xref: /netbsd-src/sys/netinet/ip_output.c (revision fc4f42693f9b1c31f39f9cf50af1bf2010325808)
1 /*	$NetBSD: ip_output.c,v 1.302 2018/04/13 09:00:29 maxv 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 1.302 2018/04/13 09:00:29 maxv 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 len, hlen, error = 0;
236 	struct route iproute;
237 	const struct sockaddr_in *dst;
238 	struct in_ifaddr *ia = NULL;
239 	struct ifaddr *ifa;
240 	int isbroadcast;
241 	int sw_csum;
242 	u_long mtu;
243 	bool natt_frag = false;
244 	bool rtmtu_nolock;
245 	union {
246 		struct sockaddr		sa;
247 		struct sockaddr_in	sin;
248 	} udst, usrc;
249 	struct sockaddr *rdst = &udst.sa;	/* real IP destination, as
250 						 * opposed to the nexthop
251 						 */
252 	struct psref psref, psref_ia;
253 	int bound;
254 	bool bind_need_restore = false;
255 
256 	len = 0;
257 
258 	MCLAIM(m, &ip_tx_mowner);
259 
260 	KASSERT((m->m_flags & M_PKTHDR) != 0);
261 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
262 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
263 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
264 	KASSERT(m->m_len >= sizeof(struct ip));
265 
266 	hlen = sizeof(struct ip);
267 	if (opt) {
268 		m = ip_insertoptions(m, opt, &len);
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 	 * RFC1112.
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 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 			 * TSO capable interfaces (typically?) increment
581 			 * ip_id for each segment.
582 			 * "allocate" enough ids here to increase the chance
583 			 * for them to be unique.
584 			 *
585 			 * note that the following calculation is not
586 			 * needed to be precise.  wasting some ip_id is fine.
587 			 */
588 
589 			unsigned int segsz = m->m_pkthdr.segsz;
590 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
591 			unsigned int num = howmany(datasz, segsz);
592 
593 			ip->ip_id = ip_newid_range(ia, num);
594 		}
595 	}
596 	if (ia != NULL) {
597 		ia4_release(ia, &psref_ia);
598 		ia = NULL;
599 	}
600 
601 	/*
602 	 * If we're doing Path MTU Discovery, we need to set DF unless
603 	 * the route's MTU is locked.
604 	 */
605 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
606 		ip->ip_off |= htons(IP_DF);
607 	}
608 
609 #ifdef IPSEC
610 	if (ipsec_used) {
611 		bool ipsec_done = false;
612 
613 		/* Perform IPsec processing, if any. */
614 		error = ipsec4_output(m, inp, flags, &mtu, &natt_frag,
615 		    &ipsec_done);
616 		if (error || ipsec_done)
617 			goto done;
618 	}
619 #endif
620 
621 	/*
622 	 * Run through list of hooks for output packets.
623 	 */
624 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
625 	if (error)
626 		goto done;
627 	if (m == NULL)
628 		goto done;
629 
630 	ip = mtod(m, struct ip *);
631 	hlen = ip->ip_hl << 2;
632 
633 	m->m_pkthdr.csum_data |= hlen << 16;
634 
635 	/*
636 	 * search for the source address structure to
637 	 * maintain output statistics, and verify address
638 	 * validity
639 	 */
640 	KASSERT(ia == NULL);
641 	sockaddr_in_init(&usrc.sin, &ip->ip_src, 0);
642 	ifa = ifaof_ifpforaddr_psref(&usrc.sa, ifp, &psref_ia);
643 	if (ifa != NULL)
644 		ia = ifatoia(ifa);
645 
646 	/*
647 	 * Ensure we only send from a valid address.
648 	 * A NULL address is valid because the packet could be
649 	 * generated from a packet filter.
650 	 */
651 	if (ia != NULL && (flags & IP_FORWARDING) == 0 &&
652 	    (error = ip_ifaddrvalid(ia)) != 0)
653 	{
654 		ARPLOG(LOG_ERR,
655 		    "refusing to send from invalid address %s (pid %d)\n",
656 		    ARPLOGADDR(&ip->ip_src), curproc->p_pid);
657 		IP_STATINC(IP_STAT_ODROPPED);
658 		if (error == 1)
659 			/*
660 			 * Address exists, but is tentative or detached.
661 			 * We can't send from it because it's invalid,
662 			 * so we drop the packet.
663 			 */
664 			error = 0;
665 		else
666 			error = EADDRNOTAVAIL;
667 		goto bad;
668 	}
669 
670 	/* Maybe skip checksums on loopback interfaces. */
671 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
672 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
673 	}
674 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
675 
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 fragment the packet.
730 	 *
731 	 * XXX Some hardware can do this.
732 	 */
733 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
734 		if (IN_NEED_CHECKSUM(ifp,
735 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
736 			in_delayed_cksum(m);
737 		}
738 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
739 	}
740 
741 	/*
742 	 * Too large for interface; fragment if possible.
743 	 * Must be able to put at least 8 bytes per fragment.
744 	 */
745 	if (ntohs(ip->ip_off) & IP_DF) {
746 		if (flags & IP_RETURNMTU) {
747 			KASSERT(inp != NULL);
748 			inp->inp_errormtu = mtu;
749 		}
750 		error = EMSGSIZE;
751 		IP_STATINC(IP_STAT_CANTFRAG);
752 		goto bad;
753 	}
754 
755 	error = ip_fragment(m, ifp, mtu);
756 	if (error) {
757 		m = NULL;
758 		goto bad;
759 	}
760 
761 	for (; m; m = m0) {
762 		m0 = m->m_nextpkt;
763 		m->m_nextpkt = NULL;
764 		if (error) {
765 			m_freem(m);
766 			continue;
767 		}
768 #if IFA_STATS
769 		if (ia)
770 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
771 #endif
772 		/*
773 		 * If we get there, the packet has not been handled by
774 		 * IPsec whereas it should have. Now that it has been
775 		 * fragmented, re-inject it in ip_output so that IPsec
776 		 * processing can occur.
777 		 */
778 		if (natt_frag) {
779 			error = ip_output(m, opt, ro,
780 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
781 			    imo, inp);
782 		} else {
783 			KASSERT((m->m_pkthdr.csum_flags &
784 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
785 			error = ip_if_output(ifp, m,
786 			    (m->m_flags & M_MCAST) ?
