xref: /netbsd-src/sys/netinet/ip_output.c (revision b7ae68fde0d8ef1c03714e8bbb1ee7c6118ea93b)
1 /*	$NetBSD: ip_output.c,v 1.165 2006/07/23 22:06:13 ad 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 1.165 2006/07/23 22:06:13 ad 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 *, struct sockaddr_in *);
159 static int ip_getoptval(struct mbuf *, u_int8_t *, u_int);
160 
161 #ifdef PFIL_HOOKS
162 extern struct pfil_head inet_pfil_hook;			/* XXX */
163 #endif
164 
165 int	ip_do_loopback_cksum = 0;
166 
167 #define	IN_NEED_CHECKSUM(ifp, csum_flags) \
168 	(__predict_true(((ifp)->if_flags & IFF_LOOPBACK) == 0 || \
169 	(((csum_flags) & M_CSUM_UDPv4) != 0 && udp_do_loopback_cksum) || \
170 	(((csum_flags) & M_CSUM_TCPv4) != 0 && tcp_do_loopback_cksum) || \
171 	(((csum_flags) & M_CSUM_IPv4) != 0 && ip_do_loopback_cksum)))
172 
173 struct ip_tso_output_args {
174 	struct ifnet *ifp;
175 	struct sockaddr *sa;
176 	struct rtentry *rt;
177 };
178 
179 static int ip_tso_output_callback(void *, struct mbuf *);
180 static int ip_tso_output(struct ifnet *, struct mbuf *, struct sockaddr *,
181     struct rtentry *);
182 
183 static int
184 ip_tso_output_callback(void *vp, struct mbuf *m)
185 {
186 	struct ip_tso_output_args *args = vp;
187 	struct ifnet *ifp = args->ifp;
188 
189 	return (*ifp->if_output)(ifp, m, args->sa, args->rt);
190 }
191 
192 static int
193 ip_tso_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa,
194     struct rtentry *rt)
195 {
196 	struct ip_tso_output_args args;
197 
198 	args.ifp = ifp;
199 	args.sa = sa;
200 	args.rt = rt;
201 
202 	return tcp4_segment(m, ip_tso_output_callback, &args);
203 }
204 
205 /*
206  * IP output.  The packet in mbuf chain m contains a skeletal IP
207  * header (with len, off, ttl, proto, tos, src, dst).
208  * The mbuf chain containing the packet will be freed.
209  * The mbuf opt, if present, will not be freed.
210  */
211 int
212 ip_output(struct mbuf *m0, ...)
213 {
214 	struct ip *ip;
215 	struct ifnet *ifp;
216 	struct mbuf *m = m0;
217 	int hlen = sizeof (struct ip);
218 	int len, error = 0;
219 	struct route iproute;
220 	struct sockaddr_in *dst;
221 	struct in_ifaddr *ia;
222 	struct mbuf *opt;
223 	struct route *ro;
224 	int flags, sw_csum;
225 	int *mtu_p;
226 	u_long mtu;
227 	struct ip_moptions *imo;
228 	struct socket *so;
229 	va_list ap;
230 #ifdef IPSEC_NAT_T
231 	int natt_frag = 0;
232 #endif
233 #ifdef IPSEC
234 	struct secpolicy *sp = NULL;
235 #endif /*IPSEC*/
236 #ifdef FAST_IPSEC
237 	struct inpcb *inp;
238 	struct m_tag *mtag;
239 	struct secpolicy *sp = NULL;
240 	struct tdb_ident *tdbi;
241 	int s;
242 #endif
243 	u_int16_t ip_len;
244 
245 	len = 0;
246 	va_start(ap, m0);
247 	opt = va_arg(ap, struct mbuf *);
248 	ro = va_arg(ap, struct route *);
249 	flags = va_arg(ap, int);
250 	imo = va_arg(ap, struct ip_moptions *);
251 	so = va_arg(ap, struct socket *);
252 	if (flags & IP_RETURNMTU)
253 		mtu_p = va_arg(ap, int *);
254 	else
255 		mtu_p = NULL;
256 	va_end(ap);
257 
258 	MCLAIM(m, &ip_tx_mowner);
259 #ifdef FAST_IPSEC
260 	if (so != NULL && so->so_proto->pr_domain->dom_family == AF_INET)
261 		inp = (struct inpcb *)so->so_pcb;
262 	else
263 		inp = NULL;
264 #endif /* FAST_IPSEC */
265 
266 #ifdef	DIAGNOSTIC
267 	if ((m->m_flags & M_PKTHDR) == 0)
268 		panic("ip_output: no HDR");
269 
270 	if ((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) != 0) {
271 		panic("ip_output: IPv6 checksum offload flags: %d",
272 		    m->m_pkthdr.csum_flags);
273 	}
274 
275 	if ((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) ==
276 	    (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
277 		panic("ip_output: conflicting checksum offload flags: %d",
278 		    m->m_pkthdr.csum_flags);
279 	}
280 #endif
281 	if (opt) {
282 		m = ip_insertoptions(m, opt, &len);
283 		if (len >= sizeof(struct ip))
284 			hlen = len;
285 	}
286 	ip = mtod(m, struct ip *);
287 	/*
288 	 * Fill in IP header.
289 	 */
290 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
291 		ip->ip_v = IPVERSION;
292 		ip->ip_off = htons(0);
293 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
294 			ip->ip_id = ip_newid();
295 		} else {
296 
297 			/*
298 			 * TSO capable interfaces (typically?) increment
299 			 * ip_id for each segment.
300 			 * "allocate" enough ids here to increase the chance
301 			 * for them to be unique.
302 			 *
303 			 * note that the following calculation is not
304 			 * needed to be precise.  wasting some ip_id is fine.
305 			 */
306 
307 			unsigned int segsz = m->m_pkthdr.segsz;
308 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
309 			unsigned int num = howmany(datasz, segsz);
310 
311 			ip->ip_id = ip_newid_range(num);
312 		}
313 		ip->ip_hl = hlen >> 2;
314 		ipstat.ips_localout++;
315 	} else {
316 		hlen = ip->ip_hl << 2;
317 	}
318 	/*
319 	 * Route packet.
320 	 */
321 	if (ro == 0) {
322 		ro = &iproute;
323 		bzero((caddr_t)ro, sizeof (*ro));
324 	}
325 	dst = satosin(&ro->ro_dst);
326 	/*
327 	 * If there is a cached route,
328 	 * check that it is to the same destination
329 	 * and is still up.  If not, free it and try again.
330 	 * The address family should also be checked in case of sharing the
331 	 * cache with IPv6.
332 	 */
333 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
334 	    dst->sin_family != AF_INET ||
335 	    !in_hosteq(dst->sin_addr, ip->ip_dst))) {
336 		RTFREE(ro->ro_rt);
337 		ro->ro_rt = (struct rtentry *)0;
338 	}
339 	if (ro->ro_rt == 0) {
340 		bzero(dst, sizeof(*dst));
341 		dst->sin_family = AF_INET;
342 		dst->sin_len = sizeof(*dst);
343 		dst->sin_addr = ip->ip_dst;
344 	}
345 	/*
346 	 * If routing to interface only,
347 	 * short circuit routing lookup.
