xref: /netbsd-src/sys/netinet/ip_icmp.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: ip_icmp.c,v 1.172 2018/06/21 10:37:50 knakahara Exp $	*/
2 
3 /*
4  * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Public Access Networks Corporation ("Panix").  It was developed under
9  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
10  *
11  * This code is derived from software contributed to The NetBSD Foundation
12  * by Jason R. Thorpe of Zembu Labs, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
25  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
26  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
27  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /*
37  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
38  * All rights reserved.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the project nor the names of its contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  */
64 
65 /*
66  * Copyright (c) 1982, 1986, 1988, 1993
67  *	The Regents of the University of California.  All rights reserved.
68  *
69  * Redistribution and use in source and binary forms, with or without
70  * modification, are permitted provided that the following conditions
71  * are met:
72  * 1. Redistributions of source code must retain the above copyright
73  *    notice, this list of conditions and the following disclaimer.
74  * 2. Redistributions in binary form must reproduce the above copyright
75  *    notice, this list of conditions and the following disclaimer in the
76  *    documentation and/or other materials provided with the distribution.
77  * 3. Neither the name of the University nor the names of its contributors
78  *    may be used to endorse or promote products derived from this software
79  *    without specific prior written permission.
80  *
81  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
82  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
83  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
84  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
85  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
86  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
87  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
88  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
89  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
90  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
91  * SUCH DAMAGE.
92  *
93  *	@(#)ip_icmp.c	8.2 (Berkeley) 1/4/94
94  */
95 
96 #include <sys/cdefs.h>
97 __KERNEL_RCSID(0, "$NetBSD: ip_icmp.c,v 1.172 2018/06/21 10:37:50 knakahara Exp $");
98 
99 #ifdef _KERNEL_OPT
100 #include "opt_ipsec.h"
101 #endif
102 
103 #include <sys/param.h>
104 #include <sys/systm.h>
105 #include <sys/mbuf.h>
106 #include <sys/protosw.h>
107 #include <sys/socket.h>
108 #include <sys/socketvar.h> /* For softnet_lock */
109 #include <sys/kmem.h>
110 #include <sys/time.h>
111 #include <sys/kernel.h>
112 #include <sys/syslog.h>
113 #include <sys/sysctl.h>
114 
115 #include <net/if.h>
116 #include <net/route.h>
117 
118 #include <netinet/in.h>
119 #include <netinet/in_systm.h>
120 #include <netinet/in_var.h>
121 #include <netinet/ip.h>
122 #include <netinet/ip_icmp.h>
123 #include <netinet/ip_var.h>
124 #include <netinet/in_pcb.h>
125 #include <netinet/in_proto.h>
126 #include <netinet/icmp_var.h>
127 #include <netinet/icmp_private.h>
128 #include <netinet/wqinput.h>
129 
130 #ifdef IPSEC
131 #include <netipsec/ipsec.h>
132 #include <netipsec/key.h>
133 #endif
134 
135 /*
136  * ICMP routines: error generation, receive packet processing, and
137  * routines to turnaround packets back to the originator, and
138  * host table maintenance routines.
139  */
140 
141 int icmpmaskrepl = 0;
142 int icmpbmcastecho = 0;
143 int icmpreturndatabytes = 8;
144 
145 percpu_t *icmpstat_percpu;
146 
147 /*
148  * List of callbacks to notify when Path MTU changes are made.
149  */
150 struct icmp_mtudisc_callback {
151 	LIST_ENTRY(icmp_mtudisc_callback) mc_list;
152 	void (*mc_func)(struct in_addr);
153 };
154 
155 LIST_HEAD(, icmp_mtudisc_callback) icmp_mtudisc_callbacks =
156     LIST_HEAD_INITIALIZER(&icmp_mtudisc_callbacks);
157 
158 /* unused... */
159 u_int ip_next_mtu(u_int, int);
160 
161 static int icmperrppslim = 100;			/* 100pps */
162 static int icmperrpps_count = 0;
163 static struct timeval icmperrppslim_last;
164 static int icmp_rediraccept = 1;
165 static int icmp_redirtimeout = 600;
166 static struct rttimer_queue *icmp_redirect_timeout_q = NULL;
167 
168 /* Protect mtudisc and redirect stuff */
169 static kmutex_t icmp_mtx __cacheline_aligned;
170 
171 static void icmp_send(struct mbuf *, struct mbuf *);
172 static void icmp_mtudisc_timeout(struct rtentry *, struct rttimer *);
173 static void icmp_redirect_timeout(struct rtentry *, struct rttimer *);
174 
175 static void sysctl_netinet_icmp_setup(struct sysctllog **);
176 
177 /* workqueue-based pr_input */
178 static struct wqinput *icmp_wqinput;
179 static void _icmp_input(struct mbuf *, int, int);
180 
181 void
182 icmp_init(void)
183 {
184 
185 	sysctl_netinet_icmp_setup(NULL);
186 
187 	mutex_init(&icmp_mtx, MUTEX_DEFAULT, IPL_NONE);
188 	/*
189 	 * This is only useful if the user initializes redirtimeout to
190 	 * something other than zero.
191 	 */
192 	mutex_enter(&icmp_mtx);
193 	icmp_redirect_timeout_q = rt_timer_queue_create(icmp_redirtimeout);
194 	mutex_exit(&icmp_mtx);
195 
196 	icmpstat_percpu = percpu_alloc(sizeof(uint64_t) * ICMP_NSTATS);
197 	icmp_wqinput = wqinput_create("icmp", _icmp_input);
198 }
199 
200 void
201 icmp_mtudisc_lock(void)
202 {
203 
204 	mutex_enter(&icmp_mtx);
205 }
206 
207 void
208 icmp_mtudisc_unlock(void)
209 {
210 
211 	mutex_exit(&icmp_mtx);
212 }
213 
214 /*
215  * Register a Path MTU Discovery callback.
