xref: /openbsd-src/sys/netinet/ip_output.c (revision 43003dfe3ad45d1698bed8a37f2b0f5b14f20d4f)
1 /*	$OpenBSD: ip_output.c,v 1.195 2009/10/06 21:21:48 claudio Exp $	*/
2 /*	$NetBSD: ip_output.c,v 1.28 1996/02/13 23:43:07 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
33  */
34 
35 #include "pf.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mbuf.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/proc.h>
44 #include <sys/kernel.h>
45 
46 #include <net/if.h>
47 #include <net/if_enc.h>
48 #include <net/route.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/ip.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/in_var.h>
55 #include <netinet/ip_var.h>
56 #include <netinet/ip_icmp.h>
57 #include <netinet/tcp.h>
58 #include <netinet/udp.h>
59 #include <netinet/tcp_timer.h>
60 #include <netinet/tcp_var.h>
61 #include <netinet/udp_var.h>
62 
63 #if NPF > 0
64 #include <net/pfvar.h>
65 #endif
66 
67 #ifdef IPSEC
68 #ifdef ENCDEBUG
69 #define DPRINTF(x)    do { if (encdebug) printf x ; } while (0)
70 #else
71 #define DPRINTF(x)
72 #endif
73 
74 extern u_int8_t get_sa_require(struct inpcb *);
75 
76 extern int ipsec_auth_default_level;
77 extern int ipsec_esp_trans_default_level;
78 extern int ipsec_esp_network_default_level;
79 extern int ipsec_ipcomp_default_level;
80 extern int ipforwarding;
81 #endif /* IPSEC */
82 
83 #ifdef MROUTING
84 extern int ipmforwarding;
85 #endif
86 
87 struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
88 void ip_mloopback(struct ifnet *, struct mbuf *, struct sockaddr_in *);
89 
90 /*
91  * IP output.  The packet in mbuf chain m contains a skeletal IP
92  * header (with len, off, ttl, proto, tos, src, dst).
93  * The mbuf chain containing the packet will be freed.
94  * The mbuf opt, if present, will not be freed.
95  */
96 int
97 ip_output(struct mbuf *m0, ...)
98 {
99 	struct ip *ip;
100 	struct ifnet *ifp;
101 	struct mbuf *m = m0;
102 	int hlen = sizeof (struct ip);
103 	int len, error = 0;
104 	struct route iproute;
105 	struct sockaddr_in *dst;
106 	struct in_ifaddr *ia;
107 	struct mbuf *opt;
108 	struct route *ro;
109 	int flags;
110 	struct ip_moptions *imo;
111 	va_list ap;
112 	u_int8_t sproto = 0, donerouting = 0;
113 	u_long mtu;
114 #ifdef IPSEC
115 	u_int32_t icmp_mtu = 0;
116 	union sockaddr_union sdst;
117 	u_int32_t sspi;
118 	struct m_tag *mtag;
119 	struct tdb_ident *tdbi;
120 
121 	struct inpcb *inp;
122 	struct tdb *tdb;
123 	int s;
124 #endif /* IPSEC */
125 
126 	va_start(ap, m0);
127 	opt = va_arg(ap, struct mbuf *);
128 	ro = va_arg(ap, struct route *);
129 	flags = va_arg(ap, int);
130 	imo = va_arg(ap, struct ip_moptions *);
131 #ifdef IPSEC
132 	inp = va_arg(ap, struct inpcb *);
133 	if (inp && (inp->inp_flags & INP_IPV6) != 0)
134 		panic("ip_output: IPv6 pcb is passed");
135 #endif /* IPSEC */
136 	va_end(ap);
137 
138 #ifdef	DIAGNOSTIC
139 	if ((m->m_flags & M_PKTHDR) == 0)
140 		panic("ip_output no HDR");
141 #endif
142 	if (opt) {
143 		m = ip_insertoptions(m, opt, &len);
144 		hlen = len;
145 	}
146 
147 	ip = mtod(m, struct ip *);
148 
149 	/*
150 	 * Fill in IP header.
151 	 */
152 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
153 		ip->ip_v = IPVERSION;
154 		ip->ip_off &= htons(IP_DF);
155 		ip->ip_id = htons(ip_randomid());
156 		ip->ip_hl = hlen >> 2;
157 		ipstat.ips_localout++;
158 	} else {
159 		hlen = ip->ip_hl << 2;
160 	}
161 
162 	/*
163 	 * If we're missing the IP source address, do a route lookup. We'll
164 	 * remember this result, in case we don't need to do any IPsec
165 	 * processing on the packet. We need the source address so we can
166 	 * do an SPD lookup in IPsec; for most packets, the source address
167 	 * is set at a higher level protocol. ICMPs and other packets
168 	 * though (e.g., traceroute) have a source address of zeroes.
169 	 */
170 	if (ip->ip_src.s_addr == INADDR_ANY) {
171 		if (flags & IP_ROUTETOETHER) {
172 			error = EINVAL;
173 			goto bad;
174 		}
175 		donerouting = 1;
176 
177 		if (ro == 0) {
178 			ro = &iproute;
179 			bzero((caddr_t)ro, sizeof (*ro));
180 		}
181 
182 		dst = satosin(&ro->ro_dst);
183 
184 		/*
185 		 * If there is a cached route, check that it is to the same
186 		 * destination and is still up.  If not, free it and try again.
187 		 */
188 		if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
189 				  dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
190 			RTFREE(ro->ro_rt);
191 			ro->ro_rt = (struct rtentry *)0;
192 		}
193 
194 		if (ro->ro_rt == 0) {
195 			dst->sin_family = AF_INET;
196 			dst->sin_len = sizeof(*dst);
197 			dst->sin_addr = ip->ip_dst;
198 		}
199 
200 		/*
201 		 * If routing to interface only, short-circuit routing lookup.
202 		 */
203 		if (flags & IP_ROUTETOIF) {
204 			if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
205 			    m->m_pkthdr.rdomain))) == 0 &&
206 			    (ia = ifatoia(ifa_ifwithnet(sintosa(dst),
207 			    m->m_pkthdr.rdomain))) == 0) {
208 				ipstat.ips_noroute++;
209 				error = ENETUNREACH;
210 				goto bad;
211 			}
212 
213 			ifp = ia->ia_ifp;
214 			mtu = ifp->if_mtu;
215 			ip->ip_ttl = 1;
216 		} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
217 		    (ip->ip_dst.s_addr == INADDR_BROADCAST)) &&
218 		    imo != NULL && imo->imo_multicast_ifp != NULL) {
219 			ifp = imo->imo_multicast_ifp;
220 			mtu = ifp->if_mtu;
221 			IFP_TO_IA(ifp, ia);
222 		} else {
223 			if (ro->ro_rt == 0)
224 				rtalloc_mpath(ro, NULL,
225 				    m->m_pkthdr.rdomain);
226 
227 			if (ro->ro_rt == 0) {
228 				ipstat.ips_noroute++;
229 				error = EHOSTUNREACH;
230 				goto bad;
231 			}
232 
233 			ia = ifatoia(ro->ro_rt->rt_ifa);
234 			ifp = ro->ro_rt->rt_ifp;
235 			if ((mtu = ro->ro_rt->rt_rmx.rmx_mtu) == 0)
236 				mtu = ifp->if_mtu;
237 			ro->ro_rt->rt_use++;
238 
239 			if (ro->ro_rt->rt_flags & RTF_GATEWAY)
240 				dst = satosin(ro->ro_rt->rt_gateway);
241 		}
242 
243 		/* Set the source IP address */
244 		if (!IN_MULTICAST(ip->ip_dst.s_addr))
245 			ip->ip_src = ia->ia_addr.sin_addr;
246 	}
247 
248 #if NPF > 0
249 reroute:
250 #endif
251 
252 #ifdef IPSEC
253 	if (!ipsec_in_use && inp == NULL)
254 		goto done_spd;
255 
256 	/*
257 	 * splnet is chosen over spltdb because we are not allowed to
258 	 * lower the level, and udp_output calls us in splnet().
259 	 */
260 	s = splnet();
261 
262 	/* Do we have any pending SAs to apply ? */
263 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
264 	if (mtag != NULL) {
265 #ifdef DIAGNOSTIC
266 		if (mtag->m_tag_len != sizeof (struct tdb_ident))
267 			panic("ip_output: tag of length %d (should be %d",
268 			    mtag->m_tag_len, sizeof (struct tdb_ident));
269 #endif
270 		tdbi = (struct tdb_ident *)(mtag + 1);
271 		tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto);
272 		if (tdb == NULL)
273 			error = -EINVAL;
274 		m_tag_delete(m, mtag);
275 	}
276 	else
277 		tdb = ipsp_spd_lookup(m, AF_INET, hlen, &error,
278 		    IPSP_DIRECTION_OUT, NULL, inp);
279 
280 	if (tdb == NULL) {
281 		splx(s);
282 
283 		if (error == 0) {
284 			/*
285 			 * No IPsec processing required, we'll just send the
286 			 * packet out.
287 			 */
288 			sproto = 0;
289 
290 			/* Fall through to routing/multicast handling */
291 		} else {
292 			/*
293 			 * -EINVAL is used to indicate that the packet should
294 			 * be silently dropped, typically because we've asked
295 			 * key management for an SA.
