xref: /netbsd-src/sys/netinet/ip_output.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: ip_output.c,v 1.247 2015/09/02 11:35:11 ozaki-r 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 1.247 2015/09/02 11:35:11 ozaki-r 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 <sys/param.h>
105 #include <sys/kmem.h>
106 #include <sys/mbuf.h>
107 #include <sys/protosw.h>
108 #include <sys/socket.h>
109 #include <sys/socketvar.h>
110 #include <sys/kauth.h>
111 #ifdef IPSEC
112 #include <sys/domain.h>
113 #endif
114 #include <sys/systm.h>
115 
116 #include <net/if.h>
117 #include <net/if_types.h>
118 #include <net/route.h>
119 #include <net/pfil.h>
120 
121 #include <netinet/in.h>
122 #include <netinet/in_systm.h>
123 #include <netinet/ip.h>
124 #include <netinet/in_pcb.h>
125 #include <netinet/in_var.h>
126 #include <netinet/ip_var.h>
127 #include <netinet/ip_private.h>
128 #include <netinet/in_offload.h>
129 #include <netinet/portalgo.h>
130 #include <netinet/udp.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 
156 extern pfil_head_t *inet_pfil_hook;			/* XXX */
157 
158 int	ip_do_loopback_cksum = 0;
159 
160 static bool
161 ip_hresolv_needed(const struct ifnet * const ifp)
162 {
163 	switch (ifp->if_type) {
164 	case IFT_ARCNET:
165 	case IFT_ATM:
166 	case IFT_ECONET:
167 	case IFT_ETHER:
168 	case IFT_FDDI:
169 	case IFT_HIPPI:
170 	case IFT_IEEE1394:
171 	case IFT_ISO88025:
172 	case IFT_SLIP:
173 		return true;
174 	default:
175 		return false;
176 	}
177 }
178 
179 static int
180 klock_if_output(struct ifnet * const ifp, struct mbuf * const m,
181     const struct sockaddr * const dst, struct rtentry *rt)
182 {
183 	int error;
184 
185 #ifndef NET_MPSAFE
186 	KERNEL_LOCK(1, NULL);
187 #endif
188 
189 	error = (*ifp->if_output)(ifp, m, dst, rt);
190 
191 #ifndef NET_MPSAFE
192 	KERNEL_UNLOCK_ONE(NULL);
193 #endif
194 
195 	return error;
196 }
197 
198 /*
199  * Send an IP packet to a host.
200  *
201  * If necessary, resolve the arbitrary IP route, rt0, to an IP host route before
202  * calling ifp's output routine.
203  */
204 int
205 ip_hresolv_output(struct ifnet * const ifp0, struct mbuf * const m,
206     const struct sockaddr * const dst, struct rtentry *rt00)
207 {
208 	int error = 0;
209 	struct ifnet *ifp = ifp0;
210 	struct rtentry *rt, *rt0, *gwrt;
211 
212 #define RTFREE_IF_NEEDED(_rt) \
213 	if ((_rt) != NULL && (_rt) != rt00) \
214 		rtfree((_rt));
215 
216 	rt0 = rt00;
217 retry:
218 	if (!ip_hresolv_needed(ifp)) {
219 		rt = rt0;
220 		goto out;
221 	}
222 
223 	if (rt0 == NULL) {
224 		rt = NULL;
225 		goto out;
226 	}
227 
228 	rt = rt0;
229 
230 	/*
231 	 * The following block is highly questionable.  How did we get here
232 	 * with a !RTF_UP route?  Does rtalloc1() always return an RTF_UP
233 	 * route?
234 	 */
235 	if ((rt->rt_flags & RTF_UP) == 0) {
236 		rt = rtalloc1(dst, 1);
237 		if (rt == NULL) {
238 			error = EHOSTUNREACH;
239 			goto bad;
240 		}
241 		rt0 = rt;
242 		if (rt->rt_ifp != ifp) {
243 			ifp = rt->rt_ifp;
244 			goto retry;
245 		}
246 	}
247 
248 	if ((rt->rt_flags & RTF_GATEWAY) == 0)
249 		goto out;
250 
251 	gwrt = rt_get_gwroute(rt);
252 	RTFREE_IF_NEEDED(rt);
253 	rt = gwrt;
254 	if (rt == NULL || (rt->rt_flags & RTF_UP) == 0) {
255 		if (rt != NULL) {
256 			RTFREE_IF_NEEDED(rt);
257 			rt = rt0;
258 		}
259 		if (rt == NULL) {
260 			error = EHOSTUNREACH;
261 			goto bad;
262 		}
263 		gwrt = rtalloc1(rt->rt_gateway, 1);
264 		rt_set_gwroute(rt, gwrt);
265 		RTFREE_IF_NEEDED(rt);
266 		rt = gwrt;
267 		if (rt == NULL) {
268 			error = EHOSTUNREACH;
269 			goto bad;
270 		}
271 		/* the "G" test below also prevents rt == rt0 */
272 		if ((rt->rt_flags & RTF_GATEWAY) != 0 || rt->rt_ifp != ifp) {
273 			if (rt0->rt_gwroute != NULL)
274 				rtfree(rt0->rt_gwroute);
275 			rt0->rt_gwroute = NULL;
276 			error = EHOSTUNREACH;
277 			goto bad;
278 		}
279 	}
280 	if ((rt->rt_flags & RTF_REJECT) != 0) {
281 		if (rt->rt_rmx.rmx_expire == 0 ||
282 		    time_uptime < rt->rt_rmx.rmx_expire) {
283 			error = (rt == rt0) ? EHOSTDOWN : EHOSTUNREACH;
284 			goto bad;
285 		}
286 	}
287 
288 out:
289 #ifdef MPLS
290 	if (rt0 != NULL && rt_gettag(rt0) != NULL &&
291 	    rt_gettag(rt0)->sa_family == AF_MPLS &&
292 	    (m->m_flags & (M_MCAST | M_BCAST)) == 0 &&
293 	    ifp->if_type == IFT_ETHER) {
294 		union mpls_shim msh;
295 		msh.s_addr = MPLS_GETSADDR(rt0);
296 		if (msh.shim.label != MPLS_LABEL_IMPLNULL) {
297 			struct m_tag *mtag;
298 			/*
299 			 * XXX tentative solution to tell ether_output
300 			 * it's MPLS. Need some more efficient solution.
301 			 */
302 			mtag = m_tag_get(PACKET_TAG_MPLS,
303 			    sizeof(int) /* dummy */,
304 			    M_NOWAIT);
305 			if (mtag == NULL) {
306 				error = ENOMEM;
307 				goto bad;
308 			}
309 			m_tag_prepend(m, mtag);
310 		}
311 	}
312 #endif
313 
314 	error = klock_if_output(ifp, m, dst, rt);
315 	goto exit;
316 
317 bad:
318 	if (m != NULL)
319 		m_freem(m);
320 exit:
321 	RTFREE_IF_NEEDED(rt);
322 
323 	return error;
324 
325 #undef RTFREE_IF_NEEDED
326 }
327 
328 /*
329  * IP output.  The packet in mbuf chain m contains a skeletal IP
330  * header (with len, off, ttl, proto, tos, src, dst).
