xref: /netbsd-src/sys/netinet/ip_output.c (revision 87d689fb734c654d2486f87f7be32f1b53ecdbec)
1 /*	$NetBSD: ip_output.c,v 1.292 2018/01/10 18:51:31 christos 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  *
49  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
50  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
51  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
52  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
53  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
54  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
55  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
56  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
57  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
58  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
59  * POSSIBILITY OF SUCH DAMAGE.
60  */
61 
62 /*
63  * Copyright (c) 1982, 1986, 1988, 1990, 1993
64  *	The Regents of the University of California.  All rights reserved.
65  *
66  * Redistribution and use in source and binary forms, with or without
67  * modification, are permitted provided that the following conditions
68  * are met:
69  * 1. Redistributions of source code must retain the above copyright
70  *    notice, this list of conditions and the following disclaimer.
71  * 2. Redistributions in binary form must reproduce the above copyright
72  *    notice, this list of conditions and the following disclaimer in the
73  *    documentation and/or other materials provided with the distribution.
74  * 3. Neither the name of the University nor the names of its contributors
75  *    may be used to endorse or promote products derived from this software
76  *    without specific prior written permission.
77  *
78  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88  * SUCH DAMAGE.
89  *
90  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: ip_output.c,v 1.292 2018/01/10 18:51:31 christos Exp $");
95 
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_mrouting.h"
100 #include "opt_net_mpsafe.h"
101 #include "opt_mpls.h"
102 #endif
103 
104 #include "arp.h"
105 
106 #include <sys/param.h>
107 #include <sys/kmem.h>
108 #include <sys/mbuf.h>
109 #include <sys/socket.h>
110 #include <sys/socketvar.h>
111 #include <sys/kauth.h>
112 #include <sys/systm.h>
113 #include <sys/syslog.h>
114 
115 #include <net/if.h>
116 #include <net/if_types.h>
117 #include <net/route.h>
118 #include <net/pfil.h>
119 
120 #include <netinet/in.h>
121 #include <netinet/in_systm.h>
122 #include <netinet/ip.h>
123 #include <netinet/in_pcb.h>
124 #include <netinet/in_var.h>
125 #include <netinet/ip_var.h>
126 #include <netinet/ip_private.h>
127 #include <netinet/in_offload.h>
128 #include <netinet/portalgo.h>
129 #include <netinet/udp.h>
130 #include <netinet/udp_var.h>
131 
132 #ifdef INET6
133 #include <netinet6/ip6_var.h>
134 #endif
135 
136 #ifdef MROUTING
137 #include <netinet/ip_mroute.h>
138 #endif
139 
140 #ifdef IPSEC
141 #include <netipsec/ipsec.h>
142 #include <netipsec/key.h>
143 #endif
144 
145 #ifdef MPLS
146 #include <netmpls/mpls.h>
147 #include <netmpls/mpls_var.h>
148 #endif
149 
150 static int ip_pcbopts(struct inpcb *, const struct sockopt *);
151 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
152 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
153 static void ip_mloopback(struct ifnet *, struct mbuf *,
154     const struct sockaddr_in *);
155 static int ip_ifaddrvalid(const struct in_ifaddr *);
156 
157 extern pfil_head_t *inet_pfil_hook;			/* XXX */
158 
159 int	ip_do_loopback_cksum = 0;
160 
161 static int
162 ip_mark_mpls(struct ifnet * const ifp, struct mbuf * const m,
163     const struct rtentry *rt)
164 {
165 	int error = 0;
166 #ifdef MPLS
167 	union mpls_shim msh;
168 
169 	if (rt == NULL || rt_gettag(rt) == NULL ||
170 	    rt_gettag(rt)->sa_family != AF_MPLS ||
171 	    (m->m_flags & (M_MCAST | M_BCAST)) != 0 ||
172 	    ifp->if_type != IFT_ETHER)
173 		return 0;
174 
175 	msh.s_addr = MPLS_GETSADDR(rt);
176 	if (msh.shim.label != MPLS_LABEL_IMPLNULL) {
177 		struct m_tag *mtag;
178 		/*
179 		 * XXX tentative solution to tell ether_output
180 		 * it's MPLS. Need some more efficient solution.
181 		 */
182 		mtag = m_tag_get(PACKET_TAG_MPLS,
183 		    sizeof(int) /* dummy */,
184 		    M_NOWAIT);
185 		if (mtag == NULL)
186 			return ENOMEM;
187 		m_tag_prepend(m, mtag);
188 	}
189 #endif
190 	return error;
191 }
192 
193 /*
194  * Send an IP packet to a host.
195  */
196 int
197 ip_if_output(struct ifnet * const ifp, struct mbuf * const m,
198     const struct sockaddr * const dst, const struct rtentry *rt)
199 {
200 	int error = 0;
201 
202 	if (rt != NULL) {
203 		error = rt_check_reject_route(rt, ifp);
204 		if (error != 0) {
205 			m_freem(m);
206 			return error;
207 		}
208 	}
209 
210 	error = ip_mark_mpls(ifp, m, rt);
211 	if (error != 0) {
212 		m_freem(m);
213 		return error;
214 	}
215 
216 	error = if_output_lock(ifp, ifp, m, dst, rt);
217 
218 	return error;
219 }
220 
221 /*
222  * IP output.  The packet in mbuf chain m contains a skeletal IP
223  * header (with len, off, ttl, proto, tos, src, dst).
224  * The mbuf chain containing the packet will be freed.
225  * The mbuf opt, if present, will not be freed.
226  */
227 int
228 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro, int flags,
229     struct ip_moptions *imo, struct inpcb *inp)
230 {
231 	struct rtentry *rt;
232 	struct ip *ip;
233 	struct ifnet *ifp, *mifp = NULL;
234 	struct mbuf *m = m0;
235 	int hlen = sizeof (struct ip);
236 	int len, error = 0;
237 	struct route iproute;
238 	const struct sockaddr_in *dst;
239 	struct in_ifaddr *ia = NULL;
240 	struct ifaddr *ifa;
241 	int isbroadcast;
242 	int sw_csum;
243 	u_long mtu;
244 	bool natt_frag = false;
245 	bool rtmtu_nolock;
246 	union {
247 		struct sockaddr		sa;
248 		struct sockaddr_in	sin;
249 	} udst, usrc;
250 	struct sockaddr *rdst = &udst.sa;	/* real IP destination, as
251 						 * opposed to the nexthop
252 						 */
253 	struct psref psref, psref_ia;
254 	int bound;
255 	bool bind_need_restore = false;
256 
257 	len = 0;
258 
259 	MCLAIM(m, &ip_tx_mowner);
260 
261 	KASSERT((m->m_flags & M_PKTHDR) != 0);
262 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
263 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
264 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
265 
266 	if (opt) {
267 		m = ip_insertoptions(m, opt, &len);
268 		if (len >= sizeof(struct ip))
269 			hlen = len;
270 	}
271 	ip = mtod(m, struct ip *);
272 
273 	/*
274 	 * Fill in IP header.
275 	 */
276 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
277 		ip->ip_v = IPVERSION;
278 		ip->ip_off = htons(0);
279 		/* ip->ip_id filled in after we find out source ia */
280 		ip->ip_hl = hlen >> 2;
281 		IP_STATINC(IP_STAT_LOCALOUT);
282 	} else {
283 		hlen = ip->ip_hl << 2;
284 	}
285 
286 	/*
287 	 * Route packet.
288 	 */
289 	if (ro == NULL) {
290 		memset(&iproute, 0, sizeof(iproute));
291 		ro = &iproute;
292 	}
293 	sockaddr_in_init(&udst.sin, &ip->ip_dst, 0);
294 	dst = satocsin(rtcache_getdst(ro));
295 
296 	/*
297 	 * If there is a cached route, check that it is to the same
298 	 * destination and is still up.  If not, free it and try again.
299 	 * The address family should also be checked in case of sharing
300 	 * the cache with IPv6.
301 	 */
302 	if (dst && (dst->sin_family != AF_INET ||
303 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
304 		rtcache_free(ro);
305 
306 	/* XXX must be before rtcache operations */
307 	bound = curlwp_bind();
308 	bind_need_restore = true;
309 
310 	if ((rt = rtcache_validate(ro)) == NULL &&
311 	    (rt = rtcache_update(ro, 1)) == NULL) {
312 		dst = &udst.sin;
313 		error = rtcache_setdst(ro, &udst.sa);
314 		if (error != 0)
315 			goto bad;
316 	}
317 
318 	/*
319 	 * If routing to interface only, short circuit routing lookup.
