xref: /netbsd-src/sys/netinet/ip_output.c (revision f21b7d7f2cbdd5c14b3882c4e8a3d43580d460a6)
1 /*	$NetBSD: ip_output.c,v 1.263 2016/09/20 14:30:13 roy 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 1.263 2016/09/20 14:30:13 roy Exp $");
95 
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_mrouting.h"
100 #include "opt_net_mpsafe.h"
101 #include "opt_mpls.h"
102 #endif
103 
104 #include "arp.h"
105 
106 #include <sys/param.h>
107 #include <sys/kmem.h>
108 #include <sys/mbuf.h>
109 #include <sys/protosw.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/kauth.h>
113 #ifdef IPSEC
114 #include <sys/domain.h>
115 #endif
116 #include <sys/systm.h>
117 #include <sys/syslog.h>
118 
119 #include <net/if.h>
120 #include <net/if_types.h>
121 #include <net/route.h>
122 #include <net/pfil.h>
123 
124 #include <netinet/in.h>
125 #include <netinet/in_systm.h>
126 #include <netinet/ip.h>
127 #include <netinet/in_pcb.h>
128 #include <netinet/in_var.h>
129 #include <netinet/ip_var.h>
130 #include <netinet/ip_private.h>
131 #include <netinet/in_offload.h>
132 #include <netinet/portalgo.h>
133 #include <netinet/udp.h>
134 
135 #ifdef INET6
136 #include <netinet6/ip6_var.h>
137 #endif
138 
139 #ifdef MROUTING
140 #include <netinet/ip_mroute.h>
141 #endif
142 
143 #ifdef IPSEC
144 #include <netipsec/ipsec.h>
145 #include <netipsec/key.h>
146 #endif
147 
148 #ifdef MPLS
149 #include <netmpls/mpls.h>
150 #include <netmpls/mpls_var.h>
151 #endif
152 
153 static int ip_pcbopts(struct inpcb *, const struct sockopt *);
154 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
155 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
156 static void ip_mloopback(struct ifnet *, struct mbuf *,
157     const struct sockaddr_in *);
158 static int ip_ifaddrvalid(const struct in_ifaddr *);
159 
160 extern pfil_head_t *inet_pfil_hook;			/* XXX */
161 
162 int	ip_do_loopback_cksum = 0;
163 
164 static int
165 ip_mark_mpls(struct ifnet * const ifp, struct mbuf * const m,
166     const struct rtentry *rt)
167 {
168 	int error = 0;
169 #ifdef MPLS
170 	union mpls_shim msh;
171 
172 	if (rt == NULL || rt_gettag(rt) == NULL ||
173 	    rt_gettag(rt)->sa_family != AF_MPLS ||
174 	    (m->m_flags & (M_MCAST | M_BCAST)) != 0 ||
175 	    ifp->if_type != IFT_ETHER)
176 		return 0;
177 
178 	msh.s_addr = MPLS_GETSADDR(rt);
179 	if (msh.shim.label != MPLS_LABEL_IMPLNULL) {
180 		struct m_tag *mtag;
181 		/*
182 		 * XXX tentative solution to tell ether_output
183 		 * it's MPLS. Need some more efficient solution.
184 		 */
185 		mtag = m_tag_get(PACKET_TAG_MPLS,
186 		    sizeof(int) /* dummy */,
187 		    M_NOWAIT);
188 		if (mtag == NULL)
189 			return ENOMEM;
190 		m_tag_prepend(m, mtag);
191 	}
192 #endif
193 	return error;
194 }
195 
196 /*
197  * Send an IP packet to a host.
198  */
199 int
200 ip_if_output(struct ifnet * const ifp, struct mbuf * const m,
201     const struct sockaddr * const dst, const struct rtentry *rt)
202 {
203 	int error = 0;
204 
205 	if (rt != NULL) {
206 		error = rt_check_reject_route(rt, ifp);
207 		if (error != 0) {
208 			m_freem(m);
209 			return error;
210 		}
211 	}
212 
213 	error = ip_mark_mpls(ifp, m, rt);
214 	if (error != 0) {
215 		m_freem(m);
216 		return error;
217 	}
218 
219 	error = if_output_lock(ifp, ifp, m, dst, rt);
220 
221 	return error;
222 }
223 
224 /*
225  * IP output.  The packet in mbuf chain m contains a skeletal IP
226  * header (with len, off, ttl, proto, tos, src, dst).
227  * The mbuf chain containing the packet will be freed.
228  * The mbuf opt, if present, will not be freed.
229  */
230 int
231 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro, int flags,
232     struct ip_moptions *imo, struct socket *so)
233 {
234 	struct rtentry *rt;
235 	struct ip *ip;
236 	struct ifnet *ifp, *mifp = NULL;
237 	struct mbuf *m = m0;
238 	int hlen = sizeof (struct ip);
239 	int len, error = 0;
240 	struct route iproute;
241 	const struct sockaddr_in *dst;
242 	struct in_ifaddr *ia = NULL;
243 	int isbroadcast;
244 	int sw_csum;
245 	u_long mtu;
246 #ifdef IPSEC
247 	struct secpolicy *sp = NULL;
248 #endif
249 	bool natt_frag = false;
250 	bool rtmtu_nolock;
251 	union {
252 		struct sockaddr		dst;
253 		struct sockaddr_in	dst4;
254 	} u;
255 	struct sockaddr *rdst = &u.dst;	/* real IP destination, as opposed
256 					 * to the nexthop
257 					 */
258 	struct psref psref, psref_ia;
259 	int bound;
260 	bool bind_need_restore = false;
261 
262 	len = 0;
263 
264 	MCLAIM(m, &ip_tx_mowner);
265 
266 	KASSERT((m->m_flags & M_PKTHDR) != 0);
267 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
268 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
269 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
270 
271 	if (opt) {
272 		m = ip_insertoptions(m, opt, &len);
273 		if (len >= sizeof(struct ip))
274 			hlen = len;
275 	}
276 	ip = mtod(m, struct ip *);
277 
278 	/*
279 	 * Fill in IP header.
280 	 */
281 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
282 		ip->ip_v = IPVERSION;
283 		ip->ip_off = htons(0);
284 		/* ip->ip_id filled in after we find out source ia */
285 		ip->ip_hl = hlen >> 2;
286 		IP_STATINC(IP_STAT_LOCALOUT);
287 	} else {
288 		hlen = ip->ip_hl << 2;
289 	}
290 
291 	/*
292 	 * Route packet.
293 	 */
294 	if (ro == NULL) {
295 		memset(&iproute, 0, sizeof(iproute));
296 		ro = &iproute;
297 	}
298 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
299 	dst = satocsin(rtcache_getdst(ro));
300 
301 	/*
302 	 * If there is a cached route, check that it is to the same
303 	 * destination and is still up.  If not, free it and try again.
