xref: /netbsd-src/sys/netinet/ip_output.c (revision d909946ca08dceb44d7d0f22ec9488679695d976)
1 /*	$NetBSD: ip_output.c,v 1.260 2016/08/01 03:15:30 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 1.260 2016/08/01 03:15:30 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 #include <sys/syslog.h>
116 
117 #include <net/if.h>
118 #include <net/if_types.h>
119 #include <net/route.h>
120 #include <net/pfil.h>
121 
122 #include <netinet/in.h>
123 #include <netinet/in_systm.h>
124 #include <netinet/ip.h>
125 #include <netinet/in_pcb.h>
126 #include <netinet/in_var.h>
127 #include <netinet/ip_var.h>
128 #include <netinet/ip_private.h>
129 #include <netinet/in_offload.h>
130 #include <netinet/portalgo.h>
131 #include <netinet/udp.h>
132 
133 #ifdef INET6
134 #include <netinet6/ip6_var.h>
135 #endif
136 
137 #ifdef MROUTING
138 #include <netinet/ip_mroute.h>
139 #endif
140 
141 #ifdef IPSEC
142 #include <netipsec/ipsec.h>
143 #include <netipsec/key.h>
144 #endif
145 
146 #ifdef MPLS
147 #include <netmpls/mpls.h>
148 #include <netmpls/mpls_var.h>
149 #endif
150 
151 static int ip_pcbopts(struct inpcb *, const struct sockopt *);
152 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
153 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
154 static void ip_mloopback(struct ifnet *, struct mbuf *,
155     const struct sockaddr_in *);
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 socket *so)
230 {
231 	struct rtentry *rt;
232 	struct ip *ip;
233 	struct ifnet *ifp, *mifp = NULL;
234 	struct mbuf *m = m0;
235 	int hlen = sizeof (struct ip);
236 	int len, error = 0;
237 	struct route iproute;
238 	const struct sockaddr_in *dst;
239 	struct in_ifaddr *ia = NULL;
240 	int isbroadcast;
241 	int sw_csum;
242 	u_long mtu;
243 #ifdef IPSEC
244 	struct secpolicy *sp = NULL;
245 #endif
246 	bool natt_frag = false;
247 	bool rtmtu_nolock;
248 	union {
249 		struct sockaddr		dst;
250 		struct sockaddr_in	dst4;
251 	} u;
252 	struct sockaddr *rdst = &u.dst;	/* real IP destination, as opposed
253 					 * to the nexthop
254 					 */
255 	struct psref psref, psref_ia;
256 	int bound;
257 	bool bind_need_restore = false;
258 
259 	len = 0;
260 
261 	MCLAIM(m, &ip_tx_mowner);
262 
263 	KASSERT((m->m_flags & M_PKTHDR) != 0);
264 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
265 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
266 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
267 
268 	if (opt) {
269 		m = ip_insertoptions(m, opt, &len);
270 		if (len >= sizeof(struct ip))
271 			hlen = len;
272 	}
273 	ip = mtod(m, struct ip *);
274 
275 	/*
276 	 * Fill in IP header.
277 	 */
278 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
279 		ip->ip_v = IPVERSION;
280 		ip->ip_off = htons(0);
281 		/* ip->ip_id filled in after we find out source ia */
282 		ip->ip_hl = hlen >> 2;
283 		IP_STATINC(IP_STAT_LOCALOUT);
284 	} else {
285 		hlen = ip->ip_hl << 2;
286 	}
287 
288 	/*
289 	 * Route packet.
290 	 */
291 	if (ro == NULL) {
292 		memset(&iproute, 0, sizeof(iproute));
293 		ro = &iproute;
294 	}
295 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
296 	dst = satocsin(rtcache_getdst(ro));
297 
298 	/*
299 	 * If there is a cached route, check that it is to the same
300 	 * destination and is still up.  If not, free it and try again.
301 	 * The address family should also be checked in case of sharing
302 	 * the cache with IPv6.
303 	 */
304 	if (dst && (dst->sin_family != AF_INET ||
305 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
306 		rtcache_free(ro);
307 
308 	if ((rt = rtcache_validate(ro)) == NULL &&
309 	    (rt = rtcache_update(ro, 1)) == NULL) {
310 		dst = &u.dst4;
311 		error = rtcache_setdst(ro, &u.dst);
312 		if (error != 0)
313 			goto bad;
314 	}
315 
316 	bound = curlwp_bind();
317 	bind_need_restore = true;
318 	/*
319 	 * If routing to interface only, short circuit routing lookup.
320 	 */
321 	if (flags & IP_ROUTETOIF) {
322 		struct ifaddr *ifa;
323 
324 		ifa = ifa_ifwithladdr_psref(sintocsa(dst), &psref_ia);
325 		if (ifa == NULL) {
326 			IP_STATINC(IP_STAT_NOROUTE);
327 			error = ENETUNREACH;
328 			goto bad;
329 		}
330 		/* ia is already referenced by psref_ia */
331 		ia = ifatoia(ifa);
332 
333 		ifp = ia->ia_ifp;
334 		mtu = ifp->if_mtu;
335 		ip->ip_ttl = 1;
336 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
337 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
338 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
339 	    imo != NULL && imo->imo_multicast_if_index != 0) {
340 		ifp = mifp = if_get_byindex(imo->imo_multicast_if_index, &psref);
341 		if (ifp == NULL) {
342 			IP_STATINC(IP_STAT_NOROUTE);
343 			error = ENETUNREACH;
344 			goto bad;
345 		}
346 		mtu = ifp->if_mtu;
347 		ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
348 		if (ia == NULL) {
349 			error = EADDRNOTAVAIL;
350 			goto bad;
351 		}
352 		isbroadcast = 0;
353 	} else {
354 		if (rt == NULL)
355 			rt = rtcache_init(ro);
356 		if (rt == NULL) {
357 			IP_STATINC(IP_STAT_NOROUTE);
358 			error = EHOSTUNREACH;
359 			goto bad;
360 		}
361 		/*
362 		 * XXX NOMPSAFE: depends on accessing rt->rt_ifa isn't racy.
363 		 * Revisit when working on rtentry MP-ification.