320 	 */
321 	if (flags & IP_ROUTETOIF) {
322 		ifa = ifa_ifwithladdr_psref(sintocsa(dst), &psref_ia);
323 		if (ifa == NULL) {
324 			IP_STATINC(IP_STAT_NOROUTE);
325 			error = ENETUNREACH;
326 			goto bad;
327 		}
328 		/* ia is already referenced by psref_ia */
329 		ia = ifatoia(ifa);
330 
331 		ifp = ia->ia_ifp;
332 		mtu = ifp->if_mtu;
333 		ip->ip_ttl = 1;
334 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
335 	} else if (((IN_MULTICAST(ip->ip_dst.s_addr) ||
336 	    ip->ip_dst.s_addr == INADDR_BROADCAST) ||
337 	    (flags & IP_ROUTETOIFINDEX)) &&
338 	    imo != NULL && imo->imo_multicast_if_index != 0) {
339 		ifp = mifp = if_get_byindex(imo->imo_multicast_if_index, &psref);
340 		if (ifp == NULL) {
341 			IP_STATINC(IP_STAT_NOROUTE);
342 			error = ENETUNREACH;
343 			goto bad;
344 		}
345 		mtu = ifp->if_mtu;
346 		ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
347 		if (ia == NULL) {
348 			error = EADDRNOTAVAIL;
349 			goto bad;
350 		}
351 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
352 		    ip->ip_dst.s_addr == INADDR_BROADCAST) {
353 			isbroadcast = 0;
354 		} else {
355 			/* IP_ROUTETOIFINDEX */
356 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
357 			if ((isbroadcast == 0) && ((ifp->if_flags &
358 			    (IFF_LOOPBACK | IFF_POINTOPOINT)) == 0) &&
359 			    (in_direct(dst->sin_addr, ifp) == 0)) {
360 				/* gateway address required */
361 				if (rt == NULL)
362 					rt = rtcache_init(ro);
363 				if (rt == NULL || rt->rt_ifp != ifp) {
364 					IP_STATINC(IP_STAT_NOROUTE);
365 					error = EHOSTUNREACH;
366 					goto bad;
367 				}
368 				rt->rt_use++;
369 				if (rt->rt_flags & RTF_GATEWAY)
370 					dst = satosin(rt->rt_gateway);
371 				if (rt->rt_flags & RTF_HOST)
372 					isbroadcast =
373 					    rt->rt_flags & RTF_BROADCAST;
374 			}
375 		}
376 	} else {
377 		if (rt == NULL)
378 			rt = rtcache_init(ro);
379 		if (rt == NULL) {
380 			IP_STATINC(IP_STAT_NOROUTE);
381 			error = EHOSTUNREACH;
382 			goto bad;
383 		}
384 		if (ifa_is_destroying(rt->rt_ifa)) {
385 			rtcache_unref(rt, ro);
386 			rt = NULL;
387 			IP_STATINC(IP_STAT_NOROUTE);
388 			error = EHOSTUNREACH;
389 			goto bad;
390 		}
391 		ifa_acquire(rt->rt_ifa, &psref_ia);
392 		ia = ifatoia(rt->rt_ifa);
393 		ifp = rt->rt_ifp;
394 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
395 			mtu = ifp->if_mtu;
396 		rt->rt_use++;
397 		if (rt->rt_flags & RTF_GATEWAY)
398 			dst = satosin(rt->rt_gateway);
399 		if (rt->rt_flags & RTF_HOST)
400 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
401 		else
402 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
403 	}
404 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
405 
406 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
407 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
408 		bool inmgroup;
409 
410 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
411 		    M_BCAST : M_MCAST;
412 		/*
413 		 * See if the caller provided any multicast options
414 		 */
415 		if (imo != NULL)
416 			ip->ip_ttl = imo->imo_multicast_ttl;
417 		else
418 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
419 
420 		/*
421 		 * if we don't know the outgoing ifp yet, we can't generate
422 		 * output
423 		 */
424 		if (!ifp) {
425 			IP_STATINC(IP_STAT_NOROUTE);
426 			error = ENETUNREACH;
427 			goto bad;
428 		}
429 
430 		/*
431 		 * If the packet is multicast or broadcast, confirm that
432 		 * the outgoing interface can transmit it.
433 		 */
434 		if (((m->m_flags & M_MCAST) &&
435 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
436 		    ((m->m_flags & M_BCAST) &&
437 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
438 			IP_STATINC(IP_STAT_NOROUTE);
439 			error = ENETUNREACH;
440 			goto bad;
441 		}
442 		/*
443 		 * If source address not specified yet, use an address
444 		 * of outgoing interface.
445 		 */
446 		if (in_nullhost(ip->ip_src)) {
447 			struct in_ifaddr *xia;
448 			struct ifaddr *xifa;
449 			struct psref _psref;
450 
451 			xia = in_get_ia_from_ifp_psref(ifp, &_psref);
452 			if (!xia) {
453 				error = EADDRNOTAVAIL;
454 				goto bad;
455 			}
456 			xifa = &xia->ia_ifa;
457 			if (xifa->ifa_getifa != NULL) {
458 				ia4_release(xia, &_psref);
459 				/* FIXME ifa_getifa is NOMPSAFE */
460 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
461 				if (xia == NULL) {
462 					error = EADDRNOTAVAIL;
463 					goto bad;
464 				}
465 				ia4_acquire(xia, &_psref);
466 			}
467 			ip->ip_src = xia->ia_addr.sin_addr;
468 			ia4_release(xia, &_psref);
469 		}
470 
471 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
472 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
473 			/*
474 			 * If we belong to the destination multicast group
475 			 * on the outgoing interface, and the caller did not
476 			 * forbid loopback, loop back a copy.
477 			 */
478 			ip_mloopback(ifp, m, &udst.sin);
479 		}
480 #ifdef MROUTING
481 		else {
482 			/*
483 			 * If we are acting as a multicast router, perform
484 			 * multicast forwarding as if the packet had just
485 			 * arrived on the interface to which we are about
486 			 * to send.  The multicast forwarding function
487 			 * recursively calls this function, using the
488 			 * IP_FORWARDING flag to prevent infinite recursion.
489 			 *
490 			 * Multicasts that are looped back by ip_mloopback(),
491 			 * above, will be forwarded by the ip_input() routine,
492 			 * if necessary.
493 			 */
494 			extern struct socket *ip_mrouter;
495 
496 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
497 				if (ip_mforward(m, ifp) != 0) {
498 					m_freem(m);
499 					goto done;
500 				}
501 			}
502 		}
503 #endif
504 		/*
505 		 * Multicasts with a time-to-live of zero may be looped-
506 		 * back, above, but must not be transmitted on a network.
507 		 * Also, multicasts addressed to the loopback interface
508 		 * are not sent -- the above call to ip_mloopback() will
509 		 * loop back a copy if this host actually belongs to the
510 		 * destination group on the loopback interface.
511 		 */
512 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
513 			m_freem(m);
514 			goto done;
515 		}
516 		goto sendit;
517 	}
518 
519 	/*
520 	 * If source address not specified yet, use address
521 	 * of outgoing interface.
522 	 */
523 	if (in_nullhost(ip->ip_src)) {
524 		struct ifaddr *xifa;
525 
526 		xifa = &ia->ia_ifa;
527 		if (xifa->ifa_getifa != NULL) {
528 			ia4_release(ia, &psref_ia);
529 			/* FIXME ifa_getifa is NOMPSAFE */
530 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
531 			if (ia == NULL) {
532 				error = EADDRNOTAVAIL;
533 				goto bad;
534 			}
535 			ia4_acquire(ia, &psref_ia);
536 		}
537 		ip->ip_src = ia->ia_addr.sin_addr;
538 	}
539 
540 	/*
541 	 * packets with Class-D address as source are not valid per
542 	 * RFC 1112
543 	 */
544 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
545 		IP_STATINC(IP_STAT_ODROPPED);
546 		error = EADDRNOTAVAIL;
547 		goto bad;
548 	}
549 
550 	/*
551 	 * Look for broadcast address and and verify user is allowed to
552 	 * send such a packet.
