xref: /openbmc/linux/net/ipv4/icmp.c (revision 5e012745)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	NET3:	Implementation of the ICMP protocol layer.
4  *
5  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
6  *
7  *	Some of the function names and the icmp unreach table for this
8  *	module were derived from [icmp.c 1.0.11 06/02/93] by
9  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
10  *	Other than that this module is a complete rewrite.
11  *
12  *	Fixes:
13  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
14  *					with icmp type masking ability instead
15  *					of broken per type icmp timeouts.
16  *		Mike Shaver	:	RFC1122 checks.
17  *		Alan Cox	:	Multicast ping reply as self.
18  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
19  *					call.
20  *		Alan Cox	:	Added 216,128 byte paths to the MTU
21  *					code.
22  *		Martin Mares	:	RFC1812 checks.
23  *		Martin Mares	:	Can be configured to follow redirects
24  *					if acting as a router _without_ a
25  *					routing protocol (RFC 1812).
26  *		Martin Mares	:	Echo requests may be configured to
27  *					be ignored (RFC 1812).
28  *		Martin Mares	:	Limitation of ICMP error message
29  *					transmit rate (RFC 1812).
30  *		Martin Mares	:	TOS and Precedence set correctly
31  *					(RFC 1812).
32  *		Martin Mares	:	Now copying as much data from the
33  *					original packet as we can without
34  *					exceeding 576 bytes (RFC 1812).
35  *	Willy Konynenberg	:	Transparent proxying support.
36  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
37  *					path MTU bug.
38  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
39  *					valid (RFC 1812).
40  *		Andi Kleen	:	Check all packet lengths properly
41  *					and moved all kfree_skb() up to
42  *					icmp_rcv.
43  *		Andi Kleen	:	Move the rate limit bookkeeping
44  *					into the dest entry and use a token
45  *					bucket filter (thanks to ANK). Make
46  *					the rates sysctl configurable.
47  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
48  *					- IP option length was accounted wrongly
49  *					- ICMP header length was not accounted
50  *					  at all.
51  *              Tristan Greaves :       Added sysctl option to ignore bogus
52  *              			broadcast responses from broken routers.
53  *
54  * To Fix:
55  *
56  *	- Should use skb_pull() instead of all the manual checking.
57  *	  This would also greatly simply some upper layer error handlers. --AK
58  */
59 
60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
61 
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/jiffies.h>
65 #include <linux/kernel.h>
66 #include <linux/fcntl.h>
67 #include <linux/socket.h>
68 #include <linux/in.h>
69 #include <linux/inet.h>
70 #include <linux/inetdevice.h>
71 #include <linux/netdevice.h>
72 #include <linux/string.h>
73 #include <linux/netfilter_ipv4.h>
74 #include <linux/slab.h>
75 #include <net/snmp.h>
76 #include <net/ip.h>
77 #include <net/route.h>
78 #include <net/protocol.h>
79 #include <net/icmp.h>
80 #include <net/tcp.h>
81 #include <net/udp.h>
82 #include <net/raw.h>
83 #include <net/ping.h>
84 #include <linux/skbuff.h>
85 #include <net/sock.h>
86 #include <linux/errno.h>
87 #include <linux/timer.h>
88 #include <linux/init.h>
89 #include <linux/uaccess.h>
90 #include <net/checksum.h>
91 #include <net/xfrm.h>
92 #include <net/inet_common.h>
93 #include <net/ip_fib.h>
94 #include <net/l3mdev.h>
95 
96 /*
97  *	Build xmit assembly blocks
98  */
99 
100 struct icmp_bxm {
101 	struct sk_buff *skb;
102 	int offset;
103 	int data_len;
104 
105 	struct {
106 		struct icmphdr icmph;
107 		__be32	       times[3];
108 	} data;
109 	int head_len;
110 	struct ip_options_data replyopts;
111 };
112 
113 /* An array of errno for error messages from dest unreach. */
114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
115 
116 const struct icmp_err icmp_err_convert[] = {
117 	{
118 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
119 		.fatal = 0,
120 	},
121 	{
122 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
123 		.fatal = 0,
124 	},
125 	{
126 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
127 		.fatal = 1,
128 	},
129 	{
130 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
131 		.fatal = 1,
132 	},
133 	{
134 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
135 		.fatal = 0,
136 	},
137 	{
138 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
139 		.fatal = 0,
140 	},
141 	{
142 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
143 		.fatal = 1,
144 	},
145 	{
146 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
147 		.fatal = 1,
148 	},
149 	{
150 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
163 		.fatal = 0,
164 	},
165 	{
166 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
167 		.fatal = 0,
168 	},
169 	{
170 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
171 		.fatal = 1,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
175 		.fatal = 1,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
179 		.fatal = 1,
180 	},
181 };
182 EXPORT_SYMBOL(icmp_err_convert);
183 
184 /*
185  *	ICMP control array. This specifies what to do with each ICMP.
