xref: /freebsd/sys/netinet/ip_icmp.c (revision 7562eaabc01a48e6b11d5b558c41e3b92dae5c2d)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_icmp.c	8.2 (Berkeley) 1/4/94
30  * $FreeBSD$
31  */
32 
33 #include "opt_ipsec.h"
34 #include "opt_mac.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/mac.h>
39 #include <sys/mbuf.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/time.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 
46 #include <net/if.h>
47 #include <net/if_types.h>
48 #include <net/route.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_pcb.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/in_var.h>
54 #include <netinet/ip.h>
55 #include <netinet/ip_icmp.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/tcp.h>
58 #include <netinet/tcp_var.h>
59 #include <netinet/tcpip.h>
60 #include <netinet/icmp_var.h>
61 
62 #ifdef IPSEC
63 #include <netinet6/ipsec.h>
64 #include <netkey/key.h>
65 #endif
66 
67 #ifdef FAST_IPSEC
68 #include <netipsec/ipsec.h>
69 #include <netipsec/key.h>
70 #define	IPSEC
71 #endif
72 
73 #include <machine/in_cksum.h>
74 
75 /*
76  * ICMP routines: error generation, receive packet processing, and
77  * routines to turnaround packets back to the originator, and
78  * host table maintenance routines.
79  */
80 
81 struct	icmpstat icmpstat;
82 SYSCTL_STRUCT(_net_inet_icmp, ICMPCTL_STATS, stats, CTLFLAG_RW,
83 	&icmpstat, icmpstat, "");
84 
85 static int	icmpmaskrepl = 0;
86 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_RW,
87 	&icmpmaskrepl, 0, "Reply to ICMP Address Mask Request packets.");
88 
89 static u_int	icmpmaskfake = 0;
90 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_RW,
91 	&icmpmaskfake, 0, "Fake reply to ICMP Address Mask Request packets.");
92 
93 static int	drop_redirect = 0;
94 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_RW,
95 	&drop_redirect, 0, "");
96 
97 static int	log_redirect = 0;
98 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_RW,
99 	&log_redirect, 0, "");
100 
101 static int      icmplim = 200;
102 SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_RW,
103 	&icmplim, 0, "");
104 
105 static int	icmplim_output = 1;
106 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_RW,
107 	&icmplim_output, 0, "");
108 
109 static char	reply_src[IFNAMSIZ];
110 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_RW,
111 	&reply_src, IFNAMSIZ, "icmp reply source for non-local packets.");
112 
113 /*
114  * ICMP broadcast echo sysctl
115  */
116 
117 static int	icmpbmcastecho = 0;
118 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_RW,
119 	&icmpbmcastecho, 0, "");
120 
121 
122 #ifdef ICMPPRINTFS
123 int	icmpprintfs = 0;
124 #endif
125 
126 static void	icmp_reflect(struct mbuf *);
127 static void	icmp_send(struct mbuf *, struct mbuf *);
128 static int	ip_next_mtu(int, int);
129 
130 extern	struct protosw inetsw[];
131 
132 /*
133  * Generate an error packet of type error
134  * in response to bad packet ip.
135  */
136 void
137 icmp_error(n, type, code, dest, destifp)
138 	struct mbuf *n;
139 	int type, code;
140 	n_long dest;
141 	struct ifnet *destifp;
142 {
143 	register struct ip *oip = mtod(n, struct ip *), *nip;
144 	register unsigned oiplen = oip->ip_hl << 2;
145 	register struct icmp *icp;
146 	register struct mbuf *m;
147 	unsigned icmplen;
148 
149 #ifdef ICMPPRINTFS
150 	if (icmpprintfs)
151 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
152 #endif
153 	if (type != ICMP_REDIRECT)
154 		icmpstat.icps_error++;
155 	/*
156 	 * Don't send error if the original packet was encrypted.
157 	 * Don't send error if not the first fragment of message.
158 	 * Don't error if the old packet protocol was ICMP
159 	 * error message, only known informational types.