304 	 * The address family should also be checked in case of sharing
305 	 * the cache with IPv6.
306 	 */
307 	if (dst && (dst->sin_family != AF_INET ||
308 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
309 		rtcache_free(ro);
310 
311 	if ((rt = rtcache_validate(ro)) == NULL &&
312 	    (rt = rtcache_update(ro, 1)) == NULL) {
313 		dst = &u.dst4;
314 		error = rtcache_setdst(ro, &u.dst);
315 		if (error != 0)
316 			goto bad;
317 	}
318 
319 	bound = curlwp_bind();
320 	bind_need_restore = true;
321 	/*
322 	 * If routing to interface only, short circuit routing lookup.
323 	 */
324 	if (flags & IP_ROUTETOIF) {
325 		struct ifaddr *ifa;
326 
327 		ifa = ifa_ifwithladdr_psref(sintocsa(dst), &psref_ia);
328 		if (ifa == NULL) {
329 			IP_STATINC(IP_STAT_NOROUTE);
330 			error = ENETUNREACH;
331 			goto bad;
332 		}
333 		/* ia is already referenced by psref_ia */
334 		ia = ifatoia(ifa);
335 
336 		ifp = ia->ia_ifp;
337 		mtu = ifp->if_mtu;
338 		ip->ip_ttl = 1;
339 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
340 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
341 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
342 	    imo != NULL && imo->imo_multicast_if_index != 0) {
343 		ifp = mifp = if_get_byindex(imo->imo_multicast_if_index, &psref);
344 		if (ifp == NULL) {
345 			IP_STATINC(IP_STAT_NOROUTE);
346 			error = ENETUNREACH;
347 			goto bad;
348 		}
349 		mtu = ifp->if_mtu;
350 		ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
351 		if (ia == NULL) {
352 			error = EADDRNOTAVAIL;
353 			goto bad;
354 		}
355 		isbroadcast = 0;
356 	} else {
357 		if (rt == NULL)
358 			rt = rtcache_init(ro);
359 		if (rt == NULL) {
360 			IP_STATINC(IP_STAT_NOROUTE);
361 			error = EHOSTUNREACH;
362 			goto bad;
363 		}
364 		/*
365 		 * XXX NOMPSAFE: depends on accessing rt->rt_ifa isn't racy.
366 		 * Revisit when working on rtentry MP-ification.
367 		 */
368 		ifa_acquire(rt->rt_ifa, &psref_ia);
369 		ia = ifatoia(rt->rt_ifa);
370 		ifp = rt->rt_ifp;
371 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
372 			mtu = ifp->if_mtu;
373 		rt->rt_use++;
374 		if (rt->rt_flags & RTF_GATEWAY)
375 			dst = satosin(rt->rt_gateway);
376 		if (rt->rt_flags & RTF_HOST)
377 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
378 		else
379 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
380 	}
381 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
382 
383 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
384 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
385 		bool inmgroup;
386 
387 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
388 		    M_BCAST : M_MCAST;
389 		/*
390 		 * See if the caller provided any multicast options
391 		 */
392 		if (imo != NULL)
393 			ip->ip_ttl = imo->imo_multicast_ttl;
394 		else
395 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
396 
397 		/*
398 		 * if we don't know the outgoing ifp yet, we can't generate
399 		 * output
400 		 */
401 		if (!ifp) {
402 			IP_STATINC(IP_STAT_NOROUTE);
403 			error = ENETUNREACH;
404 			goto bad;
405 		}
406 
407 		/*
408 		 * If the packet is multicast or broadcast, confirm that
409 		 * the outgoing interface can transmit it.
410 		 */
411 		if (((m->m_flags & M_MCAST) &&
412 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
413 		    ((m->m_flags & M_BCAST) &&
414 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
415 			IP_STATINC(IP_STAT_NOROUTE);
416 			error = ENETUNREACH;
417 			goto bad;
418 		}
419 		/*
420 		 * If source address not specified yet, use an address
421 		 * of outgoing interface.
422 		 */
423 		if (in_nullhost(ip->ip_src)) {
424 			struct in_ifaddr *xia;
425 			struct ifaddr *xifa;
426 			struct psref _psref;
427 
428 			xia = in_get_ia_from_ifp_psref(ifp, &_psref);
429 			if (!xia) {
430 				error = EADDRNOTAVAIL;
431 				goto bad;
432 			}
433 			xifa = &xia->ia_ifa;
434 			if (xifa->ifa_getifa != NULL) {
435 				ia4_release(xia, &_psref);
436 				/* FIXME NOMPSAFE */
437 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
438 				if (xia == NULL) {
439 					error = EADDRNOTAVAIL;
440 					goto bad;
441 				}
442 				ia4_acquire(xia, &_psref);
443 			}
444 			ip->ip_src = xia->ia_addr.sin_addr;
445 			ia4_release(xia, &_psref);
446 		}
447 
448 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
449 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
450 			/*
451 			 * If we belong to the destination multicast group
452 			 * on the outgoing interface, and the caller did not
453 			 * forbid loopback, loop back a copy.
454 			 */
455 			ip_mloopback(ifp, m, &u.dst4);
456 		}
457 #ifdef MROUTING
458 		else {
459 			/*
460 			 * If we are acting as a multicast router, perform
461 			 * multicast forwarding as if the packet had just
462 			 * arrived on the interface to which we are about
463 			 * to send.  The multicast forwarding function
464 			 * recursively calls this function, using the
465 			 * IP_FORWARDING flag to prevent infinite recursion.
466 			 *
467 			 * Multicasts that are looped back by ip_mloopback(),
468 			 * above, will be forwarded by the ip_input() routine,
469 			 * if necessary.
470 			 */
471 			extern struct socket *ip_mrouter;
472 
473 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
474 				if (ip_mforward(m, ifp) != 0) {
475 					m_freem(m);
476 					goto done;
477 				}
478 			}
479 		}
480 #endif
481 		/*
482 		 * Multicasts with a time-to-live of zero may be looped-
483 		 * back, above, but must not be transmitted on a network.
484 		 * Also, multicasts addressed to the loopback interface
485 		 * are not sent -- the above call to ip_mloopback() will
486 		 * loop back a copy if this host actually belongs to the
487 		 * destination group on the loopback interface.
488 		 */
489 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
490 			m_freem(m);
491 			goto done;
492 		}
493 		goto sendit;
494 	}
495 
496 	/*
497 	 * If source address not specified yet, use address
498 	 * of outgoing interface.