553 	 */
554 	if (isbroadcast) {
555 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
556 			error = EADDRNOTAVAIL;
557 			goto bad;
558 		}
559 		if ((flags & IP_ALLOWBROADCAST) == 0) {
560 			error = EACCES;
561 			goto bad;
562 		}
563 		/* don't allow broadcast messages to be fragmented */
564 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
565 			error = EMSGSIZE;
566 			goto bad;
567 		}
568 		m->m_flags |= M_BCAST;
569 	} else
570 		m->m_flags &= ~M_BCAST;
571 
572 sendit:
573 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
574 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
575 			ip->ip_id = 0;
576 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
577 			ip->ip_id = ip_newid(ia);
578 		} else {
579 
580 			/*
581 			 * TSO capable interfaces (typically?) increment
582 			 * ip_id for each segment.
583 			 * "allocate" enough ids here to increase the chance
584 			 * for them to be unique.
585 			 *
586 			 * note that the following calculation is not
587 			 * needed to be precise.  wasting some ip_id is fine.
588 			 */
589 
590 			unsigned int segsz = m->m_pkthdr.segsz;
591 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
592 			unsigned int num = howmany(datasz, segsz);
593 
594 			ip->ip_id = ip_newid_range(ia, num);
595 		}
596 	}
597 	if (ia != NULL) {
598 		ia4_release(ia, &psref_ia);
599 		ia = NULL;
600 	}
601 
602 	/*
603 	 * If we're doing Path MTU Discovery, we need to set DF unless
604 	 * the route's MTU is locked.
605 	 */
606 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
607 		ip->ip_off |= htons(IP_DF);
608 	}
609 
610 #ifdef IPSEC
611 	if (ipsec_used) {
612 		bool ipsec_done = false;
613 
614 		/* Perform IPsec processing, if any. */
615 		error = ipsec4_output(m, inp, flags, &mtu, &natt_frag,
616 		    &ipsec_done);
617 		if (error || ipsec_done)
618 			goto done;
619 	}
620 #endif
621 
622 	/*
623 	 * Run through list of hooks for output packets.
624 	 */
625 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
626 	if (error)
627 		goto done;
628 	if (m == NULL)
629 		goto done;
630 
631 	ip = mtod(m, struct ip *);
632 	hlen = ip->ip_hl << 2;
633 
634 	m->m_pkthdr.csum_data |= hlen << 16;
635 
636 	/*
637 	 * search for the source address structure to
638 	 * maintain output statistics, and verify address
639 	 * validity
640 	 */
641 	KASSERT(ia == NULL);
642 	sockaddr_in_init(&usrc.sin, &ip->ip_src, 0);
643 	ifa = ifaof_ifpforaddr_psref(&usrc.sa, ifp, &psref_ia);
644 	if (ifa != NULL)
645 		ia = ifatoia(ifa);
646 
647 	/*
648 	 * Ensure we only send from a valid address.
649 	 * A NULL address is valid because the packet could be
650 	 * generated from a packet filter.
651 	 */
652 	if (ia != NULL && (flags & IP_FORWARDING) == 0 &&
653 	    (error = ip_ifaddrvalid(ia)) != 0)
654 	{
655 		ARPLOG(LOG_ERR,
656 		    "refusing to send from invalid address %s (pid %d)\n",
657 		    ARPLOGADDR(&ip->ip_src), curproc->p_pid);
658 		IP_STATINC(IP_STAT_ODROPPED);
659 		if (error == 1)
660 			/*
661 			 * Address exists, but is tentative or detached.
662 			 * We can't send from it because it's invalid,
663 			 * so we drop the packet.
664 			 */
665 			error = 0;
666 		else
667 			error = EADDRNOTAVAIL;
668 		goto bad;
669 	}
670 
671 	/* Maybe skip checksums on loopback interfaces. */
672 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
673 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
674 	}
675 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
676 	/*
677 	 * If small enough for mtu of path, or if using TCP segmentation
678 	 * offload, can just send directly.
679 	 */
680 	if (ntohs(ip->ip_len) <= mtu ||
681 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
682 		const struct sockaddr *sa;
683 
684 #if IFA_STATS
685 		if (ia)
686 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
687 #endif
688 		/*
689 		 * Always initialize the sum to 0!  Some HW assisted
690 		 * checksumming requires this.
691 		 */
692 		ip->ip_sum = 0;
693 
694 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
695 			/*
696 			 * Perform any checksums that the hardware can't do
697 			 * for us.
698 			 *
699 			 * XXX Does any hardware require the {th,uh}_sum
700 			 * XXX fields to be 0?
701 			 */
702 			if (sw_csum & M_CSUM_IPv4) {
703 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
704 				ip->ip_sum = in_cksum(m, hlen);
705 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
706 			}
707 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
708 				if (IN_NEED_CHECKSUM(ifp,
709 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
710 					in_delayed_cksum(m);
711 				}
712 				m->m_pkthdr.csum_flags &=
713 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
714 			}
715 		}
716 
717 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
718 		if (__predict_true(
719 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
720 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
721 			error = ip_if_output(ifp, m, sa, rt);
722 		} else {
723 			error = ip_tso_output(ifp, m, sa, rt);
724 		}
725 		goto done;
726 	}
727 
728 	/*
729 	 * We can't use HW checksumming if we're about to
730 	 * fragment the packet.
731 	 *
732 	 * XXX Some hardware can do this.
733 	 */
734 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
735 		if (IN_NEED_CHECKSUM(ifp,
736 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
737 			in_delayed_cksum(m);
738 		}
739 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
740 	}
741 
742 	/*
743 	 * Too large for interface; fragment if possible.
744 	 * Must be able to put at least 8 bytes per fragment.
745 	 */
746 	if (ntohs(ip->ip_off) & IP_DF) {
747 		if (flags & IP_RETURNMTU) {
748 			KASSERT(inp != NULL);
749 			inp->inp_errormtu = mtu;
750 		}
751 		error = EMSGSIZE;
752 		IP_STATINC(IP_STAT_CANTFRAG);
753 		goto bad;
754 	}
755 
756 	error = ip_fragment(m, ifp, mtu);
757 	if (error) {
758 		m = NULL;
759 		goto bad;
760 	}
761 
762 	for (; m; m = m0) {
763 		m0 = m->m_nextpkt;
764 		m->m_nextpkt = 0;
765 		if (error) {
766 			m_freem(m);
767 			continue;
768 		}
769 #if IFA_STATS
770 		if (ia)
771 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
772 #endif
773 		/*
774 		 * If we get there, the packet has not been handled by
775 		 * IPsec whereas it should have. Now that it has been
776 		 * fragmented, re-inject it in ip_output so that IPsec
777 		 * processing can occur.
778 		 */
779 		if (natt_frag) {
780 			error = ip_output(m, opt, ro,
781 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
782 			    imo, inp);
783 		} else {
784 			KASSERT((m->m_pkthdr.csum_flags &
785 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
786 			error = ip_if_output(ifp, m,
787 			    (m->m_flags & M_MCAST) ?
788 			    sintocsa(rdst) : sintocsa(dst), rt);
789 		}
790 	}
791 	if (error == 0) {
792 		IP_STATINC(IP_STAT_FRAGMENTED);
793 	}
794 done:
795 	ia4_release(ia, &psref_ia);
796 	rtcache_unref(rt, ro);
797 	if (ro == &iproute) {
798 		rtcache_free(&iproute);
799 	}
800 	if (mifp != NULL) {
801 		if_put(mifp, &psref);
802 	}
803 	if (bind_need_restore)
804 		curlwp_bindx(bound);
805 	return error;
806 bad:
807 	m_freem(m);
808 	goto done;
809 }
810 
811 int
812 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
813 {
814 	struct ip *ip, *mhip;
815 	struct mbuf *m0;
816 	int len, hlen, off;
817 	int mhlen, firstlen;
818 	struct mbuf **mnext;
819 	int sw_csum = m->m_pkthdr.csum_flags;
820 	int fragments = 0;
821 	int error = 0;
822 
823 	ip = mtod(m, struct ip *);
824 	hlen = ip->ip_hl << 2;
825 	if (ifp != NULL)
826 		sw_csum &= ~ifp->if_csum_flags_tx;
827 
828 	len = (mtu - hlen) &~ 7;
829 	if (len < 8) {
830 		m_freem(m);
831 		return (EMSGSIZE);
832 	}
833 
834 	firstlen = len;
835 	mnext = &m->m_nextpkt;
836 
837 	/*
838 	 * Loop through length of segment after first fragment,
839 	 * make new header and copy data of each part and link onto chain.