186  */
187 
188 struct icmp_control {
189 	bool (*handler)(struct sk_buff *skb);
190 	short   error;		/* This ICMP is classed as an error message */
191 };
192 
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194 
195 /*
196  *	The ICMP socket(s). This is the most convenient way to flow control
197  *	our ICMP output as well as maintain a clean interface throughout
198  *	all layers. All Socketless IP sends will soon be gone.
199  *
200  *	On SMP we have one ICMP socket per-cpu.
201  */
202 static struct sock *icmp_sk(struct net *net)
203 {
204 	return this_cpu_read(*net->ipv4.icmp_sk);
205 }
206 
207 /* Called with BH disabled */
208 static inline struct sock *icmp_xmit_lock(struct net *net)
209 {
210 	struct sock *sk;
211 
212 	sk = icmp_sk(net);
213 
214 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
215 		/* This can happen if the output path signals a
216 		 * dst_link_failure() for an outgoing ICMP packet.
217 		 */
218 		return NULL;
219 	}
220 	return sk;
221 }
222 
223 static inline void icmp_xmit_unlock(struct sock *sk)
224 {
225 	spin_unlock(&sk->sk_lock.slock);
226 }
227 
228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229 int sysctl_icmp_msgs_burst __read_mostly = 50;
230 
231 static struct {
232 	spinlock_t	lock;
233 	u32		credit;
234 	u32		stamp;
235 } icmp_global = {
236 	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
237 };
238 
239 /**
240  * icmp_global_allow - Are we allowed to send one more ICMP message ?
241  *
242  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
243  * Returns false if we reached the limit and can not send another packet.
244  * Note: called with BH disabled
245  */
246 bool icmp_global_allow(void)
247 {
248 	u32 credit, delta, incr = 0, now = (u32)jiffies;
249 	bool rc = false;
250 
251 	/* Check if token bucket is empty and cannot be refilled
252 	 * without taking the spinlock.
253 	 */
254 	if (!icmp_global.credit) {
255 		delta = min_t(u32, now - icmp_global.stamp, HZ);
256 		if (delta < HZ / 50)
257 			return false;
258 	}
259 
260 	spin_lock(&icmp_global.lock);
261 	delta = min_t(u32, now - icmp_global.stamp, HZ);
262 	if (delta >= HZ / 50) {
263 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
264 		if (incr)
265 			icmp_global.stamp = now;
266 	}
267 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
268 	if (credit) {
269 		credit--;
270 		rc = true;
271 	}
272 	icmp_global.credit = credit;
273 	spin_unlock(&icmp_global.lock);
274 	return rc;
275 }
276 EXPORT_SYMBOL(icmp_global_allow);
277 
278 static bool icmpv4_mask_allow(struct net *net, int type, int code)
279 {
280 	if (type > NR_ICMP_TYPES)
281 		return true;
282 
283 	/* Don't limit PMTU discovery. */
284 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 		return true;
286 
287 	/* Limit if icmp type is enabled in ratemask. */
288 	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
289 		return true;
290 
291 	return false;
292 }
293 
294 static bool icmpv4_global_allow(struct net *net, int type, int code)
295 {
296 	if (icmpv4_mask_allow(net, type, code))
297 		return true;
298 
299 	if (icmp_global_allow())
300 		return true;
301 
302 	return false;
303 }
304 
305 /*
306  *	Send an ICMP frame.
307  */
308 
309 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
310 			       struct flowi4 *fl4, int type, int code)
311 {
312 	struct dst_entry *dst = &rt->dst;
313 	struct inet_peer *peer;
314 	bool rc = true;
315 	int vif;
316 
317 	if (icmpv4_mask_allow(net, type, code))
318 		goto out;
319 
320 	/* No rate limit on loopback */
321 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
322 		goto out;
323 
324 	vif = l3mdev_master_ifindex(dst->dev);
325 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
326 	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
327 	if (peer)
328 		inet_putpeer(peer);
329 out:
330 	return rc;
331 }
332 
333 /*
334  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
335  */
336 void icmp_out_count(struct net *net, unsigned char type)
337 {
338 	ICMPMSGOUT_INC_STATS(net, type);
339 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
340 }
341 
342 /*
343  *	Checksum each fragment, and on the first include the headers and final
344  *	checksum.
