xref: /netbsd-src/sys/netinet/ip_icmp.c (revision 3fe138c1461e710931a50b66f08982c5f52c371f)
1 /*	$NetBSD: ip_icmp.c,v 1.26 1997/10/29 05:28:44 kml Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)ip_icmp.c	8.2 (Berkeley) 1/4/94
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/time.h>
45 #include <sys/kernel.h>
46 #include <sys/proc.h>
47 
48 #include <vm/vm.h>
49 #include <sys/sysctl.h>
50 
51 #include <net/if.h>
52 #include <net/route.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/in_var.h>
57 #include <netinet/ip.h>
58 #include <netinet/ip_icmp.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/icmp_var.h>
61 
62 #include <machine/stdarg.h>
63 
64 /*
65  * ICMP routines: error generation, receive packet processing, and
66  * routines to turnaround packets back to the originator, and
67  * host table maintenance routines.
68  */
69 
70 int	icmpmaskrepl = 0;
71 #ifdef ICMPPRINTFS
72 int	icmpprintfs = 0;
73 #endif
74 
75 extern	struct protosw inetsw[];
76 
77 static void icmp_mtudisc __P((struct icmp *));
78 
79 /*
80  * Generate an error packet of type error
81  * in response to bad packet ip.
82  */
83 void
84 icmp_error(n, type, code, dest, destifp)
85 	struct mbuf *n;
86 	int type, code;
87 	n_long dest;
88 	struct ifnet *destifp;
89 {
90 	register struct ip *oip = mtod(n, struct ip *), *nip;
91 	register unsigned oiplen = oip->ip_hl << 2;
92 	register struct icmp *icp;
93 	register struct mbuf *m;
94 	unsigned icmplen;
95 
96 #ifdef ICMPPRINTFS
97 	if (icmpprintfs)
98 		printf("icmp_error(%x, %d, %d)\n", oip, type, code);
99 #endif
100 	if (type != ICMP_REDIRECT)
101 		icmpstat.icps_error++;
102 	/*
103 	 * Don't send error if not the first fragment of message.
104 	 * Don't error if the old packet protocol was ICMP
105 	 * error message, only known informational types.
106 	 */
107 	if (oip->ip_off &~ (IP_MF|IP_DF))
108 		goto freeit;
109 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
110 	  n->m_len >= oiplen + ICMP_MINLEN &&
111 	  !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiplen))->icmp_type)) {
112 		icmpstat.icps_oldicmp++;
113 		goto freeit;
114 	}
115 	/* Don't send error in response to a multicast or broadcast packet */
116 	if (n->m_flags & (M_BCAST|M_MCAST))
117 		goto freeit;
118 	/*
119 	 * First, formulate icmp message
120 	 */
121 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
122 	if (m == NULL)
123 		goto freeit;
124 	icmplen = oiplen + min(8, oip->ip_len);
125 	m->m_len = icmplen + ICMP_MINLEN;
126 	MH_ALIGN(m, m->m_len);
127 	icp = mtod(m, struct icmp *);
128 	if ((u_int)type > ICMP_MAXTYPE)
129 		panic("icmp_error");
130 	icmpstat.icps_outhist[type]++;
131 	icp->icmp_type = type;
132 	if (type == ICMP_REDIRECT)
133 		icp->icmp_gwaddr.s_addr = dest;
134 	else {
135 		icp->icmp_void = 0;
136 		/*
137 		 * The following assignments assume an overlay with the
138 		 * zeroed icmp_void field.
139 		 */
140 		if (type == ICMP_PARAMPROB) {
141 			icp->icmp_pptr = code;
142 			code = 0;
143 		} else if (type == ICMP_UNREACH &&
144 		    code == ICMP_UNREACH_NEEDFRAG && destifp)
145 			icp->icmp_nextmtu = htons(destifp->if_mtu);
146 	}
147 
148 	icp->icmp_code = code;
149 	bcopy((caddr_t)oip, (caddr_t)&icp->icmp_ip, icmplen);
150 	nip = &icp->icmp_ip;
151 	nip->ip_len = htons((u_int16_t)(nip->ip_len + oiplen));
152 
153 	/*
154 	 * Now, copy old ip header (without options)
155 	 * in front of icmp message.