160 	 */
161 	if (n->m_flags & M_DECRYPTED)
162 		goto freeit;
163 	if (oip->ip_off &~ (IP_MF|IP_DF))
164 		goto freeit;
165 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
166 	  n->m_len >= oiplen + ICMP_MINLEN &&
167 	  !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiplen))->icmp_type)) {
168 		icmpstat.icps_oldicmp++;
169 		goto freeit;
170 	}
171 	/* Don't send error in response to a multicast or broadcast packet */
172 	if (n->m_flags & (M_BCAST|M_MCAST))
173 		goto freeit;
174 	/*
175 	 * First, formulate icmp message
176 	 */
177 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
178 	if (m == NULL)
179 		goto freeit;
180 #ifdef MAC
181 	mac_create_mbuf_netlayer(n, m);
182 #endif
183 	icmplen = min(oiplen + 8, oip->ip_len);
184 	if (icmplen < sizeof(struct ip))
185 		panic("icmp_error: bad length");
186 	m->m_len = icmplen + ICMP_MINLEN;
187 	MH_ALIGN(m, m->m_len);
188 	icp = mtod(m, struct icmp *);
189 	if ((u_int)type > ICMP_MAXTYPE)
190 		panic("icmp_error");
191 	icmpstat.icps_outhist[type]++;
192 	icp->icmp_type = type;
193 	if (type == ICMP_REDIRECT)
194 		icp->icmp_gwaddr.s_addr = dest;
195 	else {
196 		icp->icmp_void = 0;
197 		/*
198 		 * The following assignments assume an overlay with the
199 		 * zeroed icmp_void field.
200 		 */
201 		if (type == ICMP_PARAMPROB) {
202 			icp->icmp_pptr = code;
203 			code = 0;
204 		} else if (type == ICMP_UNREACH &&
205 			code == ICMP_UNREACH_NEEDFRAG && destifp) {
206 			icp->icmp_nextmtu = htons(destifp->if_mtu);
207 		}
208 	}
209 
210 	icp->icmp_code = code;
211 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
212 	nip = &icp->icmp_ip;
213 
214 	/*
215 	 * Convert fields to network representation.
216 	 */
217 	nip->ip_len = htons(nip->ip_len);
218 	nip->ip_off = htons(nip->ip_off);
219 
220 	/*
221 	 * Now, copy old ip header (without options)
222 	 * in front of icmp message.
223 	 */
224 	if (m->m_data - sizeof(struct ip) < m->m_pktdat)
225 		panic("icmp len");
226 	/*
227 	 * If the original mbuf was meant to bypass the firewall, the error
228 	 * reply should bypass as well.
229 	 */
230 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
231 	m->m_data -= sizeof(struct ip);
232 	m->m_len += sizeof(struct ip);
233 	m->m_pkthdr.len = m->m_len;
234 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
235 	nip = mtod(m, struct ip *);
236 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
237 	nip->ip_len = m->m_len;
238 	nip->ip_v = IPVERSION;
239 	nip->ip_hl = 5;
240 	nip->ip_p = IPPROTO_ICMP;
241 	nip->ip_tos = 0;
242 	icmp_reflect(m);
243 
244 freeit:
245 	m_freem(n);
246 }
247 
248 /*
249  * Process a received ICMP message.
250  */
251 void
252 icmp_input(m, off)
253 	struct mbuf *m;
254 	int off;
255 {
256 	struct icmp *icp;
257 	struct in_ifaddr *ia;
258 	struct ip *ip = mtod(m, struct ip *);
259 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
260 	int hlen = off;
261 	int icmplen = ip->ip_len;
262 	int i, code;
263 	void (*ctlfunc)(int, struct sockaddr *, void *);
264 
265 	/*
266 	 * Locate icmp structure in mbuf, and check
267 	 * that not corrupted and of at least minimum length.
268 	 */
269 #ifdef ICMPPRINTFS
270 	if (icmpprintfs) {
271 		char buf[4 * sizeof "123"];
272 		strcpy(buf, inet_ntoa(ip->ip_src));
273 		printf("icmp_input from %s to %s, len %d\n",
274 		       buf, inet_ntoa(ip->ip_dst), icmplen);
275 	}
276 #endif
277 	if (icmplen < ICMP_MINLEN) {
278 		icmpstat.icps_tooshort++;
279 		goto freeit;
280 	}
281 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
282 	if (m->m_len < i && (m = m_pullup(m, i)) == 0)  {
283 		icmpstat.icps_tooshort++;
284 		return;
285 	}
286 	ip = mtod(m, struct ip *);
287 	m->m_len -= hlen;
288 	m->m_data += hlen;
289 	icp = mtod(m, struct icmp *);
290 	if (in_cksum(m, icmplen)) {
291 		icmpstat.icps_checksum++;
292 		goto freeit;
293 	}
294 	m->m_len += hlen;
295 	m->m_data -= hlen;
296 
297 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
298 		/*
299 		 * Deliver very specific ICMP type only.