296 			 */
297 			if (error == -EINVAL) /* Should silently drop packet */
298 			  error = 0;
299 
300 			m_freem(m);
301 			goto done;
302 		}
303 	} else {
304 		/* Loop detection */
305 		for (mtag = m_tag_first(m); mtag != NULL;
306 		    mtag = m_tag_next(m, mtag)) {
307 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
308 			    mtag->m_tag_id !=
309 			    PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
310 				continue;
311 			tdbi = (struct tdb_ident *)(mtag + 1);
312 			if (tdbi->spi == tdb->tdb_spi &&
313 			    tdbi->proto == tdb->tdb_sproto &&
314 			    !bcmp(&tdbi->dst, &tdb->tdb_dst,
315 			    sizeof(union sockaddr_union))) {
316 				splx(s);
317 				sproto = 0; /* mark as no-IPsec-needed */
318 				goto done_spd;
319 			}
320 		}
321 
322 		/* We need to do IPsec */
323 		bcopy(&tdb->tdb_dst, &sdst, sizeof(sdst));
324 		sspi = tdb->tdb_spi;
325 		sproto = tdb->tdb_sproto;
326 		splx(s);
327 
328 		/*
329 		 * If it needs TCP/UDP hardware-checksumming, do the
330 		 * computation now.
331 		 */
332 		if (m->m_pkthdr.csum_flags & (M_TCPV4_CSUM_OUT | M_UDPV4_CSUM_OUT)) {
333 			in_delayed_cksum(m);
334 			m->m_pkthdr.csum_flags &=
335 			    ~(M_UDPV4_CSUM_OUT | M_TCPV4_CSUM_OUT);
336 		}
337 
338 		/* If it's not a multicast packet, try to fast-path */
339 		if (!IN_MULTICAST(ip->ip_dst.s_addr)) {
340 			goto sendit;
341 		}
342 	}
343 
344 	/* Fall through to the routing/multicast handling code */
345  done_spd:
346 #endif /* IPSEC */
347 
348 	if (flags & IP_ROUTETOETHER) {
349 		dst = satosin(&ro->ro_dst);
350 		ifp = ro->ro_rt->rt_ifp;
351 		mtu = ifp->if_mtu;
352 		ro->ro_rt = NULL;
353 	} else if (donerouting == 0) {
354 		if (ro == 0) {
355 			ro = &iproute;
356 			bzero((caddr_t)ro, sizeof (*ro));
357 		}
358 
359 		dst = satosin(&ro->ro_dst);
360 
361 		/*
362 		 * If there is a cached route, check that it is to the same
363 		 * destination and is still up.  If not, free it and try again.
364 		 */
365 		if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
366 				  dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
367 			RTFREE(ro->ro_rt);
368 			ro->ro_rt = (struct rtentry *)0;
369 		}
370 
371 		if (ro->ro_rt == 0) {
372 			dst->sin_family = AF_INET;
373 			dst->sin_len = sizeof(*dst);
374 			dst->sin_addr = ip->ip_dst;
375 		}
376 
377 		/*
378 		 * If routing to interface only, short-circuit routing lookup.
379 		 */
380 		if (flags & IP_ROUTETOIF) {
381 			if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
382 			    m->m_pkthdr.rdomain))) == 0 &&
383 			    (ia = ifatoia(ifa_ifwithnet(sintosa(dst),
384 			    m->m_pkthdr.rdomain))) == 0) {
385 				ipstat.ips_noroute++;
386 				error = ENETUNREACH;
387 				goto bad;
388 			}
389 
390 			ifp = ia->ia_ifp;
391 			mtu = ifp->if_mtu;
392 			ip->ip_ttl = 1;
393 		} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
394 		    (ip->ip_dst.s_addr == INADDR_BROADCAST)) &&
395 		    imo != NULL && imo->imo_multicast_ifp != NULL) {
396 			ifp = imo->imo_multicast_ifp;
397 			mtu = ifp->if_mtu;
398 			IFP_TO_IA(ifp, ia);
399 		} else {
400 			if (ro->ro_rt == 0)
401 				rtalloc_mpath(ro, &ip->ip_src.s_addr,
402 				    m->m_pkthdr.rdomain);
403 
404 			if (ro->ro_rt == 0) {
405 				ipstat.ips_noroute++;
406 				error = EHOSTUNREACH;
407 				goto bad;
408 			}
409 
410 			ia = ifatoia(ro->ro_rt->rt_ifa);
411 			ifp = ro->ro_rt->rt_ifp;
412 			if ((mtu = ro->ro_rt->rt_rmx.rmx_mtu) == 0)
413 				mtu = ifp->if_mtu;
414 			ro->ro_rt->rt_use++;
415 
416 			if (ro->ro_rt->rt_flags & RTF_GATEWAY)
417 				dst = satosin(ro->ro_rt->rt_gateway);
418 		}
419 
420 		/* Set the source IP address */
421 		if (ip->ip_src.s_addr == INADDR_ANY)
422 			ip->ip_src = ia->ia_addr.sin_addr;
423 	}
424 
425 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
426 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
427 		struct in_multi *inm;
428 
429 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
430 			M_BCAST : M_MCAST;
431 
432 		/*
433 		 * IP destination address is multicast.  Make sure "dst"
434 		 * still points to the address in "ro".  (It may have been
435 		 * changed to point to a gateway address, above.)
436 		 */
437 		dst = satosin(&ro->ro_dst);
438 
439 		/*
440 		 * See if the caller provided any multicast options
441 		 */
442 		if (imo != NULL)
443 			ip->ip_ttl = imo->imo_multicast_ttl;
444 		else
445 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
446 
447 		/*
448 		 * if we don't know the outgoing ifp yet, we can't generate
449 		 * output
450 		 */
451 		if (!ifp) {
452 			ipstat.ips_noroute++;
453 			error = EHOSTUNREACH;
454 			goto bad;
455 		}
456 
457 		/*
458 		 * Confirm that the outgoing interface supports multicast,
459 		 * but only if the packet actually is going out on that
460 		 * interface (i.e., no IPsec is applied).
461 		 */
462 		if ((((m->m_flags & M_MCAST) &&
463 		      (ifp->if_flags & IFF_MULTICAST) == 0) ||
464 		     ((m->m_flags & M_BCAST) &&
465 		      (ifp->if_flags & IFF_BROADCAST) == 0)) && (sproto == 0)) {
466 			ipstat.ips_noroute++;
467 			error = ENETUNREACH;
468 			goto bad;
469 		}
470 
471 		/*
472 		 * If source address not specified yet, use address
473 		 * of outgoing interface.
474 		 */
475 		if (ip->ip_src.s_addr == INADDR_ANY) {
476 			struct in_ifaddr *ia;
477 
478 			TAILQ_FOREACH(ia, &in_ifaddr, ia_list)
479 				if (ia->ia_ifp == ifp) {
480 					ip->ip_src = ia->ia_addr.sin_addr;
481 					break;
482 				}
483 		}
484 
485 		IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm);
486 		if (inm != NULL &&
487 		   (imo == NULL || imo->imo_multicast_loop)) {
488 			/*
489 			 * If we belong to the destination multicast group
490 			 * on the outgoing interface, and the caller did not
491 			 * forbid loopback, loop back a copy.
492 			 * Can't defer TCP/UDP checksumming, do the
493 			 * computation now.
494 			 */
495 			if (m->m_pkthdr.csum_flags &
496 			    (M_TCPV4_CSUM_OUT | M_UDPV4_CSUM_OUT)) {
497 				in_delayed_cksum(m);
498 				m->m_pkthdr.csum_flags &=
499 				    ~(M_UDPV4_CSUM_OUT | M_TCPV4_CSUM_OUT);
500 			}
501 			ip_mloopback(ifp, m, dst);
502 		}
503 #ifdef MROUTING
504 		else {
505 			/*
506 			 * If we are acting as a multicast router, perform
507 			 * multicast forwarding as if the packet had just
508 			 * arrived on the interface to which we are about
509 			 * to send.  The multicast forwarding function
510 			 * recursively calls this function, using the
511 			 * IP_FORWARDING flag to prevent infinite recursion.
512 			 *
513 			 * Multicasts that are looped back by ip_mloopback(),
514 			 * above, will be forwarded by the ip_input() routine,
515 			 * if necessary.
516 			 */
517 			extern struct socket *ip_mrouter;
518 
519 			if (ipmforwarding && ip_mrouter &&
520 			    (flags & IP_FORWARDING) == 0) {
521 				if (ip_mforward(m, ifp) != 0) {
522 					m_freem(m);
523 					goto done;
524 				}
525 			}
526 		}
527 #endif
528 		/*
529 		 * Multicasts with a time-to-live of zero may be looped-
530 		 * back, above, but must not be transmitted on a network.
531 		 * Also, multicasts addressed to the loopback interface
532 		 * are not sent -- the above call to ip_mloopback() will
533 		 * loop back a copy if this host actually belongs to the
534 		 * destination group on the loopback interface.
535 		 */
536 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
537 			m_freem(m);
538 			goto done;
539 		}
540 
541 		goto sendit;
542 	}
543 
544 	/*
545 	 * Look for broadcast address and and verify user is allowed to send
546 	 * such a packet; if the packet is going in an IPsec tunnel, skip
547 	 * this check.
548 	 */
549 	if ((sproto == 0) && (in_broadcast(dst->sin_addr, ifp))) {
550 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
551 			error = EADDRNOTAVAIL;
552 			goto bad;
553 		}
554 		if ((flags & IP_ALLOWBROADCAST) == 0) {
555 			error = EACCES;
556 			goto bad;
557 		}
558 
559 		/* Don't allow broadcast messages to be fragmented */
560 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
561 			error = EMSGSIZE;
562 			goto bad;
563 		}
564 		m->m_flags |= M_BCAST;
565 	} else
566 		m->m_flags &= ~M_BCAST;
567 
568 sendit:
569 	/*
570 	 * If we're doing Path MTU discovery, we need to set DF unless
571 	 * the route's MTU is locked.
572 	 */
573 	if ((flags & IP_MTUDISC) && ro && ro->ro_rt &&
574 	    (ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
575 		ip->ip_off |= htons(IP_DF);
576 
577 #ifdef IPSEC
578 	/*
579 	 * Check if the packet needs encapsulation.
