xref: /netbsd-src/sys/netinet/ip_output.c (revision cc576e1d8e4f4078fd4e81238abca9fca216f6ec)
1 /*	$NetBSD: ip_output.c,v 1.269 2017/01/16 15:14:16 christos Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1998 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Public Access Networks Corporation ("Panix").  It was developed under
38  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
50  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
51  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
52  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
53  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
54  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
55  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
56  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
57  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
58  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
59  * POSSIBILITY OF SUCH DAMAGE.
60  */
61 
62 /*
63  * Copyright (c) 1982, 1986, 1988, 1990, 1993
64  *	The Regents of the University of California.  All rights reserved.
65  *
66  * Redistribution and use in source and binary forms, with or without
67  * modification, are permitted provided that the following conditions
68  * are met:
69  * 1. Redistributions of source code must retain the above copyright
70  *    notice, this list of conditions and the following disclaimer.
71  * 2. Redistributions in binary form must reproduce the above copyright
72  *    notice, this list of conditions and the following disclaimer in the
73  *    documentation and/or other materials provided with the distribution.
74  * 3. Neither the name of the University nor the names of its contributors
75  *    may be used to endorse or promote products derived from this software
76  *    without specific prior written permission.
77  *
78  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88  * SUCH DAMAGE.
89  *
90  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: ip_output.c,v 1.269 2017/01/16 15:14:16 christos Exp $");
95 
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_mrouting.h"
100 #include "opt_net_mpsafe.h"
101 #include "opt_mpls.h"
102 #endif
103 
104 #include "arp.h"
105 
106 #include <sys/param.h>
107 #include <sys/kmem.h>
108 #include <sys/mbuf.h>
109 #include <sys/socket.h>
110 #include <sys/socketvar.h>
111 #include <sys/kauth.h>
112 #include <sys/systm.h>
113 #include <sys/syslog.h>
114 
115 #include <net/if.h>
116 #include <net/if_types.h>
117 #include <net/route.h>
118 #include <net/pfil.h>
119 
120 #include <netinet/in.h>
121 #include <netinet/in_systm.h>
122 #include <netinet/ip.h>
123 #include <netinet/in_pcb.h>
124 #include <netinet/in_var.h>
125 #include <netinet/ip_var.h>
126 #include <netinet/ip_private.h>
127 #include <netinet/in_offload.h>
128 #include <netinet/portalgo.h>
129 #include <netinet/udp.h>
130 
131 #ifdef INET6
132 #include <netinet6/ip6_var.h>
133 #endif
134 
135 #ifdef MROUTING
136 #include <netinet/ip_mroute.h>
137 #endif
138 
139 #ifdef IPSEC
140 #include <netipsec/ipsec.h>
141 #include <netipsec/key.h>
142 #endif
143 
144 #ifdef MPLS
145 #include <netmpls/mpls.h>
146 #include <netmpls/mpls_var.h>
147 #endif
148 
149 static int ip_pcbopts(struct inpcb *, const struct sockopt *);
150 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
151 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
152 static void ip_mloopback(struct ifnet *, struct mbuf *,
153     const struct sockaddr_in *);
154 static int ip_ifaddrvalid(const struct in_ifaddr *);
155 
156 extern pfil_head_t *inet_pfil_hook;			/* XXX */
157 
158 int	ip_do_loopback_cksum = 0;
159 
160 static int
161 ip_mark_mpls(struct ifnet * const ifp, struct mbuf * const m,
162     const struct rtentry *rt)
163 {
164 	int error = 0;
165 #ifdef MPLS
166 	union mpls_shim msh;
167 
168 	if (rt == NULL || rt_gettag(rt) == NULL ||
169 	    rt_gettag(rt)->sa_family != AF_MPLS ||
170 	    (m->m_flags & (M_MCAST | M_BCAST)) != 0 ||
171 	    ifp->if_type != IFT_ETHER)
172 		return 0;
173 
174 	msh.s_addr = MPLS_GETSADDR(rt);
175 	if (msh.shim.label != MPLS_LABEL_IMPLNULL) {
176 		struct m_tag *mtag;
177 		/*
178 		 * XXX tentative solution to tell ether_output
179 		 * it's MPLS. Need some more efficient solution.