840 	 */
841 	m0 = m;
842 	mhlen = sizeof (struct ip);
843 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
844 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
845 		if (m == 0) {
846 			error = ENOBUFS;
847 			IP_STATINC(IP_STAT_ODROPPED);
848 			goto sendorfree;
849 		}
850 		MCLAIM(m, m0->m_owner);
851 		*mnext = m;
852 		mnext = &m->m_nextpkt;
853 		m->m_data += max_linkhdr;
854 		mhip = mtod(m, struct ip *);
855 		*mhip = *ip;
856 		/* we must inherit MCAST and BCAST flags */
857 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
858 		if (hlen > sizeof (struct ip)) {
859 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
860 			mhip->ip_hl = mhlen >> 2;
861 		}
862 		m->m_len = mhlen;
863 		mhip->ip_off = ((off - hlen) >> 3) +
864 		    (ntohs(ip->ip_off) & ~IP_MF);
865 		if (ip->ip_off & htons(IP_MF))
866 			mhip->ip_off |= IP_MF;
867 		if (off + len >= ntohs(ip->ip_len))
868 			len = ntohs(ip->ip_len) - off;
869 		else
870 			mhip->ip_off |= IP_MF;
871 		HTONS(mhip->ip_off);
872 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
873 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
874 		if (m->m_next == 0) {
875 			error = ENOBUFS;	/* ??? */
876 			IP_STATINC(IP_STAT_ODROPPED);
877 			goto sendorfree;
878 		}
879 		m->m_pkthdr.len = mhlen + len;
880 		m_reset_rcvif(m);
881 		mhip->ip_sum = 0;
882 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
883 		if (sw_csum & M_CSUM_IPv4) {
884 			mhip->ip_sum = in_cksum(m, mhlen);
885 		} else {
886 			/*
887 			 * checksum is hw-offloaded or not necessary.
888 			 */
889 			m->m_pkthdr.csum_flags |=
890 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
891 			m->m_pkthdr.csum_data |= mhlen << 16;
892 			KASSERT(!(ifp != NULL &&
893 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
894 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
895 		}
896 		IP_STATINC(IP_STAT_OFRAGMENTS);
897 		fragments++;
898 	}
899 	/*
900 	 * Update first fragment by trimming what's been copied out
901 	 * and updating header, then send each fragment (in order).
902 	 */
903 	m = m0;
904 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
905 	m->m_pkthdr.len = hlen + firstlen;
906 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
907 	ip->ip_off |= htons(IP_MF);
908 	ip->ip_sum = 0;
909 	if (sw_csum & M_CSUM_IPv4) {
910 		ip->ip_sum = in_cksum(m, hlen);
911 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
912 	} else {
913 		/*
914 		 * checksum is hw-offloaded or not necessary.
915 		 */
916 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
917 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
918 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
919 		    sizeof(struct ip));
920 	}
921 sendorfree:
922 	/*
923 	 * If there is no room for all the fragments, don't queue
924 	 * any of them.
925 	 */
926 	if (ifp != NULL) {
927 		IFQ_LOCK(&ifp->if_snd);
928 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
929 		    error == 0) {
930 			error = ENOBUFS;
931 			IP_STATINC(IP_STAT_ODROPPED);
932 			IFQ_INC_DROPS(&ifp->if_snd);
933 		}
934 		IFQ_UNLOCK(&ifp->if_snd);
935 	}
936 	if (error) {
937 		for (m = m0; m; m = m0) {
938 			m0 = m->m_nextpkt;
939 			m->m_nextpkt = NULL;
940 			m_freem(m);
941 		}
942 	}
943 	return (error);
944 }
945 
946 /*
947  * Process a delayed payload checksum calculation.
948  */
949 void
950 in_delayed_cksum(struct mbuf *m)
951 {
952 	struct ip *ip;
953 	u_int16_t csum, offset;
954 
955 	ip = mtod(m, struct ip *);
956 	offset = ip->ip_hl << 2;
957 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
958 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
959 		csum = 0xffff;
960 
961 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
962 
963 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
964 		/* This happen when ip options were inserted
965 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
966 		    m->m_len, offset, ip->ip_p);
967 		 */
968 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
969 	} else
970 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
971 }
972 
973 /*
974  * Determine the maximum length of the options to be inserted;
975  * we would far rather allocate too much space rather than too little.
976  */
977 
978 u_int
979 ip_optlen(struct inpcb *inp)
980 {
981 	struct mbuf *m = inp->inp_options;
982 
983 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
984 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
985 	}
986 	return 0;
987 }
988 
989 /*
990  * Insert IP options into preformed packet.
991  * Adjust IP destination as required for IP source routing,
992  * as indicated by a non-zero in_addr at the start of the options.
993  */
994 static struct mbuf *
995 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
996 {
997 	struct ipoption *p = mtod(opt, struct ipoption *);
998 	struct mbuf *n;
999 	struct ip *ip = mtod(m, struct ip *);
1000 	unsigned optlen;
1001 
1002 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1003 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
1004 		return (m);		/* XXX should fail */
1005 	if (!in_nullhost(p->ipopt_dst))
1006 		ip->ip_dst = p->ipopt_dst;
1007 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
1008 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1009 		if (n == 0)
1010 			return (m);
1011 		MCLAIM(n, m->m_owner);
1012 		M_MOVE_PKTHDR(n, m);
1013 		m->m_len -= sizeof(struct ip);
1014 		m->m_data += sizeof(struct ip);
1015 		n->m_next = m;
1016 		m = n;
1017 		m->m_len = optlen + sizeof(struct ip);
1018 		m->m_data += max_linkhdr;
1019 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
1020 	} else {
1021 		m->m_data -= optlen;
1022 		m->m_len += optlen;
1023 		memmove(mtod(m, void *), ip, sizeof(struct ip));
1024 	}
1025 	m->m_pkthdr.len += optlen;
1026 	ip = mtod(m, struct ip *);
1027 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
1028 	*phlen = sizeof(struct ip) + optlen;
1029 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1030 	return (m);
1031 }
1032 
1033 /*
1034  * Copy options from ip to jp,
1035  * omitting those not copied during fragmentation.
1036  */
1037 int
1038 ip_optcopy(struct ip *ip, struct ip *jp)
1039 {
1040 	u_char *cp, *dp;
1041 	int opt, optlen, cnt;
1042 
1043 	cp = (u_char *)(ip + 1);
1044 	dp = (u_char *)(jp + 1);
1045 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1046 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1047 		opt = cp[0];
1048 		if (opt == IPOPT_EOL)
1049 			break;
1050 		if (opt == IPOPT_NOP) {
1051 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1052 			*dp++ = IPOPT_NOP;
1053 			optlen = 1;
1054 			continue;
1055 		}
1056 
1057 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
1058 		optlen = cp[IPOPT_OLEN];
1059 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
1060 
1061 		/* Invalid lengths should have been caught by ip_dooptions. */
1062 		if (optlen > cnt)
1063 			optlen = cnt;
1064 		if (IPOPT_COPIED(opt)) {
1065 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1066 			dp += optlen;
1067 		}
1068 	}
1069 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1070 		*dp++ = IPOPT_EOL;
1071 	return (optlen);
1072 }
1073 
1074 /*
1075  * IP socket option processing.
