xref: /netbsd-src/sys/netinet/ip_output.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: ip_output.c,v 1.185 2007/11/28 04:14:11 dyoung Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * 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. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1998 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Public Access Networks Corporation ("Panix").  It was developed under
38  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed by the NetBSD
51  *	Foundation, Inc. and its contributors.
52  * 4. Neither the name of The NetBSD Foundation nor the names of its
53  *    contributors may be used to endorse or promote products derived
54  *    from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 /*
70  * Copyright (c) 1982, 1986, 1988, 1990, 1993
71  *	The Regents of the University of California.  All rights reserved.
72  *
73  * Redistribution and use in source and binary forms, with or without
74  * modification, are permitted provided that the following conditions
75  * are met:
76  * 1. Redistributions of source code must retain the above copyright
77  *    notice, this list of conditions and the following disclaimer.
78  * 2. Redistributions in binary form must reproduce the above copyright
79  *    notice, this list of conditions and the following disclaimer in the
80  *    documentation and/or other materials provided with the distribution.
81  * 3. Neither the name of the University nor the names of its contributors
82  *    may be used to endorse or promote products derived from this software
83  *    without specific prior written permission.
84  *
85  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
86  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
87  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
88  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
89  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
90  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
91  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
92  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
93  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
94  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
95  * SUCH DAMAGE.
96  *
97  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
98  */
99 
100 #include <sys/cdefs.h>
101 __KERNEL_RCSID(0, "$NetBSD: ip_output.c,v 1.185 2007/11/28 04:14:11 dyoung Exp $");
102 
103 #include "opt_pfil_hooks.h"
104 #include "opt_inet.h"
105 #include "opt_ipsec.h"
106 #include "opt_mrouting.h"
107 
108 #include <sys/param.h>
109 #include <sys/malloc.h>
110 #include <sys/mbuf.h>
111 #include <sys/errno.h>
112 #include <sys/protosw.h>
113 #include <sys/socket.h>
114 #include <sys/socketvar.h>
115 #include <sys/kauth.h>
116 #ifdef FAST_IPSEC
117 #include <sys/domain.h>
118 #endif
119 #include <sys/systm.h>
120 #include <sys/proc.h>
121 
122 #include <net/if.h>
123 #include <net/route.h>
124 #include <net/pfil.h>
125 
126 #include <netinet/in.h>
127 #include <netinet/in_systm.h>
128 #include <netinet/ip.h>
129 #include <netinet/in_pcb.h>
130 #include <netinet/in_var.h>
131 #include <netinet/ip_var.h>
132 #include <netinet/in_offload.h>
133 
134 #ifdef MROUTING
135 #include <netinet/ip_mroute.h>
136 #endif
137 
138 #include <machine/stdarg.h>
139 
140 #ifdef IPSEC
141 #include <netinet6/ipsec.h>
142 #include <netkey/key.h>
143 #include <netkey/key_debug.h>
144 #endif /*IPSEC*/
145 
146 #ifdef FAST_IPSEC
147 #include <netipsec/ipsec.h>
148 #include <netipsec/key.h>
149 #include <netipsec/xform.h>
150 #endif	/* FAST_IPSEC*/
151 
152 #ifdef IPSEC_NAT_T
153 #include <netinet/udp.h>
154 #endif
155 
156 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
157 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
158 static void ip_mloopback(struct ifnet *, struct mbuf *,
159     const struct sockaddr_in *);
160 static int ip_getoptval(struct mbuf *, u_int8_t *, u_int);
161 
162 #ifdef PFIL_HOOKS
163 extern struct pfil_head inet_pfil_hook;			/* XXX */
164 #endif
165 
166 int	ip_do_loopback_cksum = 0;
167 
168 /*
169  * IP output.  The packet in mbuf chain m contains a skeletal IP
170  * header (with len, off, ttl, proto, tos, src, dst).
171  * The mbuf chain containing the packet will be freed.
172  * The mbuf opt, if present, will not be freed.
173  */
174 int
175 ip_output(struct mbuf *m0, ...)
176 {
177 	struct ip *ip;
178 	struct ifnet *ifp;
179 	struct mbuf *m = m0;
180 	int hlen = sizeof (struct ip);
181 	int len, error = 0;
182 	struct route iproute;
183 	const struct sockaddr_in *dst;
184 	struct in_ifaddr *ia;
185 	struct ifaddr *xifa;
186 	struct mbuf *opt;
187 	struct route *ro;
188 	int flags, sw_csum;
189 	int *mtu_p;
190 	u_long mtu;
191 	struct ip_moptions *imo;
192 	struct socket *so;
193 	va_list ap;
194 #ifdef IPSEC_NAT_T
195 	int natt_frag = 0;
196 #endif
197 #ifdef IPSEC
198 	struct secpolicy *sp = NULL;
199 #endif /*IPSEC*/
200 #ifdef FAST_IPSEC
201 	struct inpcb *inp;
202 	struct m_tag *mtag;
203 	struct secpolicy *sp = NULL;
204 	struct tdb_ident *tdbi;
205 	int s;
206 #endif
207 	u_int16_t ip_len;
208 	union {
209 		struct sockaddr		dst;
210 		struct sockaddr_in	dst4;
211 	} u;
212 	struct sockaddr *rdst = &u.dst;	/* real IP destination, as opposed
213 					 * to the nexthop
214 					 */
215 
216 	len = 0;
217 	va_start(ap, m0);
218 	opt = va_arg(ap, struct mbuf *);
219 	ro = va_arg(ap, struct route *);
220 	flags = va_arg(ap, int);
221 	imo = va_arg(ap, struct ip_moptions *);
222 	so = va_arg(ap, struct socket *);
223 	if (flags & IP_RETURNMTU)
224 		mtu_p = va_arg(ap, int *);
225 	else
226 		mtu_p = NULL;
227 	va_end(ap);
228 
229 	MCLAIM(m, &ip_tx_mowner);
230 #ifdef FAST_IPSEC
231 	if (so != NULL && so->so_proto->pr_domain->dom_family == AF_INET)
232 		inp = (struct inpcb *)so->so_pcb;
233 	else
234 		inp = NULL;
235 #endif /* FAST_IPSEC */
236 
237 #ifdef	DIAGNOSTIC
238 	if ((m->m_flags & M_PKTHDR) == 0)
239 		panic("ip_output: no HDR");
240 
241 	if ((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) != 0) {
242 		panic("ip_output: IPv6 checksum offload flags: %d",
243 		    m->m_pkthdr.csum_flags);
244 	}
245 
246 	if ((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) ==
247 	    (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
248 		panic("ip_output: conflicting checksum offload flags: %d",
249 		    m->m_pkthdr.csum_flags);
250 	}
251 #endif
252 	if (opt) {
253 		m = ip_insertoptions(m, opt, &len);
254 		if (len >= sizeof(struct ip))
255 			hlen = len;
256 	}
257 	ip = mtod(m, struct ip *);
258 	/*
259 	 * Fill in IP header.
260 	 */
261 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
262 		ip->ip_v = IPVERSION;
263 		ip->ip_off = htons(0);
264 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
265 			ip->ip_id = ip_newid();
266 		} else {
267 
268 			/*
269 			 * TSO capable interfaces (typically?) increment
270 			 * ip_id for each segment.
271 			 * "allocate" enough ids here to increase the chance
272 			 * for them to be unique.
273 			 *
274 			 * note that the following calculation is not
275 			 * needed to be precise.  wasting some ip_id is fine.
276 			 */
277 
278 			unsigned int segsz = m->m_pkthdr.segsz;
279 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
280 			unsigned int num = howmany(datasz, segsz);
281 
282 			ip->ip_id = ip_newid_range(num);
283 		}
284 		ip->ip_hl = hlen >> 2;
285 		ipstat.ips_localout++;
286 	} else {
287 		hlen = ip->ip_hl << 2;
288 	}
289 	/*
290 	 * Route packet.
