xref: /linux/net/ipv4/icmp.c (revision d8f87aa5fa0a4276491fa8ef436cd22605a3f9ba)
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 
116 	/* Must be last as it ends in a flexible-array member. */
117 	struct ip_options_rcu replyopts;
118 };
119 
120 /* An array of errno for error messages from dest unreach. */
121 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
122 
123 const struct icmp_err icmp_err_convert[] = {
124 	{
125 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
126 		.fatal = 0,
127 	},
128 	{
129 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
130 		.fatal = 0,
131 	},
132 	{
133 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
134 		.fatal = 1,
135 	},
136 	{
137 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
138 		.fatal = 1,
139 	},
140 	{
141 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
142 		.fatal = 0,
143 	},
144 	{
145 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
146 		.fatal = 0,
147 	},
148 	{
149 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
150 		.fatal = 1,
151 	},
152 	{
153 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
154 		.fatal = 1,
155 	},
156 	{
157 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
158 		.fatal = 1,
159 	},
160 	{
161 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
162 		.fatal = 1,
163 	},
164 	{
165 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
166 		.fatal = 1,
167 	},
168 	{
169 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
170 		.fatal = 0,
171 	},
172 	{
173 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
174 		.fatal = 0,
175 	},
176 	{
177 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
178 		.fatal = 1,
179 	},
180 	{
181 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
182 		.fatal = 1,
183 	},
184 	{
185 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
186 		.fatal = 1,
187 	},
188 };
189 EXPORT_SYMBOL(icmp_err_convert);
190 
191 /*
192  *	ICMP control array. This specifies what to do with each ICMP.
193  */
194 
195 struct icmp_control {
196 	enum skb_drop_reason (*handler)(struct sk_buff *skb);
197 	short   error;		/* This ICMP is classed as an error message */
198 };
199 
200 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
201 
202 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
203 
204 /* Called with BH disabled */
205 static inline struct sock *icmp_xmit_lock(struct net *net)
206 {
207 	struct sock *sk;
208 
209 	sk = this_cpu_read(ipv4_icmp_sk);
210 
211 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
212 		/* This can happen if the output path signals a
213 		 * dst_link_failure() for an outgoing ICMP packet.
214 		 */
215 		return NULL;
216 	}
217 	sock_net_set(sk, net);
218 	return sk;
219 }
220 
221 static inline void icmp_xmit_unlock(struct sock *sk)
222 {
223 	sock_net_set(sk, &init_net);
224 	spin_unlock(&sk->sk_lock.slock);
225 }
226 
227 /**
228  * icmp_global_allow - Are we allowed to send one more ICMP message ?
229  * @net: network namespace
230  *
231  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
232  * Returns false if we reached the limit and can not send another packet.
233  * Works in tandem with icmp_global_consume().
234  */
235 bool icmp_global_allow(struct net *net)
236 {
237 	u32 delta, now, oldstamp;
238 	int incr, new, old;
239 
240 	/* Note: many cpus could find this condition true.
241 	 * Then later icmp_global_consume() could consume more credits,
242 	 * this is an acceptable race.
243 	 */
244 	if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
245 		return true;
246 
247 	now = jiffies;
248 	oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
249 	delta = min_t(u32, now - oldstamp, HZ);
250 	if (delta < HZ / 50)
251 		return false;
252 
253 	incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ;
254 	if (!incr)
255 		return false;
256 
257 	if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
258 		old = atomic_read(&net->ipv4.icmp_global_credit);
259 		do {
260 			new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
261 		} while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
262 	}
263 	return true;
264 }
265 EXPORT_SYMBOL(icmp_global_allow);
266 
267 void icmp_global_consume(struct net *net)
268 {
269 	int credits = get_random_u32_below(3);
270 
271 	/* Note: this might make icmp_global.credit negative. */
272 	if (credits)
273 		atomic_sub(credits, &net->ipv4.icmp_global_credit);
274 }
275 EXPORT_SYMBOL(icmp_global_consume);
276 
277 static bool icmpv4_mask_allow(struct net *net, int type, int code)
278 {
279 	if (type > NR_ICMP_TYPES)
280 		return true;
281 
282 	/* Don't limit PMTU discovery. */
283 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
284 		return true;
285 
286 	/* Limit if icmp type is enabled in ratemask. */
287 	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
288 		return true;
289 
290 	return false;
291 }
292 
293 static bool icmpv4_global_allow(struct net *net, int type, int code,
294 				bool *apply_ratelimit)
295 {
296 	if (icmpv4_mask_allow(net, type, code))
297 		return true;
298 
299 	if (icmp_global_allow(net)) {
300 		*apply_ratelimit = true;
301 		return true;
302 	}
303 	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
304 	return false;
305 }
306 
307 /*
308  *	Send an ICMP frame.
309  */
310 
311 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
312 			       struct flowi4 *fl4, int type, int code,
313 			       bool apply_ratelimit)
314 {
315 	struct dst_entry *dst = &rt->dst;
316 	struct inet_peer *peer;
317 	struct net_device *dev;
318 	bool rc = true;
319 
320 	if (!apply_ratelimit)
321 		return true;
322 
323 	/* No rate limit on loopback */
324 	rcu_read_lock();
325 	dev = dst_dev_rcu(dst);
326 	if (dev && (dev->flags & IFF_LOOPBACK))
327 		goto out;
328 
329 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr,
330 			       l3mdev_master_ifindex_rcu(dev));
331 	rc = inet_peer_xrlim_allow(peer,
332 				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
333 out:
334 	rcu_read_unlock();
335 	if (!rc)
336 		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
337 	else
338 		icmp_global_consume(net);
339 	return rc;
340 }
341 
342 /*
343  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
344  */
345 void icmp_out_count(struct net *net, unsigned char type)
346 {
347 	ICMPMSGOUT_INC_STATS(net, type);
348 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
349 }
350 
351 /*
352  *	Checksum each fragment, and on the first include the headers and final
353  *	checksum.