364 		 */
365 		ifa_acquire(rt->rt_ifa, &psref_ia);
366 		ia = ifatoia(rt->rt_ifa);
367 		ifp = rt->rt_ifp;
368 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
369 			mtu = ifp->if_mtu;
370 		rt->rt_use++;
371 		if (rt->rt_flags & RTF_GATEWAY)
372 			dst = satosin(rt->rt_gateway);
373 		if (rt->rt_flags & RTF_HOST)
374 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
375 		else
376 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
377 	}
378 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
379 
380 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
381 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
382 		bool inmgroup;
383 
384 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
385 		    M_BCAST : M_MCAST;
386 		/*
387 		 * See if the caller provided any multicast options
388 		 */
389 		if (imo != NULL)
390 			ip->ip_ttl = imo->imo_multicast_ttl;
391 		else
392 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
393 
394 		/*
395 		 * if we don't know the outgoing ifp yet, we can't generate
396 		 * output
397 		 */
398 		if (!ifp) {
399 			IP_STATINC(IP_STAT_NOROUTE);
400 			error = ENETUNREACH;
401 			goto bad;
402 		}
403 
404 		/*
405 		 * If the packet is multicast or broadcast, confirm that
406 		 * the outgoing interface can transmit it.
407 		 */
408 		if (((m->m_flags & M_MCAST) &&
409 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
410 		    ((m->m_flags & M_BCAST) &&
411 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
412 			IP_STATINC(IP_STAT_NOROUTE);
413 			error = ENETUNREACH;
414 			goto bad;
415 		}
416 		/*
417 		 * If source address not specified yet, use an address
418 		 * of outgoing interface.
419 		 */
420 		if (in_nullhost(ip->ip_src)) {
421 			struct in_ifaddr *xia;
422 			struct ifaddr *xifa;
423 			struct psref _psref;
424 
425 			xia = in_get_ia_from_ifp_psref(ifp, &_psref);
426 			if (!xia) {
427 				error = EADDRNOTAVAIL;
428 				goto bad;
429 			}
430 			xifa = &xia->ia_ifa;
431 			if (xifa->ifa_getifa != NULL) {
432 				ia4_release(xia, &_psref);
433 				/* FIXME NOMPSAFE */
434 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
435 				if (xia == NULL) {
436 					error = EADDRNOTAVAIL;
437 					goto bad;
438 				}
439 				ia4_acquire(xia, &_psref);
440 			}
441 			ip->ip_src = xia->ia_addr.sin_addr;
442 			ia4_release(xia, &_psref);
443 		}
444 
445 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
446 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
447 			/*
448 			 * If we belong to the destination multicast group
449 			 * on the outgoing interface, and the caller did not
450 			 * forbid loopback, loop back a copy.
451 			 */
452 			ip_mloopback(ifp, m, &u.dst4);
453 		}
454 #ifdef MROUTING
455 		else {
456 			/*
457 			 * If we are acting as a multicast router, perform
458 			 * multicast forwarding as if the packet had just
459 			 * arrived on the interface to which we are about
460 			 * to send.  The multicast forwarding function
461 			 * recursively calls this function, using the
462 			 * IP_FORWARDING flag to prevent infinite recursion.
463 			 *
464 			 * Multicasts that are looped back by ip_mloopback(),
465 			 * above, will be forwarded by the ip_input() routine,
466 			 * if necessary.
467 			 */
468 			extern struct socket *ip_mrouter;
469 
470 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
471 				if (ip_mforward(m, ifp) != 0) {
472 					m_freem(m);
473 					goto done;
474 				}
475 			}
476 		}
477 #endif
478 		/*
479 		 * Multicasts with a time-to-live of zero may be looped-
480 		 * back, above, but must not be transmitted on a network.
481 		 * Also, multicasts addressed to the loopback interface
482 		 * are not sent -- the above call to ip_mloopback() will
483 		 * loop back a copy if this host actually belongs to the
484 		 * destination group on the loopback interface.
485 		 */
486 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
487 			m_freem(m);
488 			goto done;
489 		}
490 		goto sendit;
491 	}
492 
493 	/*
494 	 * If source address not specified yet, use address
495 	 * of outgoing interface.
496 	 */
497 	if (in_nullhost(ip->ip_src)) {
498 		struct ifaddr *xifa;
499 
500 		xifa = &ia->ia_ifa;
501 		if (xifa->ifa_getifa != NULL) {
502 			ia4_release(ia, &psref_ia);
503 			/* FIXME NOMPSAFE */
504 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
505 			if (ia == NULL) {
506 				error = EADDRNOTAVAIL;
507 				goto bad;
508 			}
509 			ia4_acquire(ia, &psref_ia);
510 		}
511 		ip->ip_src = ia->ia_addr.sin_addr;
512 	}
513 
514 	/*
515 	 * packets with Class-D address as source are not valid per
516 	 * RFC 1112
517 	 */
518 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
519 		IP_STATINC(IP_STAT_ODROPPED);
520 		error = EADDRNOTAVAIL;
521 		goto bad;
522 	}
523 
524 	/*
525 	 * Look for broadcast address and and verify user is allowed to
526 	 * send such a packet.
527 	 */
528 	if (isbroadcast) {
529 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
530 			error = EADDRNOTAVAIL;
531 			goto bad;
532 		}
533 		if ((flags & IP_ALLOWBROADCAST) == 0) {
534 			error = EACCES;
535 			goto bad;
536 		}
537 		/* don't allow broadcast messages to be fragmented */
538 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
539 			error = EMSGSIZE;
540 			goto bad;
541 		}
542 		m->m_flags |= M_BCAST;
543 	} else
544 		m->m_flags &= ~M_BCAST;
545 
546 sendit:
547 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
548 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
549 			ip->ip_id = 0;
550 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
551 			ip->ip_id = ip_newid(ia);
552 		} else {
553 
554 			/*
555 			 * TSO capable interfaces (typically?) increment
556 			 * ip_id for each segment.
557 			 * "allocate" enough ids here to increase the chance
558 			 * for them to be unique.
559 			 *
560 			 * note that the following calculation is not
561 			 * needed to be precise.  wasting some ip_id is fine.
562 			 */
563 
564 			unsigned int segsz = m->m_pkthdr.segsz;
565 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
566 			unsigned int num = howmany(datasz, segsz);
567 
568 			ip->ip_id = ip_newid_range(ia, num);
569 		}
570 	}
571 	if (ia != NULL) {
572 		ia4_release(ia, &psref_ia);
573 		ia = NULL;
574 	}
575 
576 	/*
577 	 * If we're doing Path MTU Discovery, we need to set DF unless
578 	 * the route's MTU is locked.