180 		 */
181 		mtag = m_tag_get(PACKET_TAG_MPLS,
182 		    sizeof(int) /* dummy */,
183 		    M_NOWAIT);
184 		if (mtag == NULL)
185 			return ENOMEM;
186 		m_tag_prepend(m, mtag);
187 	}
188 #endif
189 	return error;
190 }
191 
192 /*
193  * Send an IP packet to a host.
194  */
195 int
196 ip_if_output(struct ifnet * const ifp, struct mbuf * const m,
197     const struct sockaddr * const dst, const struct rtentry *rt)
198 {
199 	int error = 0;
200 
201 	if (rt != NULL) {
202 		error = rt_check_reject_route(rt, ifp);
203 		if (error != 0) {
204 			m_freem(m);
205 			return error;
206 		}
207 	}
208 
209 	error = ip_mark_mpls(ifp, m, rt);
210 	if (error != 0) {
211 		m_freem(m);
212 		return error;
213 	}
214 
215 	error = if_output_lock(ifp, ifp, m, dst, rt);
216 
217 	return error;
218 }
219 
220 /*
221  * IP output.  The packet in mbuf chain m contains a skeletal IP
222  * header (with len, off, ttl, proto, tos, src, dst).
223  * The mbuf chain containing the packet will be freed.
224  * The mbuf opt, if present, will not be freed.
225  */
226 int
227 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro, int flags,
228     struct ip_moptions *imo, struct socket *so)
229 {
230 	struct rtentry *rt;
231 	struct ip *ip;
232 	struct ifnet *ifp, *mifp = NULL;
233 	struct mbuf *m = m0;
234 	int hlen = sizeof (struct ip);
235 	int len, error = 0;
236 	struct route iproute;
237 	const struct sockaddr_in *dst;
238 	struct in_ifaddr *ia = NULL;
239 	int isbroadcast;
240 	int sw_csum;
241 	u_long mtu;
242 #ifdef IPSEC
243 	struct secpolicy *sp = NULL;
244 #endif
245 	bool natt_frag = false;
246 	bool rtmtu_nolock;
247 	union {
248 		struct sockaddr		dst;
249 		struct sockaddr_in	dst4;
250 	} u;
251 	struct sockaddr *rdst = &u.dst;	/* real IP destination, as opposed
252 					 * to the nexthop
253 					 */
254 	struct psref psref, psref_ia;
255 	int bound;
256 	bool bind_need_restore = false;
257 
258 	len = 0;
259 
260 	MCLAIM(m, &ip_tx_mowner);
261 
262 	KASSERT((m->m_flags & M_PKTHDR) != 0);
263 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
264 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
265 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
266 
267 	if (opt) {
268 		m = ip_insertoptions(m, opt, &len);
269 		if (len >= sizeof(struct ip))
270 			hlen = len;
271 	}
272 	ip = mtod(m, struct ip *);
273 
274 	/*
275 	 * Fill in IP header.
276 	 */
277 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
278 		ip->ip_v = IPVERSION;
279 		ip->ip_off = htons(0);
280 		/* ip->ip_id filled in after we find out source ia */
281 		ip->ip_hl = hlen >> 2;
282 		IP_STATINC(IP_STAT_LOCALOUT);
283 	} else {
284 		hlen = ip->ip_hl << 2;
285 	}
286 
287 	/*
288 	 * Route packet.
289 	 */
290 	if (ro == NULL) {
291 		memset(&iproute, 0, sizeof(iproute));
292 		ro = &iproute;
293 	}
294 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
295 	dst = satocsin(rtcache_getdst(ro));
296 
297 	/*
298 	 * If there is a cached route, check that it is to the same
299 	 * destination and is still up.  If not, free it and try again.
300 	 * The address family should also be checked in case of sharing
301 	 * the cache with IPv6.