580 	 */
581 	if (sproto != 0) {
582 		s = splnet();
583 
584 		/*
585 		 * Packet filter
586 		 */
587 #if NPF > 0
588 
589 		if (pf_test(PF_OUT, &encif[0].sc_if, &m, NULL) != PF_PASS) {
590 			error = EHOSTUNREACH;
591 			splx(s);
592 			m_freem(m);
593 			goto done;
594 		}
595 		if (m == NULL) {
596 			splx(s);
597 			goto done;
598 		}
599 		ip = mtod(m, struct ip *);
600 		hlen = ip->ip_hl << 2;
601 		/*
602 		 * PF_TAG_REROUTE handling or not...
603 		 * Packet is entering IPsec so the routing is
604 		 * already overruled by the IPsec policy.
605 		 * Until now the change was not reconsidered.
606 		 * What's the behaviour?
607 		 */
608 #endif
609 
610 		tdb = gettdb(sspi, &sdst, sproto);
611 		if (tdb == NULL) {
612 			DPRINTF(("ip_output: unknown TDB"));
613 			error = EHOSTUNREACH;
614 			splx(s);
615 			m_freem(m);
616 			goto done;
617 		}
618 
619 		/* Check if we are allowed to fragment */
620 		if (ip_mtudisc && (ip->ip_off & htons(IP_DF)) && tdb->tdb_mtu &&
621 		    ntohs(ip->ip_len) > tdb->tdb_mtu &&
622 		    tdb->tdb_mtutimeout > time_second) {
623 			struct rtentry *rt = NULL;
624 			int rt_mtucloned = 0;
625 			int transportmode = 0;
626 
627 			transportmode = (tdb->tdb_dst.sa.sa_family == AF_INET) &&
628 			    (tdb->tdb_dst.sin.sin_addr.s_addr ==
629 			    ip->ip_dst.s_addr);
630 			icmp_mtu = tdb->tdb_mtu;
631 			splx(s);
632 
633 			/* Find a host route to store the mtu in */
634 			if (ro != NULL)
635 				rt = ro->ro_rt;
636 			/* but don't add a PMTU route for transport mode SAs */
637 			if (transportmode)
638 				rt = NULL;
639 			else if (rt == NULL || (rt->rt_flags & RTF_HOST) == 0) {
640 				struct sockaddr_in dst = {
641 					sizeof(struct sockaddr_in), AF_INET};
642 				dst.sin_addr = ip->ip_dst;
643 				rt = icmp_mtudisc_clone((struct sockaddr *)&dst,
644 				    m->m_pkthdr.rdomain);
645 				rt_mtucloned = 1;
646 			}
647 			DPRINTF(("ip_output: spi %08x mtu %d rt %p cloned %d\n",
648 			    ntohl(tdb->tdb_spi), icmp_mtu, rt, rt_mtucloned));
649 			if (rt != NULL) {
650 				rt->rt_rmx.rmx_mtu = icmp_mtu;
651 				if (ro && ro->ro_rt != NULL) {
652 					RTFREE(ro->ro_rt);
653 					ro->ro_rt = NULL;
654 					rtalloc1(&ro->ro_dst, 1,
655 					    m->m_pkthdr.rdomain);
656 				}
657 				if (rt_mtucloned)
658 					rtfree(rt);
659 			}
660 			error = EMSGSIZE;
661 			goto bad;
662 		}
663 
664 		/*
665 		 * Clear these -- they'll be set in the recursive invocation
666 		 * as needed.
667 		 */
668 		m->m_flags &= ~(M_MCAST | M_BCAST);
669 
670 		/* Callee frees mbuf */
671 		error = ipsp_process_packet(m, tdb, AF_INET, 0);
672 		splx(s);
673 		return error;  /* Nothing more to be done */
674 	}
675 
676 	/*
677 	 * If deferred crypto processing is needed, check that the
678 	 * interface supports it.
679 	 */
680 	if (ipsec_in_use && (mtag = m_tag_find(m,
681 	    PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL)) != NULL &&
682 	    (ifp->if_capabilities & IFCAP_IPSEC) == 0) {
683 		/* Notify IPsec to do its own crypto. */
684 		ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
685 		m_freem(m);
686 		error = EHOSTUNREACH;
687 		goto done;
688 	}
689 #endif /* IPSEC */
690 
691 	/* Catch routing changes wrt. hardware checksumming for TCP or UDP. */
692 	if (m->m_pkthdr.csum_flags & M_TCPV4_CSUM_OUT) {
693 		if (!(ifp->if_capabilities & IFCAP_CSUM_TCPv4) ||
694 		    ifp->if_bridge != NULL) {
695 			in_delayed_cksum(m);
696 			m->m_pkthdr.csum_flags &= ~M_TCPV4_CSUM_OUT; /* Clear */
697 		}
698 	} else if (m->m_pkthdr.csum_flags & M_UDPV4_CSUM_OUT) {
699 		if (!(ifp->if_capabilities & IFCAP_CSUM_UDPv4) ||
700 		    ifp->if_bridge != NULL) {
701 			in_delayed_cksum(m);
702 			m->m_pkthdr.csum_flags &= ~M_UDPV4_CSUM_OUT; /* Clear */
703 		}
704 	}
705 
706 	/*
707 	 * Packet filter
708 	 */
709 #if NPF > 0
710 	if (pf_test(PF_OUT, ifp, &m, NULL) != PF_PASS) {
711 		error = EHOSTUNREACH;
712 		m_freem(m);
713 		goto done;
714 	}
715 	if (m == NULL)
716 		goto done;
717 	ip = mtod(m, struct ip *);
718 	hlen = ip->ip_hl << 2;
719 	if ((m->m_pkthdr.pf.flags & (PF_TAG_REROUTE | PF_TAG_GENERATED)) ==
720 	    (PF_TAG_REROUTE | PF_TAG_GENERATED))
721 		/* already rerun the route lookup, go on */
722 		m->m_pkthdr.pf.flags &= ~(PF_TAG_GENERATED | PF_TAG_REROUTE);
723 	else if (m->m_pkthdr.pf.flags & PF_TAG_REROUTE) {
724 		/* tag as generated to skip over pf_test on rerun */
725 		m->m_pkthdr.pf.flags |= PF_TAG_GENERATED;
726 		ro = NULL;
727 		donerouting = 0;
728 		goto reroute;
729 	}
730 #endif
731 
732 #ifdef IPSEC
733 	if (ipsec_in_use && (flags & IP_FORWARDING) && (ipforwarding == 2) &&
734 	    (m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) == NULL)) {
735 		error = EHOSTUNREACH;
736 		m_freem(m);
737 		goto done;
738 	}
739 #endif
740 
741 	/* XXX
742 	 * Try to use jumbograms based on socket option, or the route
743 	 * or... for other reasons later on.
744 	 */
745 	if ((flags & IP_JUMBO) && ro->ro_rt && (ro->ro_rt->rt_flags & RTF_JUMBO) &&
746 	    ro->ro_rt->rt_ifp)
747 		mtu = ro->ro_rt->rt_ifp->if_hardmtu;
748 
749 	/*
750 	 * If small enough for interface, can just send directly.
751 	 */
752 	if (ntohs(ip->ip_len) <= mtu) {
753 		ip->ip_sum = 0;
754 		if ((ifp->if_capabilities & IFCAP_CSUM_IPv4) &&
755 		    ifp->if_bridge == NULL) {
756 			m->m_pkthdr.csum_flags |= M_IPV4_CSUM_OUT;
757 			ipstat.ips_outhwcsum++;
758 		} else
759 			ip->ip_sum = in_cksum(m, hlen);
760 		/* Update relevant hardware checksum stats for TCP/UDP */
761 		if (m->m_pkthdr.csum_flags & M_TCPV4_CSUM_OUT)
762 			tcpstat.tcps_outhwcsum++;
763 		else if (m->m_pkthdr.csum_flags & M_UDPV4_CSUM_OUT)
764 			udpstat.udps_outhwcsum++;
765 		error = (*ifp->if_output)(ifp, m, sintosa(dst), ro->ro_rt);
766 		goto done;
767 	}
768 
769 	/*
770 	 * Too large for interface; fragment if possible.
771 	 * Must be able to put at least 8 bytes per fragment.
772 	 */
773 	if (ip->ip_off & htons(IP_DF)) {
774 #ifdef IPSEC
775 		icmp_mtu = ifp->if_mtu;
776 #endif
777 		error = EMSGSIZE;
778 		/*
779 		 * This case can happen if the user changed the MTU
780 		 * of an interface after enabling IP on it.  Because
781 		 * most netifs don't keep track of routes pointing to
782 		 * them, there is no way for one to update all its
783 		 * routes when the MTU is changed.