345  */
346 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
347 			  struct sk_buff *skb)
348 {
349 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
350 	__wsum csum;
351 
352 	csum = skb_copy_and_csum_bits(icmp_param->skb,
353 				      icmp_param->offset + offset,
354 				      to, len, 0);
355 
356 	skb->csum = csum_block_add(skb->csum, csum, odd);
357 	if (icmp_pointers[icmp_param->data.icmph.type].error)
358 		nf_ct_attach(skb, icmp_param->skb);
359 	return 0;
360 }
361 
362 static void icmp_push_reply(struct icmp_bxm *icmp_param,
363 			    struct flowi4 *fl4,
364 			    struct ipcm_cookie *ipc, struct rtable **rt)
365 {
366 	struct sock *sk;
367 	struct sk_buff *skb;
368 
369 	sk = icmp_sk(dev_net((*rt)->dst.dev));
370 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
371 			   icmp_param->data_len+icmp_param->head_len,
372 			   icmp_param->head_len,
373 			   ipc, rt, MSG_DONTWAIT) < 0) {
374 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
375 		ip_flush_pending_frames(sk);
376 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
377 		struct icmphdr *icmph = icmp_hdr(skb);
378 		__wsum csum = 0;
379 		struct sk_buff *skb1;
380 
381 		skb_queue_walk(&sk->sk_write_queue, skb1) {
382 			csum = csum_add(csum, skb1->csum);
383 		}
384 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
385 						 (char *)icmph,
386 						 icmp_param->head_len, csum);
387 		icmph->checksum = csum_fold(csum);
388 		skb->ip_summed = CHECKSUM_NONE;
389 		ip_push_pending_frames(sk, fl4);
390 	}
391 }
392 
393 /*
394  *	Driving logic for building and sending ICMP messages.
395  */
396 
397 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
398 {
399 	struct ipcm_cookie ipc;
400 	struct rtable *rt = skb_rtable(skb);
401 	struct net *net = dev_net(rt->dst.dev);
402 	struct flowi4 fl4;
403 	struct sock *sk;
404 	struct inet_sock *inet;
405 	__be32 daddr, saddr;
406 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
407 	int type = icmp_param->data.icmph.type;
408 	int code = icmp_param->data.icmph.code;
409 
410 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
411 		return;
412 
413 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
414 	local_bh_disable();
415 
416 	/* global icmp_msgs_per_sec */
417 	if (!icmpv4_global_allow(net, type, code))
418 		goto out_bh_enable;
419 
420 	sk = icmp_xmit_lock(net);
421 	if (!sk)
422 		goto out_bh_enable;
423 	inet = inet_sk(sk);
424 
425 	icmp_param->data.icmph.checksum = 0;
426 
427 	ipcm_init(&ipc);
428 	inet->tos = ip_hdr(skb)->tos;
429 	sk->sk_mark = mark;
430 	daddr = ipc.addr = ip_hdr(skb)->saddr;
431 	saddr = fib_compute_spec_dst(skb);
432 
433 	if (icmp_param->replyopts.opt.opt.optlen) {
434 		ipc.opt = &icmp_param->replyopts.opt;
435 		if (ipc.opt->opt.srr)
436 			daddr = icmp_param->replyopts.opt.opt.faddr;
437 	}
438 	memset(&fl4, 0, sizeof(fl4));
439 	fl4.daddr = daddr;
440 	fl4.saddr = saddr;
441 	fl4.flowi4_mark = mark;
442 	fl4.flowi4_uid = sock_net_uid(net, NULL);
443 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
444 	fl4.flowi4_proto = IPPROTO_ICMP;
445 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
446 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
447 	rt = ip_route_output_key(net, &fl4);
448 	if (IS_ERR(rt))
449 		goto out_unlock;
450 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
451 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
452 	ip_rt_put(rt);
453 out_unlock:
454 	icmp_xmit_unlock(sk);
455 out_bh_enable:
456 	local_bh_enable();
457 }
458 
459 static struct rtable *icmp_route_lookup(struct net *net,
460 					struct flowi4 *fl4,
461 					struct sk_buff *skb_in,
462 					const struct iphdr *iph,
463 					__be32 saddr, u8 tos, u32 mark,
464 					int type, int code,
465 					struct icmp_bxm *param)
466 {
467 	struct rtable *rt, *rt2;
468 	struct flowi4 fl4_dec;
469 	int err;
470 
471 	memset(fl4, 0, sizeof(*fl4));
472 	fl4->daddr = (param->replyopts.opt.opt.srr ?