156 	 */
157 	if (m->m_data - sizeof(struct ip) < m->m_pktdat)
158 		panic("icmp len");
159 	m->m_data -= sizeof(struct ip);
160 	m->m_len += sizeof(struct ip);
161 	m->m_pkthdr.len = m->m_len;
162 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
163 	nip = mtod(m, struct ip *);
164 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
165 	nip->ip_len = m->m_len;
166 	nip->ip_hl = sizeof(struct ip) >> 2;
167 	nip->ip_p = IPPROTO_ICMP;
168 	nip->ip_tos = 0;
169 	icmp_reflect(m);
170 
171 freeit:
172 	m_freem(n);
173 }
174 
175 static struct sockaddr_in icmpsrc = { sizeof (struct sockaddr_in), AF_INET };
176 static struct sockaddr_in icmpdst = { sizeof (struct sockaddr_in), AF_INET };
177 static struct sockaddr_in icmpgw = { sizeof (struct sockaddr_in), AF_INET };
178 struct sockaddr_in icmpmask = { 8, 0 };
179 
180 /*
181  * Process a received ICMP message.
182  */
183 void
184 #if __STDC__
185 icmp_input(struct mbuf *m, ...)
186 #else
187 icmp_input(m, va_alist)
188 	struct mbuf *m;
189 	va_dcl
190 #endif
191 {
192 	register struct icmp *icp;
193 	register struct ip *ip = mtod(m, struct ip *);
194 	int icmplen = ip->ip_len;
195 	register int i;
196 	struct in_ifaddr *ia;
197 	void *(*ctlfunc) __P((int, struct sockaddr *, void *));
198 	int code;
199 	extern u_char ip_protox[];
200 	int hlen;
201 	va_list ap;
202 
203 	va_start(ap, m);
204 	hlen = va_arg(ap, int);
205 	va_end(ap);
206 
207 	/*
208 	 * Locate icmp structure in mbuf, and check
209 	 * that not corrupted and of at least minimum length.
210 	 */
211 #ifdef ICMPPRINTFS
212 	if (icmpprintfs)
213 		printf("icmp_input from %x to %x, len %d\n",
214 		    ntohl(ip->ip_src.s_addr), ntohl(ip->ip_dst.s_addr),
215 		    icmplen);
216 #endif
217 	if (icmplen < ICMP_MINLEN) {
218 		icmpstat.icps_tooshort++;
219 		goto freeit;
220 	}
221 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
222 	if (m->m_len < i && (m = m_pullup(m, i)) == 0) {
223 		icmpstat.icps_tooshort++;
224 		return;
225 	}
226 	ip = mtod(m, struct ip *);
227 	m->m_len -= hlen;
228 	m->m_data += hlen;
229 	icp = mtod(m, struct icmp *);
230 	if (in_cksum(m, icmplen)) {
231 		icmpstat.icps_checksum++;
232 		goto freeit;
233 	}
234 	m->m_len += hlen;
235 	m->m_data -= hlen;
236 
237 #ifdef ICMPPRINTFS
238 	/*
239 	 * Message type specific processing.