302 	 */
303 	if (dst && (dst->sin_family != AF_INET ||
304 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
305 		rtcache_free(ro);
306 
307 	if ((rt = rtcache_validate(ro)) == NULL &&
308 	    (rt = rtcache_update(ro, 1)) == NULL) {
309 		dst = &u.dst4;
310 		error = rtcache_setdst(ro, &u.dst);
311 		if (error != 0)
312 			goto bad;
313 	}
314 
315 	bound = curlwp_bind();
316 	bind_need_restore = true;
317 	/*
318 	 * If routing to interface only, short circuit routing lookup.
319 	 */
320 	if (flags & IP_ROUTETOIF) {
321 		struct ifaddr *ifa;
322 
323 		ifa = ifa_ifwithladdr_psref(sintocsa(dst), &psref_ia);
324 		if (ifa == NULL) {
325 			IP_STATINC(IP_STAT_NOROUTE);
326 			error = ENETUNREACH;
327 			goto bad;
328 		}
329 		/* ia is already referenced by psref_ia */
330 		ia = ifatoia(ifa);
331 
332 		ifp = ia->ia_ifp;
333 		mtu = ifp->if_mtu;
334 		ip->ip_ttl = 1;
335 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
336 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
337 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
338 	    imo != NULL && imo->imo_multicast_if_index != 0) {
339 		ifp = mifp = if_get_byindex(imo->imo_multicast_if_index, &psref);
340 		if (ifp == NULL) {
341 			IP_STATINC(IP_STAT_NOROUTE);
342 			error = ENETUNREACH;
343 			goto bad;
344 		}
345 		mtu = ifp->if_mtu;
346 		ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
347 		if (ia == NULL) {
348 			error = EADDRNOTAVAIL;
349 			goto bad;
350 		}
351 		isbroadcast = 0;
352 	} else {
353 		if (rt == NULL)
354 			rt = rtcache_init(ro);
355 		if (rt == NULL) {
356 			IP_STATINC(IP_STAT_NOROUTE);
357 			error = EHOSTUNREACH;
358 			goto bad;
359 		}
360 		if (ifa_is_destroying(rt->rt_ifa)) {
361 			rtcache_unref(rt, ro);
362 			rt = NULL;
363 			IP_STATINC(IP_STAT_NOROUTE);
364 			error = EHOSTUNREACH;
365 			goto bad;
366 		}
367 		ifa_acquire(rt->rt_ifa, &psref_ia);
368 		ia = ifatoia(rt->rt_ifa);
369 		ifp = rt->rt_ifp;
370 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
371 			mtu = ifp->if_mtu;
372 		rt->rt_use++;
373 		if (rt->rt_flags & RTF_GATEWAY)
374 			dst = satosin(rt->rt_gateway);
375 		if (rt->rt_flags & RTF_HOST)
376 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
377 		else
378 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
379 	}
380 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
381 
382 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
383 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
384 		bool inmgroup;
385 
386 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
387 		    M_BCAST : M_MCAST;
388 		/*
389 		 * See if the caller provided any multicast options
390 		 */
391 		if (imo != NULL)
392 			ip->ip_ttl = imo->imo_multicast_ttl;
393 		else
394 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
395 
396 		/*
397 		 * if we don't know the outgoing ifp yet, we can't generate
398 		 * output
399 		 */
400 		if (!ifp) {
401 			IP_STATINC(IP_STAT_NOROUTE);
402 			error = ENETUNREACH;
403 			goto bad;
404 		}
405 
406 		/*
407 		 * If the packet is multicast or broadcast, confirm that
408 		 * the outgoing interface can transmit it.
409 		 */
410 		if (((m->m_flags & M_MCAST) &&
411 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
412 		    ((m->m_flags & M_BCAST) &&
413 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
414 			IP_STATINC(IP_STAT_NOROUTE);
415 			error = ENETUNREACH;
416 			goto bad;
417 		}
418 		/*
419 		 * If source address not specified yet, use an address
420 		 * of outgoing interface.