348 	 */
349 	if (flags & IP_ROUTETOIF) {
350 		if ((ia = ifatoia(ifa_ifwithladdr(sintosa(dst)))) == 0) {
351 			ipstat.ips_noroute++;
352 			error = ENETUNREACH;
353 			goto bad;
354 		}
355 		ifp = ia->ia_ifp;
356 		mtu = ifp->if_mtu;
357 		ip->ip_ttl = 1;
358 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
359 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
360 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
361 		ifp = imo->imo_multicast_ifp;
362 		mtu = ifp->if_mtu;
363 		IFP_TO_IA(ifp, ia);
364 	} else {
365 		if (ro->ro_rt == 0)
366 			rtalloc(ro);
367 		if (ro->ro_rt == 0) {
368 			ipstat.ips_noroute++;
369 			error = EHOSTUNREACH;
370 			goto bad;
371 		}
372 		ia = ifatoia(ro->ro_rt->rt_ifa);
373 		ifp = ro->ro_rt->rt_ifp;
374 		if ((mtu = ro->ro_rt->rt_rmx.rmx_mtu) == 0)
375 			mtu = ifp->if_mtu;
376 		ro->ro_rt->rt_use++;
377 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
378 			dst = satosin(ro->ro_rt->rt_gateway);
379 	}
380 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
381 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
382 		struct in_multi *inm;
383 
384 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
385 			M_BCAST : M_MCAST;
386 		/*
387 		 * IP destination address is multicast.  Make sure "dst"
388 		 * still points to the address in "ro".  (It may have been
389 		 * changed to point to a gateway address, above.)
390 		 */
391 		dst = satosin(&ro->ro_dst);
392 		/*
393 		 * See if the caller provided any multicast options
394 		 */
395 		if (imo != NULL)
396 			ip->ip_ttl = imo->imo_multicast_ttl;
397 		else
398 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
399 
400 		/*
401 		 * if we don't know the outgoing ifp yet, we can't generate
402 		 * output
403 		 */
404 		if (!ifp) {
405 			ipstat.ips_noroute++;
406 			error = ENETUNREACH;
407 			goto bad;
408 		}
409 
410 		/*
411 		 * If the packet is multicast or broadcast, confirm that
412 		 * the outgoing interface can transmit it.
413 		 */
414 		if (((m->m_flags & M_MCAST) &&
415 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
416 		    ((m->m_flags & M_BCAST) &&
417 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
418 			ipstat.ips_noroute++;
419 			error = ENETUNREACH;
420 			goto bad;
421 		}
422 		/*
423 		 * If source address not specified yet, use an address
424 		 * of outgoing interface.
425 		 */
426 		if (in_nullhost(ip->ip_src)) {
427 			struct in_ifaddr *xia;
428 
429 			IFP_TO_IA(ifp, xia);
430 			if (!xia) {
431 				error = EADDRNOTAVAIL;
432 				goto bad;
433 			}
434 			ip->ip_src = xia->ia_addr.sin_addr;
435 		}
436 
437 		IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm);
438 		if (inm != NULL &&
439 		   (imo == NULL || imo->imo_multicast_loop)) {
440 			/*
441 			 * If we belong to the destination multicast group
442 			 * on the outgoing interface, and the caller did not
443 			 * forbid loopback, loop back a copy.
444 			 */
445 			ip_mloopback(ifp, m, dst);
446 		}
447 #ifdef MROUTING
448 		else {
449 			/*
450 			 * If we are acting as a multicast router, perform
451 			 * multicast forwarding as if the packet had just
452 			 * arrived on the interface to which we are about
453 			 * to send.  The multicast forwarding function
454 			 * recursively calls this function, using the
455 			 * IP_FORWARDING flag to prevent infinite recursion.
456 			 *
457 			 * Multicasts that are looped back by ip_mloopback(),
458 			 * above, will be forwarded by the ip_input() routine,
459 			 * if necessary.
460 			 */
461 			extern struct socket *ip_mrouter;
462 
463 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
464 				if (ip_mforward(m, ifp) != 0) {
465 					m_freem(m);
466 					goto done;
467 				}
468 			}
469 		}
470 #endif
471 		/*
472 		 * Multicasts with a time-to-live of zero may be looped-
473 		 * back, above, but must not be transmitted on a network.
474 		 * Also, multicasts addressed to the loopback interface
475 		 * are not sent -- the above call to ip_mloopback() will
476 		 * loop back a copy if this host actually belongs to the
477 		 * destination group on the loopback interface.
478 		 */
479 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
480 			m_freem(m);
481 			goto done;
482 		}
483 
484 		goto sendit;
485 	}
486 	/*
487 	 * If source address not specified yet, use address
488 	 * of outgoing interface.
489 	 */
490 	if (in_nullhost(ip->ip_src))
491 		ip->ip_src = ia->ia_addr.sin_addr;
492 
493 	/*
494 	 * packets with Class-D address as source are not valid per
495 	 * RFC 1112
496 	 */
497 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
498 		ipstat.ips_odropped++;
499 		error = EADDRNOTAVAIL;
500 		goto bad;
501 	}
502 
503 	/*
504 	 * Look for broadcast address and
505 	 * and verify user is allowed to send
506 	 * such a packet.
507 	 */
508 	if (in_broadcast(dst->sin_addr, ifp)) {
509 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
510 			error = EADDRNOTAVAIL;
511 			goto bad;
512 		}
513 		if ((flags & IP_ALLOWBROADCAST) == 0) {
514 			error = EACCES;
515 			goto bad;
516 		}
517 		/* don't allow broadcast messages to be fragmented */
518 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
519 			error = EMSGSIZE;
520 			goto bad;
521 		}
522 		m->m_flags |= M_BCAST;
523 	} else
524 		m->m_flags &= ~M_BCAST;
525 
526 sendit:
527 	/*
528 	 * If we're doing Path MTU Discovery, we need to set DF unless
529 	 * the route's MTU is locked.
530 	 */
531 	if ((flags & IP_MTUDISC) != 0 && ro->ro_rt != NULL &&
532 	    (ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
533 		ip->ip_off |= htons(IP_DF);
534 
535 	/* Remember the current ip_len */
536 	ip_len = ntohs(ip->ip_len);
537 
538 #ifdef IPSEC
539 	/* get SP for this packet */
540 	if (so == NULL)
541 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND,
542 		    flags, &error);
543 	else {
544 		if (IPSEC_PCB_SKIP_IPSEC(sotoinpcb_hdr(so)->inph_sp,
545 					 IPSEC_DIR_OUTBOUND))
546 			goto skip_ipsec;
547 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
548 	}
549 
550 	if (sp == NULL) {
551 		ipsecstat.out_inval++;
552 		goto bad;
553 	}
554 
555 	error = 0;
556 
557 	/* check policy */
558 	switch (sp->policy) {
559 	case IPSEC_POLICY_DISCARD:
560 		/*
561 		 * This packet is just discarded.
562 		 */
563 		ipsecstat.out_polvio++;
564 		goto bad;
565 
566 	case IPSEC_POLICY_BYPASS:
567 	case IPSEC_POLICY_NONE:
568 		/* no need to do IPsec. */
569 		goto skip_ipsec;
570 
571 	case IPSEC_POLICY_IPSEC:
572 		if (sp->req == NULL) {
573 			/* XXX should be panic ? */
574 			printf("ip_output: No IPsec request specified.\n");
575 			error = EINVAL;
576 			goto bad;
577 		}
578 		break;
579 
580 	case IPSEC_POLICY_ENTRUST:
581 	default:
582 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
583 	}
584 
585 #ifdef IPSEC_NAT_T
586 	/*
587 	 * NAT-T ESP fragmentation: don't do IPSec processing now,
588 	 * we'll do it on each fragmented packet.
589 	 */
590 	if (sp->req->sav &&
591 	    ((sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP) ||
592 	     (sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP_NON_IKE))) {
593 		if (ntohs(ip->ip_len) > sp->req->sav->esp_frag) {
594 			natt_frag = 1;
595 			mtu = sp->req->sav->esp_frag;
596 			goto skip_ipsec;
597 		}
598 	}
599 #endif /* IPSEC_NAT_T */
600 
601 	/*
602 	 * ipsec4_output() expects ip_len and ip_off in network
603 	 * order.  They have been set to network order above.