291 	 */
292 	memset(&iproute, 0, sizeof(iproute));
293 	if (ro == NULL)
294 		ro = &iproute;
295 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
296 	dst = satocsin(rtcache_getdst(ro));
297 	/*
298 	 * If there is a cached route,
299 	 * check that it is to the same destination
300 	 * and is still up.  If not, free it and try again.
301 	 * The address family should also be checked in case of sharing the
302 	 * cache with IPv6.
303 	 */
304 	if (dst == NULL)
305 		;
306 	else if (dst->sin_family != AF_INET ||
307 		 !in_hosteq(dst->sin_addr, ip->ip_dst))
308 		rtcache_free(ro);
309 	else
310 		rtcache_check(ro);
311 	if (ro->ro_rt == NULL) {
312 		dst = &u.dst4;
313 		rtcache_setdst(ro, &u.dst);
314 	}
315 	/*
316 	 * If routing to interface only,
317 	 * short circuit routing lookup.
318 	 */
319 	if (flags & IP_ROUTETOIF) {
320 		if ((ia = ifatoia(ifa_ifwithladdr(sintocsa(dst)))) == NULL) {
321 			ipstat.ips_noroute++;
322 			error = ENETUNREACH;
323 			goto bad;
324 		}
325 		ifp = ia->ia_ifp;
326 		mtu = ifp->if_mtu;
327 		ip->ip_ttl = 1;
328 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
329 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
330 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
331 		ifp = imo->imo_multicast_ifp;
332 		mtu = ifp->if_mtu;
333 		IFP_TO_IA(ifp, ia);
334 	} else {
335 		if (ro->ro_rt == NULL)
336 			rtcache_init(ro);
337 		if (ro->ro_rt == NULL) {
338 			ipstat.ips_noroute++;
339 			error = EHOSTUNREACH;
340 			goto bad;
341 		}
342 		ia = ifatoia(ro->ro_rt->rt_ifa);
343 		ifp = ro->ro_rt->rt_ifp;
344 		if ((mtu = ro->ro_rt->rt_rmx.rmx_mtu) == 0)
345 			mtu = ifp->if_mtu;
346 		ro->ro_rt->rt_use++;
347 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
348 			dst = satosin(ro->ro_rt->rt_gateway);
349 	}
350 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
351 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
352 		struct in_multi *inm;
353 
354 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
355 			M_BCAST : M_MCAST;
356 		/*
357 		 * See if the caller provided any multicast options
358 		 */
359 		if (imo != NULL)
360 			ip->ip_ttl = imo->imo_multicast_ttl;
361 		else
362 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
363 
364 		/*
365 		 * if we don't know the outgoing ifp yet, we can't generate
366 		 * output
367 		 */
368 		if (!ifp) {
369 			ipstat.ips_noroute++;
370 			error = ENETUNREACH;
371 			goto bad;
372 		}
373 
374 		/*
375 		 * If the packet is multicast or broadcast, confirm that
376 		 * the outgoing interface can transmit it.
377 		 */
378 		if (((m->m_flags & M_MCAST) &&
379 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
380 		    ((m->m_flags & M_BCAST) &&
381 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
382 			ipstat.ips_noroute++;
383 			error = ENETUNREACH;
384 			goto bad;
385 		}
386 		/*
387 		 * If source address not specified yet, use an address
388 		 * of outgoing interface.
389 		 */
390 		if (in_nullhost(ip->ip_src)) {
391 			struct in_ifaddr *xia;
392 
393 			IFP_TO_IA(ifp, xia);
394 			if (!xia) {
395 				error = EADDRNOTAVAIL;
396 				goto bad;
397 			}
398 			xifa = &xia->ia_ifa;
399 			if (xifa->ifa_getifa != NULL) {
400 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
401 			}
402 			ip->ip_src = xia->ia_addr.sin_addr;
403 		}
404 
405 		IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm);
406 		if (inm != NULL &&
407 		   (imo == NULL || imo->imo_multicast_loop)) {
408 			/*
409 			 * If we belong to the destination multicast group
410 			 * on the outgoing interface, and the caller did not
411 			 * forbid loopback, loop back a copy.
412 			 */
413 			ip_mloopback(ifp, m, &u.dst4);
414 		}
415 #ifdef MROUTING
416 		else {
417 			/*
418 			 * If we are acting as a multicast router, perform
419 			 * multicast forwarding as if the packet had just
420 			 * arrived on the interface to which we are about
421 			 * to send.  The multicast forwarding function
422 			 * recursively calls this function, using the
423 			 * IP_FORWARDING flag to prevent infinite recursion.
424 			 *
425 			 * Multicasts that are looped back by ip_mloopback(),
426 			 * above, will be forwarded by the ip_input() routine,
427 			 * if necessary.
428 			 */
429 			extern struct socket *ip_mrouter;
430 
431 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
432 				if (ip_mforward(m, ifp) != 0) {
433 					m_freem(m);
434 					goto done;
435 				}
436 			}
437 		}
438 #endif
439 		/*
440 		 * Multicasts with a time-to-live of zero may be looped-
441 		 * back, above, but must not be transmitted on a network.
442 		 * Also, multicasts addressed to the loopback interface
443 		 * are not sent -- the above call to ip_mloopback() will
444 		 * loop back a copy if this host actually belongs to the
445 		 * destination group on the loopback interface.
446 		 */
447 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
448 			m_freem(m);
449 			goto done;
450 		}
451 
452 		goto sendit;
453 	}
454 	/*
455 	 * If source address not specified yet, use address
456 	 * of outgoing interface.
457 	 */
458 	if (in_nullhost(ip->ip_src)) {
459 		xifa = &ia->ia_ifa;
460 		if (xifa->ifa_getifa != NULL)
461 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
462 		ip->ip_src = ia->ia_addr.sin_addr;
463 	}
464 
465 	/*
466 	 * packets with Class-D address as source are not valid per
467 	 * RFC 1112
468 	 */
469 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
470 		ipstat.ips_odropped++;
471 		error = EADDRNOTAVAIL;
472 		goto bad;
473 	}
474 
475 	/*
476 	 * Look for broadcast address and
477 	 * and verify user is allowed to send
478 	 * such a packet.
479 	 */
480 	if (in_broadcast(dst->sin_addr, ifp)) {
481 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
482 			error = EADDRNOTAVAIL;
483 			goto bad;
484 		}
485 		if ((flags & IP_ALLOWBROADCAST) == 0) {
486 			error = EACCES;
487 			goto bad;
488 		}
489 		/* don't allow broadcast messages to be fragmented */
490 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
491 			error = EMSGSIZE;
492 			goto bad;
493 		}
494 		m->m_flags |= M_BCAST;
495 	} else
496 		m->m_flags &= ~M_BCAST;
497 
498 sendit:
499 	/*
500 	 * If we're doing Path MTU Discovery, we need to set DF unless
501 	 * the route's MTU is locked.
502 	 */
503 	if ((flags & IP_MTUDISC) != 0 && ro->ro_rt != NULL &&
504 	    (ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
505 		ip->ip_off |= htons(IP_DF);
506 
507 	/* Remember the current ip_len */
508 	ip_len = ntohs(ip->ip_len);
509 
510 #ifdef IPSEC
511 	/* get SP for this packet */
512 	if (so == NULL)
513 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND,
514 		    flags, &error);
515 	else {
516 		if (IPSEC_PCB_SKIP_IPSEC(sotoinpcb_hdr(so)->inph_sp,
517 					 IPSEC_DIR_OUTBOUND))
518 			goto skip_ipsec;
519 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
520 	}
521 
522 	if (sp == NULL) {
523 		ipsecstat.out_inval++;
524 		goto bad;
525 	}
526 
527 	error = 0;
528 
529 	/* check policy */
530 	switch (sp->policy) {
531 	case IPSEC_POLICY_DISCARD:
532 		/*
533 		 * This packet is just discarded.
