xref: /netbsd-src/sys/netinet/ip_output.c (revision c34236556bea94afcaca1782d7d228301edc3ea0)
1 /*	$NetBSD: ip_output.c,v 1.265 2016/12/12 03:55:57 ozaki-r Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1998 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Public Access Networks Corporation ("Panix").  It was developed under
38  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
50  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
51  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
52  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
53  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
54  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
55  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
56  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
57  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
58  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
59  * POSSIBILITY OF SUCH DAMAGE.
60  */
61 
62 /*
63  * Copyright (c) 1982, 1986, 1988, 1990, 1993
64  *	The Regents of the University of California.  All rights reserved.
65  *
66  * Redistribution and use in source and binary forms, with or without
67  * modification, are permitted provided that the following conditions
68  * are met:
69  * 1. Redistributions of source code must retain the above copyright
70  *    notice, this list of conditions and the following disclaimer.
71  * 2. Redistributions in binary form must reproduce the above copyright
72  *    notice, this list of conditions and the following disclaimer in the
73  *    documentation and/or other materials provided with the distribution.
74  * 3. Neither the name of the University nor the names of its contributors
75  *    may be used to endorse or promote products derived from this software
76  *    without specific prior written permission.
77  *
78  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88  * SUCH DAMAGE.
89  *
90  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: ip_output.c,v 1.265 2016/12/12 03:55:57 ozaki-r Exp $");
95 
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_mrouting.h"
100 #include "opt_net_mpsafe.h"
101 #include "opt_mpls.h"
102 #endif
103 
104 #include "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 		if (ifa_is_destroying(rt->rt_ifa)) {
365 			rtcache_unref(rt, ro);
366 			IP_STATINC(IP_STAT_NOROUTE);
367 			error = EHOSTUNREACH;
368 			goto bad;
369 		}
370 		ifa_acquire(rt->rt_ifa, &psref_ia);
371 		ia = ifatoia(rt->rt_ifa);
372 		ifp = rt->rt_ifp;
373 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
374 			mtu = ifp->if_mtu;
375 		rt->rt_use++;
376 		if (rt->rt_flags & RTF_GATEWAY)
377 			dst = satosin(rt->rt_gateway);
378 		if (rt->rt_flags & RTF_HOST)
379 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
380 		else
381 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
382 	}
383 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
384 
385 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
386 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
387 		bool inmgroup;
388 
389 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
390 		    M_BCAST : M_MCAST;
391 		/*
392 		 * See if the caller provided any multicast options
393 		 */
394 		if (imo != NULL)
395 			ip->ip_ttl = imo->imo_multicast_ttl;
396 		else
397 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
398 
399 		/*
400 		 * if we don't know the outgoing ifp yet, we can't generate
401 		 * output
402 		 */
403 		if (!ifp) {
404 			IP_STATINC(IP_STAT_NOROUTE);
405 			error = ENETUNREACH;
406 			goto bad;
407 		}
408 
409 		/*
410 		 * If the packet is multicast or broadcast, confirm that
411 		 * the outgoing interface can transmit it.
412 		 */
413 		if (((m->m_flags & M_MCAST) &&
414 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
415 		    ((m->m_flags & M_BCAST) &&
416 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
417 			IP_STATINC(IP_STAT_NOROUTE);
418 			error = ENETUNREACH;
419 			goto bad;
420 		}
421 		/*
422 		 * If source address not specified yet, use an address
423 		 * of outgoing interface.
424 		 */
425 		if (in_nullhost(ip->ip_src)) {
426 			struct in_ifaddr *xia;
427 			struct ifaddr *xifa;
428 			struct psref _psref;
429 
430 			xia = in_get_ia_from_ifp_psref(ifp, &_psref);
431 			if (!xia) {
432 				error = EADDRNOTAVAIL;
433 				goto bad;
434 			}
435 			xifa = &xia->ia_ifa;
436 			if (xifa->ifa_getifa != NULL) {
437 				ia4_release(xia, &_psref);
438 				/* FIXME NOMPSAFE */
439 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
440 				if (xia == NULL) {
441 					error = EADDRNOTAVAIL;
442 					goto bad;
443 				}
444 				ia4_acquire(xia, &_psref);
445 			}
446 			ip->ip_src = xia->ia_addr.sin_addr;
447 			ia4_release(xia, &_psref);
448 		}
449 
450 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
451 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
452 			/*
453 			 * If we belong to the destination multicast group
454 			 * on the outgoing interface, and the caller did not
455 			 * forbid loopback, loop back a copy.
456 			 */
457 			ip_mloopback(ifp, m, &u.dst4);
458 		}
459 #ifdef MROUTING
460 		else {
461 			/*
462 			 * If we are acting as a multicast router, perform
463 			 * multicast forwarding as if the packet had just
464 			 * arrived on the interface to which we are about
465 			 * to send.  The multicast forwarding function
466 			 * recursively calls this function, using the
467 			 * IP_FORWARDING flag to prevent infinite recursion.