604 	 */
605 
606     {
607 	struct ipsec_output_state state;
608 	bzero(&state, sizeof(state));
609 	state.m = m;
610 	if (flags & IP_ROUTETOIF) {
611 		state.ro = &iproute;
612 		bzero(&iproute, sizeof(iproute));
613 	} else
614 		state.ro = ro;
615 	state.dst = (struct sockaddr *)dst;
616 
617 	/*
618 	 * We can't defer the checksum of payload data if
619 	 * we're about to encrypt/authenticate it.
620 	 *
621 	 * XXX When we support crypto offloading functions of
622 	 * XXX network interfaces, we need to reconsider this,
623 	 * XXX since it's likely that they'll support checksumming,
624 	 * XXX as well.
625 	 */
626 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
627 		in_delayed_cksum(m);
628 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
629 	}
630 
631 	error = ipsec4_output(&state, sp, flags);
632 
633 	m = state.m;
634 	if (flags & IP_ROUTETOIF) {
635 		/*
636 		 * if we have tunnel mode SA, we may need to ignore
637 		 * IP_ROUTETOIF.
638 		 */
639 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
640 			flags &= ~IP_ROUTETOIF;
641 			ro = state.ro;
642 		}
643 	} else
644 		ro = state.ro;
645 	dst = (struct sockaddr_in *)state.dst;
646 	if (error) {
647 		/* mbuf is already reclaimed in ipsec4_output. */
648 		m0 = NULL;
649 		switch (error) {
650 		case EHOSTUNREACH:
651 		case ENETUNREACH:
652 		case EMSGSIZE:
653 		case ENOBUFS:
654 		case ENOMEM:
655 			break;
656 		default:
657 			printf("ip4_output (ipsec): error code %d\n", error);
658 			/*fall through*/
659 		case ENOENT:
660 			/* don't show these error codes to the user */
661 			error = 0;
662 			break;
663 		}
664 		goto bad;
665 	}
666 
667 	/* be sure to update variables that are affected by ipsec4_output() */
668 	ip = mtod(m, struct ip *);
669 	hlen = ip->ip_hl << 2;
670 	ip_len = ntohs(ip->ip_len);
671 
672 	if (ro->ro_rt == NULL) {
673 		if ((flags & IP_ROUTETOIF) == 0) {
674 			printf("ip_output: "
675 				"can't update route after IPsec processing\n");
676 			error = EHOSTUNREACH;	/*XXX*/
677 			goto bad;
678 		}
679 	} else {
680 		/* nobody uses ia beyond here */
681 		if (state.encap) {
682 			ifp = ro->ro_rt->rt_ifp;
683 			if ((mtu = ro->ro_rt->rt_rmx.rmx_mtu) == 0)
684 				mtu = ifp->if_mtu;
685 		}
686 	}
687     }
688 skip_ipsec:
689 #endif /*IPSEC*/
690 #ifdef FAST_IPSEC
691 	/*
692 	 * Check the security policy (SP) for the packet and, if
693 	 * required, do IPsec-related processing.  There are two
694 	 * cases here; the first time a packet is sent through
695 	 * it will be untagged and handled by ipsec4_checkpolicy.
696 	 * If the packet is resubmitted to ip_output (e.g. after
697 	 * AH, ESP, etc. processing), there will be a tag to bypass
698 	 * the lookup and related policy checking.
699 	 */
700 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
701 	s = splsoftnet();
702 	if (mtag != NULL) {
703 		tdbi = (struct tdb_ident *)(mtag + 1);
704 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
705 		if (sp == NULL)
706 			error = -EINVAL;	/* force silent drop */
707 		m_tag_delete(m, mtag);
708 	} else {
709 		if (inp != NULL &&
710 		    IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND))
711 			goto spd_done;
712 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
713 					&error, inp);
714 	}
715 	/*
716 	 * There are four return cases:
717 	 *    sp != NULL	 	    apply IPsec policy
718 	 *    sp == NULL, error == 0	    no IPsec handling needed
719 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
720 	 *    sp == NULL, error != 0	    discard packet, report error
721 	 */
722 	if (sp != NULL) {
723 #ifdef IPSEC_NAT_T
724 		/*
725 		 * NAT-T ESP fragmentation: don't do IPSec processing now,
726 		 * we'll do it on each fragmented packet.
727 		 */
728 		if (sp->req->sav &&
729 		    ((sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP) ||
730 		     (sp->req->sav->natt_type & UDP_ENCAP_ESPINUDP_NON_IKE))) {
731 			if (ntohs(ip->ip_len) > sp->req->sav->esp_frag) {
732 				natt_frag = 1;
733 				mtu = sp->req->sav->esp_frag;
734 				goto spd_done;
735 			}
736 		}
737 #endif /* IPSEC_NAT_T */
738 		/* Loop detection, check if ipsec processing already done */
739 		IPSEC_ASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
740 		for (mtag = m_tag_first(m); mtag != NULL;
741 		     mtag = m_tag_next(m, mtag)) {
742 #ifdef MTAG_ABI_COMPAT
743 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
744 				continue;
745 #endif
746 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
747 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
748 				continue;
749 			/*
750 			 * Check if policy has an SA associated with it.
751 			 * This can happen when an SP has yet to acquire
752 			 * an SA; e.g. on first reference.  If it occurs,
753 			 * then we let ipsec4_process_packet do its thing.
754 			 */
755 			if (sp->req->sav == NULL)
756 				break;
757 			tdbi = (struct tdb_ident *)(mtag + 1);
758 			if (tdbi->spi == sp->req->sav->spi &&
759 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
760 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
761 				 sizeof (union sockaddr_union)) == 0) {
762 				/*
763 				 * No IPsec processing is needed, free
764 				 * reference to SP.
765 				 *
766 				 * NB: null pointer to avoid free at
767 				 *     done: below.
768 				 */
769 				KEY_FREESP(&sp), sp = NULL;
770 				splx(s);
771 				goto spd_done;
772 			}
773 		}
774 
775 		/*
776 		 * Do delayed checksums now because we send before
777 		 * this is done in the normal processing path.
778 		 */
779 		if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
780 			in_delayed_cksum(m);
781 			m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
782 		}
783 
784 #ifdef __FreeBSD__
785 		ip->ip_len = htons(ip->ip_len);
786 		ip->ip_off = htons(ip->ip_off);
787 #endif
788 
789 		/* NB: callee frees mbuf */
790 		error = ipsec4_process_packet(m, sp->req, flags, 0);
791 		/*
792 		 * Preserve KAME behaviour: ENOENT can be returned
793 		 * when an SA acquire is in progress.  Don't propagate
794 		 * this to user-level; it confuses applications.
795 		 *
796 		 * XXX this will go away when the SADB is redone.
797 		 */
798 		if (error == ENOENT)
799 			error = 0;
800 		splx(s);
801 		goto done;
802 	} else {
803 		splx(s);
804 
805 		if (error != 0) {
806 			/*
807 			 * Hack: -EINVAL is used to signal that a packet
808 			 * should be silently discarded.  This is typically
809 			 * because we asked key management for an SA and
810 			 * it was delayed (e.g. kicked up to IKE).
811 			 */
812 			if (error == -EINVAL)
813 				error = 0;
814 			goto bad;
815 		} else {
816 			/* No IPsec processing for this packet. */
817 		}
818 #ifdef notyet
819 		/*
820 		 * If deferred crypto processing is needed, check that
821 		 * the interface supports it.
822 		 */
823 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
824 		if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
825 			/* notify IPsec to do its own crypto */
826 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
827 			error = EHOSTUNREACH;
828 			goto bad;
829 		}
830 #endif
831 	}
832 spd_done:
833 #endif /* FAST_IPSEC */
834 
835 #ifdef PFIL_HOOKS
836 	/*
837 	 * Run through list of hooks for output packets.