331  * The mbuf chain containing the packet will be freed.
332  * The mbuf opt, if present, will not be freed.
333  */
334 int
335 ip_output(struct mbuf *m0, ...)
336 {
337 	struct rtentry *rt;
338 	struct ip *ip;
339 	struct ifnet *ifp;
340 	struct mbuf *m = m0;
341 	int hlen = sizeof (struct ip);
342 	int len, error = 0;
343 	struct route iproute;
344 	const struct sockaddr_in *dst;
345 	struct in_ifaddr *ia;
346 	int isbroadcast;
347 	struct mbuf *opt;
348 	struct route *ro;
349 	int flags, sw_csum;
350 	u_long mtu;
351 	struct ip_moptions *imo;
352 	struct socket *so;
353 	va_list ap;
354 #ifdef IPSEC
355 	struct secpolicy *sp = NULL;
356 #endif
357 	bool natt_frag = false;
358 	bool rtmtu_nolock;
359 	union {
360 		struct sockaddr		dst;
361 		struct sockaddr_in	dst4;
362 	} u;
363 	struct sockaddr *rdst = &u.dst;	/* real IP destination, as opposed
364 					 * to the nexthop
365 					 */
366 
367 	len = 0;
368 	va_start(ap, m0);
369 	opt = va_arg(ap, struct mbuf *);
370 	ro = va_arg(ap, struct route *);
371 	flags = va_arg(ap, int);
372 	imo = va_arg(ap, struct ip_moptions *);
373 	so = va_arg(ap, struct socket *);
374 	va_end(ap);
375 
376 	MCLAIM(m, &ip_tx_mowner);
377 
378 	KASSERT((m->m_flags & M_PKTHDR) != 0);
379 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
380 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
381 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
382 
383 	if (opt) {
384 		m = ip_insertoptions(m, opt, &len);
385 		if (len >= sizeof(struct ip))
386 			hlen = len;
387 	}
388 	ip = mtod(m, struct ip *);
389 
390 	/*
391 	 * Fill in IP header.
392 	 */
393 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
394 		ip->ip_v = IPVERSION;
395 		ip->ip_off = htons(0);
396 		/* ip->ip_id filled in after we find out source ia */
397 		ip->ip_hl = hlen >> 2;
398 		IP_STATINC(IP_STAT_LOCALOUT);
399 	} else {
400 		hlen = ip->ip_hl << 2;
401 	}
402 
403 	/*
404 	 * Route packet.
405 	 */
406 	if (ro == NULL) {
407 		memset(&iproute, 0, sizeof(iproute));
408 		ro = &iproute;
409 	}
410 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
411 	dst = satocsin(rtcache_getdst(ro));
412 
413 	/*
414 	 * If there is a cached route, check that it is to the same
415 	 * destination and is still up.  If not, free it and try again.
416 	 * The address family should also be checked in case of sharing
417 	 * the cache with IPv6.
418 	 */
419 	if (dst && (dst->sin_family != AF_INET ||
420 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
421 		rtcache_free(ro);
422 
423 	if ((rt = rtcache_validate(ro)) == NULL &&
424 	    (rt = rtcache_update(ro, 1)) == NULL) {
425 		dst = &u.dst4;
426 		error = rtcache_setdst(ro, &u.dst);
427 		if (error != 0)
428 			goto bad;
429 	}
430 
431 	/*
432 	 * If routing to interface only, short circuit routing lookup.
433 	 */
434 	if (flags & IP_ROUTETOIF) {
435 		if ((ia = ifatoia(ifa_ifwithladdr(sintocsa(dst)))) == NULL) {
436 			IP_STATINC(IP_STAT_NOROUTE);
437 			error = ENETUNREACH;
438 			goto bad;
439 		}
440 		ifp = ia->ia_ifp;
441 		mtu = ifp->if_mtu;
442 		ip->ip_ttl = 1;
443 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
444 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
445 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
446 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
447 		ifp = imo->imo_multicast_ifp;
448 		mtu = ifp->if_mtu;
449 		IFP_TO_IA(ifp, ia);
450 		isbroadcast = 0;
451 	} else {
452 		if (rt == NULL)
453 			rt = rtcache_init(ro);
454 		if (rt == NULL) {
455 			IP_STATINC(IP_STAT_NOROUTE);
456 			error = EHOSTUNREACH;
457 			goto bad;
458 		}
459 		ia = ifatoia(rt->rt_ifa);
460 		ifp = rt->rt_ifp;
461 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
462 			mtu = ifp->if_mtu;
463 		rt->rt_use++;
464 		if (rt->rt_flags & RTF_GATEWAY)
465 			dst = satosin(rt->rt_gateway);
466 		if (rt->rt_flags & RTF_HOST)
467 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
468 		else
469 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
470 	}
471 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
472 
473 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
474 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
475 		bool inmgroup;
476 
477 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
478 		    M_BCAST : M_MCAST;
479 		/*
480 		 * See if the caller provided any multicast options
481 		 */
482 		if (imo != NULL)
483 			ip->ip_ttl = imo->imo_multicast_ttl;
484 		else
485 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
486 
487 		/*
488 		 * if we don't know the outgoing ifp yet, we can't generate
489 		 * output
490 		 */
491 		if (!ifp) {
492 			IP_STATINC(IP_STAT_NOROUTE);
493 			error = ENETUNREACH;
494 			goto bad;
495 		}
496 
497 		/*
498 		 * If the packet is multicast or broadcast, confirm that
499 		 * the outgoing interface can transmit it.
500 		 */
501 		if (((m->m_flags & M_MCAST) &&
502 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
503 		    ((m->m_flags & M_BCAST) &&
504 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
505 			IP_STATINC(IP_STAT_NOROUTE);
506 			error = ENETUNREACH;
507 			goto bad;
508 		}
509 		/*
510 		 * If source address not specified yet, use an address
511 		 * of outgoing interface.
512 		 */
513 		if (in_nullhost(ip->ip_src)) {
514 			struct in_ifaddr *xia;
515 			struct ifaddr *xifa;
516 
517 			IFP_TO_IA(ifp, xia);
518 			if (!xia) {
519 				error = EADDRNOTAVAIL;
520 				goto bad;
521 			}
522 			xifa = &xia->ia_ifa;
523 			if (xifa->ifa_getifa != NULL) {
524 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
525 				if (xia == NULL) {
526 					error = EADDRNOTAVAIL;
527 					goto bad;
528 				}
529 			}
530 			ip->ip_src = xia->ia_addr.sin_addr;
531 		}
532 
533 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
534 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
535 			/*
536 			 * If we belong to the destination multicast group
537 			 * on the outgoing interface, and the caller did not
538 			 * forbid loopback, loop back a copy.
539 			 */
540 			ip_mloopback(ifp, m, &u.dst4);
541 		}
542 #ifdef MROUTING
543 		else {
544 			/*
545 			 * If we are acting as a multicast router, perform
546 			 * multicast forwarding as if the packet had just
547 			 * arrived on the interface to which we are about
548 			 * to send.  The multicast forwarding function
549 			 * recursively calls this function, using the
550 			 * IP_FORWARDING flag to prevent infinite recursion.
551 			 *
552 			 * Multicasts that are looped back by ip_mloopback(),
553 			 * above, will be forwarded by the ip_input() routine,
554 			 * if necessary.
555 			 */
556 			extern struct socket *ip_mrouter;
557 
558 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
559 				if (ip_mforward(m, ifp) != 0) {
560 					m_freem(m);
561 					goto done;
562 				}
563 			}
564 		}
565 #endif
566 		/*
567 		 * Multicasts with a time-to-live of zero may be looped-
568 		 * back, above, but must not be transmitted on a network.
569 		 * Also, multicasts addressed to the loopback interface
570 		 * are not sent -- the above call to ip_mloopback() will
571 		 * loop back a copy if this host actually belongs to the
572 		 * destination group on the loopback interface.
573 		 */
574 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
575 			m_freem(m);
576 			goto done;
577 		}
578 		goto sendit;
579 	}
580 
581 	/*
582 	 * If source address not specified yet, use address
583 	 * of outgoing interface.
584 	 */
585 	if (in_nullhost(ip->ip_src)) {
586 		struct ifaddr *xifa;
587 
588 		xifa = &ia->ia_ifa;
589 		if (xifa->ifa_getifa != NULL) {
590 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
591 			if (ia == NULL) {
592 				error = EADDRNOTAVAIL;
593 				goto bad;
594 			}
595 		}
596 		ip->ip_src = ia->ia_addr.sin_addr;
597 	}
598 
599 	/*
600 	 * packets with Class-D address as source are not valid per
601 	 * RFC 1112
602 	 */
603 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
604 		IP_STATINC(IP_STAT_ODROPPED);
605 		error = EADDRNOTAVAIL;
606 		goto bad;
607 	}
608 
609 	/*
610 	 * Look for broadcast address and and verify user is allowed to
611 	 * send such a packet.
612 	 */
613 	if (isbroadcast) {
614 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
615 			error = EADDRNOTAVAIL;
616 			goto bad;
617 		}
618 		if ((flags & IP_ALLOWBROADCAST) == 0) {
619 			error = EACCES;
620 			goto bad;
621 		}
622 		/* don't allow broadcast messages to be fragmented */
623 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
624 			error = EMSGSIZE;
625 			goto bad;
626 		}
627 		m->m_flags |= M_BCAST;
628 	} else
629 		m->m_flags &= ~M_BCAST;
630 
631 sendit:
632 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
633 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
634 			ip->ip_id = 0;
635 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
636 			ip->ip_id = ip_newid(ia);
637 		} else {
638 
639 			/*
640 			 * TSO capable interfaces (typically?) increment
641 			 * ip_id for each segment.
642 			 * "allocate" enough ids here to increase the chance
643 			 * for them to be unique.
644 			 *
645 			 * note that the following calculation is not
646 			 * needed to be precise.  wasting some ip_id is fine.
647 			 */
648 
649 			unsigned int segsz = m->m_pkthdr.segsz;
650 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
651 			unsigned int num = howmany(datasz, segsz);
652 
653 			ip->ip_id = ip_newid_range(ia, num);
654 		}
655 	}
656 
657 	/*
658 	 * If we're doing Path MTU Discovery, we need to set DF unless
659 	 * the route's MTU is locked.
660 	 */
661 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
662 		ip->ip_off |= htons(IP_DF);
663 	}
664 
665 #ifdef IPSEC
666 	if (ipsec_used) {
667 		bool ipsec_done = false;
668 
669 		/* Perform IPsec processing, if any. */
670 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
671 		    &ipsec_done);
672 		if (error || ipsec_done)
673 			goto done;
674 	}
675 #endif
676 
677 	/*
678 	 * Run through list of hooks for output packets.
679 	 */
680 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
681 	if (error)
682 		goto done;
683 	if (m == NULL)
684 		goto done;
685 
686 	ip = mtod(m, struct ip *);
687 	hlen = ip->ip_hl << 2;
688 
689 	m->m_pkthdr.csum_data |= hlen << 16;
690 
691 #if IFA_STATS
692 	/*
693 	 * search for the source address structure to
694 	 * maintain output statistics.
695 	 */
696 	INADDR_TO_IA(ip->ip_src, ia);
697 #endif
698 
699 	/* Maybe skip checksums on loopback interfaces. */
700 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
701 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
702 	}
703 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
704 	/*
705 	 * If small enough for mtu of path, or if using TCP segmentation
706 	 * offload, can just send directly.
707 	 */
708 	if (ntohs(ip->ip_len) <= mtu ||
709 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
710 		const struct sockaddr *sa;
711 
712 #if IFA_STATS
713 		if (ia)
714 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
715 #endif
716 		/*
717 		 * Always initialize the sum to 0!  Some HW assisted
718 		 * checksumming requires this.
719 		 */
720 		ip->ip_sum = 0;
721 
722 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
723 			/*
724 			 * Perform any checksums that the hardware can't do
725 			 * for us.
726 			 *
727 			 * XXX Does any hardware require the {th,uh}_sum
728 			 * XXX fields to be 0?
729 			 */
730 			if (sw_csum & M_CSUM_IPv4) {
731 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
732 				ip->ip_sum = in_cksum(m, hlen);
733 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
734 			}
735 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
736 				if (IN_NEED_CHECKSUM(ifp,
737 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
738 					in_delayed_cksum(m);
739 				}
740 				m->m_pkthdr.csum_flags &=
741 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
742 			}
743 		}
744 
745 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
746 		if (__predict_true(
747 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
748 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
749 			error = ip_hresolv_output(ifp, m, sa, rt);
750 		} else {
751 			error = ip_tso_output(ifp, m, sa, rt);
752 		}
753 		goto done;
754 	}
755 
756 	/*
757 	 * We can't use HW checksumming if we're about to
758 	 * to fragment the packet.
759 	 *
760 	 * XXX Some hardware can do this.
761 	 */
762 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
763 		if (IN_NEED_CHECKSUM(ifp,
764 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
765 			in_delayed_cksum(m);
766 		}
767 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
768 	}
769 
770 	/*
771 	 * Too large for interface; fragment if possible.
772 	 * Must be able to put at least 8 bytes per fragment.