499 	 */
500 	if (in_nullhost(ip->ip_src)) {
501 		struct ifaddr *xifa;
502 
503 		xifa = &ia->ia_ifa;
504 		if (xifa->ifa_getifa != NULL) {
505 			ia4_release(ia, &psref_ia);
506 			/* FIXME NOMPSAFE */
507 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
508 			if (ia == NULL) {
509 				error = EADDRNOTAVAIL;
510 				goto bad;
511 			}
512 			ia4_acquire(ia, &psref_ia);
513 		}
514 		ip->ip_src = ia->ia_addr.sin_addr;
515 	}
516 
517 	/*
518 	 * packets with Class-D address as source are not valid per
519 	 * RFC 1112
520 	 */
521 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
522 		IP_STATINC(IP_STAT_ODROPPED);
523 		error = EADDRNOTAVAIL;
524 		goto bad;
525 	}
526 
527 	/*
528 	 * Look for broadcast address and and verify user is allowed to
529 	 * send such a packet.
530 	 */
531 	if (isbroadcast) {
532 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
533 			error = EADDRNOTAVAIL;
534 			goto bad;
535 		}
536 		if ((flags & IP_ALLOWBROADCAST) == 0) {
537 			error = EACCES;
538 			goto bad;
539 		}
540 		/* don't allow broadcast messages to be fragmented */
541 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
542 			error = EMSGSIZE;
543 			goto bad;
544 		}
545 		m->m_flags |= M_BCAST;
546 	} else
547 		m->m_flags &= ~M_BCAST;
548 
549 sendit:
550 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
551 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
552 			ip->ip_id = 0;
553 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
554 			ip->ip_id = ip_newid(ia);
555 		} else {
556 
557 			/*
558 			 * TSO capable interfaces (typically?) increment
559 			 * ip_id for each segment.
560 			 * "allocate" enough ids here to increase the chance
561 			 * for them to be unique.
562 			 *
563 			 * note that the following calculation is not
564 			 * needed to be precise.  wasting some ip_id is fine.
565 			 */
566 
567 			unsigned int segsz = m->m_pkthdr.segsz;
568 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
569 			unsigned int num = howmany(datasz, segsz);
570 
571 			ip->ip_id = ip_newid_range(ia, num);
572 		}
573 	}
574 	if (ia != NULL) {
575 		ia4_release(ia, &psref_ia);
576 		ia = NULL;
577 	}
578 
579 	/*
580 	 * If we're doing Path MTU Discovery, we need to set DF unless
581 	 * the route's MTU is locked.
582 	 */
583 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
584 		ip->ip_off |= htons(IP_DF);
585 	}
586 
587 #ifdef IPSEC
588 	if (ipsec_used) {
589 		bool ipsec_done = false;
590 
591 		/* Perform IPsec processing, if any. */
592 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
593 		    &ipsec_done);
594 		if (error || ipsec_done)
595 			goto done;
596 	}
597 #endif
598 
599 	/*
600 	 * Run through list of hooks for output packets.
601 	 */
602 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
603 	if (error)
604 		goto done;
605 	if (m == NULL)
606 		goto done;
607 
608 	ip = mtod(m, struct ip *);
609 	hlen = ip->ip_hl << 2;
610 
611 	m->m_pkthdr.csum_data |= hlen << 16;
612 
613 	/*
614 	 * search for the source address structure to
615 	 * maintain output statistics.
616 	 */
617 	KASSERT(ia == NULL);
618 	ia = in_get_ia_psref(ip->ip_src, &psref_ia);
619 
620 	/* Ensure we only send from a valid address. */
621 	if ((ia != NULL || (flags & IP_FORWARDING) == 0) &&
622 	    (error = ip_ifaddrvalid(ia)) != 0)
623 	{
624 		arplog(LOG_ERR,
625 		    "refusing to send from invalid address %s (pid %d)\n",
626 		    in_fmtaddr(ip->ip_src), curproc->p_pid);
627 		IP_STATINC(IP_STAT_ODROPPED);
628 		if (error == 1)
629 			/*
630 			 * Address exists, but is tentative or detached.
631 			 * We can't send from it because it's invalid,
632 			 * so we drop the packet.
633 			 */
634 			error = 0;
635 		else
636 			error = EADDRNOTAVAIL;
637 		goto bad;
638 	}
639 
640 	/* Maybe skip checksums on loopback interfaces. */
641 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
642 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
643 	}
644 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
645 	/*
646 	 * If small enough for mtu of path, or if using TCP segmentation
647 	 * offload, can just send directly.
648 	 */
649 	if (ntohs(ip->ip_len) <= mtu ||
650 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
651 		const struct sockaddr *sa;
652 
653 #if IFA_STATS
654 		if (ia)
655 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
656 #endif
657 		/*
658 		 * Always initialize the sum to 0!  Some HW assisted
659 		 * checksumming requires this.
660 		 */
661 		ip->ip_sum = 0;
662 
663 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
664 			/*
665 			 * Perform any checksums that the hardware can't do
666 			 * for us.
667 			 *
668 			 * XXX Does any hardware require the {th,uh}_sum
669 			 * XXX fields to be 0?
670 			 */
671 			if (sw_csum & M_CSUM_IPv4) {
672 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
673 				ip->ip_sum = in_cksum(m, hlen);
674 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
675 			}
676 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
677 				if (IN_NEED_CHECKSUM(ifp,
678 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
679 					in_delayed_cksum(m);
680 				}
681 				m->m_pkthdr.csum_flags &=
682 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
683 			}
684 		}
685 
686 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
687 		if (__predict_true(
688 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
689 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
690 			error = ip_if_output(ifp, m, sa, rt);
691 		} else {
692 			error = ip_tso_output(ifp, m, sa, rt);
693 		}
694 		goto done;
695 	}
696 
697 	/*
698 	 * We can't use HW checksumming if we're about to
699 	 * to fragment the packet.
700 	 *
701 	 * XXX Some hardware can do this.
702 	 */
703 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
704 		if (IN_NEED_CHECKSUM(ifp,
705 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
706 			in_delayed_cksum(m);
707 		}
708 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
709 	}
710 
711 	/*
712 	 * Too large for interface; fragment if possible.
713 	 * Must be able to put at least 8 bytes per fragment.
714 	 */
715 	if (ntohs(ip->ip_off) & IP_DF) {
716 		if (flags & IP_RETURNMTU) {
717 			struct inpcb *inp;
718 
719 			KASSERT(so && solocked(so));
720 			inp = sotoinpcb(so);
721 			inp->inp_errormtu = mtu;
722 		}
723 		error = EMSGSIZE;
724 		IP_STATINC(IP_STAT_CANTFRAG);
725 		goto bad;
726 	}
727 
728 	error = ip_fragment(m, ifp, mtu);
729 	if (error) {
730 		m = NULL;
731 		goto bad;
732 	}
733 
734 	for (; m; m = m0) {
735 		m0 = m->m_nextpkt;
736 		m->m_nextpkt = 0;
737 		if (error) {
738 			m_freem(m);
739 			continue;
740 		}
741 #if IFA_STATS
742 		if (ia)
743 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
744 #endif
745 		/*
746 		 * If we get there, the packet has not been handled by
747 		 * IPsec whereas it should have. Now that it has been
748 		 * fragmented, re-inject it in ip_output so that IPsec
749 		 * processing can occur.
