xref: /linux/net/ipv4/icmp.c (revision 8f7aa3d3c7323f4ca2768a9e74ebbe359c4f8f88)
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/flow.h>
76 #include <net/snmp.h>
77 #include <net/ip.h>
78 #include <net/route.h>
79 #include <net/protocol.h>
80 #include <net/icmp.h>
81 #include <net/tcp.h>
82 #include <net/udp.h>
83 #include <net/raw.h>
84 #include <net/ping.h>
85 #include <linux/skbuff.h>
86 #include <net/sock.h>
87 #include <linux/errno.h>
88 #include <linux/timer.h>
89 #include <linux/init.h>
90 #include <linux/uaccess.h>
91 #include <net/checksum.h>
92 #include <net/xfrm.h>
93 #include <net/inet_common.h>
94 #include <net/ip_fib.h>
95 #include <net/l3mdev.h>
96 #include <net/addrconf.h>
97 #include <net/inet_dscp.h>
98 #define CREATE_TRACE_POINTS
99 #include <trace/events/icmp.h>
100 
101 /*
102  *	Build xmit assembly blocks
103  */
104 
105 struct icmp_bxm {
106 	struct sk_buff *skb;
107 	int offset;
108 	int data_len;
109 
110 	struct {
111 		struct icmphdr icmph;
112 		__be32	       times[3];
113 	} data;
114 	int head_len;
115 	struct ip_options_data replyopts;
116 };
117 
118 /* An array of errno for error messages from dest unreach. */
119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
120 
121 const struct icmp_err icmp_err_convert[] = {
122 	{
123 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
124 		.fatal = 0,
125 	},
126 	{
127 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
128 		.fatal = 0,
129 	},
130 	{
131 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
132 		.fatal = 1,
133 	},
134 	{
135 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
136 		.fatal = 1,
137 	},
138 	{
139 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
140 		.fatal = 0,
141 	},
142 	{
143 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
144 		.fatal = 0,
145 	},
146 	{
147 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
148 		.fatal = 1,
149 	},
150 	{
151 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
152 		.fatal = 1,
153 	},
154 	{
155 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
156 		.fatal = 1,
157 	},
158 	{
159 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
160 		.fatal = 1,
161 	},
162 	{
163 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
164 		.fatal = 1,
165 	},
166 	{
167 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
168 		.fatal = 0,
169 	},
170 	{
171 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
172 		.fatal = 0,
173 	},
174 	{
175 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
176 		.fatal = 1,
177 	},
178 	{
179 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
180 		.fatal = 1,
181 	},
182 	{
183 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
184 		.fatal = 1,
185 	},
186 };
187 EXPORT_SYMBOL(icmp_err_convert);
188 
189 /*
190  *	ICMP control array. This specifies what to do with each ICMP.
191  */
192 
193 struct icmp_control {
194 	enum skb_drop_reason (*handler)(struct sk_buff *skb);
195 	short   error;		/* This ICMP is classed as an error message */
196 };
197 
198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
199 
200 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
201 
202 /* Called with BH disabled */
203 static inline struct sock *icmp_xmit_lock(struct net *net)
204 {
205 	struct sock *sk;
206 
207 	sk = this_cpu_read(ipv4_icmp_sk);
208 
209 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
210 		/* This can happen if the output path signals a
211 		 * dst_link_failure() for an outgoing ICMP packet.
212 		 */
213 		return NULL;
214 	}
215 	sock_net_set(sk, net);
216 	return sk;
217 }
218 
219 static inline void icmp_xmit_unlock(struct sock *sk)
220 {
221 	sock_net_set(sk, &init_net);
222 	spin_unlock(&sk->sk_lock.slock);
223 }
224 
225 /**
226  * icmp_global_allow - Are we allowed to send one more ICMP message ?
227  * @net: network namespace
228  *
229  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
230  * Returns false if we reached the limit and can not send another packet.
231  * Works in tandem with icmp_global_consume().
232  */
233 bool icmp_global_allow(struct net *net)
234 {
235 	u32 delta, now, oldstamp;
236 	int incr, new, old;
237 
238 	/* Note: many cpus could find this condition true.
239 	 * Then later icmp_global_consume() could consume more credits,
240 	 * this is an acceptable race.
241 	 */
242 	if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
243 		return true;
244 
245 	now = jiffies;
246 	oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
247 	delta = min_t(u32, now - oldstamp, HZ);
248 	if (delta < HZ / 50)
249 		return false;
250 
251 	incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ;
252 	if (!incr)
253 		return false;
254 
255 	if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
256 		old = atomic_read(&net->ipv4.icmp_global_credit);
257 		do {
258 			new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
259 		} while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
260 	}
261 	return true;
262 }
263 EXPORT_SYMBOL(icmp_global_allow);
264 
265 void icmp_global_consume(struct net *net)
266 {
267 	int credits = get_random_u32_below(3);
268 
269 	/* Note: this might make icmp_global.credit negative. */
270 	if (credits)
271 		atomic_sub(credits, &net->ipv4.icmp_global_credit);
272 }
273 EXPORT_SYMBOL(icmp_global_consume);
274 
275 static bool icmpv4_mask_allow(struct net *net, int type, int code)
276 {
277 	if (type > NR_ICMP_TYPES)
278 		return true;
279 
280 	/* Don't limit PMTU discovery. */
281 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
282 		return true;
283 
284 	/* Limit if icmp type is enabled in ratemask. */
285 	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
286 		return true;
287 
288 	return false;
289 }
290 
291 static bool icmpv4_global_allow(struct net *net, int type, int code,
292 				bool *apply_ratelimit)
293 {
294 	if (icmpv4_mask_allow(net, type, code))
295 		return true;
296 
297 	if (icmp_global_allow(net)) {
298 		*apply_ratelimit = true;
299 		return true;
300 	}
301 	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
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 			       bool apply_ratelimit)
312 {
313 	struct dst_entry *dst = &rt->dst;
314 	struct inet_peer *peer;
315 	struct net_device *dev;
316 	bool rc = true;
317 
318 	if (!apply_ratelimit)
319 		return true;
320 
321 	/* No rate limit on loopback */
322 	rcu_read_lock();
323 	dev = dst_dev_rcu(dst);
324 	if (dev && (dev->flags & IFF_LOOPBACK))
325 		goto out;
326 
327 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr,
328 			       l3mdev_master_ifindex_rcu(dev));
329 	rc = inet_peer_xrlim_allow(peer,
330 				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
331 out:
332 	rcu_read_unlock();
333 	if (!rc)
334 		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
335 	else
336 		icmp_global_consume(net);
337 	return rc;
338 }
339 
340 /*
341  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
342  */
343 void icmp_out_count(struct net *net, unsigned char type)
344 {
345 	ICMPMSGOUT_INC_STATS(net, type);
346 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
347 }
348 
349 /*
350  *	Checksum each fragment, and on the first include the headers and final
351  *	checksum.