579 	 */
580 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
581 		ip->ip_off |= htons(IP_DF);
582 	}
583 
584 #ifdef IPSEC
585 	if (ipsec_used) {
586 		bool ipsec_done = false;
587 
588 		/* Perform IPsec processing, if any. */
589 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
590 		    &ipsec_done);
591 		if (error || ipsec_done)
592 			goto done;
593 	}
594 #endif
595 
596 	/*
597 	 * Run through list of hooks for output packets.
598 	 */
599 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
600 	if (error)
601 		goto done;
602 	if (m == NULL)
603 		goto done;
604 
605 	ip = mtod(m, struct ip *);
606 	hlen = ip->ip_hl << 2;
607 
608 	m->m_pkthdr.csum_data |= hlen << 16;
609 
610 #if IFA_STATS
611 	/*
612 	 * search for the source address structure to
613 	 * maintain output statistics.
614 	 */
615 	KASSERT(ia == NULL);
616 	ia = in_get_ia_psref(ip->ip_src, &psref_ia);
617 #endif
618 
619 	/* Maybe skip checksums on loopback interfaces. */
620 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
621 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
622 	}
623 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
624 	/*
625 	 * If small enough for mtu of path, or if using TCP segmentation
626 	 * offload, can just send directly.
627 	 */
628 	if (ntohs(ip->ip_len) <= mtu ||
629 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
630 		const struct sockaddr *sa;
631 
632 #if IFA_STATS
633 		if (ia)
634 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
635 #endif
636 		/*
637 		 * Always initialize the sum to 0!  Some HW assisted
638 		 * checksumming requires this.
639 		 */
640 		ip->ip_sum = 0;
641 
642 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
643 			/*
644 			 * Perform any checksums that the hardware can't do
645 			 * for us.
646 			 *
647 			 * XXX Does any hardware require the {th,uh}_sum
648 			 * XXX fields to be 0?
649 			 */
650 			if (sw_csum & M_CSUM_IPv4) {
651 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
652 				ip->ip_sum = in_cksum(m, hlen);
653 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
654 			}
655 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
656 				if (IN_NEED_CHECKSUM(ifp,
657 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
658 					in_delayed_cksum(m);
659 				}
660 				m->m_pkthdr.csum_flags &=
661 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
662 			}
663 		}
664 
665 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
666 		if (__predict_true(
667 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
668 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
669 			error = ip_if_output(ifp, m, sa, rt);
670 		} else {
671 			error = ip_tso_output(ifp, m, sa, rt);
672 		}
673 		goto done;
674 	}
675 
676 	/*
677 	 * We can't use HW checksumming if we're about to
678 	 * to fragment the packet.
679 	 *
680 	 * XXX Some hardware can do this.
681 	 */
682 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
683 		if (IN_NEED_CHECKSUM(ifp,
684 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
685 			in_delayed_cksum(m);
686 		}
687 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
688 	}
689 
690 	/*
691 	 * Too large for interface; fragment if possible.
692 	 * Must be able to put at least 8 bytes per fragment.
693 	 */
694 	if (ntohs(ip->ip_off) & IP_DF) {
695 		if (flags & IP_RETURNMTU) {
696 			struct inpcb *inp;
697 
698 			KASSERT(so && solocked(so));
699 			inp = sotoinpcb(so);
700 			inp->inp_errormtu = mtu;
701 		}
702 		error = EMSGSIZE;
703 		IP_STATINC(IP_STAT_CANTFRAG);
704 		goto bad;
705 	}
706 
707 	error = ip_fragment(m, ifp, mtu);
708 	if (error) {
709 		m = NULL;
710 		goto bad;
711 	}
712 
713 	for (; m; m = m0) {
714 		m0 = m->m_nextpkt;
715 		m->m_nextpkt = 0;
716 		if (error) {
717 			m_freem(m);
718 			continue;
719 		}
720 #if IFA_STATS
721 		if (ia)
722 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
723 #endif
724 		/*
725 		 * If we get there, the packet has not been handled by
726 		 * IPsec whereas it should have. Now that it has been
727 		 * fragmented, re-inject it in ip_output so that IPsec
728 		 * processing can occur.
729 		 */
730 		if (natt_frag) {
731 			error = ip_output(m, opt, ro,
732 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
733 			    imo, so);
734 		} else {
735 			KASSERT((m->m_pkthdr.csum_flags &
736 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
737 			error = ip_if_output(ifp, m,
738 			    (m->m_flags & M_MCAST) ?
739 			    sintocsa(rdst) : sintocsa(dst), rt);
740 		}
741 	}
742 	if (error == 0) {
743 		IP_STATINC(IP_STAT_FRAGMENTED);
744 	}
745 done:
746 	ia4_release(ia, &psref_ia);
747 	if (ro == &iproute) {
748 		rtcache_free(&iproute);
749 	}
750 #ifdef IPSEC
751 	if (sp) {
752 		KEY_FREESP(&sp);
753 	}
754 #endif
755 	if (mifp != NULL) {
756 		if_put(mifp, &psref);
757 	}
758 	if (bind_need_restore)
759 		curlwp_bindx(bound);
760 	return error;
761 bad:
762 	m_freem(m);
763 	goto done;
764 }
765 
766 int
767 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
768 {
769 	struct ip *ip, *mhip;
770 	struct mbuf *m0;
771 	int len, hlen, off;
772 	int mhlen, firstlen;
773 	struct mbuf **mnext;
774 	int sw_csum = m->m_pkthdr.csum_flags;
775 	int fragments = 0;
776 	int s;
777 	int error = 0;
778 
779 	ip = mtod(m, struct ip *);
780 	hlen = ip->ip_hl << 2;
781 	if (ifp != NULL)
782 		sw_csum &= ~ifp->if_csum_flags_tx;
783 
784 	len = (mtu - hlen) &~ 7;
785 	if (len < 8) {
786 		m_freem(m);
787 		return (EMSGSIZE);
788 	}
789 
790 	firstlen = len;
791 	mnext = &m->m_nextpkt;
792 
793 	/*
794 	 * Loop through length of segment after first fragment,
795 	 * make new header and copy data of each part and link onto chain.