216  */
217 void
218 icmp_mtudisc_callback_register(void (*func)(struct in_addr))
219 {
220 	struct icmp_mtudisc_callback *mc, *new;
221 
222 	new = kmem_alloc(sizeof(*mc), KM_SLEEP);
223 
224 	mutex_enter(&icmp_mtx);
225 	for (mc = LIST_FIRST(&icmp_mtudisc_callbacks); mc != NULL;
226 	     mc = LIST_NEXT(mc, mc_list)) {
227 		if (mc->mc_func == func) {
228 			mutex_exit(&icmp_mtx);
229 			kmem_free(new, sizeof(*mc));
230 			return;
231 		}
232 	}
233 
234 	new->mc_func = func;
235 	LIST_INSERT_HEAD(&icmp_mtudisc_callbacks, new, mc_list);
236 	mutex_exit(&icmp_mtx);
237 }
238 
239 /*
240  * Generate an error packet of type error in response to a bad IP packet. 'n'
241  * contains this packet. We create 'm' and send it.
242  *
243  * As we are not required to return everything we have, we return whatever
244  * we can return at ease.
245  *
246  * Note that ICMP datagrams longer than 576 octets are out of spec according
247  * to RFC1812; the limit on icmpreturndatabytes will keep things below that
248  * limit.
249  */
250 void
251 icmp_error(struct mbuf *n, int type, int code, n_long dest, int destmtu)
252 {
253 	struct ip *oip = mtod(n, struct ip *), *nip;
254 	const unsigned oiphlen = oip->ip_hl << 2;
255 	struct icmp *icp;
256 	struct mbuf *m;
257 	struct m_tag *mtag;
258 	unsigned datalen, mblen;
259 	int totlen;
260 
261 	if (type != ICMP_REDIRECT)
262 		ICMP_STATINC(ICMP_STAT_ERROR);
263 
264 	/*
265 	 * Don't send error if:
266 	 *  - The original packet was encrypted.
267 	 *  - The packet is multicast or broadcast.
268 	 *  - The packet is not the first fragment of the message.
269 	 *  - The packet is an ICMP message with an unknown type.
270 	 */
271 	if (n->m_flags & M_DECRYPTED)
272 		goto freeit;
273 	if (n->m_flags & (M_BCAST|M_MCAST))
274 		goto freeit;
275 	if (oip->ip_off &~ htons(IP_MF|IP_DF))
276 		goto freeit;
277 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
278 	    n->m_len >= oiphlen + ICMP_MINLEN) {
279 		struct icmp *oicp = (struct icmp *)((char *)oip + oiphlen);
280 		if (!ICMP_INFOTYPE(oicp->icmp_type)) {
281 			ICMP_STATINC(ICMP_STAT_OLDICMP);
282 			goto freeit;
283 		}
284 	}
285 
286 	/*
287 	 * First, do a rate limitation check.
288 	 */
289 	if (icmp_ratelimit(&oip->ip_src, type, code)) {
290 		/* XXX stat */
291 		goto freeit;
292 	}
293 
294 	/*
295 	 * Compute the number of bytes we will put in 'icmp_ip'. Truncate
296 	 * it to the size of the mbuf, if it's too big.
297 	 */
298 	datalen = oiphlen + min(icmpreturndatabytes,
299 	    ntohs(oip->ip_len) - oiphlen);
300 	mblen = 0;
301 	for (m = n; m && (mblen < datalen); m = m->m_next)
302 		mblen += m->m_len;
303 	datalen = min(mblen, datalen);
304 
305 	/*
306 	 * Compute the total length of the new packet. Truncate it if it's
307 	 * bigger than the size of a cluster.
308 	 */
309 	CTASSERT(ICMP_MINLEN + sizeof(struct ip) <= MCLBYTES);
310 	totlen = sizeof(struct ip) + ICMP_MINLEN + datalen;
311 	if (totlen > MCLBYTES) {
312 		datalen = MCLBYTES - ICMP_MINLEN - sizeof(struct ip);
313 		totlen = MCLBYTES;
314 	}
315 
316 	/*
317 	 * Allocate the mbuf for the new packet.
318 	 */
319 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
320 	if (m && (totlen > MHLEN)) {
321 		MCLGET(m, M_DONTWAIT);
322 		if ((m->m_flags & M_EXT) == 0) {
323 			m_freem(m);
324 			m = NULL;
325 		}
326 	}
327 	if (m == NULL)
328 		goto freeit;
329 	MCLAIM(m, n->m_owner);
330 	m->m_len = totlen;
331 	m->m_pkthdr.len = m->m_len;
332 	m_copy_rcvif(m, n);
333 
334 	if ((u_int)type > ICMP_MAXTYPE)
335 		panic("icmp_error");
336 	ICMP_STATINC(ICMP_STAT_OUTHIST + type);
337 
338 	if ((m->m_flags & M_EXT) == 0)
339 		MH_ALIGN(m, m->m_len);
340 
341 	/*
342 	 * Get pointers on the IP header and the ICMP header.
343 	 */
344 	nip = mtod(m, struct ip *);
345 	icp = (struct icmp *)(nip + 1);
346 
347 	/*
348 	 * Fill in the fields of the ICMP header: icmp_type, icmp_code
349 	 * and icmp_ip. icmp_cksum gets filled later.
350 	 */
351 	icp->icmp_type = type;
352 	if (type == ICMP_REDIRECT) {
353 		icp->icmp_gwaddr.s_addr = dest;
354 	} else {
355 		icp->icmp_void = 0;
356 		/*
357 		 * The following assignments assume an overlay with the
358 		 * zeroed icmp_void field.
359 		 */
360 		if (type == ICMP_PARAMPROB) {
361 			icp->icmp_pptr = code;
362 			code = 0;
363 		} else if (type == ICMP_UNREACH &&
364 		    code == ICMP_UNREACH_NEEDFRAG && destmtu)
365 			icp->icmp_nextmtu = htons(destmtu);
366 	}
367 	icp->icmp_code = code;
368 	m_copydata(n, 0, datalen, (void *)&icp->icmp_ip);
369 
370 	/*
371 	 * Now, copy the old IP header (without options) in front of the
372 	 * ICMP message. The src/dst fields will be swapped in icmp_reflect.