787 			    sintocsa(rdst) : sintocsa(dst), rt);
788 		}
789 	}
790 	if (error == 0) {
791 		IP_STATINC(IP_STAT_FRAGMENTED);
792 	}
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 
807 bad:
808 	m_freem(m);
809 	goto done;
810 }
811 
812 int
813 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
814 {
815 	struct ip *ip, *mhip;
816 	struct mbuf *m0;
817 	int len, hlen, off;
818 	int mhlen, firstlen;
819 	struct mbuf **mnext;
820 	int sw_csum = m->m_pkthdr.csum_flags;
821 	int fragments = 0;
822 	int error = 0;
823 	int ipoff;
824 	bool mff;
825 
826 	ip = mtod(m, struct ip *);
827 	hlen = ip->ip_hl << 2;
828 
829 	/* XXX: Why don't we remove IP_RF? */
830 	ipoff = ntohs(ip->ip_off) & ~IP_MF;
831 
832 	mff = (ip->ip_off & htons(IP_MF)) != 0;
833 
834 	if (ifp != NULL)
835 		sw_csum &= ~ifp->if_csum_flags_tx;
836 
837 	len = (mtu - hlen) &~ 7;
838 	if (len < 8) {
839 		m_freem(m);
840 		return EMSGSIZE;
841 	}
842 
843 	firstlen = len;
844 	mnext = &m->m_nextpkt;
845 
846 	/*
847 	 * Loop through length of segment after first fragment,
848 	 * make new header and copy data of each part and link onto chain.
849 	 */
850 	m0 = m;
851 	mhlen = sizeof(struct ip);
852 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
853 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
854 		if (m == NULL) {
855 			error = ENOBUFS;
856 			IP_STATINC(IP_STAT_ODROPPED);
857 			goto sendorfree;
858 		}
859 		MCLAIM(m, m0->m_owner);
860 
861 		*mnext = m;
862 		mnext = &m->m_nextpkt;
863 
864 		m->m_data += max_linkhdr;
865 		mhip = mtod(m, struct ip *);
866 		*mhip = *ip;
867 
868 		/* we must inherit MCAST and BCAST flags */
869 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
870 
871 		if (hlen > sizeof(struct ip)) {
872 			mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
873 			mhip->ip_hl = mhlen >> 2;
874 		}
875 		m->m_len = mhlen;
876 
877 		mhip->ip_off = ((off - hlen) >> 3) + ipoff;
878 		if (mff)
879 			mhip->ip_off |= IP_MF;
880 		if (off + len >= ntohs(ip->ip_len))
881 			len = ntohs(ip->ip_len) - off;
882 		else
883 			mhip->ip_off |= IP_MF;
884 		HTONS(mhip->ip_off);
885 
886 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
887 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
888 		if (m->m_next == NULL) {
889 			error = ENOBUFS;
890 			IP_STATINC(IP_STAT_ODROPPED);
891 			goto sendorfree;
892 		}
893 
894 		m->m_pkthdr.len = mhlen + len;
895 		m_reset_rcvif(m);
896 
897 		mhip->ip_sum = 0;
898 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
899 		if (sw_csum & M_CSUM_IPv4) {
900 			mhip->ip_sum = in_cksum(m, mhlen);
901 		} else {
902 			/*
903 			 * checksum is hw-offloaded or not necessary.
904 			 */
905 			m->m_pkthdr.csum_flags |=
906 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
907 			m->m_pkthdr.csum_data |= mhlen << 16;
908 			KASSERT(!(ifp != NULL &&
909 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
910 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
911 		}
912 		IP_STATINC(IP_STAT_OFRAGMENTS);
913 		fragments++;
914 	}
915 
916 	/*
917 	 * Update first fragment by trimming what's been copied out
918 	 * and updating header, then send each fragment (in order).
919 	 */
920 	m = m0;
921 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
922 	m->m_pkthdr.len = hlen + firstlen;
923 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
924 	ip->ip_off |= htons(IP_MF);
925 	ip->ip_sum = 0;
926 	if (sw_csum & M_CSUM_IPv4) {
927 		ip->ip_sum = in_cksum(m, hlen);
928 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
929 	} else {
930 		/*
931 		 * checksum is hw-offloaded or not necessary.
932 		 */
933 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
934 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
935 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
936 		    sizeof(struct ip));
937 	}
938 
939 sendorfree:
940 	/*
941 	 * If there is no room for all the fragments, don't queue
942 	 * any of them.
943 	 */
944 	if (ifp != NULL) {
945 		IFQ_LOCK(&ifp->if_snd);
946 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
947 		    error == 0) {
948 			error = ENOBUFS;
949 			IP_STATINC(IP_STAT_ODROPPED);
950 			IFQ_INC_DROPS(&ifp->if_snd);
951 		}
952 		IFQ_UNLOCK(&ifp->if_snd);
953 	}
954 	if (error) {
955 		for (m = m0; m; m = m0) {
956 			m0 = m->m_nextpkt;
957 			m->m_nextpkt = NULL;
958 			m_freem(m);
959 		}
960 	}
961 
962 	return error;
963 }
964 
965 /*
966  * Process a delayed payload checksum calculation.
967  */
968 void
969 in_delayed_cksum(struct mbuf *m)
970 {
971 	struct ip *ip;
972 	u_int16_t csum, offset;
973 
974 	ip = mtod(m, struct ip *);
975 	offset = ip->ip_hl << 2;
976 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
977 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
978 		csum = 0xffff;
979 
980 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
981 
982 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
983 		/* This happens when ip options were inserted */
984 		m_copyback(m, offset, sizeof(csum), (void *)&csum);
985 	} else {
986 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
987 	}
988 }
989 
990 /*
991  * Determine the maximum length of the options to be inserted;
992  * we would far rather allocate too much space rather than too little.
993  */
994 u_int
995 ip_optlen(struct inpcb *inp)
996 {
997 	struct mbuf *m = inp->inp_options;
998 
999 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
1000 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
1001 	}
1002 	return 0;
1003 }
1004 
1005 /*
1006  * Insert IP options into preformed packet.
1007  * Adjust IP destination as required for IP source routing,
1008  * as indicated by a non-zero in_addr at the start of the options.