784 		 */
785 		if (ro->ro_rt != NULL &&
786 		    (ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
787 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
788 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
789 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
790 		}
791 		ipstat.ips_cantfrag++;
792 		goto bad;
793 	}
794 
795 	error = ip_fragment(m, ifp, mtu);
796 	if (error) {
797 		m = m0 = NULL;
798 		goto bad;
799 	}
800 
801 	for (; m; m = m0) {
802 		m0 = m->m_nextpkt;
803 		m->m_nextpkt = 0;
804 		if (error == 0)
805 			error = (*ifp->if_output)(ifp, m, sintosa(dst),
806 			    ro->ro_rt);
807 		else
808 			m_freem(m);
809 	}
810 
811 	if (error == 0)
812 		ipstat.ips_fragmented++;
813 
814 done:
815 	if (ro == &iproute && (flags & IP_ROUTETOIF) == 0 && ro->ro_rt)
816 		RTFREE(ro->ro_rt);
817 	return (error);
818 bad:
819 #ifdef IPSEC
820 	if (error == EMSGSIZE && ip_mtudisc && icmp_mtu != 0 && m != NULL)
821 		ipsec_adjust_mtu(m, icmp_mtu);
822 #endif
823 	m_freem(m0);
824 	goto done;
825 }
826 
827 int
828 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
829 {
830 	struct ip *ip, *mhip;
831 	struct mbuf *m0;
832 	int len, hlen, off;
833 	int mhlen, firstlen;
834 	struct mbuf **mnext;
835 	int fragments = 0;
836 	int s;
837 	int error = 0;
838 
839 	ip = mtod(m, struct ip *);
840 	hlen = ip->ip_hl << 2;
841 
842 	len = (mtu - hlen) &~ 7;
843 	if (len < 8) {
844 		m_freem(m);
845 		return (EMSGSIZE);
846 	}
847 
848 	/*
849 	 * If we are doing fragmentation, we can't defer TCP/UDP
850 	 * checksumming; compute the checksum and clear the flag.
851 	 */
852 	if (m->m_pkthdr.csum_flags & (M_TCPV4_CSUM_OUT | M_UDPV4_CSUM_OUT)) {
853 		in_delayed_cksum(m);
854 		m->m_pkthdr.csum_flags &= ~(M_UDPV4_CSUM_OUT | M_TCPV4_CSUM_OUT);
855 	}
856 
857 	firstlen = len;
858 	mnext = &m->m_nextpkt;
859 
860 	/*
861 	 * Loop through length of segment after first fragment,
862 	 * make new header and copy data of each part and link onto chain.
863 	 */
864 	m0 = m;
865 	mhlen = sizeof (struct ip);
866 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
867 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
868 		if (m == 0) {
869 			ipstat.ips_odropped++;
870 			error = ENOBUFS;
871 			goto sendorfree;
872 		}
873 		*mnext = m;
874 		mnext = &m->m_nextpkt;
875 		m->m_data += max_linkhdr;
876 		mhip = mtod(m, struct ip *);
877 		*mhip = *ip;
878 		/* we must inherit MCAST and BCAST flags */
879 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
880 		if (hlen > sizeof (struct ip)) {
881 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
882 			mhip->ip_hl = mhlen >> 2;
883 		}
884 		m->m_len = mhlen;
885 		mhip->ip_off = ((off - hlen) >> 3) +
886 		    (ntohs(ip->ip_off) & ~IP_MF);
887 		if (ip->ip_off & htons(IP_MF))
888 			mhip->ip_off |= IP_MF;
889 		if (off + len >= ntohs(ip->ip_len))
890 			len = ntohs(ip->ip_len) - off;
891 		else
892 			mhip->ip_off |= IP_MF;
893 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
894 		m->m_next = m_copy(m0, off, len);
895 		if (m->m_next == 0) {
896 			ipstat.ips_odropped++;
897 			error = ENOBUFS;
898 			goto sendorfree;
899 		}
900 		m->m_pkthdr.len = mhlen + len;
901 		m->m_pkthdr.rcvif = (struct ifnet *)0;
902 		mhip->ip_off = htons((u_int16_t)mhip->ip_off);
903 		mhip->ip_sum = 0;
904 		if ((ifp != NULL) &&
905 		    (ifp->if_capabilities & IFCAP_CSUM_IPv4) &&
906 		    ifp->if_bridge == NULL) {
907 			m->m_pkthdr.csum_flags |= M_IPV4_CSUM_OUT;
908 			ipstat.ips_outhwcsum++;
909 		} else
910 			mhip->ip_sum = in_cksum(m, mhlen);
911 		ipstat.ips_ofragments++;
912 		fragments++;
913 	}
914 	/*
915 	 * Update first fragment by trimming what's been copied out
916 	 * and updating header, then send each fragment (in order).
917 	 */
918 	m = m0;
919 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
920 	m->m_pkthdr.len = hlen + firstlen;
921 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
922 	ip->ip_off |= htons(IP_MF);
923 	ip->ip_sum = 0;
924 	if ((ifp != NULL) &&
925 	    (ifp->if_capabilities & IFCAP_CSUM_IPv4) &&
926 	    ifp->if_bridge == NULL) {
927 		m->m_pkthdr.csum_flags |= M_IPV4_CSUM_OUT;
928 		ipstat.ips_outhwcsum++;
929 	} else
930 		ip->ip_sum = in_cksum(m, hlen);
931 sendorfree:
932 	/*
933 	 * If there is no room for all the fragments, don't queue
934 	 * any of them.
935 	 */
936 	if (ifp != NULL) {
937 		s = splnet();
938 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
939 		    error == 0) {
940 			error = ENOBUFS;
941 			ipstat.ips_odropped++;
942 			IFQ_INC_DROPS(&ifp->if_snd);
943 		}
944 		splx(s);
945 	}
946 	if (error) {
947 		for (m = m0; m; m = m0) {
948 			m0 = m->m_nextpkt;
949 			m->m_nextpkt = NULL;
950 			m_freem(m);
951 		}
952 	}
953 
954 	return (error);
955 }
956 
957 /*
958  * Insert IP options into preformed packet.
959  * Adjust IP destination as required for IP source routing,
960  * as indicated by a non-zero in_addr at the start of the options.
961  */
962 struct mbuf *
963 ip_insertoptions(m, opt, phlen)
964 	struct mbuf *m;
965 	struct mbuf *opt;
966 	int *phlen;
967 {
968 	struct ipoption *p = mtod(opt, struct ipoption *);
969 	struct mbuf *n;
970 	struct ip *ip = mtod(m, struct ip *);
971 	unsigned optlen;
972 
973 	optlen = opt->m_len - sizeof(p->ipopt_dst);
974 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
975 		return (m);		/* XXX should fail */
976 	if (p->ipopt_dst.s_addr)
977 		ip->ip_dst = p->ipopt_dst;
978 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
979 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
980 		if (n == 0)
981 			return (m);
982 		M_MOVE_HDR(n, m);
983 		n->m_pkthdr.len += optlen;
984 		m->m_len -= sizeof(struct ip);
985 		m->m_data += sizeof(struct ip);
986 		n->m_next = m;
987 		m = n;
988 		m->m_len = optlen + sizeof(struct ip);
989 		m->m_data += max_linkhdr;
990 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
991 	} else {
992 		m->m_data -= optlen;
993 		m->m_len += optlen;
994 		m->m_pkthdr.len += optlen;
995 		ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
996 	}
997 	ip = mtod(m, struct ip *);
998 	bcopy((caddr_t)p->ipopt_list, (caddr_t)(ip + 1), (unsigned)optlen);
999 	*phlen = sizeof(struct ip) + optlen;
1000 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1001 	return (m);
1002 }
1003 
1004 /*
1005  * Copy options from ip to jp,
1006  * omitting those not copied during fragmentation.
1007  */
1008 int
1009 ip_optcopy(ip, jp)
1010 	struct ip *ip, *jp;
1011 {
1012 	u_char *cp, *dp;
1013 	int opt, optlen, cnt;
1014 
1015 	cp = (u_char *)(ip + 1);
1016 	dp = (u_char *)(jp + 1);
1017 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1018 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1019 		opt = cp[0];
1020 		if (opt == IPOPT_EOL)
1021 			break;
1022 		if (opt == IPOPT_NOP) {
1023 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1024 			*dp++ = IPOPT_NOP;
1025 			optlen = 1;
1026 			continue;
1027 		}
1028 #ifdef DIAGNOSTIC
1029 		if (cnt < IPOPT_OLEN + sizeof(*cp))
1030 			panic("malformed IPv4 option passed to ip_optcopy");
1031 #endif
1032 		optlen = cp[IPOPT_OLEN];
1033 #ifdef DIAGNOSTIC
1034 		if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1035 			panic("malformed IPv4 option passed to ip_optcopy");
1036 #endif
1037 		/* bogus lengths should have been caught by ip_dooptions */
1038 		if (optlen > cnt)
1039 			optlen = cnt;
1040 		if (IPOPT_COPIED(opt)) {
1041 			bcopy((caddr_t)cp, (caddr_t)dp, (unsigned)optlen);
1042 			dp += optlen;
1043 		}
1044 	}
1045 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1046 		*dp++ = IPOPT_EOL;
1047 	return (optlen);
1048 }
1049 
1050 /*
1051  * IP socket option processing.