373 	 */
374 	/* ip_v set in ip_output */
375 	nip->ip_hl = sizeof(struct ip) >> 2;
376 	nip->ip_tos = 0;
377 	nip->ip_len = htons(m->m_len);
378 	/* ip_id set in ip_output */
379 	nip->ip_off = htons(0);
380 	/* ip_ttl set in icmp_reflect */
381 	nip->ip_p = IPPROTO_ICMP;
382 	nip->ip_src = oip->ip_src;
383 	nip->ip_dst = oip->ip_dst;
384 	/* move PF m_tag to new packet, if it exists */
385 	mtag = m_tag_find(n, PACKET_TAG_PF, NULL);
386 	if (mtag != NULL) {
387 		m_tag_unlink(n, mtag);
388 		m_tag_prepend(m, mtag);
389 	}
390 
391 	icmp_reflect(m);
392 
393 freeit:
394 	m_freem(n);
395 }
396 
397 struct sockaddr_in icmpsrc = {
398 	.sin_len = sizeof(struct sockaddr_in),
399 	.sin_family = AF_INET,
400 };
401 
402 /*
403  * Process a received ICMP message.
404  */
405 static void
406 _icmp_input(struct mbuf *m, int hlen, int proto)
407 {
408 	struct icmp *icp;
409 	struct ip *ip = mtod(m, struct ip *);
410 	int icmplen;
411 	int i;
412 	struct in_ifaddr *ia;
413 	void *(*ctlfunc)(int, const struct sockaddr *, void *);
414 	int code;
415 	struct rtentry *rt;
416 	struct sockaddr_in icmpdst = {
417 		.sin_len = sizeof(struct sockaddr_in),
418 		.sin_family = AF_INET,
419 	};
420 	struct sockaddr_in icmpgw = {
421 		.sin_len = sizeof(struct sockaddr_in),
422 		.sin_family = AF_INET,
423 	};
424 
425 	/*
426 	 * Locate icmp structure in mbuf, and check
427 	 * that not corrupted and of at least minimum length.
428 	 */
429 	icmplen = ntohs(ip->ip_len) - hlen;
430 	if (icmplen < ICMP_MINLEN) {
431 		ICMP_STATINC(ICMP_STAT_TOOSHORT);
432 		goto freeit;
433 	}
434 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
435 	if (M_UNWRITABLE(m, i) && (m = m_pullup(m, i)) == NULL) {
436 		ICMP_STATINC(ICMP_STAT_TOOSHORT);
437 		return;
438 	}
439 	ip = mtod(m, struct ip *);
440 	m->m_len -= hlen;
441 	m->m_data += hlen;
442 	icp = mtod(m, struct icmp *);
443 	/* Don't need to assert alignment, here. */
444 	if (in_cksum(m, icmplen)) {
445 		ICMP_STATINC(ICMP_STAT_CHECKSUM);
446 		goto freeit;
447 	}
448 	m->m_len += hlen;
449 	m->m_data -= hlen;
450 
451 	if (icp->icmp_type > ICMP_MAXTYPE)
452 		goto raw;
453 	ICMP_STATINC(ICMP_STAT_INHIST + icp->icmp_type);
454 	code = icp->icmp_code;
455 
456 	switch (icp->icmp_type) {
457 	case ICMP_UNREACH:
458 		switch (code) {
459 		case ICMP_UNREACH_PROTOCOL:
460 			code = PRC_UNREACH_PROTOCOL;
461 			break;
462 
463 		case ICMP_UNREACH_PORT:
464 			code = PRC_UNREACH_PORT;
465 			break;
466 
467 		case ICMP_UNREACH_SRCFAIL:
468 			code = PRC_UNREACH_SRCFAIL;
469 			break;
470 
471 		case ICMP_UNREACH_NEEDFRAG:
472 			code = PRC_MSGSIZE;
473 			break;
474 
475 		case ICMP_UNREACH_NET:
476 		case ICMP_UNREACH_NET_UNKNOWN:
477 		case ICMP_UNREACH_NET_PROHIB:
478 		case ICMP_UNREACH_TOSNET:
479 			code = PRC_UNREACH_NET;
480 			break;
481 
482 		case ICMP_UNREACH_HOST:
483 		case ICMP_UNREACH_HOST_UNKNOWN:
484 		case ICMP_UNREACH_ISOLATED:
485 		case ICMP_UNREACH_HOST_PROHIB:
486 		case ICMP_UNREACH_TOSHOST:
487 		case ICMP_UNREACH_ADMIN_PROHIBIT:
488 		case ICMP_UNREACH_HOST_PREC:
489 		case ICMP_UNREACH_PREC_CUTOFF:
490 			code = PRC_UNREACH_HOST;
491 			break;
492 
493 		default:
494 			goto badcode;
495 		}
496 		goto deliver;
497 
498 	case ICMP_TIMXCEED:
499 		if (code > 1)
500 			goto badcode;
501 		code += PRC_TIMXCEED_INTRANS;
502 		goto deliver;
503 
504 	case ICMP_PARAMPROB:
505 		if (code > 1)
506 			goto badcode;
507 		code = PRC_PARAMPROB;
508 		goto deliver;
509 
510 	case ICMP_SOURCEQUENCH:
511 		if (code)
512 			goto badcode;
513 		code = PRC_QUENCH;
514 		goto deliver;
515 
516 	deliver:
517 		/*
518 		 * Problem with datagram; advise higher level routines.