240 	 */
241 	if (icmpprintfs)
242 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
243 		    icp->icmp_code);
244 #endif
245 	if (icp->icmp_type > ICMP_MAXTYPE)
246 		goto raw;
247 	icmpstat.icps_inhist[icp->icmp_type]++;
248 	code = icp->icmp_code;
249 	switch (icp->icmp_type) {
250 
251 	case ICMP_UNREACH:
252 		switch (code) {
253 			case ICMP_UNREACH_NET:
254 			case ICMP_UNREACH_HOST:
255 			case ICMP_UNREACH_PROTOCOL:
256 			case ICMP_UNREACH_PORT:
257 			case ICMP_UNREACH_SRCFAIL:
258 				code += PRC_UNREACH_NET;
259 				break;
260 
261 			case ICMP_UNREACH_NEEDFRAG:
262 				code = PRC_MSGSIZE;
263 				break;
264 
265 			case ICMP_UNREACH_NET_UNKNOWN:
266 			case ICMP_UNREACH_NET_PROHIB:
267 			case ICMP_UNREACH_TOSNET:
268 				code = PRC_UNREACH_NET;
269 				break;
270 
271 			case ICMP_UNREACH_HOST_UNKNOWN:
272 			case ICMP_UNREACH_ISOLATED:
273 			case ICMP_UNREACH_HOST_PROHIB:
274 			case ICMP_UNREACH_TOSHOST:
275 				code = PRC_UNREACH_HOST;
276 				break;
277 
278 			default:
279 				goto badcode;
280 		}
281 		goto deliver;
282 
283 	case ICMP_TIMXCEED:
284 		if (code > 1)
285 			goto badcode;
286 		code += PRC_TIMXCEED_INTRANS;
287 		goto deliver;
288 
289 	case ICMP_PARAMPROB:
290 		if (code > 1)
291 			goto badcode;
292 		code = PRC_PARAMPROB;
293 		goto deliver;
294 
295 	case ICMP_SOURCEQUENCH:
296 		if (code)
297 			goto badcode;
298 		code = PRC_QUENCH;
299 		goto deliver;
300 
301 	deliver:
302 		/*
303 		 * Problem with datagram; advise higher level routines.
304 		 */
305 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
306 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
307 			icmpstat.icps_badlen++;
308 			goto freeit;
309 		}
310 		if (IN_MULTICAST(icp->icmp_ip.ip_dst.s_addr))
311 			goto badcode;
312 		NTOHS(icp->icmp_ip.ip_len);
313 #ifdef ICMPPRINTFS
314 		if (icmpprintfs)
315 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
316 #endif
317 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
318 		if (code == PRC_MSGSIZE && ip_mtudisc)
319 			icmp_mtudisc(icp);
320 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
321 		if (ctlfunc)
322 			(*ctlfunc)(code, sintosa(&icmpsrc), &icp->icmp_ip);
323 		break;
324 
325 	badcode:
326 		icmpstat.icps_badcode++;
327 		break;
328 
329 	case ICMP_ECHO:
330 		icp->icmp_type = ICMP_ECHOREPLY;
331 		goto reflect;
332 
333 	case ICMP_TSTAMP:
334 		if (icmplen < ICMP_TSLEN) {
335 			icmpstat.icps_badlen++;
336 			break;
337 		}
338 		icp->icmp_type = ICMP_TSTAMPREPLY;
339 		icp->icmp_rtime = iptime();
340 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
341 		goto reflect;
342 
343 	case ICMP_MASKREQ:
344 		if (icmpmaskrepl == 0)
345 			break;
346 		/*
347 		 * We are not able to respond with all ones broadcast
348 		 * unless we receive it over a point-to-point interface.
349 		 */
350 		if (icmplen < ICMP_MASKLEN) {
351 			icmpstat.icps_badlen++;
352 			break;
353 		}
354 		if (ip->ip_dst.s_addr == INADDR_BROADCAST ||
355 		    in_nullhost(ip->ip_dst))
356 			icmpdst.sin_addr = ip->ip_src;
357 		else
358 			icmpdst.sin_addr = ip->ip_dst;
359 		ia = ifatoia(ifaof_ifpforaddr(sintosa(&icmpdst),
360 		    m->m_pkthdr.rcvif));
361 		if (ia == 0)
362 			break;
363 		icp->icmp_type = ICMP_MASKREPLY;
364 		icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
365 		if (in_nullhost(ip->ip_src)) {
366 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
367 				ip->ip_src = ia->ia_broadaddr.sin_addr;
368 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
369 				ip->ip_src = ia->ia_dstaddr.sin_addr;
370 		}
371 reflect:
372 		ip->ip_len += hlen;	/* since ip_input deducts this */
373 		icmpstat.icps_reflect++;
374 		icmpstat.icps_outhist[icp->icmp_type]++;
375 		icmp_reflect(m);
376 		return;
377 
378 	case ICMP_REDIRECT:
379 		if (code > 3)
380 			goto badcode;
381 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
382 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
383 			icmpstat.icps_badlen++;
384 			break;
385 		}
386 		/*
387 		 * Short circuit routing redirects to force
388 		 * immediate change in the kernel's routing
389 		 * tables.  The message is also handed to anyone
390 		 * listening on a raw socket (e.g. the routing
391 		 * daemon for use in updating its tables).