1052  */
1053 int
1054 ip_ctloutput(op, so, level, optname, mp)
1055 	int op;
1056 	struct socket *so;
1057 	int level, optname;
1058 	struct mbuf **mp;
1059 {
1060 	struct inpcb *inp = sotoinpcb(so);
1061 	struct mbuf *m = *mp;
1062 	int optval = 0;
1063 #ifdef IPSEC
1064 	struct proc *p = curproc; /* XXX */
1065 	struct ipsec_ref *ipr;
1066 	u_int16_t opt16val;
1067 #endif
1068 	int error = 0;
1069 	u_int rtid = 0;
1070 
1071 	if (level != IPPROTO_IP) {
1072 		error = EINVAL;
1073 		if (op == PRCO_SETOPT && *mp)
1074 			(void) m_free(*mp);
1075 	} else switch (op) {
1076 	case PRCO_SETOPT:
1077 		switch (optname) {
1078 		case IP_OPTIONS:
1079 #ifdef notyet
1080 		case IP_RETOPTS:
1081 			return (ip_pcbopts(optname, &inp->inp_options, m));
1082 #else
1083 			return (ip_pcbopts(&inp->inp_options, m));
1084 #endif
1085 
1086 		case IP_TOS:
1087 		case IP_TTL:
1088 		case IP_MINTTL:
1089 		case IP_RECVOPTS:
1090 		case IP_RECVRETOPTS:
1091 		case IP_RECVDSTADDR:
1092 		case IP_RECVIF:
1093 		case IP_RECVTTL:
1094 		case IP_RECVDSTPORT:
1095 			if (m == NULL || m->m_len != sizeof(int))
1096 				error = EINVAL;
1097 			else {
1098 				optval = *mtod(m, int *);
1099 				switch (optname) {
1100 
1101 				case IP_TOS:
1102 					inp->inp_ip.ip_tos = optval;
1103 					break;
1104 
1105 				case IP_TTL:
1106 					if (optval > 0 && optval <= MAXTTL)
1107 						inp->inp_ip.ip_ttl = optval;
1108 					else
1109 						error = EINVAL;
1110 					break;
1111 
1112 				case IP_MINTTL:
1113 					if (optval > 0 && optval <= MAXTTL)
1114 						inp->inp_ip_minttl = optval;
1115 					else
1116 						error = EINVAL;
1117 					break;
1118 #define	OPTSET(bit) \
1119 	if (optval) \
1120 		inp->inp_flags |= bit; \
1121 	else \
1122 		inp->inp_flags &= ~bit;
1123 
1124 				case IP_RECVOPTS:
1125 					OPTSET(INP_RECVOPTS);
1126 					break;
1127 
1128 				case IP_RECVRETOPTS:
1129 					OPTSET(INP_RECVRETOPTS);
1130 					break;
1131 
1132 				case IP_RECVDSTADDR:
1133 					OPTSET(INP_RECVDSTADDR);
1134 					break;
1135 				case IP_RECVIF:
1136 					OPTSET(INP_RECVIF);
1137 					break;
1138 				case IP_RECVTTL:
1139 					OPTSET(INP_RECVTTL);
1140 					break;
1141 				case IP_RECVDSTPORT:
1142 					OPTSET(INP_RECVDSTPORT);
1143 					break;
1144 				}
1145 			}
1146 			break;
1147 #undef OPTSET
1148 
1149 		case IP_MULTICAST_IF:
1150 		case IP_MULTICAST_TTL:
1151 		case IP_MULTICAST_LOOP:
1152 		case IP_ADD_MEMBERSHIP:
1153 		case IP_DROP_MEMBERSHIP:
1154 			error = ip_setmoptions(optname, &inp->inp_moptions, m,
1155 			    inp->inp_rdomain);
1156 			break;
1157 
1158 		case IP_PORTRANGE:
1159 			if (m == 0 || m->m_len != sizeof(int))
1160 				error = EINVAL;
1161 			else {
1162 				optval = *mtod(m, int *);
1163 
1164 				switch (optval) {
1165 
1166 				case IP_PORTRANGE_DEFAULT:
1167 					inp->inp_flags &= ~(INP_LOWPORT);
1168 					inp->inp_flags &= ~(INP_HIGHPORT);
1169 					break;
1170 
1171 				case IP_PORTRANGE_HIGH:
1172 					inp->inp_flags &= ~(INP_LOWPORT);
1173 					inp->inp_flags |= INP_HIGHPORT;
1174 					break;
1175 
1176 				case IP_PORTRANGE_LOW:
1177 					inp->inp_flags &= ~(INP_HIGHPORT);
1178 					inp->inp_flags |= INP_LOWPORT;
1179 					break;
1180 
1181 				default:
1182 
1183 					error = EINVAL;
1184 					break;
1185 				}
1186 			}
1187 			break;
1188 		case IP_AUTH_LEVEL:
1189 		case IP_ESP_TRANS_LEVEL:
1190 		case IP_ESP_NETWORK_LEVEL:
1191 		case IP_IPCOMP_LEVEL:
1192 #ifndef IPSEC
1193 			error = EOPNOTSUPP;
1194 #else
1195 			if (m == 0 || m->m_len != sizeof(int)) {
1196 				error = EINVAL;
1197 				break;
1198 			}
1199 			optval = *mtod(m, int *);
1200 
1201 			if (optval < IPSEC_LEVEL_BYPASS ||
1202 			    optval > IPSEC_LEVEL_UNIQUE) {
1203 				error = EINVAL;
1204 				break;
1205 			}
1206 
1207 			/* Unlink cached output TDB to force a re-search */
1208 			if (inp->inp_tdb_out) {
1209 				int s = spltdb();
1210 				TAILQ_REMOVE(&inp->inp_tdb_out->tdb_inp_out,
1211 				    inp, inp_tdb_out_next);
1212 				splx(s);
1213 			}
1214 
1215 			if (inp->inp_tdb_in) {
1216 				int s = spltdb();
1217 				TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in,
1218 				    inp, inp_tdb_in_next);
1219 				splx(s);
1220 			}
1221 
1222 			switch (optname) {
1223 			case IP_AUTH_LEVEL:
1224 				if (optval < ipsec_auth_default_level &&
1225 				    suser(p, 0)) {
1226 					error = EACCES;
1227 					break;
1228 				}
1229 				inp->inp_seclevel[SL_AUTH] = optval;
1230 				break;
1231 
1232 			case IP_ESP_TRANS_LEVEL:
1233 				if (optval < ipsec_esp_trans_default_level &&
1234 				    suser(p, 0)) {
1235 					error = EACCES;
1236 					break;
1237 				}
1238 				inp->inp_seclevel[SL_ESP_TRANS] = optval;
1239 				break;
1240 
1241 			case IP_ESP_NETWORK_LEVEL:
1242 				if (optval < ipsec_esp_network_default_level &&
1243 				    suser(p, 0)) {
1244 					error = EACCES;
1245 					break;
1246 				}
1247 				inp->inp_seclevel[SL_ESP_NETWORK] = optval;
1248 				break;
1249 			case IP_IPCOMP_LEVEL:
1250 				if (optval < ipsec_ipcomp_default_level &&
1251 				    suser(p, 0)) {
1252 					error = EACCES;
1253 					break;
1254 				}
1255 				inp->inp_seclevel[SL_IPCOMP] = optval;
1256 				break;
1257 			}
1258 			if (!error)
1259 				inp->inp_secrequire = get_sa_require(inp);
1260 #endif
1261 			break;
1262 
1263 		case IP_IPSEC_REMOTE_CRED:
1264 		case IP_IPSEC_REMOTE_AUTH:
1265 			/* Can't set the remote credential or key */
1266 			error = EOPNOTSUPP;
1267 			break;
1268 
1269 		case IP_IPSEC_LOCAL_ID:
1270 		case IP_IPSEC_REMOTE_ID:
1271 		case IP_IPSEC_LOCAL_CRED:
1272 		case IP_IPSEC_LOCAL_AUTH:
1273 #ifndef IPSEC
1274 			error = EOPNOTSUPP;
1275 #else
1276 			if (m->m_len < 2) {
1277 				error = EINVAL;
1278 				break;
1279 			}
1280 
1281 			m_copydata(m, 0, 2, (caddr_t) &opt16val);
1282 
1283 			/* If the type is 0, then we cleanup and return */
1284 			if (opt16val == 0) {
1285 				switch (optname) {
1286 				case IP_IPSEC_LOCAL_ID:
1287 					if (inp->inp_ipo != NULL &&
1288 					    inp->inp_ipo->ipo_srcid != NULL) {
1289 						ipsp_reffree(inp->inp_ipo->ipo_srcid);
1290 						inp->inp_ipo->ipo_srcid = NULL;
1291 					}
1292 					break;
1293 
1294 				case IP_IPSEC_REMOTE_ID:
1295 					if (inp->inp_ipo != NULL &&
1296 					    inp->inp_ipo->ipo_dstid != NULL) {
1297 						ipsp_reffree(inp->inp_ipo->ipo_dstid);
1298 						inp->inp_ipo->ipo_dstid = NULL;
1299 					}
1300 					break;
1301 
1302 				case IP_IPSEC_LOCAL_CRED:
1303 					if (inp->inp_ipo != NULL &&
1304 					    inp->inp_ipo->ipo_local_cred != NULL) {
1305 						ipsp_reffree(inp->inp_ipo->ipo_local_cred);
1306 						inp->inp_ipo->ipo_local_cred = NULL;
1307 					}
1308 					break;
1309 
1310 				case IP_IPSEC_LOCAL_AUTH:
1311 					if (inp->inp_ipo != NULL &&
1312 					    inp->inp_ipo->ipo_local_auth != NULL) {
1313 						ipsp_reffree(inp->inp_ipo->ipo_local_auth);
1314 						inp->inp_ipo->ipo_local_auth = NULL;
1315 					}
1316 					break;
1317 				}
1318 
1319 				error = 0;
1320 				break;
1321 			}
1322 
1323 			/* Can't have an empty payload */
1324 			if (m->m_len == 2) {
1325 				error = EINVAL;
1326 				break;
1327 			}
1328 
1329 			/* Allocate if needed */
1330 			if (inp->inp_ipo == NULL) {
1331 				inp->inp_ipo = ipsec_add_policy(inp,
1332 				    AF_INET, IPSP_DIRECTION_OUT);
1333 				if (inp->inp_ipo == NULL) {
1334 					error = ENOBUFS;
1335 					break;
1336 				}
1337 			}
1338 
1339 			ipr = malloc(sizeof(struct ipsec_ref) + m->m_len - 2,
1340 			       M_CREDENTIALS, M_NOWAIT);
1341 			if (ipr == NULL) {
1342 				error = ENOBUFS;
1343 				break;
1344 			}
1345 
1346 			ipr->ref_count = 1;
1347 			ipr->ref_malloctype = M_CREDENTIALS;
1348 			ipr->ref_len = m->m_len - 2;
1349 			ipr->ref_type = opt16val;
1350 			m_copydata(m, 2, m->m_len - 2, (caddr_t)(ipr + 1));
1351 
1352 			switch (optname) {
1353 			case IP_IPSEC_LOCAL_ID:
1354 				/* Check valid types and NUL-termination */