519 		 */
520 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
521 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
522 			ICMP_STATINC(ICMP_STAT_BADLEN);
523 			goto freeit;
524 		}
525 		if (m->m_len < hlen + ICMP_ADVLEN(icp)) {
526 			m = m_pullup(m, hlen + ICMP_ADVLEN(icp));
527 			if (m == NULL)
528 				goto freeit;
529 		}
530 		ip = mtod(m, struct ip *);
531 		icp = (struct icmp *)(mtod(m, uint8_t *) + hlen);
532 
533 		if (IN_MULTICAST(icp->icmp_ip.ip_dst.s_addr))
534 			goto badcode;
535 
536 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
537 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
538 		if (ctlfunc)
539 			(void) (*ctlfunc)(code, sintosa(&icmpsrc),
540 			    &icp->icmp_ip);
541 		break;
542 
543 	badcode:
544 		ICMP_STATINC(ICMP_STAT_BADCODE);
545 		break;
546 
547 	case ICMP_ECHO:
548 		if (!icmpbmcastecho &&
549 		    (m->m_flags & (M_MCAST | M_BCAST)) != 0)  {
550 			ICMP_STATINC(ICMP_STAT_BMCASTECHO);
551 			break;
552 		}
553 		icp->icmp_type = ICMP_ECHOREPLY;
554 		goto reflect;
555 
556 	case ICMP_TSTAMP:
557 		if (icmplen < ICMP_TSLEN) {
558 			ICMP_STATINC(ICMP_STAT_BADLEN);
559 			break;
560 		}
561 		if (!icmpbmcastecho &&
562 		    (m->m_flags & (M_MCAST | M_BCAST)) != 0)  {
563 			ICMP_STATINC(ICMP_STAT_BMCASTTSTAMP);
564 			break;
565 		}
566 		icp->icmp_type = ICMP_TSTAMPREPLY;
567 		icp->icmp_rtime = iptime();
568 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
569 		goto reflect;
570 
571 	case ICMP_MASKREQ: {
572 		struct ifnet *rcvif;
573 		int s, ss;
574 		struct ifaddr *ifa = NULL;
575 
576 		if (icmpmaskrepl == 0)
577 			break;
578 		/*
579 		 * We are not able to respond with all ones broadcast
580 		 * unless we receive it over a point-to-point interface.
581 		 */
582 		if (icmplen < ICMP_MASKLEN) {
583 			ICMP_STATINC(ICMP_STAT_BADLEN);
584 			break;
585 		}
586 		if (ip->ip_dst.s_addr == INADDR_BROADCAST ||
587 		    in_nullhost(ip->ip_dst))
588 			icmpdst.sin_addr = ip->ip_src;
589 		else
590 			icmpdst.sin_addr = ip->ip_dst;
591 		ss = pserialize_read_enter();
592 		rcvif = m_get_rcvif(m, &s);
593 		if (__predict_true(rcvif != NULL))
594 			ifa = ifaof_ifpforaddr(sintosa(&icmpdst), rcvif);
595 		m_put_rcvif(rcvif, &s);
596 		if (ifa == NULL) {
597 			pserialize_read_exit(ss);
598 			break;
599 		}
600 		ia = ifatoia(ifa);
601 		icp->icmp_type = ICMP_MASKREPLY;
602 		icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
603 		if (in_nullhost(ip->ip_src)) {
604 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
605 				ip->ip_src = ia->ia_broadaddr.sin_addr;
606 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
607 				ip->ip_src = ia->ia_dstaddr.sin_addr;
608 		}
609 		pserialize_read_exit(ss);
610 reflect:
611 		{
612 			uint64_t *icps = percpu_getref(icmpstat_percpu);
613 			icps[ICMP_STAT_REFLECT]++;
614 			icps[ICMP_STAT_OUTHIST + icp->icmp_type]++;
615 			percpu_putref(icmpstat_percpu);
616 		}
617 		icmp_reflect(m);
618 		return;
619 	}
620 
621 	case ICMP_REDIRECT:
622 		if (code > 3)
623 			goto badcode;
624 		if (icmp_rediraccept == 0)
625 			goto freeit;
626 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
627 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
628 			ICMP_STATINC(ICMP_STAT_BADLEN);
629 			break;
630 		}
631 		/*
632 		 * Short circuit routing redirects to force
633 		 * immediate change in the kernel's routing
634 		 * tables.  The message is also handed to anyone
635 		 * listening on a raw socket (e.g. the routing
636 		 * daemon for use in updating its tables).
637 		 */
638 		icmpgw.sin_addr = ip->ip_src;
639 		icmpdst.sin_addr = icp->icmp_gwaddr;
640 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
641 		rt = NULL;
642 		rtredirect(sintosa(&icmpsrc), sintosa(&icmpdst),
643 		    NULL, RTF_GATEWAY | RTF_HOST, sintosa(&icmpgw), &rt);
644 		mutex_enter(&icmp_mtx);
645 		if (rt != NULL && icmp_redirtimeout != 0) {
646 			i = rt_timer_add(rt, icmp_redirect_timeout,
647 					 icmp_redirect_timeout_q);
648 			if (i) {
649 				char buf[INET_ADDRSTRLEN];
650 				log(LOG_ERR, "ICMP:  redirect failed to "
651 				    "register timeout for route to %s, "
652 				    "code %d\n",
653 				    IN_PRINT(buf, &icp->icmp_ip.ip_dst), i);
654 			}
655 		}
656 		mutex_exit(&icmp_mtx);
657 		if (rt != NULL)
658 			rt_unref(rt);
659 
660 		pfctlinput(PRC_REDIRECT_HOST, sintosa(&icmpsrc));
661 #if defined(IPSEC)
662 		if (ipsec_used)
663 			key_sa_routechange((struct sockaddr *)&icmpsrc);
664 #endif
665 		break;
666 
667 	/*
668 	 * No kernel processing for the following;
669 	 * just fall through to send to raw listener.
670 	 */
671 	case ICMP_ECHOREPLY:
672 	case ICMP_ROUTERADVERT:
673 	case ICMP_ROUTERSOLICIT:
674 	case ICMP_TSTAMPREPLY:
675 	case ICMP_IREQREPLY:
676 	case ICMP_MASKREPLY:
677 	default:
678 		break;
679 	}
680 
681 raw:
682 	/*
683 	 * Currently, pim_input() is always called holding softnet_lock
684 	 * by ipintr()(!NET_MPSAFE) or PR_INPUT_WRAP()(NET_MPSAFE).
685 	 */
686 	KASSERT(mutex_owned(softnet_lock));
687 	rip_input(m, hlen, proto);
688 	return;
689 
690 freeit:
691 	m_freem(m);
692 	return;
693 }
694 
695 void
696 icmp_input(struct mbuf *m, ...)