421 		 */
422 		if (in_nullhost(ip->ip_src)) {
423 			struct in_ifaddr *xia;
424 			struct ifaddr *xifa;
425 			struct psref _psref;
426 
427 			xia = in_get_ia_from_ifp_psref(ifp, &_psref);
428 			if (!xia) {
429 				error = EADDRNOTAVAIL;
430 				goto bad;
431 			}
432 			xifa = &xia->ia_ifa;
433 			if (xifa->ifa_getifa != NULL) {
434 				ia4_release(xia, &_psref);
435 				/* FIXME NOMPSAFE */
436 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
437 				if (xia == NULL) {
438 					error = EADDRNOTAVAIL;
439 					goto bad;
440 				}
441 				ia4_acquire(xia, &_psref);
442 			}
443 			ip->ip_src = xia->ia_addr.sin_addr;
444 			ia4_release(xia, &_psref);
445 		}
446 
447 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
448 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
449 			/*
450 			 * If we belong to the destination multicast group
451 			 * on the outgoing interface, and the caller did not
452 			 * forbid loopback, loop back a copy.
453 			 */
454 			ip_mloopback(ifp, m, &u.dst4);
455 		}
456 #ifdef MROUTING
457 		else {
458 			/*
459 			 * If we are acting as a multicast router, perform
460 			 * multicast forwarding as if the packet had just
461 			 * arrived on the interface to which we are about
462 			 * to send.  The multicast forwarding function
463 			 * recursively calls this function, using the
464 			 * IP_FORWARDING flag to prevent infinite recursion.
465 			 *
466 			 * Multicasts that are looped back by ip_mloopback(),
467 			 * above, will be forwarded by the ip_input() routine,
468 			 * if necessary.
469 			 */
470 			extern struct socket *ip_mrouter;
471 
472 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
473 				if (ip_mforward(m, ifp) != 0) {
474 					m_freem(m);
475 					goto done;
476 				}
477 			}
478 		}
479 #endif
480 		/*
481 		 * Multicasts with a time-to-live of zero may be looped-
482 		 * back, above, but must not be transmitted on a network.
483 		 * Also, multicasts addressed to the loopback interface
484 		 * are not sent -- the above call to ip_mloopback() will
485 		 * loop back a copy if this host actually belongs to the
486 		 * destination group on the loopback interface.
487 		 */
488 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
489 			m_freem(m);
490 			goto done;
491 		}
492 		goto sendit;
493 	}
494 
495 	/*
496 	 * If source address not specified yet, use address
497 	 * of outgoing interface.
498 	 */
499 	if (in_nullhost(ip->ip_src)) {
500 		struct ifaddr *xifa;
501 
502 		xifa = &ia->ia_ifa;
503 		if (xifa->ifa_getifa != NULL) {
504 			ia4_release(ia, &psref_ia);
505 			/* FIXME NOMPSAFE */
506 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
507 			if (ia == NULL) {
508 				error = EADDRNOTAVAIL;
509 				goto bad;
510 			}
511 			ia4_acquire(ia, &psref_ia);
512 		}
513 		ip->ip_src = ia->ia_addr.sin_addr;
514 	}
515 
516 	/*
517 	 * packets with Class-D address as source are not valid per
518 	 * RFC 1112
519 	 */
520 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
521 		IP_STATINC(IP_STAT_ODROPPED);
522 		error = EADDRNOTAVAIL;
523 		goto bad;
524 	}
525 
526 	/*
527 	 * Look for broadcast address and and verify user is allowed to
528 	 * send such a packet.
529 	 */
530 	if (isbroadcast) {
531 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
532 			error = EADDRNOTAVAIL;
533 			goto bad;
534 		}
535 		if ((flags & IP_ALLOWBROADCAST) == 0) {
536 			error = EACCES;
537 			goto bad;
538 		}
539 		/* don't allow broadcast messages to be fragmented */
540 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
541 			error = EMSGSIZE;
542 			goto bad;
543 		}
544 		m->m_flags |= M_BCAST;
545 	} else
546 		m->m_flags &= ~M_BCAST;
547 
548 sendit:
549 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
550 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
551 			ip->ip_id = 0;
552 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
553 			ip->ip_id = ip_newid(ia);
554 		} else {
555 
556 			/*
557 			 * TSO capable interfaces (typically?) increment
558 			 * ip_id for each segment.
559 			 * "allocate" enough ids here to increase the chance
560 			 * for them to be unique.
561 			 *
562 			 * note that the following calculation is not
563 			 * needed to be precise.  wasting some ip_id is fine.