838 	 */
839 	if ((error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT)) != 0)
840 		goto done;
841 	if (m == NULL)
842 		goto done;
843 
844 	ip = mtod(m, struct ip *);
845 	hlen = ip->ip_hl << 2;
846 #endif /* PFIL_HOOKS */
847 
848 	m->m_pkthdr.csum_data |= hlen << 16;
849 
850 #if IFA_STATS
851 	/*
852 	 * search for the source address structure to
853 	 * maintain output statistics.
854 	 */
855 	INADDR_TO_IA(ip->ip_src, ia);
856 #endif
857 
858 	/* Maybe skip checksums on loopback interfaces. */
859 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
860 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
861 	}
862 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
863 	/*
864 	 * If small enough for mtu of path, or if using TCP segmentation
865 	 * offload, can just send directly.
866 	 */
867 	if (ip_len <= mtu ||
868 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
869 #if IFA_STATS
870 		if (ia)
871 			ia->ia_ifa.ifa_data.ifad_outbytes += ip_len;
872 #endif
873 		/*
874 		 * Always initialize the sum to 0!  Some HW assisted
875 		 * checksumming requires this.
876 		 */
877 		ip->ip_sum = 0;
878 
879 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
880 			/*
881 			 * Perform any checksums that the hardware can't do
882 			 * for us.
883 			 *
884 			 * XXX Does any hardware require the {th,uh}_sum
885 			 * XXX fields to be 0?
886 			 */
887 			if (sw_csum & M_CSUM_IPv4) {
888 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
889 				ip->ip_sum = in_cksum(m, hlen);
890 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
891 			}
892 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
893 				if (IN_NEED_CHECKSUM(ifp,
894 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
895 					in_delayed_cksum(m);
896 				}
897 				m->m_pkthdr.csum_flags &=
898 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
899 			}
900 		}
901 
902 #ifdef IPSEC
903 		/* clean ipsec history once it goes out of the node */
904 		ipsec_delaux(m);
905 #endif
906 
907 		if (__predict_true(
908 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
909 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
910 			error =
911 			    (*ifp->if_output)(ifp, m, sintosa(dst), ro->ro_rt);
912 		} else {
913 			error =
914 			    ip_tso_output(ifp, m, sintosa(dst), ro->ro_rt);
915 		}
916 		goto done;
917 	}
918 
919 	/*
920 	 * We can't use HW checksumming if we're about to
921 	 * to fragment the packet.
922 	 *
923 	 * XXX Some hardware can do this.
924 	 */
925 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
926 		if (IN_NEED_CHECKSUM(ifp,
927 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
928 			in_delayed_cksum(m);
929 		}
930 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
931 	}
932 
933 	/*
934 	 * Too large for interface; fragment if possible.
935 	 * Must be able to put at least 8 bytes per fragment.
936 	 */
937 	if (ntohs(ip->ip_off) & IP_DF) {
938 		if (flags & IP_RETURNMTU)
939 			*mtu_p = mtu;
940 		error = EMSGSIZE;
941 		ipstat.ips_cantfrag++;
942 		goto bad;
943 	}
944 
945 	error = ip_fragment(m, ifp, mtu);
946 	if (error) {
947 		m = NULL;
948 		goto bad;
949 	}
950 
951 	for (; m; m = m0) {
952 		m0 = m->m_nextpkt;
953 		m->m_nextpkt = 0;
954 		if (error == 0) {
955 #if IFA_STATS
956 			if (ia)
957 				ia->ia_ifa.ifa_data.ifad_outbytes +=
958 				    ntohs(ip->ip_len);
959 #endif
960 #ifdef IPSEC
961 			/* clean ipsec history once it goes out of the node */
962 			ipsec_delaux(m);
963 #endif /* IPSEC */
964 
965 #ifdef IPSEC_NAT_T
966 			/*
967 			 * If we get there, the packet has not been handeld by
968 			 * IPSec whereas it should have. Now that it has been
969 			 * fragmented, re-inject it in ip_output so that IPsec
970 			 * processing can occur.
971 			 */
972 			if (natt_frag) {
973 				error = ip_output(m, opt,
974 				    ro, flags, imo, so, mtu_p);
975 			} else
976 #endif /* IPSEC_NAT_T */
977 			{
978 				KASSERT((m->m_pkthdr.csum_flags &
979 				    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
980 				error = (*ifp->if_output)(ifp, m, sintosa(dst),
981 				    ro->ro_rt);
982 			}
983 		} else
984 			m_freem(m);
985 	}
986 
987 	if (error == 0)
988 		ipstat.ips_fragmented++;
989 done:
990 	if (ro == &iproute && (flags & IP_ROUTETOIF) == 0 && ro->ro_rt) {
991 		RTFREE(ro->ro_rt);
992 		ro->ro_rt = 0;
993 	}
994 
995 #ifdef IPSEC
996 	if (sp != NULL) {
997 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
998 			printf("DP ip_output call free SP:%p\n", sp));
999 		key_freesp(sp);
1000 	}
1001 #endif /* IPSEC */
1002 #ifdef FAST_IPSEC
1003 	if (sp != NULL)
1004 		KEY_FREESP(&sp);
1005 #endif /* FAST_IPSEC */
1006 
1007 	return (error);
1008 bad:
1009 	m_freem(m);
1010 	goto done;
1011 }
1012 
1013 int
1014 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
1015 {
1016 	struct ip *ip, *mhip;
1017 	struct mbuf *m0;
1018 	int len, hlen, off;
1019 	int mhlen, firstlen;
1020 	struct mbuf **mnext;
1021 	int sw_csum = m->m_pkthdr.csum_flags;
1022 	int fragments = 0;
1023 	int s;
1024 	int error = 0;
1025 
1026 	ip = mtod(m, struct ip *);
1027 	hlen = ip->ip_hl << 2;
1028 	if (ifp != NULL)
1029 		sw_csum &= ~ifp->if_csum_flags_tx;
1030 
1031 	len = (mtu - hlen) &~ 7;
1032 	if (len < 8) {
1033 		m_freem(m);
1034 		return (EMSGSIZE);
1035 	}
1036 
1037 	firstlen = len;
1038 	mnext = &m->m_nextpkt;
1039 
1040 	/*
1041 	 * Loop through length of segment after first fragment,
1042 	 * make new header and copy data of each part and link onto chain.
1043 	 */
1044 	m0 = m;
1045 	mhlen = sizeof (struct ip);
1046 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
1047 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
1048 		if (m == 0) {
1049 			error = ENOBUFS;
1050 			ipstat.ips_odropped++;
1051 			goto sendorfree;
1052 		}
1053 		MCLAIM(m, m0->m_owner);
1054 		*mnext = m;
1055 		mnext = &m->m_nextpkt;
1056 		m->m_data += max_linkhdr;
1057 		mhip = mtod(m, struct ip *);
1058 		*mhip = *ip;
1059 		/* we must inherit MCAST and BCAST flags */
1060 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
1061 		if (hlen > sizeof (struct ip)) {
1062 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1063 			mhip->ip_hl = mhlen >> 2;
1064 		}
1065 		m->m_len = mhlen;
1066 		mhip->ip_off = ((off - hlen) >> 3) +
1067 		    (ntohs(ip->ip_off) & ~IP_MF);
1068 		if (ip->ip_off & htons(IP_MF))
1069 			mhip->ip_off |= IP_MF;
1070 		if (off + len >= ntohs(ip->ip_len))
1071 			len = ntohs(ip->ip_len) - off;
1072 		else
1073 			mhip->ip_off |= IP_MF;
1074 		HTONS(mhip->ip_off);
1075 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
1076 		m->m_next = m_copy(m0, off, len);
1077 		if (m->m_next == 0) {
1078 			error = ENOBUFS;	/* ??? */
1079 			ipstat.ips_odropped++;
1080 			goto sendorfree;
1081 		}
1082 		m->m_pkthdr.len = mhlen + len;
1083 		m->m_pkthdr.rcvif = (struct ifnet *)0;
1084 		mhip->ip_sum = 0;
1085 		if (sw_csum & M_CSUM_IPv4) {
1086 			mhip->ip_sum = in_cksum(m, mhlen);
1087 			KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
1088 		} else {
1089 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1090 			m->m_pkthdr.csum_data |= mhlen << 16;
1091 		}
1092 		ipstat.ips_ofragments++;
1093 		fragments++;
1094 	}
1095 	/*
1096 	 * Update first fragment by trimming what's been copied out
1097 	 * and updating header, then send each fragment (in order).