354  */
355 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
356 			  struct sk_buff *skb)
357 {
358 	DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
359 			IP_OPTIONS_DATA_FIXED_SIZE);
360 	__wsum csum;
361 
362 	icmp_param = from;
363 
364 	csum = skb_copy_and_csum_bits(icmp_param->skb,
365 				      icmp_param->offset + offset,
366 				      to, len);
367 
368 	skb->csum = csum_block_add(skb->csum, csum, odd);
369 	if (icmp_pointers[icmp_param->data.icmph.type].error)
370 		nf_ct_attach(skb, icmp_param->skb);
371 	return 0;
372 }
373 
374 static void icmp_push_reply(struct sock *sk,
375 			    struct icmp_bxm *icmp_param,
376 			    struct flowi4 *fl4,
377 			    struct ipcm_cookie *ipc, struct rtable **rt)
378 {
379 	struct sk_buff *skb;
380 
381 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
382 			   icmp_param->data_len+icmp_param->head_len,
383 			   icmp_param->head_len,
384 			   ipc, rt, MSG_DONTWAIT) < 0) {
385 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
386 		ip_flush_pending_frames(sk);
387 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
388 		struct icmphdr *icmph = icmp_hdr(skb);
389 		__wsum csum;
390 		struct sk_buff *skb1;
391 
392 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
393 						 (char *)icmph,
394 						 icmp_param->head_len);
395 		skb_queue_walk(&sk->sk_write_queue, skb1) {
396 			csum = csum_add(csum, skb1->csum);
397 		}
398 		icmph->checksum = csum_fold(csum);
399 		skb->ip_summed = CHECKSUM_NONE;
400 		ip_push_pending_frames(sk, fl4);
401 	}
402 }
403 
404 /*
405  *	Driving logic for building and sending ICMP messages.
406  */
407 
408 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
409 {
410 	struct rtable *rt = skb_rtable(skb);
411 	struct net *net = dev_net_rcu(rt->dst.dev);
412 	bool apply_ratelimit = false;
413 	struct ipcm_cookie ipc;
414 	struct flowi4 fl4;
415 	struct sock *sk;
416 	__be32 daddr, saddr;
417 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
418 	int type = icmp_param->data.icmph.type;
419 	int code = icmp_param->data.icmph.code;
420 
421 	if (ip_options_echo(net, &icmp_param->replyopts.opt, skb))
422 		return;
423 
424 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
425 	local_bh_disable();
426 
427 	/* is global icmp_msgs_per_sec exhausted ? */
428 	if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
429 		goto out_bh_enable;
430 
431 	sk = icmp_xmit_lock(net);
432 	if (!sk)
433 		goto out_bh_enable;
434 
435 	icmp_param->data.icmph.checksum = 0;
436 
437 	ipcm_init(&ipc);
438 	ipc.tos = ip_hdr(skb)->tos;
439 	ipc.sockc.mark = mark;
440 	daddr = ipc.addr = ip_hdr(skb)->saddr;
441 	saddr = fib_compute_spec_dst(skb);
442 
443 	if (icmp_param->replyopts.opt.optlen) {
444 		ipc.opt = &icmp_param->replyopts;
445 		if (ipc.opt->opt.srr)
446 			daddr = icmp_param->replyopts.opt.faddr;
447 	}
448 	memset(&fl4, 0, sizeof(fl4));
449 	fl4.daddr = daddr;
450 	fl4.saddr = saddr;
451 	fl4.flowi4_mark = mark;
452 	fl4.flowi4_uid = sock_net_uid(net, NULL);
453 	fl4.flowi4_dscp = ip4h_dscp(ip_hdr(skb));
454 	fl4.flowi4_proto = IPPROTO_ICMP;
455 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
456 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
457 	rt = ip_route_output_key(net, &fl4);
458 	if (IS_ERR(rt))
459 		goto out_unlock;
460 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
461 		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
462 	ip_rt_put(rt);
463 out_unlock:
464 	icmp_xmit_unlock(sk);
465 out_bh_enable:
466 	local_bh_enable();
467 }
468 
469 /*
470  * The device used for looking up which routing table to use for sending an ICMP
471  * error is preferably the source whenever it is set, which should ensure the
472  * icmp error can be sent to the source host, else lookup using the routing
473  * table of the destination device, else use the main routing table (index 0).
474  */
475 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
476 {
477 	struct net_device *dev = skb->dev;
478 	const struct dst_entry *dst;
479 
480 	if (dev)
481 		return dev;
482 	dst = skb_dst(skb);
483 	return dst ? dst_dev(dst) : NULL;
484 }
485 
486 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4,
487 					struct sk_buff *skb_in,
488 					const struct iphdr *iph, __be32 saddr,
489 					dscp_t dscp, u32 mark, int type,
490 					int code, struct icmp_bxm *param)
491 {
492 	struct net_device *route_lookup_dev;
493 	struct dst_entry *dst, *dst2;
494 	struct rtable *rt, *rt2;
495 	struct flowi4 fl4_dec;
496 	int err;
497 
498 	memset(fl4, 0, sizeof(*fl4));
499 	fl4->daddr = (param->replyopts.opt.srr ?
500 		      param->replyopts.opt.faddr : iph->saddr);
501 	fl4->saddr = saddr;
502 	fl4->flowi4_mark = mark;
503 	fl4->flowi4_uid = sock_net_uid(net, NULL);
504 	fl4->flowi4_dscp = dscp;
505 	fl4->flowi4_proto = IPPROTO_ICMP;
506 	fl4->fl4_icmp_type = type;
507 	fl4->fl4_icmp_code = code;
508 	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
509 	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
510 
511 	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
512 	rt = ip_route_output_key_hash(net, fl4, skb_in);
513 	if (IS_ERR(rt))
514 		return rt;
515 
516 	/* No need to clone since we're just using its address. */
517 	rt2 = rt;
518 
519 	dst = xfrm_lookup(net, &rt->dst,
520 			  flowi4_to_flowi(fl4), NULL, 0);
521 	rt = dst_rtable(dst);
522 	if (!IS_ERR(dst)) {
523 		if (rt != rt2)
524 			return rt;
525 		if (inet_addr_type_dev_table(net, route_lookup_dev,
526 					     fl4->daddr) == RTN_LOCAL)
527 			return rt;
528 	} else if (PTR_ERR(dst) == -EPERM) {
529 		rt = NULL;
530 	} else {
531 		return rt;
532 	}
533 	err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
534 	if (err)
535 		goto relookup_failed;
536 
537 	if (inet_addr_type_dev_table(net, route_lookup_dev,
538 				     fl4_dec.saddr) == RTN_LOCAL) {
539 		rt2 = __ip_route_output_key(net, &fl4_dec);
540 		if (IS_ERR(rt2))
541 			err = PTR_ERR(rt2);
542 	} else {
543 		struct flowi4 fl4_2 = {};
544 		unsigned long orefdst;
545 
546 		fl4_2.daddr = fl4_dec.saddr;
547 		rt2 = ip_route_output_key(net, &fl4_2);
548 		if (IS_ERR(rt2)) {
549 			err = PTR_ERR(rt2);
550 			goto relookup_failed;
551 		}
552 		/* Ugh! */
553 		orefdst = skb_dstref_steal(skb_in);
554 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
555 				     dscp, rt2->dst.dev) ? -EINVAL : 0;
556 
557 		dst_release(&rt2->dst);
558 		rt2 = skb_rtable(skb_in);
559 		/* steal dst entry from skb_in, don't drop refcnt */
560 		skb_dstref_steal(skb_in);
561 		skb_dstref_restore(skb_in, orefdst);
562 	}
563 
564 	if (err)
565 		goto relookup_failed;
566 
567 	dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
568 			   XFRM_LOOKUP_ICMP);
569 	rt2 = dst_rtable(dst2);
570 	if (!IS_ERR(dst2)) {
571 		dst_release(&rt->dst);
572 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
573 		rt = rt2;
574 	} else if (PTR_ERR(dst2) == -EPERM) {
575 		if (rt)
576 			dst_release(&rt->dst);
577 		return rt2;
578 	} else {
579 		err = PTR_ERR(dst2);
580 		goto relookup_failed;
581 	}
582 	return rt;
583 
584 relookup_failed:
585 	if (rt)
586 		return rt;
587 	return ERR_PTR(err);
588 }
589 
590 struct icmp_ext_iio_addr4_subobj {
591 	__be16 afi;
592 	__be16 reserved;
593 	__be32 addr4;
594 };
595 
596 static unsigned int icmp_ext_iio_len(void)
597 {
598 	return sizeof(struct icmp_extobj_hdr) +
599 		/* ifIndex */
600 		sizeof(__be32) +
601 		/* Interface Address Sub-Object */
602 		sizeof(struct icmp_ext_iio_addr4_subobj) +
603 		/* Interface Name Sub-Object. Length must be a multiple of 4
604 		 * bytes.
605 		 */
606 		ALIGN(sizeof(struct icmp_ext_iio_name_subobj), 4) +
607 		/* MTU */
608 		sizeof(__be32);
609 }
610 
611 static unsigned int icmp_ext_max_len(u8 ext_objs)
612 {
613 	unsigned int ext_max_len;
614 
615 	ext_max_len = sizeof(struct icmp_ext_hdr);
616 
617 	if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
618 		ext_max_len += icmp_ext_iio_len();
619 
620 	return ext_max_len;
621 }
622 
623 static __be32 icmp_ext_iio_addr4_find(const struct net_device *dev)
624 {
625 	struct in_device *in_dev;
626 	struct in_ifaddr *ifa;
627 
628 	in_dev = __in_dev_get_rcu(dev);
629 	if (!in_dev)
630 		return 0;
631 
632 	/* It is unclear from RFC 5837 which IP address should be chosen, but
633 	 * it makes sense to choose a global unicast address.
634 	 */
635 	in_dev_for_each_ifa_rcu(ifa, in_dev) {
636 		if (READ_ONCE(ifa->ifa_flags) & IFA_F_SECONDARY)
637 			continue;
638 		if (ifa->ifa_scope != RT_SCOPE_UNIVERSE ||
639 		    ipv4_is_multicast(ifa->ifa_address))
640 			continue;
641 		return ifa->ifa_address;
642 	}
643 
644 	return 0;
645 }
646 
647 static void icmp_ext_iio_iif_append(struct net *net, struct sk_buff *skb,
648 				    int iif)
649 {
650 	struct icmp_ext_iio_name_subobj *name_subobj;
651 	struct icmp_extobj_hdr *objh;
652 	struct net_device *dev;
653 	__be32 data;
654 
655 	if (!iif)
656 		return;
657 
658 	/* Add the fields in the order specified by RFC 5837. */
659 	objh = skb_put(skb, sizeof(*objh));
660 	objh->class_num = ICMP_EXT_OBJ_CLASS_IIO;
661 	objh->class_type = ICMP_EXT_CTYPE_IIO_ROLE(ICMP_EXT_CTYPE_IIO_ROLE_IIF);
662 
663 	data = htonl(iif);
664 	skb_put_data(skb, &data, sizeof(__be32));
665 	objh->class_type |= ICMP_EXT_CTYPE_IIO_IFINDEX;
666 
667 	rcu_read_lock();
668 
669 	dev = dev_get_by_index_rcu(net, iif);
670 	if (!dev)
671 		goto out;
672 
673 	data = icmp_ext_iio_addr4_find(dev);
674 	if (data) {
675 		struct icmp_ext_iio_addr4_subobj *addr4_subobj;
676 
677 		addr4_subobj = skb_put_zero(skb, sizeof(*addr4_subobj));
678 		addr4_subobj->afi = htons(ICMP_AFI_IP);
679 		addr4_subobj->addr4 = data;
680 		objh->class_type |= ICMP_EXT_CTYPE_IIO_IPADDR;
681 	}
682 
683 	name_subobj = skb_put_zero(skb, ALIGN(sizeof(*name_subobj), 4));
684 	name_subobj->len = ALIGN(sizeof(*name_subobj), 4);
685 	netdev_copy_name(dev, name_subobj->name);
686 	objh->class_type |= ICMP_EXT_CTYPE_IIO_NAME;
687 
688 	data = htonl(READ_ONCE(dev->mtu));
689 	skb_put_data(skb, &data, sizeof(__be32));
690 	objh->class_type |= ICMP_EXT_CTYPE_IIO_MTU;
691 
692 out:
693 	rcu_read_unlock();
694 	objh->length = htons(skb_tail_pointer(skb) - (unsigned char *)objh);
695 }
696 
697 static void icmp_ext_objs_append(struct net *net, struct sk_buff *skb,
698 				 u8 ext_objs, int iif)
699 {
700 	if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
701 		icmp_ext_iio_iif_append(net, skb, iif);
702 }
703 
704 static struct sk_buff *
705 icmp_ext_append(struct net *net, struct sk_buff *skb_in, struct icmphdr *icmph,
706 		unsigned int room, int iif)
707 {
708 	unsigned int payload_len, ext_max_len, ext_len;
709 	struct icmp_ext_hdr *ext_hdr;
710 	struct sk_buff *skb;
711 	u8 ext_objs;
712 	int nhoff;
713 
714 	switch (icmph->type) {
715 	case ICMP_DEST_UNREACH:
716 	case ICMP_TIME_EXCEEDED:
717 	case ICMP_PARAMETERPROB:
718 		break;
719 	default:
720 		return NULL;
721 	}
722 
723 	ext_objs = READ_ONCE(net->ipv4.sysctl_icmp_errors_extension_mask);
724 	if (!ext_objs)
725 		return NULL;
726 
727 	ext_max_len = icmp_ext_max_len(ext_objs);
728 	if (ICMP_EXT_ORIG_DGRAM_MIN_LEN + ext_max_len > room)
729 		return NULL;
730 
731 	skb = skb_clone(skb_in, GFP_ATOMIC);
732 	if (!skb)
733 		return NULL;
734 
735 	nhoff = skb_network_offset(skb);
736 	payload_len = min(skb->len - nhoff, ICMP_EXT_ORIG_DGRAM_MIN_LEN);
737 
738 	if (!pskb_network_may_pull(skb, payload_len))
739 		goto free_skb;
740 
741 	if (pskb_trim(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN) ||
742 	    __skb_put_padto(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN, false))
743 		goto free_skb;
744 
745 	if (pskb_expand_head(skb, 0, ext_max_len, GFP_ATOMIC))
746 		goto free_skb;
747 
748 	ext_hdr = skb_put_zero(skb, sizeof(*ext_hdr));
749 	ext_hdr->version = ICMP_EXT_VERSION_2;
750 
751 	icmp_ext_objs_append(net, skb, ext_objs, iif);
752 
753 	/* Do not send an empty extension structure. */
754 	ext_len = skb_tail_pointer(skb) - (unsigned char *)ext_hdr;
755 	if (ext_len == sizeof(*ext_hdr))
756 		goto free_skb;
757 
758 	ext_hdr->checksum = ip_compute_csum(ext_hdr, ext_len);
759 	/* The length of the original datagram in 32-bit words (RFC 4884). */
760 	icmph->un.reserved[1] = ICMP_EXT_ORIG_DGRAM_MIN_LEN / sizeof(u32);
761 
762 	return skb;
763 
764 free_skb:
765 	consume_skb(skb);
766 	return NULL;
767 }
768 
769 /*
770  *	Send an ICMP message in response to a situation
771  *
772  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
773  *		  MAY send more (we do).