534 		 */
535 		ipsecstat.out_polvio++;
536 		goto bad;
537 
538 	case IPSEC_POLICY_BYPASS:
539 	case IPSEC_POLICY_NONE:
540 		/* no need to do IPsec. */
541 		goto skip_ipsec;
542 
543 	case IPSEC_POLICY_IPSEC:
544 		if (sp->req == NULL) {
545 			/* XXX should be panic ? */
546 			printf("ip_output: No IPsec request specified.\n");
547 			error = EINVAL;
548 			goto bad;
549 		}
550 		break;
551 
552 	case IPSEC_POLICY_ENTRUST:
553 	default:
554 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
555 	}
556 
557 #ifdef IPSEC_NAT_T
558 	/*
559 	 * NAT-T ESP fragmentation: don't do IPSec processing now,
560 	 * we'll do it on each fragmented packet.
561 	 */
562 	if (sp->req->sav &&
563 	    ((sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP) ||
564 	     (sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP_NON_IKE))) {
565 		if (ntohs(ip->ip_len) > sp->req->sav->esp_frag) {
566 			natt_frag = 1;
567 			mtu = sp->req->sav->esp_frag;
568 			goto skip_ipsec;
569 		}
570 	}
571 #endif /* IPSEC_NAT_T */
572 
573 	/*
574 	 * ipsec4_output() expects ip_len and ip_off in network
575 	 * order.  They have been set to network order above.
576 	 */
577 
578     {
579 	struct ipsec_output_state state;
580 	bzero(&state, sizeof(state));
581 	state.m = m;
582 	if (flags & IP_ROUTETOIF) {
583 		state.ro = &iproute;
584 		memset(&iproute, 0, sizeof(iproute));
585 	} else
586 		state.ro = ro;
587 	state.dst = sintocsa(dst);
588 
589 	/*
590 	 * We can't defer the checksum of payload data if
591 	 * we're about to encrypt/authenticate it.
592 	 *
593 	 * XXX When we support crypto offloading functions of
594 	 * XXX network interfaces, we need to reconsider this,
595 	 * XXX since it's likely that they'll support checksumming,
596 	 * XXX as well.
597 	 */
598 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
599 		in_delayed_cksum(m);
600 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
601 	}
602 
603 	error = ipsec4_output(&state, sp, flags);
604 
605 	m = state.m;
606 	if (flags & IP_ROUTETOIF) {
607 		/*
608 		 * if we have tunnel mode SA, we may need to ignore
609 		 * IP_ROUTETOIF.
610 		 */
611 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
612 			flags &= ~IP_ROUTETOIF;
613 			ro = state.ro;
614 		}
615 	} else
616 		ro = state.ro;
617 	dst = satocsin(state.dst);
618 	if (error) {
619 		/* mbuf is already reclaimed in ipsec4_output. */
620 		m0 = NULL;
621 		switch (error) {
622 		case EHOSTUNREACH:
623 		case ENETUNREACH:
624 		case EMSGSIZE:
625 		case ENOBUFS:
626 		case ENOMEM:
627 			break;
628 		default:
629 			printf("ip4_output (ipsec): error code %d\n", error);
630 			/*fall through*/
631 		case ENOENT:
632 			/* don't show these error codes to the user */
633 			error = 0;
634 			break;
635 		}
636 		goto bad;
637 	}
638 
639 	/* be sure to update variables that are affected by ipsec4_output() */
640 	ip = mtod(m, struct ip *);
641 	hlen = ip->ip_hl << 2;
642 	ip_len = ntohs(ip->ip_len);
643 
644 	if (ro->ro_rt == NULL) {
645 		if ((flags & IP_ROUTETOIF) == 0) {
646 			printf("ip_output: "
647 				"can't update route after IPsec processing\n");
648 			error = EHOSTUNREACH;	/*XXX*/
649 			goto bad;
650 		}
651 	} else {
652 		/* nobody uses ia beyond here */
653 		if (state.encap) {
654 			ifp = ro->ro_rt->rt_ifp;
655 			if ((mtu = ro->ro_rt->rt_rmx.rmx_mtu) == 0)
656 				mtu = ifp->if_mtu;
657 		}
658 	}
659     }
660 skip_ipsec:
661 #endif /*IPSEC*/
662 #ifdef FAST_IPSEC
663 	/*
664 	 * Check the security policy (SP) for the packet and, if
665 	 * required, do IPsec-related processing.  There are two
666 	 * cases here; the first time a packet is sent through
667 	 * it will be untagged and handled by ipsec4_checkpolicy.
668 	 * If the packet is resubmitted to ip_output (e.g. after
669 	 * AH, ESP, etc. processing), there will be a tag to bypass
670 	 * the lookup and related policy checking.
671 	 */
672 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
673 	s = splsoftnet();
674 	if (mtag != NULL) {
675 		tdbi = (struct tdb_ident *)(mtag + 1);
676 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
677 		if (sp == NULL)
678 			error = -EINVAL;	/* force silent drop */
679 		m_tag_delete(m, mtag);
680 	} else {
681 		if (inp != NULL &&
682 		    IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND))
683 			goto spd_done;
684 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
685 					&error, inp);
686 	}
687 	/*
688 	 * There are four return cases:
689 	 *    sp != NULL	 	    apply IPsec policy
690 	 *    sp == NULL, error == 0	    no IPsec handling needed
691 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
692 	 *    sp == NULL, error != 0	    discard packet, report error
693 	 */
694 	if (sp != NULL) {
695 #ifdef IPSEC_NAT_T
696 		/*
697 		 * NAT-T ESP fragmentation: don't do IPSec processing now,
698 		 * we'll do it on each fragmented packet.
699 		 */
700 		if (sp->req->sav &&
701 		    ((sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP) ||
702 		     (sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP_NON_IKE))) {
703 			if (ntohs(ip->ip_len) > sp->req->sav->esp_frag) {
704 				natt_frag = 1;
705 				mtu = sp->req->sav->esp_frag;
706 				goto spd_done;
707 			}
708 		}
709 #endif /* IPSEC_NAT_T */
710 		/* Loop detection, check if ipsec processing already done */
711 		IPSEC_ASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
712 		for (mtag = m_tag_first(m); mtag != NULL;
713 		     mtag = m_tag_next(m, mtag)) {
714 #ifdef MTAG_ABI_COMPAT
715 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
716 				continue;
717 #endif
718 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
719 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
720 				continue;
721 			/*
722 			 * Check if policy has an SA associated with it.
723 			 * This can happen when an SP has yet to acquire
724 			 * an SA; e.g. on first reference.  If it occurs,
725 			 * then we let ipsec4_process_packet do its thing.
726 			 */
727 			if (sp->req->sav == NULL)
728 				break;
729 			tdbi = (struct tdb_ident *)(mtag + 1);
730 			if (tdbi->spi == sp->req->sav->spi &&
731 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
732 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
733 				 sizeof (union sockaddr_union)) == 0) {
734 				/*
735 				 * No IPsec processing is needed, free
736 				 * reference to SP.
737 				 *
738 				 * NB: null pointer to avoid free at
739 				 *     done: below.
740 				 */
741 				KEY_FREESP(&sp), sp = NULL;
742 				splx(s);
743 				goto spd_done;
744 			}
745 		}
746 
747 		/*
748 		 * Do delayed checksums now because we send before
749 		 * this is done in the normal processing path.
750 		 */
751 		if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
752 			in_delayed_cksum(m);
753 			m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
754 		}
755 
756 #ifdef __FreeBSD__
757 		ip->ip_len = htons(ip->ip_len);
758 		ip->ip_off = htons(ip->ip_off);
759 #endif
760 
761 		/* NB: callee frees mbuf */
762 		error = ipsec4_process_packet(m, sp->req, flags, 0);
763 		/*
764 		 * Preserve KAME behaviour: ENOENT can be returned
765 		 * when an SA acquire is in progress.  Don't propagate
766 		 * this to user-level; it confuses applications.
767 		 *
768 		 * XXX this will go away when the SADB is redone.