352  */
353 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
354 			  struct sk_buff *skb)
355 {
356 	struct icmp_bxm *icmp_param = from;
357 	__wsum csum;
358 
359 	csum = skb_copy_and_csum_bits(icmp_param->skb,
360 				      icmp_param->offset + offset,
361 				      to, len);
362 
363 	skb->csum = csum_block_add(skb->csum, csum, odd);
364 	if (icmp_pointers[icmp_param->data.icmph.type].error)
365 		nf_ct_attach(skb, icmp_param->skb);
366 	return 0;
367 }
368 
369 static void icmp_push_reply(struct sock *sk,
370 			    struct icmp_bxm *icmp_param,
371 			    struct flowi4 *fl4,
372 			    struct ipcm_cookie *ipc, struct rtable **rt)
373 {
374 	struct sk_buff *skb;
375 
376 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
377 			   icmp_param->data_len+icmp_param->head_len,
378 			   icmp_param->head_len,
379 			   ipc, rt, MSG_DONTWAIT) < 0) {
380 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
381 		ip_flush_pending_frames(sk);
382 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
383 		struct icmphdr *icmph = icmp_hdr(skb);
384 		__wsum csum;
385 		struct sk_buff *skb1;
386 
387 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
388 						 (char *)icmph,
389 						 icmp_param->head_len);
390 		skb_queue_walk(&sk->sk_write_queue, skb1) {
391 			csum = csum_add(csum, skb1->csum);
392 		}
393 		icmph->checksum = csum_fold(csum);
394 		skb->ip_summed = CHECKSUM_NONE;
395 		ip_push_pending_frames(sk, fl4);
396 	}
397 }
398 
399 /*
400  *	Driving logic for building and sending ICMP messages.
401  */
402 
403 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
404 {
405 	struct rtable *rt = skb_rtable(skb);
406 	struct net *net = dev_net_rcu(rt->dst.dev);
407 	bool apply_ratelimit = false;
408 	struct ipcm_cookie ipc;
409 	struct flowi4 fl4;
410 	struct sock *sk;
411 	__be32 daddr, saddr;
412 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
413 	int type = icmp_param->data.icmph.type;
414 	int code = icmp_param->data.icmph.code;
415 
416 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
417 		return;
418 
419 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
420 	local_bh_disable();
421 
422 	/* is global icmp_msgs_per_sec exhausted ? */
423 	if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
424 		goto out_bh_enable;
425 
426 	sk = icmp_xmit_lock(net);
427 	if (!sk)
428 		goto out_bh_enable;
429 
430 	icmp_param->data.icmph.checksum = 0;
431 
432 	ipcm_init(&ipc);
433 	ipc.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_dscp = 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 *dev = skb->dev;
473 	const struct dst_entry *dst;
474 
475 	if (dev)
476 		return dev;
477 	dst = skb_dst(skb);
478 	return dst ? dst_dev(dst) : NULL;
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_dscp = 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 		if (inet_addr_type_dev_table(net, route_lookup_dev,
521 					     fl4->daddr) == RTN_LOCAL)
522 			return rt;
523 	} else if (PTR_ERR(dst) == -EPERM) {
524 		rt = NULL;
525 	} else {
526 		return rt;
527 	}
528 	err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
529 	if (err)
530 		goto relookup_failed;
531 
532 	if (inet_addr_type_dev_table(net, route_lookup_dev,
533 				     fl4_dec.saddr) == RTN_LOCAL) {
534 		rt2 = __ip_route_output_key(net, &fl4_dec);
535 		if (IS_ERR(rt2))
536 			err = PTR_ERR(rt2);
537 	} else {
538 		struct flowi4 fl4_2 = {};
539 		unsigned long orefdst;
540 
541 		fl4_2.daddr = fl4_dec.saddr;
542 		rt2 = ip_route_output_key(net, &fl4_2);
543 		if (IS_ERR(rt2)) {
544 			err = PTR_ERR(rt2);
545 			goto relookup_failed;
546 		}
547 		/* Ugh! */
548 		orefdst = skb_dstref_steal(skb_in);
549 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
550 				     dscp, rt2->dst.dev) ? -EINVAL : 0;
551 
552 		dst_release(&rt2->dst);
553 		rt2 = skb_rtable(skb_in);
554 		/* steal dst entry from skb_in, don't drop refcnt */
555 		skb_dstref_steal(skb_in);
556 		skb_dstref_restore(skb_in, orefdst);
557 	}
558 
559 	if (err)
560 		goto relookup_failed;
561 
562 	dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
563 			   XFRM_LOOKUP_ICMP);
564 	rt2 = dst_rtable(dst2);
565 	if (!IS_ERR(dst2)) {
566 		dst_release(&rt->dst);
567 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
568 		rt = rt2;
569 	} else if (PTR_ERR(dst2) == -EPERM) {
570 		if (rt)
571 			dst_release(&rt->dst);
572 		return rt2;
573 	} else {
574 		err = PTR_ERR(dst2);
575 		goto relookup_failed;
576 	}
577 	return rt;
578 
579 relookup_failed:
580 	if (rt)
581 		return rt;
582 	return ERR_PTR(err);
583 }
584 
585 struct icmp_ext_iio_addr4_subobj {
586 	__be16 afi;
587 	__be16 reserved;
588 	__be32 addr4;
589 };
590 
591 static unsigned int icmp_ext_iio_len(void)
592 {
593 	return sizeof(struct icmp_extobj_hdr) +
594 		/* ifIndex */
595 		sizeof(__be32) +
596 		/* Interface Address Sub-Object */
597 		sizeof(struct icmp_ext_iio_addr4_subobj) +
598 		/* Interface Name Sub-Object. Length must be a multiple of 4
599 		 * bytes.