300 		 */
301 		switch (icp->icmp_type) {
302 		case ICMP_UNREACH:
303 		case ICMP_TIMXCEED:
304 			break;
305 		default:
306 			goto freeit;
307 		}
308 	}
309 
310 #ifdef ICMPPRINTFS
311 	if (icmpprintfs)
312 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
313 		    icp->icmp_code);
314 #endif
315 
316 	/*
317 	 * Message type specific processing.
318 	 */
319 	if (icp->icmp_type > ICMP_MAXTYPE)
320 		goto raw;
321 
322 	/* Initialize */
323 	bzero(&icmpsrc, sizeof(icmpsrc));
324 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
325 	icmpsrc.sin_family = AF_INET;
326 	bzero(&icmpdst, sizeof(icmpdst));
327 	icmpdst.sin_len = sizeof(struct sockaddr_in);
328 	icmpdst.sin_family = AF_INET;
329 	bzero(&icmpgw, sizeof(icmpgw));
330 	icmpgw.sin_len = sizeof(struct sockaddr_in);
331 	icmpgw.sin_family = AF_INET;
332 
333 	icmpstat.icps_inhist[icp->icmp_type]++;
334 	code = icp->icmp_code;
335 	switch (icp->icmp_type) {
336 
337 	case ICMP_UNREACH:
338 		switch (code) {
339 			case ICMP_UNREACH_NET:
340 			case ICMP_UNREACH_HOST:
341 			case ICMP_UNREACH_SRCFAIL:
342 			case ICMP_UNREACH_NET_UNKNOWN:
343 			case ICMP_UNREACH_HOST_UNKNOWN:
344 			case ICMP_UNREACH_ISOLATED:
345 			case ICMP_UNREACH_TOSNET:
346 			case ICMP_UNREACH_TOSHOST:
347 			case ICMP_UNREACH_HOST_PRECEDENCE:
348 			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
349 				code = PRC_UNREACH_NET;
350 				break;
351 
352 			case ICMP_UNREACH_NEEDFRAG:
353 				code = PRC_MSGSIZE;
354 				break;
355 
356 			/*
357 			 * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
358 			 * Treat subcodes 2,3 as immediate RST
359 			 */
360 			case ICMP_UNREACH_PROTOCOL:
361 			case ICMP_UNREACH_PORT:
362 				code = PRC_UNREACH_PORT;
363 				break;
364 
365 			case ICMP_UNREACH_NET_PROHIB:
366 			case ICMP_UNREACH_HOST_PROHIB:
367 			case ICMP_UNREACH_FILTER_PROHIB:
368 				code = PRC_UNREACH_ADMIN_PROHIB;
369 				break;
370 
371 			default:
372 				goto badcode;
373 		}
374 		goto deliver;
375 
376 	case ICMP_TIMXCEED:
377 		if (code > 1)
378 			goto badcode;
379 		code += PRC_TIMXCEED_INTRANS;
380 		goto deliver;
381 
382 	case ICMP_PARAMPROB:
383 		if (code > 1)
384 			goto badcode;
385 		code = PRC_PARAMPROB;
386 		goto deliver;
387 
388 	case ICMP_SOURCEQUENCH:
389 		if (code)
390 			goto badcode;
391 		code = PRC_QUENCH;
392 	deliver:
393 		/*
394 		 * Problem with datagram; advise higher level routines.
395 		 */
396 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
397 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
398 			icmpstat.icps_badlen++;
399 			goto freeit;
400 		}
401 		icp->icmp_ip.ip_len = ntohs(icp->icmp_ip.ip_len);
402 		/* Discard ICMP's in response to multicast packets */
403 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
404 			goto badcode;
405 #ifdef ICMPPRINTFS
406 		if (icmpprintfs)
407 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
408 #endif
409 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
410 
411 		/*
412 		 * MTU discovery:
413 		 * If we got a needfrag and there is a host route to the
414 		 * original destination, and the MTU is not locked, then
415 		 * set the MTU in the route to the suggested new value
416 		 * (if given) and then notify as usual.  The ULPs will
417 		 * notice that the MTU has changed and adapt accordingly.