769 		 */
770 		if (error == ENOENT)
771 			error = 0;
772 		splx(s);
773 		goto done;
774 	} else {
775 		splx(s);
776 
777 		if (error != 0) {
778 			/*
779 			 * Hack: -EINVAL is used to signal that a packet
780 			 * should be silently discarded.  This is typically
781 			 * because we asked key management for an SA and
782 			 * it was delayed (e.g. kicked up to IKE).
783 			 */
784 			if (error == -EINVAL)
785 				error = 0;
786 			goto bad;
787 		} else {
788 			/* No IPsec processing for this packet. */
789 		}
790 #ifdef notyet
791 		/*
792 		 * If deferred crypto processing is needed, check that
793 		 * the interface supports it.
794 		 */
795 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
796 		if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
797 			/* notify IPsec to do its own crypto */
798 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
799 			error = EHOSTUNREACH;
800 			goto bad;
801 		}
802 #endif
803 	}
804 spd_done:
805 #endif /* FAST_IPSEC */
806 
807 #ifdef PFIL_HOOKS
808 	/*
809 	 * Run through list of hooks for output packets.
810 	 */
811 	if ((error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT)) != 0)
812 		goto done;
813 	if (m == NULL)
814 		goto done;
815 
816 	ip = mtod(m, struct ip *);
817 	hlen = ip->ip_hl << 2;
818 	ip_len = ntohs(ip->ip_len);
819 #endif /* PFIL_HOOKS */
820 
821 	m->m_pkthdr.csum_data |= hlen << 16;
822 
823 #if IFA_STATS
824 	/*
825 	 * search for the source address structure to
826 	 * maintain output statistics.
827 	 */
828 	INADDR_TO_IA(ip->ip_src, ia);
829 #endif
830 
831 	/* Maybe skip checksums on loopback interfaces. */
832 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
833 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
834 	}
835 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
836 	/*
837 	 * If small enough for mtu of path, or if using TCP segmentation
838 	 * offload, can just send directly.
839 	 */
840 	if (ip_len <= mtu ||
841 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
842 #if IFA_STATS
843 		if (ia)
844 			ia->ia_ifa.ifa_data.ifad_outbytes += ip_len;
845 #endif
846 		/*
847 		 * Always initialize the sum to 0!  Some HW assisted
848 		 * checksumming requires this.
849 		 */
850 		ip->ip_sum = 0;
851 
852 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
853 			/*
854 			 * Perform any checksums that the hardware can't do
855 			 * for us.
856 			 *
857 			 * XXX Does any hardware require the {th,uh}_sum
858 			 * XXX fields to be 0?
859 			 */
860 			if (sw_csum & M_CSUM_IPv4) {
861 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
862 				ip->ip_sum = in_cksum(m, hlen);
863 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
864 			}
865 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
866 				if (IN_NEED_CHECKSUM(ifp,
867 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
868 					in_delayed_cksum(m);
869 				}
870 				m->m_pkthdr.csum_flags &=
871 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
872 			}
873 		}
874 
875 #ifdef IPSEC
876 		/* clean ipsec history once it goes out of the node */
877 		ipsec_delaux(m);
878 #endif
879 
880 		if (__predict_true(
881 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
882 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
883 			error =
884 			    (*ifp->if_output)(ifp, m,
885 				(m->m_flags & M_MCAST) ?
886 				    sintocsa(rdst) : sintocsa(dst),
887 				ro->ro_rt);
888 		} else {
889 			error =
890 			    ip_tso_output(ifp, m,
891 				(m->m_flags & M_MCAST) ?
892 				    sintocsa(rdst) : sintocsa(dst),
893 				ro->ro_rt);
894 		}
895 		goto done;
896 	}
897 
898 	/*
899 	 * We can't use HW checksumming if we're about to
900 	 * to fragment the packet.
901 	 *
902 	 * XXX Some hardware can do this.
903 	 */
904 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
905 		if (IN_NEED_CHECKSUM(ifp,
906 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
907 			in_delayed_cksum(m);
908 		}
909 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
910 	}
911 
912 	/*
913 	 * Too large for interface; fragment if possible.
914 	 * Must be able to put at least 8 bytes per fragment.
915 	 */
916 	if (ntohs(ip->ip_off) & IP_DF) {
917 		if (flags & IP_RETURNMTU)
918 			*mtu_p = mtu;
919 		error = EMSGSIZE;
920 		ipstat.ips_cantfrag++;
921 		goto bad;
922 	}
923 
924 	error = ip_fragment(m, ifp, mtu);
925 	if (error) {
926 		m = NULL;
927 		goto bad;
928 	}
929 
930 	for (; m; m = m0) {
931 		m0 = m->m_nextpkt;
932 		m->m_nextpkt = 0;
933 		if (error == 0) {
934 #if IFA_STATS
935 			if (ia)
936 				ia->ia_ifa.ifa_data.ifad_outbytes +=
937 				    ntohs(ip->ip_len);
938 #endif
939 #ifdef IPSEC
940 			/* clean ipsec history once it goes out of the node */
941 			ipsec_delaux(m);
942 #endif /* IPSEC */
943 
944 #ifdef IPSEC_NAT_T
945 			/*
946 			 * If we get there, the packet has not been handeld by
947 			 * IPSec whereas it should have. Now that it has been
948 			 * fragmented, re-inject it in ip_output so that IPsec
949 			 * processing can occur.
950 			 */
951 			if (natt_frag) {
952 				error = ip_output(m, opt,
953 				    ro, flags, imo, so, mtu_p);
954 			} else
955 #endif /* IPSEC_NAT_T */
956 			{
957 				KASSERT((m->m_pkthdr.csum_flags &
958 				    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
959 				error = (*ifp->if_output)(ifp, m,
960 				    (m->m_flags & M_MCAST) ?
961 					sintocsa(rdst) : sintocsa(dst),
962 				    ro->ro_rt);
963 			}
964 		} else
965 			m_freem(m);
966 	}
967 
968 	if (error == 0)
969 		ipstat.ips_fragmented++;
970 done:
971 	rtcache_free(&iproute);
972 
973 #ifdef IPSEC
974 	if (sp != NULL) {
975 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
976 			printf("DP ip_output call free SP:%p\n", sp));
977 		key_freesp(sp);
978 	}
979 #endif /* IPSEC */
980 #ifdef FAST_IPSEC
981 	if (sp != NULL)
982 		KEY_FREESP(&sp);
983 #endif /* FAST_IPSEC */
984 
985 	return (error);
986 bad:
987 	m_freem(m);
988 	goto done;
989 }
990 
991 int
992 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
993 {
994 	struct ip *ip, *mhip;
995 	struct mbuf *m0;
996 	int len, hlen, off;
997 	int mhlen, firstlen;
998 	struct mbuf **mnext;
999 	int sw_csum = m->m_pkthdr.csum_flags;
1000 	int fragments = 0;
1001 	int s;
1002 	int error = 0;
1003 
1004 	ip = mtod(m, struct ip *);
1005 	hlen = ip->ip_hl << 2;
1006 	if (ifp != NULL)
1007 		sw_csum &= ~ifp->if_csum_flags_tx;
1008 
1009 	len = (mtu - hlen) &~ 7;
1010 	if (len < 8) {
1011 		m_freem(m);
1012 		return (EMSGSIZE);
1013 	}
1014 
1015 	firstlen = len;
1016 	mnext = &m->m_nextpkt;
1017 
1018 	/*
1019 	 * Loop through length of segment after first fragment,
1020 	 * make new header and copy data of each part and link onto chain.