1098 	 */
1099 	m = m0;
1100 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
1101 	m->m_pkthdr.len = hlen + firstlen;
1102 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
1103 	ip->ip_off |= htons(IP_MF);
1104 	ip->ip_sum = 0;
1105 	if (sw_csum & M_CSUM_IPv4) {
1106 		ip->ip_sum = in_cksum(m, hlen);
1107 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
1108 	} else {
1109 		KASSERT(m->m_pkthdr.csum_flags & M_CSUM_IPv4);
1110 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
1111 			sizeof(struct ip));
1112 	}
1113 sendorfree:
1114 	/*
1115 	 * If there is no room for all the fragments, don't queue
1116 	 * any of them.
1117 	 */
1118 	if (ifp != NULL) {
1119 		s = splnet();
1120 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
1121 		    error == 0) {
1122 			error = ENOBUFS;
1123 			ipstat.ips_odropped++;
1124 			IFQ_INC_DROPS(&ifp->if_snd);
1125 		}
1126 		splx(s);
1127 	}
1128 	if (error) {
1129 		for (m = m0; m; m = m0) {
1130 			m0 = m->m_nextpkt;
1131 			m->m_nextpkt = NULL;
1132 			m_freem(m);
1133 		}
1134 	}
1135 	return (error);
1136 }
1137 
1138 /*
1139  * Process a delayed payload checksum calculation.
1140  */
1141 void
1142 in_delayed_cksum(struct mbuf *m)
1143 {
1144 	struct ip *ip;
1145 	u_int16_t csum, offset;
1146 
1147 	ip = mtod(m, struct ip *);
1148 	offset = ip->ip_hl << 2;
1149 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
1150 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
1151 		csum = 0xffff;
1152 
1153 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
1154 
1155 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
1156 		/* This happen when ip options were inserted
1157 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
1158 		    m->m_len, offset, ip->ip_p);
1159 		 */
1160 		m_copyback(m, offset, sizeof(csum), (caddr_t) &csum);
1161 	} else
1162 		*(u_int16_t *)(mtod(m, caddr_t) + offset) = csum;
1163 }
1164 
1165 /*
1166  * Determine the maximum length of the options to be inserted;
1167  * we would far rather allocate too much space rather than too little.
1168  */
1169 
1170 u_int
1171 ip_optlen(struct inpcb *inp)
1172 {
1173 	struct mbuf *m = inp->inp_options;
1174 
1175 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst))
1176 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
1177 	else
1178 		return 0;
1179 }
1180 
1181 
1182 /*
1183  * Insert IP options into preformed packet.
1184  * Adjust IP destination as required for IP source routing,
1185  * as indicated by a non-zero in_addr at the start of the options.
1186  */
1187 static struct mbuf *
1188 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1189 {
1190 	struct ipoption *p = mtod(opt, struct ipoption *);
1191 	struct mbuf *n;
1192 	struct ip *ip = mtod(m, struct ip *);
1193 	unsigned optlen;
1194 
1195 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1196 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
1197 		return (m);		/* XXX should fail */
1198 	if (!in_nullhost(p->ipopt_dst))
1199 		ip->ip_dst = p->ipopt_dst;
1200 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
1201 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1202 		if (n == 0)
1203 			return (m);
1204 		MCLAIM(n, m->m_owner);
1205 		M_MOVE_PKTHDR(n, m);
1206 		m->m_len -= sizeof(struct ip);
1207 		m->m_data += sizeof(struct ip);
1208 		n->m_next = m;
1209 		m = n;
1210 		m->m_len = optlen + sizeof(struct ip);
1211 		m->m_data += max_linkhdr;
1212 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1213 	} else {
1214 		m->m_data -= optlen;
1215 		m->m_len += optlen;
1216 		memmove(mtod(m, caddr_t), ip, sizeof(struct ip));
1217 	}
1218 	m->m_pkthdr.len += optlen;
1219 	ip = mtod(m, struct ip *);
1220 	bcopy((caddr_t)p->ipopt_list, (caddr_t)(ip + 1), (unsigned)optlen);
1221 	*phlen = sizeof(struct ip) + optlen;
1222 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1223 	return (m);
1224 }
1225 
1226 /*
1227  * Copy options from ip to jp,
1228  * omitting those not copied during fragmentation.
1229  */
1230 int
1231 ip_optcopy(struct ip *ip, struct ip *jp)
1232 {
1233 	u_char *cp, *dp;
1234 	int opt, optlen, cnt;
1235 
1236 	cp = (u_char *)(ip + 1);
1237 	dp = (u_char *)(jp + 1);
1238 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1239 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1240 		opt = cp[0];
1241 		if (opt == IPOPT_EOL)
1242 			break;
1243 		if (opt == IPOPT_NOP) {
1244 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1245 			*dp++ = IPOPT_NOP;
1246 			optlen = 1;
1247 			continue;
1248 		}
1249 #ifdef DIAGNOSTIC
1250 		if (cnt < IPOPT_OLEN + sizeof(*cp))
1251 			panic("malformed IPv4 option passed to ip_optcopy");
1252 #endif
1253 		optlen = cp[IPOPT_OLEN];
1254 #ifdef DIAGNOSTIC
1255 		if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1256 			panic("malformed IPv4 option passed to ip_optcopy");
1257 #endif
1258 		/* bogus lengths should have been caught by ip_dooptions */
1259 		if (optlen > cnt)
1260 			optlen = cnt;
1261 		if (IPOPT_COPIED(opt)) {
1262 			bcopy((caddr_t)cp, (caddr_t)dp, (unsigned)optlen);
1263 			dp += optlen;
1264 		}
1265 	}
1266 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1267 		*dp++ = IPOPT_EOL;
1268 	return (optlen);
1269 }
1270 
1271 /*
1272  * IP socket option processing.
1273  */
1274 int
1275 ip_ctloutput(int op, struct socket *so, int level, int optname,
1276     struct mbuf **mp)
1277 {
1278 	struct inpcb *inp = sotoinpcb(so);
1279 	struct mbuf *m = *mp;
1280 	int optval = 0;
1281 	int error = 0;
1282 #if defined(IPSEC) || defined(FAST_IPSEC)
1283 	struct lwp *l = curlwp;	/*XXX*/
1284 #endif
1285 
1286 	if (level != IPPROTO_IP) {
1287 		error = EINVAL;
1288 		if (op == PRCO_SETOPT && *mp)
1289 			(void) m_free(*mp);
1290 	} else switch (op) {
1291 
1292 	case PRCO_SETOPT:
1293 		switch (optname) {
1294 		case IP_OPTIONS:
1295 #ifdef notyet
1296 		case IP_RETOPTS:
1297 			return (ip_pcbopts(optname, &inp->inp_options, m));
1298 #else
1299 			return (ip_pcbopts(&inp->inp_options, m));
1300 #endif
1301 
1302 		case IP_TOS:
1303 		case IP_TTL:
1304 		case IP_RECVOPTS:
1305 		case IP_RECVRETOPTS:
1306 		case IP_RECVDSTADDR:
1307 		case IP_RECVIF:
1308 			if (m == NULL || m->m_len != sizeof(int))
1309 				error = EINVAL;
1310 			else {
1311 				optval = *mtod(m, int *);
1312 				switch (optname) {
1313 
1314 				case IP_TOS:
1315 					inp->inp_ip.ip_tos = optval;
1316 					break;
1317 
1318 				case IP_TTL:
1319 					inp->inp_ip.ip_ttl = optval;
1320 					break;
1321 #define	OPTSET(bit) \
1322 	if (optval) \
1323 		inp->inp_flags |= bit; \
1324 	else \
1325 		inp->inp_flags &= ~bit;
1326 
1327 				case IP_RECVOPTS:
1328 					OPTSET(INP_RECVOPTS);
1329 					break;
1330 
1331 				case IP_RECVRETOPTS:
1332 					OPTSET(INP_RECVRETOPTS);
1333 					break;
1334 
1335 				case IP_RECVDSTADDR:
1336 					OPTSET(INP_RECVDSTADDR);
1337 					break;
1338 
1339 				case IP_RECVIF:
1340 					OPTSET(INP_RECVIF);
1341 					break;
1342 				}
1343 			}
1344 			break;
1345 #undef OPTSET
1346 
1347 		case IP_MULTICAST_IF:
1348 		case IP_MULTICAST_TTL:
1349 		case IP_MULTICAST_LOOP:
1350 		case IP_ADD_MEMBERSHIP:
1351 		case IP_DROP_MEMBERSHIP:
1352 			error = ip_setmoptions(optname, &inp->inp_moptions, m);
1353 			break;
1354 
1355 		case IP_PORTRANGE:
1356 			if (m == 0 || m->m_len != sizeof(int))
1357 				error = EINVAL;
1358 			else {
1359 				optval = *mtod(m, int *);
1360 
1361 				switch (optval) {
1362 
1363 				case IP_PORTRANGE_DEFAULT:
1364 				case IP_PORTRANGE_HIGH:
1365 					inp->inp_flags &= ~(INP_LOWPORT);
1366 					break;
1367 
1368 				case IP_PORTRANGE_LOW:
1369 					inp->inp_flags |= INP_LOWPORT;
1370 					break;
1371 
1372 				default:
1373 					error = EINVAL;
1374 					break;
1375 				}
1376 			}
1377 			break;
1378 
1379 #if defined(IPSEC) || defined(FAST_IPSEC)
1380 		case IP_IPSEC_POLICY:
1381 		{
1382 			caddr_t req = NULL;
1383 			size_t len = 0;
1384 			int priv = 0;
1385 
1386 #ifdef __NetBSD__
1387 			if (l == 0 || kauth_authorize_generic(l->l_cred,
1388 			    KAUTH_GENERIC_ISSUSER, &l->l_acflag))
1389 				priv = 0;
1390 			else
1391 				priv = 1;
1392 #else
1393 			priv = (in6p->in6p_socket->so_state & SS_PRIV);
1394 #endif
1395 			if (m) {
1396 				req = mtod(m, caddr_t);
1397 				len = m->m_len;
1398 			}
1399 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1400 			break;
1401 		    }
1402 #endif /*IPSEC*/
1403 
1404 		default:
1405 			error = ENOPROTOOPT;
1406 			break;
1407 		}
1408 		if (m)
1409 			(void)m_free(m);
1410 		break;
1411 
1412 	case PRCO_GETOPT:
1413 		switch (optname) {
1414 		case IP_OPTIONS:
1415 		case IP_RETOPTS:
1416 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1417 			MCLAIM(m, so->so_mowner);
1418 			if (inp->inp_options) {
1419 				m->m_len = inp->inp_options->m_len;
1420 				bcopy(mtod(inp->inp_options, caddr_t),
1421 				    mtod(m, caddr_t), (unsigned)m->m_len);
1422 			} else
1423 				m->m_len = 0;
1424 			break;
1425 
1426 		case IP_TOS:
1427 		case IP_TTL:
1428 		case IP_RECVOPTS:
1429 		case IP_RECVRETOPTS:
1430 		case IP_RECVDSTADDR:
1431 		case IP_RECVIF:
1432 		case IP_ERRORMTU:
1433 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1434 			MCLAIM(m, so->so_mowner);
1435 			m->m_len = sizeof(int);
1436 			switch (optname) {
1437 
1438 			case IP_TOS:
1439 				optval = inp->inp_ip.ip_tos;
1440 				break;
1441 
1442 			case IP_TTL:
1443 				optval = inp->inp_ip.ip_ttl;
1444 				break;
1445 
1446 			case IP_ERRORMTU:
1447 				optval = inp->inp_errormtu;
1448 				break;
1449 
1450 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1451 
1452 			case IP_RECVOPTS:
1453 				optval = OPTBIT(INP_RECVOPTS);
1454 				break;
1455 
1456 			case IP_RECVRETOPTS:
1457 				optval = OPTBIT(INP_RECVRETOPTS);
1458 				break;
1459 
1460 			case IP_RECVDSTADDR:
1461 				optval = OPTBIT(INP_RECVDSTADDR);
1462 				break;
1463 
1464 			case IP_RECVIF:
1465 				optval = OPTBIT(INP_RECVIF);
1466 				break;
1467 			}
1468 			*mtod(m, int *) = optval;
1469 			break;
1470 
1471 #if 0	/* defined(IPSEC) || defined(FAST_IPSEC) */
1472 		/* XXX: code broken */
1473 		case IP_IPSEC_POLICY:
1474 		{
1475 			caddr_t req = NULL;
1476 			size_t len = 0;
1477 
1478 			if (m) {
1479 				req = mtod(m, caddr_t);
1480 				len = m->m_len;
1481 			}
1482 			error = ipsec4_get_policy(inp, req, len, mp);
1483 			break;
1484 		}
1485 #endif /*IPSEC*/
1486 
1487 		case IP_MULTICAST_IF:
1488 		case IP_MULTICAST_TTL:
1489 		case IP_MULTICAST_LOOP:
1490 		case IP_ADD_MEMBERSHIP:
1491 		case IP_DROP_MEMBERSHIP:
1492 			error = ip_getmoptions(optname, inp->inp_moptions, mp);
1493 			if (*mp)
1494 				MCLAIM(*mp, so->so_mowner);
1495 			break;
1496 
1497 		case IP_PORTRANGE:
1498 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
1499 			MCLAIM(m, so->so_mowner);
1500 			m->m_len = sizeof(int);
1501 
1502 			if (inp->inp_flags & INP_LOWPORT)
1503 				optval = IP_PORTRANGE_LOW;
1504 			else
1505 				optval = IP_PORTRANGE_DEFAULT;
1506 
1507 			*mtod(m, int *) = optval;
1508 			break;
1509 
1510 		default:
1511 			error = ENOPROTOOPT;
1512 			break;
1513 		}
1514 		break;
1515 	}
1516 	return (error);
1517 }
1518 
1519 /*
1520  * Set up IP options in pcb for insertion in output packets.
1521  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1522  * with destination address if source routed.
1523  */
1524 int
1525 #ifdef notyet
1526 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1527 #else
1528 ip_pcbopts(struct mbuf **pcbopt, struct mbuf *m)
1529 #endif
1530 {
1531 	int cnt, optlen;
1532 	u_char *cp;
1533 	u_char opt;
1534 
1535 	/* turn off any old options */
1536 	if (*pcbopt)
1537 		(void)m_free(*pcbopt);
1538 	*pcbopt = 0;
1539 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1540 		/*
1541 		 * Only turning off any previous options.
1542 		 */
1543 		if (m)
1544 			(void)m_free(m);
1545 		return (0);
1546 	}
1547 
1548 #ifndef	__vax__
1549 	if (m->m_len % sizeof(int32_t))
1550 		goto bad;
1551 #endif
1552 	/*
1553 	 * IP first-hop destination address will be stored before
1554 	 * actual options; move other options back
1555 	 * and clear it when none present.
1556 	 */
1557 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1558 		goto bad;
1559 	cnt = m->m_len;
1560 	m->m_len += sizeof(struct in_addr);
1561 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1562 	memmove(cp, mtod(m, caddr_t), (unsigned)cnt);
1563 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1564 
1565 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1566 		opt = cp[IPOPT_OPTVAL];
1567 		if (opt == IPOPT_EOL)
1568 			break;
1569 		if (opt == IPOPT_NOP)
1570 			optlen = 1;
1571 		else {
1572 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1573 				goto bad;
1574 			optlen = cp[IPOPT_OLEN];
1575 			if (optlen < IPOPT_OLEN  + sizeof(*cp) || optlen > cnt)
1576 				goto bad;
1577 		}
1578 		switch (opt) {
1579 
1580 		default:
1581 			break;
1582 
1583 		case IPOPT_LSRR:
1584 		case IPOPT_SSRR:
1585 			/*
1586 			 * user process specifies route as:
1587 			 *	->A->B->C->D
1588 			 * D must be our final destination (but we can't
1589 			 * check that since we may not have connected yet).
1590 			 * A is first hop destination, which doesn't appear in
1591 			 * actual IP option, but is stored before the options.
1592 			 */
1593 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1594 				goto bad;
1595 			m->m_len -= sizeof(struct in_addr);
1596 			cnt -= sizeof(struct in_addr);
1597 			optlen -= sizeof(struct in_addr);
1598 			cp[IPOPT_OLEN] = optlen;
1599 			/*
1600 			 * Move first hop before start of options.
1601 			 */
1602 			bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1603 			    sizeof(struct in_addr));
1604 			/*
1605 			 * Then copy rest of options back
1606 			 * to close up the deleted entry.
1607 			 */
1608 			(void)memmove(&cp[IPOPT_OFFSET+1],
1609 			    &cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1610 			    (unsigned)cnt - (IPOPT_MINOFF - 1));
1611 			break;
1612 		}
1613 	}
1614 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1615 		goto bad;
1616 	*pcbopt = m;
1617 	return (0);
1618 
1619 bad:
1620 	(void)m_free(m);
1621 	return (EINVAL);
1622 }
1623 
1624 /*
1625  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1626  */
1627 static struct ifnet *
1628 ip_multicast_if(struct in_addr *a, int *ifindexp)
1629 {
1630 	int ifindex;
1631 	struct ifnet *ifp = NULL;
1632 	struct in_ifaddr *ia;
1633 
1634 	if (ifindexp)
1635 		*ifindexp = 0;
1636 	if (ntohl(a->s_addr) >> 24 == 0) {
1637 		ifindex = ntohl(a->s_addr) & 0xffffff;
1638 		if (ifindex < 0 || if_indexlim <= ifindex)
1639 			return NULL;
1640 		ifp = ifindex2ifnet[ifindex];
1641 		if (!ifp)
1642 			return NULL;
1643 		if (ifindexp)
1644 			*ifindexp = ifindex;
1645 	} else {
1646 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1647 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1648 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1649 				ifp = ia->ia_ifp;
1650 				break;
1651 			}
1652 		}
1653 	}
1654 	return ifp;
1655 }
1656 
1657 static int
1658 ip_getoptval(struct mbuf *m, u_int8_t *val, u_int maxval)
1659 {
1660 	u_int tval;
1661 
1662 	if (m == NULL)
1663 		return EINVAL;
1664 
1665 	switch (m->m_len) {
1666 	case sizeof(u_char):
1667 		tval = *(mtod(m, u_char *));
1668 		break;
1669 	case sizeof(u_int):
1670 		tval = *(mtod(m, u_int *));
1671 		break;
1672 	default:
1673 		return EINVAL;
1674 	}
1675 
1676 	if (tval > maxval)
1677 		return EINVAL;
1678 
1679 	*val = tval;
1680 	return 0;
1681 }
1682 
1683 /*
1684  * Set the IP multicast options in response to user setsockopt().
1685  */
1686 int
1687 ip_setmoptions(int optname, struct ip_moptions **imop, struct mbuf *m)
1688 {
1689 	int error = 0;
1690 	int i;
1691 	struct in_addr addr;
1692 	struct ip_mreq *mreq;
1693 	struct ifnet *ifp;
1694 	struct ip_moptions *imo = *imop;
1695 	struct route ro;
1696 	struct sockaddr_in *dst;
1697 	int ifindex;
1698 
1699 	if (imo == NULL) {
1700 		/*
1701 		 * No multicast option buffer attached to the pcb;
1702 		 * allocate one and initialize to default values.
1703 		 */
1704 		imo = (struct ip_moptions *)malloc(sizeof(*imo), M_IPMOPTS,
1705 		    M_WAITOK);
1706 
1707 		if (imo == NULL)
1708 			return (ENOBUFS);
1709 		*imop = imo;
1710 		imo->imo_multicast_ifp = NULL;
1711 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1712 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1713 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1714 		imo->imo_num_memberships = 0;
1715 	}
1716 
1717 	switch (optname) {
1718 
1719 	case IP_MULTICAST_IF:
1720 		/*
1721 		 * Select the interface for outgoing multicast packets.
1722 		 */
1723 		if (m == NULL || m->m_len != sizeof(struct in_addr)) {
1724 			error = EINVAL;
1725 			break;
1726 		}
1727 		addr = *(mtod(m, struct in_addr *));
1728 		/*
1729 		 * INADDR_ANY is used to remove a previous selection.
1730 		 * When no interface is selected, a default one is
1731 		 * chosen every time a multicast packet is sent.
1732 		 */
1733 		if (in_nullhost(addr)) {
1734 			imo->imo_multicast_ifp = NULL;
1735 			break;
1736 		}
1737 		/*
1738 		 * The selected interface is identified by its local
1739 		 * IP address.  Find the interface and confirm that
1740 		 * it supports multicasting.
1741 		 */
1742 		ifp = ip_multicast_if(&addr, &ifindex);
1743 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1744 			error = EADDRNOTAVAIL;
1745 			break;
1746 		}
1747 		imo->imo_multicast_ifp = ifp;
1748 		if (ifindex)
1749 			imo->imo_multicast_addr = addr;
1750 		else
1751 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1752 		break;
1753 
1754 	case IP_MULTICAST_TTL:
1755 		/*
1756 		 * Set the IP time-to-live for outgoing multicast packets.
1757 		 */
1758 		error = ip_getoptval(m, &imo->imo_multicast_ttl, MAXTTL);
1759 		break;
1760 
1761 	case IP_MULTICAST_LOOP:
1762 		/*
1763 		 * Set the loopback flag for outgoing multicast packets.
1764 		 * Must be zero or one.
1765 		 */
1766 		error = ip_getoptval(m, &imo->imo_multicast_loop, 1);
1767 		break;
1768 
1769 	case IP_ADD_MEMBERSHIP:
1770 		/*
1771 		 * Add a multicast group membership.
1772 		 * Group must be a valid IP multicast address.
1773 		 */
1774 		if (m == NULL || m->m_len != sizeof(struct ip_mreq)) {
1775 			error = EINVAL;
1776 			break;
1777 		}
1778 		mreq = mtod(m, struct ip_mreq *);
1779 		if (!IN_MULTICAST(mreq->imr_multiaddr.s_addr)) {
1780 			error = EINVAL;
1781 			break;
1782 		}
1783 		/*
1784 		 * If no interface address was provided, use the interface of
1785 		 * the route to the given multicast address.
1786 		 */
1787 		if (in_nullhost(mreq->imr_interface)) {
1788 			bzero((caddr_t)&ro, sizeof(ro));
1789 			ro.ro_rt = NULL;
1790 			dst = satosin(&ro.ro_dst);
1791 			dst->sin_len = sizeof(*dst);
1792 			dst->sin_family = AF_INET;
1793 			dst->sin_addr = mreq->imr_multiaddr;
1794 			rtalloc(&ro);
1795 			if (ro.ro_rt == NULL) {
1796 				error = EADDRNOTAVAIL;
1797 				break;
1798 			}
1799 			ifp = ro.ro_rt->rt_ifp;
1800 			rtfree(ro.ro_rt);
1801 		} else {
1802 			ifp = ip_multicast_if(&mreq->imr_interface, NULL);
1803 		}
1804 		/*
1805 		 * See if we found an interface, and confirm that it
1806 		 * supports multicast.
1807 		 */
1808 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1809 			error = EADDRNOTAVAIL;
1810 			break;
1811 		}
1812 		/*
1813 		 * See if the membership already exists or if all the
1814 		 * membership slots are full.
1815 		 */
1816 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1817 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1818 			    in_hosteq(imo->imo_membership[i]->inm_addr,
1819 				      mreq->imr_multiaddr))
1820 				break;
1821 		}
1822 		if (i < imo->imo_num_memberships) {
1823 			error = EADDRINUSE;
1824 			break;
1825 		}
1826 		if (i == IP_MAX_MEMBERSHIPS) {
1827 			error = ETOOMANYREFS;
1828 			break;
1829 		}
1830 		/*
1831 		 * Everything looks good; add a new record to the multicast
1832 		 * address list for the given interface.
1833 		 */
1834 		if ((imo->imo_membership[i] =
1835 		    in_addmulti(&mreq->imr_multiaddr, ifp)) == NULL) {
1836 			error = ENOBUFS;
1837 			break;
1838 		}
1839 		++imo->imo_num_memberships;
1840 		break;
1841 
1842 	case IP_DROP_MEMBERSHIP:
1843 		/*
1844 		 * Drop a multicast group membership.
1845 		 * Group must be a valid IP multicast address.
1846 		 */
1847 		if (m == NULL || m->m_len != sizeof(struct ip_mreq)) {
1848 			error = EINVAL;
1849 			break;
1850 		}
1851 		mreq = mtod(m, struct ip_mreq *);
1852 		if (!IN_MULTICAST(mreq->imr_multiaddr.s_addr)) {
1853 			error = EINVAL;
1854 			break;
1855 		}
1856 		/*
1857 		 * If an interface address was specified, get a pointer
1858 		 * to its ifnet structure.
1859 		 */
1860 		if (in_nullhost(mreq->imr_interface))
1861 			ifp = NULL;
1862 		else {
1863 			ifp = ip_multicast_if(&mreq->imr_interface, NULL);
1864 			if (ifp == NULL) {
1865 				error = EADDRNOTAVAIL;
1866 				break;
1867 			}
1868 		}
1869 		/*
1870 		 * Find the membership in the membership array.
1871 		 */
1872 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1873 			if ((ifp == NULL ||
1874 			     imo->imo_membership[i]->inm_ifp == ifp) &&
1875 			     in_hosteq(imo->imo_membership[i]->inm_addr,
1876 				       mreq->imr_multiaddr))
1877 				break;
1878 		}
1879 		if (i == imo->imo_num_memberships) {
1880 			error = EADDRNOTAVAIL;
1881 			break;
1882 		}
1883 		/*
1884 		 * Give up the multicast address record to which the
1885 		 * membership points.
1886 		 */
1887 		in_delmulti(imo->imo_membership[i]);
1888 		/*
1889 		 * Remove the gap in the membership array.
1890 		 */
1891 		for (++i; i < imo->imo_num_memberships; ++i)
1892 			imo->imo_membership[i-1] = imo->imo_membership[i];
1893 		--imo->imo_num_memberships;
1894 		break;
1895 
1896 	default:
1897 		error = EOPNOTSUPP;
1898 		break;
1899 	}
1900 
1901 	/*
1902 	 * If all options have default values, no need to keep the mbuf.
1903 	 */
1904 	if (imo->imo_multicast_ifp == NULL &&
1905 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1906 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1907 	    imo->imo_num_memberships == 0) {
1908 		free(*imop, M_IPMOPTS);
1909 		*imop = NULL;
1910 	}
1911 
1912 	return (error);
1913 }
1914 
1915 /*
1916  * Return the IP multicast options in response to user getsockopt().
1917  */
1918 int
1919 ip_getmoptions(int optname, struct ip_moptions *imo, struct mbuf **mp)
1920 {
1921 	u_char *ttl;
1922 	u_char *loop;
1923 	struct in_addr *addr;
1924 	struct in_ifaddr *ia;
1925 
1926 	*mp = m_get(M_WAIT, MT_SOOPTS);
1927 
1928 	switch (optname) {
1929 
1930 	case IP_MULTICAST_IF:
1931 		addr = mtod(*mp, struct in_addr *);
1932 		(*mp)->m_len = sizeof(struct in_addr);
1933 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
1934 			*addr = zeroin_addr;
1935 		else if (imo->imo_multicast_addr.s_addr) {
1936 			/* return the value user has set */
1937 			*addr = imo->imo_multicast_addr;
1938 		} else {
1939 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
1940 			*addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1941 		}
1942 		return (0);
1943 
1944 	case IP_MULTICAST_TTL:
1945 		ttl = mtod(*mp, u_char *);
1946 		(*mp)->m_len = 1;
1947 		*ttl = imo ? imo->imo_multicast_ttl
1948 			   : IP_DEFAULT_MULTICAST_TTL;
1949 		return (0);
1950 
1951 	case IP_MULTICAST_LOOP:
1952 		loop = mtod(*mp, u_char *);
1953 		(*mp)->m_len = 1;
1954 		*loop = imo ? imo->imo_multicast_loop
1955 			    : IP_DEFAULT_MULTICAST_LOOP;
1956 		return (0);
1957 
1958 	default:
1959 		return (EOPNOTSUPP);
1960 	}
1961 }
1962 
1963 /*
1964  * Discard the IP multicast options.
1965  */
1966 void
1967 ip_freemoptions(struct ip_moptions *imo)
1968 {
1969 	int i;
1970 
1971 	if (imo != NULL) {
1972 		for (i = 0; i < imo->imo_num_memberships; ++i)
1973 			in_delmulti(imo->imo_membership[i]);
1974 		free(imo, M_IPMOPTS);
1975 	}
1976 }
1977 
1978 /*
1979  * Routine called from ip_output() to loop back a copy of an IP multicast
1980  * packet to the input queue of a specified interface.  Note that this
1981  * calls the output routine of the loopback "driver", but with an interface
1982  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1983  */
1984 static void
1985 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst)
1986 {
1987 	struct ip *ip;
1988 	struct mbuf *copym;
1989 
1990 	copym = m_copy(m, 0, M_COPYALL);
1991 	if (copym != NULL
1992 	 && (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1993 		copym = m_pullup(copym, sizeof(struct ip));
1994 	if (copym != NULL) {
1995 		/*
1996 		 * We don't bother to fragment if the IP length is greater
1997 		 * than the interface's MTU.  Can this possibly matter?
1998 		 */
1999 		ip = mtod(copym, struct ip *);
2000 
2001 		if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
2002 			in_delayed_cksum(copym);
2003 			copym->m_pkthdr.csum_flags &=
2004 			    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
2005 		}
2006 
2007 		ip->ip_sum = 0;
2008 		ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
2009 		(void) looutput(ifp, copym, sintosa(dst), NULL);
2010 	}
2011 }
2012