750 		 */
751 		if (natt_frag) {
752 			error = ip_output(m, opt, ro,
753 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
754 			    imo, so);
755 		} else {
756 			KASSERT((m->m_pkthdr.csum_flags &
757 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
758 			error = ip_if_output(ifp, m,
759 			    (m->m_flags & M_MCAST) ?
760 			    sintocsa(rdst) : sintocsa(dst), rt);
761 		}
762 	}
763 	if (error == 0) {
764 		IP_STATINC(IP_STAT_FRAGMENTED);
765 	}
766 done:
767 	ia4_release(ia, &psref_ia);
768 	if (ro == &iproute) {
769 		rtcache_free(&iproute);
770 	}
771 #ifdef IPSEC
772 	if (sp) {
773 		KEY_FREESP(&sp);
774 	}
775 #endif
776 	if (mifp != NULL) {
777 		if_put(mifp, &psref);
778 	}
779 	if (bind_need_restore)
780 		curlwp_bindx(bound);
781 	return error;
782 bad:
783 	m_freem(m);
784 	goto done;
785 }
786 
787 int
788 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
789 {
790 	struct ip *ip, *mhip;
791 	struct mbuf *m0;
792 	int len, hlen, off;
793 	int mhlen, firstlen;
794 	struct mbuf **mnext;
795 	int sw_csum = m->m_pkthdr.csum_flags;
796 	int fragments = 0;
797 	int s;
798 	int error = 0;
799 
800 	ip = mtod(m, struct ip *);
801 	hlen = ip->ip_hl << 2;
802 	if (ifp != NULL)
803 		sw_csum &= ~ifp->if_csum_flags_tx;
804 
805 	len = (mtu - hlen) &~ 7;
806 	if (len < 8) {
807 		m_freem(m);
808 		return (EMSGSIZE);
809 	}
810 
811 	firstlen = len;
812 	mnext = &m->m_nextpkt;
813 
814 	/*
815 	 * Loop through length of segment after first fragment,
816 	 * make new header and copy data of each part and link onto chain.
817 	 */
818 	m0 = m;
819 	mhlen = sizeof (struct ip);
820 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
821 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
822 		if (m == 0) {
823 			error = ENOBUFS;
824 			IP_STATINC(IP_STAT_ODROPPED);
825 			goto sendorfree;
826 		}
827 		MCLAIM(m, m0->m_owner);
828 		*mnext = m;
829 		mnext = &m->m_nextpkt;
830 		m->m_data += max_linkhdr;
831 		mhip = mtod(m, struct ip *);
832 		*mhip = *ip;
833 		/* we must inherit MCAST and BCAST flags */
834 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
835 		if (hlen > sizeof (struct ip)) {
836 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
837 			mhip->ip_hl = mhlen >> 2;
838 		}
839 		m->m_len = mhlen;
840 		mhip->ip_off = ((off - hlen) >> 3) +
841 		    (ntohs(ip->ip_off) & ~IP_MF);
842 		if (ip->ip_off & htons(IP_MF))
843 			mhip->ip_off |= IP_MF;
844 		if (off + len >= ntohs(ip->ip_len))
845 			len = ntohs(ip->ip_len) - off;
846 		else
847 			mhip->ip_off |= IP_MF;
848 		HTONS(mhip->ip_off);
849 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
850 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
851 		if (m->m_next == 0) {
852 			error = ENOBUFS;	/* ??? */
853 			IP_STATINC(IP_STAT_ODROPPED);
854 			goto sendorfree;
855 		}
856 		m->m_pkthdr.len = mhlen + len;
857 		m_reset_rcvif(m);
858 		mhip->ip_sum = 0;
859 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
860 		if (sw_csum & M_CSUM_IPv4) {
861 			mhip->ip_sum = in_cksum(m, mhlen);
862 		} else {
863 			/*
864 			 * checksum is hw-offloaded or not necessary.
865 			 */
866 			m->m_pkthdr.csum_flags |=
867 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
868 			m->m_pkthdr.csum_data |= mhlen << 16;
869 			KASSERT(!(ifp != NULL &&
870 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
871 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
872 		}
873 		IP_STATINC(IP_STAT_OFRAGMENTS);
874 		fragments++;
875 	}
876 	/*
877 	 * Update first fragment by trimming what's been copied out
878 	 * and updating header, then send each fragment (in order).
879 	 */
880 	m = m0;
881 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
882 	m->m_pkthdr.len = hlen + firstlen;
883 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
884 	ip->ip_off |= htons(IP_MF);
885 	ip->ip_sum = 0;
886 	if (sw_csum & M_CSUM_IPv4) {
887 		ip->ip_sum = in_cksum(m, hlen);
888 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
889 	} else {
890 		/*
891 		 * checksum is hw-offloaded or not necessary.
892 		 */
893 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
894 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
895 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
896 		    sizeof(struct ip));
897 	}
898 sendorfree:
899 	/*
900 	 * If there is no room for all the fragments, don't queue
901 	 * any of them.
902 	 */
903 	if (ifp != NULL) {
904 		s = splnet();
905 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
906 		    error == 0) {
907 			error = ENOBUFS;
908 			IP_STATINC(IP_STAT_ODROPPED);
909 			IFQ_INC_DROPS(&ifp->if_snd);
910 		}
911 		splx(s);
912 	}
913 	if (error) {
914 		for (m = m0; m; m = m0) {
915 			m0 = m->m_nextpkt;
916 			m->m_nextpkt = NULL;
917 			m_freem(m);
918 		}
919 	}
920 	return (error);
921 }
922 
923 /*
924  * Process a delayed payload checksum calculation.
925  */
926 void
927 in_delayed_cksum(struct mbuf *m)
928 {
929 	struct ip *ip;
930 	u_int16_t csum, offset;
931 
932 	ip = mtod(m, struct ip *);
933 	offset = ip->ip_hl << 2;
934 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
935 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
936 		csum = 0xffff;
937 
938 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
939 
940 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
941 		/* This happen when ip options were inserted
942 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
943 		    m->m_len, offset, ip->ip_p);
944 		 */
945 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
946 	} else
947 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
948 }
949 
950 /*
951  * Determine the maximum length of the options to be inserted;
952  * we would far rather allocate too much space rather than too little.
953  */
954 
955 u_int
956 ip_optlen(struct inpcb *inp)
957 {
958 	struct mbuf *m = inp->inp_options;
959 
960 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
961 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
962 	}
963 	return 0;
964 }
965 
966 /*
967  * Insert IP options into preformed packet.
968  * Adjust IP destination as required for IP source routing,
969  * as indicated by a non-zero in_addr at the start of the options.
