xref: /linux/net/ipv4/icmp.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 #include <net/addrconf.h>
96 #include <net/inet_dscp.h>
97 #define CREATE_TRACE_POINTS
98 #include <trace/events/icmp.h>
99 
100 /*
101  *	Build xmit assembly blocks
102  */
103 
104 struct icmp_bxm {
105 	struct sk_buff *skb;
106 	int offset;
107 	int data_len;
108 
109 	struct {
110 		struct icmphdr icmph;
111 		__be32	       times[3];
112 	} data;
113 	int head_len;
114 	struct ip_options_data replyopts;
115 };
116 
117 /* An array of errno for error messages from dest unreach. */
118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
119 
120 const struct icmp_err icmp_err_convert[] = {
121 	{
122 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
123 		.fatal = 0,
124 	},
125 	{
126 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
127 		.fatal = 0,
128 	},
129 	{
130 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
131 		.fatal = 1,
132 	},
133 	{
134 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
135 		.fatal = 1,
136 	},
137 	{
138 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
139 		.fatal = 0,
140 	},
141 	{
142 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
143 		.fatal = 0,
144 	},
145 	{
146 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
147 		.fatal = 1,
148 	},
149 	{
150 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
163 		.fatal = 1,
164 	},
165 	{
166 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
167 		.fatal = 0,
168 	},
169 	{
170 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
171 		.fatal = 0,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
175 		.fatal = 1,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
179 		.fatal = 1,
180 	},
181 	{
182 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
183 		.fatal = 1,
184 	},
185 };
186 EXPORT_SYMBOL(icmp_err_convert);
187 
188 /*
189  *	ICMP control array. This specifies what to do with each ICMP.
190  */
191 
192 struct icmp_control {
193 	enum skb_drop_reason (*handler)(struct sk_buff *skb);
194 	short   error;		/* This ICMP is classed as an error message */
195 };
196 
197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
198 
199 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
200 
201 /* Called with BH disabled */
202 static inline struct sock *icmp_xmit_lock(struct net *net)
203 {
204 	struct sock *sk;
205 
206 	sk = this_cpu_read(ipv4_icmp_sk);
207 
208 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
209 		/* This can happen if the output path signals a
210 		 * dst_link_failure() for an outgoing ICMP packet.
211 		 */
212 		return NULL;
213 	}
214 	sock_net_set(sk, net);
215 	return sk;
216 }
217 
218 static inline void icmp_xmit_unlock(struct sock *sk)
219 {
220 	sock_net_set(sk, &init_net);
221 	spin_unlock(&sk->sk_lock.slock);
222 }
223 
224 /**
225  * icmp_global_allow - Are we allowed to send one more ICMP message ?
226  * @net: network namespace
227  *
228  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
229  * Returns false if we reached the limit and can not send another packet.
230  * Works in tandem with icmp_global_consume().
231  */
232 bool icmp_global_allow(struct net *net)
233 {
234 	u32 delta, now, oldstamp;
235 	int incr, new, old;
236 
237 	/* Note: many cpus could find this condition true.
238 	 * Then later icmp_global_consume() could consume more credits,
239 	 * this is an acceptable race.
240 	 */
241 	if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
242 		return true;
243 
244 	now = jiffies;
245 	oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
246 	delta = min_t(u32, now - oldstamp, HZ);
247 	if (delta < HZ / 50)
248 		return false;
249 
250 	incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ;
251 	if (!incr)
252 		return false;
253 
254 	if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
255 		old = atomic_read(&net->ipv4.icmp_global_credit);
256 		do {
257 			new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
258 		} while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
259 	}
260 	return true;
261 }
262 EXPORT_SYMBOL(icmp_global_allow);
263 
264 void icmp_global_consume(struct net *net)
265 {
266 	int credits = get_random_u32_below(3);
267 
268 	/* Note: this might make icmp_global.credit negative. */
269 	if (credits)
270 		atomic_sub(credits, &net->ipv4.icmp_global_credit);
271 }
272 EXPORT_SYMBOL(icmp_global_consume);
273 
274 static bool icmpv4_mask_allow(struct net *net, int type, int code)
275 {
276 	if (type > NR_ICMP_TYPES)
277 		return true;
278 
279 	/* Don't limit PMTU discovery. */
280 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
281 		return true;
282 
283 	/* Limit if icmp type is enabled in ratemask. */
284 	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
285 		return true;
286 
287 	return false;
288 }
289 
290 static bool icmpv4_global_allow(struct net *net, int type, int code,
291 				bool *apply_ratelimit)
292 {
293 	if (icmpv4_mask_allow(net, type, code))
294 		return true;
295 
296 	if (icmp_global_allow(net)) {
297 		*apply_ratelimit = true;
298 		return true;
299 	}
300 	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
301 	return false;
302 }
303 
304 /*
305  *	Send an ICMP frame.