1009  */
1010 static struct mbuf *
1011 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1012 {
1013 	struct ipoption *p = mtod(opt, struct ipoption *);
1014 	struct mbuf *n;
1015 	struct ip *ip = mtod(m, struct ip *);
1016 	unsigned optlen;
1017 
1018 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1019 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
1020 		return m;		/* XXX should fail */
1021 	if (!in_nullhost(p->ipopt_dst))
1022 		ip->ip_dst = p->ipopt_dst;
1023 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
1024 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1025 		if (n == NULL)
1026 			return m;
1027 		MCLAIM(n, m->m_owner);
1028 		M_MOVE_PKTHDR(n, m);
1029 		m->m_len -= sizeof(struct ip);
1030 		m->m_data += sizeof(struct ip);
1031 		n->m_next = m;
1032 		n->m_len = optlen + sizeof(struct ip);
1033 		n->m_data += max_linkhdr;
1034 		memcpy(mtod(n, void *), ip, sizeof(struct ip));
1035 		m = n;
1036 	} else {
1037 		m->m_data -= optlen;
1038 		m->m_len += optlen;
1039 		memmove(mtod(m, void *), ip, sizeof(struct ip));
1040 	}
1041 	m->m_pkthdr.len += optlen;
1042 	ip = mtod(m, struct ip *);
1043 	memcpy(ip + 1, p->ipopt_list, optlen);
1044 	*phlen = sizeof(struct ip) + optlen;
1045 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1046 	return m;
1047 }
1048 
1049 /*
1050  * Copy options from ipsrc to ipdst, omitting those not copied during
1051  * fragmentation.
1052  */
1053 int
1054 ip_optcopy(struct ip *ipsrc, struct ip *ipdst)
1055 {
1056 	u_char *cp, *dp;
1057 	int opt, optlen, cnt;
1058 
1059 	cp = (u_char *)(ipsrc + 1);
1060 	dp = (u_char *)(ipdst + 1);
1061 	cnt = (ipsrc->ip_hl << 2) - sizeof(struct ip);
1062 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1063 		opt = cp[0];
1064 		if (opt == IPOPT_EOL)
1065 			break;
1066 		if (opt == IPOPT_NOP) {
1067 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1068 			*dp++ = IPOPT_NOP;
1069 			optlen = 1;
1070 			continue;
1071 		}
1072 
1073 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
1074 		optlen = cp[IPOPT_OLEN];
1075 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
1076 
1077 		/* Invalid lengths should have been caught by ip_dooptions. */
1078 		if (optlen > cnt)
1079 			optlen = cnt;
1080 		if (IPOPT_COPIED(opt)) {
1081 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1082 			dp += optlen;
1083 		}
1084 	}
1085 
1086 	for (optlen = dp - (u_char *)(ipdst+1); optlen & 0x3; optlen++) {
1087 		*dp++ = IPOPT_EOL;
1088 	}
1089 
1090 	return optlen;
1091 }
1092 
1093 /*
1094  * IP socket option processing.
1095  */
1096 int
1097 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1098 {
1099 	struct inpcb *inp = sotoinpcb(so);
1100 	struct ip *ip = &inp->inp_ip;
1101 	int inpflags = inp->inp_flags;
1102 	int optval = 0, error = 0;
1103 	struct in_pktinfo pktinfo;
1104 
1105 	KASSERT(solocked(so));
1106 
1107 	if (sopt->sopt_level != IPPROTO_IP) {
1108 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1109 			return 0;
1110 		return ENOPROTOOPT;
1111 	}
1112 
1113 	switch (op) {
1114 	case PRCO_SETOPT:
1115 		switch (sopt->sopt_name) {
1116 		case IP_OPTIONS:
1117 #ifdef notyet
1118 		case IP_RETOPTS:
1119 #endif
1120 			error = ip_pcbopts(inp, sopt);
1121 			break;
1122 
1123 		case IP_TOS:
1124 		case IP_TTL:
1125 		case IP_MINTTL:
1126 		case IP_RECVOPTS:
1127 		case IP_RECVRETOPTS:
1128 		case IP_RECVDSTADDR:
1129 		case IP_RECVIF:
1130 		case IP_RECVPKTINFO:
1131 		case IP_RECVTTL:
1132 			error = sockopt_getint(sopt, &optval);
1133 			if (error)
1134 				break;
1135 
1136 			switch (sopt->sopt_name) {
1137 			case IP_TOS:
1138 				ip->ip_tos = optval;
1139 				break;
1140 
1141 			case IP_TTL:
1142 				ip->ip_ttl = optval;
1143 				break;
1144 
1145 			case IP_MINTTL:
1146 				if (optval > 0 && optval <= MAXTTL)
1147 					inp->inp_ip_minttl = optval;
1148 				else
1149 					error = EINVAL;
1150 				break;
1151 #define	OPTSET(bit) \
1152 	if (optval) \
1153 		inpflags |= bit; \
1154 	else \
1155 		inpflags &= ~bit;
1156 
1157 			case IP_RECVOPTS:
1158 				OPTSET(INP_RECVOPTS);
1159 				break;
1160 
1161 			case IP_RECVPKTINFO:
1162 				OPTSET(INP_RECVPKTINFO);
1163 				break;
1164 
1165 			case IP_RECVRETOPTS:
1166 				OPTSET(INP_RECVRETOPTS);
1167 				break;
1168 
1169 			case IP_RECVDSTADDR:
1170 				OPTSET(INP_RECVDSTADDR);
1171 				break;
1172 
1173 			case IP_RECVIF:
1174 				OPTSET(INP_RECVIF);
1175 				break;
1176 
1177 			case IP_RECVTTL:
1178 				OPTSET(INP_RECVTTL);
1179 				break;
1180 			}
1181 			break;
1182 		case IP_PKTINFO:
1183 			error = sockopt_getint(sopt, &optval);
1184 			if (!error) {
1185 				/* Linux compatibility */
1186 				OPTSET(INP_RECVPKTINFO);
1187 				break;
1188 			}
1189 			error = sockopt_get(sopt, &pktinfo, sizeof(pktinfo));
1190 			if (error)
1191 				break;
1192 