564 			 */
565 
566 			unsigned int segsz = m->m_pkthdr.segsz;
567 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
568 			unsigned int num = howmany(datasz, segsz);
569 
570 			ip->ip_id = ip_newid_range(ia, num);
571 		}
572 	}
573 	if (ia != NULL) {
574 		ia4_release(ia, &psref_ia);
575 		ia = NULL;
576 	}
577 
578 	/*
579 	 * If we're doing Path MTU Discovery, we need to set DF unless
580 	 * the route's MTU is locked.
581 	 */
582 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
583 		ip->ip_off |= htons(IP_DF);
584 	}
585 
586 #ifdef IPSEC
587 	if (ipsec_used) {
588 		bool ipsec_done = false;
589 
590 		/* Perform IPsec processing, if any. */
591 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
592 		    &ipsec_done);
593 		if (error || ipsec_done)
594 			goto done;
595 	}
596 #endif
597 
598 	/*
599 	 * Run through list of hooks for output packets.
600 	 */
601 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
602 	if (error)
603 		goto done;
604 	if (m == NULL)
605 		goto done;
606 
607 	ip = mtod(m, struct ip *);
608 	hlen = ip->ip_hl << 2;
609 
610 	m->m_pkthdr.csum_data |= hlen << 16;
611 
612 	/*
613 	 * search for the source address structure to
614 	 * maintain output statistics.
615 	 */
616 	KASSERT(ia == NULL);
617 	ia = in_get_ia_psref(ip->ip_src, &psref_ia);
618 
619 	/* Ensure we only send from a valid address. */
620 	if ((ia != NULL || (flags & IP_FORWARDING) == 0) &&
621 	    (error = ip_ifaddrvalid(ia)) != 0)
622 	{
623 		ARPLOG(LOG_ERR,
624 		    "refusing to send from invalid address %s (pid %d)\n",
625 		    ARPLOGADDR(ip->ip_src), curproc->p_pid);
626 		IP_STATINC(IP_STAT_ODROPPED);
627 		if (error == 1)
628 			/*
629 			 * Address exists, but is tentative or detached.
630 			 * We can't send from it because it's invalid,
631 			 * so we drop the packet.
632 			 */
633 			error = 0;
634 		else
635 			error = EADDRNOTAVAIL;
636 		goto bad;
637 	}
638 
639 	/* Maybe skip checksums on loopback interfaces. */
640 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
641 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
642 	}
643 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
644 	/*
645 	 * If small enough for mtu of path, or if using TCP segmentation
646 	 * offload, can just send directly.
647 	 */
648 	if (ntohs(ip->ip_len) <= mtu ||
649 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
650 		const struct sockaddr *sa;
651 
652 #if IFA_STATS
653 		if (ia)
654 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
655 #endif
656 		/*
657 		 * Always initialize the sum to 0!  Some HW assisted
658 		 * checksumming requires this.
659 		 */
660 		ip->ip_sum = 0;
661 
662 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
663 			/*
664 			 * Perform any checksums that the hardware can't do
665 			 * for us.
666 			 *
667 			 * XXX Does any hardware require the {th,uh}_sum
668 			 * XXX fields to be 0?
669 			 */
670 			if (sw_csum & M_CSUM_IPv4) {
671 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
672 				ip->ip_sum = in_cksum(m, hlen);
673 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
674 			}
675 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
676 				if (IN_NEED_CHECKSUM(ifp,
677 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
678 					in_delayed_cksum(m);
679 				}
680 				m->m_pkthdr.csum_flags &=
681 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
682 			}
683 		}
684 
685 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
686 		if (__predict_true(
687 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
688 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
689 			error = ip_if_output(ifp, m, sa, rt);
690 		} else {
691 			error = ip_tso_output(ifp, m, sa, rt);
692 		}
693 		goto done;
694 	}
695 
696 	/*
697 	 * We can't use HW checksumming if we're about to
698 	 * to fragment the packet.
699 	 *
700 	 * XXX Some hardware can do this.