392 		 */
393 		icmpgw.sin_addr = ip->ip_src;
394 		icmpdst.sin_addr = icp->icmp_gwaddr;
395 #ifdef	ICMPPRINTFS
396 		if (icmpprintfs)
397 			printf("redirect dst %x to %x\n", icp->icmp_ip.ip_dst,
398 			    icp->icmp_gwaddr);
399 #endif
400 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
401 		rtredirect(sintosa(&icmpsrc), sintosa(&icmpdst),
402 		    (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
403 		    sintosa(&icmpgw), (struct rtentry **)0);
404 		pfctlinput(PRC_REDIRECT_HOST, sintosa(&icmpsrc));
405 		break;
406 
407 	/*
408 	 * No kernel processing for the following;
409 	 * just fall through to send to raw listener.
410 	 */
411 	case ICMP_ECHOREPLY:
412 	case ICMP_ROUTERADVERT:
413 	case ICMP_ROUTERSOLICIT:
414 	case ICMP_TSTAMPREPLY:
415 	case ICMP_IREQREPLY:
416 	case ICMP_MASKREPLY:
417 	default:
418 		break;
419 	}
420 
421 raw:
422 	rip_input(m);
423 	return;
424 
425 freeit:
426 	m_freem(m);
427 }
428 
429 /*
430  * Reflect the ip packet back to the source
431  */
432 void
433 icmp_reflect(m)
434 	struct mbuf *m;
435 {
436 	register struct ip *ip = mtod(m, struct ip *);
437 	register struct in_ifaddr *ia;
438 	struct in_addr t;
439 	struct mbuf *opts = 0;
440 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
441 
442 	if (!in_canforward(ip->ip_src) &&
443 	    ((ip->ip_src.s_addr & IN_CLASSA_NET) !=
444 	     htonl(IN_LOOPBACKNET << IN_CLASSA_NSHIFT))) {
445 		m_freem(m);	/* Bad return address */
446 		goto done;	/* ip_output() will check for broadcast */
447 	}
448 	t = ip->ip_dst;
449 	ip->ip_dst = ip->ip_src;
450 	/*
451 	 * If the incoming packet was addressed directly to us,
452 	 * use dst as the src for the reply.  Otherwise (broadcast
453 	 * or anonymous), use the address which corresponds
454 	 * to the incoming interface.
455 	 */
456 	for (ia = in_ifaddr.tqh_first; ia; ia = ia->ia_list.tqe_next) {
457 		if (in_hosteq(t, ia->ia_addr.sin_addr))
458 			break;
459 		if ((ia->ia_ifp->if_flags & IFF_BROADCAST) &&
460 		    in_hosteq(t, ia->ia_broadaddr.sin_addr))
461 			break;
462 	}
463 	icmpdst.sin_addr = t;
464 	if (ia == (struct in_ifaddr *)0)
465 		ia = ifatoia(ifaof_ifpforaddr(sintosa(&icmpdst),
466 		    m->m_pkthdr.rcvif));
467 	/*
468 	 * The following happens if the packet was not addressed to us,
469 	 * and was received on an interface with no IP address.