600 		 */
601 		ALIGN(sizeof(struct icmp_ext_iio_name_subobj), 4) +
602 		/* MTU */
603 		sizeof(__be32);
604 }
605 
606 static unsigned int icmp_ext_max_len(u8 ext_objs)
607 {
608 	unsigned int ext_max_len;
609 
610 	ext_max_len = sizeof(struct icmp_ext_hdr);
611 
612 	if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
613 		ext_max_len += icmp_ext_iio_len();
614 
615 	return ext_max_len;
616 }
617 
618 static __be32 icmp_ext_iio_addr4_find(const struct net_device *dev)
619 {
620 	struct in_device *in_dev;
621 	struct in_ifaddr *ifa;
622 
623 	in_dev = __in_dev_get_rcu(dev);
624 	if (!in_dev)
625 		return 0;
626 
627 	/* It is unclear from RFC 5837 which IP address should be chosen, but
628 	 * it makes sense to choose a global unicast address.
629 	 */
630 	in_dev_for_each_ifa_rcu(ifa, in_dev) {
631 		if (READ_ONCE(ifa->ifa_flags) & IFA_F_SECONDARY)
632 			continue;
633 		if (ifa->ifa_scope != RT_SCOPE_UNIVERSE ||
634 		    ipv4_is_multicast(ifa->ifa_address))
635 			continue;
636 		return ifa->ifa_address;
637 	}
638 
639 	return 0;
640 }
641 
642 static void icmp_ext_iio_iif_append(struct net *net, struct sk_buff *skb,
643 				    int iif)
644 {
645 	struct icmp_ext_iio_name_subobj *name_subobj;
646 	struct icmp_extobj_hdr *objh;
647 	struct net_device *dev;
648 	__be32 data;
649 
650 	if (!iif)
651 		return;
652 
653 	/* Add the fields in the order specified by RFC 5837. */
654 	objh = skb_put(skb, sizeof(*objh));
655 	objh->class_num = ICMP_EXT_OBJ_CLASS_IIO;
656 	objh->class_type = ICMP_EXT_CTYPE_IIO_ROLE(ICMP_EXT_CTYPE_IIO_ROLE_IIF);
657 
658 	data = htonl(iif);
659 	skb_put_data(skb, &data, sizeof(__be32));
660 	objh->class_type |= ICMP_EXT_CTYPE_IIO_IFINDEX;
661 
662 	rcu_read_lock();
663 
664 	dev = dev_get_by_index_rcu(net, iif);
665 	if (!dev)
666 		goto out;
667 
668 	data = icmp_ext_iio_addr4_find(dev);
669 	if (data) {
670 		struct icmp_ext_iio_addr4_subobj *addr4_subobj;
671 
672 		addr4_subobj = skb_put_zero(skb, sizeof(*addr4_subobj));
673 		addr4_subobj->afi = htons(ICMP_AFI_IP);
674 		addr4_subobj->addr4 = data;
675 		objh->class_type |= ICMP_EXT_CTYPE_IIO_IPADDR;
676 	}
677 
678 	name_subobj = skb_put_zero(skb, ALIGN(sizeof(*name_subobj), 4));
679 	name_subobj->len = ALIGN(sizeof(*name_subobj), 4);
680 	netdev_copy_name(dev, name_subobj->name);
681 	objh->class_type |= ICMP_EXT_CTYPE_IIO_NAME;
682 
683 	data = htonl(READ_ONCE(dev->mtu));
684 	skb_put_data(skb, &data, sizeof(__be32));
685 	objh->class_type |= ICMP_EXT_CTYPE_IIO_MTU;
686 
687 out:
688 	rcu_read_unlock();
689 	objh->length = htons(skb_tail_pointer(skb) - (unsigned char *)objh);
690 }
691 
692 static void icmp_ext_objs_append(struct net *net, struct sk_buff *skb,
693 				 u8 ext_objs, int iif)
694 {
695 	if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
696 		icmp_ext_iio_iif_append(net, skb, iif);
697 }
698 
699 static struct sk_buff *
700 icmp_ext_append(struct net *net, struct sk_buff *skb_in, struct icmphdr *icmph,
701 		unsigned int room, int iif)
702 {
703 	unsigned int payload_len, ext_max_len, ext_len;
704 	struct icmp_ext_hdr *ext_hdr;
705 	struct sk_buff *skb;
706 	u8 ext_objs;
707 	int nhoff;
708 
709 	switch (icmph->type) {
710 	case ICMP_DEST_UNREACH:
711 	case ICMP_TIME_EXCEEDED:
712 	case ICMP_PARAMETERPROB:
713 		break;
714 	default:
715 		return NULL;
716 	}
717 
718 	ext_objs = READ_ONCE(net->ipv4.sysctl_icmp_errors_extension_mask);
719 	if (!ext_objs)
720 		return NULL;
721 
722 	ext_max_len = icmp_ext_max_len(ext_objs);
723 	if (ICMP_EXT_ORIG_DGRAM_MIN_LEN + ext_max_len > room)
724 		return NULL;
725 
726 	skb = skb_clone(skb_in, GFP_ATOMIC);
727 	if (!skb)
728 		return NULL;
729 
730 	nhoff = skb_network_offset(skb);
731 	payload_len = min(skb->len - nhoff, ICMP_EXT_ORIG_DGRAM_MIN_LEN);
732 
733 	if (!pskb_network_may_pull(skb, payload_len))
734 		goto free_skb;
735 
736 	if (pskb_trim(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN) ||
737 	    __skb_put_padto(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN, false))
738 		goto free_skb;
739 
740 	if (pskb_expand_head(skb, 0, ext_max_len, GFP_ATOMIC))
741 		goto free_skb;
742 
743 	ext_hdr = skb_put_zero(skb, sizeof(*ext_hdr));
744 	ext_hdr->version = ICMP_EXT_VERSION_2;
745 
746 	icmp_ext_objs_append(net, skb, ext_objs, iif);
747 
748 	/* Do not send an empty extension structure. */
749 	ext_len = skb_tail_pointer(skb) - (unsigned char *)ext_hdr;
750 	if (ext_len == sizeof(*ext_hdr))
751 		goto free_skb;
752 
753 	ext_hdr->checksum = ip_compute_csum(ext_hdr, ext_len);
754 	/* The length of the original datagram in 32-bit words (RFC 4884). */
755 	icmph->un.reserved[1] = ICMP_EXT_ORIG_DGRAM_MIN_LEN / sizeof(u32);
756 
757 	return skb;
758 
759 free_skb:
760 	consume_skb(skb);
761 	return NULL;
762 }
763 
764 /*
765  *	Send an ICMP message in response to a situation
766  *
767  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
768  *		  MAY send more (we do).