1193 			if (pktinfo.ipi_ifindex == 0) {
1194 				inp->inp_prefsrcip = pktinfo.ipi_addr;
1195 				break;
1196 			}
1197 
1198 			/* Solaris compatibility */
1199 			struct ifnet *ifp;
1200 			struct in_ifaddr *ia;
1201 			int s;
1202 
1203 			/* pick up primary address */
1204 			s = pserialize_read_enter();
1205 			ifp = if_byindex(pktinfo.ipi_ifindex);
1206 			if (ifp == NULL) {
1207 				pserialize_read_exit(s);
1208 				error = EADDRNOTAVAIL;
1209 				break;
1210 			}
1211 			ia = in_get_ia_from_ifp(ifp);
1212 			if (ia == NULL) {
1213 				pserialize_read_exit(s);
1214 				error = EADDRNOTAVAIL;
1215 				break;
1216 			}
1217 			inp->inp_prefsrcip = IA_SIN(ia)->sin_addr;
1218 			pserialize_read_exit(s);
1219 			break;
1220 		break;
1221 #undef OPTSET
1222 
1223 		case IP_MULTICAST_IF:
1224 		case IP_MULTICAST_TTL:
1225 		case IP_MULTICAST_LOOP:
1226 		case IP_ADD_MEMBERSHIP:
1227 		case IP_DROP_MEMBERSHIP:
1228 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1229 			break;
1230 
1231 		case IP_PORTRANGE:
1232 			error = sockopt_getint(sopt, &optval);
1233 			if (error)
1234 				break;
1235 
1236 			switch (optval) {
1237 			case IP_PORTRANGE_DEFAULT:
1238 			case IP_PORTRANGE_HIGH:
1239 				inpflags &= ~(INP_LOWPORT);
1240 				break;
1241 
1242 			case IP_PORTRANGE_LOW:
1243 				inpflags |= INP_LOWPORT;
1244 				break;
1245 
1246 			default:
1247 				error = EINVAL;
1248 				break;
1249 			}
1250 			break;
1251 
1252 		case IP_PORTALGO:
1253 			error = sockopt_getint(sopt, &optval);
1254 			if (error)
1255 				break;
1256 
1257 			error = portalgo_algo_index_select(
1258 			    (struct inpcb_hdr *)inp, optval);
1259 			break;
1260 
1261 #if defined(IPSEC)
1262 		case IP_IPSEC_POLICY:
1263 			if (ipsec_enabled) {
1264 				error = ipsec_set_policy(inp, sopt->sopt_name,
1265 				    sopt->sopt_data, sopt->sopt_size,
1266 				    curlwp->l_cred);
1267 				break;
1268 			}
1269 			/*FALLTHROUGH*/
1270 #endif /* IPSEC */
1271 
1272 		default:
1273 			error = ENOPROTOOPT;
1274 			break;
1275 		}
1276 		break;
1277 
1278 	case PRCO_GETOPT:
1279 		switch (sopt->sopt_name) {
1280 		case IP_OPTIONS:
1281 		case IP_RETOPTS: {
1282 			struct mbuf *mopts = inp->inp_options;
1283 
1284 			if (mopts) {
1285 				struct mbuf *m;
1286 
1287 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1288 				if (m == NULL) {
1289 					error = ENOBUFS;
1290 					break;
1291 				}
1292 				error = sockopt_setmbuf(sopt, m);
1293 			}
1294 			break;
1295 		}
1296 		case IP_TOS:
1297 		case IP_TTL:
1298 		case IP_MINTTL:
1299 		case IP_RECVOPTS:
1300 		case IP_RECVRETOPTS:
1301 		case IP_RECVDSTADDR:
1302 		case IP_RECVIF:
1303 		case IP_RECVPKTINFO:
1304 		case IP_RECVTTL:
1305 		case IP_ERRORMTU:
1306 			switch (sopt->sopt_name) {
1307 			case IP_TOS:
1308 				optval = ip->ip_tos;
1309 				break;
1310 
1311 			case IP_TTL:
1312 				optval = ip->ip_ttl;
1313 				break;
1314 
1315 			case IP_MINTTL:
1316 				optval = inp->inp_ip_minttl;
1317 				break;
1318 
1319 			case IP_ERRORMTU:
1320 				optval = inp->inp_errormtu;
1321 				break;
1322 
1323 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1324 
1325 			case IP_RECVOPTS:
1326 				optval = OPTBIT(INP_RECVOPTS);
1327 				break;
1328 
1329 			case IP_RECVPKTINFO:
1330 				optval = OPTBIT(INP_RECVPKTINFO);
1331 				break;
1332 
1333 			case IP_RECVRETOPTS:
1334 				optval = OPTBIT(INP_RECVRETOPTS);
1335 				break;
1336 
1337 			case IP_RECVDSTADDR:
1338 				optval = OPTBIT(INP_RECVDSTADDR);
1339 				break;
1340 
1341 			case IP_RECVIF:
1342 				optval = OPTBIT(INP_RECVIF);
1343 				break;
1344 
1345 			case IP_RECVTTL:
1346 				optval = OPTBIT(INP_RECVTTL);
1347 				break;
1348 			}
1349 			error = sockopt_setint(sopt, optval);
1350 			break;
1351 
1352 		case IP_PKTINFO:
1353 			switch (sopt->sopt_size) {
1354 			case sizeof(int):
1355 				/* Linux compatibility */
1356 				optval = OPTBIT(INP_RECVPKTINFO);
1357 				error = sockopt_setint(sopt, optval);
1358 				break;
1359 			case sizeof(struct in_pktinfo):
1360 				/* Solaris compatibility */
1361 				pktinfo.ipi_ifindex = 0;
1362 				pktinfo.ipi_addr = inp->inp_prefsrcip;
1363 				error = sockopt_set(sopt, &pktinfo,
1364 				    sizeof(pktinfo));
1365 				break;
1366 			default:
1367 				/*
1368 				 * While size is stuck at 0, and, later, if
1369 				 * the caller doesn't use an exactly sized
1370 				 * recipient for the data, default to Linux
1371 				 * compatibility
1372 				 */
1373 				optval = OPTBIT(INP_RECVPKTINFO);
1374 				error = sockopt_setint(sopt, optval);
1375 				break;
1376 			}
1377 			break;
1378 
1379 #if 0	/* defined(IPSEC) */
1380 		case IP_IPSEC_POLICY:
1381 		{
1382 			struct mbuf *m = NULL;
1383 
1384 			/* XXX this will return EINVAL as sopt is empty */
1385 			error = ipsec_get_policy(inp, sopt->sopt_data,
1386 			    sopt->sopt_size, &m);