796 	 */
797 	m0 = m;
798 	mhlen = sizeof (struct ip);
799 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
800 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
801 		if (m == 0) {
802 			error = ENOBUFS;
803 			IP_STATINC(IP_STAT_ODROPPED);
804 			goto sendorfree;
805 		}
806 		MCLAIM(m, m0->m_owner);
807 		*mnext = m;
808 		mnext = &m->m_nextpkt;
809 		m->m_data += max_linkhdr;
810 		mhip = mtod(m, struct ip *);
811 		*mhip = *ip;
812 		/* we must inherit MCAST and BCAST flags */
813 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
814 		if (hlen > sizeof (struct ip)) {
815 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
816 			mhip->ip_hl = mhlen >> 2;
817 		}
818 		m->m_len = mhlen;
819 		mhip->ip_off = ((off - hlen) >> 3) +
820 		    (ntohs(ip->ip_off) & ~IP_MF);
821 		if (ip->ip_off & htons(IP_MF))
822 			mhip->ip_off |= IP_MF;
823 		if (off + len >= ntohs(ip->ip_len))
824 			len = ntohs(ip->ip_len) - off;
825 		else
826 			mhip->ip_off |= IP_MF;
827 		HTONS(mhip->ip_off);
828 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
829 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
830 		if (m->m_next == 0) {
831 			error = ENOBUFS;	/* ??? */
832 			IP_STATINC(IP_STAT_ODROPPED);
833 			goto sendorfree;
834 		}
835 		m->m_pkthdr.len = mhlen + len;
836 		m_reset_rcvif(m);
837 		mhip->ip_sum = 0;
838 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
839 		if (sw_csum & M_CSUM_IPv4) {
840 			mhip->ip_sum = in_cksum(m, mhlen);
841 		} else {
842 			/*
843 			 * checksum is hw-offloaded or not necessary.
844 			 */
845 			m->m_pkthdr.csum_flags |=
846 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
847 			m->m_pkthdr.csum_data |= mhlen << 16;
848 			KASSERT(!(ifp != NULL &&
849 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
850 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
851 		}
852 		IP_STATINC(IP_STAT_OFRAGMENTS);
853 		fragments++;
854 	}
855 	/*
856 	 * Update first fragment by trimming what's been copied out
857 	 * and updating header, then send each fragment (in order).
858 	 */
859 	m = m0;
860 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
861 	m->m_pkthdr.len = hlen + firstlen;
862 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
863 	ip->ip_off |= htons(IP_MF);
864 	ip->ip_sum = 0;
865 	if (sw_csum & M_CSUM_IPv4) {
866 		ip->ip_sum = in_cksum(m, hlen);
867 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
868 	} else {
869 		/*
870 		 * checksum is hw-offloaded or not necessary.
871 		 */
872 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
873 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
874 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
875 		    sizeof(struct ip));
876 	}
877 sendorfree:
878 	/*
879 	 * If there is no room for all the fragments, don't queue
880 	 * any of them.
881 	 */
882 	if (ifp != NULL) {
883 		s = splnet();
884 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
885 		    error == 0) {
886 			error = ENOBUFS;
887 			IP_STATINC(IP_STAT_ODROPPED);
888 			IFQ_INC_DROPS(&ifp->if_snd);
889 		}
890 		splx(s);
891 	}
892 	if (error) {
893 		for (m = m0; m; m = m0) {
894 			m0 = m->m_nextpkt;
895 			m->m_nextpkt = NULL;
896 			m_freem(m);
897 		}
898 	}
899 	return (error);
900 }
901 
902 /*
903  * Process a delayed payload checksum calculation.
904  */
905 void
906 in_delayed_cksum(struct mbuf *m)
907 {
908 	struct ip *ip;
909 	u_int16_t csum, offset;
910 
911 	ip = mtod(m, struct ip *);
912 	offset = ip->ip_hl << 2;
913 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
914 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
915 		csum = 0xffff;
916 
917 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
918 
919 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
920 		/* This happen when ip options were inserted
921 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
922 		    m->m_len, offset, ip->ip_p);
923 		 */
924 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
925 	} else
926 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
927 }
928 
929 /*
930  * Determine the maximum length of the options to be inserted;
931  * we would far rather allocate too much space rather than too little.
932  */
933 
934 u_int
935 ip_optlen(struct inpcb *inp)
936 {
937 	struct mbuf *m = inp->inp_options;
938 
939 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
940 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
941 	}
942 	return 0;
943 }
944 
945 /*
946  * Insert IP options into preformed packet.
947  * Adjust IP destination as required for IP source routing,
948  * as indicated by a non-zero in_addr at the start of the options.
949  */
950 static struct mbuf *
951 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
952 {
953 	struct ipoption *p = mtod(opt, struct ipoption *);
954 	struct mbuf *n;
955 	struct ip *ip = mtod(m, struct ip *);
956 	unsigned optlen;
957 
958 	optlen = opt->m_len - sizeof(p->ipopt_dst);
959 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
960 		return (m);		/* XXX should fail */
961 	if (!in_nullhost(p->ipopt_dst))
962 		ip->ip_dst = p->ipopt_dst;
963 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
964 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
965 		if (n == 0)
966 			return (m);
967 		MCLAIM(n, m->m_owner);
968 		M_MOVE_PKTHDR(n, m);
969 		m->m_len -= sizeof(struct ip);
970 		m->m_data += sizeof(struct ip);
971 		n->m_next = m;
972 		m = n;
973 		m->m_len = optlen + sizeof(struct ip);
974 		m->m_data += max_linkhdr;
975 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
976 	} else {
977 		m->m_data -= optlen;
978 		m->m_len += optlen;
979 		memmove(mtod(m, void *), ip, sizeof(struct ip));
980 	}
981 	m->m_pkthdr.len += optlen;
982 	ip = mtod(m, struct ip *);
983 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
984 	*phlen = sizeof(struct ip) + optlen;
985 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
986 	return (m);
987 }
988 
989 /*
990  * Copy options from ip to jp,
991  * omitting those not copied during fragmentation.
992  */
993 int
994 ip_optcopy(struct ip *ip, struct ip *jp)
995 {
996 	u_char *cp, *dp;
997 	int opt, optlen, cnt;
998 
999 	cp = (u_char *)(ip + 1);
1000 	dp = (u_char *)(jp + 1);
1001 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1002 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1003 		opt = cp[0];
1004 		if (opt == IPOPT_EOL)
1005 			break;
1006 		if (opt == IPOPT_NOP) {
1007 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1008 			*dp++ = IPOPT_NOP;
1009 			optlen = 1;
1010 			continue;
1011 		}
1012 
1013 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
1014 		optlen = cp[IPOPT_OLEN];
1015 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
1016 
1017 		/* Invalid lengths should have been caught by ip_dooptions. */
1018 		if (optlen > cnt)
1019 			optlen = cnt;
1020 		if (IPOPT_COPIED(opt)) {
1021 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1022 			dp += optlen;
1023 		}
1024 	}
1025 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1026 		*dp++ = IPOPT_EOL;
1027 	return (optlen);
1028 }
1029 
1030 /*
1031  * IP socket option processing.
1032  */
1033 int
1034 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1035 {
1036 	struct inpcb *inp = sotoinpcb(so);
1037 	struct ip *ip = &inp->inp_ip;
1038 	int inpflags = inp->inp_flags;
1039 	int optval = 0, error = 0;
1040 
1041 	if (sopt->sopt_level != IPPROTO_IP) {
1042 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1043 			return 0;
1044 		return ENOPROTOOPT;
1045 	}
1046 
1047 	switch (op) {
1048 	case PRCO_SETOPT:
1049 		switch (sopt->sopt_name) {
1050 		case IP_OPTIONS:
1051 #ifdef notyet
1052 		case IP_RETOPTS:
1053 #endif
1054 			error = ip_pcbopts(inp, sopt);
1055 			break;
1056 
1057 		case IP_TOS:
1058 		case IP_TTL:
1059 		case IP_MINTTL:
1060 		case IP_PKTINFO:
1061 		case IP_RECVOPTS:
1062 		case IP_RECVRETOPTS:
1063 		case IP_RECVDSTADDR:
1064 		case IP_RECVIF:
1065 		case IP_RECVPKTINFO:
1066 		case IP_RECVTTL:
1067 			error = sockopt_getint(sopt, &optval);
1068 			if (error)
1069 				break;
1070 
1071 			switch (sopt->sopt_name) {
1072 			case IP_TOS:
1073 				ip->ip_tos = optval;
1074 				break;
1075 
1076 			case IP_TTL:
1077 				ip->ip_ttl = optval;
1078 				break;
1079 
1080 			case IP_MINTTL:
1081 				if (optval > 0 && optval <= MAXTTL)
1082 					inp->inp_ip_minttl = optval;
1083 				else
1084 					error = EINVAL;
1085 				break;
1086 #define	OPTSET(bit) \
1087 	if (optval) \
1088 		inpflags |= bit; \
1089 	else \
1090 		inpflags &= ~bit;
1091 
1092 			case IP_PKTINFO:
1093 				OPTSET(INP_PKTINFO);
1094 				break;
1095 
1096 			case IP_RECVOPTS:
1097 				OPTSET(INP_RECVOPTS);
1098 				break;
1099 
1100 			case IP_RECVPKTINFO:
1101 				OPTSET(INP_RECVPKTINFO);
1102 				break;
1103 
1104 			case IP_RECVRETOPTS:
1105 				OPTSET(INP_RECVRETOPTS);
1106 				break;
1107 
1108 			case IP_RECVDSTADDR:
1109 				OPTSET(INP_RECVDSTADDR);
1110 				break;
1111 
1112 			case IP_RECVIF:
1113 				OPTSET(INP_RECVIF);
1114 				break;
1115 
1116 			case IP_RECVTTL:
1117 				OPTSET(INP_RECVTTL);
1118 				break;
1119 			}
1120 		break;
1121 #undef OPTSET
1122 
1123 		case IP_MULTICAST_IF:
1124 		case IP_MULTICAST_TTL:
1125 		case IP_MULTICAST_LOOP:
1126 		case IP_ADD_MEMBERSHIP:
1127 		case IP_DROP_MEMBERSHIP:
1128 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1129 			break;
1130 
1131 		case IP_PORTRANGE:
1132 			error = sockopt_getint(sopt, &optval);
1133 			if (error)
1134 				break;
1135 
1136 			switch (optval) {
1137 			case IP_PORTRANGE_DEFAULT:
1138 			case IP_PORTRANGE_HIGH:
1139 				inpflags &= ~(INP_LOWPORT);
1140 				break;
1141 
1142 			case IP_PORTRANGE_LOW:
1143 				inpflags |= INP_LOWPORT;
1144 				break;
1145 
1146 			default:
1147 				error = EINVAL;
1148 				break;
1149 			}
1150 			break;
1151 
1152 		case IP_PORTALGO:
1153 			error = sockopt_getint(sopt, &optval);
1154 			if (error)
1155 				break;
1156 
1157 			error = portalgo_algo_index_select(
1158 			    (struct inpcb_hdr *)inp, optval);
1159 			break;
1160 
1161 #if defined(IPSEC)
1162 		case IP_IPSEC_POLICY:
1163 			if (ipsec_enabled) {
1164 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1165 				    sopt->sopt_data, sopt->sopt_size,
1166 				    curlwp->l_cred);
1167 				break;
1168 			}
1169 			/*FALLTHROUGH*/
1170 #endif /* IPSEC */
1171 
1172 		default:
1173 			error = ENOPROTOOPT;
1174 			break;
1175 		}
1176 		break;
1177 
1178 	case PRCO_GETOPT:
1179 		switch (sopt->sopt_name) {
1180 		case IP_OPTIONS:
1181 		case IP_RETOPTS: {
1182 			struct mbuf *mopts = inp->inp_options;
1183 
1184 			if (mopts) {
1185 				struct mbuf *m;
1186 
1187 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1188 				if (m == NULL) {
1189 					error = ENOBUFS;
1190 					break;
1191 				}