1355 				if (ipr->ref_type < IPSP_IDENTITY_PREFIX ||
1356 				    ipr->ref_type > IPSP_IDENTITY_CONNECTION ||
1357 				    ((char *)(ipr + 1))[ipr->ref_len - 1]) {
1358 					free(ipr, M_CREDENTIALS);
1359 					error = EINVAL;
1360 				} else {
1361 					if (inp->inp_ipo->ipo_srcid != NULL)
1362 						ipsp_reffree(inp->inp_ipo->ipo_srcid);
1363 					inp->inp_ipo->ipo_srcid = ipr;
1364 				}
1365 				break;
1366 			case IP_IPSEC_REMOTE_ID:
1367 				/* Check valid types and NUL-termination */
1368 				if (ipr->ref_type < IPSP_IDENTITY_PREFIX ||
1369 				    ipr->ref_type > IPSP_IDENTITY_CONNECTION ||
1370 				    ((char *)(ipr + 1))[ipr->ref_len - 1]) {
1371 					free(ipr, M_CREDENTIALS);
1372 					error = EINVAL;
1373 				} else {
1374 					if (inp->inp_ipo->ipo_dstid != NULL)
1375 						ipsp_reffree(inp->inp_ipo->ipo_dstid);
1376 					inp->inp_ipo->ipo_dstid = ipr;
1377 				}
1378 				break;
1379 			case IP_IPSEC_LOCAL_CRED:
1380 				if (ipr->ref_type < IPSP_CRED_KEYNOTE ||
1381 				    ipr->ref_type > IPSP_CRED_X509) {
1382 					free(ipr, M_CREDENTIALS);
1383 					error = EINVAL;
1384 				} else {
1385 					if (inp->inp_ipo->ipo_local_cred != NULL)
1386 						ipsp_reffree(inp->inp_ipo->ipo_local_cred);
1387 					inp->inp_ipo->ipo_local_cred = ipr;
1388 				}
1389 				break;
1390 			case IP_IPSEC_LOCAL_AUTH:
1391 				if (ipr->ref_type < IPSP_AUTH_PASSPHRASE ||
1392 				    ipr->ref_type > IPSP_AUTH_RSA) {
1393 					free(ipr, M_CREDENTIALS);
1394 					error = EINVAL;
1395 				} else {
1396 					if (inp->inp_ipo->ipo_local_auth != NULL)
1397 						ipsp_reffree(inp->inp_ipo->ipo_local_auth);
1398 					inp->inp_ipo->ipo_local_auth = ipr;
1399 				}
1400 				break;
1401 			}
1402 
1403 			/* Unlink cached output TDB to force a re-search */
1404 			if (inp->inp_tdb_out) {
1405 				int s = spltdb();
1406 				TAILQ_REMOVE(&inp->inp_tdb_out->tdb_inp_out,
1407 				    inp, inp_tdb_out_next);
1408 				splx(s);
1409 			}
1410 
1411 			if (inp->inp_tdb_in) {
1412 				int s = spltdb();
1413 				TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in,
1414 				    inp, inp_tdb_in_next);
1415 				splx(s);
1416 			}
1417 #endif
1418 			break;
1419 		case SO_RDOMAIN:
1420 			if (m == NULL || m->m_len < sizeof(u_int)) {
1421 				error = EINVAL;
1422 				break;
1423 			}
1424 			rtid = *mtod(m, u_int *);
1425 			if (!rtable_exists(rtid)) {
1426 				error = EINVAL;
1427 				break;
1428 			}
1429 			inp->inp_rdomain = rtid;
1430 			break;
1431 		default:
1432 			error = ENOPROTOOPT;
1433 			break;
1434 		}
1435 		if (m)
1436 			(void)m_free(m);
1437 		break;
1438 
1439 	case PRCO_GETOPT:
1440 		switch (optname) {
1441 		case IP_OPTIONS:
1442 		case IP_RETOPTS:
1443 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1444 			if (inp->inp_options) {
1445 				m->m_len = inp->inp_options->m_len;
1446 				bcopy(mtod(inp->inp_options, caddr_t),
1447 				    mtod(m, caddr_t), (unsigned)m->m_len);
1448 			} else
1449 				m->m_len = 0;
1450 			break;
1451 
1452 		case IP_TOS:
1453 		case IP_TTL:
1454 		case IP_MINTTL:
1455 		case IP_RECVOPTS:
1456 		case IP_RECVRETOPTS:
1457 		case IP_RECVDSTADDR:
1458 		case IP_RECVIF:
1459 		case IP_RECVTTL:
1460 		case IP_RECVDSTPORT:
1461 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1462 			m->m_len = sizeof(int);
1463 			switch (optname) {
1464 
1465 			case IP_TOS:
1466 				optval = inp->inp_ip.ip_tos;
1467 				break;
1468 
1469 			case IP_TTL:
1470 				optval = inp->inp_ip.ip_ttl;
1471 				break;
1472 
1473 			case IP_MINTTL:
1474 				optval = inp->inp_ip_minttl;
1475 				break;
1476 
1477 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1478 
1479 			case IP_RECVOPTS:
1480 				optval = OPTBIT(INP_RECVOPTS);
1481 				break;
1482 
1483 			case IP_RECVRETOPTS:
1484 				optval = OPTBIT(INP_RECVRETOPTS);
1485 				break;
1486 
1487 			case IP_RECVDSTADDR:
1488 				optval = OPTBIT(INP_RECVDSTADDR);
1489 				break;
1490 			case IP_RECVIF:
1491 				optval = OPTBIT(INP_RECVIF);
1492 				break;
1493 			case IP_RECVTTL:
1494 				optval = OPTBIT(INP_RECVTTL);
1495 				break;
1496 			case IP_RECVDSTPORT:
1497 				optval = OPTBIT(INP_RECVDSTPORT);
1498 				break;
1499 			}
1500 			*mtod(m, int *) = optval;
1501 			break;
1502 
1503 		case IP_MULTICAST_IF:
1504 		case IP_MULTICAST_TTL:
1505 		case IP_MULTICAST_LOOP:
1506 		case IP_ADD_MEMBERSHIP:
1507 		case IP_DROP_MEMBERSHIP:
1508 			error = ip_getmoptions(optname, inp->inp_moptions, mp);
1509 			break;
1510 
1511 		case IP_PORTRANGE:
1512 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1513 			m->m_len = sizeof(int);
1514 
1515 			if (inp->inp_flags & INP_HIGHPORT)
1516 				optval = IP_PORTRANGE_HIGH;
1517 			else if (inp->inp_flags & INP_LOWPORT)
1518 				optval = IP_PORTRANGE_LOW;
1519 			else
1520 				optval = 0;
1521 
1522 			*mtod(m, int *) = optval;
1523 			break;
1524 
1525 		case IP_AUTH_LEVEL:
1526 		case IP_ESP_TRANS_LEVEL:
1527 		case IP_ESP_NETWORK_LEVEL:
1528 		case IP_IPCOMP_LEVEL:
1529 #ifndef IPSEC
1530 			m->m_len = sizeof(int);
1531 			*mtod(m, int *) = IPSEC_LEVEL_NONE;
1532 #else
1533 			m->m_len = sizeof(int);
1534 			switch (optname) {
1535 			case IP_AUTH_LEVEL:
1536 				optval = inp->inp_seclevel[SL_AUTH];
1537 				break;
1538 
1539 			case IP_ESP_TRANS_LEVEL:
1540 				optval = inp->inp_seclevel[SL_ESP_TRANS];
1541 				break;
1542 
1543 			case IP_ESP_NETWORK_LEVEL:
1544 				optval = inp->inp_seclevel[SL_ESP_NETWORK];
1545 				break;
1546 			case IP_IPCOMP_LEVEL:
1547 				optval = inp->inp_seclevel[SL_IPCOMP];
1548 				break;
1549 			}
1550 			*mtod(m, int *) = optval;
1551 #endif
1552 			break;
1553 		case IP_IPSEC_LOCAL_ID:
1554 		case IP_IPSEC_REMOTE_ID:
1555 		case IP_IPSEC_LOCAL_CRED:
1556 		case IP_IPSEC_REMOTE_CRED:
1557 		case IP_IPSEC_LOCAL_AUTH:
1558 		case IP_IPSEC_REMOTE_AUTH:
1559 #ifndef IPSEC
1560 			error = EOPNOTSUPP;
1561 #else
1562 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1563 			m->m_len = sizeof(u_int16_t);
1564 			ipr = NULL;
1565 			switch (optname) {
1566 			case IP_IPSEC_LOCAL_ID:
1567 				if (inp->inp_ipo != NULL)
1568 					ipr = inp->inp_ipo->ipo_srcid;
1569 				opt16val = IPSP_IDENTITY_NONE;
1570 				break;
1571 			case IP_IPSEC_REMOTE_ID:
1572 				if (inp->inp_ipo != NULL)
1573 					ipr = inp->inp_ipo->ipo_dstid;
1574 				opt16val = IPSP_IDENTITY_NONE;
1575 				break;
1576 			case IP_IPSEC_LOCAL_CRED:
1577 				if (inp->inp_ipo != NULL)
1578 					ipr = inp->inp_ipo->ipo_local_cred;
1579 				opt16val = IPSP_CRED_NONE;
1580 				break;
1581 			case IP_IPSEC_REMOTE_CRED:
1582 				ipr = inp->inp_ipsec_remotecred;
1583 				opt16val = IPSP_CRED_NONE;
1584 				break;
1585 			case IP_IPSEC_LOCAL_AUTH:
1586 				if (inp->inp_ipo != NULL)
1587 					ipr = inp->inp_ipo->ipo_local_auth;
1588 				break;
1589 			case IP_IPSEC_REMOTE_AUTH:
1590 				ipr = inp->inp_ipsec_remoteauth;
1591 				break;
1592 			}
1593 			if (ipr == NULL)
1594 				*mtod(m, u_int16_t *) = opt16val;
1595 			else {
1596 				size_t len;
1597 
1598 				len = m->m_len + ipr->ref_len;
1599 				if (len > MCLBYTES) {
1600 					 m_free(m);
1601 					 error = EINVAL;
1602 					 break;
1603 				}
1604 				/* allocate mbuf cluster for larger option */
1605 				if (len > MLEN) {
1606 					 MCLGET(m, M_WAITOK);
1607 					 if ((m->m_flags & M_EXT) == 0) {
1608 						 m_free(m);
1609 						 error = ENOBUFS;
1610 						 break;
1611 					 }
1612 
1613 				}
1614 				m->m_len = len;
1615 				*mtod(m, u_int16_t *) = ipr->ref_type;
1616 				m_copyback(m, sizeof(u_int16_t), ipr->ref_len,
1617 				    ipr + 1);
1618 			}
1619 #endif
1620 			break;
1621 		case SO_RDOMAIN:
1622 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1623 			m->m_len = sizeof(u_int);
1624 			*mtod(m, u_int *) = inp->inp_rdomain;
1625 			break;
1626 		default:
1627 			error = ENOPROTOOPT;
1628 			break;
1629 		}
1630 		break;
1631 	}
1632 	return (error);
1633 }
1634 
1635 /*
1636  * Set up IP options in pcb for insertion in output packets.