418 		 * If no new MTU was suggested, then we guess a new one
419 		 * less than the current value.  If the new MTU is
420 		 * unreasonably small (defined by sysctl tcp_minmss), then
421 		 * we don't update the MTU value.
422 		 *
423 		 * XXX: All this should be done in tcp_mtudisc() because
424 		 * the way we do it now, everyone can send us bogus ICMP
425 		 * MSGSIZE packets for any destination. By doing this far
426 		 * higher in the chain we have a matching tcp connection.
427 		 * Thus spoofing is much harder. However there is no easy
428 		 * non-hackish way to pass the new MTU up to tcp_mtudisc().
429 		 * Also see next XXX regarding IPv4 AH TCP.
430 		 */
431 		if (code == PRC_MSGSIZE) {
432 			int mtu;
433 			struct in_conninfo inc;
434 
435 			bzero(&inc, sizeof(inc));
436 			inc.inc_flags = 0; /* IPv4 */
437 			inc.inc_faddr = icmpsrc.sin_addr;
438 
439 			mtu = ntohs(icp->icmp_nextmtu);
440 			if (!mtu)
441 				mtu = ip_next_mtu(mtu, 1);
442 
443 			if (mtu >= max(296, (tcp_minmss +
444 					sizeof(struct tcpiphdr))))
445 				tcp_hc_updatemtu(&inc, mtu);
446 
447 #ifdef DEBUG_MTUDISC
448 			printf("MTU for %s reduced to %d\n",
449 				inet_ntoa(icmpsrc.sin_addr), mtu);
450 #endif
451 		}
452 
453 		/*
454 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
455 		 * notification to TCP layer.
456 		 */
457 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
458 		if (ctlfunc)
459 			(*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
460 				   (void *)&icp->icmp_ip);
461 		break;
462 
463 	badcode:
464 		icmpstat.icps_badcode++;
465 		break;
466 
467 	case ICMP_ECHO:
468 		if (!icmpbmcastecho
469 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
470 			icmpstat.icps_bmcastecho++;
471 			break;
472 		}
473 		icp->icmp_type = ICMP_ECHOREPLY;
474 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
475 			goto freeit;
476 		else
477 			goto reflect;
478 
479 	case ICMP_TSTAMP:
480 		if (!icmpbmcastecho
481 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
482 			icmpstat.icps_bmcasttstamp++;
483 			break;
484 		}
485 		if (icmplen < ICMP_TSLEN) {
486 			icmpstat.icps_badlen++;
487 			break;
488 		}
489 		icp->icmp_type = ICMP_TSTAMPREPLY;
490 		icp->icmp_rtime = iptime();
491 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
492 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
493 			goto freeit;
494 		else
495 			goto reflect;
496 
497 	case ICMP_MASKREQ:
498 		if (icmpmaskrepl == 0)
499 			break;
500 		/*
501 		 * We are not able to respond with all ones broadcast
502 		 * unless we receive it over a point-to-point interface.
503 		 */
504 		if (icmplen < ICMP_MASKLEN)
505 			break;
506 		switch (ip->ip_dst.s_addr) {
507 
508 		case INADDR_BROADCAST:
509 		case INADDR_ANY:
510 			icmpdst.sin_addr = ip->ip_src;
511 			break;
512 
513 		default:
514 			icmpdst.sin_addr = ip->ip_dst;
515 		}
516 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
517 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
518 		if (ia == 0)
519 			break;
520 		if (ia->ia_ifp == 0)
521 			break;
522 		icp->icmp_type = ICMP_MASKREPLY;
523 		if (icmpmaskfake == 0)
524 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
525 		else
526 			icp->icmp_mask = icmpmaskfake;
527 		if (ip->ip_src.s_addr == 0) {
528 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
529 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
530 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
531 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
532 		}
533 reflect:
534 		ip->ip_len += hlen;	/* since ip_input deducts this */
535 		icmpstat.icps_reflect++;
536 		icmpstat.icps_outhist[icp->icmp_type]++;
537 		icmp_reflect(m);
538 		return;
539 
540 	case ICMP_REDIRECT:
541 		if (log_redirect) {
542 			u_long src, dst, gw;
543 
544 			src = ntohl(ip->ip_src.s_addr);
545 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
546 			gw = ntohl(icp->icmp_gwaddr.s_addr);
547 			printf("icmp redirect from %d.%d.%d.%d: "
548 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
549 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
550 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
551 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
552 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
553 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
554 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
555 		}
556 		/*
557 		 * RFC1812 says we must ignore ICMP redirects if we
558 		 * are acting as router.