769  *			MUST NOT change this header information.
770  *			MUST NOT reply to a multicast/broadcast IP address.
771  *			MUST NOT reply to a multicast/broadcast MAC address.
772  *			MUST reply to only the first fragment.
773  */
774 
775 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
776 		 const struct inet_skb_parm *parm)
777 {
778 	struct iphdr *iph;
779 	int room;
780 	struct icmp_bxm icmp_param;
781 	struct rtable *rt = skb_rtable(skb_in);
782 	bool apply_ratelimit = false;
783 	struct sk_buff *ext_skb;
784 	struct ipcm_cookie ipc;
785 	struct flowi4 fl4;
786 	__be32 saddr;
787 	u8  tos;
788 	u32 mark;
789 	struct net *net;
790 	struct sock *sk;
791 
792 	if (!rt)
793 		return;
794 
795 	rcu_read_lock();
796 
797 	if (rt->dst.dev)
798 		net = dev_net_rcu(rt->dst.dev);
799 	else if (skb_in->dev)
800 		net = dev_net_rcu(skb_in->dev);
801 	else
802 		goto out;
803 
804 	/*
805 	 *	Find the original header. It is expected to be valid, of course.
806 	 *	Check this, icmp_send is called from the most obscure devices
807 	 *	sometimes.
808 	 */
809 	iph = ip_hdr(skb_in);
810 
811 	if ((u8 *)iph < skb_in->head ||
812 	    (skb_network_header(skb_in) + sizeof(*iph)) >
813 	    skb_tail_pointer(skb_in))
814 		goto out;
815 
816 	/*
817 	 *	No replies to physical multicast/broadcast
818 	 */
819 	if (skb_in->pkt_type != PACKET_HOST)
820 		goto out;
821 
822 	/*
823 	 *	Now check at the protocol level
824 	 */
825 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
826 		goto out;
827 
828 	/*
829 	 *	Only reply to fragment 0. We byte re-order the constant
830 	 *	mask for efficiency.
831 	 */
832 	if (iph->frag_off & htons(IP_OFFSET))
833 		goto out;
834 
835 	/*
836 	 *	If we send an ICMP error to an ICMP error a mess would result..
837 	 */
838 	if (icmp_pointers[type].error) {
839 		/*
840 		 *	We are an error, check if we are replying to an
841 		 *	ICMP error
842 		 */
843 		if (iph->protocol == IPPROTO_ICMP) {
844 			u8 _inner_type, *itp;
845 
846 			itp = skb_header_pointer(skb_in,
847 						 skb_network_header(skb_in) +
848 						 (iph->ihl << 2) +
849 						 offsetof(struct icmphdr,
850 							  type) -
851 						 skb_in->data,
852 						 sizeof(_inner_type),
853 						 &_inner_type);
854 			if (!itp)
855 				goto out;
856 
857 			/*
858 			 *	Assume any unknown ICMP type is an error. This
859 			 *	isn't specified by the RFC, but think about it..
860 			 */
861 			if (*itp > NR_ICMP_TYPES ||
862 			    icmp_pointers[*itp].error)
863 				goto out;
864 		}
865 	}
866 
867 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
868 	local_bh_disable();
869 
870 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
871 	 * incoming dev is loopback.  If outgoing dev change to not be
872 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
873 	 */
874 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
875 	      !icmpv4_global_allow(net, type, code, &apply_ratelimit))
876 		goto out_bh_enable;
877 
878 	sk = icmp_xmit_lock(net);
879 	if (!sk)
880 		goto out_bh_enable;
881 
882 	/*
883 	 *	Construct source address and options.
884 	 */
885 
886 	saddr = iph->daddr;
887 	if (!(rt->rt_flags & RTCF_LOCAL)) {
888 		struct net_device *dev = NULL;
889 
890 		rcu_read_lock();
891 		if (rt_is_input_route(rt) &&
892 		    READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
893 			dev = dev_get_by_index_rcu(net, parm->iif ? parm->iif :
894 						   inet_iif(skb_in));
895 
896 		if (dev)
897 			saddr = inet_select_addr(dev, iph->saddr,
898 						 RT_SCOPE_LINK);
899 		else
900 			saddr = 0;
901 		rcu_read_unlock();
902 	}
903 
904 	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
905 					   IPTOS_PREC_INTERNETCONTROL) :
906 					   iph->tos;
907 	mark = IP4_REPLY_MARK(net, skb_in->mark);
908 
909 	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in,
910 			      &parm->opt))
911 		goto out_unlock;
912 
913 
914 	/*
915 	 *	Prepare data for ICMP header.
916 	 */
917 
918 	icmp_param.data.icmph.type	 = type;
919 	icmp_param.data.icmph.code	 = code;
920 	icmp_param.data.icmph.un.gateway = info;
921 	icmp_param.data.icmph.checksum	 = 0;
922 	icmp_param.skb	  = skb_in;
923 	icmp_param.offset = skb_network_offset(skb_in);
924 	ipcm_init(&ipc);
925 	ipc.tos = tos;
926 	ipc.addr = iph->saddr;
927 	ipc.opt = &icmp_param.replyopts.opt;
928 	ipc.sockc.mark = mark;
929 
930 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
931 			       inet_dsfield_to_dscp(tos), mark, type, code,
932 			       &icmp_param);
933 	if (IS_ERR(rt))
934 		goto out_unlock;
935 
936 	/* peer icmp_ratelimit */
937 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
938 		goto ende;
939 
940 	/* RFC says return as much as we can without exceeding 576 bytes. */
941 
942 	room = dst_mtu(&rt->dst);
943 	if (room > 576)
944 		room = 576;
945 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
946 	room -= sizeof(struct icmphdr);
947 	/* Guard against tiny mtu. We need to include at least one
948 	 * IP network header for this message to make any sense.