468 			 *
469 			 * Multicasts that are looped back by ip_mloopback(),
470 			 * above, will be forwarded by the ip_input() routine,
471 			 * if necessary.
472 			 */
473 			extern struct socket *ip_mrouter;
474 
475 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
476 				if (ip_mforward(m, ifp) != 0) {
477 					m_freem(m);
478 					goto done;
479 				}
480 			}
481 		}
482 #endif
483 		/*
484 		 * Multicasts with a time-to-live of zero may be looped-
485 		 * back, above, but must not be transmitted on a network.
486 		 * Also, multicasts addressed to the loopback interface
487 		 * are not sent -- the above call to ip_mloopback() will
488 		 * loop back a copy if this host actually belongs to the
489 		 * destination group on the loopback interface.
490 		 */
491 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
492 			m_freem(m);
493 			goto done;
494 		}
495 		goto sendit;
496 	}
497 
498 	/*
499 	 * If source address not specified yet, use address
500 	 * of outgoing interface.
501 	 */
502 	if (in_nullhost(ip->ip_src)) {
503 		struct ifaddr *xifa;
504 
505 		xifa = &ia->ia_ifa;
506 		if (xifa->ifa_getifa != NULL) {
507 			ia4_release(ia, &psref_ia);
508 			/* FIXME NOMPSAFE */
509 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
510 			if (ia == NULL) {
511 				error = EADDRNOTAVAIL;
512 				goto bad;
513 			}
514 			ia4_acquire(ia, &psref_ia);
515 		}
516 		ip->ip_src = ia->ia_addr.sin_addr;
517 	}
518 
519 	/*
520 	 * packets with Class-D address as source are not valid per
521 	 * RFC 1112
522 	 */
523 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
524 		IP_STATINC(IP_STAT_ODROPPED);
525 		error = EADDRNOTAVAIL;
526 		goto bad;
527 	}
528 
529 	/*
530 	 * Look for broadcast address and and verify user is allowed to
531 	 * send such a packet.
532 	 */
533 	if (isbroadcast) {
534 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
535 			error = EADDRNOTAVAIL;
536 			goto bad;
537 		}
538 		if ((flags & IP_ALLOWBROADCAST) == 0) {
539 			error = EACCES;
540 			goto bad;
541 		}
542 		/* don't allow broadcast messages to be fragmented */
543 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
544 			error = EMSGSIZE;
545 			goto bad;
546 		}
547 		m->m_flags |= M_BCAST;
548 	} else
549 		m->m_flags &= ~M_BCAST;
550 
551 sendit:
552 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
553 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
554 			ip->ip_id = 0;
555 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
556 			ip->ip_id = ip_newid(ia);
557 		} else {
558 
559 			/*
560 			 * TSO capable interfaces (typically?) increment
561 			 * ip_id for each segment.
562 			 * "allocate" enough ids here to increase the chance
563 			 * for them to be unique.
564 			 *
565 			 * note that the following calculation is not
566 			 * needed to be precise.  wasting some ip_id is fine.
567 			 */
568 
569 			unsigned int segsz = m->m_pkthdr.segsz;
570 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
571 			unsigned int num = howmany(datasz, segsz);
572 
573 			ip->ip_id = ip_newid_range(ia, num);
574 		}
575 	}
576 	if (ia != NULL) {
577 		ia4_release(ia, &psref_ia);
578 		ia = NULL;
579 	}
580 
581 	/*
582 	 * If we're doing Path MTU Discovery, we need to set DF unless
583 	 * the route's MTU is locked.
584 	 */
585 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
586 		ip->ip_off |= htons(IP_DF);
587 	}
588 
589 #ifdef IPSEC
590 	if (ipsec_used) {
591 		bool ipsec_done = false;
592 
593 		/* Perform IPsec processing, if any. */
594 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
595 		    &ipsec_done);
596 		if (error || ipsec_done)
597 			goto done;
598 	}
599 #endif
600 
601 	/*
602 	 * Run through list of hooks for output packets.
603 	 */
604 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
605 	if (error)
606 		goto done;
607 	if (m == NULL)
608 		goto done;
609 
610 	ip = mtod(m, struct ip *);
611 	hlen = ip->ip_hl << 2;
612 
613 	m->m_pkthdr.csum_data |= hlen << 16;
614 
615 	/*
616 	 * search for the source address structure to
617 	 * maintain output statistics.
618 	 */
619 	KASSERT(ia == NULL);
620 	ia = in_get_ia_psref(ip->ip_src, &psref_ia);
621 
622 	/* Ensure we only send from a valid address. */
623 	if ((ia != NULL || (flags & IP_FORWARDING) == 0) &&
624 	    (error = ip_ifaddrvalid(ia)) != 0)
625 	{
626 		arplog(LOG_ERR,
627 		    "refusing to send from invalid address %s (pid %d)\n",
628 		    in_fmtaddr(ip->ip_src), curproc->p_pid);
629 		IP_STATINC(IP_STAT_ODROPPED);
630 		if (error == 1)
631 			/*
632 			 * Address exists, but is tentative or detached.