1021 	 */
1022 	m0 = m;
1023 	mhlen = sizeof (struct ip);
1024 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
1025 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
1026 		if (m == 0) {
1027 			error = ENOBUFS;
1028 			ipstat.ips_odropped++;
1029 			goto sendorfree;
1030 		}
1031 		MCLAIM(m, m0->m_owner);
1032 		*mnext = m;
1033 		mnext = &m->m_nextpkt;
1034 		m->m_data += max_linkhdr;
1035 		mhip = mtod(m, struct ip *);
1036 		*mhip = *ip;
1037 		/* we must inherit MCAST and BCAST flags */
1038 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
1039 		if (hlen > sizeof (struct ip)) {
1040 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1041 			mhip->ip_hl = mhlen >> 2;
1042 		}
1043 		m->m_len = mhlen;
1044 		mhip->ip_off = ((off - hlen) >> 3) +
1045 		    (ntohs(ip->ip_off) & ~IP_MF);
1046 		if (ip->ip_off & htons(IP_MF))
1047 			mhip->ip_off |= IP_MF;
1048 		if (off + len >= ntohs(ip->ip_len))
1049 			len = ntohs(ip->ip_len) - off;
1050 		else
1051 			mhip->ip_off |= IP_MF;
1052 		HTONS(mhip->ip_off);
1053 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
1054 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
1055 		if (m->m_next == 0) {
1056 			error = ENOBUFS;	/* ??? */
1057 			ipstat.ips_odropped++;
1058 			goto sendorfree;
1059 		}
1060 		m->m_pkthdr.len = mhlen + len;
1061 		m->m_pkthdr.rcvif = (struct ifnet *)0;
1062 		mhip->ip_sum = 0;
1063 		if (sw_csum & M_CSUM_IPv4) {
1064 			mhip->ip_sum = in_cksum(m, mhlen);
1065 			KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
1066 		} else {
1067 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1068 			m->m_pkthdr.csum_data |= mhlen << 16;
1069 		}
1070 		ipstat.ips_ofragments++;
1071 		fragments++;
1072 	}
1073 	/*
1074 	 * Update first fragment by trimming what's been copied out
1075 	 * and updating header, then send each fragment (in order).
1076 	 */
1077 	m = m0;
1078 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
1079 	m->m_pkthdr.len = hlen + firstlen;
1080 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
1081 	ip->ip_off |= htons(IP_MF);
1082 	ip->ip_sum = 0;
1083 	if (sw_csum & M_CSUM_IPv4) {
1084 		ip->ip_sum = in_cksum(m, hlen);
1085 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
1086 	} else {
1087 		KASSERT(m->m_pkthdr.csum_flags & M_CSUM_IPv4);
1088 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
1089 			sizeof(struct ip));
1090 	}
1091 sendorfree:
1092 	/*
1093 	 * If there is no room for all the fragments, don't queue
1094 	 * any of them.
1095 	 */
1096 	if (ifp != NULL) {
1097 		s = splnet();
1098 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
1099 		    error == 0) {
1100 			error = ENOBUFS;
1101 			ipstat.ips_odropped++;
1102 			IFQ_INC_DROPS(&ifp->if_snd);
1103 		}
1104 		splx(s);
1105 	}
1106 	if (error) {
1107 		for (m = m0; m; m = m0) {
1108 			m0 = m->m_nextpkt;
1109 			m->m_nextpkt = NULL;
1110 			m_freem(m);
1111 		}
1112 	}
1113 	return (error);
1114 }
1115 
1116 /*
1117  * Process a delayed payload checksum calculation.
1118  */
1119 void
1120 in_delayed_cksum(struct mbuf *m)
1121 {
1122 	struct ip *ip;
1123 	u_int16_t csum, offset;
1124 
1125 	ip = mtod(m, struct ip *);
1126 	offset = ip->ip_hl << 2;
1127 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
1128 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
1129 		csum = 0xffff;
1130 
1131 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
1132 
1133 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
1134 		/* This happen when ip options were inserted
1135 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
1136 		    m->m_len, offset, ip->ip_p);
1137 		 */
1138 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
1139 	} else
1140 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
1141 }
1142 
1143 /*
1144  * Determine the maximum length of the options to be inserted;
1145  * we would far rather allocate too much space rather than too little.
1146  */
1147 
1148 u_int
1149 ip_optlen(struct inpcb *inp)
1150 {
1151 	struct mbuf *m = inp->inp_options;
1152 
1153 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst))
1154 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
1155 	else
1156 		return 0;
1157 }
1158 
1159 
1160 /*
1161  * Insert IP options into preformed packet.
1162  * Adjust IP destination as required for IP source routing,
1163  * as indicated by a non-zero in_addr at the start of the options.
1164  */
1165 static struct mbuf *
1166 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1167 {
1168 	struct ipoption *p = mtod(opt, struct ipoption *);
1169 	struct mbuf *n;
1170 	struct ip *ip = mtod(m, struct ip *);
1171 	unsigned optlen;
1172 
1173 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1174 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
1175 		return (m);		/* XXX should fail */
1176 	if (!in_nullhost(p->ipopt_dst))
1177 		ip->ip_dst = p->ipopt_dst;
1178 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
1179 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1180 		if (n == 0)
1181 			return (m);
1182 		MCLAIM(n, m->m_owner);
1183 		M_MOVE_PKTHDR(n, m);
1184 		m->m_len -= sizeof(struct ip);
1185 		m->m_data += sizeof(struct ip);
1186 		n->m_next = m;
1187 		m = n;
1188 		m->m_len = optlen + sizeof(struct ip);
1189 		m->m_data += max_linkhdr;
1190 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
1191 	} else {
1192 		m->m_data -= optlen;
1193 		m->m_len += optlen;
1194 		memmove(mtod(m, void *), ip, sizeof(struct ip));
1195 	}
1196 	m->m_pkthdr.len += optlen;
1197 	ip = mtod(m, struct ip *);
1198 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
1199 	*phlen = sizeof(struct ip) + optlen;
1200 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1201 	return (m);
1202 }
1203 
1204 /*
1205  * Copy options from ip to jp,
1206  * omitting those not copied during fragmentation.
1207  */
1208 int
1209 ip_optcopy(struct ip *ip, struct ip *jp)
1210 {
1211 	u_char *cp, *dp;
1212 	int opt, optlen, cnt;
1213 
1214 	cp = (u_char *)(ip + 1);
1215 	dp = (u_char *)(jp + 1);
1216 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1217 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1218 		opt = cp[0];
1219 		if (opt == IPOPT_EOL)
1220 			break;
1221 		if (opt == IPOPT_NOP) {
1222 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1223 			*dp++ = IPOPT_NOP;
1224 			optlen = 1;
1225 			continue;
1226 		}
1227 #ifdef DIAGNOSTIC
1228 		if (cnt < IPOPT_OLEN + sizeof(*cp))
1229 			panic("malformed IPv4 option passed to ip_optcopy");
1230 #endif
1231 		optlen = cp[IPOPT_OLEN];
1232 #ifdef DIAGNOSTIC
1233 		if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1234 			panic("malformed IPv4 option passed to ip_optcopy");
1235 #endif
1236 		/* bogus lengths should have been caught by ip_dooptions */
1237 		if (optlen > cnt)
1238 			optlen = cnt;
1239 		if (IPOPT_COPIED(opt)) {
1240 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1241 			dp += optlen;
1242 		}
1243 	}
1244 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1245 		*dp++ = IPOPT_EOL;
1246 	return (optlen);
1247 }
1248 
1249 /*
1250  * IP socket option processing.