949 	 */
950 	if (room <= (int)sizeof(struct iphdr))
951 		goto ende;
952 
953 	ext_skb = icmp_ext_append(net, skb_in, &icmp_param.data.icmph, room,
954 				  parm->iif);
955 	if (ext_skb)
956 		icmp_param.skb = ext_skb;
957 
958 	icmp_param.data_len = icmp_param.skb->len - icmp_param.offset;
959 	if (icmp_param.data_len > room)
960 		icmp_param.data_len = room;
961 	icmp_param.head_len = sizeof(struct icmphdr);
962 
963 	/* if we don't have a source address at this point, fall back to the
964 	 * dummy address instead of sending out a packet with a source address
965 	 * of 0.0.0.0
966 	 */
967 	if (!fl4.saddr)
968 		fl4.saddr = htonl(INADDR_DUMMY);
969 
970 	trace_icmp_send(skb_in, type, code);
971 
972 	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
973 
974 	if (ext_skb)
975 		consume_skb(ext_skb);
976 ende:
977 	ip_rt_put(rt);
978 out_unlock:
979 	icmp_xmit_unlock(sk);
980 out_bh_enable:
981 	local_bh_enable();
982 out:
983 	rcu_read_unlock();
984 }
985 EXPORT_SYMBOL(__icmp_send);
986 
987 #if IS_ENABLED(CONFIG_NF_NAT)
988 #include <net/netfilter/nf_conntrack.h>
989 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
990 {
991 	struct sk_buff *cloned_skb = NULL;
992 	enum ip_conntrack_info ctinfo;
993 	enum ip_conntrack_dir dir;
994 	struct inet_skb_parm parm;
995 	struct nf_conn *ct;
996 	__be32 orig_ip;
997 
998 	memset(&parm, 0, sizeof(parm));
999 	ct = nf_ct_get(skb_in, &ctinfo);
1000 	if (!ct || !(READ_ONCE(ct->status) & IPS_NAT_MASK)) {
1001 		__icmp_send(skb_in, type, code, info, &parm);
1002 		return;
1003 	}
1004 
1005 	if (skb_shared(skb_in))
1006 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
1007 
1008 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
1009 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
1010 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
1011 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
1012 		goto out;
1013 
1014 	orig_ip = ip_hdr(skb_in)->saddr;
1015 	dir = CTINFO2DIR(ctinfo);
1016 	ip_hdr(skb_in)->saddr = ct->tuplehash[dir].tuple.src.u3.ip;
1017 	__icmp_send(skb_in, type, code, info, &parm);
1018 	ip_hdr(skb_in)->saddr = orig_ip;
1019 out:
1020 	consume_skb(cloned_skb);
1021 }
1022 EXPORT_SYMBOL(icmp_ndo_send);
1023 #endif
1024 
1025 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
1026 {
1027 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1028 	const struct net_protocol *ipprot;
1029 	int protocol = iph->protocol;
1030 
1031 	/* Checkin full IP header plus 8 bytes of protocol to
1032 	 * avoid additional coding at protocol handlers.
1033 	 */
1034 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
1035 		__ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1036 		return;
1037 	}
1038 
1039 	raw_icmp_error(skb, protocol, info);
1040 
1041 	ipprot = rcu_dereference(inet_protos[protocol]);
1042 	if (ipprot && ipprot->err_handler)
1043 		ipprot->err_handler(skb, info);
1044 }
1045 
1046 static bool icmp_tag_validation(int proto)
1047 {
1048 	bool ok;
1049 
1050 	rcu_read_lock();
1051 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
1052 	rcu_read_unlock();
1053 	return ok;
1054 }
1055 
1056 /*
1057  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
1058  *	ICMP_PARAMETERPROB.
1059  */
1060 
1061 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
1062 {
1063 	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
1064 	const struct iphdr *iph;
1065 	struct icmphdr *icmph;
1066 	struct net *net;
1067 	u32 info = 0;
1068 
1069 	net = skb_dst_dev_net_rcu(skb);
1070 
1071 	/*
1072 	 *	Incomplete header ?
1073 	 * 	Only checks for the IP header, there should be an
1074 	 *	additional check for longer headers in upper levels.
1075 	 */
1076 
1077 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1078 		goto out_err;
1079 
1080 	icmph = icmp_hdr(skb);
1081 	iph   = (const struct iphdr *)skb->data;
1082 
1083 	if (iph->ihl < 5)  { /* Mangled header, drop. */
1084 		reason = SKB_DROP_REASON_IP_INHDR;
1085 		goto out_err;
1086 	}
1087 
1088 	switch (icmph->type) {
1089 	case ICMP_DEST_UNREACH:
1090 		switch (icmph->code & 15) {
1091 		case ICMP_NET_UNREACH:
1092 		case ICMP_HOST_UNREACH:
1093 		case ICMP_PROT_UNREACH:
1094 		case ICMP_PORT_UNREACH:
1095 			break;
1096 		case ICMP_FRAG_NEEDED:
1097 			/* for documentation of the ip_no_pmtu_disc
1098 			 * values please see
1099 			 * Documentation/networking/ip-sysctl.rst
1100 			 */
1101 			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
1102 			default:
1103 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
1104 						    &iph->daddr);
1105 				break;
1106 			case 2:
1107 				goto out;
1108 			case 3:
1109 				if (!icmp_tag_validation(iph->protocol))
1110 					goto out;
1111 				fallthrough;
1112 			case 0:
1113 				info = ntohs(icmph->un.frag.mtu);
1114 			}
1115 			break;
1116 		case ICMP_SR_FAILED:
1117 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
1118 					    &iph->daddr);
1119 			break;
1120 		default:
1121 			break;
1122 		}
1123 		if (icmph->code > NR_ICMP_UNREACH)
1124 			goto out;
1125 		break;
1126 	case ICMP_PARAMETERPROB:
1127 		info = ntohl(icmph->un.gateway) >> 24;
1128 		break;
1129 	case ICMP_TIME_EXCEEDED:
1130 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
1131 		if (icmph->code == ICMP_EXC_FRAGTIME)
1132 			goto out;
1133 		break;
1134 	}
1135 
1136 	/*
1137 	 *	Throw it at our lower layers
1138 	 *
1139 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
1140 	 *		  header.