473 		      param->replyopts.opt.opt.faddr : iph->saddr);
474 	fl4->saddr = saddr;
475 	fl4->flowi4_mark = mark;
476 	fl4->flowi4_uid = sock_net_uid(net, NULL);
477 	fl4->flowi4_tos = RT_TOS(tos);
478 	fl4->flowi4_proto = IPPROTO_ICMP;
479 	fl4->fl4_icmp_type = type;
480 	fl4->fl4_icmp_code = code;
481 	fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
482 
483 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
484 	rt = ip_route_output_key_hash(net, fl4, skb_in);
485 	if (IS_ERR(rt))
486 		return rt;
487 
488 	/* No need to clone since we're just using its address. */
489 	rt2 = rt;
490 
491 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
492 					   flowi4_to_flowi(fl4), NULL, 0);
493 	if (!IS_ERR(rt)) {
494 		if (rt != rt2)
495 			return rt;
496 	} else if (PTR_ERR(rt) == -EPERM) {
497 		rt = NULL;
498 	} else
499 		return rt;
500 
501 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
502 	if (err)
503 		goto relookup_failed;
504 
505 	if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
506 				     fl4_dec.saddr) == RTN_LOCAL) {
507 		rt2 = __ip_route_output_key(net, &fl4_dec);
508 		if (IS_ERR(rt2))
509 			err = PTR_ERR(rt2);
510 	} else {
511 		struct flowi4 fl4_2 = {};
512 		unsigned long orefdst;
513 
514 		fl4_2.daddr = fl4_dec.saddr;
515 		rt2 = ip_route_output_key(net, &fl4_2);
516 		if (IS_ERR(rt2)) {
517 			err = PTR_ERR(rt2);
518 			goto relookup_failed;
519 		}
520 		/* Ugh! */
521 		orefdst = skb_in->_skb_refdst; /* save old refdst */
522 		skb_dst_set(skb_in, NULL);
523 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
524 				     RT_TOS(tos), rt2->dst.dev);
525 
526 		dst_release(&rt2->dst);
527 		rt2 = skb_rtable(skb_in);
528 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
529 	}
530 
531 	if (err)
532 		goto relookup_failed;
533 
534 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
535 					    flowi4_to_flowi(&fl4_dec), NULL,
536 					    XFRM_LOOKUP_ICMP);
537 	if (!IS_ERR(rt2)) {
538 		dst_release(&rt->dst);
539 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
540 		rt = rt2;
541 	} else if (PTR_ERR(rt2) == -EPERM) {
542 		if (rt)
543 			dst_release(&rt->dst);
544 		return rt2;
545 	} else {
546 		err = PTR_ERR(rt2);
547 		goto relookup_failed;
548 	}
549 	return rt;
550 
551 relookup_failed:
552 	if (rt)
553 		return rt;
554 	return ERR_PTR(err);
555 }
556 
557 /*
558  *	Send an ICMP message in response to a situation
559  *
560  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
561  *		  MAY send more (we do).
562  *			MUST NOT change this header information.
563  *			MUST NOT reply to a multicast/broadcast IP address.
564  *			MUST NOT reply to a multicast/broadcast MAC address.
565  *			MUST reply to only the first fragment.
566  */
567 
568 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
569 		 const struct ip_options *opt)
570 {
571 	struct iphdr *iph;
572 	int room;
573 	struct icmp_bxm icmp_param;
574 	struct rtable *rt = skb_rtable(skb_in);
575 	struct ipcm_cookie ipc;
576 	struct flowi4 fl4;
577 	__be32 saddr;
578 	u8  tos;
579 	u32 mark;
580 	struct net *net;
581 	struct sock *sk;
582 
583 	if (!rt)
584 		goto out;
585 	net = dev_net(rt->dst.dev);
586 
587 	/*
588 	 *	Find the original header. It is expected to be valid, of course.
589 	 *	Check this, icmp_send is called from the most obscure devices
590 	 *	sometimes.
591 	 */
592 	iph = ip_hdr(skb_in);
593 
594 	if ((u8 *)iph < skb_in->head ||
595 	    (skb_network_header(skb_in) + sizeof(*iph)) >
596 	    skb_tail_pointer(skb_in))
597 		goto out;
598 
599 	/*
600 	 *	No replies to physical multicast/broadcast
601 	 */
602 	if (skb_in->pkt_type != PACKET_HOST)
603 		goto out;
604 
605 	/*
606 	 *	Now check at the protocol level
607 	 */
608 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
609 		goto out;
610 
611 	/*
612 	 *	Only reply to fragment 0. We byte re-order the constant
613 	 *	mask for efficiency.
614 	 */
615 	if (iph->frag_off & htons(IP_OFFSET))
616 		goto out;
617 
618 	/*
619 	 *	If we send an ICMP error to an ICMP error a mess would result..