697 {
698 	int hlen, proto;
699 	va_list ap;
700 
701 	va_start(ap, m);
702 	hlen = va_arg(ap, int);
703 	proto = va_arg(ap, int);
704 	va_end(ap);
705 
706 	wqinput_input(icmp_wqinput, m, hlen, proto);
707 }
708 
709 /*
710  * Reflect the ip packet back to the source
711  */
712 void
713 icmp_reflect(struct mbuf *m)
714 {
715 	struct ip *ip = mtod(m, struct ip *);
716 	struct in_ifaddr *ia;
717 	struct ifaddr *ifa;
718 	struct sockaddr_in *sin;
719 	struct in_addr t;
720 	struct mbuf *opts = NULL;
721 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
722 	struct ifnet *rcvif;
723 	struct psref psref, psref_ia;
724 	int s;
725 	int bound;
726 
727 	bound = curlwp_bind();
728 
729 	if (!in_canforward(ip->ip_src) &&
730 	    ((ip->ip_src.s_addr & IN_CLASSA_NET) !=
731 	     htonl(IN_LOOPBACKNET << IN_CLASSA_NSHIFT))) {
732 		m_freem(m);	/* Bad return address */
733 		goto done;	/* ip_output() will check for broadcast */
734 	}
735 	t = ip->ip_dst;
736 	ip->ip_dst = ip->ip_src;
737 
738 	/*
739 	 * If the incoming packet was addressed directly to us, use
740 	 * dst as the src for the reply.  Otherwise (broadcast or
741 	 * anonymous), use an address which corresponds to the
742 	 * incoming interface, with a preference for the address which
743 	 * corresponds to the route to the destination of the ICMP.
744 	 */
745 
746 	/* Look for packet addressed to us */
747 	ia = in_get_ia_psref(t, &psref_ia);
748 	if (ia && (ia->ia4_flags & IN_IFF_NOTREADY)) {
749 		ia4_release(ia, &psref_ia);
750 		ia = NULL;
751 	}
752 
753 	rcvif = m_get_rcvif_psref(m, &psref);
754 
755 	/* look for packet sent to broadcast address */
756 	if (ia == NULL && rcvif &&
757 	    (rcvif->if_flags & IFF_BROADCAST)) {
758 		s = pserialize_read_enter();
759 		IFADDR_READER_FOREACH(ifa, rcvif) {
760 			if (ifa->ifa_addr->sa_family != AF_INET)
761 				continue;
762 			if (in_hosteq(t,ifatoia(ifa)->ia_broadaddr.sin_addr)) {
763 				ia = ifatoia(ifa);
764 				if ((ia->ia4_flags & IN_IFF_NOTREADY) == 0)
765 					break;
766 				ia = NULL;
767 			}
768 		}
769 		if (ia != NULL)
770 			ia4_acquire(ia, &psref_ia);
771 		pserialize_read_exit(s);
772 	}
773 
774 	sin = ia ? &ia->ia_addr : NULL;
775 
776 	/*
777 	 * if the packet is addressed somewhere else, compute the
778 	 * source address for packets routed back to the source, and
779 	 * use that, if it's an address on the interface which
780 	 * received the packet
781 	 */
782 	if (sin == NULL && rcvif) {
783 		struct sockaddr_in sin_dst;
784 		struct route icmproute;
785 		int errornum;
786 
787 		sockaddr_in_init(&sin_dst, &ip->ip_dst, 0);
788 		memset(&icmproute, 0, sizeof(icmproute));
789 		errornum = 0;
790 		ia = in_selectsrc(&sin_dst, &icmproute, 0, NULL, &errornum,
791 		    &psref_ia);
792 		/* errornum is never used */
793 		rtcache_free(&icmproute);
794 		/* check to make sure sin is a source address on rcvif */
795 		if (ia != NULL) {
796 			sin = &ia->ia_addr;
797 			t = sin->sin_addr;
798 			sin = NULL;
799 			ia4_release(ia, &psref_ia);
800 			ia = in_get_ia_on_iface_psref(t, rcvif, &psref_ia);
801 			if (ia != NULL)
802 				sin = &ia->ia_addr;
803 		}
804 	}
805 
806 	/*
807 	 * if it was not addressed to us, but the route doesn't go out
808 	 * the source interface, pick an address on the source
809 	 * interface.  This can happen when routing is asymmetric, or
810 	 * when the incoming packet was encapsulated
811 	 */
812 	if (sin == NULL && rcvif) {
813 		KASSERT(ia == NULL);
814 		s = pserialize_read_enter();
815 		IFADDR_READER_FOREACH(ifa, rcvif) {
816 			if (ifa->ifa_addr->sa_family != AF_INET)
817 				continue;
818 			sin = &(ifatoia(ifa)->ia_addr);
819 			ia = ifatoia(ifa);
820 			ia4_acquire(ia, &psref_ia);
821 			break;
822 		}
823 		pserialize_read_exit(s);
824 	}
825 
826 	m_put_rcvif_psref(rcvif, &psref);
827 
828 	/*
829 	 * The following happens if the packet was not addressed to us,
830 	 * and was received on an interface with no IP address:
831 	 * We find the first AF_INET address on the first non-loopback
832 	 * interface.
833 	 */
834 	if (sin == NULL) {
835 		KASSERT(ia == NULL);
836 		s = pserialize_read_enter();
837 		IN_ADDRLIST_READER_FOREACH(ia) {
838 			if (ia->ia_ifp->if_flags & IFF_LOOPBACK)
839 				continue;
840 			sin = &ia->ia_addr;
841 			ia4_acquire(ia, &psref_ia);
842 			break;
843 		}
844 		pserialize_read_exit(s);
845 	}
846 
847 	/*
848 	 * If we still didn't find an address, punt.  We could have an
849 	 * interface up (and receiving packets) with no address.
850 	 */
851 	if (sin == NULL) {
852 		KASSERT(ia == NULL);
853 		m_freem(m);
854 		goto done;
855 	}
856 
857 	ip->ip_src = sin->sin_addr;
858 	ip->ip_ttl = MAXTTL;
859 
860 	if (ia != NULL)
861 		ia4_release(ia, &psref_ia);
862 
863 	if (optlen > 0) {
864 		u_char *cp;
865 		int opt, cnt;
866 		u_int len;
867 
868 		/*
869 		 * Retrieve any source routing from the incoming packet;
870 		 * add on any record-route or timestamp options.