633 			 * We can't send from it because it's invalid,
634 			 * so we drop the packet.
635 			 */
636 			error = 0;
637 		else
638 			error = EADDRNOTAVAIL;
639 		goto bad;
640 	}
641 
642 	/* Maybe skip checksums on loopback interfaces. */
643 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
644 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
645 	}
646 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
647 	/*
648 	 * If small enough for mtu of path, or if using TCP segmentation
649 	 * offload, can just send directly.
650 	 */
651 	if (ntohs(ip->ip_len) <= mtu ||
652 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
653 		const struct sockaddr *sa;
654 
655 #if IFA_STATS
656 		if (ia)
657 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
658 #endif
659 		/*
660 		 * Always initialize the sum to 0!  Some HW assisted
661 		 * checksumming requires this.
662 		 */
663 		ip->ip_sum = 0;
664 
665 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
666 			/*
667 			 * Perform any checksums that the hardware can't do
668 			 * for us.
669 			 *
670 			 * XXX Does any hardware require the {th,uh}_sum
671 			 * XXX fields to be 0?
672 			 */
673 			if (sw_csum & M_CSUM_IPv4) {
674 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
675 				ip->ip_sum = in_cksum(m, hlen);
676 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
677 			}
678 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
679 				if (IN_NEED_CHECKSUM(ifp,
680 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
681 					in_delayed_cksum(m);
682 				}
683 				m->m_pkthdr.csum_flags &=
684 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
685 			}
686 		}
687 
688 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
689 		if (__predict_true(
690 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
691 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
692 			error = ip_if_output(ifp, m, sa, rt);
693 		} else {
694 			error = ip_tso_output(ifp, m, sa, rt);
695 		}
696 		goto done;
697 	}
698 
699 	/*
700 	 * We can't use HW checksumming if we're about to
701 	 * to fragment the packet.
702 	 *
703 	 * XXX Some hardware can do this.
704 	 */
705 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
706 		if (IN_NEED_CHECKSUM(ifp,
707 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
708 			in_delayed_cksum(m);
709 		}
710 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
711 	}
712 
713 	/*
714 	 * Too large for interface; fragment if possible.
715 	 * Must be able to put at least 8 bytes per fragment.
716 	 */
717 	if (ntohs(ip->ip_off) & IP_DF) {
718 		if (flags & IP_RETURNMTU) {
719 			struct inpcb *inp;
720 
721 			KASSERT(so && solocked(so));
722 			inp = sotoinpcb(so);
723 			inp->inp_errormtu = mtu;
724 		}
725 		error = EMSGSIZE;
726 		IP_STATINC(IP_STAT_CANTFRAG);
727 		goto bad;
728 	}
729 
730 	error = ip_fragment(m, ifp, mtu);
731 	if (error) {
732 		m = NULL;
733 		goto bad;
734 	}
735 
736 	for (; m; m = m0) {
737 		m0 = m->m_nextpkt;
738 		m->m_nextpkt = 0;
739 		if (error) {
740 			m_freem(m);
741 			continue;
742 		}
743 #if IFA_STATS
744 		if (ia)
745 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
746 #endif
747 		/*
748 		 * If we get there, the packet has not been handled by
749 		 * IPsec whereas it should have. Now that it has been
750 		 * fragmented, re-inject it in ip_output so that IPsec
751 		 * processing can occur.
752 		 */
753 		if (natt_frag) {
754 			error = ip_output(m, opt, ro,
755 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
756 			    imo, so);
757 		} else {
758 			KASSERT((m->m_pkthdr.csum_flags &
759 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
760 			error = ip_if_output(ifp, m,
761 			    (m->m_flags & M_MCAST) ?
762 			    sintocsa(rdst) : sintocsa(dst), rt);
763 		}
764 	}
765 	if (error == 0) {
766 		IP_STATINC(IP_STAT_FRAGMENTED);
767 	}
768 done:
769 	ia4_release(ia, &psref_ia);
770 	rtcache_unref(rt, ro);
771 	if (ro == &iproute) {
772 		rtcache_free(&iproute);
773 	}
774 #ifdef IPSEC
775 	if (sp) {
776 		KEY_FREESP(&sp);
777 	}
778 #endif
779 	if (mifp != NULL) {
780 		if_put(mifp, &psref);
781 	}
782 	if (bind_need_restore)
783 		curlwp_bindx(bound);
784 	return error;
785 bad:
786 	m_freem(m);
787 	goto done;
788 }
789 
790 int
791 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
792 {
793 	struct ip *ip, *mhip;
794 	struct mbuf *m0;
795 	int len, hlen, off;
796 	int mhlen, firstlen;
797 	struct mbuf **mnext;
798 	int sw_csum = m->m_pkthdr.csum_flags;
799 	int fragments = 0;
800 	int s;
801 	int error = 0;
802 
803 	ip = mtod(m, struct ip *);
804 	hlen = ip->ip_hl << 2;
805 	if (ifp != NULL)
806 		sw_csum &= ~ifp->if_csum_flags_tx;
807 
808 	len = (mtu - hlen) &~ 7;
809 	if (len < 8) {
810 		m_freem(m);
811 		return (EMSGSIZE);
812 	}
813 
814 	firstlen = len;
815 	mnext = &m->m_nextpkt;
816 
817 	/*
818 	 * Loop through length of segment after first fragment,
819 	 * make new header and copy data of each part and link onto chain.