970  */
971 static struct mbuf *
972 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
973 {
974 	struct ipoption *p = mtod(opt, struct ipoption *);
975 	struct mbuf *n;
976 	struct ip *ip = mtod(m, struct ip *);
977 	unsigned optlen;
978 
979 	optlen = opt->m_len - sizeof(p->ipopt_dst);
980 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
981 		return (m);		/* XXX should fail */
982 	if (!in_nullhost(p->ipopt_dst))
983 		ip->ip_dst = p->ipopt_dst;
984 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
985 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
986 		if (n == 0)
987 			return (m);
988 		MCLAIM(n, m->m_owner);
989 		M_MOVE_PKTHDR(n, m);
990 		m->m_len -= sizeof(struct ip);
991 		m->m_data += sizeof(struct ip);
992 		n->m_next = m;
993 		m = n;
994 		m->m_len = optlen + sizeof(struct ip);
995 		m->m_data += max_linkhdr;
996 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
997 	} else {
998 		m->m_data -= optlen;
999 		m->m_len += optlen;
1000 		memmove(mtod(m, void *), ip, sizeof(struct ip));
1001 	}
1002 	m->m_pkthdr.len += optlen;
1003 	ip = mtod(m, struct ip *);
1004 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
1005 	*phlen = sizeof(struct ip) + optlen;
1006 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1007 	return (m);
1008 }
1009 
1010 /*
1011  * Copy options from ip to jp,
1012  * omitting those not copied during fragmentation.
1013  */
1014 int
1015 ip_optcopy(struct ip *ip, struct ip *jp)
1016 {
1017 	u_char *cp, *dp;
1018 	int opt, optlen, cnt;
1019 
1020 	cp = (u_char *)(ip + 1);
1021 	dp = (u_char *)(jp + 1);
1022 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1023 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1024 		opt = cp[0];
1025 		if (opt == IPOPT_EOL)
1026 			break;
1027 		if (opt == IPOPT_NOP) {
1028 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1029 			*dp++ = IPOPT_NOP;
1030 			optlen = 1;
1031 			continue;
1032 		}
1033 
1034 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
1035 		optlen = cp[IPOPT_OLEN];
1036 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
1037 
1038 		/* Invalid lengths should have been caught by ip_dooptions. */
1039 		if (optlen > cnt)
1040 			optlen = cnt;
1041 		if (IPOPT_COPIED(opt)) {
1042 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1043 			dp += optlen;
1044 		}
1045 	}
1046 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1047 		*dp++ = IPOPT_EOL;
1048 	return (optlen);
1049 }
1050 
1051 /*
1052  * IP socket option processing.
1053  */
1054 int
1055 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1056 {
1057 	struct inpcb *inp = sotoinpcb(so);
1058 	struct ip *ip = &inp->inp_ip;
1059 	int inpflags = inp->inp_flags;
1060 	int optval = 0, error = 0;
1061 
1062 	if (sopt->sopt_level != IPPROTO_IP) {
1063 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1064 			return 0;
1065 		return ENOPROTOOPT;
1066 	}
1067 
1068 	switch (op) {
1069 	case PRCO_SETOPT:
1070 		switch (sopt->sopt_name) {
1071 		case IP_OPTIONS:
1072 #ifdef notyet
1073 		case IP_RETOPTS:
1074 #endif
1075 			error = ip_pcbopts(inp, sopt);
1076 			break;
1077 
1078 		case IP_TOS:
1079 		case IP_TTL:
1080 		case IP_MINTTL:
1081 		case IP_PKTINFO:
1082 		case IP_RECVOPTS:
1083 		case IP_RECVRETOPTS:
1084 		case IP_RECVDSTADDR:
1085 		case IP_RECVIF:
1086 		case IP_RECVPKTINFO:
1087 		case IP_RECVTTL:
1088 			error = sockopt_getint(sopt, &optval);
1089 			if (error)
1090 				break;
1091 
1092 			switch (sopt->sopt_name) {
1093 			case IP_TOS:
1094 				ip->ip_tos = optval;
1095 				break;
1096 
1097 			case IP_TTL:
1098 				ip->ip_ttl = optval;
1099 				break;
1100 
1101 			case IP_MINTTL:
1102 				if (optval > 0 && optval <= MAXTTL)
1103 					inp->inp_ip_minttl = optval;
1104 				else
1105 					error = EINVAL;
1106 				break;
1107 #define	OPTSET(bit) \
1108 	if (optval) \
1109 		inpflags |= bit; \
1110 	else \
1111 		inpflags &= ~bit;
1112 
1113 			case IP_PKTINFO:
1114 				OPTSET(INP_PKTINFO);
1115 				break;
1116 
1117 			case IP_RECVOPTS:
1118 				OPTSET(INP_RECVOPTS);
1119 				break;
1120 
1121 			case IP_RECVPKTINFO:
1122 				OPTSET(INP_RECVPKTINFO);
1123 				break;
1124 
1125 			case IP_RECVRETOPTS:
1126 				OPTSET(INP_RECVRETOPTS);
1127 				break;
1128 
1129 			case IP_RECVDSTADDR:
1130 				OPTSET(INP_RECVDSTADDR);
1131 				break;
1132 
1133 			case IP_RECVIF:
1134 				OPTSET(INP_RECVIF);
1135 				break;
1136 
1137 			case IP_RECVTTL:
1138 				OPTSET(INP_RECVTTL);
1139 				break;
1140 			}
1141 		break;
1142 #undef OPTSET
1143 
1144 		case IP_MULTICAST_IF:
1145 		case IP_MULTICAST_TTL:
1146 		case IP_MULTICAST_LOOP:
1147 		case IP_ADD_MEMBERSHIP:
1148 		case IP_DROP_MEMBERSHIP:
1149 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1150 			break;
1151 
1152 		case IP_PORTRANGE:
1153 			error = sockopt_getint(sopt, &optval);
1154 			if (error)
1155 				break;
1156 
1157 			switch (optval) {
1158 			case IP_PORTRANGE_DEFAULT:
1159 			case IP_PORTRANGE_HIGH:
1160 				inpflags &= ~(INP_LOWPORT);
1161 				break;
1162 
1163 			case IP_PORTRANGE_LOW:
1164 				inpflags |= INP_LOWPORT;
1165 				break;
1166 
1167 			default:
1168 				error = EINVAL;
1169 				break;
1170 			}
1171 			break;
1172 
1173 		case IP_PORTALGO:
1174 			error = sockopt_getint(sopt, &optval);
1175 			if (error)
1176 				break;
1177 
1178 			error = portalgo_algo_index_select(
1179 			    (struct inpcb_hdr *)inp, optval);
1180 			break;
1181 
1182 #if defined(IPSEC)