871 		 */
872 		cp = (u_char *)(ip + 1);
873 		if ((opts = ip_srcroute(m)) == NULL &&
874 		    (opts = m_gethdr(M_DONTWAIT, MT_HEADER))) {
875 			MCLAIM(opts, m->m_owner);
876 			opts->m_len = sizeof(struct in_addr);
877 			*mtod(opts, struct in_addr *) = zeroin_addr;
878 		}
879 
880 		if (opts) {
881 			for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
882 				opt = cp[IPOPT_OPTVAL];
883 				if (opt == IPOPT_EOL)
884 					break;
885 				if (opt == IPOPT_NOP)
886 					len = 1;
887 				else {
888 					if (cnt < IPOPT_OLEN + sizeof(*cp))
889 						break;
890 					len = cp[IPOPT_OLEN];
891 					if (len < IPOPT_OLEN + sizeof(*cp) ||
892 					    len > cnt)
893 						break;
894 				}
895 
896 				/* Overflows can't happen */
897 				KASSERT(opts->m_len + len <= MHLEN);
898 
899 				if (opt == IPOPT_RR || opt == IPOPT_TS ||
900 				    opt == IPOPT_SECURITY) {
901 					memmove(mtod(opts, char *) +
902 					    opts->m_len, cp, len);
903 					opts->m_len += len;
904 				}
905 			}
906 
907 			/* Terminate & pad, if necessary */
908 			if ((cnt = opts->m_len % 4) != 0) {
909 				for (; cnt < 4; cnt++) {
910 					*(mtod(opts, char *) + opts->m_len) =
911 					    IPOPT_EOL;
912 					opts->m_len++;
913 				}
914 			}
915 		}
916 
917 		/*
918 		 * Now strip out original options by copying rest of first
919 		 * mbuf's data back, and adjust the IP length.
920 		 */
921 		ip->ip_len = htons(ntohs(ip->ip_len) - optlen);
922 		ip->ip_hl = sizeof(struct ip) >> 2;
923 		m->m_len -= optlen;
924 		if (m->m_flags & M_PKTHDR)
925 			m->m_pkthdr.len -= optlen;
926 		optlen += sizeof(struct ip);
927 		memmove(ip + 1, (char *)ip + optlen,
928 		    (unsigned)(m->m_len - sizeof(struct ip)));
929 	}
930 	m_tag_delete_nonpersistent(m);
931 	m->m_flags &= ~(M_BCAST|M_MCAST);
932 
933 	/*
934 	 * Clear any in-bound checksum flags for this packet.
935 	 */
936 	if (m->m_flags & M_PKTHDR)
937 		m->m_pkthdr.csum_flags = 0;
938 
939 	icmp_send(m, opts);
940 done:
941 	curlwp_bindx(bound);
942 	if (opts)
943 		(void)m_free(opts);
944 }
945 
946 /*
947  * Send an icmp packet back to the ip level,
948  * after supplying a checksum.
949  */
950 static void
951 icmp_send(struct mbuf *m, struct mbuf *opts)
952 {
953 	struct ip *ip = mtod(m, struct ip *);
954 	int hlen;
955 	struct icmp *icp;
956 
957 	hlen = ip->ip_hl << 2;
958 	m->m_data += hlen;
959 	m->m_len -= hlen;
960 	icp = mtod(m, struct icmp *);
961 	icp->icmp_cksum = 0;
962 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
963 	m->m_data -= hlen;
964 	m->m_len += hlen;
965 
966 	(void)ip_output(m, opts, NULL, 0, NULL, NULL);
967 }
968 
969 n_time
970 iptime(void)
971 {
972 	struct timeval atv;
973 	u_long t;
974 
975 	microtime(&atv);
976 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
977 	return (htonl(t));
978 }
979 
980 /*
981  * sysctl helper routine for net.inet.icmp.returndatabytes.  ensures
982  * that the new value is in the correct range.
983  */
984 static int
985 sysctl_net_inet_icmp_returndatabytes(SYSCTLFN_ARGS)
986 {
987 	int error, t;
988 	struct sysctlnode node;
989 
990 	node = *rnode;
991 	node.sysctl_data = &t;
992 	t = icmpreturndatabytes;
993 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
994 	if (error || newp == NULL)
995 		return error;
996 
997 	if (t < 8 || t > 512)
998 		return EINVAL;
999 	icmpreturndatabytes = t;
1000 
1001 	return 0;
1002 }
1003 
1004 /*
1005  * sysctl helper routine for net.inet.icmp.redirtimeout.  ensures that
1006  * the given value is not less than zero and then resets the timeout
1007  * queue.
1008  */
1009 static int
1010 sysctl_net_inet_icmp_redirtimeout(SYSCTLFN_ARGS)
1011 {
1012 	int error, tmp;
1013 	struct sysctlnode node;
1014 
1015 	mutex_enter(&icmp_mtx);
1016 
1017 	node = *rnode;
1018 	node.sysctl_data = &tmp;
1019 	tmp = icmp_redirtimeout;
1020 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1021 	if (error || newp == NULL)
1022 		goto out;
1023 	if (tmp < 0) {
1024 		error = EINVAL;
1025 		goto out;
1026 	}
1027 	icmp_redirtimeout = tmp;
1028 
1029 	/*
1030 	 * was it a *defined* side-effect that anyone even *reading*
1031 	 * this value causes these things to happen?