1637  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1638  * with destination address if source routed.
1639  */
1640 int
1641 #ifdef notyet
1642 ip_pcbopts(optname, pcbopt, m)
1643 	int optname;
1644 #else
1645 ip_pcbopts(pcbopt, m)
1646 #endif
1647 	struct mbuf **pcbopt;
1648 	struct mbuf *m;
1649 {
1650 	int cnt, optlen;
1651 	u_char *cp;
1652 	u_char opt;
1653 
1654 	/* turn off any old options */
1655 	if (*pcbopt)
1656 		(void)m_free(*pcbopt);
1657 	*pcbopt = 0;
1658 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1659 		/*
1660 		 * Only turning off any previous options.
1661 		 */
1662 		if (m)
1663 			(void)m_free(m);
1664 		return (0);
1665 	}
1666 
1667 	if (m->m_len % sizeof(int32_t))
1668 		goto bad;
1669 
1670 	/*
1671 	 * IP first-hop destination address will be stored before
1672 	 * actual options; move other options back
1673 	 * and clear it when none present.
1674 	 */
1675 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1676 		goto bad;
1677 	cnt = m->m_len;
1678 	m->m_len += sizeof(struct in_addr);
1679 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1680 	ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1681 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1682 
1683 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1684 		opt = cp[IPOPT_OPTVAL];
1685 		if (opt == IPOPT_EOL)
1686 			break;
1687 		if (opt == IPOPT_NOP)
1688 			optlen = 1;
1689 		else {
1690 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1691 				goto bad;
1692 			optlen = cp[IPOPT_OLEN];
1693 			if (optlen < IPOPT_OLEN  + sizeof(*cp) || optlen > cnt)
1694 				goto bad;
1695 		}
1696 		switch (opt) {
1697 
1698 		default:
1699 			break;
1700 
1701 		case IPOPT_LSRR:
1702 		case IPOPT_SSRR:
1703 			/*
1704 			 * user process specifies route as:
1705 			 *	->A->B->C->D
1706 			 * D must be our final destination (but we can't
1707 			 * check that since we may not have connected yet).
1708 			 * A is first hop destination, which doesn't appear in
1709 			 * actual IP option, but is stored before the options.
1710 			 */
1711 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1712 				goto bad;
1713 			m->m_len -= sizeof(struct in_addr);
1714 			cnt -= sizeof(struct in_addr);
1715 			optlen -= sizeof(struct in_addr);
1716 			cp[IPOPT_OLEN] = optlen;
1717 			/*
1718 			 * Move first hop before start of options.
1719 			 */
1720 			bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1721 			    sizeof(struct in_addr));
1722 			/*
1723 			 * Then copy rest of options back
1724 			 * to close up the deleted entry.
1725 			 */
1726 			ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1727 			    sizeof(struct in_addr)),
1728 			    (caddr_t)&cp[IPOPT_OFFSET+1],
1729 			    (unsigned)cnt - (IPOPT_OFFSET+1));
1730 			break;
1731 		}
1732 	}
1733 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1734 		goto bad;
1735 	*pcbopt = m;
1736 	return (0);
1737 
1738 bad:
1739 	(void)m_free(m);
1740 	return (EINVAL);
1741 }
1742 
1743 /*
1744  * Set the IP multicast options in response to user setsockopt().
1745  */
1746 int
1747 ip_setmoptions(int optname, struct ip_moptions **imop, struct mbuf *m,
1748     u_int rdomain)
1749 {
1750 	int error = 0;
1751 	u_char loop;
1752 	int i;
1753 	struct in_addr addr;
1754 	struct ip_mreq *mreq;
1755 	struct ifnet *ifp;
1756 	struct ip_moptions *imo = *imop;
1757 	struct in_multi **immp;
1758 	struct route ro;
1759 	struct sockaddr_in *dst;
1760 
1761 	if (imo == NULL) {
1762 		/*
1763 		 * No multicast option buffer attached to the pcb;
1764 		 * allocate one and initialize to default values.
1765 		 */
1766 		imo = (struct ip_moptions *)malloc(sizeof(*imo), M_IPMOPTS,
1767 		    M_WAITOK|M_ZERO);
1768 		immp = (struct in_multi **)malloc(
1769 		    (sizeof(*immp) * IP_MIN_MEMBERSHIPS), M_IPMOPTS,
1770 		    M_WAITOK|M_ZERO);
1771 		*imop = imo;
1772 		imo->imo_multicast_ifp = NULL;
1773 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1774 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1775 		imo->imo_num_memberships = 0;
1776 		imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1777 		imo->imo_membership = immp;
1778 	}
1779 
1780 	switch (optname) {
1781 
1782 	case IP_MULTICAST_IF:
1783 		/*
1784 		 * Select the interface for outgoing multicast packets.
1785 		 */
1786 		if (m == NULL || m->m_len != sizeof(struct in_addr)) {
1787 			error = EINVAL;
1788 			break;
1789 		}
1790 		addr = *(mtod(m, struct in_addr *));
1791 		/*
1792 		 * INADDR_ANY is used to remove a previous selection.
1793 		 * When no interface is selected, a default one is
1794 		 * chosen every time a multicast packet is sent.
1795 		 */
1796 		if (addr.s_addr == INADDR_ANY) {
1797 			imo->imo_multicast_ifp = NULL;
1798 			break;
1799 		}
1800 		/*
1801 		 * The selected interface is identified by its local
1802 		 * IP address.  Find the interface and confirm that
1803 		 * it supports multicasting.
1804 		 */
1805 		INADDR_TO_IFP(addr, ifp, rdomain);
1806 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1807 			error = EADDRNOTAVAIL;
1808 			break;
1809 		}
1810 		imo->imo_multicast_ifp = ifp;
1811 		break;
1812 
1813 	case IP_MULTICAST_TTL:
1814 		/*
1815 		 * Set the IP time-to-live for outgoing multicast packets.
1816 		 */
1817 		if (m == NULL || m->m_len != 1) {
1818 			error = EINVAL;
1819 			break;
1820 		}
1821 		imo->imo_multicast_ttl = *(mtod(m, u_char *));
1822 		break;
1823 
1824 	case IP_MULTICAST_LOOP:
1825 		/*
1826 		 * Set the loopback flag for outgoing multicast packets.
1827 		 * Must be zero or one.
1828 		 */
1829 		if (m == NULL || m->m_len != 1 ||
1830 		   (loop = *(mtod(m, u_char *))) > 1) {
1831 			error = EINVAL;
1832 			break;
1833 		}
1834 		imo->imo_multicast_loop = loop;
1835 		break;
1836 
1837 	case IP_ADD_MEMBERSHIP:
1838 		/*
1839 		 * Add a multicast group membership.
1840 		 * Group must be a valid IP multicast address.
1841 		 */
1842 		if (m == NULL || m->m_len != sizeof(struct ip_mreq)) {
1843 			error = EINVAL;
1844 			break;
1845 		}
1846 		mreq = mtod(m, struct ip_mreq *);
1847 		if (!IN_MULTICAST(mreq->imr_multiaddr.s_addr)) {
1848 			error = EINVAL;
1849 			break;
1850 		}
1851 		/*
1852 		 * If no interface address was provided, use the interface of
1853 		 * the route to the given multicast address.
1854 		 */
1855 		if (mreq->imr_interface.s_addr == INADDR_ANY) {
1856 			ro.ro_rt = NULL;
1857 			dst = satosin(&ro.ro_dst);
1858 			dst->sin_len = sizeof(*dst);
1859 			dst->sin_family = AF_INET;
1860 			dst->sin_addr = mreq->imr_multiaddr;
1861 			if (!(ro.ro_rt && ro.ro_rt->rt_ifp &&
1862 			    (ro.ro_rt->rt_flags & RTF_UP)))
1863 				ro.ro_rt = rtalloc1(&ro.ro_dst, 1, rdomain);
1864 			if (ro.ro_rt == NULL) {
1865 				error = EADDRNOTAVAIL;
1866 				break;
1867 			}
1868 			ifp = ro.ro_rt->rt_ifp;
1869 			rtfree(ro.ro_rt);
1870 		} else {
1871 			INADDR_TO_IFP(mreq->imr_interface, ifp, rdomain);
1872 		}
1873 		/*
1874 		 * See if we found an interface, and confirm that it
1875 		 * supports multicast.