1183 		case IP_IPSEC_POLICY:
1184 			if (ipsec_enabled) {
1185 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1186 				    sopt->sopt_data, sopt->sopt_size,
1187 				    curlwp->l_cred);
1188 				break;
1189 			}
1190 			/*FALLTHROUGH*/
1191 #endif /* IPSEC */
1192 
1193 		default:
1194 			error = ENOPROTOOPT;
1195 			break;
1196 		}
1197 		break;
1198 
1199 	case PRCO_GETOPT:
1200 		switch (sopt->sopt_name) {
1201 		case IP_OPTIONS:
1202 		case IP_RETOPTS: {
1203 			struct mbuf *mopts = inp->inp_options;
1204 
1205 			if (mopts) {
1206 				struct mbuf *m;
1207 
1208 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1209 				if (m == NULL) {
1210 					error = ENOBUFS;
1211 					break;
1212 				}
1213 				error = sockopt_setmbuf(sopt, m);
1214 			}
1215 			break;
1216 		}
1217 		case IP_PKTINFO:
1218 		case IP_TOS:
1219 		case IP_TTL:
1220 		case IP_MINTTL:
1221 		case IP_RECVOPTS:
1222 		case IP_RECVRETOPTS:
1223 		case IP_RECVDSTADDR:
1224 		case IP_RECVIF:
1225 		case IP_RECVPKTINFO:
1226 		case IP_RECVTTL:
1227 		case IP_ERRORMTU:
1228 			switch (sopt->sopt_name) {
1229 			case IP_TOS:
1230 				optval = ip->ip_tos;
1231 				break;
1232 
1233 			case IP_TTL:
1234 				optval = ip->ip_ttl;
1235 				break;
1236 
1237 			case IP_MINTTL:
1238 				optval = inp->inp_ip_minttl;
1239 				break;
1240 
1241 			case IP_ERRORMTU:
1242 				optval = inp->inp_errormtu;
1243 				break;
1244 
1245 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1246 
1247 			case IP_PKTINFO:
1248 				optval = OPTBIT(INP_PKTINFO);
1249 				break;
1250 
1251 			case IP_RECVOPTS:
1252 				optval = OPTBIT(INP_RECVOPTS);
1253 				break;
1254 
1255 			case IP_RECVPKTINFO:
1256 				optval = OPTBIT(INP_RECVPKTINFO);
1257 				break;
1258 
1259 			case IP_RECVRETOPTS:
1260 				optval = OPTBIT(INP_RECVRETOPTS);
1261 				break;
1262 
1263 			case IP_RECVDSTADDR:
1264 				optval = OPTBIT(INP_RECVDSTADDR);
1265 				break;
1266 
1267 			case IP_RECVIF:
1268 				optval = OPTBIT(INP_RECVIF);
1269 				break;
1270 
1271 			case IP_RECVTTL:
1272 				optval = OPTBIT(INP_RECVTTL);
1273 				break;
1274 			}
1275 			error = sockopt_setint(sopt, optval);
1276 			break;
1277 
1278 #if 0	/* defined(IPSEC) */
1279 		case IP_IPSEC_POLICY:
1280 		{
1281 			struct mbuf *m = NULL;
1282 
1283 			/* XXX this will return EINVAL as sopt is empty */
1284 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1285 			    sopt->sopt_size, &m);
1286 			if (error == 0)
1287 				error = sockopt_setmbuf(sopt, m);
1288 			break;
1289 		}
1290 #endif /*IPSEC*/
1291 
1292 		case IP_MULTICAST_IF:
1293 		case IP_MULTICAST_TTL:
1294 		case IP_MULTICAST_LOOP:
1295 		case IP_ADD_MEMBERSHIP:
1296 		case IP_DROP_MEMBERSHIP:
1297 			error = ip_getmoptions(inp->inp_moptions, sopt);
1298 			break;
1299 
1300 		case IP_PORTRANGE:
1301 			if (inpflags & INP_LOWPORT)
1302 				optval = IP_PORTRANGE_LOW;
1303 			else
1304 				optval = IP_PORTRANGE_DEFAULT;
1305 			error = sockopt_setint(sopt, optval);
1306 			break;
1307 
1308 		case IP_PORTALGO:
1309 			optval = inp->inp_portalgo;
1310 			error = sockopt_setint(sopt, optval);
1311 			break;
1312 
1313 		default:
1314 			error = ENOPROTOOPT;
1315 			break;
1316 		}
1317 		break;
1318 	}
1319 
1320 	if (!error) {
1321 		inp->inp_flags = inpflags;
1322 	}
1323 	return error;
1324 }
1325 
1326 /*
1327  * Set up IP options in pcb for insertion in output packets.
1328  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1329  * with destination address if source routed.
1330  */
1331 static int
1332 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1333 {
1334 	struct mbuf *m;
1335 	const u_char *cp;
1336 	u_char *dp;
1337 	int cnt;
1338 
1339 	/* Turn off any old options. */
1340 	if (inp->inp_options) {
1341 		m_free(inp->inp_options);
1342 	}
1343 	inp->inp_options = NULL;
1344 	if ((cnt = sopt->sopt_size) == 0) {
1345 		/* Only turning off any previous options. */
1346 		return 0;
1347 	}
1348 	cp = sopt->sopt_data;
1349 
1350 #ifndef	__vax__
1351 	if (cnt % sizeof(int32_t))
1352 		return (EINVAL);
1353 #endif
1354 
1355 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1356 	if (m == NULL)
1357 		return (ENOBUFS);
1358 
1359 	dp = mtod(m, u_char *);
1360 	memset(dp, 0, sizeof(struct in_addr));
1361 	dp += sizeof(struct in_addr);
1362 	m->m_len = sizeof(struct in_addr);
1363 
1364 	/*
1365 	 * IP option list according to RFC791. Each option is of the form
1366 	 *
1367 	 *	[optval] [olen] [(olen - 2) data bytes]
1368 	 *
1369 	 * We validate the list and copy options to an mbuf for prepending
1370 	 * to data packets. The IP first-hop destination address will be
1371 	 * stored before actual options and is zero if unset.
1372 	 */
1373 	while (cnt > 0) {
1374 		uint8_t optval, olen, offset;
1375 
1376 		optval = cp[IPOPT_OPTVAL];
1377 
1378 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1379 			olen = 1;
1380 		} else {
1381 			if (cnt < IPOPT_OLEN + 1)
1382 				goto bad;
1383 
1384 			olen = cp[IPOPT_OLEN];
1385 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1386 				goto bad;
1387 		}
1388 
1389 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1390 			/*
1391 			 * user process specifies route as:
1392 			 *	->A->B->C->D
1393 			 * D must be our final destination (but we can't
1394 			 * check that since we may not have connected yet).
1395 			 * A is first hop destination, which doesn't appear in
1396 			 * actual IP option, but is stored before the options.