773 	 */
774 	if (ntohs(ip->ip_off) & IP_DF) {
775 		if (flags & IP_RETURNMTU) {
776 			struct inpcb *inp;
777 
778 			KASSERT(so && solocked(so));
779 			inp = sotoinpcb(so);
780 			inp->inp_errormtu = mtu;
781 		}
782 		error = EMSGSIZE;
783 		IP_STATINC(IP_STAT_CANTFRAG);
784 		goto bad;
785 	}
786 
787 	error = ip_fragment(m, ifp, mtu);
788 	if (error) {
789 		m = NULL;
790 		goto bad;
791 	}
792 
793 	for (; m; m = m0) {
794 		m0 = m->m_nextpkt;
795 		m->m_nextpkt = 0;
796 		if (error) {
797 			m_freem(m);
798 			continue;
799 		}
800 #if IFA_STATS
801 		if (ia)
802 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
803 #endif
804 		/*
805 		 * If we get there, the packet has not been handled by
806 		 * IPsec whereas it should have. Now that it has been
807 		 * fragmented, re-inject it in ip_output so that IPsec
808 		 * processing can occur.
809 		 */
810 		if (natt_frag) {
811 			error = ip_output(m, opt, ro,
812 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
813 			    imo, so);
814 		} else {
815 			KASSERT((m->m_pkthdr.csum_flags &
816 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
817 			error = ip_hresolv_output(ifp, m,
818 			    (m->m_flags & M_MCAST) ?
819 			    sintocsa(rdst) : sintocsa(dst), rt);
820 		}
821 	}
822 	if (error == 0) {
823 		IP_STATINC(IP_STAT_FRAGMENTED);
824 	}
825 done:
826 	if (ro == &iproute) {
827 		rtcache_free(&iproute);
828 	}
829 #ifdef IPSEC
830 	if (sp) {
831 		KEY_FREESP(&sp);
832 	}
833 #endif
834 	return error;
835 bad:
836 	m_freem(m);
837 	goto done;
838 }
839 
840 int
841 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
842 {
843 	struct ip *ip, *mhip;
844 	struct mbuf *m0;
845 	int len, hlen, off;
846 	int mhlen, firstlen;
847 	struct mbuf **mnext;
848 	int sw_csum = m->m_pkthdr.csum_flags;
849 	int fragments = 0;
850 	int s;
851 	int error = 0;
852 
853 	ip = mtod(m, struct ip *);
854 	hlen = ip->ip_hl << 2;
855 	if (ifp != NULL)
856 		sw_csum &= ~ifp->if_csum_flags_tx;
857 
858 	len = (mtu - hlen) &~ 7;
859 	if (len < 8) {
860 		m_freem(m);
861 		return (EMSGSIZE);
862 	}
863 
864 	firstlen = len;
865 	mnext = &m->m_nextpkt;
866 
867 	/*
868 	 * Loop through length of segment after first fragment,
869 	 * make new header and copy data of each part and link onto chain.
870 	 */
871 	m0 = m;
872 	mhlen = sizeof (struct ip);
873 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
874 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
875 		if (m == 0) {
876 			error = ENOBUFS;
877 			IP_STATINC(IP_STAT_ODROPPED);
878 			goto sendorfree;
879 		}
880 		MCLAIM(m, m0->m_owner);
881 		*mnext = m;
882 		mnext = &m->m_nextpkt;
883 		m->m_data += max_linkhdr;
884 		mhip = mtod(m, struct ip *);
885 		*mhip = *ip;
886 		/* we must inherit MCAST and BCAST flags */
887 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
888 		if (hlen > sizeof (struct ip)) {
889 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
890 			mhip->ip_hl = mhlen >> 2;
891 		}
892 		m->m_len = mhlen;
893 		mhip->ip_off = ((off - hlen) >> 3) +
894 		    (ntohs(ip->ip_off) & ~IP_MF);
895 		if (ip->ip_off & htons(IP_MF))
896 			mhip->ip_off |= IP_MF;
897 		if (off + len >= ntohs(ip->ip_len))
898 			len = ntohs(ip->ip_len) - off;
899 		else
900 			mhip->ip_off |= IP_MF;
901 		HTONS(mhip->ip_off);
902 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
903 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
904 		if (m->m_next == 0) {
905 			error = ENOBUFS;	/* ??? */
906 			IP_STATINC(IP_STAT_ODROPPED);
907 			goto sendorfree;
908 		}
909 		m->m_pkthdr.len = mhlen + len;
910 		m->m_pkthdr.rcvif = NULL;
911 		mhip->ip_sum = 0;
912 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
913 		if (sw_csum & M_CSUM_IPv4) {
914 			mhip->ip_sum = in_cksum(m, mhlen);
915 		} else {
916 			/*
917 			 * checksum is hw-offloaded or not necessary.
918 			 */
919 			m->m_pkthdr.csum_flags |=
920 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
921 			m->m_pkthdr.csum_data |= mhlen << 16;
922 			KASSERT(!(ifp != NULL &&
923 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
924 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
925 		}
926 		IP_STATINC(IP_STAT_OFRAGMENTS);
927 		fragments++;
928 	}
929 	/*
930 	 * Update first fragment by trimming what's been copied out
931 	 * and updating header, then send each fragment (in order).
932 	 */
933 	m = m0;
934 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
935 	m->m_pkthdr.len = hlen + firstlen;
936 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
937 	ip->ip_off |= htons(IP_MF);
938 	ip->ip_sum = 0;
939 	if (sw_csum & M_CSUM_IPv4) {
940 		ip->ip_sum = in_cksum(m, hlen);
941 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
942 	} else {
943 		/*
944 		 * checksum is hw-offloaded or not necessary.
945 		 */
946 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
947 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
948 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
949 		    sizeof(struct ip));
950 	}
951 sendorfree:
952 	/*
953 	 * If there is no room for all the fragments, don't queue
954 	 * any of them.
955 	 */
956 	if (ifp != NULL) {
957 		s = splnet();
958 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
959 		    error == 0) {
960 			error = ENOBUFS;
961 			IP_STATINC(IP_STAT_ODROPPED);
962 			IFQ_INC_DROPS(&ifp->if_snd);
963 		}
964 		splx(s);
965 	}
966 	if (error) {
967 		for (m = m0; m; m = m0) {
968 			m0 = m->m_nextpkt;
969 			m->m_nextpkt = NULL;
970 			m_freem(m);
971 		}
972 	}
973 	return (error);
974 }
975 
976 /*
977  * Process a delayed payload checksum calculation.