559 		 */
560 		if (drop_redirect || ipforwarding)
561 			break;
562 		if (code > 3)
563 			goto badcode;
564 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
565 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
566 			icmpstat.icps_badlen++;
567 			break;
568 		}
569 		/*
570 		 * Short circuit routing redirects to force
571 		 * immediate change in the kernel's routing
572 		 * tables.  The message is also handed to anyone
573 		 * listening on a raw socket (e.g. the routing
574 		 * daemon for use in updating its tables).
575 		 */
576 		icmpgw.sin_addr = ip->ip_src;
577 		icmpdst.sin_addr = icp->icmp_gwaddr;
578 #ifdef	ICMPPRINTFS
579 		if (icmpprintfs) {
580 			char buf[4 * sizeof "123"];
581 			strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
582 
583 			printf("redirect dst %s to %s\n",
584 			       buf, inet_ntoa(icp->icmp_gwaddr));
585 		}
586 #endif
587 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
588 		rtredirect((struct sockaddr *)&icmpsrc,
589 		  (struct sockaddr *)&icmpdst,
590 		  (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
591 		  (struct sockaddr *)&icmpgw);
592 		pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
593 #ifdef IPSEC
594 		key_sa_routechange((struct sockaddr *)&icmpsrc);
595 #endif
596 		break;
597 
598 	/*
599 	 * No kernel processing for the following;
600 	 * just fall through to send to raw listener.
601 	 */
602 	case ICMP_ECHOREPLY:
603 	case ICMP_ROUTERADVERT:
604 	case ICMP_ROUTERSOLICIT:
605 	case ICMP_TSTAMPREPLY:
606 	case ICMP_IREQREPLY:
607 	case ICMP_MASKREPLY:
608 	default:
609 		break;
610 	}
611 
612 raw:
613 	rip_input(m, off);
614 	return;
615 
616 freeit:
617 	m_freem(m);
618 }
619 
620 /*
621  * Reflect the ip packet back to the source
622  */
623 static void
624 icmp_reflect(m)
625 	struct mbuf *m;
626 {
627 	struct ip *ip = mtod(m, struct ip *);
628 	struct ifaddr *ifa;
629 	struct ifnet *ifn;
630 	struct in_ifaddr *ia;
631 	struct in_addr t;
632 	struct mbuf *opts = 0;
633 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
634 
635 	if (!in_canforward(ip->ip_src) &&
636 	    ((ntohl(ip->ip_src.s_addr) & IN_CLASSA_NET) !=
637 	     (IN_LOOPBACKNET << IN_CLASSA_NSHIFT))) {
638 		m_freem(m);	/* Bad return address */
639 		icmpstat.icps_badaddr++;
640 		goto done;	/* Ip_output() will check for broadcast */
641 	}
642 	t = ip->ip_dst;
643 	ip->ip_dst = ip->ip_src;
644 
645 	/*
646 	 * Source selection for ICMP replies:
647 	 *
648 	 * If the incoming packet was addressed directly to one of our
649 	 * own addresses, use dst as the src for the reply.
650 	 */
651 	LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash)
652 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr)
653 			goto match;
654 	/*
655 	 * If the incoming packet was addressed to one of our broadcast
656 	 * addresses, use the first non-broadcast address which corresponds
657 	 * to the incoming interface.
658 	 */
659 	if (m->m_pkthdr.rcvif != NULL &&
660 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
661 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
662 			if (ifa->ifa_addr->sa_family != AF_INET)
663 				continue;
664 			ia = ifatoia(ifa);
665 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
666 			    t.s_addr)
667 				goto match;
668 		}
669 	}
670 	/*
671 	 * If the incoming packet was not addressed directly to us, use
672 	 * designated interface for icmp replies specified by sysctl
673 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
674 	 * with normal source selection.
675 	 */
676 	if (reply_src[0] != '\0' && (ifn = ifunit(reply_src))) {
677 		TAILQ_FOREACH(ifa, &ifn->if_addrhead, ifa_link) {
678 			if (ifa->ifa_addr->sa_family != AF_INET)
679 				continue;
680 			ia = ifatoia(ifa);
681 			goto match;
682 		}
683 	}
684 	/*
685 	 * If the packet was transiting through us, use the address of
686 	 * the interface that is the closest to the packet source.
687 	 * When we don't have a route back to the packet source, stop here
688 	 * and drop the packet.
689 	 */
690 	ia = ip_rtaddr(ip->ip_dst);
691 	if (ia == NULL) {
692 		m_freem(m);
693 		icmpstat.icps_noroute++;
694 		goto done;
695 	}
696 match:
697 #ifdef MAC
698 	mac_reflect_mbuf_icmp(m);
699 #endif
700 	t = IA_SIN(ia)->sin_addr;
701 	ip->ip_src = t;
702 	ip->ip_ttl = ip_defttl;
703 
704 	if (optlen > 0) {
705 		register u_char *cp;
706 		int opt, cnt;
707 		u_int len;
708 
709 		/*
710 		 * Retrieve any source routing from the incoming packet;
711 		 * add on any record-route or timestamp options.
712 		 */
713 		cp = (u_char *) (ip + 1);
714 		if ((opts = ip_srcroute(m)) == 0 &&
715 		    (opts = m_gethdr(M_DONTWAIT, MT_HEADER))) {
716 			opts->m_len = sizeof(struct in_addr);
717 			mtod(opts, struct in_addr *)->s_addr = 0;
718 		}
719 		if (opts) {
720 #ifdef ICMPPRINTFS
721 		    if (icmpprintfs)
722 			    printf("icmp_reflect optlen %d rt %d => ",
723 				optlen, opts->m_len);
724 #endif
725 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
726 			    opt = cp[IPOPT_OPTVAL];
727 			    if (opt == IPOPT_EOL)
728 				    break;
729 			    if (opt == IPOPT_NOP)
730 				    len = 1;
731 			    else {
732 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
733 					    break;
734 				    len = cp[IPOPT_OLEN];
735 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
736 				        len > cnt)
737 					    break;
738 			    }
739 			    /*
740 			     * Should check for overflow, but it "can't happen"
741 			     */
742 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
743 				opt == IPOPT_SECURITY) {
744 				    bcopy((caddr_t)cp,
745 					mtod(opts, caddr_t) + opts->m_len, len);
746 				    opts->m_len += len;
747 			    }
748 		    }
749 		    /* Terminate & pad, if necessary */
750 		    cnt = opts->m_len % 4;
751 		    if (cnt) {
752 			    for (; cnt < 4; cnt++) {
753 				    *(mtod(opts, caddr_t) + opts->m_len) =
754 					IPOPT_EOL;
755 				    opts->m_len++;
756 			    }
757 		    }
758 #ifdef ICMPPRINTFS
759 		    if (icmpprintfs)
760 			    printf("%d\n", opts->m_len);
761 #endif
762 		}
763 		/*
764 		 * Now strip out original options by copying rest of first
765 		 * mbuf's data back, and adjust the IP length.
766 		 */
767 		ip->ip_len -= optlen;
768 		ip->ip_v = IPVERSION;
769 		ip->ip_hl = 5;
770 		m->m_len -= optlen;
771 		if (m->m_flags & M_PKTHDR)
772 			m->m_pkthdr.len -= optlen;
773 		optlen += sizeof(struct ip);
774 		bcopy((caddr_t)ip + optlen, (caddr_t)(ip + 1),
775 			 (unsigned)(m->m_len - sizeof(struct ip)));
776 	}
777 	m_tag_delete_nonpersistent(m);
778 	m->m_flags &= ~(M_BCAST|M_MCAST);
779 	icmp_send(m, opts);
780 done:
781 	if (opts)
782 		(void)m_free(opts);
783 }
784 
785 /*
786  * Send an icmp packet back to the ip level,
787  * after supplying a checksum.