470 	 */
471 	if (ia == (struct in_ifaddr *)0)
472 		ia = in_ifaddr.tqh_first;
473 	t = ia->ia_addr.sin_addr;
474 	ip->ip_src = t;
475 	ip->ip_ttl = MAXTTL;
476 
477 	if (optlen > 0) {
478 		register u_char *cp;
479 		int opt, cnt;
480 		u_int len;
481 
482 		/*
483 		 * Retrieve any source routing from the incoming packet;
484 		 * add on any record-route or timestamp options.
485 		 */
486 		cp = (u_char *) (ip + 1);
487 		if ((opts = ip_srcroute()) == 0 &&
488 		    (opts = m_gethdr(M_DONTWAIT, MT_HEADER))) {
489 			opts->m_len = sizeof(struct in_addr);
490 			*mtod(opts, struct in_addr *) = zeroin_addr;
491 		}
492 		if (opts) {
493 #ifdef ICMPPRINTFS
494 		    if (icmpprintfs)
495 			    printf("icmp_reflect optlen %d rt %d => ",
496 				optlen, opts->m_len);
497 #endif
498 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
499 			    opt = cp[IPOPT_OPTVAL];
500 			    if (opt == IPOPT_EOL)
501 				    break;
502 			    if (opt == IPOPT_NOP)
503 				    len = 1;
504 			    else {
505 				    len = cp[IPOPT_OLEN];
506 				    if (len <= 0 || len > cnt)
507 					    break;
508 			    }
509 			    /*
510 			     * Should check for overflow, but it "can't happen"
511 			     */
512 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
513 				opt == IPOPT_SECURITY) {
514 				    bcopy((caddr_t)cp,
515 					mtod(opts, caddr_t) + opts->m_len, len);
516 				    opts->m_len += len;
517 			    }
518 		    }
519 		    /* Terminate & pad, if necessary */
520 		    if ((cnt = opts->m_len % 4) != 0) {
521 			    for (; cnt < 4; cnt++) {
522 				    *(mtod(opts, caddr_t) + opts->m_len) =
523 					IPOPT_EOL;
524 				    opts->m_len++;
525 			    }
526 		    }
527 #ifdef ICMPPRINTFS
528 		    if (icmpprintfs)
529 			    printf("%d\n", opts->m_len);
530 #endif
531 		}
532 		/*
533 		 * Now strip out original options by copying rest of first
534 		 * mbuf's data back, and adjust the IP length.
535 		 */
536 		ip->ip_len -= optlen;
537 		ip->ip_hl = sizeof(struct ip) >> 2;
538 		m->m_len -= optlen;
539 		if (m->m_flags & M_PKTHDR)
540 			m->m_pkthdr.len -= optlen;
541 		optlen += sizeof(struct ip);
542 		bcopy((caddr_t)ip + optlen, (caddr_t)(ip + 1),
543 			 (unsigned)(m->m_len - sizeof(struct ip)));
544 	}
545 	m->m_flags &= ~(M_BCAST|M_MCAST);
546 	icmp_send(m, opts);
547 done:
548 	if (opts)
549 		(void)m_free(opts);
550 }
551 
552 /*
553  * Send an icmp packet back to the ip level,
554  * after supplying a checksum.
