
1092 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1093 	 *
1094 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1095 	 *		  discarded.
1096 	 */
1097 	if (icmph->type > NR_ICMP_TYPES)
1098 		goto error;
1099 
1100 
1101 	/*
1102 	 *	Parse the ICMP message
1103 	 */
1104 
1105 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1106 		/*
1107 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1108 		 *	  silently ignored (we let user decide with a sysctl).
1109 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1110 		 *	  discarded if to broadcast/multicast.
1111 		 */
1112 		if ((icmph->type == ICMP_ECHO ||
1113 		     icmph->type == ICMP_TIMESTAMP) &&
1114 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1115 			goto error;
1116 		}
1117 		if (icmph->type != ICMP_ECHO &&
1118 		    icmph->type != ICMP_TIMESTAMP &&
1119 		    icmph->type != ICMP_ADDRESS &&
1120 		    icmph->type != ICMP_ADDRESSREPLY) {
1121 			goto error;
1122 		}
1123 	}
1124 
1125 	success = icmp_pointers[icmph->type].handler(skb);
1126 
1127 	if (success)  {
1128 		consume_skb(skb);
1129 		return NET_RX_SUCCESS;
1130 	}
1131 
1132 drop:
1133 	kfree_skb(skb);
1134 	return NET_RX_DROP;
1135 csum_error:
1136 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1137 error:
1138 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1139 	goto drop;
1140 }
1141 
1142 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1143 {
1144 	struct icmp_extobj_hdr *objh, _objh;
1145 	struct icmp_ext_hdr *exth, _exth;
1146 	u16 olen;
1147 
1148 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1149 	if (!exth)
1150 		return false;
1151 	if (exth->version != 2)
1152 		return true;
1153 
1154 	if (exth->checksum &&
1155 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1156 		return false;
1157 
1158 	off += sizeof(_exth);
1159 	while (off < skb->len) {
1160 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1161 		if (!objh)
1162 			return false;
1163 
1164 		olen = ntohs(objh->length);
1165 		if (olen < sizeof(_objh))
1166 			return false;
1167 
1168 		off += olen;
1169 		if (off > skb->len)
1170 			return false;
1171 	}
1172 
1173 	return true;
1174 }
1175 
1176 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1177 			   struct sock_ee_data_rfc4884 *out,
1178 			   int thlen, int off)
1179 {
1180 	int hlen;
1181 
1182 	/* original datagram headers: end of icmph to payload (skb->data) */
1183 	hlen = -skb_transport_offset(skb) - thlen;
1184 
1185 	/* per rfc 4884: minimal datagram length of 128 bytes */
1186 	if (off < 128 || off < hlen)
1187 		return;
1188 
1189 	/* kernel has stripped headers: return payload offset in bytes */
1190 	off -= hlen;
1191 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1192 		return;
1193 
1194 	out->len = off;
1195 
1196 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1197 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1198 }
1199 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1200 
1201 int icmp_err(struct sk_buff *skb, u32 info)
1202 {
1203 	struct iphdr *iph = (struct iphdr *)skb->data;
1204 	int offset = iph->ihl<<2;
1205 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1206 	int type = icmp_hdr(skb)->type;
1207 	int code = icmp_hdr(skb)->code;
1208 	struct net *net = dev_net(skb->dev);
1209 
1210 	/*
1211 	 * Use ping_err to handle all icmp errors except those
1212 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1213 	 */
1214 	if (icmph->type != ICMP_ECHOREPLY) {
1215 		ping_err(skb, offset, info);
1216 		return 0;
1217 	}
1218 
1219 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1220 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1221 	else if (type == ICMP_REDIRECT)
1222 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1223 
1224 	return 0;
1225 }
1226 
1227 /*
1228  *	This table is the definition of how we handle ICMP.
1229  */
1230 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1231 	[ICMP_ECHOREPLY] = {
1232 		.handler = ping_rcv,
1233 	},
1234 	[1] = {
1235 		.handler = icmp_discard,
1236 		.error = 1,
1237 	},
1238 	[2] = {
1239 		.handler = icmp_discard,
1240 		.error = 1,
1241 	},
1242 	[ICMP_DEST_UNREACH] = {
1243 		.handler = icmp_unreach,
1244 		.error = 1,
1245 	},
1246 	[ICMP_SOURCE_QUENCH] = {
1247 		.handler = icmp_unreach,
1248 		.error = 1,
1249 	},
1250 	[ICMP_REDIRECT] = {
1251 		.handler = icmp_redirect,
1252 		.error = 1,
1253 	},
1254 	[6] = {
1255 		.handler = icmp_discard,
1256 		.error = 1,
1257 	},
1258 	[7] = {
1259 		.handler = icmp_discard,
1260 		.error = 1,
1261 	},
1262 	[ICMP_ECHO] = {
1263 		.handler = icmp_echo,
1264 	},
1265 	[9] = {
1266 		.handler = icmp_discard,
1267 		.error = 1,
1268 	},
1269 	[10] = {
1270 		.handler = icmp_discard,
1271 		.error = 1,
1272 	},
1273 	[ICMP_TIME_EXCEEDED] = {
1274 		.handler = icmp_unreach,
1275 		.error = 1,
1276 	},
1277 	[ICMP_PARAMETERPROB] = {
1278 		.handler = icmp_unreach,
1279 		.error = 1,
1280 	},
1281 	[ICMP_TIMESTAMP] = {
1282 		.handler = icmp_timestamp,
1283 	},
1284 	[ICMP_TIMESTAMPREPLY] = {
1285 		.handler = icmp_discard,
1286 	},
1287 	[ICMP_INFO_REQUEST] = {
1288 		.handler = icmp_discard,
1289 	},
1290 	[ICMP_INFO_REPLY] = {
1291 		.handler = icmp_discard,
1292 	},
1293 	[ICMP_ADDRESS] = {
1294 		.handler = icmp_discard,
1295 	},
1296 	[ICMP_ADDRESSREPLY] = {
1297 		.handler = icmp_discard,
1298 	},
1299 };
1300 
1301 static void __net_exit icmp_sk_exit(struct net *net)
1302 {
1303 	int i;
1304 
1305 	for_each_possible_cpu(i)
1306 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1307 	free_percpu(net->ipv4.icmp_sk);
1308 	net->ipv4.icmp_sk = NULL;
1309 }
1310 
1311 static int __net_init icmp_sk_init(struct net *net)
1312 {
1313 	int i, err;
1314 
1315 	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1316 	if (!net->ipv4.icmp_sk)
1317 		return -ENOMEM;
1318 
1319 	for_each_possible_cpu(i) {
1320 		struct sock *sk;
1321 
1322 		err = inet_ctl_sock_create(&sk, PF_INET,
1323 					   SOCK_RAW, IPPROTO_ICMP, net);
1324 		if (err < 0)
1325 			goto fail;
1326 
1327 		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1328 
1329 		/* Enough space for 2 64K ICMP packets, including
1330 		 * sk_buff/skb_shared_info struct overhead.
1331 		 */
1332 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1333 
1334 		/*
1335 		 * Speedup sock_wfree()
1336 		 */
1337 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1338 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1339 	}
1340 
1341 	/* Control parameters for ECHO replies. */
1342 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1343 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1344 
1345 	/* Control parameter - ignore bogus broadcast responses? */
1346 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1347 
1348 	/*
1349 	 * 	Configurable global rate limit.
1350 	 *
1351 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1352 	 *	bucket ratemask defines which icmp types are ratelimited by
1353 	 *	setting	it's bit position.
1354 	 *
1355 	 *	default:
1356 	 *	dest unreachable (3), source quench (4),
1357 	 *	time exceeded (11), parameter problem (12)
1358 	 */
1359 
1360 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1361 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1362 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1363 
1364 	return 0;
1365 
1366 fail:
1367 	icmp_sk_exit(net);
1368 	return err;
1369 }
1370 
1371 static struct pernet_operations __net_initdata icmp_sk_ops = {
1372        .init = icmp_sk_init,
1373        .exit = icmp_sk_exit,
1374 };
1375 
1376 int __init icmp_init(void)
1377 {
1378 	return register_pernet_subsys(&icmp_sk_ops);
1379 }
1380