1076  */
1077 int
1078 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1079 {
1080 	struct inpcb *inp = sotoinpcb(so);
1081 	struct ip *ip = &inp->inp_ip;
1082 	int inpflags = inp->inp_flags;
1083 	int optval = 0, error = 0;
1084 	struct in_pktinfo pktinfo;
1085 
1086 	KASSERT(solocked(so));
1087 
1088 	if (sopt->sopt_level != IPPROTO_IP) {
1089 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1090 			return 0;
1091 		return ENOPROTOOPT;
1092 	}
1093 
1094 	switch (op) {
1095 	case PRCO_SETOPT:
1096 		switch (sopt->sopt_name) {
1097 		case IP_OPTIONS:
1098 #ifdef notyet
1099 		case IP_RETOPTS:
1100 #endif
1101 			error = ip_pcbopts(inp, sopt);
1102 			break;
1103 
1104 		case IP_TOS:
1105 		case IP_TTL:
1106 		case IP_MINTTL:
1107 		case IP_RECVOPTS:
1108 		case IP_RECVRETOPTS:
1109 		case IP_RECVDSTADDR:
1110 		case IP_RECVIF:
1111 		case IP_RECVPKTINFO:
1112 		case IP_RECVTTL:
1113 			error = sockopt_getint(sopt, &optval);
1114 			if (error)
1115 				break;
1116 
1117 			switch (sopt->sopt_name) {
1118 			case IP_TOS:
1119 				ip->ip_tos = optval;
1120 				break;
1121 
1122 			case IP_TTL:
1123 				ip->ip_ttl = optval;
1124 				break;
1125 
1126 			case IP_MINTTL:
1127 				if (optval > 0 && optval <= MAXTTL)
1128 					inp->inp_ip_minttl = optval;
1129 				else
1130 					error = EINVAL;
1131 				break;
1132 #define	OPTSET(bit) \
1133 	if (optval) \
1134 		inpflags |= bit; \
1135 	else \
1136 		inpflags &= ~bit;
1137 
1138 			case IP_RECVOPTS:
1139 				OPTSET(INP_RECVOPTS);
1140 				break;
1141 
1142 			case IP_RECVPKTINFO:
1143 				OPTSET(INP_RECVPKTINFO);
1144 				break;
1145 
1146 			case IP_RECVRETOPTS:
1147 				OPTSET(INP_RECVRETOPTS);
1148 				break;
1149 
1150 			case IP_RECVDSTADDR:
1151 				OPTSET(INP_RECVDSTADDR);
1152 				break;
1153 
1154 			case IP_RECVIF:
1155 				OPTSET(INP_RECVIF);
1156 				break;
1157 
1158 			case IP_RECVTTL:
1159 				OPTSET(INP_RECVTTL);
1160 				break;
1161 			}
1162 			break;
1163 		case IP_PKTINFO:
1164 			error = sockopt_getint(sopt, &optval);
1165 			if (!error) {
1166 				/* Linux compatibility */
1167 				OPTSET(INP_RECVPKTINFO);
1168 				break;
1169 			}
1170 			error = sockopt_get(sopt, &pktinfo, sizeof(pktinfo));
1171 			if (error)
1172 				break;
1173 
1174 			if (pktinfo.ipi_ifindex == 0) {
1175 				inp->inp_prefsrcip = pktinfo.ipi_addr;
1176 				break;
1177 			}
1178 
1179 			/* Solaris compatibility */
1180 			struct ifnet *ifp;
1181 			struct in_ifaddr *ia;
1182 			int s;
1183 
1184 			/* pick up primary address */
1185 			s = pserialize_read_enter();
1186 			ifp = if_byindex(pktinfo.ipi_ifindex);
1187 			if (ifp == NULL) {
1188 				pserialize_read_exit(s);
1189 				error = EADDRNOTAVAIL;
1190 				break;
1191 			}
1192 			ia = in_get_ia_from_ifp(ifp);
1193 			if (ia == NULL) {
1194 				pserialize_read_exit(s);
1195 				error = EADDRNOTAVAIL;
1196 				break;
1197 			}
1198 			inp->inp_prefsrcip = IA_SIN(ia)->sin_addr;
1199 			pserialize_read_exit(s);
1200 			break;
1201 		break;
1202 #undef OPTSET
1203 
1204 		case IP_MULTICAST_IF:
1205 		case IP_MULTICAST_TTL:
1206 		case IP_MULTICAST_LOOP:
1207 		case IP_ADD_MEMBERSHIP:
1208 		case IP_DROP_MEMBERSHIP:
1209 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1210 			break;
1211 
1212 		case IP_PORTRANGE:
1213 			error = sockopt_getint(sopt, &optval);
1214 			if (error)
1215 				break;
1216 
1217 			switch (optval) {
1218 			case IP_PORTRANGE_DEFAULT:
1219 			case IP_PORTRANGE_HIGH:
1220 				inpflags &= ~(INP_LOWPORT);
1221 				break;
1222 
1223 			case IP_PORTRANGE_LOW:
1224 				inpflags |= INP_LOWPORT;
1225 				break;
1226 
1227 			default:
1228 				error = EINVAL;
1229 				break;
1230 			}
1231 			break;
1232 
1233 		case IP_PORTALGO:
1234 			error = sockopt_getint(sopt, &optval);
1235 			if (error)
1236 				break;
1237 
1238 			error = portalgo_algo_index_select(
1239 			    (struct inpcb_hdr *)inp, optval);
1240 			break;
1241 
1242 #if defined(IPSEC)
1243 		case IP_IPSEC_POLICY:
1244 			if (ipsec_enabled) {
1245 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1246 				    sopt->sopt_data, sopt->sopt_size,
1247 				    curlwp->l_cred);
1248 				break;
1249 			}
1250 			/*FALLTHROUGH*/
1251 #endif /* IPSEC */
1252 
1253 		default:
1254 			error = ENOPROTOOPT;
1255 			break;
1256 		}
1257 		break;
1258 
1259 	case PRCO_GETOPT:
1260 		switch (sopt->sopt_name) {
1261 		case IP_OPTIONS:
1262 		case IP_RETOPTS: {
1263 			struct mbuf *mopts = inp->inp_options;
1264 
1265 			if (mopts) {
1266 				struct mbuf *m;
1267 
1268 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1269 				if (m == NULL) {
1270 					error = ENOBUFS;
1271 					break;
1272 				}
1273 				error = sockopt_setmbuf(sopt, m);
1274 			}
1275 			break;
1276 		}
1277 		case IP_TOS:
1278 		case IP_TTL:
1279 		case IP_MINTTL:
1280 		case IP_RECVOPTS:
1281 		case IP_RECVRETOPTS:
1282 		case IP_RECVDSTADDR:
1283 		case IP_RECVIF:
1284 		case IP_RECVPKTINFO:
1285 		case IP_RECVTTL:
1286 		case IP_ERRORMTU:
1287 			switch (sopt->sopt_name) {
1288 			case IP_TOS:
1289 				optval = ip->ip_tos;
1290 				break;
1291 
1292 			case IP_TTL:
1293 				optval = ip->ip_ttl;
1294 				break;
1295 
1296 			case IP_MINTTL:
1297 				optval = inp->inp_ip_minttl;
1298 				break;
1299 
1300 			case IP_ERRORMTU:
1301 				optval = inp->inp_errormtu;
1302 				break;
1303 
1304 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1305 
1306 			case IP_RECVOPTS:
1307 				optval = OPTBIT(INP_RECVOPTS);
1308 				break;
1309 
1310 			case IP_RECVPKTINFO:
1311 				optval = OPTBIT(INP_RECVPKTINFO);
1312 				break;
1313 
1314 			case IP_RECVRETOPTS:
1315 				optval = OPTBIT(INP_RECVRETOPTS);
1316 				break;
1317 
1318 			case IP_RECVDSTADDR:
1319 				optval = OPTBIT(INP_RECVDSTADDR);
1320 				break;
1321 
1322 			case IP_RECVIF:
1323 				optval = OPTBIT(INP_RECVIF);
1324 				break;
1325 
1326 			case IP_RECVTTL:
1327 				optval = OPTBIT(INP_RECVTTL);
1328 				break;
1329 			}
1330 			error = sockopt_setint(sopt, optval);
1331 			break;
1332 
1333 		case IP_PKTINFO:
1334 			switch (sopt->sopt_size) {
1335 			case sizeof(int):
1336 				/* Linux compatibility */
1337 				optval = OPTBIT(INP_RECVPKTINFO);
1338 				error = sockopt_setint(sopt, optval);
1339 				break;
1340 			case sizeof(struct in_pktinfo):
1341 				/* Solaris compatibility */
1342 				pktinfo.ipi_ifindex = 0;
1343 				pktinfo.ipi_addr = inp->inp_prefsrcip;
1344 				error = sockopt_set(sopt, &pktinfo,
1345 				    sizeof(pktinfo));
1346 				break;
1347 			default:
1348 				/*
1349 				 * While size is stuck at 0, and, later, if
1350 				 * the caller doesn't use an exactly sized
1351 				 * recipient for the data, default to Linux
1352 				 * compatibility
1353 				 */
1354 				optval = OPTBIT(INP_RECVPKTINFO);
1355 				error = sockopt_setint(sopt, optval);
1356 				break;
1357 			}
1358 			break;
1359 
1360 #if 0	/* defined(IPSEC) */
1361 		case IP_IPSEC_POLICY:
1362 		{
1363 			struct mbuf *m = NULL;
1364 