701 	 */
702 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
703 		if (IN_NEED_CHECKSUM(ifp,
704 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
705 			in_delayed_cksum(m);
706 		}
707 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
708 	}
709 
710 	/*
711 	 * Too large for interface; fragment if possible.
712 	 * Must be able to put at least 8 bytes per fragment.
713 	 */
714 	if (ntohs(ip->ip_off) & IP_DF) {
715 		if (flags & IP_RETURNMTU) {
716 			struct inpcb *inp;
717 
718 			KASSERT(so && solocked(so));
719 			inp = sotoinpcb(so);
720 			inp->inp_errormtu = mtu;
721 		}
722 		error = EMSGSIZE;
723 		IP_STATINC(IP_STAT_CANTFRAG);
724 		goto bad;
725 	}
726 
727 	error = ip_fragment(m, ifp, mtu);
728 	if (error) {
729 		m = NULL;
730 		goto bad;
731 	}
732 
733 	for (; m; m = m0) {
734 		m0 = m->m_nextpkt;
735 		m->m_nextpkt = 0;
736 		if (error) {
737 			m_freem(m);
738 			continue;
739 		}
740 #if IFA_STATS
741 		if (ia)
742 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
743 #endif
744 		/*
745 		 * If we get there, the packet has not been handled by
746 		 * IPsec whereas it should have. Now that it has been
747 		 * fragmented, re-inject it in ip_output so that IPsec
748 		 * processing can occur.
749 		 */
750 		if (natt_frag) {
751 			error = ip_output(m, opt, ro,
752 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
753 			    imo, so);
754 		} else {
755 			KASSERT((m->m_pkthdr.csum_flags &
756 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
757 			error = ip_if_output(ifp, m,
758 			    (m->m_flags & M_MCAST) ?
759 			    sintocsa(rdst) : sintocsa(dst), rt);
760 		}
761 	}
762 	if (error == 0) {
763 		IP_STATINC(IP_STAT_FRAGMENTED);
764 	}
765 done:
766 	ia4_release(ia, &psref_ia);
767 	rtcache_unref(rt, ro);
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_unref(rt, &ro);
1547 		rtcache_free(&ro);
1548 	} else {
1549 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1550 		if (!add && *ifp == NULL)
1551 			return EADDRNOTAVAIL;
1552 	}
1553 	return 0;
1554 }
1555 
1556 /*
1557  * Add a multicast group membership.
1558  * Group must be a valid IP multicast address.
1559  */
1560 static int
1561 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1562 {
1563 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1564 	struct in_addr ia;
1565 	int i, error;
1566 
1567 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1568 		error = ip_get_membership(sopt, &ifp, &ia, true);
1569 	else
1570 #ifdef INET6
1571 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1572 #else
1573 		return EINVAL;
1574 #endif
1575 
1576 	if (error)
1577 		return error;
1578 
1579 	/*
1580 	 * See if we found an interface, and confirm that it
1581 	 * supports multicast.
1582 	 */
1583 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1584 		return EADDRNOTAVAIL;
1585 
1586 	/*
1587 	 * See if the membership already exists or if all the
1588 	 * membership slots are full.
1589 	 */
1590 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1591 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1592 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1593 			break;
1594 	}
1595 	if (i < imo->imo_num_memberships)
1596 		return EADDRINUSE;
1597 
1598 	if (i == IP_MAX_MEMBERSHIPS)
1599 		return ETOOMANYREFS;
1600 
1601 	/*
1602 	 * Everything looks good; add a new record to the multicast
1603 	 * address list for the given interface.
1604 	 */
1605 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1606 		return ENOBUFS;
1607 
1608 	++imo->imo_num_memberships;
1609 	return 0;
1610 }
1611 
1612 /*
1613  * Drop a multicast group membership.
1614  * Group must be a valid IP multicast address.
1615  */
1616 static int
1617 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1618 {
1619 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1620 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1621 	int i, error;
1622 
1623 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1624 		error = ip_get_membership(sopt, &ifp, &ia, false);
1625 	else
1626 #ifdef INET6
1627 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1628 #else
1629 		return EINVAL;
1630 #endif
1631 
1632 	if (error)
1633 		return error;
1634 
1635 	/*
1636 	 * Find the membership in the membership array.