978  */
979 void
980 in_delayed_cksum(struct mbuf *m)
981 {
982 	struct ip *ip;
983 	u_int16_t csum, offset;
984 
985 	ip = mtod(m, struct ip *);
986 	offset = ip->ip_hl << 2;
987 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
988 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
989 		csum = 0xffff;
990 
991 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
992 
993 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
994 		/* This happen when ip options were inserted
995 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
996 		    m->m_len, offset, ip->ip_p);
997 		 */
998 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
999 	} else
1000 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
1001 }
1002 
1003 /*
1004  * Determine the maximum length of the options to be inserted;
1005  * we would far rather allocate too much space rather than too little.
1006  */
1007 
1008 u_int
1009 ip_optlen(struct inpcb *inp)
1010 {
1011 	struct mbuf *m = inp->inp_options;
1012 
1013 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
1014 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
1015 	}
1016 	return 0;
1017 }
1018 
1019 /*
1020  * Insert IP options into preformed packet.
1021  * Adjust IP destination as required for IP source routing,
1022  * as indicated by a non-zero in_addr at the start of the options.
1023  */
1024 static struct mbuf *
1025 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1026 {
1027 	struct ipoption *p = mtod(opt, struct ipoption *);
1028 	struct mbuf *n;
1029 	struct ip *ip = mtod(m, struct ip *);
1030 	unsigned optlen;
1031 
1032 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1033 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
1034 		return (m);		/* XXX should fail */
1035 	if (!in_nullhost(p->ipopt_dst))
1036 		ip->ip_dst = p->ipopt_dst;
1037 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
1038 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1039 		if (n == 0)
1040 			return (m);
1041 		MCLAIM(n, m->m_owner);
1042 		M_MOVE_PKTHDR(n, m);
1043 		m->m_len -= sizeof(struct ip);
1044 		m->m_data += sizeof(struct ip);
1045 		n->m_next = m;
1046 		m = n;
1047 		m->m_len = optlen + sizeof(struct ip);
1048 		m->m_data += max_linkhdr;
1049 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
1050 	} else {
1051 		m->m_data -= optlen;
1052 		m->m_len += optlen;
1053 		memmove(mtod(m, void *), ip, sizeof(struct ip));
1054 	}
1055 	m->m_pkthdr.len += optlen;
1056 	ip = mtod(m, struct ip *);
1057 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
1058 	*phlen = sizeof(struct ip) + optlen;
1059 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1060 	return (m);
1061 }
1062 
1063 /*
1064  * Copy options from ip to jp,
1065  * omitting those not copied during fragmentation.
1066  */
1067 int
1068 ip_optcopy(struct ip *ip, struct ip *jp)
1069 {
1070 	u_char *cp, *dp;
1071 	int opt, optlen, cnt;
1072 
1073 	cp = (u_char *)(ip + 1);
1074 	dp = (u_char *)(jp + 1);
1075 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1076 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1077 		opt = cp[0];
1078 		if (opt == IPOPT_EOL)
1079 			break;
1080 		if (opt == IPOPT_NOP) {
1081 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1082 			*dp++ = IPOPT_NOP;
1083 			optlen = 1;
1084 			continue;
1085 		}
1086 
1087 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
1088 		optlen = cp[IPOPT_OLEN];
1089 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
1090 
1091 		/* Invalid lengths should have been caught by ip_dooptions. */
1092 		if (optlen > cnt)
1093 			optlen = cnt;
1094 		if (IPOPT_COPIED(opt)) {
1095 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1096 			dp += optlen;
1097 		}
1098 	}
1099 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1100 		*dp++ = IPOPT_EOL;
1101 	return (optlen);
1102 }
1103 
1104 /*
1105  * IP socket option processing.
1106  */
1107 int
1108 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1109 {
1110 	struct inpcb *inp = sotoinpcb(so);
1111 	struct ip *ip = &inp->inp_ip;
1112 	int inpflags = inp->inp_flags;
1113 	int optval = 0, error = 0;
1114 
1115 	if (sopt->sopt_level != IPPROTO_IP) {
1116 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1117 			return 0;
1118 		return ENOPROTOOPT;
1119 	}
1120 
1121 	switch (op) {
1122 	case PRCO_SETOPT:
1123 		switch (sopt->sopt_name) {
1124 		case IP_OPTIONS:
1125 #ifdef notyet
1126 		case IP_RETOPTS:
1127 #endif
1128 			error = ip_pcbopts(inp, sopt);
1129 			break;
1130 
1131 		case IP_TOS:
1132 		case IP_TTL:
1133 		case IP_MINTTL:
1134 		case IP_PKTINFO:
1135 		case IP_RECVOPTS:
1136 		case IP_RECVRETOPTS:
1137 		case IP_RECVDSTADDR:
1138 		case IP_RECVIF:
1139 		case IP_RECVPKTINFO:
1140 		case IP_RECVTTL:
1141 			error = sockopt_getint(sopt, &optval);
1142 			if (error)
1143 				break;
1144 
1145 			switch (sopt->sopt_name) {
1146 			case IP_TOS:
1147 				ip->ip_tos = optval;
1148 				break;
1149 
1150 			case IP_TTL:
1151 				ip->ip_ttl = optval;
1152 				break;
1153 
1154 			case IP_MINTTL:
1155 				if (optval > 0 && optval <= MAXTTL)
1156 					inp->inp_ip_minttl = optval;
1157 				else
1158 					error = EINVAL;
1159 				break;
1160 #define	OPTSET(bit) \
1161 	if (optval) \
1162 		inpflags |= bit; \
1163 	else \
1164 		inpflags &= ~bit;
1165 
1166 			case IP_PKTINFO:
1167 				OPTSET(INP_PKTINFO);
1168 				break;
1169 
1170 			case IP_RECVOPTS:
1171 				OPTSET(INP_RECVOPTS);
1172 				break;
1173 
1174 			case IP_RECVPKTINFO:
1175 				OPTSET(INP_RECVPKTINFO);
1176 				break;
1177 
1178 			case IP_RECVRETOPTS:
1179 				OPTSET(INP_RECVRETOPTS);
1180 				break;
1181 
1182 			case IP_RECVDSTADDR:
1183 				OPTSET(INP_RECVDSTADDR);
1184 				break;
1185 
1186 			case IP_RECVIF:
1187 				OPTSET(INP_RECVIF);