1365 			/* XXX this will return EINVAL as sopt is empty */
1366 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1367 			    sopt->sopt_size, &m);
1368 			if (error == 0)
1369 				error = sockopt_setmbuf(sopt, m);
1370 			break;
1371 		}
1372 #endif /*IPSEC*/
1373 
1374 		case IP_MULTICAST_IF:
1375 		case IP_MULTICAST_TTL:
1376 		case IP_MULTICAST_LOOP:
1377 		case IP_ADD_MEMBERSHIP:
1378 		case IP_DROP_MEMBERSHIP:
1379 			error = ip_getmoptions(inp->inp_moptions, sopt);
1380 			break;
1381 
1382 		case IP_PORTRANGE:
1383 			if (inpflags & INP_LOWPORT)
1384 				optval = IP_PORTRANGE_LOW;
1385 			else
1386 				optval = IP_PORTRANGE_DEFAULT;
1387 			error = sockopt_setint(sopt, optval);
1388 			break;
1389 
1390 		case IP_PORTALGO:
1391 			optval = inp->inp_portalgo;
1392 			error = sockopt_setint(sopt, optval);
1393 			break;
1394 
1395 		default:
1396 			error = ENOPROTOOPT;
1397 			break;
1398 		}
1399 		break;
1400 	}
1401 
1402 	if (!error) {
1403 		inp->inp_flags = inpflags;
1404 	}
1405 	return error;
1406 }
1407 
1408 static int
1409 ip_pktinfo_prepare(const struct in_pktinfo *pktinfo, struct ip_pktopts *pktopts,
1410     int *flags, kauth_cred_t cred)
1411 {
1412 	struct ip_moptions *imo;
1413 	int error = 0;
1414 	bool addrset = false;
1415 
1416 	if (!in_nullhost(pktinfo->ipi_addr)) {
1417 		pktopts->ippo_laddr.sin_addr = pktinfo->ipi_addr;
1418 		/* EADDRNOTAVAIL? */
1419 		error = in_pcbbindableaddr(&pktopts->ippo_laddr, cred);
1420 		if (error != 0)
1421 			return error;
1422 		addrset = true;
1423 	}
1424 
1425 	if (pktinfo->ipi_ifindex != 0) {
1426 		if (!addrset) {
1427 			struct ifnet *ifp;
1428 			struct in_ifaddr *ia;
1429 			int s;
1430 
1431 			/* pick up primary address */
1432 			s = pserialize_read_enter();
1433 			ifp = if_byindex(pktinfo->ipi_ifindex);
1434 			if (ifp == NULL) {
1435 				pserialize_read_exit(s);
1436 				return EADDRNOTAVAIL;
1437 			}
1438 			ia = in_get_ia_from_ifp(ifp);
1439 			if (ia == NULL) {
1440 				pserialize_read_exit(s);
1441 				return EADDRNOTAVAIL;
1442 			}
1443 			pktopts->ippo_laddr.sin_addr = IA_SIN(ia)->sin_addr;
1444 			pserialize_read_exit(s);
1445 		}
1446 
1447 		/*
1448 		 * If specified ipi_ifindex,
1449 		 * use copied or locally initialized ip_moptions.
1450 		 * Original ip_moptions must not be modified.
1451 		 */
1452 		imo = &pktopts->ippo_imobuf;	/* local buf in pktopts */
1453 		if (pktopts->ippo_imo != NULL) {
1454 			memcpy(imo, pktopts->ippo_imo, sizeof(*imo));
1455 		} else {
1456 			memset(imo, 0, sizeof(*imo));
1457 			imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1458 			imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1459 		}
1460 		imo->imo_multicast_if_index = pktinfo->ipi_ifindex;
1461 		pktopts->ippo_imo = imo;
1462 		*flags |= IP_ROUTETOIFINDEX;
1463 	}
1464 	return error;
1465 }
1466 
1467 /*
1468  * Set up IP outgoing packet options. Even if control is NULL,
1469  * pktopts->ippo_laddr and pktopts->ippo_imo are set and used.
1470  */
1471 int
1472 ip_setpktopts(struct mbuf *control, struct ip_pktopts *pktopts, int *flags,
1473     struct inpcb *inp, kauth_cred_t cred)
1474 {
1475 	struct cmsghdr *cm;
1476 	struct in_pktinfo pktinfo;
1477 	int error;
1478 
1479 	pktopts->ippo_imo = inp->inp_moptions;
1480 
1481 	struct in_addr *ia = in_nullhost(inp->inp_prefsrcip) ? &inp->inp_laddr :
1482 	    &inp->inp_prefsrcip;
1483 	sockaddr_in_init(&pktopts->ippo_laddr, ia, 0);
1484 
1485 	if (control == NULL)
1486 		return 0;
1487 
1488 	/*
1489 	 * XXX: Currently, we assume all the optional information is
1490 	 * stored in a single mbuf.
1491 	 */
1492 	if (control->m_next)
1493 		return EINVAL;
1494 
1495 	for (; control->m_len > 0;
1496 	    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1497 	    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1498 		cm = mtod(control, struct cmsghdr *);
1499 		if ((control->m_len < sizeof(*cm)) ||
1500 		    (cm->cmsg_len == 0) ||
1501 		    (cm->cmsg_len > control->m_len)) {
1502 			return EINVAL;
1503 		}
1504 		if (cm->cmsg_level != IPPROTO_IP)
1505 			continue;
1506 
1507 		switch (cm->cmsg_type) {
1508 		case IP_PKTINFO:
1509 			if (cm->cmsg_len != CMSG_LEN(sizeof(pktinfo)))
1510 				return EINVAL;
1511 			memcpy(&pktinfo, CMSG_DATA(cm), sizeof(pktinfo));
1512 			error = ip_pktinfo_prepare(&pktinfo, pktopts, flags,
1513 			    cred);
1514 			if (error)
1515 				return error;
1516 			break;
1517 		case IP_SENDSRCADDR: /* FreeBSD compatibility */
1518 			if (cm->cmsg_len != CMSG_LEN(sizeof(struct in_addr)))
1519 				return EINVAL;
1520 			pktinfo.ipi_ifindex = 0;
1521 			pktinfo.ipi_addr =
1522 			    ((struct in_pktinfo *)CMSG_DATA(cm))->ipi_addr;
1523 			error = ip_pktinfo_prepare(&pktinfo, pktopts, flags,
1524 			    cred);
1525 			if (error)
1526 				return error;
1527 			break;
1528 		default:
1529 			return ENOPROTOOPT;
1530 		}
1531 	}
1532 	return 0;
1533 }
1534 
1535 /*
1536  * Set up IP options in pcb for insertion in output packets.
1537  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1538  * with destination address if source routed.
1539  */
1540 static int
1541 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1542 {
1543 	struct mbuf *m;
1544 	const u_char *cp;
1545 	u_char *dp;
1546 	int cnt;
1547 
1548 	KASSERT(inp_locked(inp));
1549 
1550 	/* Turn off any old options. */
1551 	if (inp->inp_options) {
1552 		m_free(inp->inp_options);
1553 	}
1554 	inp->inp_options = NULL;
1555 	if ((cnt = sopt->sopt_size) == 0) {
1556 		/* Only turning off any previous options. */
1557 		return 0;
1558 	}
1559 	cp = sopt->sopt_data;
1560 
1561 #ifndef	__vax__
1562 	if (cnt % sizeof(int32_t))
1563 		return (EINVAL);
1564 #endif
1565 
1566 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1567 	if (m == NULL)
1568 		return (ENOBUFS);
1569 
1570 	dp = mtod(m, u_char *);
1571 	memset(dp, 0, sizeof(struct in_addr));
1572 	dp += sizeof(struct in_addr);
1573 	m->m_len = sizeof(struct in_addr);
1574 
1575 	/*
1576 	 * IP option list according to RFC791. Each option is of the form
1577 	 *
1578 	 *	[optval] [olen] [(olen - 2) data bytes]
1579 	 *
1580 	 * We validate the list and copy options to an mbuf for prepending
1581 	 * to data packets. The IP first-hop destination address will be
1582 	 * stored before actual options and is zero if unset.
1583 	 */
1584 	while (cnt > 0) {
1585 		uint8_t optval, olen, offset;
1586 
1587 		optval = cp[IPOPT_OPTVAL];
1588 
1589 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1590 			olen = 1;
1591 		} else {
1592 			if (cnt < IPOPT_OLEN + 1)
1593 				goto bad;
1594 
1595 			olen = cp[IPOPT_OLEN];
1596 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1597 				goto bad;
1598 		}
1599 
1600 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1601 			/*
1602 			 * user process specifies route as:
1603 			 *	->A->B->C->D
1604 			 * D must be our final destination (but we can't
1605 			 * check that since we may not have connected yet).