1387 			if (error == 0)
1388 				error = sockopt_setmbuf(sopt, m);
1389 			break;
1390 		}
1391 #endif /*IPSEC*/
1392 
1393 		case IP_MULTICAST_IF:
1394 		case IP_MULTICAST_TTL:
1395 		case IP_MULTICAST_LOOP:
1396 		case IP_ADD_MEMBERSHIP:
1397 		case IP_DROP_MEMBERSHIP:
1398 			error = ip_getmoptions(inp->inp_moptions, sopt);
1399 			break;
1400 
1401 		case IP_PORTRANGE:
1402 			if (inpflags & INP_LOWPORT)
1403 				optval = IP_PORTRANGE_LOW;
1404 			else
1405 				optval = IP_PORTRANGE_DEFAULT;
1406 			error = sockopt_setint(sopt, optval);
1407 			break;
1408 
1409 		case IP_PORTALGO:
1410 			optval = inp->inp_portalgo;
1411 			error = sockopt_setint(sopt, optval);
1412 			break;
1413 
1414 		default:
1415 			error = ENOPROTOOPT;
1416 			break;
1417 		}
1418 		break;
1419 	}
1420 
1421 	if (!error) {
1422 		inp->inp_flags = inpflags;
1423 	}
1424 	return error;
1425 }
1426 
1427 static int
1428 ip_pktinfo_prepare(const struct in_pktinfo *pktinfo, struct ip_pktopts *pktopts,
1429     int *flags, kauth_cred_t cred)
1430 {
1431 	struct ip_moptions *imo;
1432 	int error = 0;
1433 	bool addrset = false;
1434 
1435 	if (!in_nullhost(pktinfo->ipi_addr)) {
1436 		pktopts->ippo_laddr.sin_addr = pktinfo->ipi_addr;
1437 		/* EADDRNOTAVAIL? */
1438 		error = in_pcbbindableaddr(&pktopts->ippo_laddr, cred);
1439 		if (error != 0)
1440 			return error;
1441 		addrset = true;
1442 	}
1443 
1444 	if (pktinfo->ipi_ifindex != 0) {
1445 		if (!addrset) {
1446 			struct ifnet *ifp;
1447 			struct in_ifaddr *ia;
1448 			int s;
1449 
1450 			/* pick up primary address */
1451 			s = pserialize_read_enter();
1452 			ifp = if_byindex(pktinfo->ipi_ifindex);
1453 			if (ifp == NULL) {
1454 				pserialize_read_exit(s);
1455 				return EADDRNOTAVAIL;
1456 			}
1457 			ia = in_get_ia_from_ifp(ifp);
1458 			if (ia == NULL) {
1459 				pserialize_read_exit(s);
1460 				return EADDRNOTAVAIL;
1461 			}
1462 			pktopts->ippo_laddr.sin_addr = IA_SIN(ia)->sin_addr;
1463 			pserialize_read_exit(s);
1464 		}
1465 
1466 		/*
1467 		 * If specified ipi_ifindex,
1468 		 * use copied or locally initialized ip_moptions.
1469 		 * Original ip_moptions must not be modified.
1470 		 */
1471 		imo = &pktopts->ippo_imobuf;	/* local buf in pktopts */
1472 		if (pktopts->ippo_imo != NULL) {
1473 			memcpy(imo, pktopts->ippo_imo, sizeof(*imo));
1474 		} else {
1475 			memset(imo, 0, sizeof(*imo));
1476 			imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1477 			imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1478 		}
1479 		imo->imo_multicast_if_index = pktinfo->ipi_ifindex;
1480 		pktopts->ippo_imo = imo;
1481 		*flags |= IP_ROUTETOIFINDEX;
1482 	}
1483 	return error;
1484 }
1485 
1486 /*
1487  * Set up IP outgoing packet options. Even if control is NULL,
1488  * pktopts->ippo_laddr and pktopts->ippo_imo are set and used.
1489  */
1490 int
1491 ip_setpktopts(struct mbuf *control, struct ip_pktopts *pktopts, int *flags,
1492     struct inpcb *inp, kauth_cred_t cred)
1493 {
1494 	struct cmsghdr *cm;
1495 	struct in_pktinfo pktinfo;
1496 	int error;
1497 
1498 	pktopts->ippo_imo = inp->inp_moptions;
1499 
1500 	struct in_addr *ia = in_nullhost(inp->inp_prefsrcip) ? &inp->inp_laddr :
1501 	    &inp->inp_prefsrcip;
1502 	sockaddr_in_init(&pktopts->ippo_laddr, ia, 0);
1503 
1504 	if (control == NULL)
1505 		return 0;
1506 
1507 	/*
1508 	 * XXX: Currently, we assume all the optional information is
1509 	 * stored in a single mbuf.
1510 	 */
1511 	if (control->m_next)
1512 		return EINVAL;
1513 
1514 	for (; control->m_len > 0;
1515 	    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1516 	    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1517 		cm = mtod(control, struct cmsghdr *);
1518 		if ((control->m_len < sizeof(*cm)) ||
1519 		    (cm->cmsg_len == 0) ||
1520 		    (cm->cmsg_len > control->m_len)) {
1521 			return EINVAL;
1522 		}
1523 		if (cm->cmsg_level != IPPROTO_IP)
1524 			continue;
1525 
1526 		switch (cm->cmsg_type) {
1527 		case IP_PKTINFO:
1528 			if (cm->cmsg_len != CMSG_LEN(sizeof(pktinfo)))
1529 				return EINVAL;
1530 			memcpy(&pktinfo, CMSG_DATA(cm), sizeof(pktinfo));
1531 			error = ip_pktinfo_prepare(&pktinfo, pktopts, flags,
1532 			    cred);
1533 			if (error)
1534 				return error;
1535 			break;
1536 		case IP_SENDSRCADDR: /* FreeBSD compatibility */
1537 			if (cm->cmsg_len != CMSG_LEN(sizeof(struct in_addr)))
1538 				return EINVAL;
1539 			pktinfo.ipi_ifindex = 0;
1540 			pktinfo.ipi_addr =
1541 			    ((struct in_pktinfo *)CMSG_DATA(cm))->ipi_addr;
1542 			error = ip_pktinfo_prepare(&pktinfo, pktopts, flags,
1543 			    cred);
1544 			if (error)
1545 				return error;
1546 			break;
1547 		default:
1548 			return ENOPROTOOPT;
1549 		}
1550 	}
1551 	return 0;
1552 }
1553 
1554 /*
1555  * Set up IP options in pcb for insertion in output packets.