1637 	 */
1638 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1639 		if ((ifp == NULL ||
1640 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1641 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1642 			break;
1643 	}
1644 	if (i == imo->imo_num_memberships)
1645 		return EADDRNOTAVAIL;
1646 
1647 	/*
1648 	 * Give up the multicast address record to which the
1649 	 * membership points.
1650 	 */
1651 	in_delmulti(imo->imo_membership[i]);
1652 
1653 	/*
1654 	 * Remove the gap in the membership array.
1655 	 */
1656 	for (++i; i < imo->imo_num_memberships; ++i)
1657 		imo->imo_membership[i-1] = imo->imo_membership[i];
1658 	--imo->imo_num_memberships;
1659 	return 0;
1660 }
1661 
1662 /*
1663  * Set the IP multicast options in response to user setsockopt().
1664  */
1665 int
1666 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1667 {
1668 	struct ip_moptions *imo = *pimo;
1669 	struct in_addr addr;
1670 	struct ifnet *ifp;
1671 	int ifindex, error = 0;
1672 
1673 	if (!imo) {
1674 		/*
1675 		 * No multicast option buffer attached to the pcb;
1676 		 * allocate one and initialize to default values.
1677 		 */
1678 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1679 		if (imo == NULL)
1680 			return ENOBUFS;
1681 
1682 		imo->imo_multicast_if_index = 0;
1683 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1684 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1685 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1686 		imo->imo_num_memberships = 0;
1687 		*pimo = imo;
1688 	}
1689 
1690 	switch (sopt->sopt_name) {
1691 	case IP_MULTICAST_IF:
1692 		/*
1693 		 * Select the interface for outgoing multicast packets.
1694 		 */
1695 		error = sockopt_get(sopt, &addr, sizeof(addr));
1696 		if (error)
1697 			break;
1698 
1699 		/*
1700 		 * INADDR_ANY is used to remove a previous selection.
1701 		 * When no interface is selected, a default one is
1702 		 * chosen every time a multicast packet is sent.
1703 		 */
1704 		if (in_nullhost(addr)) {
1705 			imo->imo_multicast_if_index = 0;
1706 			break;
1707 		}
1708 		/*
1709 		 * The selected interface is identified by its local
1710 		 * IP address.  Find the interface and confirm that
1711 		 * it supports multicasting.
1712 		 */
1713 		ifp = ip_multicast_if(&addr, &ifindex);
1714 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1715 			error = EADDRNOTAVAIL;
1716 			break;
1717 		}
1718 		imo->imo_multicast_if_index = ifp->if_index;
1719 		if (ifindex)
1720 			imo->imo_multicast_addr = addr;
1721 		else
1722 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1723 		break;
1724 
1725 	case IP_MULTICAST_TTL:
1726 		/*
1727 		 * Set the IP time-to-live for outgoing multicast packets.
1728 		 */
1729 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1730 		break;
1731 
1732 	case IP_MULTICAST_LOOP:
1733 		/*
1734 		 * Set the loopback flag for outgoing multicast packets.
1735 		 * Must be zero or one.
1736 		 */
1737 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1738 		break;
1739 
1740 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1741 		error = ip_add_membership(imo, sopt);
1742 		break;
1743 
1744 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1745 		error = ip_drop_membership(imo, sopt);
1746 		break;
1747 
1748 	default:
1749 		error = EOPNOTSUPP;
1750 		break;
1751 	}
1752 
1753 	/*
1754 	 * If all options have default values, no need to keep the mbuf.
1755 	 */
1756 	if (imo->imo_multicast_if_index == 0 &&
1757 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1758 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1759 	    imo->imo_num_memberships == 0) {
1760 		kmem_free(imo, sizeof(*imo));
1761 		*pimo = NULL;
1762 	}
1763 
1764 	return error;
1765 }
1766 
1767 /*
1768  * Return the IP multicast options in response to user getsockopt().