1188 				break;
1189 
1190 			case IP_RECVTTL:
1191 				OPTSET(INP_RECVTTL);
1192 				break;
1193 			}
1194 		break;
1195 #undef OPTSET
1196 
1197 		case IP_MULTICAST_IF:
1198 		case IP_MULTICAST_TTL:
1199 		case IP_MULTICAST_LOOP:
1200 		case IP_ADD_MEMBERSHIP:
1201 		case IP_DROP_MEMBERSHIP:
1202 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1203 			break;
1204 
1205 		case IP_PORTRANGE:
1206 			error = sockopt_getint(sopt, &optval);
1207 			if (error)
1208 				break;
1209 
1210 			switch (optval) {
1211 			case IP_PORTRANGE_DEFAULT:
1212 			case IP_PORTRANGE_HIGH:
1213 				inpflags &= ~(INP_LOWPORT);
1214 				break;
1215 
1216 			case IP_PORTRANGE_LOW:
1217 				inpflags |= INP_LOWPORT;
1218 				break;
1219 
1220 			default:
1221 				error = EINVAL;
1222 				break;
1223 			}
1224 			break;
1225 
1226 		case IP_PORTALGO:
1227 			error = sockopt_getint(sopt, &optval);
1228 			if (error)
1229 				break;
1230 
1231 			error = portalgo_algo_index_select(
1232 			    (struct inpcb_hdr *)inp, optval);
1233 			break;
1234 
1235 #if defined(IPSEC)
1236 		case IP_IPSEC_POLICY:
1237 			if (ipsec_enabled) {
1238 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1239 				    sopt->sopt_data, sopt->sopt_size,
1240 				    curlwp->l_cred);
1241 				break;
1242 			}
1243 			/*FALLTHROUGH*/
1244 #endif /* IPSEC */
1245 
1246 		default:
1247 			error = ENOPROTOOPT;
1248 			break;
1249 		}
1250 		break;
1251 
1252 	case PRCO_GETOPT:
1253 		switch (sopt->sopt_name) {
1254 		case IP_OPTIONS:
1255 		case IP_RETOPTS: {
1256 			struct mbuf *mopts = inp->inp_options;
1257 
1258 			if (mopts) {
1259 				struct mbuf *m;
1260 
1261 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1262 				if (m == NULL) {
1263 					error = ENOBUFS;
1264 					break;
1265 				}
1266 				error = sockopt_setmbuf(sopt, m);
1267 			}
1268 			break;
1269 		}
1270 		case IP_PKTINFO:
1271 		case IP_TOS:
1272 		case IP_TTL:
1273 		case IP_MINTTL:
1274 		case IP_RECVOPTS:
1275 		case IP_RECVRETOPTS:
1276 		case IP_RECVDSTADDR:
1277 		case IP_RECVIF:
1278 		case IP_RECVPKTINFO:
1279 		case IP_RECVTTL:
1280 		case IP_ERRORMTU:
1281 			switch (sopt->sopt_name) {
1282 			case IP_TOS:
1283 				optval = ip->ip_tos;
1284 				break;
1285 
1286 			case IP_TTL:
1287 				optval = ip->ip_ttl;
1288 				break;
1289 
1290 			case IP_MINTTL:
1291 				optval = inp->inp_ip_minttl;
1292 				break;
1293 
1294 			case IP_ERRORMTU:
1295 				optval = inp->inp_errormtu;
1296 				break;
1297 
1298 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1299 
1300 			case IP_PKTINFO:
1301 				optval = OPTBIT(INP_PKTINFO);
1302 				break;
1303 
1304 			case IP_RECVOPTS:
1305 				optval = OPTBIT(INP_RECVOPTS);
1306 				break;
1307 
1308 			case IP_RECVPKTINFO:
1309 				optval = OPTBIT(INP_RECVPKTINFO);
1310 				break;
1311 
1312 			case IP_RECVRETOPTS:
1313 				optval = OPTBIT(INP_RECVRETOPTS);
1314 				break;
1315 
1316 			case IP_RECVDSTADDR:
1317 				optval = OPTBIT(INP_RECVDSTADDR);
1318 				break;
1319 
1320 			case IP_RECVIF:
1321 				optval = OPTBIT(INP_RECVIF);
1322 				break;
1323 
1324 			case IP_RECVTTL:
1325 				optval = OPTBIT(INP_RECVTTL);
1326 				break;
1327 			}
1328 			error = sockopt_setint(sopt, optval);
1329 			break;
1330 
1331 #if 0	/* defined(IPSEC) */
1332 		case IP_IPSEC_POLICY:
1333 		{
1334 			struct mbuf *m = NULL;
1335 
1336 			/* XXX this will return EINVAL as sopt is empty */
1337 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1338 			    sopt->sopt_size, &m);
1339 			if (error == 0)
1340 				error = sockopt_setmbuf(sopt, m);
1341 			break;
1342 		}
1343 #endif /*IPSEC*/
1344 
1345 		case IP_MULTICAST_IF:
1346 		case IP_MULTICAST_TTL:
1347 		case IP_MULTICAST_LOOP:
1348 		case IP_ADD_MEMBERSHIP:
1349 		case IP_DROP_MEMBERSHIP:
1350 			error = ip_getmoptions(inp->inp_moptions, sopt);
1351 			break;
1352 
1353 		case IP_PORTRANGE:
1354 			if (inpflags & INP_LOWPORT)
1355 				optval = IP_PORTRANGE_LOW;
1356 			else
1357 				optval = IP_PORTRANGE_DEFAULT;
1358 			error = sockopt_setint(sopt, optval);
1359 			break;
1360 
1361 		case IP_PORTALGO:
1362 			optval = inp->inp_portalgo;
1363 			error = sockopt_setint(sopt, optval);
1364 			break;
1365 
1366 		default:
1367 			error = ENOPROTOOPT;
1368 			break;
1369 		}
1370 		break;
1371 	}
1372 
1373 	if (!error) {
1374 		inp->inp_flags = inpflags;
1375 	}
1376 	return error;
1377 }
1378 
1379 /*
1380  * Set up IP options in pcb for insertion in output packets.
1381  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1382  * with destination address if source routed.
1383  */
1384 static int
1385 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1386 {
1387 	struct mbuf *m;
1388 	const u_char *cp;
1389 	u_char *dp;
1390 	int cnt;
1391 
1392 	/* Turn off any old options. */
1393 	if (inp->inp_options) {
1394 		m_free(inp->inp_options);
1395 	}
1396 	inp->inp_options = NULL;
1397 	if ((cnt = sopt->sopt_size) == 0) {
1398 		/* Only turning off any previous options. */
1399 		return 0;
1400 	}
1401 	cp = sopt->sopt_data;
1402 
1403 #ifndef	__vax__
1404 	if (cnt % sizeof(int32_t))
1405 		return (EINVAL);
1406 #endif
1407 
1408 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1409 	if (m == NULL)
1410 		return (ENOBUFS);
1411 
1412 	dp = mtod(m, u_char *);
1413 	memset(dp, 0, sizeof(struct in_addr));
1414 	dp += sizeof(struct in_addr);
1415 	m->m_len = sizeof(struct in_addr);
1416 
1417 	/*
1418 	 * IP option list according to RFC791. Each option is of the form
1419 	 *
1420 	 *	[optval] [olen] [(olen - 2) data bytes]
1421 	 *
1422 	 * We validate the list and copy options to an mbuf for prepending
1423 	 * to data packets. The IP first-hop destination address will be
1424 	 * stored before actual options and is zero if unset.