1556  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1557  * with destination address if source routed.
1558  */
1559 static int
1560 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1561 {
1562 	struct mbuf *m;
1563 	const u_char *cp;
1564 	u_char *dp;
1565 	int cnt;
1566 
1567 	KASSERT(inp_locked(inp));
1568 
1569 	/* Turn off any old options. */
1570 	if (inp->inp_options) {
1571 		m_free(inp->inp_options);
1572 	}
1573 	inp->inp_options = NULL;
1574 	if ((cnt = sopt->sopt_size) == 0) {
1575 		/* Only turning off any previous options. */
1576 		return 0;
1577 	}
1578 	cp = sopt->sopt_data;
1579 
1580 #ifndef	__vax__
1581 	if (cnt % sizeof(int32_t))
1582 		return EINVAL;
1583 #endif
1584 
1585 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1586 	if (m == NULL)
1587 		return ENOBUFS;
1588 
1589 	dp = mtod(m, u_char *);
1590 	memset(dp, 0, sizeof(struct in_addr));
1591 	dp += sizeof(struct in_addr);
1592 	m->m_len = sizeof(struct in_addr);
1593 
1594 	/*
1595 	 * IP option list according to RFC791. Each option is of the form
1596 	 *
1597 	 *	[optval] [olen] [(olen - 2) data bytes]
1598 	 *
1599 	 * We validate the list and copy options to an mbuf for prepending
1600 	 * to data packets. The IP first-hop destination address will be
1601 	 * stored before actual options and is zero if unset.
1602 	 */
1603 	while (cnt > 0) {
1604 		uint8_t optval, olen, offset;
1605 
1606 		optval = cp[IPOPT_OPTVAL];
1607 
1608 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1609 			olen = 1;
1610 		} else {
1611 			if (cnt < IPOPT_OLEN + 1)
1612 				goto bad;
1613 
1614 			olen = cp[IPOPT_OLEN];
1615 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1616 				goto bad;
1617 		}
1618 
1619 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1620 			/*
1621 			 * user process specifies route as:
1622 			 *	->A->B->C->D
1623 			 * D must be our final destination (but we can't
1624 			 * check that since we may not have connected yet).
1625 			 * A is first hop destination, which doesn't appear in
1626 			 * actual IP option, but is stored before the options.
1627 			 */
1628 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1629 				goto bad;
1630 
1631 			offset = cp[IPOPT_OFFSET];
1632 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1633 			    sizeof(struct in_addr));
1634 
1635 			cp += sizeof(struct in_addr);
1636 			cnt -= sizeof(struct in_addr);
1637 			olen -= sizeof(struct in_addr);
1638 
1639 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1640 				goto bad;
1641 
1642 			memcpy(dp, cp, olen);
1643 			dp[IPOPT_OPTVAL] = optval;
1644 			dp[IPOPT_OLEN] = olen;
1645 			dp[IPOPT_OFFSET] = offset;
1646 			break;
1647 		} else {
1648 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1649 				goto bad;
1650 
1651 			memcpy(dp, cp, olen);
1652 			break;
1653 		}
1654 
1655 		dp += olen;
1656 		m->m_len += olen;
1657 
1658 		if (optval == IPOPT_EOL)
1659 			break;
1660 
1661 		cp += olen;
1662 		cnt -= olen;
1663 	}
1664 
1665 	inp->inp_options = m;
1666 	return 0;
1667 
1668 bad:
1669 	(void)m_free(m);
1670 	return EINVAL;
1671 }
1672 
1673 /*
1674  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1675  * Must be called in a pserialize critical section.
1676  */
1677 static struct ifnet *
1678 ip_multicast_if(struct in_addr *a, int *ifindexp)
1679 {
1680 	int ifindex;
1681 	struct ifnet *ifp = NULL;
1682 	struct in_ifaddr *ia;
1683 
1684 	if (ifindexp)
1685 		*ifindexp = 0;
1686 	if (ntohl(a->s_addr) >> 24 == 0) {
1687 		ifindex = ntohl(a->s_addr) & 0xffffff;
1688 		ifp = if_byindex(ifindex);
1689 		if (!ifp)
1690 			return NULL;
1691 		if (ifindexp)
1692 			*ifindexp = ifindex;
1693 	} else {
1694 		IN_ADDRHASH_READER_FOREACH(ia, a->s_addr) {
1695 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1696 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1697 				ifp = ia->ia_ifp;
1698 				if (if_is_deactivated(ifp))
1699 					ifp = NULL;
1700 				break;
1701 			}
1702 		}
1703 	}
1704 	return ifp;
1705 }
1706 
1707 static int
1708 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1709 {
1710 	u_int tval;
1711 	u_char cval;
1712 	int error;
1713 
1714 	if (sopt == NULL)
1715 		return EINVAL;
1716 
1717 	switch (sopt->sopt_size) {
1718 	case sizeof(u_char):
1719 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1720 		tval = cval;
1721 		break;
1722 
1723 	case sizeof(u_int):
1724 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1725 		break;
1726 
1727 	default:
1728 		error = EINVAL;
1729 	}
1730 
1731 	if (error)
1732 		return error;
1733 
1734 	if (tval > maxval)
1735 		return EINVAL;
1736 
1737 	*val = tval;
1738 	return 0;
1739 }
1740 
1741 static int
1742 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1743     struct psref *psref, struct in_addr *ia, bool add)
1744 {
1745 	int error;
1746 	struct ip_mreq mreq;
1747 
1748 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1749 	if (error)
1750 		return error;
1751 
1752 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1753 		return EINVAL;
1754 
1755 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1756 
1757 	if (in_nullhost(mreq.imr_interface)) {
1758 		union {
1759 			struct sockaddr		dst;
1760 			struct sockaddr_in	dst4;
1761 		} u;
1762 		struct route ro;
1763 
1764 		if (!add) {
1765 			*ifp = NULL;
1766 			return 0;
1767 		}
1768 		/*
1769 		 * If no interface address was provided, use the interface of
1770 		 * the route to the given multicast address.