1251  */
1252 int
1253 ip_ctloutput(int op, struct socket *so, int level, int optname,
1254     struct mbuf **mp)
1255 {
1256 	struct inpcb *inp = sotoinpcb(so);
1257 	struct mbuf *m = *mp;
1258 	int optval = 0;
1259 	int error = 0;
1260 #if defined(IPSEC) || defined(FAST_IPSEC)
1261 	struct lwp *l = curlwp;	/*XXX*/
1262 #endif
1263 
1264 	if (level != IPPROTO_IP) {
1265 		if (op == PRCO_SETOPT && *mp)
1266 			(void) m_free(*mp);
1267 		if (level == SOL_SOCKET && optname == SO_NOHEADER)
1268 			return 0;
1269 		return ENOPROTOOPT;
1270 	}
1271 
1272 	switch (op) {
1273 
1274 	case PRCO_SETOPT:
1275 		switch (optname) {
1276 		case IP_OPTIONS:
1277 #ifdef notyet
1278 		case IP_RETOPTS:
1279 			return (ip_pcbopts(optname, &inp->inp_options, m));
1280 #else
1281 			return (ip_pcbopts(&inp->inp_options, m));
1282 #endif
1283 
1284 		case IP_TOS:
1285 		case IP_TTL:
1286 		case IP_RECVOPTS:
1287 		case IP_RECVRETOPTS:
1288 		case IP_RECVDSTADDR:
1289 		case IP_RECVIF:
1290 			if (m == NULL || m->m_len != sizeof(int))
1291 				error = EINVAL;
1292 			else {
1293 				optval = *mtod(m, int *);
1294 				switch (optname) {
1295 
1296 				case IP_TOS:
1297 					inp->inp_ip.ip_tos = optval;
1298 					break;
1299 
1300 				case IP_TTL:
1301 					inp->inp_ip.ip_ttl = optval;
1302 					break;
1303 #define	OPTSET(bit) \
1304 	if (optval) \
1305 		inp->inp_flags |= bit; \
1306 	else \
1307 		inp->inp_flags &= ~bit;
1308 
1309 				case IP_RECVOPTS:
1310 					OPTSET(INP_RECVOPTS);
1311 					break;
1312 
1313 				case IP_RECVRETOPTS:
1314 					OPTSET(INP_RECVRETOPTS);
1315 					break;
1316 
1317 				case IP_RECVDSTADDR:
1318 					OPTSET(INP_RECVDSTADDR);
1319 					break;
1320 
1321 				case IP_RECVIF:
1322 					OPTSET(INP_RECVIF);
1323 					break;
1324 				}
1325 			}
1326 			break;
1327 #undef OPTSET
1328 
1329 		case IP_MULTICAST_IF:
1330 		case IP_MULTICAST_TTL:
1331 		case IP_MULTICAST_LOOP:
1332 		case IP_ADD_MEMBERSHIP:
1333 		case IP_DROP_MEMBERSHIP:
1334 			error = ip_setmoptions(optname, &inp->inp_moptions, m);
1335 			break;
1336 
1337 		case IP_PORTRANGE:
1338 			if (m == 0 || m->m_len != sizeof(int))
1339 				error = EINVAL;
1340 			else {
1341 				optval = *mtod(m, int *);
1342 
1343 				switch (optval) {
1344 
1345 				case IP_PORTRANGE_DEFAULT:
1346 				case IP_PORTRANGE_HIGH:
1347 					inp->inp_flags &= ~(INP_LOWPORT);
1348 					break;
1349 
1350 				case IP_PORTRANGE_LOW:
1351 					inp->inp_flags |= INP_LOWPORT;
1352 					break;
1353 
1354 				default:
1355 					error = EINVAL;
1356 					break;
1357 				}
1358 			}
1359 			break;
1360 
1361 #if defined(IPSEC) || defined(FAST_IPSEC)
1362 		case IP_IPSEC_POLICY:
1363 		{
1364 			void *req = NULL;
1365 			size_t len = 0;
1366 			int priv = 0;
1367 
1368 #ifdef __NetBSD__
1369 			if (l == 0 || kauth_authorize_generic(l->l_cred,
1370 			    KAUTH_GENERIC_ISSUSER, NULL))
1371 				priv = 0;
1372 			else
1373 				priv = 1;
1374 #else
1375 			priv = (in6p->in6p_socket->so_state & SS_PRIV);
1376 #endif
1377 			if (m) {
1378 				req = mtod(m, void *);
1379 				len = m->m_len;
1380 			}
1381 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1382 			break;
1383 		    }
1384 #endif /*IPSEC*/
1385 
1386 		default:
1387 			error = ENOPROTOOPT;
1388 			break;
1389 		}
1390 		if (m)
1391 			(void)m_free(m);
1392 		break;
1393 
1394 	case PRCO_GETOPT:
1395 		switch (optname) {
1396 		case IP_OPTIONS:
1397 		case IP_RETOPTS:
1398 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1399 			MCLAIM(m, so->so_mowner);
1400 			if (inp->inp_options) {
1401 				m->m_len = inp->inp_options->m_len;
1402 				bcopy(mtod(inp->inp_options, void *),
1403 				    mtod(m, void *), (unsigned)m->m_len);
1404 			} else
1405 				m->m_len = 0;
1406 			break;
1407 
1408 		case IP_TOS:
1409 		case IP_TTL:
1410 		case IP_RECVOPTS:
1411 		case IP_RECVRETOPTS:
1412 		case IP_RECVDSTADDR:
1413 		case IP_RECVIF:
1414 		case IP_ERRORMTU:
1415 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1416 			MCLAIM(m, so->so_mowner);
1417 			m->m_len = sizeof(int);
1418 			switch (optname) {
1419 
1420 			case IP_TOS:
1421 				optval = inp->inp_ip.ip_tos;
1422 				break;
1423 
1424 			case IP_TTL:
1425 				optval = inp->inp_ip.ip_ttl;
1426 				break;
1427 
1428 			case IP_ERRORMTU:
1429 				optval = inp->inp_errormtu;
1430 				break;
1431 
1432 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1433 
1434 			case IP_RECVOPTS:
1435 				optval = OPTBIT(INP_RECVOPTS);
1436 				break;
1437 
1438 			case IP_RECVRETOPTS:
1439 				optval = OPTBIT(INP_RECVRETOPTS);
1440 				break;
1441 
1442 			case IP_RECVDSTADDR:
1443 				optval = OPTBIT(INP_RECVDSTADDR);
1444 				break;
1445 
1446 			case IP_RECVIF:
1447 				optval = OPTBIT(INP_RECVIF);
1448 				break;
1449 			}
1450 			*mtod(m, int *) = optval;
1451 			break;
1452 
1453 #if 0	/* defined(IPSEC) || defined(FAST_IPSEC) */
1454 		/* XXX: code broken */
1455 		case IP_IPSEC_POLICY:
1456 		{
1457 			void *req = NULL;
1458 			size_t len = 0;
1459 
1460 			if (m) {
1461 				req = mtod(m, void *);
1462 				len = m->m_len;
1463 			}
1464 			error = ipsec4_get_policy(inp, req, len, mp);
1465 			break;
1466 		}
1467 #endif /*IPSEC*/
1468 
1469 		case IP_MULTICAST_IF:
1470 		case IP_MULTICAST_TTL:
1471 		case IP_MULTICAST_LOOP:
1472 		case IP_ADD_MEMBERSHIP:
1473 		case IP_DROP_MEMBERSHIP:
1474 			error = ip_getmoptions(optname, inp->inp_moptions, mp);
1475 			if (*mp)
1476 				MCLAIM(*mp, so->so_mowner);
1477 			break;
1478 
1479 		case IP_PORTRANGE:
1480 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1481 			MCLAIM(m, so->so_mowner);
1482 			m->m_len = sizeof(int);
1483 
1484 			if (inp->inp_flags & INP_LOWPORT)
1485 				optval = IP_PORTRANGE_LOW;
1486 			else
1487 				optval = IP_PORTRANGE_DEFAULT;
1488 
1489 			*mtod(m, int *) = optval;
1490 			break;
1491 
1492 		default:
1493 			error = ENOPROTOOPT;
1494 			break;
1495 		}
1496 		break;
1497 	}
1498 	return (error);
1499 }
1500 
1501 /*
1502  * Set up IP options in pcb for insertion in output packets.
1503  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1504  * with destination address if source routed.