820 	 */
821 	m0 = m;
822 	mhlen = sizeof (struct ip);
823 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
824 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
825 		if (m == 0) {
826 			error = ENOBUFS;
827 			IP_STATINC(IP_STAT_ODROPPED);
828 			goto sendorfree;
829 		}
830 		MCLAIM(m, m0->m_owner);
831 		*mnext = m;
832 		mnext = &m->m_nextpkt;
833 		m->m_data += max_linkhdr;
834 		mhip = mtod(m, struct ip *);
835 		*mhip = *ip;
836 		/* we must inherit MCAST and BCAST flags */
837 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
838 		if (hlen > sizeof (struct ip)) {
839 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
840 			mhip->ip_hl = mhlen >> 2;
841 		}
842 		m->m_len = mhlen;
843 		mhip->ip_off = ((off - hlen) >> 3) +
844 		    (ntohs(ip->ip_off) & ~IP_MF);
845 		if (ip->ip_off & htons(IP_MF))
846 			mhip->ip_off |= IP_MF;
847 		if (off + len >= ntohs(ip->ip_len))
848 			len = ntohs(ip->ip_len) - off;
849 		else
850 			mhip->ip_off |= IP_MF;
851 		HTONS(mhip->ip_off);
852 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
853 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
854 		if (m->m_next == 0) {
855 			error = ENOBUFS;	/* ??? */
856 			IP_STATINC(IP_STAT_ODROPPED);
857 			goto sendorfree;
858 		}
859 		m->m_pkthdr.len = mhlen + len;
860 		m_reset_rcvif(m);
861 		mhip->ip_sum = 0;
862 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
863 		if (sw_csum & M_CSUM_IPv4) {
864 			mhip->ip_sum = in_cksum(m, mhlen);
865 		} else {
866 			/*
867 			 * checksum is hw-offloaded or not necessary.
868 			 */
869 			m->m_pkthdr.csum_flags |=
870 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
871 			m->m_pkthdr.csum_data |= mhlen << 16;
872 			KASSERT(!(ifp != NULL &&
873 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
874 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
875 		}
876 		IP_STATINC(IP_STAT_OFRAGMENTS);
877 		fragments++;
878 	}
879 	/*
880 	 * Update first fragment by trimming what's been copied out
881 	 * and updating header, then send each fragment (in order).
882 	 */
883 	m = m0;
884 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
885 	m->m_pkthdr.len = hlen + firstlen;
886 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
887 	ip->ip_off |= htons(IP_MF);
888 	ip->ip_sum = 0;
889 	if (sw_csum & M_CSUM_IPv4) {
890 		ip->ip_sum = in_cksum(m, hlen);
891 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
892 	} else {
893 		/*
894 		 * checksum is hw-offloaded or not necessary.
895 		 */
896 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
897 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
898 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
899 		    sizeof(struct ip));
900 	}
901 sendorfree:
902 	/*
903 	 * If there is no room for all the fragments, don't queue
904 	 * any of them.
905 	 */
906 	if (ifp != NULL) {
907 		s = splnet();
908 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
909 		    error == 0) {
910 			error = ENOBUFS;
911 			IP_STATINC(IP_STAT_ODROPPED);
912 			IFQ_INC_DROPS(&ifp->if_snd);
913 		}
914 		splx(s);
915 	}
916 	if (error) {
917 		for (m = m0; m; m = m0) {
918 			m0 = m->m_nextpkt;
919 			m->m_nextpkt = NULL;
920 			m_freem(m);
921 		}
922 	}
923 	return (error);
924 }
925 
926 /*
927  * Process a delayed payload checksum calculation.
928  */
929 void
930 in_delayed_cksum(struct mbuf *m)
931 {
932 	struct ip *ip;
933 	u_int16_t csum, offset;
934 
935 	ip = mtod(m, struct ip *);
936 	offset = ip->ip_hl << 2;
937 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
938 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
939 		csum = 0xffff;
940 
941 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
942 
943 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
944 		/* This happen when ip options were inserted
945 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
946 		    m->m_len, offset, ip->ip_p);
947 		 */
948 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
949 	} else
950 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
951 }
952 
953 /*
954  * Determine the maximum length of the options to be inserted;
955  * we would far rather allocate too much space rather than too little.