1192 				error = sockopt_setmbuf(sopt, m);
1193 			}
1194 			break;
1195 		}
1196 		case IP_PKTINFO:
1197 		case IP_TOS:
1198 		case IP_TTL:
1199 		case IP_MINTTL:
1200 		case IP_RECVOPTS:
1201 		case IP_RECVRETOPTS:
1202 		case IP_RECVDSTADDR:
1203 		case IP_RECVIF:
1204 		case IP_RECVPKTINFO:
1205 		case IP_RECVTTL:
1206 		case IP_ERRORMTU:
1207 			switch (sopt->sopt_name) {
1208 			case IP_TOS:
1209 				optval = ip->ip_tos;
1210 				break;
1211 
1212 			case IP_TTL:
1213 				optval = ip->ip_ttl;
1214 				break;
1215 
1216 			case IP_MINTTL:
1217 				optval = inp->inp_ip_minttl;
1218 				break;
1219 
1220 			case IP_ERRORMTU:
1221 				optval = inp->inp_errormtu;
1222 				break;
1223 
1224 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1225 
1226 			case IP_PKTINFO:
1227 				optval = OPTBIT(INP_PKTINFO);
1228 				break;
1229 
1230 			case IP_RECVOPTS:
1231 				optval = OPTBIT(INP_RECVOPTS);
1232 				break;
1233 
1234 			case IP_RECVPKTINFO:
1235 				optval = OPTBIT(INP_RECVPKTINFO);
1236 				break;
1237 
1238 			case IP_RECVRETOPTS:
1239 				optval = OPTBIT(INP_RECVRETOPTS);
1240 				break;
1241 
1242 			case IP_RECVDSTADDR:
1243 				optval = OPTBIT(INP_RECVDSTADDR);
1244 				break;
1245 
1246 			case IP_RECVIF:
1247 				optval = OPTBIT(INP_RECVIF);
1248 				break;
1249 
1250 			case IP_RECVTTL:
1251 				optval = OPTBIT(INP_RECVTTL);
1252 				break;
1253 			}
1254 			error = sockopt_setint(sopt, optval);
1255 			break;
1256 
1257 #if 0	/* defined(IPSEC) */
1258 		case IP_IPSEC_POLICY:
1259 		{
1260 			struct mbuf *m = NULL;
1261 
1262 			/* XXX this will return EINVAL as sopt is empty */
1263 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1264 			    sopt->sopt_size, &m);
1265 			if (error == 0)
1266 				error = sockopt_setmbuf(sopt, m);
1267 			break;
1268 		}
1269 #endif /*IPSEC*/
1270 
1271 		case IP_MULTICAST_IF:
1272 		case IP_MULTICAST_TTL:
1273 		case IP_MULTICAST_LOOP:
1274 		case IP_ADD_MEMBERSHIP:
1275 		case IP_DROP_MEMBERSHIP:
1276 			error = ip_getmoptions(inp->inp_moptions, sopt);
1277 			break;
1278 
1279 		case IP_PORTRANGE:
1280 			if (inpflags & INP_LOWPORT)
1281 				optval = IP_PORTRANGE_LOW;
1282 			else
1283 				optval = IP_PORTRANGE_DEFAULT;
1284 			error = sockopt_setint(sopt, optval);
1285 			break;
1286 
1287 		case IP_PORTALGO:
1288 			optval = inp->inp_portalgo;
1289 			error = sockopt_setint(sopt, optval);
1290 			break;
1291 
1292 		default:
1293 			error = ENOPROTOOPT;
1294 			break;
1295 		}
1296 		break;
1297 	}
1298 
1299 	if (!error) {
1300 		inp->inp_flags = inpflags;
1301 	}
1302 	return error;
1303 }
1304 
1305 /*
1306  * Set up IP options in pcb for insertion in output packets.
1307  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1308  * with destination address if source routed.
1309  */
1310 static int
1311 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1312 {
1313 	struct mbuf *m;
1314 	const u_char *cp;
1315 	u_char *dp;
1316 	int cnt;
1317 
1318 	/* Turn off any old options. */
1319 	if (inp->inp_options) {
1320 		m_free(inp->inp_options);
1321 	}
1322 	inp->inp_options = NULL;
1323 	if ((cnt = sopt->sopt_size) == 0) {
1324 		/* Only turning off any previous options. */
1325 		return 0;
1326 	}
1327 	cp = sopt->sopt_data;
1328 
1329 #ifndef	__vax__
1330 	if (cnt % sizeof(int32_t))
1331 		return (EINVAL);
1332 #endif
1333 
1334 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1335 	if (m == NULL)
1336 		return (ENOBUFS);
1337 
1338 	dp = mtod(m, u_char *);
1339 	memset(dp, 0, sizeof(struct in_addr));
1340 	dp += sizeof(struct in_addr);
1341 	m->m_len = sizeof(struct in_addr);
1342 
1343 	/*
1344 	 * IP option list according to RFC791. Each option is of the form
1345 	 *
1346 	 *	[optval] [olen] [(olen - 2) data bytes]
1347 	 *
1348 	 * We validate the list and copy options to an mbuf for prepending
1349 	 * to data packets. The IP first-hop destination address will be
1350 	 * stored before actual options and is zero if unset.
1351 	 */
1352 	while (cnt > 0) {
1353 		uint8_t optval, olen, offset;
1354 
1355 		optval = cp[IPOPT_OPTVAL];
1356 
1357 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1358 			olen = 1;
1359 		} else {
1360 			if (cnt < IPOPT_OLEN + 1)
1361 				goto bad;
1362 
1363 			olen = cp[IPOPT_OLEN];
1364 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1365 				goto bad;
1366 		}
1367 
1368 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1369 			/*
1370 			 * user process specifies route as:
1371 			 *	->A->B->C->D
1372 			 * D must be our final destination (but we can't
1373 			 * check that since we may not have connected yet).
1374 			 * A is first hop destination, which doesn't appear in
1375 			 * actual IP option, but is stored before the options.