1505  */
1506 int
1507 #ifdef notyet
1508 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1509 #else
1510 ip_pcbopts(struct mbuf **pcbopt, struct mbuf *m)
1511 #endif
1512 {
1513 	int cnt, optlen;
1514 	u_char *cp;
1515 	u_char opt;
1516 
1517 	/* turn off any old options */
1518 	if (*pcbopt)
1519 		(void)m_free(*pcbopt);
1520 	*pcbopt = 0;
1521 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1522 		/*
1523 		 * Only turning off any previous options.
1524 		 */
1525 		if (m)
1526 			(void)m_free(m);
1527 		return (0);
1528 	}
1529 
1530 #ifndef	__vax__
1531 	if (m->m_len % sizeof(int32_t))
1532 		goto bad;
1533 #endif
1534 	/*
1535 	 * IP first-hop destination address will be stored before
1536 	 * actual options; move other options back
1537 	 * and clear it when none present.
1538 	 */
1539 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1540 		goto bad;
1541 	cnt = m->m_len;
1542 	m->m_len += sizeof(struct in_addr);
1543 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1544 	memmove(cp, mtod(m, void *), (unsigned)cnt);
1545 	bzero(mtod(m, void *), sizeof(struct in_addr));
1546 
1547 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1548 		opt = cp[IPOPT_OPTVAL];
1549 		if (opt == IPOPT_EOL)
1550 			break;
1551 		if (opt == IPOPT_NOP)
1552 			optlen = 1;
1553 		else {
1554 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1555 				goto bad;
1556 			optlen = cp[IPOPT_OLEN];
1557 			if (optlen < IPOPT_OLEN  + sizeof(*cp) || optlen > cnt)
1558 				goto bad;
1559 		}
1560 		switch (opt) {
1561 
1562 		default:
1563 			break;
1564 
1565 		case IPOPT_LSRR:
1566 		case IPOPT_SSRR:
1567 			/*
1568 			 * user process specifies route as:
1569 			 *	->A->B->C->D
1570 			 * D must be our final destination (but we can't
1571 			 * check that since we may not have connected yet).
1572 			 * A is first hop destination, which doesn't appear in
1573 			 * actual IP option, but is stored before the options.
1574 			 */
1575 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1576 				goto bad;
1577 			m->m_len -= sizeof(struct in_addr);
1578 			cnt -= sizeof(struct in_addr);
1579 			optlen -= sizeof(struct in_addr);
1580 			cp[IPOPT_OLEN] = optlen;
1581 			/*
1582 			 * Move first hop before start of options.
1583 			 */
1584 			bcopy((void *)&cp[IPOPT_OFFSET+1], mtod(m, void *),
1585 			    sizeof(struct in_addr));
1586 			/*
1587 			 * Then copy rest of options back
1588 			 * to close up the deleted entry.
1589 			 */
1590 			(void)memmove(&cp[IPOPT_OFFSET+1],
1591 			    &cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1592 			    (unsigned)cnt - (IPOPT_MINOFF - 1));
1593 			break;
1594 		}
1595 	}
1596 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1597 		goto bad;
1598 	*pcbopt = m;
1599 	return (0);
1600 
1601 bad:
1602 	(void)m_free(m);
1603 	return (EINVAL);
1604 }
1605 
1606 /*
1607  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1608  */
1609 static struct ifnet *
1610 ip_multicast_if(struct in_addr *a, int *ifindexp)
1611 {
1612 	int ifindex;
1613 	struct ifnet *ifp = NULL;
1614 	struct in_ifaddr *ia;
1615 
1616 	if (ifindexp)
1617 		*ifindexp = 0;
1618 	if (ntohl(a->s_addr) >> 24 == 0) {
1619 		ifindex = ntohl(a->s_addr) & 0xffffff;
1620 		if (ifindex < 0 || if_indexlim <= ifindex)
1621 			return NULL;
1622 		ifp = ifindex2ifnet[ifindex];
1623 		if (!ifp)
1624 			return NULL;
1625 		if (ifindexp)
1626 			*ifindexp = ifindex;
1627 	} else {
1628 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1629 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1630 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1631 				ifp = ia->ia_ifp;
1632 				break;
1633 			}
1634 		}
1635 	}
1636 	return ifp;
1637 }
1638 
1639 static int
1640 ip_getoptval(struct mbuf *m, u_int8_t *val, u_int maxval)
1641 {
1642 	u_int tval;
1643 
1644 	if (m == NULL)
1645 		return EINVAL;
1646 
1647 	switch (m->m_len) {
1648 	case sizeof(u_char):
1649 		tval = *(mtod(m, u_char *));
1650 		break;
1651 	case sizeof(u_int):
1652 		tval = *(mtod(m, u_int *));
1653 		break;
1654 	default:
1655 		return EINVAL;
1656 	}
1657 
1658 	if (tval > maxval)
1659 		return EINVAL;
1660 
1661 	*val = tval;
1662 	return 0;
1663 }
1664 
1665 /*
1666  * Set the IP multicast options in response to user setsockopt().
1667  */
1668 int
1669 ip_setmoptions(int optname, struct ip_moptions **imop, struct mbuf *m)
1670 {
1671 	int error = 0;
1672 	int i;
1673 	struct in_addr addr;
1674 	struct ip_mreq *mreq;
1675 	struct ifnet *ifp;
1676 	struct ip_moptions *imo = *imop;
1677 	int ifindex;
1678 
1679 	if (imo == NULL) {
1680 		/*
1681 		 * No multicast option buffer attached to the pcb;
1682 		 * allocate one and initialize to default values.
1683 		 */
1684 		imo = (struct ip_moptions *)malloc(sizeof(*imo), M_IPMOPTS,
1685 		    M_WAITOK);
1686 
1687 		if (imo == NULL)
1688 			return (ENOBUFS);
1689 		*imop = imo;
1690 		imo->imo_multicast_ifp = NULL;
1691 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1692 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1693 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1694 		imo->imo_num_memberships = 0;
1695 	}
1696 
1697 	switch (optname) {
1698 
1699 	case IP_MULTICAST_IF:
1700 		/*
1701 		 * Select the interface for outgoing multicast packets.
1702 		 */
1703 		if (m == NULL || m->m_len != sizeof(struct in_addr)) {
1704 			error = EINVAL;
1705 			break;
1706 		}
1707 		addr = *(mtod(m, struct in_addr *));
1708 		/*
1709 		 * INADDR_ANY is used to remove a previous selection.
1710 		 * When no interface is selected, a default one is
1711 		 * chosen every time a multicast packet is sent.
1712 		 */
1713 		if (in_nullhost(addr)) {
1714 			imo->imo_multicast_ifp = NULL;
1715 			break;
1716 		}
1717 		/*
1718 		 * The selected interface is identified by its local
1719 		 * IP address.  Find the interface and confirm that
1720 		 * it supports multicasting.
1721 		 */
1722 		ifp = ip_multicast_if(&addr, &ifindex);
1723 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1724 			error = EADDRNOTAVAIL;
1725 			break;
1726 		}
1727 		imo->imo_multicast_ifp = ifp;
1728 		if (ifindex)
1729 			imo->imo_multicast_addr = addr;
1730 		else
1731 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1732 		break;
1733 
1734 	case IP_MULTICAST_TTL:
1735 		/*
1736 		 * Set the IP time-to-live for outgoing multicast packets.
1737 		 */
1738 		error = ip_getoptval(m, &imo->imo_multicast_ttl, MAXTTL);
1739 		break;
1740 
1741 	case IP_MULTICAST_LOOP:
1742 		/*
1743 		 * Set the loopback flag for outgoing multicast packets.
1744 		 * Must be zero or one.
1745 		 */
1746 		error = ip_getoptval(m, &imo->imo_multicast_loop, 1);
1747 		break;
1748 
1749 	case IP_ADD_MEMBERSHIP:
1750 		/*
1751 		 * Add a multicast group membership.
1752 		 * Group must be a valid IP multicast address.