1769  */
1770 int
1771 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1772 {
1773 	struct in_addr addr;
1774 	uint8_t optval;
1775 	int error = 0;
1776 
1777 	switch (sopt->sopt_name) {
1778 	case IP_MULTICAST_IF:
1779 		if (imo == NULL || imo->imo_multicast_if_index == 0)
1780 			addr = zeroin_addr;
1781 		else if (imo->imo_multicast_addr.s_addr) {
1782 			/* return the value user has set */
1783 			addr = imo->imo_multicast_addr;
1784 		} else {
1785 			struct ifnet *ifp;
1786 			struct in_ifaddr *ia = NULL;
1787 			int s = pserialize_read_enter();
1788 
1789 			ifp = if_byindex(imo->imo_multicast_if_index);
1790 			if (ifp != NULL) {
1791 				ia = in_get_ia_from_ifp(ifp);
1792 			}
1793 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1794 			pserialize_read_exit(s);
1795 		}
1796 		error = sockopt_set(sopt, &addr, sizeof(addr));
1797 		break;
1798 
1799 	case IP_MULTICAST_TTL:
1800 		optval = imo ? imo->imo_multicast_ttl
1801 		    : IP_DEFAULT_MULTICAST_TTL;
1802 
1803 		error = sockopt_set(sopt, &optval, sizeof(optval));
1804 		break;
1805 
1806 	case IP_MULTICAST_LOOP:
1807 		optval = imo ? imo->imo_multicast_loop
1808 		    : IP_DEFAULT_MULTICAST_LOOP;
1809 
1810 		error = sockopt_set(sopt, &optval, sizeof(optval));
1811 		break;
1812 
1813 	default:
1814 		error = EOPNOTSUPP;
1815 	}
1816 
1817 	return error;
1818 }
1819 
1820 /*
1821  * Discard the IP multicast options.
1822  */
1823 void
1824 ip_freemoptions(struct ip_moptions *imo)
1825 {
1826 	int i;
1827 
1828 	if (imo != NULL) {
1829 		for (i = 0; i < imo->imo_num_memberships; ++i)
1830 			in_delmulti(imo->imo_membership[i]);
1831 		kmem_free(imo, sizeof(*imo));
1832 	}
1833 }
1834 
1835 /*
1836  * Routine called from ip_output() to loop back a copy of an IP multicast
1837  * packet to the input queue of a specified interface.  Note that this
1838  * calls the output routine of the loopback "driver", but with an interface
1839  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1840  */
1841 static void
1842 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1843 {
1844 	struct ip *ip;
1845 	struct mbuf *copym;
1846 
1847 	copym = m_copypacket(m, M_DONTWAIT);
1848 	if (copym != NULL &&
1849 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1850 		copym = m_pullup(copym, sizeof(struct ip));
1851 	if (copym == NULL)
1852 		return;
1853 	/*
1854 	 * We don't bother to fragment if the IP length is greater
1855 	 * than the interface's MTU.  Can this possibly matter?
1856 	 */
1857 	ip = mtod(copym, struct ip *);
1858 
1859 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1860 		in_delayed_cksum(copym);
1861 		copym->m_pkthdr.csum_flags &=
1862 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1863 	}
1864 
1865 	ip->ip_sum = 0;
1866 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1867 #ifndef NET_MPSAFE
1868 	KERNEL_LOCK(1, NULL);
1869 #endif
1870 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1871 #ifndef NET_MPSAFE
1872 	KERNEL_UNLOCK_ONE(NULL);
1873 #endif
1874 }
1875 
1876 /*
1877  * Ensure sending address is valid.
1878  * Returns 0 on success, -1 if an error should be sent back or 1
1879  * if the packet could be dropped without error (protocol dependent).
1880  */
1881 static int
1882 ip_ifaddrvalid(const struct in_ifaddr *ia)
1883 {
1884 
1885 	if (ia == NULL)
1886 		return -1;
1887 
1888 	if (ia->ia_addr.sin_addr.s_addr == INADDR_ANY)
1889 		return 0;
1890 
1891 	if (ia->ia4_flags & IN_IFF_DUPLICATED)
1892 		return -1;
1893 	else if (ia->ia4_flags & (IN_IFF_TENTATIVE | IN_IFF_DETACHED))
1894 		return 1;
1895 
1896 	return 0;
1897 }
1898