1376 			 */
1377 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1378 				goto bad;
1379 
1380 			offset = cp[IPOPT_OFFSET];
1381 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1382 			    sizeof(struct in_addr));
1383 
1384 			cp += sizeof(struct in_addr);
1385 			cnt -= sizeof(struct in_addr);
1386 			olen -= sizeof(struct in_addr);
1387 
1388 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1389 				goto bad;
1390 
1391 			memcpy(dp, cp, olen);
1392 			dp[IPOPT_OPTVAL] = optval;
1393 			dp[IPOPT_OLEN] = olen;
1394 			dp[IPOPT_OFFSET] = offset;
1395 			break;
1396 		} else {
1397 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1398 				goto bad;
1399 
1400 			memcpy(dp, cp, olen);
1401 			break;
1402 		}
1403 
1404 		dp += olen;
1405 		m->m_len += olen;
1406 
1407 		if (optval == IPOPT_EOL)
1408 			break;
1409 
1410 		cp += olen;
1411 		cnt -= olen;
1412 	}
1413 
1414 	inp->inp_options = m;
1415 	return 0;
1416 bad:
1417 	(void)m_free(m);
1418 	return EINVAL;
1419 }
1420 
1421 /*
1422  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1423  */
1424 static struct ifnet *
1425 ip_multicast_if(struct in_addr *a, int *ifindexp)
1426 {
1427 	int ifindex;
1428 	struct ifnet *ifp = NULL;
1429 	struct in_ifaddr *ia;
1430 
1431 	if (ifindexp)
1432 		*ifindexp = 0;
1433 	if (ntohl(a->s_addr) >> 24 == 0) {
1434 		ifindex = ntohl(a->s_addr) & 0xffffff;
1435 		ifp = if_byindex(ifindex);
1436 		if (!ifp)
1437 			return NULL;
1438 		if (ifindexp)
1439 			*ifindexp = ifindex;
1440 	} else {
1441 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1442 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1443 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1444 				ifp = ia->ia_ifp;
1445 				break;
1446 			}
1447 		}
1448 	}
1449 	return ifp;
1450 }
1451 
1452 static int
1453 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1454 {
1455 	u_int tval;
1456 	u_char cval;
1457 	int error;
1458 
1459 	if (sopt == NULL)
1460 		return EINVAL;
1461 
1462 	switch (sopt->sopt_size) {
1463 	case sizeof(u_char):
1464 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1465 		tval = cval;
1466 		break;
1467 
1468 	case sizeof(u_int):
1469 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1470 		break;
1471 
1472 	default:
1473 		error = EINVAL;
1474 	}
1475 
1476 	if (error)
1477 		return error;
1478 
1479 	if (tval > maxval)
1480 		return EINVAL;
1481 
1482 	*val = tval;
1483 	return 0;
1484 }
1485 
1486 static int
1487 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1488     struct in_addr *ia, bool add)
1489 {
1490 	int error;
1491 	struct ip_mreq mreq;
1492 
1493 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1494 	if (error)
1495 		return error;
1496 
1497 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1498 		return EINVAL;
1499 
1500 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1501 
1502 	if (in_nullhost(mreq.imr_interface)) {
1503 		union {
1504 			struct sockaddr		dst;
1505 			struct sockaddr_in	dst4;
1506 		} u;
1507 		struct route ro;
1508 
1509 		if (!add) {
1510 			*ifp = NULL;
1511 			return 0;
1512 		}
1513 		/*
1514 		 * If no interface address was provided, use the interface of
1515 		 * the route to the given multicast address.
1516 		 */
1517 		struct rtentry *rt;
1518 		memset(&ro, 0, sizeof(ro));
1519 
1520 		sockaddr_in_init(&u.dst4, ia, 0);
1521 		error = rtcache_setdst(&ro, &u.dst);
1522 		if (error != 0)
1523 			return error;
1524 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1525 		rtcache_free(&ro);
1526 	} else {
1527 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1528 		if (!add && *ifp == NULL)
1529 			return EADDRNOTAVAIL;
1530 	}
1531 	return 0;
1532 }
1533 
1534 /*
1535  * Add a multicast group membership.
1536  * Group must be a valid IP multicast address.
1537  */
1538 static int
1539 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1540 {
1541 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1542 	struct in_addr ia;
1543 	int i, error;
1544 
1545 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1546 		error = ip_get_membership(sopt, &ifp, &ia, true);
1547 	else
1548 #ifdef INET6
1549 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1550 #else
1551 		return EINVAL;
1552 #endif
1553 
1554 	if (error)
1555 		return error;
1556 
1557 	/*
1558 	 * See if we found an interface, and confirm that it
1559 	 * supports multicast.
1560 	 */
1561 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1562 		return EADDRNOTAVAIL;
1563 
1564 	/*
1565 	 * See if the membership already exists or if all the
1566 	 * membership slots are full.
1567 	 */
1568 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1569 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1570 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1571 			break;
1572 	}
1573 	if (i < imo->imo_num_memberships)
1574 		return EADDRINUSE;
1575 
1576 	if (i == IP_MAX_MEMBERSHIPS)
1577 		return ETOOMANYREFS;
1578 
1579 	/*
1580 	 * Everything looks good; add a new record to the multicast
1581 	 * address list for the given interface.
1582 	 */
1583 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1584 		return ENOBUFS;
1585 
1586 	++imo->imo_num_memberships;
1587 	return 0;
1588 }
1589 
1590 /*
1591  * Drop a multicast group membership.
1592  * Group must be a valid IP multicast address.
1593  */
1594 static int
1595 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1596 {
1597 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1598 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1599 	int i, error;
1600 
1601 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1602 		error = ip_get_membership(sopt, &ifp, &ia, false);
1603 	else
1604 #ifdef INET6
1605 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1606 #else
1607 		return EINVAL;
1608 #endif
1609 
1610 	if (error)
1611 		return error;
1612 
1613 	/*
1614 	 * Find the membership in the membership array.
1615 	 */
1616 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1617 		if ((ifp == NULL ||
1618 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1619 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1620 			break;
1621 	}
1622 	if (i == imo->imo_num_memberships)
1623 		return EADDRNOTAVAIL;
1624 
1625 	/*
1626 	 * Give up the multicast address record to which the
1627 	 * membership points.
1628 	 */
1629 	in_delmulti(imo->imo_membership[i]);
1630 
1631 	/*
1632 	 * Remove the gap in the membership array.
1633 	 */
1634 	for (++i; i < imo->imo_num_memberships; ++i)
1635 		imo->imo_membership[i-1] = imo->imo_membership[i];
1636 	--imo->imo_num_memberships;
1637 	return 0;
1638 }
1639 
1640 /*
1641  * Set the IP multicast options in response to user setsockopt().
1642  */
1643 int
1644 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1645 {
1646 	struct ip_moptions *imo = *pimo;
1647 	struct in_addr addr;
1648 	struct ifnet *ifp;
1649 	int ifindex, error = 0;
1650 
1651 	if (!imo) {
1652 		/*
1653 		 * No multicast option buffer attached to the pcb;
1654 		 * allocate one and initialize to default values.