620 	 */
621 	if (icmp_pointers[type].error) {
622 		/*
623 		 *	We are an error, check if we are replying to an
624 		 *	ICMP error
625 		 */
626 		if (iph->protocol == IPPROTO_ICMP) {
627 			u8 _inner_type, *itp;
628 
629 			itp = skb_header_pointer(skb_in,
630 						 skb_network_header(skb_in) +
631 						 (iph->ihl << 2) +
632 						 offsetof(struct icmphdr,
633 							  type) -
634 						 skb_in->data,
635 						 sizeof(_inner_type),
636 						 &_inner_type);
637 			if (!itp)
638 				goto out;
639 
640 			/*
641 			 *	Assume any unknown ICMP type is an error. This
642 			 *	isn't specified by the RFC, but think about it..
643 			 */
644 			if (*itp > NR_ICMP_TYPES ||
645 			    icmp_pointers[*itp].error)
646 				goto out;
647 		}
648 	}
649 
650 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
651 	local_bh_disable();
652 
653 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
654 	 * incoming dev is loopback.  If outgoing dev change to not be
655 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
656 	 */
657 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
658 	      !icmpv4_global_allow(net, type, code))
659 		goto out_bh_enable;
660 
661 	sk = icmp_xmit_lock(net);
662 	if (!sk)
663 		goto out_bh_enable;
664 
665 	/*
666 	 *	Construct source address and options.
667 	 */
668 
669 	saddr = iph->daddr;
670 	if (!(rt->rt_flags & RTCF_LOCAL)) {
671 		struct net_device *dev = NULL;
672 
673 		rcu_read_lock();
674 		if (rt_is_input_route(rt) &&
675 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
676 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
677 
678 		if (dev)
679 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
680 		else
681 			saddr = 0;
682 		rcu_read_unlock();
683 	}
684 
685 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
686 					   IPTOS_PREC_INTERNETCONTROL) :
687 					  iph->tos;
688 	mark = IP4_REPLY_MARK(net, skb_in->mark);
689 
690 	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
691 		goto out_unlock;
692 
693 
694 	/*
695 	 *	Prepare data for ICMP header.
696 	 */
697 
698 	icmp_param.data.icmph.type	 = type;
699 	icmp_param.data.icmph.code	 = code;
700 	icmp_param.data.icmph.un.gateway = info;
701 	icmp_param.data.icmph.checksum	 = 0;
702 	icmp_param.skb	  = skb_in;
703 	icmp_param.offset = skb_network_offset(skb_in);
704 	inet_sk(sk)->tos = tos;
705 	sk->sk_mark = mark;
706 	ipcm_init(&ipc);
707 	ipc.addr = iph->saddr;
708 	ipc.opt = &icmp_param.replyopts.opt;
709 
710 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
711 			       type, code, &icmp_param);
712 	if (IS_ERR(rt))
713 		goto out_unlock;
714 
715 	/* peer icmp_ratelimit */
716 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
717 		goto ende;
718 
719 	/* RFC says return as much as we can without exceeding 576 bytes. */
720 
721 	room = dst_mtu(&rt->dst);
722 	if (room > 576)
723 		room = 576;
724 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
725 	room -= sizeof(struct icmphdr);
726 
727 	icmp_param.data_len = skb_in->len - icmp_param.offset;
728 	if (icmp_param.data_len > room)
729 		icmp_param.data_len = room;
730 	icmp_param.head_len = sizeof(struct icmphdr);
731 
732 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
733 ende:
734 	ip_rt_put(rt);
735 out_unlock:
736 	icmp_xmit_unlock(sk);
737 out_bh_enable:
738 	local_bh_enable();
739 out:;
740 }
741 EXPORT_SYMBOL(__icmp_send);
742 
743 
744 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
745 {
746 	const struct iphdr *iph = (const struct iphdr *) skb->data;
747 	const struct net_protocol *ipprot;
748 	int protocol = iph->protocol;
749 
750 	/* Checkin full IP header plus 8 bytes of protocol to
751 	 * avoid additional coding at protocol handlers.
752 	 */
753 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
754 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
755 		return;
756 	}
757 
758 	raw_icmp_error(skb, protocol, info);
759 
760 	ipprot = rcu_dereference(inet_protos[protocol]);
761 	if (ipprot && ipprot->err_handler)
762 		ipprot->err_handler(skb, info);
763 }
764 
765 static bool icmp_tag_validation(int proto)
766 {
767 	bool ok;
768 
769 	rcu_read_lock();
770 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
771 	rcu_read_unlock();
772 	return ok;
773 }
774 
775 /*
776  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
777  *	ICMP_PARAMETERPROB.
778  */
779 
780 static bool icmp_unreach(struct sk_buff *skb)
781 {
782 	const struct iphdr *iph;
783 	struct icmphdr *icmph;
784 	struct net *net;
785 	u32 info = 0;
786 
787 	net = dev_net(skb_dst(skb)->dev);
788 
789 	/*
790 	 *	Incomplete header ?
791 	 * 	Only checks for the IP header, there should be an
792 	 *	additional check for longer headers in upper levels.
793 	 */
794 
795 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
796 		goto out_err;
797 
798 	icmph = icmp_hdr(skb);
799 	iph   = (const struct iphdr *)skb->data;
800 
801 	if (iph->ihl < 5) /* Mangled header, drop. */
802 		goto out_err;
803 
804 	switch (icmph->type) {
805 	case ICMP_DEST_UNREACH:
806 		switch (icmph->code & 15) {
807 		case ICMP_NET_UNREACH:
808 		case ICMP_HOST_UNREACH:
809 		case ICMP_PROT_UNREACH:
810 		case ICMP_PORT_UNREACH:
811 			break;
812 		case ICMP_FRAG_NEEDED:
813 			/* for documentation of the ip_no_pmtu_disc
814 			 * values please see
815 			 * Documentation/networking/ip-sysctl.txt
816 			 */
817 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
818 			default:
819 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
820 						    &iph->daddr);
821 				break;
822 			case 2:
823 				goto out;
824 			case 3:
825 				if (!icmp_tag_validation(iph->protocol))
826 					goto out;
827 				/* fall through */
828 			case 0:
829 				info = ntohs(icmph->un.frag.mtu);
830 			}
831 			break;
832 		case ICMP_SR_FAILED:
833 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
834 					    &iph->daddr);
835 			break;
836 		default:
837 			break;
838 		}
839 		if (icmph->code > NR_ICMP_UNREACH)
840 			goto out;
841 		break;
842 	case ICMP_PARAMETERPROB:
843 		info = ntohl(icmph->un.gateway) >> 24;
844 		break;
845 	case ICMP_TIME_EXCEEDED:
846 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
847 		if (icmph->code == ICMP_EXC_FRAGTIME)
848 			goto out;
849 		break;
850 	}
851 
852 	/*
853 	 *	Throw it at our lower layers
854 	 *
855 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
856 	 *		  header.
857 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
858 	 *		  transport layer.
859 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
860 	 *		  transport layer.
861 	 */
862 
863 	/*
864 	 *	Check the other end isn't violating RFC 1122. Some routers send
865 	 *	bogus responses to broadcast frames. If you see this message
866 	 *	first check your netmask matches at both ends, if it does then
867 	 *	get the other vendor to fix their kit.
868 	 */
869 
870 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
871 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
872 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
873 				     &ip_hdr(skb)->saddr,
874 				     icmph->type, icmph->code,
875 				     &iph->daddr, skb->dev->name);
876 		goto out;
877 	}
878 
879 	icmp_socket_deliver(skb, info);
880 
881 out:
882 	return true;
883 out_err:
884 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
885 	return false;
886 }
887 
888 
889 /*
890  *	Handle ICMP_REDIRECT.
891  */
892 
893 static bool icmp_redirect(struct sk_buff *skb)
894 {
895 	if (skb->len < sizeof(struct iphdr)) {
896 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
897 		return false;
898 	}
899 
900 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
901 		/* there aught to be a stat */
902 		return false;
903 	}
904 
905 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
906 	return true;
907 }
908 
909 /*
910  *	Handle ICMP_ECHO ("ping") requests.
911  *
912  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
913  *		  requests.
914  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
915  *		  included in the reply.
916  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
917  *		  echo requests, MUST have default=NOT.
918  *	See also WRT handling of options once they are done and working.
919  */
920 
921 static bool icmp_echo(struct sk_buff *skb)
922 {
923 	struct net *net;
924 
925 	net = dev_net(skb_dst(skb)->dev);
926 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
927 		struct icmp_bxm icmp_param;
928 
929 		icmp_param.data.icmph	   = *icmp_hdr(skb);
930 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
931 		icmp_param.skb		   = skb;
932 		icmp_param.offset	   = 0;
933 		icmp_param.data_len	   = skb->len;
934 		icmp_param.head_len	   = sizeof(struct icmphdr);
935 		icmp_reply(&icmp_param, skb);
936 	}
937 	/* should there be an ICMP stat for ignored echos? */
938 	return true;
939 }
940 
941 /*
942  *	Handle ICMP Timestamp requests.
943  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
944  *		  SHOULD be in the kernel for minimum random latency.
945  *		  MUST be accurate to a few minutes.
946  *		  MUST be updated at least at 15Hz.
947  */
948 static bool icmp_timestamp(struct sk_buff *skb)
949 {
950 	struct icmp_bxm icmp_param;
951 	/*
952 	 *	Too short.
953 	 */
954 	if (skb->len < 4)
955 		goto out_err;
956 
957 	/*
958 	 *	Fill in the current time as ms since midnight UT:
959 	 */
960 	icmp_param.data.times[1] = inet_current_timestamp();
961 	icmp_param.data.times[2] = icmp_param.data.times[1];
962 
963 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
964 
965 	icmp_param.data.icmph	   = *icmp_hdr(skb);
966 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
967 	icmp_param.data.icmph.code = 0;
968 	icmp_param.skb		   = skb;
969 	icmp_param.offset	   = 0;
970 	icmp_param.data_len	   = 0;
971 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
972 	icmp_reply(&icmp_param, skb);
973 	return true;
974 
975 out_err:
976 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
977 	return false;
978 }
979 
980 static bool icmp_discard(struct sk_buff *skb)
981 {
982 	/* pretend it was a success */
983 	return true;
984 }
985 
986 /*
987  *	Deal with incoming ICMP packets.
988  */
989 int icmp_rcv(struct sk_buff *skb)
990 {
991 	struct icmphdr *icmph;
992 	struct rtable *rt = skb_rtable(skb);
993 	struct net *net = dev_net(rt->dst.dev);
994 	bool success;
995 
996 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
997 		struct sec_path *sp = skb_sec_path(skb);
998 		int nh;
999 
1000 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1001 				 XFRM_STATE_ICMP))
1002 			goto drop;
1003 
1004 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1005 			goto drop;
1006 
1007 		nh = skb_network_offset(skb);
1008 		skb_set_network_header(skb, sizeof(*icmph));
1009 
1010 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1011 			goto drop;
1012 
1013 		skb_set_network_header(skb, nh);
1014 	}
1015 
1016 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1017 
1018 	if (skb_checksum_simple_validate(skb))
1019 		goto csum_error;
1020 
1021 	if (!pskb_pull(skb, sizeof(*icmph)))
1022 		goto error;
1023 
1024 	icmph = icmp_hdr(skb);
1025 
1026 	ICMPMSGIN_INC_STATS(net, icmph->type);
1027 	/*
1028 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1029 	 *
1030 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1031 	 *		  discarded.
1032 	 */
1033 	if (icmph->type > NR_ICMP_TYPES)
1034 		goto error;
1035 
1036 
1037 	/*
1038 	 *	Parse the ICMP message
1039 	 */
1040 
1041 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1042 		/*
1043 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1044 		 *	  silently ignored (we let user decide with a sysctl).
1045 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1046 		 *	  discarded if to broadcast/multicast.
1047 		 */
1048 		if ((icmph->type == ICMP_ECHO ||
1049 		     icmph->type == ICMP_TIMESTAMP) &&
1050 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1051 			goto error;
1052 		}
1053 		if (icmph->type != ICMP_ECHO &&
1054 		    icmph->type != ICMP_TIMESTAMP &&
1055 		    icmph->type != ICMP_ADDRESS &&
1056 		    icmph->type != ICMP_ADDRESSREPLY) {
1057 			goto error;
1058 		}
1059 	}
1060 
1061 	success = icmp_pointers[icmph->type].handler(skb);
1062 
1063 	if (success)  {
1064 		consume_skb(skb);
1065 		return NET_RX_SUCCESS;
1066 	}
1067 
1068 drop:
1069 	kfree_skb(skb);
1070 	return NET_RX_DROP;
1071 csum_error:
1072 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1073 error:
1074 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1075 	goto drop;
1076 }
1077 
1078 int icmp_err(struct sk_buff *skb, u32 info)
1079 {
1080 	struct iphdr *iph = (struct iphdr *)skb->data;
1081 	int offset = iph->ihl<<2;
1082 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1083 	int type = icmp_hdr(skb)->type;
1084 	int code = icmp_hdr(skb)->code;
1085 	struct net *net = dev_net(skb->dev);
1086 
1087 	/*
1088 	 * Use ping_err to handle all icmp errors except those
1089 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1090 	 */
1091 	if (icmph->type != ICMP_ECHOREPLY) {
1092 		ping_err(skb, offset, info);
1093 		return 0;
1094 	}
1095 
1096 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1097 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1098 	else if (type == ICMP_REDIRECT)
1099 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1100 
1101 	return 0;
1102 }
1103 
1104 /*
1105  *	This table is the definition of how we handle ICMP.