956  */
957 
958 u_int
959 ip_optlen(struct inpcb *inp)
960 {
961 	struct mbuf *m = inp->inp_options;
962 
963 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
964 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
965 	}
966 	return 0;
967 }
968 
969 /*
970  * Insert IP options into preformed packet.
971  * Adjust IP destination as required for IP source routing,
972  * as indicated by a non-zero in_addr at the start of the options.
973  */
974 static struct mbuf *
975 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
976 {
977 	struct ipoption *p = mtod(opt, struct ipoption *);
978 	struct mbuf *n;
979 	struct ip *ip = mtod(m, struct ip *);
980 	unsigned optlen;
981 
982 	optlen = opt->m_len - sizeof(p->ipopt_dst);
983 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
984 		return (m);		/* XXX should fail */
985 	if (!in_nullhost(p->ipopt_dst))
986 		ip->ip_dst = p->ipopt_dst;
987 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
988 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
989 		if (n == 0)
990 			return (m);
991 		MCLAIM(n, m->m_owner);
992 		M_MOVE_PKTHDR(n, m);
993 		m->m_len -= sizeof(struct ip);
994 		m->m_data += sizeof(struct ip);
995 		n->m_next = m;
996 		m = n;
997 		m->m_len = optlen + sizeof(struct ip);
998 		m->m_data += max_linkhdr;
999 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
1000 	} else {
1001 		m->m_data -= optlen;
1002 		m->m_len += optlen;
1003 		memmove(mtod(m, void *), ip, sizeof(struct ip));
1004 	}
1005 	m->m_pkthdr.len += optlen;
1006 	ip = mtod(m, struct ip *);
1007 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
1008 	*phlen = sizeof(struct ip) + optlen;
1009 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
1010 	return (m);
1011 }
1012 
1013 /*
1014  * Copy options from ip to jp,
1015  * omitting those not copied during fragmentation.
1016  */
1017 int
1018 ip_optcopy(struct ip *ip, struct ip *jp)
1019 {
1020 	u_char *cp, *dp;
1021 	int opt, optlen, cnt;
1022 
1023 	cp = (u_char *)(ip + 1);
1024 	dp = (u_char *)(jp + 1);
1025 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1026 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1027 		opt = cp[0];
1028 		if (opt == IPOPT_EOL)
1029 			break;
1030 		if (opt == IPOPT_NOP) {
1031 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1032 			*dp++ = IPOPT_NOP;
1033 			optlen = 1;
1034 			continue;
1035 		}
1036 
1037 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
1038 		optlen = cp[IPOPT_OLEN];
1039 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
1040 
1041 		/* Invalid lengths should have been caught by ip_dooptions. */
1042 		if (optlen > cnt)
1043 			optlen = cnt;
1044 		if (IPOPT_COPIED(opt)) {
1045 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
1046 			dp += optlen;
1047 		}
1048 	}
1049 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1050 		*dp++ = IPOPT_EOL;
1051 	return (optlen);
1052 }
1053 
1054 /*
1055  * IP socket option processing.
1056  */
1057 int
1058 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1059 {
1060 	struct inpcb *inp = sotoinpcb(so);
1061 	struct ip *ip = &inp->inp_ip;
1062 	int inpflags = inp->inp_flags;
1063 	int optval = 0, error = 0;
1064 
1065 	if (sopt->sopt_level != IPPROTO_IP) {
1066 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1067 			return 0;
1068 		return ENOPROTOOPT;
1069 	}
1070 
1071 	switch (op) {
1072 	case PRCO_SETOPT:
1073 		switch (sopt->sopt_name) {
1074 		case IP_OPTIONS:
1075 #ifdef notyet
1076 		case IP_RETOPTS:
1077 #endif
1078 			error = ip_pcbopts(inp, sopt);
1079 			break;
1080 
1081 		case IP_TOS:
1082 		case IP_TTL:
1083 		case IP_MINTTL:
1084 		case IP_PKTINFO:
1085 		case IP_RECVOPTS:
1086 		case IP_RECVRETOPTS:
1087 		case IP_RECVDSTADDR:
1088 		case IP_RECVIF:
1089 		case IP_RECVPKTINFO:
1090 		case IP_RECVTTL:
1091 			error = sockopt_getint(sopt, &optval);
1092 			if (error)
1093 				break;
1094 
1095 			switch (sopt->sopt_name) {
1096 			case IP_TOS:
1097 				ip->ip_tos = optval;
1098 				break;
1099 
1100 			case IP_TTL:
1101 				ip->ip_ttl = optval;
1102 				break;
1103 
1104 			case IP_MINTTL:
1105 				if (optval > 0 && optval <= MAXTTL)
1106 					inp->inp_ip_minttl = optval;
1107 				else
1108 					error = EINVAL;
1109 				break;
1110 #define	OPTSET(bit) \
1111 	if (optval) \
1112 		inpflags |= bit; \
1113 	else \
1114 		inpflags &= ~bit;
1115 
1116 			case IP_PKTINFO:
1117 				OPTSET(INP_PKTINFO);
1118 				break;
1119 
1120 			case IP_RECVOPTS:
1121 				OPTSET(INP_RECVOPTS);
1122 				break;
1123 
1124 			case IP_RECVPKTINFO:
1125 				OPTSET(INP_RECVPKTINFO);
1126 				break;
1127 
1128 			case IP_RECVRETOPTS:
1129 				OPTSET(INP_RECVRETOPTS);
1130 				break;
1131 
1132 			case IP_RECVDSTADDR:
1133 				OPTSET(INP_RECVDSTADDR);
1134 				break;
1135 
1136 			case IP_RECVIF:
1137 				OPTSET(INP_RECVIF);
1138 				break;
1139 
1140 			case IP_RECVTTL:
1141 				OPTSET(INP_RECVTTL);
1142 				break;
1143 			}
1144 		break;
1145 #undef OPTSET
1146 
1147 		case IP_MULTICAST_IF:
1148 		case IP_MULTICAST_TTL:
1149 		case IP_MULTICAST_LOOP:
1150 		case IP_ADD_MEMBERSHIP:
1151 		case IP_DROP_MEMBERSHIP:
1152 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1153 			break;
1154 
1155 		case IP_PORTRANGE:
1156 			error = sockopt_getint(sopt, &optval);
1157 			if (error)
1158 				break;
1159 
1160 			switch (optval) {
1161 			case IP_PORTRANGE_DEFAULT:
1162 			case IP_PORTRANGE_HIGH:
1163 				inpflags &= ~(INP_LOWPORT);
1164 				break;
1165 
1166 			case IP_PORTRANGE_LOW:
1167 				inpflags |= INP_LOWPORT;
1168 				break;
1169 
1170 			default:
1171 				error = EINVAL;
1172 				break;
1173 			}
1174 			break;
1175 
1176 		case IP_PORTALGO:
1177 			error = sockopt_getint(sopt, &optval);
1178 			if (error)
1179 				break;
1180 
1181 			error = portalgo_algo_index_select(
1182 			    (struct inpcb_hdr *)inp, optval);
1183 			break;
1184 
1185 #if defined(IPSEC)
1186 		case IP_IPSEC_POLICY:
1187 			if (ipsec_enabled) {
1188 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1189 				    sopt->sopt_data, sopt->sopt_size,
1190 				    curlwp->l_cred);
1191 				break;
1192 			}
1193 			/*FALLTHROUGH*/
1194 #endif /* IPSEC */
1195 
1196 		default:
1197 			error = ENOPROTOOPT;
1198 			break;
1199 		}
1200 		break;
1201 
1202 	case PRCO_GETOPT:
1203 		switch (sopt->sopt_name) {
1204 		case IP_OPTIONS:
1205 		case IP_RETOPTS: {
1206 			struct mbuf *mopts = inp->inp_options;
1207 
1208 			if (mopts) {
1209 				struct mbuf *m;
1210 
1211 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1212 				if (m == NULL) {
1213 					error = ENOBUFS;
1214 					break;
1215 				}
1216 				error = sockopt_setmbuf(sopt, m);
1217 			}
1218 			break;
1219 		}
1220 		case IP_PKTINFO:
1221 		case IP_TOS:
1222 		case IP_TTL:
1223 		case IP_MINTTL:
1224 		case IP_RECVOPTS:
1225 		case IP_RECVRETOPTS:
1226 		case IP_RECVDSTADDR:
1227 		case IP_RECVIF:
1228 		case IP_RECVPKTINFO:
1229 		case IP_RECVTTL:
1230 		case IP_ERRORMTU:
1231 			switch (sopt->sopt_name) {
1232 			case IP_TOS:
1233 				optval = ip->ip_tos;
1234 				break;
1235 
1236 			case IP_TTL:
1237 				optval = ip->ip_ttl;
1238 				break;
1239 
1240 			case IP_MINTTL:
1241 				optval = inp->inp_ip_minttl;
1242 				break;
1243 
1244 			case IP_ERRORMTU:
1245 				optval = inp->inp_errormtu;
1246 				break;
1247 
1248 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1249 
1250 			case IP_PKTINFO:
1251 				optval = OPTBIT(INP_PKTINFO);
1252 				break;
1253 
1254 			case IP_RECVOPTS:
1255 				optval = OPTBIT(INP_RECVOPTS);
1256 				break;
1257 
1258 			case IP_RECVPKTINFO:
1259 				optval = OPTBIT(INP_RECVPKTINFO);
1260 				break;
1261 
1262 			case IP_RECVRETOPTS:
1263 				optval = OPTBIT(INP_RECVRETOPTS);
1264 				break;
1265 
1266 			case IP_RECVDSTADDR:
1267 				optval = OPTBIT(INP_RECVDSTADDR);
1268 				break;
1269 
1270 			case IP_RECVIF:
1271 				optval = OPTBIT(INP_RECVIF);
1272 				break;
1273 
1274 			case IP_RECVTTL:
1275 				optval = OPTBIT(INP_RECVTTL);
1276 				break;
1277 			}
1278 			error = sockopt_setint(sopt, optval);
1279 			break;
1280 
1281 #if 0	/* defined(IPSEC) */
1282 		case IP_IPSEC_POLICY:
1283 		{
1284 			struct mbuf *m = NULL;
1285 
1286 			/* XXX this will return EINVAL as sopt is empty */
1287 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1288 			    sopt->sopt_size, &m);
1289 			if (error == 0)
1290 				error = sockopt_setmbuf(sopt, m);
1291 			break;
1292 		}
1293 #endif /*IPSEC*/
1294 
1295 		case IP_MULTICAST_IF:
1296 		case IP_MULTICAST_TTL:
1297 		case IP_MULTICAST_LOOP:
1298 		case IP_ADD_MEMBERSHIP:
1299 		case IP_DROP_MEMBERSHIP:
1300 			error = ip_getmoptions(inp->inp_moptions, sopt);
1301 			break;
1302 
1303 		case IP_PORTRANGE:
1304 			if (inpflags & INP_LOWPORT)
1305 				optval = IP_PORTRANGE_LOW;
1306 			else
1307 				optval = IP_PORTRANGE_DEFAULT;
1308 			error = sockopt_setint(sopt, optval);
1309 			break;
1310 
1311 		case IP_PORTALGO:
1312 			optval = inp->inp_portalgo;
1313 			error = sockopt_setint(sopt, optval);
1314 			break;
1315 
1316 		default:
1317 			error = ENOPROTOOPT;
1318 			break;
1319 		}
1320 		break;
1321 	}
1322 
1323 	if (!error) {
1324 		inp->inp_flags = inpflags;
1325 	}
1326 	return error;
1327 }
1328 
1329 /*
1330  * Set up IP options in pcb for insertion in output packets.
1331  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1332  * with destination address if source routed.
1333  */
1334 static int
1335 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1336 {
1337 	struct mbuf *m;
1338 	const u_char *cp;
1339 	u_char *dp;
1340 	int cnt;
1341 
1342 	/* Turn off any old options. */
1343 	if (inp->inp_options) {
1344 		m_free(inp->inp_options);
1345 	}
1346 	inp->inp_options = NULL;
1347 	if ((cnt = sopt->sopt_size) == 0) {
1348 		/* Only turning off any previous options. */
1349 		return 0;
1350 	}
1351 	cp = sopt->sopt_data;
1352 
1353 #ifndef	__vax__
1354 	if (cnt % sizeof(int32_t))
1355 		return (EINVAL);
1356 #endif
1357 
1358 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1359 	if (m == NULL)
1360 		return (ENOBUFS);
1361 
1362 	dp = mtod(m, u_char *);
1363 	memset(dp, 0, sizeof(struct in_addr));
1364 	dp += sizeof(struct in_addr);
1365 	m->m_len = sizeof(struct in_addr);
1366 
1367 	/*
1368 	 * IP option list according to RFC791. Each option is of the form
1369 	 *
1370 	 *	[optval] [olen] [(olen - 2) data bytes]
1371 	 *
1372 	 * We validate the list and copy options to an mbuf for prepending
1373 	 * to data packets. The IP first-hop destination address will be
1374 	 * stored before actual options and is zero if unset.
1375 	 */
1376 	while (cnt > 0) {
1377 		uint8_t optval, olen, offset;
1378 
1379 		optval = cp[IPOPT_OPTVAL];
1380 
1381 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1382 			olen = 1;
1383 		} else {
1384 			if (cnt < IPOPT_OLEN + 1)
1385 				goto bad;
1386 
1387 			olen = cp[IPOPT_OLEN];
1388 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1389 				goto bad;
1390 		}
1391 
1392 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1393 			/*
1394 			 * user process specifies route as:
1395 			 *	->A->B->C->D
1396 			 * D must be our final destination (but we can't
1397 			 * check that since we may not have connected yet).
1398 			 * A is first hop destination, which doesn't appear in
1399 			 * actual IP option, but is stored before the options.