1397 			 */
1398 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1399 				goto bad;
1400 
1401 			offset = cp[IPOPT_OFFSET];
1402 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1403 			    sizeof(struct in_addr));
1404 
1405 			cp += sizeof(struct in_addr);
1406 			cnt -= sizeof(struct in_addr);
1407 			olen -= sizeof(struct in_addr);
1408 
1409 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1410 				goto bad;
1411 
1412 			memcpy(dp, cp, olen);
1413 			dp[IPOPT_OPTVAL] = optval;
1414 			dp[IPOPT_OLEN] = olen;
1415 			dp[IPOPT_OFFSET] = offset;
1416 			break;
1417 		} else {
1418 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1419 				goto bad;
1420 
1421 			memcpy(dp, cp, olen);
1422 			break;
1423 		}
1424 
1425 		dp += olen;
1426 		m->m_len += olen;
1427 
1428 		if (optval == IPOPT_EOL)
1429 			break;
1430 
1431 		cp += olen;
1432 		cnt -= olen;
1433 	}
1434 
1435 	inp->inp_options = m;
1436 	return 0;
1437 bad:
1438 	(void)m_free(m);
1439 	return EINVAL;
1440 }
1441 
1442 /*
1443  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1444  */
1445 static struct ifnet *
1446 ip_multicast_if(struct in_addr *a, int *ifindexp)
1447 {
1448 	int ifindex;
1449 	struct ifnet *ifp = NULL;
1450 	struct in_ifaddr *ia;
1451 
1452 	if (ifindexp)
1453 		*ifindexp = 0;
1454 	if (ntohl(a->s_addr) >> 24 == 0) {
1455 		ifindex = ntohl(a->s_addr) & 0xffffff;
1456 		ifp = if_byindex(ifindex);
1457 		if (!ifp)
1458 			return NULL;
1459 		if (ifindexp)
1460 			*ifindexp = ifindex;
1461 	} else {
1462 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1463 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1464 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1465 				ifp = ia->ia_ifp;
1466 				break;
1467 			}
1468 		}
1469 	}
1470 	return ifp;
1471 }
1472 
1473 static int
1474 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1475 {
1476 	u_int tval;
1477 	u_char cval;
1478 	int error;
1479 
1480 	if (sopt == NULL)
1481 		return EINVAL;
1482 
1483 	switch (sopt->sopt_size) {
1484 	case sizeof(u_char):
1485 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1486 		tval = cval;
1487 		break;
1488 
1489 	case sizeof(u_int):
1490 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1491 		break;
1492 
1493 	default:
1494 		error = EINVAL;
1495 	}
1496 
1497 	if (error)
1498 		return error;
1499 
1500 	if (tval > maxval)
1501 		return EINVAL;
1502 
1503 	*val = tval;
1504 	return 0;
1505 }
1506 
1507 static int
1508 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1509     struct in_addr *ia, bool add)
1510 {
1511 	int error;
1512 	struct ip_mreq mreq;
1513 
1514 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1515 	if (error)
1516 		return error;
1517 
1518 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1519 		return EINVAL;
1520 
1521 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1522 
1523 	if (in_nullhost(mreq.imr_interface)) {
1524 		union {
1525 			struct sockaddr		dst;
1526 			struct sockaddr_in	dst4;
1527 		} u;
1528 		struct route ro;
1529 
1530 		if (!add) {
1531 			*ifp = NULL;
1532 			return 0;
1533 		}
1534 		/*
1535 		 * If no interface address was provided, use the interface of
1536 		 * the route to the given multicast address.
1537 		 */
1538 		struct rtentry *rt;
1539 		memset(&ro, 0, sizeof(ro));
1540 
1541 		sockaddr_in_init(&u.dst4, ia, 0);
1542 		error = rtcache_setdst(&ro, &u.dst);
1543 		if (error != 0)
1544 			return error;
1545 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1546 		rtcache_free(&ro);
1547 	} else {
1548 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1549 		if (!add && *ifp == NULL)
1550 			return EADDRNOTAVAIL;
1551 	}
1552 	return 0;
1553 }
1554 
1555 /*
1556  * Add a multicast group membership.
1557  * Group must be a valid IP multicast address.
1558  */
1559 static int
1560 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1561 {
1562 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1563 	struct in_addr ia;
1564 	int i, error;
1565 
1566 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1567 		error = ip_get_membership(sopt, &ifp, &ia, true);
1568 	else
1569 #ifdef INET6
1570 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1571 #else
1572 		return EINVAL;
1573 #endif
1574 
1575 	if (error)
1576 		return error;
1577 
1578 	/*
1579 	 * See if we found an interface, and confirm that it
1580 	 * supports multicast.
1581 	 */
1582 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1583 		return EADDRNOTAVAIL;
1584 
1585 	/*
1586 	 * See if the membership already exists or if all the
1587 	 * membership slots are full.
1588 	 */
1589 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1590 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1591 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1592 			break;
1593 	}
1594 	if (i < imo->imo_num_memberships)
1595 		return EADDRINUSE;
1596 
1597 	if (i == IP_MAX_MEMBERSHIPS)
1598 		return ETOOMANYREFS;
1599 
1600 	/*
1601 	 * Everything looks good; add a new record to the multicast
1602 	 * address list for the given interface.
1603 	 */
1604 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1605 		return ENOBUFS;
1606 
1607 	++imo->imo_num_memberships;
1608 	return 0;
1609 }
1610 
1611 /*
1612  * Drop a multicast group membership.
1613  * Group must be a valid IP multicast address.
1614  */
1615 static int
1616 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1617 {
1618 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1619 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1620 	int i, error;
1621 
1622 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1623 		error = ip_get_membership(sopt, &ifp, &ia, false);
1624 	else
1625 #ifdef INET6
1626 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1627 #else
1628 		return EINVAL;
1629 #endif
1630 
1631 	if (error)
1632 		return error;
1633 
1634 	/*
1635 	 * Find the membership in the membership array.
1636 	 */
1637 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1638 		if ((ifp == NULL ||
1639 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1640 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1641 			break;
1642 	}
1643 	if (i == imo->imo_num_memberships)
1644 		return EADDRNOTAVAIL;
1645 
1646 	/*
1647 	 * Give up the multicast address record to which the
1648 	 * membership points.
1649 	 */
1650 	in_delmulti(imo->imo_membership[i]);
1651 
1652 	/*
1653 	 * Remove the gap in the membership array.
1654 	 */
1655 	for (++i; i < imo->imo_num_memberships; ++i)
1656 		imo->imo_membership[i-1] = imo->imo_membership[i];
1657 	--imo->imo_num_memberships;
1658 	return 0;
1659 }
1660 
1661 /*
1662  * Set the IP multicast options in response to user setsockopt().
1663  */
1664 int
1665 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1666 {
1667 	struct ip_moptions *imo = *pimo;
1668 	struct in_addr addr;
1669 	struct ifnet *ifp;
1670 	int ifindex, error = 0;
1671 
1672 	if (!imo) {
1673 		/*
1674 		 * No multicast option buffer attached to the pcb;
1675 		 * allocate one and initialize to default values.