1141 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
1142 	 *		  transport layer.
1143 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
1144 	 *		  transport layer.
1145 	 */
1146 
1147 	/*
1148 	 *	Check the other end isn't violating RFC 1122. Some routers send
1149 	 *	bogus responses to broadcast frames. If you see this message
1150 	 *	first check your netmask matches at both ends, if it does then
1151 	 *	get the other vendor to fix their kit.
1152 	 */
1153 
1154 	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
1155 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
1156 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
1157 				     &ip_hdr(skb)->saddr,
1158 				     icmph->type, icmph->code,
1159 				     &iph->daddr, skb->dev->name);
1160 		goto out;
1161 	}
1162 
1163 	icmp_socket_deliver(skb, info);
1164 
1165 out:
1166 	return reason;
1167 out_err:
1168 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1169 	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
1170 }
1171 
1172 
1173 /*
1174  *	Handle ICMP_REDIRECT.
1175  */
1176 
1177 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
1178 {
1179 	if (skb->len < sizeof(struct iphdr)) {
1180 		__ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1181 		return SKB_DROP_REASON_PKT_TOO_SMALL;
1182 	}
1183 
1184 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
1185 		/* there aught to be a stat */
1186 		return SKB_DROP_REASON_NOMEM;
1187 	}
1188 
1189 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
1190 	return SKB_NOT_DROPPED_YET;
1191 }
1192 
1193 /*
1194  *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1195  *
1196  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1197  *		  requests.
1198  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1199  *		  included in the reply.
1200  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1201  *		  echo requests, MUST have default=NOT.
1202  *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
1203  *		  Extended Echo Functionality, MUST be disabled by default
1204  *	See also WRT handling of options once they are done and working.
1205  */
1206 
1207 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1208 {
1209 	struct icmp_bxm icmp_param;
1210 	struct net *net;
1211 
1212 	net = skb_dst_dev_net_rcu(skb);
1213 	/* should there be an ICMP stat for ignored echos? */
1214 	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1215 		return SKB_NOT_DROPPED_YET;
1216 
1217 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1218 	icmp_param.skb		   = skb;
1219 	icmp_param.offset	   = 0;
1220 	icmp_param.data_len	   = skb->len;
1221 	icmp_param.head_len	   = sizeof(struct icmphdr);
1222 
1223 	if (icmp_param.data.icmph.type == ICMP_ECHO)
1224 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1225 	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1226 		return SKB_NOT_DROPPED_YET;
1227 
1228 	icmp_reply(&icmp_param, skb);
1229 	return SKB_NOT_DROPPED_YET;
1230 }
1231 
1232 /*	Helper for icmp_echo and icmpv6_echo_reply.
1233  *	Searches for net_device that matches PROBE interface identifier
1234  *		and builds PROBE reply message in icmphdr.
1235  *
1236  *	Returns false if PROBE responses are disabled via sysctl
1237  */
1238 
1239 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1240 {
1241 	struct net *net = dev_net_rcu(skb->dev);
1242 	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1243 	struct icmp_ext_echo_iio *iio, _iio;
1244 	struct inet6_dev *in6_dev;
1245 	struct in_device *in_dev;
1246 	struct net_device *dev;
1247 	char buff[IFNAMSIZ];
1248 	u16 ident_len;
1249 	u8 status;
1250 
1251 	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1252 		return false;
1253 
1254 	/* We currently only support probing interfaces on the proxy node
1255 	 * Check to ensure L-bit is set
1256 	 */
1257 	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1258 		return false;
1259 	/* Clear status bits in reply message */
1260 	icmphdr->un.echo.sequence &= htons(0xFF00);
1261 	if (icmphdr->type == ICMP_EXT_ECHO)
1262 		icmphdr->type = ICMP_EXT_ECHOREPLY;
1263 	else
1264 		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1265 	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1266 	/* Size of iio is class_type dependent.
1267 	 * Only check header here and assign length based on ctype in the switch statement
1268 	 */
1269 	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1270 	if (!ext_hdr || !iio)
1271 		goto send_mal_query;
1272 	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1273 	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1274 		goto send_mal_query;
1275 	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1276 	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1277 				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1278 	if (!iio)
1279 		goto send_mal_query;
1280 
1281 	status = 0;
1282 	dev = NULL;
1283 	switch (iio->extobj_hdr.class_type) {
1284 	case ICMP_EXT_ECHO_CTYPE_NAME:
1285 		if (ident_len >= IFNAMSIZ)
1286 			goto send_mal_query;
1287 		memset(buff, 0, sizeof(buff));
1288 		memcpy(buff, &iio->ident.name, ident_len);
1289 		dev = dev_get_by_name(net, buff);
1290 		break;
1291 	case ICMP_EXT_ECHO_CTYPE_INDEX:
1292 		if (ident_len != sizeof(iio->ident.ifindex))
1293 			goto send_mal_query;
1294 		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1295 		break;
1296 	case ICMP_EXT_ECHO_CTYPE_ADDR:
1297 		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1298 		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1299 				 iio->ident.addr.ctype3_hdr.addrlen)
1300 			goto send_mal_query;
1301 		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1302 		case ICMP_AFI_IP:
1303 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1304 				goto send_mal_query;
1305 			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1306 			break;
1307 #if IS_ENABLED(CONFIG_IPV6)
1308 		case ICMP_AFI_IP6:
1309 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1310 				goto send_mal_query;
1311 			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1312 			dev_hold(dev);
1313 			break;
1314 #endif
1315 		default:
1316 			goto send_mal_query;
1317 		}
1318 		break;
1319 	default:
1320 		goto send_mal_query;
1321 	}
1322 	if (!dev) {
1323 		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1324 		return true;
1325 	}
1326 	/* Fill bits in reply message */
1327 	if (dev->flags & IFF_UP)
1328 		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1329 
1330 	in_dev = __in_dev_get_rcu(dev);
1331 	if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1332 		status |= ICMP_EXT_ECHOREPLY_IPV4;
1333 
1334 	in6_dev = __in6_dev_get(dev);
1335 	if (in6_dev && !list_empty(&in6_dev->addr_list))
1336 		status |= ICMP_EXT_ECHOREPLY_IPV6;
1337 
1338 	dev_put(dev);
1339 	icmphdr->un.echo.sequence |= htons(status);
1340 	return true;
1341 send_mal_query:
1342 	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1343 	return true;
1344 }
1345 EXPORT_SYMBOL_GPL(icmp_build_probe);
1346 
1347 /*
1348  *	Handle ICMP Timestamp requests.