1771 		 */
1772 		struct rtentry *rt;
1773 		memset(&ro, 0, sizeof(ro));
1774 
1775 		sockaddr_in_init(&u.dst4, ia, 0);
1776 		error = rtcache_setdst(&ro, &u.dst);
1777 		if (error != 0)
1778 			return error;
1779 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1780 		if (*ifp != NULL) {
1781 			if (if_is_deactivated(*ifp))
1782 				*ifp = NULL;
1783 			else
1784 				if_acquire(*ifp, psref);
1785 		}
1786 		rtcache_unref(rt, &ro);
1787 		rtcache_free(&ro);
1788 	} else {
1789 		int s = pserialize_read_enter();
1790 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1791 		if (!add && *ifp == NULL) {
1792 			pserialize_read_exit(s);
1793 			return EADDRNOTAVAIL;
1794 		}
1795 		if (*ifp != NULL) {
1796 			if (if_is_deactivated(*ifp))
1797 				*ifp = NULL;
1798 			else
1799 				if_acquire(*ifp, psref);
1800 		}
1801 		pserialize_read_exit(s);
1802 	}
1803 	return 0;
1804 }
1805 
1806 /*
1807  * Add a multicast group membership.
1808  * Group must be a valid IP multicast address.
1809  */
1810 static int
1811 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1812 {
1813 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1814 	struct in_addr ia;
1815 	int i, error, bound;
1816 	struct psref psref;
1817 
1818 	/* imo is protected by solock or referenced only by the caller */
1819 
1820 	bound = curlwp_bind();
1821 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1822 		error = ip_get_membership(sopt, &ifp, &psref, &ia, true);
1823 	else {
1824 #ifdef INET6
1825 		error = ip6_get_membership(sopt, &ifp, &psref, &ia, sizeof(ia));
1826 #else
1827 		error = EINVAL;
1828 #endif
1829 	}
1830 
1831 	if (error)
1832 		goto out;
1833 
1834 	/*
1835 	 * See if we found an interface, and confirm that it
1836 	 * supports multicast.
1837 	 */
1838 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1839 		error = EADDRNOTAVAIL;
1840 		goto out;
1841 	}
1842 
1843 	/*
1844 	 * See if the membership already exists or if all the
1845 	 * membership slots are full.
1846 	 */
1847 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1848 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1849 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1850 			break;
1851 	}
1852 	if (i < imo->imo_num_memberships) {
1853 		error = EADDRINUSE;
1854 		goto out;
1855 	}
1856 
1857 	if (i == IP_MAX_MEMBERSHIPS) {
1858 		error = ETOOMANYREFS;
1859 		goto out;
1860 	}
1861 
1862 	/*
1863 	 * Everything looks good; add a new record to the multicast
1864 	 * address list for the given interface.
1865 	 */
1866 	IFNET_LOCK(ifp);
1867 	imo->imo_membership[i] = in_addmulti(&ia, ifp);
1868 	IFNET_UNLOCK(ifp);
1869 	if (imo->imo_membership[i] == NULL) {
1870 		error = ENOBUFS;
1871 		goto out;
1872 	}
1873 
1874 	++imo->imo_num_memberships;
1875 	error = 0;
1876 out:
1877 	if_put(ifp, &psref);
1878 	curlwp_bindx(bound);
1879 	return error;
1880 }
1881 
1882 /*
1883  * Drop a multicast group membership.
1884  * Group must be a valid IP multicast address.
1885  */
1886 static int
1887 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1888 {
1889 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1890 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1891 	int i, error, bound;
1892 	struct psref psref;
1893 
1894 	/* imo is protected by solock or referenced only by the caller */
1895 
1896 	bound = curlwp_bind();
1897 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1898 		error = ip_get_membership(sopt, &ifp, &psref, &ia, false);
1899 	else {
1900 #ifdef INET6
1901 		error = ip6_get_membership(sopt, &ifp, &psref, &ia, sizeof(ia));
1902 #else
1903 		error = EINVAL;
1904 #endif
1905 	}
1906 
1907 	if (error)
1908 		goto out;
1909 
1910 	/*
1911 	 * Find the membership in the membership array.
1912 	 */
1913 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1914 		if ((ifp == NULL ||
1915 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1916 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1917 			break;
1918 	}
1919 	if (i == imo->imo_num_memberships) {
1920 		error = EADDRNOTAVAIL;
1921 		goto out;
1922 	}
1923 
1924 	/*
1925 	 * Give up the multicast address record to which the
1926 	 * membership points.
1927 	 */
1928 	struct ifnet *inm_ifp = imo->imo_membership[i]->inm_ifp;
1929 	IFNET_LOCK(inm_ifp);
1930 	in_delmulti(imo->imo_membership[i]);
1931 	IFNET_UNLOCK(inm_ifp);
1932 
1933 	/*
1934 	 * Remove the gap in the membership array.
1935 	 */
1936 	for (++i; i < imo->imo_num_memberships; ++i)
1937 		imo->imo_membership[i-1] = imo->imo_membership[i];
1938 	--imo->imo_num_memberships;
1939 	error = 0;
1940 out:
1941 	if_put(ifp, &psref);
1942 	curlwp_bindx(bound);
1943 	return error;
1944 }
1945 
1946 /*
1947  * Set the IP multicast options in response to user setsockopt().
1948  */
1949 int
1950 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1951 {
1952 	struct ip_moptions *imo = *pimo;
1953 	struct in_addr addr;
1954 	struct ifnet *ifp;
1955 	int ifindex, error = 0;
1956 
1957 	/* The passed imo isn't NULL, it should be protected by solock */
1958 
1959 	if (!imo) {
1960 		/*
1961 		 * No multicast option buffer attached to the pcb;
1962 		 * allocate one and initialize to default values.
1963 		 */
1964 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1965 		if (imo == NULL)
1966 			return ENOBUFS;
1967 
1968 		imo->imo_multicast_if_index = 0;
1969 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1970 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1971 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1972 		imo->imo_num_memberships = 0;
1973 		*pimo = imo;
1974 	}
1975 
1976 	switch (sopt->sopt_name) {
1977 	case IP_MULTICAST_IF: {
1978 		int s;
1979 		/*
1980 		 * Select the interface for outgoing multicast packets.
1981 		 */
1982 		error = sockopt_get(sopt, &addr, sizeof(addr));
1983 		if (error)
1984 			break;
1985 
1986 		/*
1987 		 * INADDR_ANY is used to remove a previous selection.
1988 		 * When no interface is selected, a default one is
1989 		 * chosen every time a multicast packet is sent.