1876 		 */
1877 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1878 			error = EADDRNOTAVAIL;
1879 			break;
1880 		}
1881 		/*
1882 		 * See if the membership already exists or if all the
1883 		 * membership slots are full.
1884 		 */
1885 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1886 			if (imo->imo_membership[i]->inm_ia->ia_ifp == ifp &&
1887 			    imo->imo_membership[i]->inm_addr.s_addr
1888 						== mreq->imr_multiaddr.s_addr)
1889 				break;
1890 		}
1891 		if (i < imo->imo_num_memberships) {
1892 			error = EADDRINUSE;
1893 			break;
1894 		}
1895 		if (imo->imo_num_memberships == imo->imo_max_memberships) {
1896 			struct in_multi **nmships, **omships;
1897 			size_t newmax;
1898 			/*
1899 			 * Resize the vector to next power-of-two minus 1. If the
1900 			 * size would exceed the maximum then we know we've really
1901 			 * run out of entries. Otherwise, we reallocate the vector.
1902 			 */
1903 			nmships = NULL;
1904 			omships = imo->imo_membership;
1905 			newmax = ((imo->imo_max_memberships + 1) * 2) - 1;
1906 			if (newmax <= IP_MAX_MEMBERSHIPS) {
1907 				nmships = (struct in_multi **)malloc(
1908 				    sizeof(*nmships) * newmax, M_IPMOPTS,
1909 				    M_NOWAIT|M_ZERO);
1910 				if (nmships != NULL) {
1911 					bcopy(omships, nmships,
1912 					    sizeof(*omships) *
1913 					    imo->imo_max_memberships);
1914 					free(omships, M_IPMOPTS);
1915 					imo->imo_membership = nmships;
1916 					imo->imo_max_memberships = newmax;
1917 				}
1918 			}
1919 			if (nmships == NULL) {
1920 				error = ETOOMANYREFS;
1921 				break;
1922 			}
1923 		}
1924 		/*
1925 		 * Everything looks good; add a new record to the multicast
1926 		 * address list for the given interface.
1927 		 */
1928 		if ((imo->imo_membership[i] =
1929 		    in_addmulti(&mreq->imr_multiaddr, ifp)) == NULL) {
1930 			error = ENOBUFS;
1931 			break;
1932 		}
1933 		++imo->imo_num_memberships;
1934 		break;
1935 
1936 	case IP_DROP_MEMBERSHIP:
1937 		/*
1938 		 * Drop a multicast group membership.
1939 		 * Group must be a valid IP multicast address.
1940 		 */
1941 		if (m == NULL || m->m_len != sizeof(struct ip_mreq)) {
1942 			error = EINVAL;
1943 			break;
1944 		}
1945 		mreq = mtod(m, struct ip_mreq *);
1946 		if (!IN_MULTICAST(mreq->imr_multiaddr.s_addr)) {
1947 			error = EINVAL;
1948 			break;
1949 		}
1950 		/*
1951 		 * If an interface address was specified, get a pointer
1952 		 * to its ifnet structure.
1953 		 */
1954 		if (mreq->imr_interface.s_addr == INADDR_ANY)
1955 			ifp = NULL;
1956 		else {
1957 			INADDR_TO_IFP(mreq->imr_interface, ifp, rdomain);
1958 			if (ifp == NULL) {
1959 				error = EADDRNOTAVAIL;
1960 				break;
1961 			}
1962 		}
1963 		/*
1964 		 * Find the membership in the membership array.
1965 		 */
1966 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1967 			if ((ifp == NULL ||
1968 			     imo->imo_membership[i]->inm_ia->ia_ifp == ifp) &&
1969 			     imo->imo_membership[i]->inm_addr.s_addr ==
1970 			     mreq->imr_multiaddr.s_addr)
1971 				break;
1972 		}
1973 		if (i == imo->imo_num_memberships) {
1974 			error = EADDRNOTAVAIL;
1975 			break;
1976 		}
1977 		/*
1978 		 * Give up the multicast address record to which the
1979 		 * membership points.
1980 		 */
1981 		in_delmulti(imo->imo_membership[i]);
1982 		/*
1983 		 * Remove the gap in the membership array.
1984 		 */
1985 		for (++i; i < imo->imo_num_memberships; ++i)
1986 			imo->imo_membership[i-1] = imo->imo_membership[i];
1987 		--imo->imo_num_memberships;
1988 		break;
1989 
1990 	default:
1991 		error = EOPNOTSUPP;
1992 		break;
1993 	}
1994 
1995 	/*
1996 	 * If all options have default values, no need to keep the data.
1997 	 */
1998 	if (imo->imo_multicast_ifp == NULL &&
1999 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2000 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2001 	    imo->imo_num_memberships == 0) {
2002 		free(imo->imo_membership , M_IPMOPTS);
2003 		free(*imop, M_IPMOPTS);
2004 		*imop = NULL;
2005 	}
2006 
2007 	return (error);
2008 }
2009 
2010 /*
2011  * Return the IP multicast options in response to user getsockopt().
2012  */
2013 int
2014 ip_getmoptions(optname, imo, mp)
2015 	int optname;
2016 	struct ip_moptions *imo;
2017 	struct mbuf **mp;
2018 {
2019 	u_char *ttl;
2020 	u_char *loop;
2021 	struct in_addr *addr;
2022 	struct in_ifaddr *ia;
2023 
2024 	*mp = m_get(M_WAIT, MT_SOOPTS);
2025 
2026 	switch (optname) {
2027 
2028 	case IP_MULTICAST_IF:
2029 		addr = mtod(*mp, struct in_addr *);
2030 		(*mp)->m_len = sizeof(struct in_addr);
2031 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2032 			addr->s_addr = INADDR_ANY;
2033 		else {
2034 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2035 			addr->s_addr = (ia == NULL) ? INADDR_ANY
2036 					: ia->ia_addr.sin_addr.s_addr;
2037 		}
2038 		return (0);
2039 
2040 	case IP_MULTICAST_TTL:
2041 		ttl = mtod(*mp, u_char *);
2042 		(*mp)->m_len = 1;
2043 		*ttl = (imo == NULL) ? IP_DEFAULT_MULTICAST_TTL
2044 				     : imo->imo_multicast_ttl;
2045 		return (0);
2046 
2047 	case IP_MULTICAST_LOOP:
2048 		loop = mtod(*mp, u_char *);
2049 		(*mp)->m_len = 1;
2050 		*loop = (imo == NULL) ? IP_DEFAULT_MULTICAST_LOOP
2051 				      : imo->imo_multicast_loop;
2052 		return (0);
2053 
2054 	default:
2055 		return (EOPNOTSUPP);
2056 	}
2057 }
2058 
2059 /*
2060  * Discard the IP multicast options.
2061  */
2062 void
2063 ip_freemoptions(imo)
2064 	struct ip_moptions *imo;
2065 {
2066 	int i;
2067 
2068 	if (imo != NULL) {
2069 		for (i = 0; i < imo->imo_num_memberships; ++i)
2070 			in_delmulti(imo->imo_membership[i]);
2071 		free(imo->imo_membership, M_IPMOPTS);
2072 		free(imo, M_IPMOPTS);
2073 	}
2074 }
2075 
2076 /*
2077  * Routine called from ip_output() to loop back a copy of an IP multicast
2078  * packet to the input queue of a specified interface.  Note that this
2079  * calls the output routine of the loopback "driver", but with an interface
2080  * pointer that might NOT be &loif -- easier than replicating that code here.
2081  */
2082 void
2083 ip_mloopback(ifp, m, dst)
2084 	struct ifnet *ifp;
2085 	struct mbuf *m;
2086 	struct sockaddr_in *dst;
2087 {
2088 	struct ip *ip;
2089 	struct mbuf *copym;
2090 
2091 	copym = m_copym2(m, 0, M_COPYALL, M_DONTWAIT);
2092 	if (copym != NULL) {
2093 		/*
2094 		 * We don't bother to fragment if the IP length is greater
2095 		 * than the interface's MTU.  Can this possibly matter?
2096 		 */
2097 		ip = mtod(copym, struct ip *);
2098 		ip->ip_sum = 0;
2099 		ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
2100 		(void) looutput(ifp, copym, sintosa(dst), NULL);
2101 	}
2102 }
2103 
2104 /*
2105  * Process a delayed payload checksum calculation.
2106  */
2107 void
2108 in_delayed_cksum(struct mbuf *m)
2109 {
2110 	struct ip *ip;
2111 	u_int16_t csum, offset;
2112 
2113 	ip = mtod(m, struct ip *);
2114 	offset = ip->ip_hl << 2;
2115 	csum = in4_cksum(m, 0, offset, m->m_pkthdr.len - offset);
2116 	if (csum == 0 && ip->ip_p == IPPROTO_UDP)
2117 		csum = 0xffff;
2118 
2119 	switch (ip->ip_p) {
2120 	case IPPROTO_TCP:
2121 		offset += offsetof(struct tcphdr, th_sum);
2122 		break;
2123 
2124 	case IPPROTO_UDP:
2125 		offset += offsetof(struct udphdr, uh_sum);
2126 		break;
2127 
2128 	default:
2129 		return;
2130 	}
2131 
2132 	if ((offset + sizeof(u_int16_t)) > m->m_len)
2133 		m_copyback(m, offset, sizeof(csum), &csum);
2134 	else
2135 		*(u_int16_t *)(mtod(m, caddr_t) + offset) = csum;
2136 }
2137