1349  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1350  *		  SHOULD be in the kernel for minimum random latency.
1351  *		  MUST be accurate to a few minutes.
1352  *		  MUST be updated at least at 15Hz.
1353  */
1354 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1355 {
1356 	struct icmp_bxm icmp_param;
1357 	/*
1358 	 *	Too short.
1359 	 */
1360 	if (skb->len < 4)
1361 		goto out_err;
1362 
1363 	/*
1364 	 *	Fill in the current time as ms since midnight UT:
1365 	 */
1366 	icmp_param.data.times[1] = inet_current_timestamp();
1367 	icmp_param.data.times[2] = icmp_param.data.times[1];
1368 
1369 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1370 
1371 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1372 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1373 	icmp_param.data.icmph.code = 0;
1374 	icmp_param.skb		   = skb;
1375 	icmp_param.offset	   = 0;
1376 	icmp_param.data_len	   = 0;
1377 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1378 	icmp_reply(&icmp_param, skb);
1379 	return SKB_NOT_DROPPED_YET;
1380 
1381 out_err:
1382 	__ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS);
1383 	return SKB_DROP_REASON_PKT_TOO_SMALL;
1384 }
1385 
1386 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1387 {
1388 	/* pretend it was a success */
1389 	return SKB_NOT_DROPPED_YET;
1390 }
1391 
1392 /*
1393  *	Deal with incoming ICMP packets.
1394  */
1395 int icmp_rcv(struct sk_buff *skb)
1396 {
1397 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1398 	struct rtable *rt = skb_rtable(skb);
1399 	struct net *net = dev_net_rcu(rt->dst.dev);
1400 	struct icmphdr *icmph;
1401 
1402 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1403 		struct sec_path *sp = skb_sec_path(skb);
1404 		int nh;
1405 
1406 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1407 				 XFRM_STATE_ICMP)) {
1408 			reason = SKB_DROP_REASON_XFRM_POLICY;
1409 			goto drop;
1410 		}
1411 
1412 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1413 			goto drop;
1414 
1415 		nh = skb_network_offset(skb);
1416 		skb_set_network_header(skb, sizeof(*icmph));
1417 
1418 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1419 						skb)) {
1420 			reason = SKB_DROP_REASON_XFRM_POLICY;
1421 			goto drop;
1422 		}
1423 
1424 		skb_set_network_header(skb, nh);
1425 	}
1426 
1427 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1428 
1429 	if (skb_checksum_simple_validate(skb))
1430 		goto csum_error;
1431 
1432 	if (!pskb_pull(skb, sizeof(*icmph)))
1433 		goto error;
1434 
1435 	icmph = icmp_hdr(skb);
1436 
1437 	ICMPMSGIN_INC_STATS(net, icmph->type);
1438 
1439 	/* Check for ICMP Extended Echo (PROBE) messages */
1440 	if (icmph->type == ICMP_EXT_ECHO) {
1441 		/* We can't use icmp_pointers[].handler() because it is an array of
1442 		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1443 		 */
1444 		reason = icmp_echo(skb);
1445 		goto reason_check;
1446 	}
1447 
1448 	/*
1449 	 *	Parse the ICMP message
1450 	 */
1451 
1452 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1453 		/*
1454 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1455 		 *	  silently ignored (we let user decide with a sysctl).
1456 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1457 		 *	  discarded if to broadcast/multicast.
1458 		 */
1459 		if ((icmph->type == ICMP_ECHO ||
1460 		     icmph->type == ICMP_TIMESTAMP) &&
1461 		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1462 			reason = SKB_DROP_REASON_INVALID_PROTO;
1463 			goto error;
1464 		}
1465 		if (icmph->type != ICMP_ECHO &&
1466 		    icmph->type != ICMP_TIMESTAMP &&
1467 		    icmph->type != ICMP_ADDRESS &&
1468 		    icmph->type != ICMP_ADDRESSREPLY) {
1469 			reason = SKB_DROP_REASON_INVALID_PROTO;
1470 			goto error;
1471 		}
1472 	}
1473 
1474 	if (icmph->type == ICMP_EXT_ECHOREPLY ||
1475 	    icmph->type == ICMP_ECHOREPLY) {
1476 		reason = ping_rcv(skb);
1477 		return reason ? NET_RX_DROP : NET_RX_SUCCESS;
1478 	}
1479 
1480 	/*
1481 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1482 	 *
1483 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1484 	 *		  discarded.