1400 			 */
1401 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1402 				goto bad;
1403 
1404 			offset = cp[IPOPT_OFFSET];
1405 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1406 			    sizeof(struct in_addr));
1407 
1408 			cp += sizeof(struct in_addr);
1409 			cnt -= sizeof(struct in_addr);
1410 			olen -= sizeof(struct in_addr);
1411 
1412 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1413 				goto bad;
1414 
1415 			memcpy(dp, cp, olen);
1416 			dp[IPOPT_OPTVAL] = optval;
1417 			dp[IPOPT_OLEN] = olen;
1418 			dp[IPOPT_OFFSET] = offset;
1419 			break;
1420 		} else {
1421 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1422 				goto bad;
1423 
1424 			memcpy(dp, cp, olen);
1425 			break;
1426 		}
1427 
1428 		dp += olen;
1429 		m->m_len += olen;
1430 
1431 		if (optval == IPOPT_EOL)
1432 			break;
1433 
1434 		cp += olen;
1435 		cnt -= olen;
1436 	}
1437 
1438 	inp->inp_options = m;
1439 	return 0;
1440 bad:
1441 	(void)m_free(m);
1442 	return EINVAL;
1443 }
1444 
1445 /*
1446  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1447  */
1448 static struct ifnet *
1449 ip_multicast_if(struct in_addr *a, int *ifindexp)
1450 {
1451 	int ifindex;
1452 	struct ifnet *ifp = NULL;
1453 	struct in_ifaddr *ia;
1454 
1455 	if (ifindexp)
1456 		*ifindexp = 0;
1457 	if (ntohl(a->s_addr) >> 24 == 0) {
1458 		ifindex = ntohl(a->s_addr) & 0xffffff;
1459 		ifp = if_byindex(ifindex);
1460 		if (!ifp)
1461 			return NULL;
1462 		if (ifindexp)
1463 			*ifindexp = ifindex;
1464 	} else {
1465 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1466 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1467 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1468 				ifp = ia->ia_ifp;
1469 				break;
1470 			}
1471 		}
1472 	}
1473 	return ifp;
1474 }
1475 
1476 static int
1477 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1478 {
1479 	u_int tval;
1480 	u_char cval;
1481 	int error;
1482 
1483 	if (sopt == NULL)
1484 		return EINVAL;
1485 
1486 	switch (sopt->sopt_size) {
1487 	case sizeof(u_char):
1488 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1489 		tval = cval;
1490 		break;
1491 
1492 	case sizeof(u_int):
1493 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1494 		break;
1495 
1496 	default:
1497 		error = EINVAL;
1498 	}
1499 
1500 	if (error)
1501 		return error;
1502 
1503 	if (tval > maxval)
1504 		return EINVAL;
1505 
1506 	*val = tval;
1507 	return 0;
1508 }
1509 
1510 static int
1511 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1512     struct in_addr *ia, bool add)
1513 {
1514 	int error;
1515 	struct ip_mreq mreq;
1516 
1517 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1518 	if (error)
1519 		return error;
1520 
1521 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1522 		return EINVAL;
1523 
1524 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1525 
1526 	if (in_nullhost(mreq.imr_interface)) {
1527 		union {
1528 			struct sockaddr		dst;
1529 			struct sockaddr_in	dst4;
1530 		} u;
1531 		struct route ro;
1532 
1533 		if (!add) {
1534 			*ifp = NULL;
1535 			return 0;
1536 		}
1537 		/*
1538 		 * If no interface address was provided, use the interface of
1539 		 * the route to the given multicast address.
1540 		 */
1541 		struct rtentry *rt;
1542 		memset(&ro, 0, sizeof(ro));
1543 
1544 		sockaddr_in_init(&u.dst4, ia, 0);
1545 		error = rtcache_setdst(&ro, &u.dst);
1546 		if (error != 0)
1547 			return error;
1548 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1549 		rtcache_unref(rt, &ro);
1550 		rtcache_free(&ro);
1551 	} else {
1552 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1553 		if (!add && *ifp == NULL)
1554 			return EADDRNOTAVAIL;
1555 	}
1556 	return 0;
1557 }
1558 
1559 /*
1560  * Add a multicast group membership.
1561  * Group must be a valid IP multicast address.
1562  */
1563 static int
1564 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1565 {
1566 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1567 	struct in_addr ia;
1568 	int i, error;
1569 
1570 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1571 		error = ip_get_membership(sopt, &ifp, &ia, true);
1572 	else
1573 #ifdef INET6
1574 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1575 #else
1576 		return EINVAL;
1577 #endif
1578 
1579 	if (error)
1580 		return error;
1581 
1582 	/*
1583 	 * See if we found an interface, and confirm that it
1584 	 * supports multicast.
1585 	 */
1586 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1587 		return EADDRNOTAVAIL;
1588 
1589 	/*
1590 	 * See if the membership already exists or if all the
1591 	 * membership slots are full.
1592 	 */
1593 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1594 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1595 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1596 			break;
1597 	}
1598 	if (i < imo->imo_num_memberships)
1599 		return EADDRINUSE;
1600 
1601 	if (i == IP_MAX_MEMBERSHIPS)
1602 		return ETOOMANYREFS;
1603 
1604 	/*
1605 	 * Everything looks good; add a new record to the multicast
1606 	 * address list for the given interface.
1607 	 */
1608 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1609 		return ENOBUFS;
1610 
1611 	++imo->imo_num_memberships;
1612 	return 0;
1613 }
1614 
1615 /*
1616  * