xref: /openbmc/linux/net/ipv4/icmp.c (revision 206204a1)
1 /*
2  *	NET3:	Implementation of the ICMP protocol layer.
3  *
4  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
5  *
6  *	This program is free software; you can redistribute it and/or
7  *	modify it under the terms of the GNU General Public License
8  *	as published by the Free Software Foundation; either version
9  *	2 of the License, or (at your option) any later version.
10  *
11  *	Some of the function names and the icmp unreach table for this
12  *	module were derived from [icmp.c 1.0.11 06/02/93] by
13  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14  *	Other than that this module is a complete rewrite.
15  *
16  *	Fixes:
17  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
18  *					with icmp type masking ability instead
19  *					of broken per type icmp timeouts.
20  *		Mike Shaver	:	RFC1122 checks.
21  *		Alan Cox	:	Multicast ping reply as self.
22  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
23  *					call.
24  *		Alan Cox	:	Added 216,128 byte paths to the MTU
25  *					code.
26  *		Martin Mares	:	RFC1812 checks.
27  *		Martin Mares	:	Can be configured to follow redirects
28  *					if acting as a router _without_ a
29  *					routing protocol (RFC 1812).
30  *		Martin Mares	:	Echo requests may be configured to
31  *					be ignored (RFC 1812).
32  *		Martin Mares	:	Limitation of ICMP error message
33  *					transmit rate (RFC 1812).
34  *		Martin Mares	:	TOS and Precedence set correctly
35  *					(RFC 1812).
36  *		Martin Mares	:	Now copying as much data from the
37  *					original packet as we can without
38  *					exceeding 576 bytes (RFC 1812).
39  *	Willy Konynenberg	:	Transparent proxying support.
40  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
41  *					path MTU bug.
42  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
43  *					valid (RFC 1812).
44  *		Andi Kleen	:	Check all packet lengths properly
45  *					and moved all kfree_skb() up to
46  *					icmp_rcv.
47  *		Andi Kleen	:	Move the rate limit bookkeeping
48  *					into the dest entry and use a token
49  *					bucket filter (thanks to ANK). Make
50  *					the rates sysctl configurable.
51  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
52  *					- IP option length was accounted wrongly
53  *					- ICMP header length was not accounted
54  *					  at all.
55  *              Tristan Greaves :       Added sysctl option to ignore bogus
56  *              			broadcast responses from broken routers.
57  *
58  * To Fix:
59  *
60  *	- Should use skb_pull() instead of all the manual checking.
61  *	  This would also greatly simply some upper layer error handlers. --AK
62  *
63  */
64 
65 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
66 
67 #include <linux/module.h>
68 #include <linux/types.h>
69 #include <linux/jiffies.h>
70 #include <linux/kernel.h>
71 #include <linux/fcntl.h>
72 #include <linux/socket.h>
73 #include <linux/in.h>
74 #include <linux/inet.h>
75 #include <linux/inetdevice.h>
76 #include <linux/netdevice.h>
77 #include <linux/string.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <linux/slab.h>
80 #include <net/snmp.h>
81 #include <net/ip.h>
82 #include <net/route.h>
83 #include <net/protocol.h>
84 #include <net/icmp.h>
85 #include <net/tcp.h>
86 #include <net/udp.h>
87 #include <net/raw.h>
88 #include <net/ping.h>
89 #include <linux/skbuff.h>
90 #include <net/sock.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/init.h>
94 #include <asm/uaccess.h>
95 #include <net/checksum.h>
96 #include <net/xfrm.h>
97 #include <net/inet_common.h>
98 #include <net/ip_fib.h>
99 
100 /*
101  *	Build xmit assembly blocks
102  */
103 
104 struct icmp_bxm {
105 	struct sk_buff *skb;
106 	int offset;
107 	int data_len;
108 
109 	struct {
110 		struct icmphdr icmph;
111 		__be32	       times[3];
112 	} data;
113 	int head_len;
114 	struct ip_options_data replyopts;
115 };
116 
117 /* An array of errno for error messages from dest unreach. */
118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
119 
120 const struct icmp_err icmp_err_convert[] = {
121 	{
122 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
123 		.fatal = 0,
124 	},
125 	{
126 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
127 		.fatal = 0,
128 	},
129 	{
130 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
131 		.fatal = 1,
132 	},
133 	{
134 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
135 		.fatal = 1,
136 	},
137 	{
138 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
139 		.fatal = 0,
140 	},
141 	{
142 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
143 		.fatal = 0,
144 	},
145 	{
146 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
147 		.fatal = 1,
148 	},
149 	{
150 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
163 		.fatal = 1,
164 	},
165 	{
166 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
167 		.fatal = 0,
168 	},
169 	{
170 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
171 		.fatal = 0,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
175 		.fatal = 1,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
179 		.fatal = 1,
180 	},
181 	{
182 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
183 		.fatal = 1,
184 	},
185 };
186 EXPORT_SYMBOL(icmp_err_convert);
187 
188 /*
189  *	ICMP control array. This specifies what to do with each ICMP.
190  */
191 
192 struct icmp_control {
193 	void (*handler)(struct sk_buff *skb);
194 	short   error;		/* This ICMP is classed as an error message */
195 };
196 
197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
198 
199 /*
200  *	The ICMP socket(s). This is the most convenient way to flow control
201  *	our ICMP output as well as maintain a clean interface throughout
202  *	all layers. All Socketless IP sends will soon be gone.
203  *
204  *	On SMP we have one ICMP socket per-cpu.
205  */
206 static struct sock *icmp_sk(struct net *net)
207 {
208 	return net->ipv4.icmp_sk[smp_processor_id()];
209 }
210 
211 static inline struct sock *icmp_xmit_lock(struct net *net)
212 {
213 	struct sock *sk;
214 
215 	local_bh_disable();
216 
217 	sk = icmp_sk(net);
218 
219 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
220 		/* This can happen if the output path signals a
221 		 * dst_link_failure() for an outgoing ICMP packet.
222 		 */
223 		local_bh_enable();
224 		return NULL;
225 	}
226 	return sk;
227 }
228 
229 static inline void icmp_xmit_unlock(struct sock *sk)
230 {
231 	spin_unlock_bh(&sk->sk_lock.slock);
232 }
233 
234 /*
235  *	Send an ICMP frame.
236  */
237 
238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 				      struct flowi4 *fl4, int type, int code)
240 {
241 	struct dst_entry *dst = &rt->dst;
242 	bool rc = true;
243 
244 	if (type > NR_ICMP_TYPES)
245 		goto out;
246 
247 	/* Don't limit PMTU discovery. */
248 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
249 		goto out;
250 
251 	/* No rate limit on loopback */
252 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
253 		goto out;
254 
255 	/* Limit if icmp type is enabled in ratemask. */
256 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 		struct inet_peer *peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1);
258 		rc = inet_peer_xrlim_allow(peer,
259 					   net->ipv4.sysctl_icmp_ratelimit);
260 		if (peer)
261 			inet_putpeer(peer);
262 	}
263 out:
264 	return rc;
265 }
266 
267 /*
268  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
269  */
270 void icmp_out_count(struct net *net, unsigned char type)
271 {
272 	ICMPMSGOUT_INC_STATS(net, type);
273 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
274 }
275 
276 /*
277  *	Checksum each fragment, and on the first include the headers and final
278  *	checksum.
279  */
280 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
281 			  struct sk_buff *skb)
282 {
283 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
284 	__wsum csum;
285 
286 	csum = skb_copy_and_csum_bits(icmp_param->skb,
287 				      icmp_param->offset + offset,
288 				      to, len, 0);
289 
290 	skb->csum = csum_block_add(skb->csum, csum, odd);
291 	if (icmp_pointers[icmp_param->data.icmph.type].error)
292 		nf_ct_attach(skb, icmp_param->skb);
293 	return 0;
294 }
295 
296 static void icmp_push_reply(struct icmp_bxm *icmp_param,
297 			    struct flowi4 *fl4,
298 			    struct ipcm_cookie *ipc, struct rtable **rt)
299 {
300 	struct sock *sk;
301 	struct sk_buff *skb;
302 
303 	sk = icmp_sk(dev_net((*rt)->dst.dev));
304 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
305 			   icmp_param->data_len+icmp_param->head_len,
306 			   icmp_param->head_len,
307 			   ipc, rt, MSG_DONTWAIT) < 0) {
308 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
309 		ip_flush_pending_frames(sk);
310 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
311 		struct icmphdr *icmph = icmp_hdr(skb);
312 		__wsum csum = 0;
313 		struct sk_buff *skb1;
314 
315 		skb_queue_walk(&sk->sk_write_queue, skb1) {
316 			csum = csum_add(csum, skb1->csum);
317 		}
318 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
319 						 (char *)icmph,
320 						 icmp_param->head_len, csum);
321 		icmph->checksum = csum_fold(csum);
322 		skb->ip_summed = CHECKSUM_NONE;
323 		ip_push_pending_frames(sk, fl4);
324 	}
325 }
326 
327 /*
328  *	Driving logic for building and sending ICMP messages.
329  */
330 
331 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
332 {
333 	struct ipcm_cookie ipc;
334 	struct rtable *rt = skb_rtable(skb);
335 	struct net *net = dev_net(rt->dst.dev);
336 	struct flowi4 fl4;
337 	struct sock *sk;
338 	struct inet_sock *inet;
339 	__be32 daddr, saddr;
340 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
341 
342 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
343 		return;
344 
345 	sk = icmp_xmit_lock(net);
346 	if (sk == NULL)
347 		return;
348 	inet = inet_sk(sk);
349 
350 	icmp_param->data.icmph.checksum = 0;
351 
352 	inet->tos = ip_hdr(skb)->tos;
353 	sk->sk_mark = mark;
354 	daddr = ipc.addr = ip_hdr(skb)->saddr;
355 	saddr = fib_compute_spec_dst(skb);
356 	ipc.opt = NULL;
357 	ipc.tx_flags = 0;
358 	ipc.ttl = 0;
359 	ipc.tos = -1;
360 
361 	if (icmp_param->replyopts.opt.opt.optlen) {
362 		ipc.opt = &icmp_param->replyopts.opt;
363 		if (ipc.opt->opt.srr)
364 			daddr = icmp_param->replyopts.opt.opt.faddr;
365 	}
366 	memset(&fl4, 0, sizeof(fl4));
367 	fl4.daddr = daddr;
368 	fl4.saddr = saddr;
369 	fl4.flowi4_mark = mark;
370 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
371 	fl4.flowi4_proto = IPPROTO_ICMP;
372 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
373 	rt = ip_route_output_key(net, &fl4);
374 	if (IS_ERR(rt))
375 		goto out_unlock;
376 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
377 			       icmp_param->data.icmph.code))
378 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
379 	ip_rt_put(rt);
380 out_unlock:
381 	icmp_xmit_unlock(sk);
382 }
383 
384 static struct rtable *icmp_route_lookup(struct net *net,
385 					struct flowi4 *fl4,
386 					struct sk_buff *skb_in,
387 					const struct iphdr *iph,
388 					__be32 saddr, u8 tos, u32 mark,
389 					int type, int code,
390 					struct icmp_bxm *param)
391 {
392 	struct rtable *rt, *rt2;
393 	struct flowi4 fl4_dec;
394 	int err;
395 
396 	memset(fl4, 0, sizeof(*fl4));
397 	fl4->daddr = (param->replyopts.opt.opt.srr ?
398 		      param->replyopts.opt.opt.faddr : iph->saddr);
399 	fl4->saddr = saddr;
400 	fl4->flowi4_mark = mark;
401 	fl4->flowi4_tos = RT_TOS(tos);
402 	fl4->flowi4_proto = IPPROTO_ICMP;
403 	fl4->fl4_icmp_type = type;
404 	fl4->fl4_icmp_code = code;
405 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
406 	rt = __ip_route_output_key(net, fl4);
407 	if (IS_ERR(rt))
408 		return rt;
409 
410 	/* No need to clone since we're just using its address. */
411 	rt2 = rt;
412 
413 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
414 					   flowi4_to_flowi(fl4), NULL, 0);
415 	if (!IS_ERR(rt)) {
416 		if (rt != rt2)
417 			return rt;
418 	} else if (PTR_ERR(rt) == -EPERM) {
419 		rt = NULL;
420 	} else
421 		return rt;
422 
423 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
424 	if (err)
425 		goto relookup_failed;
426 
427 	if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
428 		rt2 = __ip_route_output_key(net, &fl4_dec);
429 		if (IS_ERR(rt2))
430 			err = PTR_ERR(rt2);
431 	} else {
432 		struct flowi4 fl4_2 = {};
433 		unsigned long orefdst;
434 
435 		fl4_2.daddr = fl4_dec.saddr;
436 		rt2 = ip_route_output_key(net, &fl4_2);
437 		if (IS_ERR(rt2)) {
438 			err = PTR_ERR(rt2);
439 			goto relookup_failed;
440 		}
441 		/* Ugh! */
442 		orefdst = skb_in->_skb_refdst; /* save old refdst */
443 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
444 				     RT_TOS(tos), rt2->dst.dev);
445 
446 		dst_release(&rt2->dst);
447 		rt2 = skb_rtable(skb_in);
448 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
449 	}
450 
451 	if (err)
452 		goto relookup_failed;
453 
454 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
455 					    flowi4_to_flowi(&fl4_dec), NULL,
456 					    XFRM_LOOKUP_ICMP);
457 	if (!IS_ERR(rt2)) {
458 		dst_release(&rt->dst);
459 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
460 		rt = rt2;
461 	} else if (PTR_ERR(rt2) == -EPERM) {
462 		if (rt)
463 			dst_release(&rt->dst);
464 		return rt2;
465 	} else {
466 		err = PTR_ERR(rt2);
467 		goto relookup_failed;
468 	}
469 	return rt;
470 
471 relookup_failed:
472 	if (rt)
473 		return rt;
474 	return ERR_PTR(err);
475 }
476 
477 /*
478  *	Send an ICMP message in response to a situation
479  *
480  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
481  *		  MAY send more (we do).
482  *			MUST NOT change this header information.
483  *			MUST NOT reply to a multicast/broadcast IP address.
484  *			MUST NOT reply to a multicast/broadcast MAC address.
485  *			MUST reply to only the first fragment.
486  */
487 
488 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
489 {
490 	struct iphdr *iph;
491 	int room;
492 	struct icmp_bxm *icmp_param;
493 	struct rtable *rt = skb_rtable(skb_in);
494 	struct ipcm_cookie ipc;
495 	struct flowi4 fl4;
496 	__be32 saddr;
497 	u8  tos;
498 	u32 mark;
499 	struct net *net;
500 	struct sock *sk;
501 
502 	if (!rt)
503 		goto out;
504 	net = dev_net(rt->dst.dev);
505 
506 	/*
507 	 *	Find the original header. It is expected to be valid, of course.
508 	 *	Check this, icmp_send is called from the most obscure devices
509 	 *	sometimes.
510 	 */
511 	iph = ip_hdr(skb_in);
512 
513 	if ((u8 *)iph < skb_in->head ||
514 	    (skb_network_header(skb_in) + sizeof(*iph)) >
515 	    skb_tail_pointer(skb_in))
516 		goto out;
517 
518 	/*
519 	 *	No replies to physical multicast/broadcast
520 	 */
521 	if (skb_in->pkt_type != PACKET_HOST)
522 		goto out;
523 
524 	/*
525 	 *	Now check at the protocol level
526 	 */
527 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
528 		goto out;
529 
530 	/*
531 	 *	Only reply to fragment 0. We byte re-order the constant
532 	 *	mask for efficiency.
533 	 */
534 	if (iph->frag_off & htons(IP_OFFSET))
535 		goto out;
536 
537 	/*
538 	 *	If we send an ICMP error to an ICMP error a mess would result..
539 	 */
540 	if (icmp_pointers[type].error) {
541 		/*
542 		 *	We are an error, check if we are replying to an
543 		 *	ICMP error
544 		 */
545 		if (iph->protocol == IPPROTO_ICMP) {
546 			u8 _inner_type, *itp;
547 
548 			itp = skb_header_pointer(skb_in,
549 						 skb_network_header(skb_in) +
550 						 (iph->ihl << 2) +
551 						 offsetof(struct icmphdr,
552 							  type) -
553 						 skb_in->data,
554 						 sizeof(_inner_type),
555 						 &_inner_type);
556 			if (itp == NULL)
557 				goto out;
558 
559 			/*
560 			 *	Assume any unknown ICMP type is an error. This
561 			 *	isn't specified by the RFC, but think about it..
562 			 */
563 			if (*itp > NR_ICMP_TYPES ||
564 			    icmp_pointers[*itp].error)
565 				goto out;
566 		}
567 	}
568 
569 	icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
570 	if (!icmp_param)
571 		return;
572 
573 	sk = icmp_xmit_lock(net);
574 	if (sk == NULL)
575 		goto out_free;
576 
577 	/*
578 	 *	Construct source address and options.
579 	 */
580 
581 	saddr = iph->daddr;
582 	if (!(rt->rt_flags & RTCF_LOCAL)) {
583 		struct net_device *dev = NULL;
584 
585 		rcu_read_lock();
586 		if (rt_is_input_route(rt) &&
587 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
588 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
589 
590 		if (dev)
591 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
592 		else
593 			saddr = 0;
594 		rcu_read_unlock();
595 	}
596 
597 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
598 					   IPTOS_PREC_INTERNETCONTROL) :
599 					  iph->tos;
600 	mark = IP4_REPLY_MARK(net, skb_in->mark);
601 
602 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
603 		goto out_unlock;
604 
605 
606 	/*
607 	 *	Prepare data for ICMP header.
608 	 */
609 
610 	icmp_param->data.icmph.type	 = type;
611 	icmp_param->data.icmph.code	 = code;
612 	icmp_param->data.icmph.un.gateway = info;
613 	icmp_param->data.icmph.checksum	 = 0;
614 	icmp_param->skb	  = skb_in;
615 	icmp_param->offset = skb_network_offset(skb_in);
616 	inet_sk(sk)->tos = tos;
617 	sk->sk_mark = mark;
618 	ipc.addr = iph->saddr;
619 	ipc.opt = &icmp_param->replyopts.opt;
620 	ipc.tx_flags = 0;
621 	ipc.ttl = 0;
622 	ipc.tos = -1;
623 
624 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
625 			       type, code, icmp_param);
626 	if (IS_ERR(rt))
627 		goto out_unlock;
628 
629 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
630 		goto ende;
631 
632 	/* RFC says return as much as we can without exceeding 576 bytes. */
633 
634 	room = dst_mtu(&rt->dst);
635 	if (room > 576)
636 		room = 576;
637 	room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
638 	room -= sizeof(struct icmphdr);
639 
640 	icmp_param->data_len = skb_in->len - icmp_param->offset;
641 	if (icmp_param->data_len > room)
642 		icmp_param->data_len = room;
643 	icmp_param->head_len = sizeof(struct icmphdr);
644 
645 	icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
646 ende:
647 	ip_rt_put(rt);
648 out_unlock:
649 	icmp_xmit_unlock(sk);
650 out_free:
651 	kfree(icmp_param);
652 out:;
653 }
654 EXPORT_SYMBOL(icmp_send);
655 
656 
657 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
658 {
659 	const struct iphdr *iph = (const struct iphdr *) skb->data;
660 	const struct net_protocol *ipprot;
661 	int protocol = iph->protocol;
662 
663 	/* Checkin full IP header plus 8 bytes of protocol to
664 	 * avoid additional coding at protocol handlers.
665 	 */
666 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
667 		return;
668 
669 	raw_icmp_error(skb, protocol, info);
670 
671 	rcu_read_lock();
672 	ipprot = rcu_dereference(inet_protos[protocol]);
673 	if (ipprot && ipprot->err_handler)
674 		ipprot->err_handler(skb, info);
675 	rcu_read_unlock();
676 }
677 
678 static bool icmp_tag_validation(int proto)
679 {
680 	bool ok;
681 
682 	rcu_read_lock();
683 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
684 	rcu_read_unlock();
685 	return ok;
686 }
687 
688 /*
689  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
690  *	ICMP_PARAMETERPROB.
691  */
692 
693 static void icmp_unreach(struct sk_buff *skb)
694 {
695 	const struct iphdr *iph;
696 	struct icmphdr *icmph;
697 	struct net *net;
698 	u32 info = 0;
699 
700 	net = dev_net(skb_dst(skb)->dev);
701 
702 	/*
703 	 *	Incomplete header ?
704 	 * 	Only checks for the IP header, there should be an
705 	 *	additional check for longer headers in upper levels.
706 	 */
707 
708 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
709 		goto out_err;
710 
711 	icmph = icmp_hdr(skb);
712 	iph   = (const struct iphdr *)skb->data;
713 
714 	if (iph->ihl < 5) /* Mangled header, drop. */
715 		goto out_err;
716 
717 	if (icmph->type == ICMP_DEST_UNREACH) {
718 		switch (icmph->code & 15) {
719 		case ICMP_NET_UNREACH:
720 		case ICMP_HOST_UNREACH:
721 		case ICMP_PROT_UNREACH:
722 		case ICMP_PORT_UNREACH:
723 			break;
724 		case ICMP_FRAG_NEEDED:
725 			/* for documentation of the ip_no_pmtu_disc
726 			 * values please see
727 			 * Documentation/networking/ip-sysctl.txt
728 			 */
729 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
730 			default:
731 				LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
732 					       &iph->daddr);
733 				break;
734 			case 2:
735 				goto out;
736 			case 3:
737 				if (!icmp_tag_validation(iph->protocol))
738 					goto out;
739 				/* fall through */
740 			case 0:
741 				info = ntohs(icmph->un.frag.mtu);
742 				if (!info)
743 					goto out;
744 			}
745 			break;
746 		case ICMP_SR_FAILED:
747 			LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
748 				       &iph->daddr);
749 			break;
750 		default:
751 			break;
752 		}
753 		if (icmph->code > NR_ICMP_UNREACH)
754 			goto out;
755 	} else if (icmph->type == ICMP_PARAMETERPROB)
756 		info = ntohl(icmph->un.gateway) >> 24;
757 
758 	/*
759 	 *	Throw it at our lower layers
760 	 *
761 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
762 	 *		  header.
763 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
764 	 *		  transport layer.
765 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
766 	 *		  transport layer.
767 	 */
768 
769 	/*
770 	 *	Check the other end isn't violating RFC 1122. Some routers send
771 	 *	bogus responses to broadcast frames. If you see this message
772 	 *	first check your netmask matches at both ends, if it does then
773 	 *	get the other vendor to fix their kit.
774 	 */
775 
776 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
777 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
778 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
779 				     &ip_hdr(skb)->saddr,
780 				     icmph->type, icmph->code,
781 				     &iph->daddr, skb->dev->name);
782 		goto out;
783 	}
784 
785 	icmp_socket_deliver(skb, info);
786 
787 out:
788 	return;
789 out_err:
790 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
791 	goto out;
792 }
793 
794 
795 /*
796  *	Handle ICMP_REDIRECT.
797  */
798 
799 static void icmp_redirect(struct sk_buff *skb)
800 {
801 	if (skb->len < sizeof(struct iphdr)) {
802 		ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
803 		return;
804 	}
805 
806 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
807 		return;
808 
809 	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
810 }
811 
812 /*
813  *	Handle ICMP_ECHO ("ping") requests.
814  *
815  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
816  *		  requests.
817  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
818  *		  included in the reply.
819  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
820  *		  echo requests, MUST have default=NOT.
821  *	See also WRT handling of options once they are done and working.
822  */
823 
824 static void icmp_echo(struct sk_buff *skb)
825 {
826 	struct net *net;
827 
828 	net = dev_net(skb_dst(skb)->dev);
829 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
830 		struct icmp_bxm icmp_param;
831 
832 		icmp_param.data.icmph	   = *icmp_hdr(skb);
833 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
834 		icmp_param.skb		   = skb;
835 		icmp_param.offset	   = 0;
836 		icmp_param.data_len	   = skb->len;
837 		icmp_param.head_len	   = sizeof(struct icmphdr);
838 		icmp_reply(&icmp_param, skb);
839 	}
840 }
841 
842 /*
843  *	Handle ICMP Timestamp requests.
844  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
845  *		  SHOULD be in the kernel for minimum random latency.
846  *		  MUST be accurate to a few minutes.
847  *		  MUST be updated at least at 15Hz.
848  */
849 static void icmp_timestamp(struct sk_buff *skb)
850 {
851 	struct timespec tv;
852 	struct icmp_bxm icmp_param;
853 	/*
854 	 *	Too short.
855 	 */
856 	if (skb->len < 4)
857 		goto out_err;
858 
859 	/*
860 	 *	Fill in the current time as ms since midnight UT:
861 	 */
862 	getnstimeofday(&tv);
863 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
864 					 tv.tv_nsec / NSEC_PER_MSEC);
865 	icmp_param.data.times[2] = icmp_param.data.times[1];
866 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
867 		BUG();
868 	icmp_param.data.icmph	   = *icmp_hdr(skb);
869 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
870 	icmp_param.data.icmph.code = 0;
871 	icmp_param.skb		   = skb;
872 	icmp_param.offset	   = 0;
873 	icmp_param.data_len	   = 0;
874 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
875 	icmp_reply(&icmp_param, skb);
876 out:
877 	return;
878 out_err:
879 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
880 	goto out;
881 }
882 
883 static void icmp_discard(struct sk_buff *skb)
884 {
885 }
886 
887 /*
888  *	Deal with incoming ICMP packets.
889  */
890 int icmp_rcv(struct sk_buff *skb)
891 {
892 	struct icmphdr *icmph;
893 	struct rtable *rt = skb_rtable(skb);
894 	struct net *net = dev_net(rt->dst.dev);
895 
896 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
897 		struct sec_path *sp = skb_sec_path(skb);
898 		int nh;
899 
900 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
901 				 XFRM_STATE_ICMP))
902 			goto drop;
903 
904 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
905 			goto drop;
906 
907 		nh = skb_network_offset(skb);
908 		skb_set_network_header(skb, sizeof(*icmph));
909 
910 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
911 			goto drop;
912 
913 		skb_set_network_header(skb, nh);
914 	}
915 
916 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
917 
918 	if (skb_checksum_simple_validate(skb))
919 		goto csum_error;
920 
921 	if (!pskb_pull(skb, sizeof(*icmph)))
922 		goto error;
923 
924 	icmph = icmp_hdr(skb);
925 
926 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
927 	/*
928 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
929 	 *
930 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
931 	 *		  discarded.
932 	 */
933 	if (icmph->type > NR_ICMP_TYPES)
934 		goto error;
935 
936 
937 	/*
938 	 *	Parse the ICMP message
939 	 */
940 
941 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
942 		/*
943 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
944 		 *	  silently ignored (we let user decide with a sysctl).
945 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
946 		 *	  discarded if to broadcast/multicast.
947 		 */
948 		if ((icmph->type == ICMP_ECHO ||
949 		     icmph->type == ICMP_TIMESTAMP) &&
950 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
951 			goto error;
952 		}
953 		if (icmph->type != ICMP_ECHO &&
954 		    icmph->type != ICMP_TIMESTAMP &&
955 		    icmph->type != ICMP_ADDRESS &&
956 		    icmph->type != ICMP_ADDRESSREPLY) {
957 			goto error;
958 		}
959 	}
960 
961 	icmp_pointers[icmph->type].handler(skb);
962 
963 drop:
964 	kfree_skb(skb);
965 	return 0;
966 csum_error:
967 	ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
968 error:
969 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
970 	goto drop;
971 }
972 
973 void icmp_err(struct sk_buff *skb, u32 info)
974 {
975 	struct iphdr *iph = (struct iphdr *)skb->data;
976 	int offset = iph->ihl<<2;
977 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
978 	int type = icmp_hdr(skb)->type;
979 	int code = icmp_hdr(skb)->code;
980 	struct net *net = dev_net(skb->dev);
981 
982 	/*
983 	 * Use ping_err to handle all icmp errors except those
984 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
985 	 */
986 	if (icmph->type != ICMP_ECHOREPLY) {
987 		ping_err(skb, offset, info);
988 		return;
989 	}
990 
991 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
992 		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
993 	else if (type == ICMP_REDIRECT)
994 		ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
995 }
996 
997 /*
998  *	This table is the definition of how we handle ICMP.
999  */
1000 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1001 	[ICMP_ECHOREPLY] = {
1002 		.handler = ping_rcv,
1003 	},
1004 	[1] = {
1005 		.handler = icmp_discard,
1006 		.error = 1,
1007 	},
1008 	[2] = {
1009 		.handler = icmp_discard,
1010 		.error = 1,
1011 	},
1012 	[ICMP_DEST_UNREACH] = {
1013 		.handler = icmp_unreach,
1014 		.error = 1,
1015 	},
1016 	[ICMP_SOURCE_QUENCH] = {
1017 		.handler = icmp_unreach,
1018 		.error = 1,
1019 	},
1020 	[ICMP_REDIRECT] = {
1021 		.handler = icmp_redirect,
1022 		.error = 1,
1023 	},
1024 	[6] = {
1025 		.handler = icmp_discard,
1026 		.error = 1,
1027 	},
1028 	[7] = {
1029 		.handler = icmp_discard,
1030 		.error = 1,
1031 	},
1032 	[ICMP_ECHO] = {
1033 		.handler = icmp_echo,
1034 	},
1035 	[9] = {
1036 		.handler = icmp_discard,
1037 		.error = 1,
1038 	},
1039 	[10] = {
1040 		.handler = icmp_discard,
1041 		.error = 1,
1042 	},
1043 	[ICMP_TIME_EXCEEDED] = {
1044 		.handler = icmp_unreach,
1045 		.error = 1,
1046 	},
1047 	[ICMP_PARAMETERPROB] = {
1048 		.handler = icmp_unreach,
1049 		.error = 1,
1050 	},
1051 	[ICMP_TIMESTAMP] = {
1052 		.handler = icmp_timestamp,
1053 	},
1054 	[ICMP_TIMESTAMPREPLY] = {
1055 		.handler = icmp_discard,
1056 	},
1057 	[ICMP_INFO_REQUEST] = {
1058 		.handler = icmp_discard,
1059 	},
1060 	[ICMP_INFO_REPLY] = {
1061 		.handler = icmp_discard,
1062 	},
1063 	[ICMP_ADDRESS] = {
1064 		.handler = icmp_discard,
1065 	},
1066 	[ICMP_ADDRESSREPLY] = {
1067 		.handler = icmp_discard,
1068 	},
1069 };
1070 
1071 static void __net_exit icmp_sk_exit(struct net *net)
1072 {
1073 	int i;
1074 
1075 	for_each_possible_cpu(i)
1076 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1077 	kfree(net->ipv4.icmp_sk);
1078 	net->ipv4.icmp_sk = NULL;
1079 }
1080 
1081 static int __net_init icmp_sk_init(struct net *net)
1082 {
1083 	int i, err;
1084 
1085 	net->ipv4.icmp_sk =
1086 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1087 	if (net->ipv4.icmp_sk == NULL)
1088 		return -ENOMEM;
1089 
1090 	for_each_possible_cpu(i) {
1091 		struct sock *sk;
1092 
1093 		err = inet_ctl_sock_create(&sk, PF_INET,
1094 					   SOCK_RAW, IPPROTO_ICMP, net);
1095 		if (err < 0)
1096 			goto fail;
1097 
1098 		net->ipv4.icmp_sk[i] = sk;
1099 
1100 		/* Enough space for 2 64K ICMP packets, including
1101 		 * sk_buff/skb_shared_info struct overhead.
1102 		 */
1103 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1104 
1105 		/*
1106 		 * Speedup sock_wfree()
1107 		 */
1108 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1109 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1110 	}
1111 
1112 	/* Control parameters for ECHO replies. */
1113 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1114 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1115 
1116 	/* Control parameter - ignore bogus broadcast responses? */
1117 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1118 
1119 	/*
1120 	 * 	Configurable global rate limit.
1121 	 *
1122 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1123 	 *	bucket ratemask defines which icmp types are ratelimited by
1124 	 *	setting	it's bit position.
1125 	 *
1126 	 *	default:
1127 	 *	dest unreachable (3), source quench (4),
1128 	 *	time exceeded (11), parameter problem (12)
1129 	 */
1130 
1131 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1132 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1133 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1134 
1135 	return 0;
1136 
1137 fail:
1138 	for_each_possible_cpu(i)
1139 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1140 	kfree(net->ipv4.icmp_sk);
1141 	return err;
1142 }
1143 
1144 static struct pernet_operations __net_initdata icmp_sk_ops = {
1145        .init = icmp_sk_init,
1146        .exit = icmp_sk_exit,
1147 };
1148 
1149 int __init icmp_init(void)
1150 {
1151 	return register_pernet_subsys(&icmp_sk_ops);
1152 }
1153