1425 	 */
1426 	while (cnt > 0) {
1427 		uint8_t optval, olen, offset;
1428 
1429 		optval = cp[IPOPT_OPTVAL];
1430 
1431 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1432 			olen = 1;
1433 		} else {
1434 			if (cnt < IPOPT_OLEN + 1)
1435 				goto bad;
1436 
1437 			olen = cp[IPOPT_OLEN];
1438 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1439 				goto bad;
1440 		}
1441 
1442 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1443 			/*
1444 			 * user process specifies route as:
1445 			 *	->A->B->C->D
1446 			 * D must be our final destination (but we can't
1447 			 * check that since we may not have connected yet).
1448 			 * A is first hop destination, which doesn't appear in
1449 			 * actual IP option, but is stored before the options.
1450 			 */
1451 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1452 				goto bad;
1453 
1454 			offset = cp[IPOPT_OFFSET];
1455 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1456 			    sizeof(struct in_addr));
1457 
1458 			cp += sizeof(struct in_addr);
1459 			cnt -= sizeof(struct in_addr);
1460 			olen -= sizeof(struct in_addr);
1461 
1462 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1463 				goto bad;
1464 
1465 			memcpy(dp, cp, olen);
1466 			dp[IPOPT_OPTVAL] = optval;
1467 			dp[IPOPT_OLEN] = olen;
1468 			dp[IPOPT_OFFSET] = offset;
1469 			break;
1470 		} else {
1471 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1472 				goto bad;
1473 
1474 			memcpy(dp, cp, olen);
1475 			break;
1476 		}
1477 
1478 		dp += olen;
1479 		m->m_len += olen;
1480 
1481 		if (optval == IPOPT_EOL)
1482 			break;
1483 
1484 		cp += olen;
1485 		cnt -= olen;
1486 	}
1487 
1488 	inp->inp_options = m;
1489 	return 0;
1490 bad:
1491 	(void)m_free(m);
1492 	return EINVAL;
1493 }
1494 
1495 /*
1496  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1497  */
1498 static struct ifnet *
1499 ip_multicast_if(struct in_addr *a, int *ifindexp)
1500 {
1501 	int ifindex;
1502 	struct ifnet *ifp = NULL;
1503 	struct in_ifaddr *ia;
1504 
1505 	if (ifindexp)
1506 		*ifindexp = 0;
1507 	if (ntohl(a->s_addr) >> 24 == 0) {
1508 		ifindex = ntohl(a->s_addr) & 0xffffff;
1509 		ifp = if_byindex(ifindex);
1510 		if (!ifp)
1511 			return NULL;
1512 		if (ifindexp)
1513 			*ifindexp = ifindex;
1514 	} else {
1515 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1516 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1517 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1518 				ifp = ia->ia_ifp;
1519 				break;
1520 			}
1521 		}
1522 	}
1523 	return ifp;
1524 }
1525 
1526 static int
1527 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1528 {
1529 	u_int tval;
1530 	u_char cval;
1531 	int error;
1532 
1533 	if (sopt == NULL)
1534 		return EINVAL;
1535 
1536 	switch (sopt->sopt_size) {
1537 	case sizeof(u_char):
1538 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1539 		tval = cval;
1540 		break;
1541 
1542 	case sizeof(u_int):
1543 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1544 		break;
1545 
1546 	default:
1547 		error = EINVAL;
1548 	}
1549 
1550 	if (error)
1551 		return error;
1552 
1553 	if (tval > maxval)
1554 		return EINVAL;
1555 
1556 	*val = tval;
1557 	return 0;
1558 }
1559 
1560 static int
1561 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1562     struct in_addr *ia, bool add)
1563 {
1564 	int error;
1565 	struct ip_mreq mreq;
1566 
1567 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1568 	if (error)
1569 		return error;
1570 
1571 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1572 		return EINVAL;
1573 
1574 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1575 
1576 	if (in_nullhost(mreq.imr_interface)) {
1577 		union {
1578 			struct sockaddr		dst;
1579 			struct sockaddr_in	dst4;
1580 		} u;
1581 		struct route ro;
1582 
1583 		if (!add) {
1584 			*ifp = NULL;
1585 			return 0;
1586 		}
1587 		/*
1588 		 * If no interface address was provided, use the interface of
1589 		 * the route to the given multicast address.
1590 		 */
1591 		struct rtentry *rt;
1592 		memset(&ro, 0, sizeof(ro));
1593 
1594 		sockaddr_in_init(&u.dst4, ia, 0);
1595 		error = rtcache_setdst(&ro, &u.dst);
1596 		if (error != 0)
1597 			return error;
1598 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1599 		rtcache_free(&ro);
1600 	} else {
1601 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1602 		if (!add && *ifp == NULL)
1603 			return EADDRNOTAVAIL;
1604 	}
1605 	return 0;
1606 }
1607 
1608 /*
1609  * Add a multicast group membership.
1610  * Group must be a valid IP multicast address.
1611  */
1612 static int
1613 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1614 {
1615 	struct ifnet *ifp;
1616 	struct in_addr ia;
1617 	int i, error;
1618 
1619 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1620 		error = ip_get_membership(sopt, &ifp, &ia, true);
1621 	else
1622 #ifdef INET6
1623 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1624 #else
1625 		return EINVAL;
1626 #endif
1627 
1628 	if (error)
1629 		return error;
1630 
1631 	/*
1632 	 * See if we found an interface, and confirm that it
1633 	 * supports multicast.
1634 	 */
1635 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1636 		return EADDRNOTAVAIL;
1637 
1638 	/*
1639 	 * See if the membership already exists or if all the
1640 	 * membership slots are full.
1641 	 */
1642 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1643 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1644 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1645 			break;
1646 	}
1647 	if (i < imo->imo_num_memberships)
1648 		return EADDRINUSE;
1649 
1650 	if (i == IP_MAX_MEMBERSHIPS)
1651 		return ETOOMANYREFS;
1652 
1653 	/*
1654 	 * Everything looks good; add a new record to the multicast
1655 	 * address list for the given interface.
1656 	 */
1657 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1658 		return ENOBUFS;
1659 
1660 	++imo->imo_num_memberships;
1661 	return 0;
1662 }
1663 
1664 /*
1665  * Drop a multicast group membership.
1666  * Group must be a valid IP multicast address.
1667  */
1668 static int
1669 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1670 {
1671 	struct in_addr ia;
1672 	struct ifnet *ifp;
1673 	int i, error;
1674 
1675 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1676 		error = ip_get_membership(sopt, &ifp, &ia, false);
1677 	else
1678 #ifdef INET6
1679 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1680 #else
1681 		return EINVAL;
1682 #endif
1683 
1684 	if (error)
1685 		return error;
1686 
1687 	/*
1688 	 * Find the membership in the membership array.