555  */
556 void
557 icmp_send(m, opts)
558 	register struct mbuf *m;
559 	struct mbuf *opts;
560 {
561 	register struct ip *ip = mtod(m, struct ip *);
562 	register int hlen;
563 	register struct icmp *icp;
564 
565 	hlen = ip->ip_hl << 2;
566 	m->m_data += hlen;
567 	m->m_len -= hlen;
568 	icp = mtod(m, struct icmp *);
569 	icp->icmp_cksum = 0;
570 	icp->icmp_cksum = in_cksum(m, ip->ip_len - hlen);
571 	m->m_data -= hlen;
572 	m->m_len += hlen;
573 #ifdef ICMPPRINTFS
574 	if (icmpprintfs)
575 		printf("icmp_send dst %x src %x\n", ip->ip_dst, ip->ip_src);
576 #endif
577 	(void) ip_output(m, opts, NULL, 0, NULL);
578 }
579 
580 n_time
581 iptime()
582 {
583 	struct timeval atv;
584 	u_long t;
585 
586 	microtime(&atv);
587 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
588 	return (htonl(t));
589 }
590 
591 int
592 icmp_sysctl(name, namelen, oldp, oldlenp, newp, newlen)
593 	int *name;
594 	u_int namelen;
595 	void *oldp;
596 	size_t *oldlenp;
597 	void *newp;
598 	size_t newlen;
599 {
600 
601 	/* All sysctl names at this level are terminal. */
602 	if (namelen != 1)
603 		return (ENOTDIR);
604 
605 	switch (name[0]) {
606 	case ICMPCTL_MASKREPL:
607 		return (sysctl_int(oldp, oldlenp, newp, newlen, &icmpmaskrepl));
608 	default:
609 		return (ENOPROTOOPT);
610 	}
611 	/* NOTREACHED */
612 }
613 
614 static void
615 icmp_mtudisc(icp)
616 	struct icmp *icp;
617 {
618 	struct rtentry *rt;
619 	struct sockaddr *dst = sintosa(&icmpsrc);
620 	u_long mtu = ntohs(icp->icmp_nextmtu);  /* Why a long?  IPv6 */
621 
622 	/* Table of common MTUs: */
623 
624 	static u_long mtu_table[] = {65535, 65280, 32000, 17914, 9180, 8166,
625 				     4352, 2002, 1492, 1006, 508, 296, 68, 0};
626 
627 	rt = rtalloc1(dst, 1);
628 	if (rt == 0)
629 		return;
630 
631 	/* If we didn't get a host route, allocate one */
632 
633 	if ((rt->rt_flags & RTF_HOST) == 0) {
634 		struct rtentry *nrt;
635 		int error;
636 
637 		error = rtrequest((int) RTM_ADD, dst,
638 		    (struct sockaddr *) rt->rt_gateway,
639 		    (struct sockaddr *) 0,
640 		    RTF_GATEWAY | RTF_HOST | RTF_DYNAMIC, &nrt);
641 		if (error) {
642 			rtfree(rt);
643 			rtfree(nrt);
644 			return;
645 		}
646 		nrt->rt_rmx = rt->rt_rmx;
647 		rtfree(rt);
648 		rt = nrt;
649 	}
650 
651 	if (mtu == 0) {
652 		int i = 0;
653 
654 		mtu = icp->icmp_ip.ip_len; /* NTOHS happened in deliver: */
655 		/* Some 4.2BSD-based routers incorrectly adjust the ip_len */
656 		if (mtu > rt->rt_rmx.rmx_mtu && rt->rt_rmx.rmx_mtu != 0)
657 			mtu -= (icp->icmp_ip.ip_hl << 2);
658 
659 		/* If we still can't guess a value, try the route */
660 
661 		if (mtu == 0) {
662 			mtu = rt->rt_rmx.rmx_mtu;
663 
664 			/* If no route mtu, default to the interface mtu */
665 
666 			if (mtu == 0)
667 				mtu = rt->rt_ifp->if_mtu;
668 		}
669 
670 		for (i = 0; i < sizeof(mtu_table) / sizeof(mtu_table[0]); i++)
671 			if (mtu > mtu_table[i]) {
672 				mtu = mtu_table[i];
673 				break;
674 			}
675 	}
676 
677 	if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0) {
678 		if (mtu < 296 || mtu > rt->rt_ifp->if_mtu)
679 			rt->rt_rmx.rmx_locks |= RTV_MTU;
680 		else if (rt->rt_rmx.rmx_mtu > mtu ||
681 			 rt->rt_rmx.rmx_mtu == 0)
682 			rt->rt_rmx.rmx_mtu = mtu;
683 	}
684 
685 	if (rt)
686 		rtfree(rt);
687 }
688