1676 		 */
1677 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1678 		if (imo == NULL)
1679 			return ENOBUFS;
1680 
1681 		imo->imo_multicast_if_index = 0;
1682 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1683 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1684 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1685 		imo->imo_num_memberships = 0;
1686 		*pimo = imo;
1687 	}
1688 
1689 	switch (sopt->sopt_name) {
1690 	case IP_MULTICAST_IF:
1691 		/*
1692 		 * Select the interface for outgoing multicast packets.
1693 		 */
1694 		error = sockopt_get(sopt, &addr, sizeof(addr));
1695 		if (error)
1696 			break;
1697 
1698 		/*
1699 		 * INADDR_ANY is used to remove a previous selection.
1700 		 * When no interface is selected, a default one is
1701 		 * chosen every time a multicast packet is sent.
1702 		 */
1703 		if (in_nullhost(addr)) {
1704 			imo->imo_multicast_if_index = 0;
1705 			break;
1706 		}
1707 		/*
1708 		 * The selected interface is identified by its local
1709 		 * IP address.  Find the interface and confirm that
1710 		 * it supports multicasting.
1711 		 */
1712 		ifp = ip_multicast_if(&addr, &ifindex);
1713 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1714 			error = EADDRNOTAVAIL;
1715 			break;
1716 		}
1717 		imo->imo_multicast_if_index = ifp->if_index;
1718 		if (ifindex)
1719 			imo->imo_multicast_addr = addr;
1720 		else
1721 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1722 		break;
1723 
1724 	case IP_MULTICAST_TTL:
1725 		/*
1726 		 * Set the IP time-to-live for outgoing multicast packets.
1727 		 */
1728 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1729 		break;
1730 
1731 	case IP_MULTICAST_LOOP:
1732 		/*
1733 		 * Set the loopback flag for outgoing multicast packets.
1734 		 * Must be zero or one.
1735 		 */
1736 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1737 		break;
1738 
1739 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1740 		error = ip_add_membership(imo, sopt);
1741 		break;
1742 
1743 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1744 		error = ip_drop_membership(imo, sopt);
1745 		break;
1746 
1747 	default:
1748 		error = EOPNOTSUPP;
1749 		break;
1750 	}
1751 
1752 	/*
1753 	 * If all options have default values, no need to keep the mbuf.
1754 	 */
1755 	if (imo->imo_multicast_if_index == 0 &&
1756 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1757 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1758 	    imo->imo_num_memberships == 0) {
1759 		kmem_free(imo, sizeof(*imo));
1760 		*pimo = NULL;
1761 	}
1762 
1763 	return error;
1764 }
1765 
1766 /*
1767  * Return the IP multicast options in response to user getsockopt().
1768  */
1769 int
1770 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1771 {
1772 	struct in_addr addr;
1773 	uint8_t optval;
1774 	int error = 0;
1775 
1776 	switch (sopt->sopt_name) {
1777 	case IP_MULTICAST_IF:
1778 		if (imo == NULL || imo->imo_multicast_if_index == 0)
1779 			addr = zeroin_addr;
1780 		else if (imo->imo_multicast_addr.s_addr) {
1781 			/* return the value user has set */
1782 			addr = imo->imo_multicast_addr;
1783 		} else {
1784 			struct ifnet *ifp;
1785 			struct in_ifaddr *ia = NULL;
1786 			int s = pserialize_read_enter();
1787 
1788 			ifp = if_byindex(imo->imo_multicast_if_index);
1789 			if (ifp != NULL) {
1790 				ia = in_get_ia_from_ifp(ifp);
1791 			}
1792 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1793 			pserialize_read_exit(s);
1794 		}
1795 		error = sockopt_set(sopt, &addr, sizeof(addr));
1796 		break;
1797 
1798 	case IP_MULTICAST_TTL:
1799 		optval = imo ? imo->imo_multicast_ttl
1800 		    : IP_DEFAULT_MULTICAST_TTL;
1801 
1802 		error = sockopt_set(sopt, &optval, sizeof(optval));
1803 		break;
1804 
1805 	case IP_MULTICAST_LOOP:
1806 		optval = imo ? imo->imo_multicast_loop
1807 		    : IP_DEFAULT_MULTICAST_LOOP;
1808 
1809 		error = sockopt_set(sopt, &optval, sizeof(optval));
1810 		break;
1811 
1812 	default:
1813 		error = EOPNOTSUPP;
1814 	}
1815 
1816 	return error;
1817 }
1818 
1819 /*
1820  * Discard the IP multicast options.
1821  */
1822 void
1823 ip_freemoptions(struct ip_moptions *imo)
1824 {
1825 	int i;
1826 
1827 	if (imo != NULL) {
1828 		for (i = 0; i < imo->imo_num_memberships; ++i)
1829 			in_delmulti(imo->imo_membership[i]);
1830 		kmem_free(imo, sizeof(*imo));
1831 	}
1832 }
1833 
1834 /*
1835  * Routine called from ip_output() to loop back a copy of an IP multicast
1836  * packet to the input queue of a specified interface.  Note that this
1837  * calls the output routine of the loopback "driver", but with an interface
1838  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1839  */
1840 static void
1841 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1842 {
1843 	struct ip *ip;
1844 	struct mbuf *copym;
1845 
1846 	copym = m_copypacket(m, M_DONTWAIT);
1847 	if (copym != NULL &&
1848 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1849 		copym = m_pullup(copym, sizeof(struct ip));
1850 	if (copym == NULL)
1851 		return;
1852 	/*
1853 	 * We don't bother to fragment if the IP length is greater
1854 	 * than the interface's MTU.  Can this possibly matter?
1855 	 */
1856 	ip = mtod(copym, struct ip *);
1857 
1858 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1859 		in_delayed_cksum(copym);
1860 		copym->m_pkthdr.csum_flags &=
1861 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1862 	}
1863 
1864 	ip->ip_sum = 0;
1865 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1866 #ifndef NET_MPSAFE
1867 	KERNEL_LOCK(1, NULL);
1868 #endif
1869 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1870 #ifndef NET_MPSAFE
1871 	KERNEL_UNLOCK_ONE(NULL);
1872 #endif
1873 }
1874 
1875 /*
1876  * Ensure sending address is valid.
1877  * Returns 0 on success, -1 if an error should be sent back or 1
1878  * if the packet could be dropped without error (protocol dependent).
1879  */
1880 static int
1881 ip_ifaddrvalid(const struct in_ifaddr *ia)
1882 {
1883 
1884 	if (ia == NULL)
1885 		return -1;
1886 
1887 	if (ia->ia_addr.sin_addr.s_addr == INADDR_ANY)
1888 		return 0;
1889 
1890 	if (ia->ia4_flags & IN_IFF_DUPLICATED)
1891 		return -1;
1892 	else if (ia->ia4_flags & (IN_IFF_TENTATIVE | IN_IFF_DETACHED))
1893 		return 1;
1894 
1895 	return 0;
1896 }
1897