1689 	 */
1690 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1691 		if ((ifp == NULL ||
1692 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1693 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1694 			break;
1695 	}
1696 	if (i == imo->imo_num_memberships)
1697 		return EADDRNOTAVAIL;
1698 
1699 	/*
1700 	 * Give up the multicast address record to which the
1701 	 * membership points.
1702 	 */
1703 	in_delmulti(imo->imo_membership[i]);
1704 
1705 	/*
1706 	 * Remove the gap in the membership array.
1707 	 */
1708 	for (++i; i < imo->imo_num_memberships; ++i)
1709 		imo->imo_membership[i-1] = imo->imo_membership[i];
1710 	--imo->imo_num_memberships;
1711 	return 0;
1712 }
1713 
1714 /*
1715  * Set the IP multicast options in response to user setsockopt().
1716  */
1717 int
1718 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1719 {
1720 	struct ip_moptions *imo = *pimo;
1721 	struct in_addr addr;
1722 	struct ifnet *ifp;
1723 	int ifindex, error = 0;
1724 
1725 	if (!imo) {
1726 		/*
1727 		 * No multicast option buffer attached to the pcb;
1728 		 * allocate one and initialize to default values.
1729 		 */
1730 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1731 		if (imo == NULL)
1732 			return ENOBUFS;
1733 
1734 		imo->imo_multicast_ifp = NULL;
1735 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1736 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1737 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1738 		imo->imo_num_memberships = 0;
1739 		*pimo = imo;
1740 	}
1741 
1742 	switch (sopt->sopt_name) {
1743 	case IP_MULTICAST_IF:
1744 		/*
1745 		 * Select the interface for outgoing multicast packets.
1746 		 */
1747 		error = sockopt_get(sopt, &addr, sizeof(addr));
1748 		if (error)
1749 			break;
1750 
1751 		/*
1752 		 * INADDR_ANY is used to remove a previous selection.
1753 		 * When no interface is selected, a default one is
1754 		 * chosen every time a multicast packet is sent.
1755 		 */
1756 		if (in_nullhost(addr)) {
1757 			imo->imo_multicast_ifp = NULL;
1758 			break;
1759 		}
1760 		/*
1761 		 * The selected interface is identified by its local
1762 		 * IP address.  Find the interface and confirm that
1763 		 * it supports multicasting.
1764 		 */
1765 		ifp = ip_multicast_if(&addr, &ifindex);
1766 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1767 			error = EADDRNOTAVAIL;
1768 			break;
1769 		}
1770 		imo->imo_multicast_ifp = ifp;
1771 		if (ifindex)
1772 			imo->imo_multicast_addr = addr;
1773 		else
1774 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1775 		break;
1776 
1777 	case IP_MULTICAST_TTL:
1778 		/*
1779 		 * Set the IP time-to-live for outgoing multicast packets.
1780 		 */
1781 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1782 		break;
1783 
1784 	case IP_MULTICAST_LOOP:
1785 		/*
1786 		 * Set the loopback flag for outgoing multicast packets.
1787 		 * Must be zero or one.
1788 		 */
1789 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1790 		break;
1791 
1792 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1793 		error = ip_add_membership(imo, sopt);
1794 		break;
1795 
1796 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1797 		error = ip_drop_membership(imo, sopt);
1798 		break;
1799 
1800 	default:
1801 		error = EOPNOTSUPP;
1802 		break;
1803 	}
1804 
1805 	/*
1806 	 * If all options have default values, no need to keep the mbuf.
1807 	 */
1808 	if (imo->imo_multicast_ifp == NULL &&
1809 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1810 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1811 	    imo->imo_num_memberships == 0) {
1812 		kmem_free(imo, sizeof(*imo));
1813 		*pimo = NULL;
1814 	}
1815 
1816 	return error;
1817 }
1818 
1819 /*
1820  * Return the IP multicast options in response to user getsockopt().
1821  */
1822 int
1823 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1824 {
1825 	struct in_addr addr;
1826 	struct in_ifaddr *ia;
1827 	uint8_t optval;
1828 	int error = 0;
1829 
1830 	switch (sopt->sopt_name) {
1831 	case IP_MULTICAST_IF:
1832 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
1833 			addr = zeroin_addr;
1834 		else if (imo->imo_multicast_addr.s_addr) {
1835 			/* return the value user has set */
1836 			addr = imo->imo_multicast_addr;
1837 		} else {
1838 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
1839 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1840 		}
1841 		error = sockopt_set(sopt, &addr, sizeof(addr));
1842 		break;
1843 
1844 	case IP_MULTICAST_TTL:
1845 		optval = imo ? imo->imo_multicast_ttl
1846 		    : IP_DEFAULT_MULTICAST_TTL;
1847 
1848 		error = sockopt_set(sopt, &optval, sizeof(optval));
1849 		break;
1850 
1851 	case IP_MULTICAST_LOOP:
1852 		optval = imo ? imo->imo_multicast_loop
1853 		    : IP_DEFAULT_MULTICAST_LOOP;
1854 
1855 		error = sockopt_set(sopt, &optval, sizeof(optval));
1856 		break;
1857 
1858 	default:
1859 		error = EOPNOTSUPP;
1860 	}
1861 
1862 	return error;
1863 }
1864 
1865 /*
1866  * Discard the IP multicast options.
1867  */
1868 void
1869 ip_freemoptions(struct ip_moptions *imo)
1870 {
1871 	int i;
1872 
1873 	if (imo != NULL) {
1874 		for (i = 0; i < imo->imo_num_memberships; ++i)
1875 			in_delmulti(imo->imo_membership[i]);
1876 		kmem_free(imo, sizeof(*imo));
1877 	}
1878 }
1879 
1880 /*
1881  * Routine called from ip_output() to loop back a copy of an IP multicast
1882  * packet to the input queue of a specified interface.  Note that this
1883  * calls the output routine of the loopback "driver", but with an interface
1884  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1885  */
1886 static void
1887 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1888 {
1889 	struct ip *ip;
1890 	struct mbuf *copym;
1891 
1892 	copym = m_copypacket(m, M_DONTWAIT);
1893 	if (copym != NULL &&
1894 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1895 		copym = m_pullup(copym, sizeof(struct ip));
1896 	if (copym == NULL)
1897 		return;
1898 	/*
1899 	 * We don't bother to fragment if the IP length is greater
1900 	 * than the interface's MTU.  Can this possibly matter?
1901 	 */
1902 	ip = mtod(copym, struct ip *);
1903 
1904 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1905 		in_delayed_cksum(copym);
1906 		copym->m_pkthdr.csum_flags &=
1907 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1908 	}
1909 
1910 	ip->ip_sum = 0;
1911 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1912 #ifndef NET_MPSAFE
1913 	KERNEL_LOCK(1, NULL);
1914 #endif
1915 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1916 #ifndef NET_MPSAFE
1917 	KERNEL_UNLOCK_ONE(NULL);
1918 #endif
1919 }
1920