1990 		 */
1991 		if (in_nullhost(addr)) {
1992 			imo->imo_multicast_if_index = 0;
1993 			break;
1994 		}
1995 		/*
1996 		 * The selected interface is identified by its local
1997 		 * IP address.  Find the interface and confirm that
1998 		 * it supports multicasting.
1999 		 */
2000 		s = pserialize_read_enter();
2001 		ifp = ip_multicast_if(&addr, &ifindex);
2002 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2003 			pserialize_read_exit(s);
2004 			error = EADDRNOTAVAIL;
2005 			break;
2006 		}
2007 		imo->imo_multicast_if_index = ifp->if_index;
2008 		pserialize_read_exit(s);
2009 		if (ifindex)
2010 			imo->imo_multicast_addr = addr;
2011 		else
2012 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
2013 		break;
2014 	    }
2015 
2016 	case IP_MULTICAST_TTL:
2017 		/*
2018 		 * Set the IP time-to-live for outgoing multicast packets.
2019 		 */
2020 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
2021 		break;
2022 
2023 	case IP_MULTICAST_LOOP:
2024 		/*
2025 		 * Set the loopback flag for outgoing multicast packets.
2026 		 * Must be zero or one.
2027 		 */
2028 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
2029 		break;
2030 
2031 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
2032 		error = ip_add_membership(imo, sopt);
2033 		break;
2034 
2035 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
2036 		error = ip_drop_membership(imo, sopt);
2037 		break;
2038 
2039 	default:
2040 		error = EOPNOTSUPP;
2041 		break;
2042 	}
2043 
2044 	/*
2045 	 * If all options have default values, no need to keep the mbuf.
2046 	 */
2047 	if (imo->imo_multicast_if_index == 0 &&
2048 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2049 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2050 	    imo->imo_num_memberships == 0) {
2051 		kmem_intr_free(imo, sizeof(*imo));
2052 		*pimo = NULL;
2053 	}
2054 
2055 	return error;
2056 }
2057 
2058 /*
2059  * Return the IP multicast options in response to user getsockopt().
2060  */
2061 int
2062 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
2063 {
2064 	struct in_addr addr;
2065 	uint8_t optval;
2066 	int error = 0;
2067 
2068 	/* imo is protected by solock or refereced only by the caller */
2069 
2070 	switch (sopt->sopt_name) {
2071 	case IP_MULTICAST_IF:
2072 		if (imo == NULL || imo->imo_multicast_if_index == 0)
2073 			addr = zeroin_addr;
2074 		else if (imo->imo_multicast_addr.s_addr) {
2075 			/* return the value user has set */
2076 			addr = imo->imo_multicast_addr;
2077 		} else {
2078 			struct ifnet *ifp;
2079 			struct in_ifaddr *ia = NULL;
2080 			int s = pserialize_read_enter();
2081 
2082 			ifp = if_byindex(imo->imo_multicast_if_index);
2083 			if (ifp != NULL) {
2084 				ia = in_get_ia_from_ifp(ifp);
2085 			}
2086 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
2087 			pserialize_read_exit(s);
2088 		}
2089 		error = sockopt_set(sopt, &addr, sizeof(addr));
2090 		break;
2091 
2092 	case IP_MULTICAST_TTL:
2093 		optval = imo ? imo->imo_multicast_ttl
2094 		    : IP_DEFAULT_MULTICAST_TTL;
2095 
2096 		error = sockopt_set(sopt, &optval, sizeof(optval));
2097 		break;
2098 
2099 	case IP_MULTICAST_LOOP:
2100 		optval = imo ? imo->imo_multicast_loop
2101 		    : IP_DEFAULT_MULTICAST_LOOP;
2102 
2103 		error = sockopt_set(sopt, &optval, sizeof(optval));
2104 		break;
2105 
2106 	default:
2107 		error = EOPNOTSUPP;
2108 	}
2109 
2110 	return error;
2111 }
2112 
2113 /*
2114  * Discard the IP multicast options.
2115  */
2116 void
2117 ip_freemoptions(struct ip_moptions *imo)
2118 {
2119 	int i;
2120 
2121 	/* The owner of imo (inp) should be protected by solock */
2122 
2123 	if (imo != NULL) {
2124 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2125 			struct in_multi *inm = imo->imo_membership[i];
2126 			struct ifnet *ifp = inm->inm_ifp;
2127 			IFNET_LOCK(ifp);
2128 			in_delmulti(inm);
2129 			/* ifp should not leave thanks to solock */
2130 			IFNET_UNLOCK(ifp);
2131 		}
2132 
2133 		kmem_intr_free(imo, sizeof(*imo));
2134 	}
2135 }
2136 
2137 /*
2138  * Routine called from ip_output() to loop back a copy of an IP multicast
2139  * packet to the input queue of a specified interface.  Note that this
2140  * calls the output routine of the loopback "driver", but with an interface
2141  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
2142  */
2143 static void
2144 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
2145 {
2146 	struct ip *ip;
2147 	struct mbuf *copym;
2148 
2149 	copym = m_copypacket(m, M_DONTWAIT);
2150 	if (copym != NULL &&
2151 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
2152 		copym = m_pullup(copym, sizeof(struct ip));
2153 	if (copym == NULL)
2154 		return;
2155 	/*
2156 	 * We don't bother to fragment if the IP length is greater
2157 	 * than the interface's MTU.  Can this possibly matter?
2158 	 */
2159 	ip = mtod(copym, struct ip *);
2160 
2161 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
2162 		in_delayed_cksum(copym);
2163 		copym->m_pkthdr.csum_flags &=
2164 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
2165 	}
2166 
2167 	ip->ip_sum = 0;
2168 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
2169 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2170 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
2171 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2172 }
2173 
2174 /*
2175  * Ensure sending address is valid.
2176  * Returns 0 on success, -1 if an error should be sent back or 1
2177  * if the packet could be dropped without error (protocol dependent).
2178  */
2179 static int
2180 ip_ifaddrvalid(const struct in_ifaddr *ia)
2181 {
2182 
2183 	if (ia->ia_addr.sin_addr.s_addr == INADDR_ANY)
2184 		return 0;
2185 
2186 	if (ia->ia4_flags & IN_IFF_DUPLICATED)
2187 		return -1;
2188 	else if (ia->ia4_flags & (IN_IFF_TENTATIVE | IN_IFF_DETACHED))
2189 		return 1;
2190 
2191 	return 0;
2192 }
2193