1753 		 */
1754 		if (m == NULL || m->m_len != sizeof(struct ip_mreq)) {
1755 			error = EINVAL;
1756 			break;
1757 		}
1758 		mreq = mtod(m, struct ip_mreq *);
1759 		if (!IN_MULTICAST(mreq->imr_multiaddr.s_addr)) {
1760 			error = EINVAL;
1761 			break;
1762 		}
1763 		/*
1764 		 * If no interface address was provided, use the interface of
1765 		 * the route to the given multicast address.
1766 		 */
1767 		if (in_nullhost(mreq->imr_interface)) {
1768 			union {
1769 				struct sockaddr		dst;
1770 				struct sockaddr_in	dst4;
1771 			} u;
1772 			struct route ro;
1773 
1774 			memset(&ro, 0, sizeof(ro));
1775 
1776 			sockaddr_in_init(&u.dst4, &mreq->imr_multiaddr, 0);
1777 			rtcache_setdst(&ro, &u.dst);
1778 			rtcache_init(&ro);
1779 			ifp = (ro.ro_rt != NULL) ? ro.ro_rt->rt_ifp : NULL;
1780 			rtcache_free(&ro);
1781 		} else {
1782 			ifp = ip_multicast_if(&mreq->imr_interface, NULL);
1783 		}
1784 		/*
1785 		 * See if we found an interface, and confirm that it
1786 		 * supports multicast.
1787 		 */
1788 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1789 			error = EADDRNOTAVAIL;
1790 			break;
1791 		}
1792 		/*
1793 		 * See if the membership already exists or if all the
1794 		 * membership slots are full.
1795 		 */
1796 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1797 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1798 			    in_hosteq(imo->imo_membership[i]->inm_addr,
1799 				      mreq->imr_multiaddr))
1800 				break;
1801 		}
1802 		if (i < imo->imo_num_memberships) {
1803 			error = EADDRINUSE;
1804 			break;
1805 		}
1806 		if (i == IP_MAX_MEMBERSHIPS) {
1807 			error = ETOOMANYREFS;
1808 			break;
1809 		}
1810 		/*
1811 		 * Everything looks good; add a new record to the multicast
1812 		 * address list for the given interface.
1813 		 */
1814 		if ((imo->imo_membership[i] =
1815 		    in_addmulti(&mreq->imr_multiaddr, ifp)) == NULL) {
1816 			error = ENOBUFS;
1817 			break;
1818 		}
1819 		++imo->imo_num_memberships;
1820 		break;
1821 
1822 	case IP_DROP_MEMBERSHIP:
1823 		/*
1824 		 * Drop a multicast group membership.
1825 		 * Group must be a valid IP multicast address.
1826 		 */
1827 		if (m == NULL || m->m_len != sizeof(struct ip_mreq)) {
1828 			error = EINVAL;
1829 			break;
1830 		}
1831 		mreq = mtod(m, struct ip_mreq *);
1832 		if (!IN_MULTICAST(mreq->imr_multiaddr.s_addr)) {
1833 			error = EINVAL;
1834 			break;
1835 		}
1836 		/*
1837 		 * If an interface address was specified, get a pointer
1838 		 * to its ifnet structure.
1839 		 */
1840 		if (in_nullhost(mreq->imr_interface))
1841 			ifp = NULL;
1842 		else {
1843 			ifp = ip_multicast_if(&mreq->imr_interface, NULL);
1844 			if (ifp == NULL) {
1845 				error = EADDRNOTAVAIL;
1846 				break;
1847 			}
1848 		}
1849 		/*
1850 		 * Find the membership in the membership array.
1851 		 */
1852 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1853 			if ((ifp == NULL ||
1854 			     imo->imo_membership[i]->inm_ifp == ifp) &&
1855 			     in_hosteq(imo->imo_membership[i]->inm_addr,
1856 				       mreq->imr_multiaddr))
1857 				break;
1858 		}
1859 		if (i == imo->imo_num_memberships) {
1860 			error = EADDRNOTAVAIL;
1861 			break;
1862 		}
1863 		/*
1864 		 * Give up the multicast address record to which the
1865 		 * membership points.
1866 		 */
1867 		in_delmulti(imo->imo_membership[i]);
1868 		/*
1869 		 * Remove the gap in the membership array.
1870 		 */
1871 		for (++i; i < imo->imo_num_memberships; ++i)
1872 			imo->imo_membership[i-1] = imo->imo_membership[i];
1873 		--imo->imo_num_memberships;
1874 		break;
1875 
1876 	default:
1877 		error = EOPNOTSUPP;
1878 		break;
1879 	}
1880 
1881 	/*
1882 	 * If all options have default values, no need to keep the mbuf.
1883 	 */
1884 	if (imo->imo_multicast_ifp == NULL &&
1885 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1886 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1887 	    imo->imo_num_memberships == 0) {
1888 		free(*imop, M_IPMOPTS);
1889 		*imop = NULL;
1890 	}
1891 
1892 	return (error);
1893 }
1894 
1895 /*
1896  * Return the IP multicast options in response to user getsockopt().
1897  */
1898 int
1899 ip_getmoptions(int optname, struct ip_moptions *imo, struct mbuf **mp)
1900 {
1901 	u_char *ttl;
1902 	u_char *loop;
1903 	struct in_addr *addr;
1904 	struct in_ifaddr *ia;
1905 
1906 	*mp = m_get(M_WAIT, MT_SOOPTS);
1907 
1908 	switch (optname) {
1909 
1910 	case IP_MULTICAST_IF:
1911 		addr = mtod(*mp, struct in_addr *);
1912 		(*mp)->m_len = sizeof(struct in_addr);
1913 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
1914 			*addr = zeroin_addr;
1915 		else if (imo->imo_multicast_addr.s_addr) {
1916 			/* return the value user has set */
1917 			*addr = imo->imo_multicast_addr;
1918 		} else {
1919 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
1920 			*addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1921 		}
1922 		return (0);
1923 
1924 	case IP_MULTICAST_TTL:
1925 		ttl = mtod(*mp, u_char *);
1926 		(*mp)->m_len = 1;
1927 		*ttl = imo ? imo->imo_multicast_ttl
1928 			   : IP_DEFAULT_MULTICAST_TTL;
1929 		return (0);
1930 
1931 	case IP_MULTICAST_LOOP:
1932 		loop = mtod(*mp, u_char *);
1933 		(*mp)->m_len = 1;
1934 		*loop = imo ? imo->imo_multicast_loop
1935 			    : IP_DEFAULT_MULTICAST_LOOP;
1936 		return (0);
1937 
1938 	default:
1939 		return (EOPNOTSUPP);
1940 	}
1941 }
1942 
1943 /*
1944  * Discard the IP multicast options.
1945  */
1946 void
1947 ip_freemoptions(struct ip_moptions *imo)
1948 {
1949 	int i;
1950 
1951 	if (imo != NULL) {
1952 		for (i = 0; i < imo->imo_num_memberships; ++i)
1953 			in_delmulti(imo->imo_membership[i]);
1954 		free(imo, M_IPMOPTS);
1955 	}
1956 }
1957 
1958 /*
1959  * Routine called from ip_output() to loop back a copy of an IP multicast
1960  * packet to the input queue of a specified interface.  Note that this
1961  * calls the output routine of the loopback "driver", but with an interface
1962  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1963  */
1964 static void
1965 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1966 {
1967 	struct ip *ip;
1968 	struct mbuf *copym;
1969 
1970 	copym = m_copypacket(m, M_DONTWAIT);
1971 	if (copym != NULL
1972 	 && (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1973 		copym = m_pullup(copym, sizeof(struct ip));
1974 	if (copym == NULL)
1975 		return;
1976 	/*
1977 	 * We don't bother to fragment if the IP length is greater
1978 	 * than the interface's MTU.  Can this possibly matter?
1979 	 */
1980 	ip = mtod(copym, struct ip *);
1981 
1982 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1983 		in_delayed_cksum(copym);
1984 		copym->m_pkthdr.csum_flags &=
1985 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1986 	}
1987 
1988 	ip->ip_sum = 0;
1989 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1990 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1991 }
1992