1106  */
1107 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1108 	[ICMP_ECHOREPLY] = {
1109 		.handler = ping_rcv,
1110 	},
1111 	[1] = {
1112 		.handler = icmp_discard,
1113 		.error = 1,
1114 	},
1115 	[2] = {
1116 		.handler = icmp_discard,
1117 		.error = 1,
1118 	},
1119 	[ICMP_DEST_UNREACH] = {
1120 		.handler = icmp_unreach,
1121 		.error = 1,
1122 	},
1123 	[ICMP_SOURCE_QUENCH] = {
1124 		.handler = icmp_unreach,
1125 		.error = 1,
1126 	},
1127 	[ICMP_REDIRECT] = {
1128 		.handler = icmp_redirect,
1129 		.error = 1,
1130 	},
1131 	[6] = {
1132 		.handler = icmp_discard,
1133 		.error = 1,
1134 	},
1135 	[7] = {
1136 		.handler = icmp_discard,
1137 		.error = 1,
1138 	},
1139 	[ICMP_ECHO] = {
1140 		.handler = icmp_echo,
1141 	},
1142 	[9] = {
1143 		.handler = icmp_discard,
1144 		.error = 1,
1145 	},
1146 	[10] = {
1147 		.handler = icmp_discard,
1148 		.error = 1,
1149 	},
1150 	[ICMP_TIME_EXCEEDED] = {
1151 		.handler = icmp_unreach,
1152 		.error = 1,
1153 	},
1154 	[ICMP_PARAMETERPROB] = {
1155 		.handler = icmp_unreach,
1156 		.error = 1,
1157 	},
1158 	[ICMP_TIMESTAMP] = {
1159 		.handler = icmp_timestamp,
1160 	},
1161 	[ICMP_TIMESTAMPREPLY] = {
1162 		.handler = icmp_discard,
1163 	},
1164 	[ICMP_INFO_REQUEST] = {
1165 		.handler = icmp_discard,
1166 	},
1167 	[ICMP_INFO_REPLY] = {
1168 		.handler = icmp_discard,
1169 	},
1170 	[ICMP_ADDRESS] = {
1171 		.handler = icmp_discard,
1172 	},
1173 	[ICMP_ADDRESSREPLY] = {
1174 		.handler = icmp_discard,
1175 	},
1176 };
1177 
1178 static void __net_exit icmp_sk_exit(struct net *net)
1179 {
1180 	int i;
1181 
1182 	for_each_possible_cpu(i)
1183 		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1184 	free_percpu(net->ipv4.icmp_sk);
1185 	net->ipv4.icmp_sk = NULL;
1186 }
1187 
1188 static int __net_init icmp_sk_init(struct net *net)
1189 {
1190 	int i, err;
1191 
1192 	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1193 	if (!net->ipv4.icmp_sk)
1194 		return -ENOMEM;
1195 
1196 	for_each_possible_cpu(i) {
1197 		struct sock *sk;
1198 
1199 		err = inet_ctl_sock_create(&sk, PF_INET,
1200 					   SOCK_RAW, IPPROTO_ICMP, net);
1201 		if (err < 0)
1202 			goto fail;
1203 
1204 		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1205 
1206 		/* Enough space for 2 64K ICMP packets, including
1207 		 * sk_buff/skb_shared_info struct overhead.
1208 		 */
1209 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1210 
1211 		/*
1212 		 * Speedup sock_wfree()
1213 		 */
1214 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1215 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1216 	}
1217 
1218 	/* Control parameters for ECHO replies. */
1219 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1220 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1221 
1222 	/* Control parameter - ignore bogus broadcast responses? */
1223 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1224 
1225 	/*
1226 	 * 	Configurable global rate limit.
1227 	 *
1228 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1229 	 *	bucket ratemask defines which icmp types are ratelimited by
1230 	 *	setting	it's bit position.
1231 	 *
1232 	 *	default:
1233 	 *	dest unreachable (3), source quench (4),
1234 	 *	time exceeded (11), parameter problem (12)
1235 	 */
1236 
1237 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1238 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1239 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1240 
1241 	return 0;
1242 
1243 fail:
1244 	icmp_sk_exit(net);
1245 	return err;
1246 }
1247 
1248 static struct pernet_operations __net_initdata icmp_sk_ops = {
1249        .init = icmp_sk_init,
1250        .exit = icmp_sk_exit,
1251 };
1252 
1253 int __init icmp_init(void)
1254 {
1255 	return register_pernet_subsys(&icmp_sk_ops);
1256 }
1257