774  *			MUST NOT change this header information.
775  *			MUST NOT reply to a multicast/broadcast IP address.
776  *			MUST NOT reply to a multicast/broadcast MAC address.
777  *			MUST reply to only the first fragment.
778  */
779 
780 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
781 		 const struct inet_skb_parm *parm)
782 {
783 	DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
784 			IP_OPTIONS_DATA_FIXED_SIZE);
785 	struct iphdr *iph;
786 	int room;
787 	struct rtable *rt = skb_rtable(skb_in);
788 	bool apply_ratelimit = false;
789 	struct sk_buff *ext_skb;
790 	struct ipcm_cookie ipc;
791 	struct flowi4 fl4;
792 	__be32 saddr;
793 	u8  tos;
794 	u32 mark;
795 	struct net *net;
796 	struct sock *sk;
797 
798 	if (!rt)
799 		return;
800 
801 	rcu_read_lock();
802 
803 	if (rt->dst.dev)
804 		net = dev_net_rcu(rt->dst.dev);
805 	else if (skb_in->dev)
806 		net = dev_net_rcu(skb_in->dev);
807 	else
808 		goto out;
809 
810 	/*
811 	 *	Find the original header. It is expected to be valid, of course.
812 	 *	Check this, icmp_send is called from the most obscure devices
813 	 *	sometimes.
814 	 */
815 	iph = ip_hdr(skb_in);
816 
817 	if ((u8 *)iph < skb_in->head ||
818 	    (skb_network_header(skb_in) + sizeof(*iph)) >
819 	    skb_tail_pointer(skb_in))
820 		goto out;
821 
822 	/*
823 	 *	No replies to physical multicast/broadcast
824 	 */
825 	if (skb_in->pkt_type != PACKET_HOST)
826 		goto out;
827 
828 	/*
829 	 *	Now check at the protocol level
830 	 */
831 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
832 		goto out;
833 
834 	/*
835 	 *	Only reply to fragment 0. We byte re-order the constant
836 	 *	mask for efficiency.
837 	 */
838 	if (iph->frag_off & htons(IP_OFFSET))
839 		goto out;
840 
841 	/*
842 	 *	If we send an ICMP error to an ICMP error a mess would result..
843 	 */
844 	if (icmp_pointers[type].error) {
845 		/*
846 		 *	We are an error, check if we are replying to an
847 		 *	ICMP error
848 		 */
849 		if (iph->protocol == IPPROTO_ICMP) {
850 			u8 _inner_type, *itp;
851 
852 			itp = skb_header_pointer(skb_in,
853 						 skb_network_header(skb_in) +
854 						 (iph->ihl << 2) +
855 						 offsetof(struct icmphdr,
856 							  type) -
857 						 skb_in->data,
858 						 sizeof(_inner_type),
859 						 &_inner_type);
860 			if (!itp)
861 				goto out;
862 
863 			/*
864 			 *	Assume any unknown ICMP type is an error. This
865 			 *	isn't specified by the RFC, but think about it..
866 			 */
867 			if (*itp > NR_ICMP_TYPES ||
868 			    icmp_pointers[*itp].error)
869 				goto out;
870 		}
871 	}
872 
873 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
874 	local_bh_disable();
875 
876 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
877 	 * incoming dev is loopback.  If outgoing dev change to not be
878 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
879 	 */
880 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
881 	      !icmpv4_global_allow(net, type, code, &apply_ratelimit))
882 		goto out_bh_enable;
883 
884 	sk = icmp_xmit_lock(net);
885 	if (!sk)
886 		goto out_bh_enable;
887 
888 	/*
889 	 *	Construct source address and options.
890 	 */
891 
892 	saddr = iph->daddr;
893 	if (!(rt->rt_flags & RTCF_LOCAL)) {
894 		struct net_device *dev = NULL;
895 
896 		rcu_read_lock();
897 		if (rt_is_input_route(rt) &&
898 		    READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
899 			dev = dev_get_by_index_rcu(net, parm->iif ? parm->iif :
900 						   inet_iif(skb_in));
901 
902 		if (dev)
903 			saddr = inet_select_addr(dev, iph->saddr,
904 						 RT_SCOPE_LINK);
905 		else
906 			saddr = 0;
907 		rcu_read_unlock();
908 	}
909 
910 	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
911 					   IPTOS_PREC_INTERNETCONTROL) :
912 					   iph->tos;
913 	mark = IP4_REPLY_MARK(net, skb_in->mark);
914 
915 	if (__ip_options_echo(net, &icmp_param->replyopts.opt, skb_in,
916 			      &parm->opt))
917 		goto out_unlock;
918 
919 
920 	/*
921 	 *	Prepare data for ICMP header.
922 	 */
923 
924 	icmp_param->data.icmph.type	 = type;
925 	icmp_param->data.icmph.code	 = code;
926 	icmp_param->data.icmph.un.gateway = info;
927 	icmp_param->data.icmph.checksum	 = 0;
928 	icmp_param->skb	  = skb_in;
929 	icmp_param->offset = skb_network_offset(skb_in);
930 	ipcm_init(&ipc);
931 	ipc.tos = tos;
932 	ipc.addr = iph->saddr;
933 	ipc.opt = &icmp_param->replyopts;
934 	ipc.sockc.mark = mark;
935 
936 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
937 			       inet_dsfield_to_dscp(tos), mark, type, code,
938 			       icmp_param);
939 	if (IS_ERR(rt))
940 		goto out_unlock;
941 
942 	/* peer icmp_ratelimit */
943 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
944 		goto ende;
945 
946 	/* RFC says return as much as we can without exceeding 576 bytes. */
947 
948 	room = dst_mtu(&rt->dst);
949 	if (room > 576)
950 		room = 576;
951 	room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.optlen;
952 	room -= sizeof(struct icmphdr);
953 	/* Guard against tiny mtu. We need to include at least one
954 	 * IP network header for this message to make any sense.
955 	 */
956 	if (room <= (int)sizeof(struct iphdr))
957 		goto ende;
958 
959 	ext_skb = icmp_ext_append(net, skb_in, &icmp_param->data.icmph, room,
960 				  parm->iif);
961 	if (ext_skb)
962 		icmp_param->skb = ext_skb;
963 
964 	icmp_param->data_len = icmp_param->skb->len - icmp_param->offset;
965 	if (icmp_param->data_len > room)
966 		icmp_param->data_len = room;
967 	icmp_param->head_len = sizeof(struct icmphdr);
968 
969 	/* if we don't have a source address at this point, fall back to the
970 	 * dummy address instead of sending out a packet with a source address
971 	 * of 0.0.0.0
972 	 */
973 	if (!fl4.saddr)
974 		fl4.saddr = htonl(INADDR_DUMMY);
975 
976 	trace_icmp_send(skb_in, type, code);
977 
978 	icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
979 
980 	if (ext_skb)
981 		consume_skb(ext_skb);
982 ende:
983 	ip_rt_put(rt);
984 out_unlock:
985 	icmp_xmit_unlock(sk);
986 out_bh_enable:
987 	local_bh_enable();
988 out:
989 	rcu_read_unlock();
990 }
991 EXPORT_SYMBOL(__icmp_send);
992 
993 #if IS_ENABLED(CONFIG_NF_NAT)
994 #include <net/netfilter/nf_conntrack.h>
995 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
996 {
997 	struct sk_buff *cloned_skb = NULL;
998 	enum ip_conntrack_info ctinfo;
999 	enum ip_conntrack_dir dir;
1000 	struct inet_skb_parm parm;
1001 	struct nf_conn *ct;
1002 	__be32 orig_ip;
1003 
1004 	memset(&parm, 0, sizeof(parm));
1005 	ct = nf_ct_get(skb_in, &ctinfo);
1006 	if (!ct || !(READ_ONCE(ct->status) & IPS_NAT_MASK)) {
1007 		__icmp_send(skb_in, type, code, info, &parm);
1008 		return;
1009 	}
1010 
1011 	if (skb_shared(skb_in))
1012 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
1013 
1014 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
1015 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
1016 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
1017 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
1018 		goto out;
1019 
1020 	orig_ip = ip_hdr(skb_in)->saddr;
1021 	dir = CTINFO2DIR(ctinfo);
1022 	ip_hdr(skb_in)->saddr = ct->tuplehash[dir].tuple.src.u3.ip;
1023 	__icmp_send(skb_in, type, code, info, &parm);
1024 	ip_hdr(skb_in)->saddr = orig_ip;
1025 out:
1026 	consume_skb(cloned_skb);
1027 }
1028 EXPORT_SYMBOL(icmp_ndo_send);
1029 #endif
1030 
1031 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
1032 {
1033 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1034 	const struct net_protocol *ipprot;
1035 	int protocol = iph->protocol;
1036 
1037 	/* Checkin full IP header plus 8 bytes of protocol to
1038 	 * avoid additional coding at protocol handlers.
1039 	 */
1040 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
1041 		__ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1042 		return;
1043 	}
1044 
1045 	raw_icmp_error(skb, protocol, info);
1046 
1047 	ipprot = rcu_dereference(inet_protos[protocol]);
1048 	if (ipprot && ipprot->err_handler)
1049 		ipprot->err_handler(skb, info);
1050 }
1051 
1052 static bool icmp_tag_validation(int proto)
1053 {
1054 	bool ok;
1055 
1056 	rcu_read_lock();
1057 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
1058 	rcu_read_unlock();
1059 	return ok;
1060 }
1061 
1062 /*
1063  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
1064  *	ICMP_PARAMETERPROB.
1065  */
1066 
1067 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
1068 {
1069 	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
1070 	const struct iphdr *iph;
1071 	struct icmphdr *icmph;
1072 	struct net *net;
1073 	u32 info = 0;
1074 
1075 	net = skb_dst_dev_net_rcu(skb);
1076 
1077 	/*
1078 	 *	Incomplete header ?
1079 	 * 	Only checks for the IP header, there should be an
1080 	 *	additional check for longer headers in upper levels.
1081 	 */
1082 
1083 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1084 		goto out_err;
1085 
1086 	icmph = icmp_hdr(skb);
1087 	iph   = (const struct iphdr *)skb->data;
1088 
1089 	if (iph->ihl < 5)  { /* Mangled header, drop. */
1090 		reason = SKB_DROP_REASON_IP_INHDR;
1091 		goto out_err;
1092 	}
1093 
1094 	switch (icmph->type) {
1095 	case ICMP_DEST_UNREACH:
1096 		switch (icmph->code & 15) {
1097 		case ICMP_NET_UNREACH:
1098 		case ICMP_HOST_UNREACH:
1099 		case ICMP_PROT_UNREACH:
1100 		case ICMP_PORT_UNREACH:
1101 			break;
1102 		case ICMP_FRAG_NEEDED:
1103 			/* for documentation of the ip_no_pmtu_disc
1104 			 * values please see
1105 			 * Documentation/networking/ip-sysctl.rst
1106 			 */
1107 			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
1108 			default:
1109 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
1110 						    &iph->daddr);
1111 				break;
1112 			case 2:
1113 				goto out;
1114 			case 3:
1115 				if (!icmp_tag_validation(iph->protocol))
1116 					goto out;
1117 				fallthrough;
1118 			case 0:
1119 				info = ntohs(icmph->un.frag.mtu);
1120 			}
1121 			break;
1122 		case ICMP_SR_FAILED:
1123 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
1124 					    &iph->daddr);
1125 			break;
1126 		default:
1127 			break;
1128 		}
1129 		if (icmph->code > NR_ICMP_UNREACH)
1130 			goto out;
1131 		break;
1132 	case ICMP_PARAMETERPROB:
1133 		info = ntohl(icmph->un.gateway) >> 24;
1134 		break;
1135 	case ICMP_TIME_EXCEEDED:
1136 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
1137 		if (icmph->code == ICMP_EXC_FRAGTIME)
1138 			goto out;
1139 		break;
1140 	}
1141 
1142 	/*
1143 	 *	Throw it at our lower layers
1144 	 *
1145 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
1146 	 *		  header.
1147 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
1148 	 *		  transport layer.
1149 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
1150 	 *		  transport layer.
1151 	 */
1152 
1153 	/*
1154 	 *	Check the other end isn't violating RFC 1122. Some routers send
1155 	 *	bogus responses to broadcast frames. If you see this message
1156 	 *	first check your netmask matches at both ends, if it does then
1157 	 *	get the other vendor to fix their kit.
1158 	 */
1159 
1160 	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
1161 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
1162 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
1163 				     &ip_hdr(skb)->saddr,
1164 				     icmph->type, icmph->code,
1165 				     &iph->daddr, skb->dev->name);
1166 		goto out;
1167 	}
1168 
1169 	icmp_socket_deliver(skb, info);
1170 
1171 out:
1172 	return reason;
1173 out_err:
1174 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1175 	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
1176 }
1177 
1178 
1179 /*
1180  *	Handle ICMP_REDIRECT.
1181  */
1182 
1183 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
1184 {
1185 	if (skb->len < sizeof(struct iphdr)) {
1186 		__ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1187 		return SKB_DROP_REASON_PKT_TOO_SMALL;
1188 	}
1189 
1190 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
1191 		/* there aught to be a stat */
1192 		return SKB_DROP_REASON_NOMEM;
1193 	}
1194 
1195 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
1196 	return SKB_NOT_DROPPED_YET;
1197 }
1198 
1199 /*
1200  *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1201  *
1202  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1203  *		  requests.
1204  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1205  *		  included in the reply.
1206  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1207  *		  echo requests, MUST have default=NOT.
1208  *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
1209  *		  Extended Echo Functionality, MUST be disabled by default
1210  *	See also WRT handling of options once they are done and working.
1211  */
1212 
1213 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1214 {
1215 	DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1216 			IP_OPTIONS_DATA_FIXED_SIZE);
1217 	struct net *net;
1218 
1219 	net = skb_dst_dev_net_rcu(skb);
1220 	/* should there be an ICMP stat for ignored echos? */
1221 	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1222 		return SKB_NOT_DROPPED_YET;
1223 
1224 	icmp_param->data.icmph	   = *icmp_hdr(skb);
1225 	icmp_param->skb		   = skb;
1226 	icmp_param->offset	   = 0;
1227 	icmp_param->data_len	   = skb->len;
1228 	icmp_param->head_len	   = sizeof(struct icmphdr);
1229 
1230 	if (icmp_param->data.icmph.type == ICMP_ECHO)
1231 		icmp_param->data.icmph.type = ICMP_ECHOREPLY;
1232 	else if (!icmp_build_probe(skb, &icmp_param->data.icmph))
1233 		return SKB_NOT_DROPPED_YET;
1234 
1235 	icmp_reply(icmp_param, skb);
1236 	return SKB_NOT_DROPPED_YET;
1237 }
1238 
1239 /*	Helper for icmp_echo and icmpv6_echo_reply.
1240  *	Searches for net_device that matches PROBE interface identifier
1241  *		and builds PROBE reply message in icmphdr.
1242  *
1243  *	Returns false if PROBE responses are disabled via sysctl
1244  */
1245 
1246 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1247 {
1248 	struct net *net = dev_net_rcu(skb->dev);
1249 	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1250 	struct icmp_ext_echo_iio *iio, _iio;
1251 	struct inet6_dev *in6_dev;
1252 	struct in_device *in_dev;
1253 	struct net_device *dev;
1254 	char buff[IFNAMSIZ];
1255 	u16 ident_len;
1256 	u8 status;
1257 
1258 	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1259 		return false;
1260 
1261 	/* We currently only support probing interfaces on the proxy node
1262 	 * Check to ensure L-bit is set
1263 	 */
1264 	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1265 		return false;
1266 	/* Clear status bits in reply message */
1267 	icmphdr->un.echo.sequence &= htons(0xFF00);
1268 	if (icmphdr->type == ICMP_EXT_ECHO)
1269 		icmphdr->type = ICMP_EXT_ECHOREPLY;
1270 	else
1271 		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1272 	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1273 	/* Size of iio is class_type dependent.
1274 	 * Only check header here and assign length based on ctype in the switch statement
1275 	 */
1276 	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1277 	if (!ext_hdr || !iio)
1278 		goto send_mal_query;
1279 	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1280 	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1281 		goto send_mal_query;
1282 	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1283 	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1284 				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1285 	if (!iio)
1286 		goto send_mal_query;
1287 
1288 	status = 0;
1289 	dev = NULL;
1290 	switch (iio->extobj_hdr.class_type) {
1291 	case ICMP_EXT_ECHO_CTYPE_NAME:
1292 		if (ident_len >= IFNAMSIZ)
1293 			goto send_mal_query;
1294 		memset(buff, 0, sizeof(buff));
1295 		memcpy(buff, &iio->ident.name, ident_len);
1296 		dev = dev_get_by_name(net, buff);
1297 		break;
1298 	case ICMP_EXT_ECHO_CTYPE_INDEX:
1299 		if (ident_len != sizeof(iio->ident.ifindex))
1300 			goto send_mal_query;
1301 		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1302 		break;
1303 	case ICMP_EXT_ECHO_CTYPE_ADDR:
1304 		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1305 		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1306 				 iio->ident.addr.ctype3_hdr.addrlen)
1307 			goto send_mal_query;
1308 		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1309 		case ICMP_AFI_IP:
1310 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1311 				goto send_mal_query;
1312 			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1313 			break;
1314 #if IS_ENABLED(CONFIG_IPV6)
1315 		case ICMP_AFI_IP6:
1316 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1317 				goto send_mal_query;
1318 			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1319 			dev_hold(dev);
1320 			break;
1321 #endif
1322 		default:
1323 			goto send_mal_query;
1324 		}
1325 		break;
1326 	default:
1327 		goto send_mal_query;
1328 	}
1329 	if (!dev) {
1330 		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1331 		return true;
1332 	}
1333 	/* Fill bits in reply message */
1334 	if (dev->flags & IFF_UP)
1335 		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1336 
1337 	in_dev = __in_dev_get_rcu(dev);
1338 	if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1339 		status |= ICMP_EXT_ECHOREPLY_IPV4;
1340 
1341 	in6_dev = __in6_dev_get(dev);
1342 	if (in6_dev && !list_empty(&in6_dev->addr_list))
1343 		status |= ICMP_EXT_ECHOREPLY_IPV6;
1344 
1345 	dev_put(dev);
1346 	icmphdr->un.echo.sequence |= htons(status);
1347 	return true;
1348 send_mal_query:
1349 	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1350 	return true;
1351 }
1352 EXPORT_SYMBOL_GPL(icmp_build_probe);
1353 
1354 /*
1355  *	Handle ICMP Timestamp requests.
1356  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1357  *		  SHOULD be in the kernel for minimum random latency.
1358  *		  MUST be accurate to a few minutes.
1359  *		  MUST be updated at least at 15Hz.
1360  */
1361 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1362 {
1363 	DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1364 			IP_OPTIONS_DATA_FIXED_SIZE);
1365 	/*
1366 	 *	Too short.
1367 	 */
1368 	if (skb->len < 4)
1369 		goto out_err;
1370 
1371 	/*
1372 	 *	Fill in the current time as ms since midnight UT:
1373 	 */
1374 	icmp_param->data.times[1] = inet_current_timestamp();
1375 	icmp_param->data.times[2] = icmp_param->data.times[1];
1376 
1377 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param->data.times[0], 4));
1378 
1379 	icmp_param->data.icmph	   = *icmp_hdr(skb);
1380 	icmp_param->data.icmph.type = ICMP_TIMESTAMPREPLY;
1381 	icmp_param->data.icmph.code = 0;
1382 	icmp_param->skb		   = skb;
1383 	icmp_param->offset	   = 0;
1384 	icmp_param->data_len	   = 0;
1385 	icmp_param->head_len	   = sizeof(struct icmphdr) + 12;
1386 	icmp_reply(icmp_param, skb);
1387 	return SKB_NOT_DROPPED_YET;
1388 
1389 out_err:
1390 	__ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS);
1391 	return SKB_DROP_REASON_PKT_TOO_SMALL;
1392 }
1393 
1394 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1395 {
1396 	/* pretend it was a success */
1397 	return SKB_NOT_DROPPED_YET;
1398 }
1399 
1400 /*
1401  *	Deal with incoming ICMP packets.
1402  */
1403 int icmp_rcv(struct sk_buff *skb)
1404 {
1405 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1406 	struct rtable *rt = skb_rtable(skb);
1407 	struct net *net = dev_net_rcu(rt->dst.dev);
1408 	struct icmphdr *icmph;
1409 
1410 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1411 		struct sec_path *sp = skb_sec_path(skb);
1412 		int nh;
1413 
1414 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1415 				 XFRM_STATE_ICMP)) {
1416 			reason = SKB_DROP_REASON_XFRM_POLICY;
1417 			goto drop;
1418 		}
1419 
1420 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1421 			goto drop;
1422 
1423 		nh = skb_network_offset(skb);
1424 		skb_set_network_header(skb, sizeof(*icmph));
1425 
1426 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1427 						skb)) {
1428 			reason = SKB_DROP_REASON_XFRM_POLICY;
1429 			goto drop;
1430 		}
1431 
1432 		skb_set_network_header(skb, nh);
1433 	}
1434 
1435 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1436 
1437 	if (skb_checksum_simple_validate(skb))
1438 		goto csum_error;
1439 
1440 	if (!pskb_pull(skb, sizeof(*icmph)))
1441 		goto error;
1442 
1443 	icmph = icmp_hdr(skb);
1444 
1445 	ICMPMSGIN_INC_STATS(net, icmph->type);
1446 
1447 	/* Check for ICMP Extended Echo (PROBE) messages */
1448 	if (icmph->type == ICMP_EXT_ECHO) {
1449 		/* We can't use icmp_pointers[].handler() because it is an array of
1450 		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1451 		 */
1452 		reason = icmp_echo(skb);
1453 		goto reason_check;
1454 	}
1455 
1456 	/*
1457 	 *	Parse the ICMP message
1458 	 */
1459 
1460 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1461 		/*
1462 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1463 		 *	  silently ignored (we let user decide with a sysctl).
1464 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1465 		 *	  discarded if to broadcast/multicast.
1466 		 */
1467 		if ((icmph->type == ICMP_ECHO ||
1468 		     icmph->type == ICMP_TIMESTAMP) &&
1469 		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1470 			reason = SKB_DROP_REASON_INVALID_PROTO;
1471 			goto error;
1472 		}
1473 		if (icmph->type != ICMP_ECHO &&
1474 		    icmph->type != ICMP_TIMESTAMP &&
1475 		    icmph->type != ICMP_ADDRESS &&
1476 		    icmph->type != ICMP_ADDRESSREPLY) {
1477 			reason = SKB_DROP_REASON_INVALID_PROTO;
1478 			goto error;
1479 		}
1480 	}
1481 
1482 	if (icmph->type == ICMP_EXT_ECHOREPLY ||
1483 	    icmph->type == ICMP_ECHOREPLY) {
1484 		reason = ping_rcv(skb);
1485 		return reason ? NET_RX_DROP : NET_RX_SUCCESS;
1486 	}
1487 
1488 	/*
1489 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1490 	 *
1491 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1492 	 *		  discarded.