1606 			 * A is first hop destination, which doesn't appear in
1607 			 * actual IP option, but is stored before the options.
1608 			 */
1609 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1610 				goto bad;
1611 
1612 			offset = cp[IPOPT_OFFSET];
1613 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1614 			    sizeof(struct in_addr));
1615 
1616 			cp += sizeof(struct in_addr);
1617 			cnt -= sizeof(struct in_addr);
1618 			olen -= sizeof(struct in_addr);
1619 
1620 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1621 				goto bad;
1622 
1623 			memcpy(dp, cp, olen);
1624 			dp[IPOPT_OPTVAL] = optval;
1625 			dp[IPOPT_OLEN] = olen;
1626 			dp[IPOPT_OFFSET] = offset;
1627 			break;
1628 		} else {
1629 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1630 				goto bad;
1631 
1632 			memcpy(dp, cp, olen);
1633 			break;
1634 		}
1635 
1636 		dp += olen;
1637 		m->m_len += olen;
1638 
1639 		if (optval == IPOPT_EOL)
1640 			break;
1641 
1642 		cp += olen;
1643 		cnt -= olen;
1644 	}
1645 
1646 	inp->inp_options = m;
1647 	return 0;
1648 bad:
1649 	(void)m_free(m);
1650 	return EINVAL;
1651 }
1652 
1653 /*
1654  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1655  * Must be called in a pserialize critical section.
1656  */
1657 static struct ifnet *
1658 ip_multicast_if(struct in_addr *a, int *ifindexp)
1659 {
1660 	int ifindex;
1661 	struct ifnet *ifp = NULL;
1662 	struct in_ifaddr *ia;
1663 
1664 	if (ifindexp)
1665 		*ifindexp = 0;
1666 	if (ntohl(a->s_addr) >> 24 == 0) {
1667 		ifindex = ntohl(a->s_addr) & 0xffffff;
1668 		ifp = if_byindex(ifindex);
1669 		if (!ifp)
1670 			return NULL;
1671 		if (ifindexp)
1672 			*ifindexp = ifindex;
1673 	} else {
1674 		IN_ADDRHASH_READER_FOREACH(ia, a->s_addr) {
1675 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1676 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1677 				ifp = ia->ia_ifp;
1678 				if (if_is_deactivated(ifp))
1679 					ifp = NULL;
1680 				break;
1681 			}
1682 		}
1683 	}
1684 	return ifp;
1685 }
1686 
1687 static int
1688 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1689 {
1690 	u_int tval;
1691 	u_char cval;
1692 	int error;
1693 
1694 	if (sopt == NULL)
1695 		return EINVAL;
1696 
1697 	switch (sopt->sopt_size) {
1698 	case sizeof(u_char):
1699 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1700 		tval = cval;
1701 		break;
1702 
1703 	case sizeof(u_int):
1704 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1705 		break;
1706 
1707 	default:
1708 		error = EINVAL;
1709 	}
1710 
1711 	if (error)
1712 		return error;
1713 
1714 	if (tval > maxval)
1715 		return EINVAL;
1716 
1717 	*val = tval;
1718 	return 0;
1719 }
1720 
1721 static int
1722 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1723     struct psref *psref, struct in_addr *ia, bool add)
1724 {
1725 	int error;
1726 	struct ip_mreq mreq;
1727 
1728 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1729 	if (error)
1730 		return error;
1731 
1732 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1733 		return EINVAL;
1734 
1735 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1736 
1737 	if (in_nullhost(mreq.imr_interface)) {
1738 		union {
1739 			struct sockaddr		dst;
1740 			struct sockaddr_in	dst4;
1741 		} u;
1742 		struct route ro;
1743 
1744 		if (!add) {
1745 			*ifp = NULL;
1746 			return 0;
1747 		}
1748 		/*
1749 		 * If no interface address was provided, use the interface of
1750 		 * the route to the given multicast address.
1751 		 */
1752 		struct rtentry *rt;
1753 		memset(&ro, 0, sizeof(ro));
1754 
1755 		sockaddr_in_init(&u.dst4, ia, 0);
1756 		error = rtcache_setdst(&ro, &u.dst);
1757 		if (error != 0)
1758 			return error;
1759 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1760 		if (*ifp != NULL) {
1761 			if (if_is_deactivated(*ifp))
1762 				*ifp = NULL;
1763 			else
1764 				if_acquire(*ifp, psref);
1765 		}
1766 		rtcache_unref(rt, &ro);
1767 		rtcache_free(&ro);
1768 	} else {
1769 		int s = pserialize_read_enter();
1770 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1771 		if (!add && *ifp == NULL) {
1772 			pserialize_read_exit(s);
1773 			return EADDRNOTAVAIL;
1774 		}
1775 		if (*ifp != NULL) {
1776 			if (if_is_deactivated(*ifp))
1777 				*ifp = NULL;
1778 			else
1779 				if_acquire(*ifp, psref);
1780 		}
1781 		pserialize_read_exit(s);
1782 	}
1783 	return 0;
1784 }
1785 
1786 /*
1787  * Add a multicast group membership.
1788  * Group must be a valid IP multicast address.
1789  */
1790 static int
1791 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1792 {
1793 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1794 	struct in_addr ia;
1795 	int i, error, bound;
1796 	struct psref psref;
1797 
1798 	/* imo is protected by solock or referenced only by the caller */
1799 
1800 	bound = curlwp_bind();
1801 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1802 		error = ip_get_membership(sopt, &ifp, &psref, &ia, true);
1803 	else
1804 #ifdef INET6
1805 		error = ip6_get_membership(sopt, &ifp, &psref, &ia, sizeof(ia));
1806 #else
1807 		error = EINVAL;
1808 		goto out;
1809 #endif
1810 
1811 	if (error)
1812 		goto out;
1813 
1814 	/*
1815 	 * See if we found an interface, and confirm that it
1816 	 * supports multicast.
1817 	 */
1818 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1819 		error = EADDRNOTAVAIL;
1820 		goto out;
1821 	}
1822 
1823 	/*
1824 	 * See if the membership already exists or if all the
1825 	 * membership slots are full.
1826 	 */
1827 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1828 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1829 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1830 			break;
1831 	}
1832 	if (i < imo->imo_num_memberships) {
1833 		error = EADDRINUSE;
1834 		goto out;
1835 	}
1836 
1837 	if (i == IP_MAX_MEMBERSHIPS) {
1838 		error = ETOOMANYREFS;
1839 		goto out;
1840 	}
1841 
1842 	/*
1843 	 * Everything looks good; add a new record to the multicast
1844 	 * address list for the given interface.
1845 	 */
1846 	IFNET_LOCK(ifp);
1847 	imo->imo_membership[i] = in_addmulti(&ia, ifp);
1848 	IFNET_UNLOCK(ifp);
1849 	if (imo->imo_membership[i] == NULL) {
1850 		error = ENOBUFS;
1851 		goto out;
1852 	}
1853 
1854 	++imo->imo_num_memberships;
1855 	error = 0;
1856 out:
1857 	if_put(ifp, &psref);
1858 	curlwp_bindx(bound);
1859 	return error;
1860 }
1861 
1862 /*
1863  * Drop a multicast group membership.
1864  * Group must be a valid IP multicast address.
1865  */
1866 static int
1867 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1868 {
1869 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1870 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1871 	int i, error, bound;
1872 	struct psref psref;
1873 
1874 	/* imo is protected by solock or referenced only by the caller */
1875 
1876 	bound = curlwp_bind();
1877 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1878 		error = ip_get_membership(sopt, &ifp, &psref, &ia, false);
1879 	else {
1880 #ifdef INET6
1881 		error = ip6_get_membership(sopt, &ifp, &psref, &ia, sizeof(ia));
1882 #else
1883 		error = EINVAL;
1884 		goto out;
1885 #endif
1886 	}
1887 
1888 	if (error)
1889 		goto out;
1890 
1891 	/*
1892 	 * Find the membership in the membership array.
1893 	 */
1894 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1895 		if ((ifp == NULL ||
1896 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1897 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1898 			break;
1899 	}
1900 	if (i == imo->imo_num_memberships) {
1901 		error = EADDRNOTAVAIL;
1902 		goto out;
1903 	}
1904 
1905 	/*
1906 	 * Give up the multicast address record to which the
1907 	 * membership points.
1908 	 */
1909 	IFNET_LOCK(imo->imo_membership[i]->inm_ifp);
1910 	in_delmulti(imo->imo_membership[i]);
1911 	IFNET_UNLOCK(imo->imo_membership[i]->inm_ifp);
1912 
1913 	/*
1914 	 * Remove the gap in the membership array.
1915 	 */
1916 	for (++i; i < imo->imo_num_memberships; ++i)
1917 		imo->imo_membership[i-1] = imo->imo_membership[i];
1918 	--imo->imo_num_memberships;
1919 	error = 0;
1920 out:
1921 	if_put(ifp, &psref);
1922 	curlwp_bindx(bound);
1923 	return error;
1924 }
1925 
1926 /*
1927  * Set the IP multicast options in response to user setsockopt().
1928  */
1929 int
1930 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1931 {
1932 	struct ip_moptions *imo = *pimo;
1933 	struct in_addr addr;
1934 	struct ifnet *ifp;
1935 	int ifindex, error = 0;
1936 
1937 	/* The passed imo isn't NULL, it should be protected by solock */
1938 
1939 	if (!imo) {
1940 		/*
1941 		 * No multicast option buffer attached to the pcb;
1942 		 * allocate one and initialize to default values.
1943 		 */
1944 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1945 		if (imo == NULL)
1946 			return ENOBUFS;
1947 
1948 		imo->imo_multicast_if_index = 0;
1949 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1950 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1951 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1952 		imo->imo_num_memberships = 0;
1953 		*pimo = imo;
1954 	}
1955 
1956 	switch (sopt->sopt_name) {
1957 	case IP_MULTICAST_IF: {
1958 		int s;
1959 		/*
1960 		 * Select the interface for outgoing multicast packets.
1961 		 */
1962 		error = sockopt_get(sopt, &addr, sizeof(addr));
1963 		if (error)
1964 			break;
1965 
1966 		/*
1967 		 * INADDR_ANY is used to remove a previous selection.
1968 		 * When no interface is selected, a default one is
1969 		 * chosen every time a multicast packet is sent.
1970 		 */
1971 		if (in_nullhost(addr)) {
1972 			imo->imo_multicast_if_index = 0;
1973 			break;
1974 		}
1975 		/*
1976 		 * The selected interface is identified by its local
1977 		 * IP address.  Find the interface and confirm that
1978 		 * it supports multicasting.
1979 		 */
1980 		s = pserialize_read_enter();
1981 		ifp = ip_multicast_if(&addr, &ifindex);
1982 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1983 			pserialize_read_exit(s);
1984 			error = EADDRNOTAVAIL;
1985 			break;
1986 		}
1987 		imo->imo_multicast_if_index = ifp->if_index;
1988 		pserialize_read_exit(s);
1989 		if (ifindex)
1990 			imo->imo_multicast_addr = addr;
1991 		else
1992 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1993 		break;
1994 	    }
1995 
1996 	case IP_MULTICAST_TTL:
1997 		/*
1998 		 * Set the IP time-to-live for outgoing multicast packets.
1999 		 */
2000 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
2001 		break;
2002 
2003 	case IP_MULTICAST_LOOP:
2004 		/*
2005 		 * Set the loopback flag for outgoing multicast packets.
2006 		 * Must be zero or one.
2007 		 */
2008 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
2009 		break;
2010 
2011 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
2012 		error = ip_add_membership(imo, sopt);
2013 		break;
2014 
2015 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
2016 		error = ip_drop_membership(imo, sopt);
2017 		break;
2018 
2019 	default:
2020 		error = EOPNOTSUPP;
2021 		break;
2022 	}
2023 
2024 	/*
2025 	 * If all options have default values, no need to keep the mbuf.
2026 	 */
2027 	if (imo->imo_multicast_if_index == 0 &&
2028 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2029 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2030 	    imo->imo_num_memberships == 0) {
2031 		kmem_intr_free(imo, sizeof(*imo));
2032 		*pimo = NULL;
2033 	}
2034 
2035 	return error;
2036 }
2037 
2038 /*
2039  * Return the IP multicast options in response to user getsockopt().
2040  */
2041 int
2042 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
2043 {
2044 	struct in_addr addr;
2045 	uint8_t optval;
2046 	int error = 0;
2047 
2048 	/* imo is protected by solock or refereced only by the caller */
2049 
2050 	switch (sopt->sopt_name) {
2051 	case IP_MULTICAST_IF:
2052 		if (imo == NULL || imo->imo_multicast_if_index == 0)
2053 			addr = zeroin_addr;
2054 		else if (imo->imo_multicast_addr.s_addr) {
2055 			/* return the value user has set */
2056 			addr = imo->imo_multicast_addr;
2057 		} else {
2058 			struct ifnet *ifp;
2059 			struct in_ifaddr *ia = NULL;
2060 			int s = pserialize_read_enter();
2061 
2062 			ifp = if_byindex(imo->imo_multicast_if_index);
2063 			if (ifp != NULL) {
2064 				ia = in_get_ia_from_ifp(ifp);
2065 			}
2066 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
2067 			pserialize_read_exit(s);
2068 		}
2069 		error = sockopt_set(sopt, &addr, sizeof(addr));
2070 		break;
2071 
2072 	case IP_MULTICAST_TTL:
2073 		optval = imo ? imo->imo_multicast_ttl
2074 		    : IP_DEFAULT_MULTICAST_TTL;
2075 
2076 		error = sockopt_set(sopt, &optval, sizeof(optval));
2077 		break;
2078 
2079 	case IP_MULTICAST_LOOP:
2080 		optval = imo ? imo->imo_multicast_loop
2081 		    : IP_DEFAULT_MULTICAST_LOOP;
2082 
2083 		error = sockopt_set(sopt, &optval, sizeof(optval));
2084 		break;
2085 
2086 	default:
2087 		error = EOPNOTSUPP;
2088 	}
2089 
2090 	return error;
2091 }
2092 
2093 /*
2094  * Discard the IP multicast options.
2095  */
2096 void
2097 ip_freemoptions(struct ip_moptions *imo)
2098 {
2099 	int i;
2100 
2101 	/* The owner of imo (inp) should be protected by solock */
2102 
2103 	if (imo != NULL) {
2104 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2105 			struct in_multi *inm = imo->imo_membership[i];
2106 			struct ifnet *ifp = inm->inm_ifp;
2107 			IFNET_LOCK(ifp);
2108 			in_delmulti(inm);
2109 			/* ifp should not leave thanks to solock */
2110 			IFNET_UNLOCK(ifp);
2111 		}
2112 
2113 		kmem_intr_free(imo, sizeof(*imo));
2114 	}
2115 }
2116 
2117 /*
2118  * Routine called from ip_output() to loop back a copy of an IP multicast
2119  * packet to the input queue of a specified interface.  Note that this
2120  * calls the output routine of the loopback "driver", but with an interface
2121  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
2122  */
2123 static void
2124 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
2125 {
2126 	struct ip *ip;
2127 	struct mbuf *copym;
2128 
2129 	copym = m_copypacket(m, M_DONTWAIT);
2130 	if (copym != NULL &&
2131 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
2132 		copym = m_pullup(copym, sizeof(struct ip));
2133 	if (copym == NULL)
2134 		return;
2135 	/*
2136 	 * We don't bother to fragment if the IP length is greater
2137 	 * than the interface's MTU.  Can this possibly matter?
2138 	 */
2139 	ip = mtod(copym, struct ip *);
2140 
2141 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
2142 		in_delayed_cksum(copym);
2143 		copym->m_pkthdr.csum_flags &=
2144 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
2145 	}
2146 
2147 	ip->ip_sum = 0;
2148 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
2149 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2150 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
2151 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2152 }
2153 
2154 /*
2155  * Ensure sending address is valid.
2156  * Returns 0 on success, -1 if an error should be sent back or 1
2157  * if the packet could be dropped without error (protocol dependent).
2158  */
2159 static int
2160 ip_ifaddrvalid(const struct in_ifaddr *ia)
2161 {
2162 
2163 	if (ia->ia_addr.sin_addr.s_addr == INADDR_ANY)
2164 		return 0;
2165 
2166 	if (ia->ia4_flags & IN_IFF_DUPLICATED)
2167 		return -1;
2168 	else if (ia->ia4_flags & (IN_IFF_TENTATIVE | IN_IFF_DETACHED))
2169 		return 1;
2170 
2171 	return 0;
2172 }
2173