1655 		 */
1656 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1657 		if (imo == NULL)
1658 			return ENOBUFS;
1659 
1660 		imo->imo_multicast_if_index = 0;
1661 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1662 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1663 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1664 		imo->imo_num_memberships = 0;
1665 		*pimo = imo;
1666 	}
1667 
1668 	switch (sopt->sopt_name) {
1669 	case IP_MULTICAST_IF:
1670 		/*
1671 		 * Select the interface for outgoing multicast packets.
1672 		 */
1673 		error = sockopt_get(sopt, &addr, sizeof(addr));
1674 		if (error)
1675 			break;
1676 
1677 		/*
1678 		 * INADDR_ANY is used to remove a previous selection.
1679 		 * When no interface is selected, a default one is
1680 		 * chosen every time a multicast packet is sent.
1681 		 */
1682 		if (in_nullhost(addr)) {
1683 			imo->imo_multicast_if_index = 0;
1684 			break;
1685 		}
1686 		/*
1687 		 * The selected interface is identified by its local
1688 		 * IP address.  Find the interface and confirm that
1689 		 * it supports multicasting.
1690 		 */
1691 		ifp = ip_multicast_if(&addr, &ifindex);
1692 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1693 			error = EADDRNOTAVAIL;
1694 			break;
1695 		}
1696 		imo->imo_multicast_if_index = ifp->if_index;
1697 		if (ifindex)
1698 			imo->imo_multicast_addr = addr;
1699 		else
1700 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1701 		break;
1702 
1703 	case IP_MULTICAST_TTL:
1704 		/*
1705 		 * Set the IP time-to-live for outgoing multicast packets.
1706 		 */
1707 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1708 		break;
1709 
1710 	case IP_MULTICAST_LOOP:
1711 		/*
1712 		 * Set the loopback flag for outgoing multicast packets.
1713 		 * Must be zero or one.
1714 		 */
1715 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1716 		break;
1717 
1718 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1719 		error = ip_add_membership(imo, sopt);
1720 		break;
1721 
1722 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1723 		error = ip_drop_membership(imo, sopt);
1724 		break;
1725 
1726 	default:
1727 		error = EOPNOTSUPP;
1728 		break;
1729 	}
1730 
1731 	/*
1732 	 * If all options have default values, no need to keep the mbuf.
1733 	 */
1734 	if (imo->imo_multicast_if_index == 0 &&
1735 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1736 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1737 	    imo->imo_num_memberships == 0) {
1738 		kmem_free(imo, sizeof(*imo));
1739 		*pimo = NULL;
1740 	}
1741 
1742 	return error;
1743 }
1744 
1745 /*
1746  * Return the IP multicast options in response to user getsockopt().
1747  */
1748 int
1749 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1750 {
1751 	struct in_addr addr;
1752 	uint8_t optval;
1753 	int error = 0;
1754 
1755 	switch (sopt->sopt_name) {
1756 	case IP_MULTICAST_IF:
1757 		if (imo == NULL || imo->imo_multicast_if_index == 0)
1758 			addr = zeroin_addr;
1759 		else if (imo->imo_multicast_addr.s_addr) {
1760 			/* return the value user has set */
1761 			addr = imo->imo_multicast_addr;
1762 		} else {
1763 			struct ifnet *ifp;
1764 			struct in_ifaddr *ia = NULL;
1765 			int s = pserialize_read_enter();
1766 
1767 			ifp = if_byindex(imo->imo_multicast_if_index);
1768 			if (ifp != NULL) {
1769 				ia = in_get_ia_from_ifp(ifp);
1770 			}
1771 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1772 			pserialize_read_exit(s);
1773 		}
1774 		error = sockopt_set(sopt, &addr, sizeof(addr));
1775 		break;
1776 
1777 	case IP_MULTICAST_TTL:
1778 		optval = imo ? imo->imo_multicast_ttl
1779 		    : IP_DEFAULT_MULTICAST_TTL;
1780 
1781 		error = sockopt_set(sopt, &optval, sizeof(optval));
1782 		break;
1783 
1784 	case IP_MULTICAST_LOOP:
1785 		optval = imo ? imo->imo_multicast_loop
1786 		    : IP_DEFAULT_MULTICAST_LOOP;
1787 
1788 		error = sockopt_set(sopt, &optval, sizeof(optval));
1789 		break;
1790 
1791 	default:
1792 		error = EOPNOTSUPP;
1793 	}
1794 
1795 	return error;
1796 }
1797 
1798 /*
1799  * Discard the IP multicast options.
1800  */
1801 void
1802 ip_freemoptions(struct ip_moptions *imo)
1803 {
1804 	int i;
1805 
1806 	if (imo != NULL) {
1807 		for (i = 0; i < imo->imo_num_memberships; ++i)
1808 			in_delmulti(imo->imo_membership[i]);
1809 		kmem_free(imo, sizeof(*imo));
1810 	}
1811 }
1812 
1813 /*
1814  * Routine called from ip_output() to loop back a copy of an IP multicast
1815  * packet to the input queue of a specified interface.  Note that this
1816  * calls the output routine of the loopback "driver", but with an interface
1817  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1818  */
1819 static void
1820 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1821 {
1822 	struct ip *ip;
1823 	struct mbuf *copym;
1824 
1825 	copym = m_copypacket(m, M_DONTWAIT);
1826 	if (copym != NULL &&
1827 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1828 		copym = m_pullup(copym, sizeof(struct ip));
1829 	if (copym == NULL)
1830 		return;
1831 	/*
1832 	 * We don't bother to fragment if the IP length is greater
1833 	 * than the interface's MTU.  Can this possibly matter?
1834 	 */
1835 	ip = mtod(copym, struct ip *);
1836 
1837 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1838 		in_delayed_cksum(copym);
1839 		copym->m_pkthdr.csum_flags &=
1840 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1841 	}
1842 
1843 	ip->ip_sum = 0;
1844 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1845 #ifndef NET_MPSAFE
1846 	KERNEL_LOCK(1, NULL);
1847 #endif
1848 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1849 #ifndef NET_MPSAFE
1850 	KERNEL_UNLOCK_ONE(NULL);
1851 #endif
1852 }
1853