1493 	 */
1494 	if (icmph->type > NR_ICMP_TYPES) {
1495 		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1496 		goto error;
1497 	}
1498 
1499 	reason = icmp_pointers[icmph->type].handler(skb);
1500 reason_check:
1501 	if (!reason)  {
1502 		consume_skb(skb);
1503 		return NET_RX_SUCCESS;
1504 	}
1505 
1506 drop:
1507 	kfree_skb_reason(skb, reason);
1508 	return NET_RX_DROP;
1509 csum_error:
1510 	reason = SKB_DROP_REASON_ICMP_CSUM;
1511 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1512 error:
1513 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1514 	goto drop;
1515 }
1516 
1517 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1518 {
1519 	struct icmp_extobj_hdr *objh, _objh;
1520 	struct icmp_ext_hdr *exth, _exth;
1521 	u16 olen;
1522 
1523 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1524 	if (!exth)
1525 		return false;
1526 	if (exth->version != 2)
1527 		return true;
1528 
1529 	if (exth->checksum &&
1530 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1531 		return false;
1532 
1533 	off += sizeof(_exth);
1534 	while (off < skb->len) {
1535 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1536 		if (!objh)
1537 			return false;
1538 
1539 		olen = ntohs(objh->length);
1540 		if (olen < sizeof(_objh))
1541 			return false;
1542 
1543 		off += olen;
1544 		if (off > skb->len)
1545 			return false;
1546 	}
1547 
1548 	return true;
1549 }
1550 
1551 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1552 			   struct sock_ee_data_rfc4884 *out,
1553 			   int thlen, int off)
1554 {
1555 	int hlen;
1556 
1557 	/* original datagram headers: end of icmph to payload (skb->data) */
1558 	hlen = -skb_transport_offset(skb) - thlen;
1559 
1560 	/* per rfc 4884: minimal datagram length of 128 bytes */
1561 	if (off < 128 || off < hlen)
1562 		return;
1563 
1564 	/* kernel has stripped headers: return payload offset in bytes */
1565 	off -= hlen;
1566 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1567 		return;
1568 
1569 	out->len = off;
1570 
1571 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1572 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1573 }
1574 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1575 
1576 int icmp_err(struct sk_buff *skb, u32 info)
1577 {
1578 	struct iphdr *iph = (struct iphdr *)skb->data;
1579 	int offset = iph->ihl<<2;
1580 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1581 	struct net *net = dev_net_rcu(skb->dev);
1582 	int type = icmp_hdr(skb)->type;
1583 	int code = icmp_hdr(skb)->code;
1584 
1585 	/*
1586 	 * Use ping_err to handle all icmp errors except those
1587 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1588 	 */
1589 	if (icmph->type != ICMP_ECHOREPLY) {
1590 		ping_err(skb, offset, info);
1591 		return 0;
1592 	}
1593 
1594 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1595 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1596 	else if (type == ICMP_REDIRECT)
1597 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1598 
1599 	return 0;
1600 }
1601 
1602 /*
1603  *	This table is the definition of how we handle ICMP.
1604  */
1605 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1606 	[ICMP_ECHOREPLY] = {
1607 		.handler = ping_rcv,
1608 	},
1609 	[1] = {
1610 		.handler = icmp_discard,
1611 		.error = 1,
1612 	},
1613 	[2] = {
1614 		.handler = icmp_discard,
1615 		.error = 1,
1616 	},
1617 	[ICMP_DEST_UNREACH] = {
1618 		.handler = icmp_unreach,
1619 		.error = 1,
1620 	},
1621 	[ICMP_SOURCE_QUENCH] = {
1622 		.handler = icmp_unreach,
1623 		.error = 1,
1624 	},
1625 	[ICMP_REDIRECT] = {
1626 		.handler = icmp_redirect,
1627 		.error = 1,
1628 	},
1629 	[6] = {
1630 		.handler = icmp_discard,
1631 		.error = 1,
1632 	},
1633 	[7] = {
1634 		.handler = icmp_discard,
1635 		.error = 1,
1636 	},
1637 	[ICMP_ECHO] = {
1638 		.handler = icmp_echo,
1639 	},
1640 	[9] = {
1641 		.handler = icmp_discard,
1642 		.error = 1,
1643 	},
1644 	[10] = {
1645 		.handler = icmp_discard,
1646 		.error = 1,
1647 	},
1648 	[ICMP_TIME_EXCEEDED] = {
1649 		.handler = icmp_unreach,
1650 		.error = 1,
1651 	},
1652 	[ICMP_PARAMETERPROB] = {
1653 		.handler = icmp_unreach,
1654 		.error = 1,
1655 	},
1656 	[ICMP_TIMESTAMP] = {
1657 		.handler = icmp_timestamp,
1658 	},
1659 	[ICMP_TIMESTAMPREPLY] = {
1660 		.handler = icmp_discard,
1661 	},
1662 	[ICMP_INFO_REQUEST] = {
1663 		.handler = icmp_discard,
1664 	},
1665 	[ICMP_INFO_REPLY] = {
1666 		.handler = icmp_discard,
1667 	},
1668 	[ICMP_ADDRESS] = {
1669 		.handler = icmp_discard,
1670 	},
1671 	[ICMP_ADDRESSREPLY] = {
1672 		.handler = icmp_discard,
1673 	},
1674 };
1675 
1676 static int __net_init icmp_sk_init(struct net *net)
1677 {
1678 	/* Control parameters for ECHO replies. */
1679 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1680 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1681 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1682 
1683 	/* Control parameter - ignore bogus broadcast responses? */
1684 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1685 
1686 	/*
1687 	 * 	Configurable global rate limit.
1688 	 *
1689 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1690 	 *	bucket ratemask defines which icmp types are ratelimited by
1691 	 *	setting	it's bit position.
1692 	 *
1693 	 *	default:
1694 	 *	dest unreachable (3), source quench (4),
1695 	 *	time exceeded (11), parameter problem (12)
1696 	 */
1697 
1698 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1699 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1700 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1701 	net->ipv4.sysctl_icmp_errors_extension_mask = 0;
1702 	net->ipv4.sysctl_icmp_msgs_per_sec = 1000;
1703 	net->ipv4.sysctl_icmp_msgs_burst = 50;
1704 
1705 	return 0;
1706 }
1707 
1708 static struct pernet_operations __net_initdata icmp_sk_ops = {
1709        .init = icmp_sk_init,
1710 };
1711 
1712 int __init icmp_init(void)
1713 {
1714 	int err, i;
1715 
1716 	for_each_possible_cpu(i) {
1717 		struct sock *sk;
1718 
1719 		err = inet_ctl_sock_create(&sk, PF_INET,
1720 					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1721 		if (err < 0)
1722 			return err;
1723 
1724 		per_cpu(ipv4_icmp_sk, i) = sk;
1725 
1726 		/* Enough space for 2 64K ICMP packets, including
1727 		 * sk_buff/skb_shared_info struct overhead.
1728 		 */
1729 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1730 
1731 		/*
1732 		 * Speedup sock_wfree()
1733 		 */
1734 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1735 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1736 	}
1737 	return register_pernet_subsys(&icmp_sk_ops);
1738 }
1739