788  */
789 static void
790 icmp_send(m, opts)
791 	register struct mbuf *m;
792 	struct mbuf *opts;
793 {
794 	register struct ip *ip = mtod(m, struct ip *);
795 	register int hlen;
796 	register struct icmp *icp;
797 
798 	hlen = ip->ip_hl << 2;
799 	m->m_data += hlen;
800 	m->m_len -= hlen;
801 	icp = mtod(m, struct icmp *);
802 	icp->icmp_cksum = 0;
803 	icp->icmp_cksum = in_cksum(m, ip->ip_len - hlen);
804 	m->m_data -= hlen;
805 	m->m_len += hlen;
806 	m->m_pkthdr.rcvif = (struct ifnet *)0;
807 #ifdef ICMPPRINTFS
808 	if (icmpprintfs) {
809 		char buf[4 * sizeof "123"];
810 		strcpy(buf, inet_ntoa(ip->ip_dst));
811 		printf("icmp_send dst %s src %s\n",
812 		       buf, inet_ntoa(ip->ip_src));
813 	}
814 #endif
815 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
816 }
817 
818 n_time
819 iptime()
820 {
821 	struct timeval atv;
822 	u_long t;
823 
824 	getmicrotime(&atv);
825 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
826 	return (htonl(t));
827 }
828 
829 /*
830  * Return the next larger or smaller MTU plateau (table from RFC 1191)
831  * given current value MTU.  If DIR is less than zero, a larger plateau
832  * is returned; otherwise, a smaller value is returned.
833  */
834 static int
835 ip_next_mtu(mtu, dir)
836 	int mtu;
837 	int dir;
838 {
839 	static int mtutab[] = {
840 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1006, 508, 296,
841 		68, 0
842 	};
843 	int i;
844 
845 	for (i = 0; i < (sizeof mtutab) / (sizeof mtutab[0]); i++) {
846 		if (mtu >= mtutab[i])
847 			break;
848 	}
849 
850 	if (dir < 0) {
851 		if (i == 0) {
852 			return 0;
853 		} else {
854 			return mtutab[i - 1];
855 		}
856 	} else {
857 		if (mtutab[i] == 0) {
858 			return 0;
859 		} else if(mtu > mtutab[i]) {
860 			return mtutab[i];
861 		} else {
862 			return mtutab[i + 1];
863 		}
864 	}
865 }
866 
867 
868 /*
869  * badport_bandlim() - check for ICMP bandwidth limit
870  *
871  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
872  *	hit our bandwidth limit and it is not ok.
873  *
874  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
875  *
876  *	For now we separate the TCP and UDP subsystems w/ different 'which'
877  *	values.  We may eventually remove this separation (and simplify the
878  *	code further).
879  *
880  *	Note that the printing of the error message is delayed so we can
881  *	properly print the icmp error rate that the system was trying to do
882  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
883  *	the 'final' error, but it doesn't make sense to solve the printing
884  *	delay with more complex code.
885  */
886 
887 int
888 badport_bandlim(int which)
889 {
890 #define	N(a)	(sizeof (a) / sizeof (a[0]))
891 	static struct rate {
892 		const char	*type;
893 		struct timeval	lasttime;
894 		int		curpps;
895 	} rates[BANDLIM_MAX+1] = {
896 		{ "icmp unreach response" },
897 		{ "icmp ping response" },
898 		{ "icmp tstamp response" },
899 		{ "closed port RST response" },
900 		{ "open port RST response" }
901 	};
902 
903 	/*
904 	 * Return ok status if feature disabled or argument out of range.
905 	 */
906 	if (icmplim > 0 && (u_int) which < N(rates)) {
907 		struct rate *r = &rates[which];
908 		int opps = r->curpps;
909 
910 		if (!ppsratecheck(&r->lasttime, &r->curpps, icmplim))
911 			return -1;	/* discard packet */
912 		/*
913 		 * If we've dropped below the threshold after having
914 		 * rate-limited traffic print the message.  This preserves
915 		 * the previous behaviour at the expense of added complexity.
916 		 */
917 		if (icmplim_output && opps > icmplim)
918 			printf("Limiting %s from %d to %d packets/sec\n",
919 				r->type, opps, icmplim);
920 	}
921 	return 0;			/* okay to send packet */
922 #undef N
923 }
924