1032 	 */
1033 	if (icmp_redirect_timeout_q != NULL) {
1034 		if (icmp_redirtimeout == 0) {
1035 			rt_timer_queue_destroy(icmp_redirect_timeout_q);
1036 			icmp_redirect_timeout_q = NULL;
1037 		} else {
1038 			rt_timer_queue_change(icmp_redirect_timeout_q,
1039 			    icmp_redirtimeout);
1040 		}
1041 	} else if (icmp_redirtimeout > 0) {
1042 		icmp_redirect_timeout_q =
1043 		    rt_timer_queue_create(icmp_redirtimeout);
1044 	}
1045 	error = 0;
1046 out:
1047 	mutex_exit(&icmp_mtx);
1048 	return error;
1049 }
1050 
1051 static int
1052 sysctl_net_inet_icmp_stats(SYSCTLFN_ARGS)
1053 {
1054 
1055 	return (NETSTAT_SYSCTL(icmpstat_percpu, ICMP_NSTATS));
1056 }
1057 
1058 static void
1059 sysctl_netinet_icmp_setup(struct sysctllog **clog)
1060 {
1061 
1062 	sysctl_createv(clog, 0, NULL, NULL,
1063 		       CTLFLAG_PERMANENT,
1064 		       CTLTYPE_NODE, "inet", NULL,
1065 		       NULL, 0, NULL, 0,
1066 		       CTL_NET, PF_INET, CTL_EOL);
1067 	sysctl_createv(clog, 0, NULL, NULL,
1068 		       CTLFLAG_PERMANENT,
1069 		       CTLTYPE_NODE, "icmp",
1070 		       SYSCTL_DESCR("ICMPv4 related settings"),
1071 		       NULL, 0, NULL, 0,
1072 		       CTL_NET, PF_INET, IPPROTO_ICMP, CTL_EOL);
1073 
1074 	sysctl_createv(clog, 0, NULL, NULL,
1075 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1076 		       CTLTYPE_INT, "maskrepl",
1077 		       SYSCTL_DESCR("Respond to ICMP_MASKREQ messages"),
1078 		       NULL, 0, &icmpmaskrepl, 0,
1079 		       CTL_NET, PF_INET, IPPROTO_ICMP,
1080 		       ICMPCTL_MASKREPL, CTL_EOL);
1081 	sysctl_createv(clog, 0, NULL, NULL,
1082 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1083 		       CTLTYPE_INT, "returndatabytes",
1084 		       SYSCTL_DESCR("Number of bytes to return in an ICMP "
1085 				    "error message"),
1086 		       sysctl_net_inet_icmp_returndatabytes, 0,
1087 		       &icmpreturndatabytes, 0,
1088 		       CTL_NET, PF_INET, IPPROTO_ICMP,
1089 		       ICMPCTL_RETURNDATABYTES, CTL_EOL);
1090 	sysctl_createv(clog, 0, NULL, NULL,
1091 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1092 		       CTLTYPE_INT, "errppslimit",
1093 		       SYSCTL_DESCR("Maximum number of outgoing ICMP error "
1094 				    "messages per second"),
1095 		       NULL, 0, &icmperrppslim, 0,
1096 		       CTL_NET, PF_INET, IPPROTO_ICMP,
1097 		       ICMPCTL_ERRPPSLIMIT, CTL_EOL);
1098 	sysctl_createv(clog, 0, NULL, NULL,
1099 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1100 		       CTLTYPE_INT, "rediraccept",
1101 		       SYSCTL_DESCR("Accept ICMP_REDIRECT messages"),
1102 		       NULL, 0, &icmp_rediraccept, 0,
1103 		       CTL_NET, PF_INET, IPPROTO_ICMP,
1104 		       ICMPCTL_REDIRACCEPT, CTL_EOL);
1105 	sysctl_createv(clog, 0, NULL, NULL,
1106 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1107 		       CTLTYPE_INT, "redirtimeout",
1108 		       SYSCTL_DESCR("Lifetime of ICMP_REDIRECT generated "
1109 				    "routes"),
1110 		       sysctl_net_inet_icmp_redirtimeout, 0,
1111 		       &icmp_redirtimeout, 0,
1112 		       CTL_NET, PF_INET, IPPROTO_ICMP,
1113 		       ICMPCTL_REDIRTIMEOUT, CTL_EOL);
1114 	sysctl_createv(clog, 0, NULL, NULL,
1115 		       CTLFLAG_PERMANENT,
1116 		       CTLTYPE_STRUCT, "stats",
1117 		       SYSCTL_DESCR("ICMP statistics"),
1118 		       sysctl_net_inet_icmp_stats, 0, NULL, 0,
1119 		       CTL_NET, PF_INET, IPPROTO_ICMP, ICMPCTL_STATS,
1120 		       CTL_EOL);
1121 	sysctl_createv(clog, 0, NULL, NULL,
1122 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1123 		       CTLTYPE_INT, "bmcastecho",
1124 		       SYSCTL_DESCR("Respond to ICMP_ECHO or ICMP_TIMESTAMP "
1125 				    "message to the broadcast or multicast"),
1126 		       NULL, 0, &icmpbmcastecho, 0,
1127 		       CTL_NET, PF_INET, IPPROTO_ICMP, ICMPCTL_BMCASTECHO,
1128 		       CTL_EOL);
1129 }
1130 
1131 void
1132 icmp_statinc(u_int stat)
1133 {
1134 
1135 	KASSERT(stat < ICMP_NSTATS);
1136 	ICMP_STATINC(stat);
1137 }
1138 
1139 /* Table of common MTUs */
1140 static const u_int mtu_table[] = {
1141 	65535, 65280, 32000, 17914, 9180, 8166,
1142 	4352, 2002, 1492, 1006, 508, 296, 68, 0
1143 };
1144 
1145 void
1146 icmp_mtudisc(struct icmp *icp, struct in_addr faddr)
1147 {
1148 	struct icmp_mtudisc_callback *mc;
1149 	struct sockaddr *dst = sintosa(&icmpsrc);
1150 	struct rtentry *rt;
1151 	u_long mtu = ntohs(icp->icmp_nextmtu);  /* Why a long?  IPv6 */
1152 	int error;
1153 
1154 	rt = rtalloc1(dst, 1);
1155 	if (rt == NULL)
1156 		return;
1157 
1158 	/* If we didn't get a host route, allocate one */
1159 	if ((rt->rt_flags & RTF_HOST) == 0) {
1160 		struct rtentry *nrt;
1161 
1162 		error = rtrequest(RTM_ADD, dst, rt->rt_gateway, NULL,
1163 		    RTF_GATEWAY | RTF_HOST | RTF_DYNAMIC, &nrt);
1164 		if (error) {
1165 			rt_unref(rt);
1166 			return;
1167 		}
1168 		nrt->rt_rmx = rt->rt_rmx;
1169 		rt_unref(rt);
1170 		rt = nrt;
1171 	}
1172 
1173 	mutex_enter(&icmp_mtx);
1174 	error = rt_timer_add(rt, icmp_mtudisc_timeout, ip_mtudisc_timeout_q);
1175 	mutex_exit(&icmp_mtx);
1176 	if (error) {
1177 		rt_unref(rt);
1178 		return;
1179 	}
1180 
1181 	if (mtu == 0) {
1182 		int i = 0;
1183 
1184 		mtu = ntohs(icp->icmp_ip.ip_len);
1185 		/* Some 4.2BSD-based routers incorrectly adjust the ip_len */
1186 		if (mtu > rt->rt_rmx.rmx_mtu && rt->rt_rmx.rmx_mtu != 0)
1187 			mtu -= (icp->icmp_ip.ip_hl << 2);
1188 
1189 		/* If we still can't guess a value, try the route */
1190 		if (mtu == 0) {
1191 			mtu = rt->rt_rmx.rmx_mtu;
1192 
1193 			/* If no route mtu, default to the interface mtu */
1194 			if (mtu == 0)
1195 				mtu = rt->rt_ifp->if_mtu;
1196 		}
1197 
1198 		for (i = 0; i < sizeof(mtu_table) / sizeof(mtu_table[0]); i++) {
1199 			if (mtu > mtu_table[i]) {
1200 				mtu = mtu_table[i];
1201 				break;
1202 			}
1203 		}
1204 	}
1205 
1206 	/*
1207 	 * XXX:   RTV_MTU is overloaded, since the admin can set it
1208 	 *	  to turn off PMTU for a route, and the kernel can
1209 	 *	  set it to indicate a serious problem with PMTU
1210 	 *	  on a route.  We should be using a separate flag
1211 	 *	  for the kernel to indicate this.
1212 	 */
1213 
1214 	if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0) {
1215 		if (mtu < 296 || mtu > rt->rt_ifp->if_mtu)
1216 			rt->rt_rmx.rmx_locks |= RTV_MTU;
1217 		else if (rt->rt_rmx.rmx_mtu > mtu ||
1218 			 rt->rt_rmx.rmx_mtu == 0) {
1219 			ICMP_STATINC(ICMP_STAT_PMTUCHG);
1220 			rt->rt_rmx.rmx_mtu = mtu;
1221 		}
1222 	}
1223 
1224 	if (rt != NULL)
1225 		rt_unref(rt);
1226 
1227 	/*
1228 	 * Notify protocols that the MTU for this destination
1229 	 * has changed.
1230 	 */
1231 	mutex_enter(&icmp_mtx);
1232 	for (mc = LIST_FIRST(&icmp_mtudisc_callbacks); mc != NULL;
1233 	     mc = LIST_NEXT(mc, mc_list))
1234 		(*mc->mc_func)(faddr);
1235 	mutex_exit(&icmp_mtx);
1236 }
1237 
1238 /*
1239  * Return the next larger or smaller MTU plateau (table from RFC 1191)
1240  * given current value MTU.  If DIR is less than zero, a larger plateau
1241  * is returned; otherwise, a smaller value is returned.
1242  */
1243 u_int
1244 ip_next_mtu(u_int mtu, int dir)	/* XXX unused */
1245 {
1246 	int i;
1247 
1248 	for (i = 0; i < (sizeof mtu_table) / (sizeof mtu_table[0]); i++) {
1249 		if (mtu >= mtu_table[i])
1250 			break;
1251 	}
1252 
1253 	if (dir < 0) {
1254 		if (i == 0) {
1255 			return 0;
1256 		} else {
1257 			return mtu_table[i - 1];
1258 		}
1259 	} else {
1260 		if (mtu_table[i] == 0) {
1261 			return 0;
1262 		} else if (mtu > mtu_table[i]) {
1263 			return mtu_table[i];
1264 		} else {
1265 			return mtu_table[i + 1];
1266 		}
1267 	}
1268 }
1269 
1270 static void
1271 icmp_mtudisc_timeout(struct rtentry *rt, struct rttimer *r)
1272 {
1273 	struct rtentry *retrt;
1274 
1275 	KASSERT(rt != NULL);
1276 	rt_assert_referenced(rt);
1277 
1278 	if ((rt->rt_flags & (RTF_DYNAMIC | RTF_HOST)) ==
1279 	    (RTF_DYNAMIC | RTF_HOST)) {
1280 		rtrequest(RTM_DELETE, rt_getkey(rt),
1281 		    rt->rt_gateway, rt_mask(rt), rt->rt_flags, &retrt);
1282 		rt_unref(rt);
1283 		rt_free(retrt);
1284 	} else {
1285 		if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0) {
1286 			rt->rt_rmx.rmx_mtu = 0;
1287 		}
1288 	}
1289 }
1290 
1291 static void
1292 icmp_redirect_timeout(struct rtentry *rt, struct rttimer *r)
1293 {
1294 	struct rtentry *retrt;
1295 
1296 	KASSERT(rt != NULL);
1297 	rt_assert_referenced(rt);
1298 
1299 	if ((rt->rt_flags & (RTF_DYNAMIC | RTF_HOST)) ==
1300 	    (RTF_DYNAMIC | RTF_HOST)) {
1301 		rtrequest(RTM_DELETE, rt_getkey(rt),
1302 		    rt->rt_gateway, rt_mask(rt), rt->rt_flags, &retrt);
1303 		rt_unref(rt);
1304 		rt_free(retrt);
1305 	}
1306 }
1307 
1308 /*
1309  * Perform rate limit check.
1310  * Returns 0 if it is okay to send the icmp packet.
1311  * Returns 1 if the router SHOULD NOT send this icmp packet due to rate
1312  * limitation.
1313  *
1314  * XXX per-destination/type check necessary?
1315  */
1316 int
1317 icmp_ratelimit(const struct in_addr *dst, const int type,
1318     const int code)
1319 {
1320 
1321 	/* PPS limit */
1322 	if (!ppsratecheck(&icmperrppslim_last, &icmperrpps_count,
1323 	    icmperrppslim)) {
1324 		/* The packet is subject to rate limit */
1325 		return 1;
1326 	}
1327 
1328 	/* okay to send */
1329 	return 0;
1330 }
1331