1485 	 */
1486 	if (icmph->type > NR_ICMP_TYPES) {
1487 		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1488 		goto error;
1489 	}
1490 
1491 	reason = icmp_pointers[icmph->type].handler(skb);
1492 reason_check:
1493 	if (!reason)  {
1494 		consume_skb(skb);
1495 		return NET_RX_SUCCESS;
1496 	}
1497 
1498 drop:
1499 	kfree_skb_reason(skb, reason);
1500 	return NET_RX_DROP;
1501 csum_error:
1502 	reason = SKB_DROP_REASON_ICMP_CSUM;
1503 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1504 error:
1505 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1506 	goto drop;
1507 }
1508 
1509 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1510 {
1511 	struct icmp_extobj_hdr *objh, _objh;
1512 	struct icmp_ext_hdr *exth, _exth;
1513 	u16 olen;
1514 
1515 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1516 	if (!exth)
1517 		return false;
1518 	if (exth->version != 2)
1519 		return true;
1520 
1521 	if (exth->checksum &&
1522 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1523 		return false;
1524 
1525 	off += sizeof(_exth);
1526 	while (off < skb->len) {
1527 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1528 		if (!objh)
1529 			return false;
1530 
1531 		olen = ntohs(objh->length);
1532 		if (olen < sizeof(_objh))
1533 			return false;
1534 
1535 		off += olen;
1536 		if (off > skb->len)
1537 			return false;
1538 	}
1539 
1540 	return true;
1541 }
1542 
1543 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1544 			   struct sock_ee_data_rfc4884 *out,
1545 			   int thlen, int off)
1546 {
1547 	int hlen;
1548 
1549 	/* original datagram headers: end of icmph to payload (skb->data) */
1550 	hlen = -skb_transport_offset(skb) - thlen;
1551 
1552 	/* per rfc 4884: minimal datagram length of 128 bytes */
1553 	if (off < 128 || off < hlen)
1554 		return;
1555 
1556 	/* kernel has stripped headers: return payload offset in bytes */
1557 	off -= hlen;
1558 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1559 		return;
1560 
1561 	out->len = off;
1562 
1563 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1564 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1565 }
1566 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1567 
1568 int icmp_err(struct sk_buff *skb, u32 info)
1569 {
1570 	struct iphdr *iph = (struct iphdr *)skb->data;
1571 	int offset = iph->ihl<<2;
1572 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1573 	struct net *net = dev_net_rcu(skb->dev);
1574 	int type = icmp_hdr(skb)->type;
1575 	int code = icmp_hdr(skb)->code;
1576 
1577 	/*
1578 	 * Use ping_err to handle all icmp errors except those
1579 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1580 	 */
1581 	if (icmph->type != ICMP_ECHOREPLY) {
1582 		ping_err(skb, offset, info);
1583 		return 0;
1584 	}
1585 
1586 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1587 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1588 	else if (type == ICMP_REDIRECT)
1589 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1590 
1591 	return 0;
1592 }
1593 
1594 /*
1595  *	This table is the definition of how we handle ICMP.
1596  */
1597 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1598 	[ICMP_ECHOREPLY] = {
1599 		.handler = ping_rcv,
1600 	},
1601 	[1] = {
1602 		.handler = icmp_discard,
1603 		.error = 1,
1604 	},
1605 	[2] = {
1606 		.handler = icmp_discard,
1607 		.error = 1,
1608 	},
1609 	[ICMP_DEST_UNREACH] = {
1610 		.handler = icmp_unreach,
1611 		.error = 1,
1612 	},
1613 	[ICMP_SOURCE_QUENCH] = {
1614 		.handler = icmp_unreach,
1615 		.error = 1,
1616 	},
1617 	[ICMP_REDIRECT] = {
1618 		.handler = icmp_redirect,
1619 		.error = 1,
1620 	},
1621 	[6] = {
1622 		.handler = icmp_discard,
1623 		.error = 1,
1624 	},
1625 	[7] = {
1626 		.handler = icmp_discard,
1627 		.error = 1,
1628 	},
1629 	[ICMP_ECHO] = {
1630 		.handler = icmp_echo,
1631 	},
1632 	[9] = {
1633 		.handler = icmp_discard,
1634 		.error = 1,
1635 	},
1636 	[10] = {
1637 		.handler = icmp_discard,
1638 		.error = 1,
1639 	},
1640 	[ICMP_TIME_EXCEEDED] = {
1641 		.handler = icmp_unreach,
1642 		.error = 1,
1643 	},
1644 	[ICMP_PARAMETERPROB] = {
1645 		.handler = icmp_unreach,
1646 		.error = 1,
1647 	},
1648 	[ICMP_TIMESTAMP] = {
1649 		.handler = icmp_timestamp,
1650 	},
1651 	[ICMP_TIMESTAMPREPLY] = {
1652 		.handler = icmp_discard,
1653 	},
1654 	[ICMP_INFO_REQUEST] = {
1655 		.handler = icmp_discard,
1656 	},
1657 	[ICMP_INFO_REPLY] = {
1658 		.handler = icmp_discard,
1659 	},
1660 	[ICMP_ADDRESS] = {
1661 		.handler = icmp_discard,
1662 	},
1663 	[ICMP_ADDRESSREPLY] = {
1664 		.handler = icmp_discard,
1665 	},
1666 };
1667 
1668 static int __net_init icmp_sk_init(struct net *net)
1669 {
1670 	/* Control parameters for ECHO replies. */
1671 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1672 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1673 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1674 
1675 	/* Control parameter - ignore bogus broadcast responses? */
1676 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1677 
1678 	/*
1679 	 * 	Configurable global rate limit.
1680 	 *
1681 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1682 	 *	bucket ratemask defines which icmp types are ratelimited by
1683 	 *	setting	it's bit position.
1684 	 *
1685 	 *	default:
1686 	 *	dest unreachable (3), source quench (4),
1687 	 *	time exceeded (11), parameter problem (12)
1688 	 */
1689 
1690 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1691 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1692 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1693 	net->ipv4.sysctl_icmp_errors_extension_mask = 0;
1694 	net->ipv4.sysctl_icmp_msgs_per_sec = 1000;
1695 	net->ipv4.sysctl_icmp_msgs_burst = 50;
1696 
1697 	return 0;
1698 }
1699 
1700 static struct pernet_operations __net_initdata icmp_sk_ops = {
1701        .init = icmp_sk_init,
1702 };
1703 
1704 int __init icmp_init(void)
1705 {
1706 	int err, i;
1707 
1708 	for_each_possible_cpu(i) {
1709 		struct sock *sk;
1710 
1711 		err = inet_ctl_sock_create(&sk, PF_INET,
1712 					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1713 		if (err < 0)
1714 			return err;
1715 
1716 		per_cpu(ipv4_icmp_sk, i) = sk;
1717 
1718 		/* Enough space for 2 64K ICMP packets, including
1719 		 * sk_buff/skb_shared_info struct overhead.
1720 		 */
1721 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1722 
1723 		/*
1724 		 * Speedup sock_wfree()
1725 		 */
1726 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1727 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1728 	}
1729 	return register_pernet_subsys(&icmp_sk_ops);
1730 }
1731