306  */
307 
308 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
309 			       struct flowi4 *fl4, int type, int code,
310 			       bool apply_ratelimit)
311 {
312 	struct dst_entry *dst = &rt->dst;
313 	struct inet_peer *peer;
314 	bool rc = true;
315 	int vif;
316 
317 	if (!apply_ratelimit)
318 		return true;
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,
327 				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
328 	if (peer)
329 		inet_putpeer(peer);
330 out:
331 	if (!rc)
332 		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
333 	else
334 		icmp_global_consume(net);
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 = from;
355 	__wsum csum;
356 
357 	csum = skb_copy_and_csum_bits(icmp_param->skb,
358 				      icmp_param->offset + offset,
359 				      to, len);
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 sock *sk,
368 			    struct icmp_bxm *icmp_param,
369 			    struct flowi4 *fl4,
370 			    struct ipcm_cookie *ipc, struct rtable **rt)
371 {
372 	struct sk_buff *skb;
373 
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 	bool apply_ratelimit = false;
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 icmpv4_global_allow and icmp_xmit_lock */
419 	local_bh_disable();
420 
421 	/* is global icmp_msgs_per_sec exhausted ? */
422 	if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
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 	ipcm_init(&ipc);
433 	inet->tos = ip_hdr(skb)->tos;
434 	ipc.sockc.mark = mark;
435 	daddr = ipc.addr = ip_hdr(skb)->saddr;
436 	saddr = fib_compute_spec_dst(skb);
437 
438 	if (icmp_param->replyopts.opt.opt.optlen) {
439 		ipc.opt = &icmp_param->replyopts.opt;
440 		if (ipc.opt->opt.srr)
441 			daddr = icmp_param->replyopts.opt.opt.faddr;
442 	}
443 	memset(&fl4, 0, sizeof(fl4));
444 	fl4.daddr = daddr;
445 	fl4.saddr = saddr;
446 	fl4.flowi4_mark = mark;
447 	fl4.flowi4_uid = sock_net_uid(net, NULL);
448 	fl4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb)));
449 	fl4.flowi4_proto = IPPROTO_ICMP;
450 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
451 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
452 	rt = ip_route_output_key(net, &fl4);
453 	if (IS_ERR(rt))
454 		goto out_unlock;
455 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
456 		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
457 	ip_rt_put(rt);
458 out_unlock:
459 	icmp_xmit_unlock(sk);
460 out_bh_enable:
461 	local_bh_enable();
462 }
463 
464 /*
465  * The device used for looking up which routing table to use for sending an ICMP
466  * error is preferably the source whenever it is set, which should ensure the
467  * icmp error can be sent to the source host, else lookup using the routing
468  * table of the destination device, else use the main routing table (index 0).
469  */
470 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
471 {
472 	struct net_device *route_lookup_dev = NULL;
473 
474 	if (skb->dev)
475 		route_lookup_dev = skb->dev;
476 	else if (skb_dst(skb))
477 		route_lookup_dev = skb_dst(skb)->dev;
478 	return route_lookup_dev;
479 }
480 
481 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4,
482 					struct sk_buff *skb_in,
483 					const struct iphdr *iph, __be32 saddr,
484 					dscp_t dscp, u32 mark, int type,
485 					int code, struct icmp_bxm *param)
486 {
487 	struct net_device *route_lookup_dev;
488 	struct dst_entry *dst, *dst2;
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 = inet_dscp_to_dsfield(dscp);
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 	dst = xfrm_lookup(net, &rt->dst,
515 			  flowi4_to_flowi(fl4), NULL, 0);
516 	rt = dst_rtable(dst);
517 	if (!IS_ERR(dst)) {
518 		if (rt != rt2)
519 			return rt;
520 	} else if (PTR_ERR(dst) == -EPERM) {
521 		rt = NULL;
522 	} else {
523 		return rt;
524 	}
525 	err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
526 	if (err)
527 		goto relookup_failed;
528 
529 	if (inet_addr_type_dev_table(net, route_lookup_dev,
530 				     fl4_dec.saddr) == RTN_LOCAL) {
531 		rt2 = __ip_route_output_key(net, &fl4_dec);
532 		if (IS_ERR(rt2))
533 			err = PTR_ERR(rt2);
534 	} else {
535 		struct flowi4 fl4_2 = {};
536 		unsigned long orefdst;
537 
538 		fl4_2.daddr = fl4_dec.saddr;
539 		rt2 = ip_route_output_key(net, &fl4_2);
540 		if (IS_ERR(rt2)) {
541 			err = PTR_ERR(rt2);
542 			goto relookup_failed;
543 		}
544 		/* Ugh! */
545 		orefdst = skb_in->_skb_refdst; /* save old refdst */
546 		skb_dst_set(skb_in, NULL);
547 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
548 				     dscp, rt2->dst.dev) ? -EINVAL : 0;
549 
550 		dst_release(&rt2->dst);
551 		rt2 = skb_rtable(skb_in);
552 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
553 	}
554 
555 	if (err)
556 		goto relookup_failed;
557 
558 	dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
559 			   XFRM_LOOKUP_ICMP);
560 	rt2 = dst_rtable(dst2);
561 	if (!IS_ERR(dst2)) {
562 		dst_release(&rt->dst);
563 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
564 		rt = rt2;
565 	} else if (PTR_ERR(dst2) == -EPERM) {
566 		if (rt)
567 			dst_release(&rt->dst);
568 		return rt2;
569 	} else {
570 		err = PTR_ERR(dst2);
571 		goto relookup_failed;
572 	}
573 	return rt;
574 
575 relookup_failed:
576 	if (rt)
577 		return rt;
578 	return ERR_PTR(err);
579 }
580 
581 /*
582  *	Send an ICMP message in response to a situation
583  *
584  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
585  *		  MAY send more (we do).
586  *			MUST NOT change this header information.
587  *			MUST NOT reply to a multicast/broadcast IP address.
588  *			MUST NOT reply to a multicast/broadcast MAC address.
589  *			MUST reply to only the first fragment.
590  */
591 
592 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
593 		 const struct ip_options *opt)
594 {
595 	struct iphdr *iph;
596 	int room;
597 	struct icmp_bxm icmp_param;
598 	struct rtable *rt = skb_rtable(skb_in);
599 	bool apply_ratelimit = false;
600 	struct ipcm_cookie ipc;
601 	struct flowi4 fl4;
602 	__be32 saddr;
603 	u8  tos;
604 	u32 mark;
605 	struct net *net;
606 	struct sock *sk;
607 
608 	if (!rt)
609 		goto out;
610 
611 	if (rt->dst.dev)
612 		net = dev_net(rt->dst.dev);
613 	else if (skb_in->dev)
614 		net = dev_net(skb_in->dev);
615 	else
616 		goto out;
617 
618 	/*
619 	 *	Find the original header. It is expected to be valid, of course.
620 	 *	Check this, icmp_send is called from the most obscure devices
621 	 *	sometimes.
622 	 */
623 	iph = ip_hdr(skb_in);
624 
625 	if ((u8 *)iph < skb_in->head ||
626 	    (skb_network_header(skb_in) + sizeof(*iph)) >
627 	    skb_tail_pointer(skb_in))
628 		goto out;
629 
630 	/*
631 	 *	No replies to physical multicast/broadcast
632 	 */
633 	if (skb_in->pkt_type != PACKET_HOST)
634 		goto out;
635 
636 	/*
637 	 *	Now check at the protocol level
638 	 */
639 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
640 		goto out;
641 
642 	/*
643 	 *	Only reply to fragment 0. We byte re-order the constant
644 	 *	mask for efficiency.
645 	 */
646 	if (iph->frag_off & htons(IP_OFFSET))
647 		goto out;
648 
649 	/*
650 	 *	If we send an ICMP error to an ICMP error a mess would result..
651 	 */
652 	if (icmp_pointers[type].error) {
653 		/*
654 		 *	We are an error, check if we are replying to an
655 		 *	ICMP error
656 		 */
657 		if (iph->protocol == IPPROTO_ICMP) {
658 			u8 _inner_type, *itp;
659 
660 			itp = skb_header_pointer(skb_in,
661 						 skb_network_header(skb_in) +
662 						 (iph->ihl << 2) +
663 						 offsetof(struct icmphdr,
664 							  type) -
665 						 skb_in->data,
666 						 sizeof(_inner_type),
667 						 &_inner_type);
668 			if (!itp)
669 				goto out;
670 
671 			/*
672 			 *	Assume any unknown ICMP type is an error. This
673 			 *	isn't specified by the RFC, but think about it..
674 			 */
675 			if (*itp > NR_ICMP_TYPES ||
676 			    icmp_pointers[*itp].error)
677 				goto out;
678 		}
679 	}
680 
681 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
682 	local_bh_disable();
683 
684 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
685 	 * incoming dev is loopback.  If outgoing dev change to not be
686 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
687 	 */
688 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
689 	      !icmpv4_global_allow(net, type, code, &apply_ratelimit))
690 		goto out_bh_enable;
691 
692 	sk = icmp_xmit_lock(net);
693 	if (!sk)
694 		goto out_bh_enable;
695 
696 	/*
697 	 *	Construct source address and options.
698 	 */
699 
700 	saddr = iph->daddr;
701 	if (!(rt->rt_flags & RTCF_LOCAL)) {
702 		struct net_device *dev = NULL;
703 
704 		rcu_read_lock();
705 		if (rt_is_input_route(rt) &&
706 		    READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
707 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
708 
709 		if (dev)
710 			saddr = inet_select_addr(dev, iph->saddr,
711 						 RT_SCOPE_LINK);
712 		else
713 			saddr = 0;
714 		rcu_read_unlock();
715 	}
716 
717 	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
718 					   IPTOS_PREC_INTERNETCONTROL) :
719 					   iph->tos;
720 	mark = IP4_REPLY_MARK(net, skb_in->mark);
721 
722 	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
723 		goto out_unlock;
724 
725 
726 	/*
727 	 *	Prepare data for ICMP header.
728 	 */
729 
730 	icmp_param.data.icmph.type	 = type;
731 	icmp_param.data.icmph.code	 = code;
732 	icmp_param.data.icmph.un.gateway = info;
733 	icmp_param.data.icmph.checksum	 = 0;
734 	icmp_param.skb	  = skb_in;
735 	icmp_param.offset = skb_network_offset(skb_in);
736 	inet_sk(sk)->tos = tos;
737 	ipcm_init(&ipc);
738 	ipc.addr = iph->saddr;
739 	ipc.opt = &icmp_param.replyopts.opt;
740 	ipc.sockc.mark = mark;
741 
742 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
743 			       inet_dsfield_to_dscp(tos), mark, type, code,
744 			       &icmp_param);
745 	if (IS_ERR(rt))
746 		goto out_unlock;
747 
748 	/* peer icmp_ratelimit */
749 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
750 		goto ende;
751 
752 	/* RFC says return as much as we can without exceeding 576 bytes. */
753 
754 	room = dst_mtu(&rt->dst);
755 	if (room > 576)
756 		room = 576;
757 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
758 	room -= sizeof(struct icmphdr);
759 	/* Guard against tiny mtu. We need to include at least one
760 	 * IP network header for this message to make any sense.
761 	 */
762 	if (room <= (int)sizeof(struct iphdr))
763 		goto ende;
764 
765 	icmp_param.data_len = skb_in->len - icmp_param.offset;
766 	if (icmp_param.data_len > room)
767 		icmp_param.data_len = room;
768 	icmp_param.head_len = sizeof(struct icmphdr);
769 
770 	/* if we don't have a source address at this point, fall back to the
771 	 * dummy address instead of sending out a packet with a source address
772 	 * of 0.0.0.0
773 	 */
774 	if (!fl4.saddr)
775 		fl4.saddr = htonl(INADDR_DUMMY);
776 
777 	trace_icmp_send(skb_in, type, code);
778 
779 	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
780 ende:
781 	ip_rt_put(rt);
782 out_unlock:
783 	icmp_xmit_unlock(sk);
784 out_bh_enable:
785 	local_bh_enable();
786 out:;
787 }
788 EXPORT_SYMBOL(__icmp_send);
789 
790 #if IS_ENABLED(CONFIG_NF_NAT)
791 #include <net/netfilter/nf_conntrack.h>
792 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
793 {
794 	struct sk_buff *cloned_skb = NULL;
795 	struct ip_options opts = { 0 };
796 	enum ip_conntrack_info ctinfo;
797 	struct nf_conn *ct;
798 	__be32 orig_ip;
799 
800 	ct = nf_ct_get(skb_in, &ctinfo);
801 	if (!ct || !(ct->status & IPS_SRC_NAT)) {
802 		__icmp_send(skb_in, type, code, info, &opts);
803 		return;
804 	}
805 
806 	if (skb_shared(skb_in))
807 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
808 
809 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
810 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
811 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
812 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
813 		goto out;
814 
815 	orig_ip = ip_hdr(skb_in)->saddr;
816 	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
817 	__icmp_send(skb_in, type, code, info, &opts);
818 	ip_hdr(skb_in)->saddr = orig_ip;
819 out:
820 	consume_skb(cloned_skb);
821 }
822 EXPORT_SYMBOL(icmp_ndo_send);
823 #endif
824 
825 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
826 {
827 	const struct iphdr *iph = (const struct iphdr *)skb->data;
828 	const struct net_protocol *ipprot;
829 	int protocol = iph->protocol;
830 
831 	/* Checkin full IP header plus 8 bytes of protocol to
832 	 * avoid additional coding at protocol handlers.
833 	 */
834 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
835 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
836 		return;
837 	}
838 
839 	raw_icmp_error(skb, protocol, info);
840 
841 	ipprot = rcu_dereference(inet_protos[protocol]);
842 	if (ipprot && ipprot->err_handler)
843 		ipprot->err_handler(skb, info);
844 }
845 
846 static bool icmp_tag_validation(int proto)
847 {
848 	bool ok;
849 
850 	rcu_read_lock();
851 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
852 	rcu_read_unlock();
853 	return ok;
854 }
855 
856 /*
857  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
858  *	ICMP_PARAMETERPROB.
859  */
860 
861 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
862 {
863 	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
864 	const struct iphdr *iph;
865 	struct icmphdr *icmph;
866 	struct net *net;
867 	u32 info = 0;
868 
869 	net = dev_net(skb_dst(skb)->dev);
870 
871 	/*
872 	 *	Incomplete header ?
873 	 * 	Only checks for the IP header, there should be an
874 	 *	additional check for longer headers in upper levels.
875 	 */
876 
877 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
878 		goto out_err;
879 
880 	icmph = icmp_hdr(skb);
881 	iph   = (const struct iphdr *)skb->data;
882 
883 	if (iph->ihl < 5)  { /* Mangled header, drop. */
884 		reason = SKB_DROP_REASON_IP_INHDR;
885 		goto out_err;
886 	}
887 
888 	switch (icmph->type) {
889 	case ICMP_DEST_UNREACH:
890 		switch (icmph->code & 15) {
891 		case ICMP_NET_UNREACH:
892 		case ICMP_HOST_UNREACH:
893 		case ICMP_PROT_UNREACH:
894 		case ICMP_PORT_UNREACH:
895 			break;
896 		case ICMP_FRAG_NEEDED:
897 			/* for documentation of the ip_no_pmtu_disc
898 			 * values please see
899 			 * Documentation/networking/ip-sysctl.rst
900 			 */
901 			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
902 			default:
903 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
904 						    &iph->daddr);
905 				break;
906 			case 2:
907 				goto out;
908 			case 3:
909 				if (!icmp_tag_validation(iph->protocol))
910 					goto out;
911 				fallthrough;
912 			case 0:
913 				info = ntohs(icmph->un.frag.mtu);
914 			}
915 			break;
916 		case ICMP_SR_FAILED:
917 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
918 					    &iph->daddr);
919 			break;
920 		default:
921 			break;
922 		}
923 		if (icmph->code > NR_ICMP_UNREACH)
924 			goto out;
925 		break;
926 	case ICMP_PARAMETERPROB:
927 		info = ntohl(icmph->un.gateway) >> 24;
928 		break;
929 	case ICMP_TIME_EXCEEDED:
930 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
931 		if (icmph->code == ICMP_EXC_FRAGTIME)
932 			goto out;
933 		break;
934 	}
935 
936 	/*
937 	 *	Throw it at our lower layers
938 	 *
939 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
940 	 *		  header.
941 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
942 	 *		  transport layer.
943 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
944 	 *		  transport layer.
945 	 */
946 
947 	/*
948 	 *	Check the other end isn't violating RFC 1122. Some routers send
949 	 *	bogus responses to broadcast frames. If you see this message
950 	 *	first check your netmask matches at both ends, if it does then
951 	 *	get the other vendor to fix their kit.
952 	 */
953 
954 	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
955 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
956 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
957 				     &ip_hdr(skb)->saddr,
958 				     icmph->type, icmph->code,
959 				     &iph->daddr, skb->dev->name);
960 		goto out;
961 	}
962 
963 	icmp_socket_deliver(skb, info);
964 
965 out:
966 	return reason;
967 out_err:
968 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
969 	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
970 }
971 
972 
973 /*
974  *	Handle ICMP_REDIRECT.
975  */
976 
977 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
978 {
979 	if (skb->len < sizeof(struct iphdr)) {
980 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
981 		return SKB_DROP_REASON_PKT_TOO_SMALL;
982 	}
983 
984 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
985 		/* there aught to be a stat */
986 		return SKB_DROP_REASON_NOMEM;
987 	}
988 
989 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
990 	return SKB_NOT_DROPPED_YET;
991 }
992 
993 /*
994  *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
995  *
996  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
997  *		  requests.
998  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
999  *		  included in the reply.
1000  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1001  *		  echo requests, MUST have default=NOT.
1002  *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
1003  *		  Extended Echo Functionality, MUST be disabled by default
1004  *	See also WRT handling of options once they are done and working.
1005  */
1006 
1007 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1008 {
1009 	struct icmp_bxm icmp_param;
1010 	struct net *net;
1011 
1012 	net = dev_net(skb_dst(skb)->dev);
1013 	/* should there be an ICMP stat for ignored echos? */
1014 	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1015 		return SKB_NOT_DROPPED_YET;
1016 
1017 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1018 	icmp_param.skb		   = skb;
1019 	icmp_param.offset	   = 0;
1020 	icmp_param.data_len	   = skb->len;
1021 	icmp_param.head_len	   = sizeof(struct icmphdr);
1022 
1023 	if (icmp_param.data.icmph.type == ICMP_ECHO)
1024 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1025 	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1026 		return SKB_NOT_DROPPED_YET;
1027 
1028 	icmp_reply(&icmp_param, skb);
1029 	return SKB_NOT_DROPPED_YET;
1030 }
1031 
1032 /*	Helper for icmp_echo and icmpv6_echo_reply.
1033  *	Searches for net_device that matches PROBE interface identifier
1034  *		and builds PROBE reply message in icmphdr.
1035  *
1036  *	Returns false if PROBE responses are disabled via sysctl
1037  */
1038 
1039 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1040 {
1041 	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1042 	struct icmp_ext_echo_iio *iio, _iio;
1043 	struct net *net = dev_net(skb->dev);
1044 	struct inet6_dev *in6_dev;
1045 	struct in_device *in_dev;
1046 	struct net_device *dev;
1047 	char buff[IFNAMSIZ];
1048 	u16 ident_len;
1049 	u8 status;
1050 
1051 	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1052 		return false;
1053 
1054 	/* We currently only support probing interfaces on the proxy node
1055 	 * Check to ensure L-bit is set
1056 	 */
1057 	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1058 		return false;
1059 	/* Clear status bits in reply message */
1060 	icmphdr->un.echo.sequence &= htons(0xFF00);
1061 	if (icmphdr->type == ICMP_EXT_ECHO)
1062 		icmphdr->type = ICMP_EXT_ECHOREPLY;
1063 	else
1064 		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1065 	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1066 	/* Size of iio is class_type dependent.
1067 	 * Only check header here and assign length based on ctype in the switch statement
1068 	 */
1069 	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1070 	if (!ext_hdr || !iio)
1071 		goto send_mal_query;
1072 	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1073 	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1074 		goto send_mal_query;
1075 	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1076 	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1077 				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1078 	if (!iio)
1079 		goto send_mal_query;
1080 
1081 	status = 0;
1082 	dev = NULL;
1083 	switch (iio->extobj_hdr.class_type) {
1084 	case ICMP_EXT_ECHO_CTYPE_NAME:
1085 		if (ident_len >= IFNAMSIZ)
1086 			goto send_mal_query;
1087 		memset(buff, 0, sizeof(buff));
1088 		memcpy(buff, &iio->ident.name, ident_len);
1089 		dev = dev_get_by_name(net, buff);
1090 		break;
1091 	case ICMP_EXT_ECHO_CTYPE_INDEX:
1092 		if (ident_len != sizeof(iio->ident.ifindex))
1093 			goto send_mal_query;
1094 		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1095 		break;
1096 	case ICMP_EXT_ECHO_CTYPE_ADDR:
1097 		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1098 		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1099 				 iio->ident.addr.ctype3_hdr.addrlen)
1100 			goto send_mal_query;
1101 		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1102 		case ICMP_AFI_IP:
1103 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1104 				goto send_mal_query;
1105 			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1106 			break;
1107 #if IS_ENABLED(CONFIG_IPV6)
1108 		case ICMP_AFI_IP6:
1109 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1110 				goto send_mal_query;
1111 			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1112 			dev_hold(dev);
1113 			break;
1114 #endif
1115 		default:
1116 			goto send_mal_query;
1117 		}
1118 		break;
1119 	default:
1120 		goto send_mal_query;
1121 	}
1122 	if (!dev) {
1123 		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1124 		return true;
1125 	}
1126 	/* Fill bits in reply message */
1127 	if (dev->flags & IFF_UP)
1128 		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1129 
1130 	in_dev = __in_dev_get_rcu(dev);
1131 	if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1132 		status |= ICMP_EXT_ECHOREPLY_IPV4;
1133 
1134 	in6_dev = __in6_dev_get(dev);
1135 	if (in6_dev && !list_empty(&in6_dev->addr_list))
1136 		status |= ICMP_EXT_ECHOREPLY_IPV6;
1137 
1138 	dev_put(dev);
1139 	icmphdr->un.echo.sequence |= htons(status);
1140 	return true;
1141 send_mal_query:
1142 	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1143 	return true;
1144 }
1145 EXPORT_SYMBOL_GPL(icmp_build_probe);
1146 
1147 /*
1148  *	Handle ICMP Timestamp requests.
1149  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1150  *		  SHOULD be in the kernel for minimum random latency.
1151  *		  MUST be accurate to a few minutes.
1152  *		  MUST be updated at least at 15Hz.
1153  */
1154 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1155 {
1156 	struct icmp_bxm icmp_param;
1157 	/*
1158 	 *	Too short.
1159 	 */
1160 	if (skb->len < 4)
1161 		goto out_err;
1162 
1163 	/*
1164 	 *	Fill in the current time as ms since midnight UT:
1165 	 */
1166 	icmp_param.data.times[1] = inet_current_timestamp();
1167 	icmp_param.data.times[2] = icmp_param.data.times[1];
1168 
1169 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1170 
1171 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1172 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1173 	icmp_param.data.icmph.code = 0;
1174 	icmp_param.skb		   = skb;
1175 	icmp_param.offset	   = 0;
1176 	icmp_param.data_len	   = 0;
1177 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1178 	icmp_reply(&icmp_param, skb);
1179 	return SKB_NOT_DROPPED_YET;
1180 
1181 out_err:
1182 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1183 	return SKB_DROP_REASON_PKT_TOO_SMALL;
1184 }
1185 
1186 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1187 {
1188 	/* pretend it was a success */
1189 	return SKB_NOT_DROPPED_YET;
1190 }
1191 
1192 /*
1193  *	Deal with incoming ICMP packets.
1194  */
1195 int icmp_rcv(struct sk_buff *skb)
1196 {
1197 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1198 	struct rtable *rt = skb_rtable(skb);
1199 	struct net *net = dev_net(rt->dst.dev);
1200 	struct icmphdr *icmph;
1201 
1202 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1203 		struct sec_path *sp = skb_sec_path(skb);
1204 		int nh;
1205 
1206 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1207 				 XFRM_STATE_ICMP)) {
1208 			reason = SKB_DROP_REASON_XFRM_POLICY;
1209 			goto drop;
1210 		}
1211 
1212 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1213 			goto drop;
1214 
1215 		nh = skb_network_offset(skb);
1216 		skb_set_network_header(skb, sizeof(*icmph));
1217 
1218 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1219 						skb)) {
1220 			reason = SKB_DROP_REASON_XFRM_POLICY;
1221 			goto drop;
1222 		}
1223 
1224 		skb_set_network_header(skb, nh);
1225 	}
1226 
1227 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1228 
1229 	if (skb_checksum_simple_validate(skb))
1230 		goto csum_error;
1231 
1232 	if (!pskb_pull(skb, sizeof(*icmph)))
1233 		goto error;
1234 
1235 	icmph = icmp_hdr(skb);
1236 
1237 	ICMPMSGIN_INC_STATS(net, icmph->type);
1238 
1239 	/* Check for ICMP Extended Echo (PROBE) messages */
1240 	if (icmph->type == ICMP_EXT_ECHO) {
1241 		/* We can't use icmp_pointers[].handler() because it is an array of
1242 		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1243 		 */
1244 		reason = icmp_echo(skb);
1245 		goto reason_check;
1246 	}
1247 
1248 	if (icmph->type == ICMP_EXT_ECHOREPLY) {
1249 		reason = ping_rcv(skb);
1250 		goto reason_check;
1251 	}
1252 
1253 	/*
1254 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1255 	 *
1256 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1257 	 *		  discarded.
1258 	 */
1259 	if (icmph->type > NR_ICMP_TYPES) {
1260 		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1261 		goto error;
1262 	}
1263 
1264 	/*
1265 	 *	Parse the ICMP message
1266 	 */
1267 
1268 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1269 		/*
1270 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1271 		 *	  silently ignored (we let user decide with a sysctl).
1272 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1273 		 *	  discarded if to broadcast/multicast.
1274 		 */
1275 		if ((icmph->type == ICMP_ECHO ||
1276 		     icmph->type == ICMP_TIMESTAMP) &&
1277 		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1278 			reason = SKB_DROP_REASON_INVALID_PROTO;
1279 			goto error;
1280 		}
1281 		if (icmph->type != ICMP_ECHO &&
1282 		    icmph->type != ICMP_TIMESTAMP &&
1283 		    icmph->type != ICMP_ADDRESS &&
1284 		    icmph->type != ICMP_ADDRESSREPLY) {
1285 			reason = SKB_DROP_REASON_INVALID_PROTO;
1286 			goto error;
1287 		}
1288 	}
1289 
1290 	reason = icmp_pointers[icmph->type].handler(skb);
1291 reason_check:
1292 	if (!reason)  {
1293 		consume_skb(skb);
1294 		return NET_RX_SUCCESS;
1295 	}
1296 
1297 drop:
1298 	kfree_skb_reason(skb, reason);
1299 	return NET_RX_DROP;
1300 csum_error:
1301 	reason = SKB_DROP_REASON_ICMP_CSUM;
1302 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1303 error:
1304 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1305 	goto drop;
1306 }
1307 
1308 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1309 {
1310 	struct icmp_extobj_hdr *objh, _objh;
1311 	struct icmp_ext_hdr *exth, _exth;
1312 	u16 olen;
1313 
1314 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1315 	if (!exth)
1316 		return false;
1317 	if (exth->version != 2)
1318 		return true;
1319 
1320 	if (exth->checksum &&
1321 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1322 		return false;
1323 
1324 	off += sizeof(_exth);
1325 	while (off < skb->len) {
1326 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1327 		if (!objh)
1328 			return false;
1329 
1330 		olen = ntohs(objh->length);
1331 		if (olen < sizeof(_objh))
1332 			return false;
1333 
1334 		off += olen;
1335 		if (off > skb->len)
1336 			return false;
1337 	}
1338 
1339 	return true;
1340 }
1341 
1342 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1343 			   struct sock_ee_data_rfc4884 *out,
1344 			   int thlen, int off)
1345 {
1346 	int hlen;
1347 
1348 	/* original datagram headers: end of icmph to payload (skb->data) */
1349 	hlen = -skb_transport_offset(skb) - thlen;
1350 
1351 	/* per rfc 4884: minimal datagram length of 128 bytes */
1352 	if (off < 128 || off < hlen)
1353 		return;
1354 
1355 	/* kernel has stripped headers: return payload offset in bytes */
1356 	off -= hlen;
1357 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1358 		return;
1359 
1360 	out->len = off;
1361 
1362 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1363 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1364 }
1365 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1366 
1367 int icmp_err(struct sk_buff *skb, u32 info)
1368 {
1369 	struct iphdr *iph = (struct iphdr *)skb->data;
1370 	int offset = iph->ihl<<2;
1371 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1372 	int type = icmp_hdr(skb)->type;
1373 	int code = icmp_hdr(skb)->code;
1374 	struct net *net = dev_net(skb->dev);
1375 
1376 	/*
1377 	 * Use ping_err to handle all icmp errors except those
1378 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1379 	 */
1380 	if (icmph->type != ICMP_ECHOREPLY) {
1381 		ping_err(skb, offset, info);
1382 		return 0;
1383 	}
1384 
1385 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1386 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1387 	else if (type == ICMP_REDIRECT)
1388 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1389 
1390 	return 0;
1391 }
1392 
1393 /*
1394  *	This table is the definition of how we handle ICMP.
1395  */
1396 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1397 	[ICMP_ECHOREPLY] = {
1398 		.handler = ping_rcv,
1399 	},
1400 	[1] = {
1401 		.handler = icmp_discard,
1402 		.error = 1,
1403 	},
1404 	[2] = {
1405 		.handler = icmp_discard,
1406 		.error = 1,
1407 	},
1408 	[ICMP_DEST_UNREACH] = {
1409 		.handler = icmp_unreach,
1410 		.error = 1,
1411 	},
1412 	[ICMP_SOURCE_QUENCH] = {
1413 		.handler = icmp_unreach,
1414 		.error = 1,
1415 	},
1416 	[ICMP_REDIRECT] = {
1417 		.handler = icmp_redirect,
1418 		.error = 1,
1419 	},
1420 	[6] = {
1421 		.handler = icmp_discard,
1422 		.error = 1,
1423 	},
1424 	[7] = {
1425 		.handler = icmp_discard,
1426 		.error = 1,
1427 	},
1428 	[ICMP_ECHO] = {
1429 		.handler = icmp_echo,
1430 	},
1431 	[9] = {
1432 		.handler = icmp_discard,
1433 		.error = 1,
1434 	},
1435 	[10] = {
1436 		.handler = icmp_discard,
1437 		.error = 1,
1438 	},
1439 	[ICMP_TIME_EXCEEDED] = {
1440 		.handler = icmp_unreach,
1441 		.error = 1,
1442 	},
1443 	[ICMP_PARAMETERPROB] = {
1444 		.handler = icmp_unreach,
1445 		.error = 1,
1446 	},
1447 	[ICMP_TIMESTAMP] = {
1448 		.handler = icmp_timestamp,
1449 	},
1450 	[ICMP_TIMESTAMPREPLY] = {
1451 		.handler = icmp_discard,
1452 	},
1453 	[ICMP_INFO_REQUEST] = {
1454 		.handler = icmp_discard,
1455 	},
1456 	[ICMP_INFO_REPLY] = {
1457 		.handler = icmp_discard,
1458 	},
1459 	[ICMP_ADDRESS] = {
1460 		.handler = icmp_discard,
1461 	},
1462 	[ICMP_ADDRESSREPLY] = {
1463 		.handler = icmp_discard,
1464 	},
1465 };
1466 
1467 static int __net_init icmp_sk_init(struct net *net)
1468 {
1469 	/* Control parameters for ECHO replies. */
1470 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1471 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1472 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1473 
1474 	/* Control parameter - ignore bogus broadcast responses? */
1475 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1476 
1477 	/*
1478 	 * 	Configurable global rate limit.
1479 	 *
1480 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1481 	 *	bucket ratemask defines which icmp types are ratelimited by
1482 	 *	setting	it's bit position.
1483 	 *
1484 	 *	default:
1485 	 *	dest unreachable (3), source quench (4),
1486 	 *	time exceeded (11), parameter problem (12)
1487 	 */
1488 
1489 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1490 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1491 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1492 	net->ipv4.sysctl_icmp_msgs_per_sec = 1000;
1493 	net->ipv4.sysctl_icmp_msgs_burst = 50;
1494 
1495 	return 0;
1496 }
1497 
1498 static struct pernet_operations __net_initdata icmp_sk_ops = {
1499        .init = icmp_sk_init,
1500 };
1501 
1502 int __init icmp_init(void)
1503 {
1504 	int err, i;
1505 
1506 	for_each_possible_cpu(i) {
1507 		struct sock *sk;
1508 
1509 		err = inet_ctl_sock_create(&sk, PF_INET,
1510 					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1511 		if (err < 0)
1512 			return err;
1513 
1514 		per_cpu(ipv4_icmp_sk, i) = sk;
1515 
1516 		/* Enough space for 2 64K ICMP packets, including
1517 		 * sk_buff/skb_shared_info struct overhead.
1518 		 */
1519 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1520 
1521 		/*
1522 		 * Speedup sock_wfree()
1523 		 */
1524 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1525 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1526 	}
1527 	return register_pernet_subsys(&icmp_sk_ops);
1528 }
1529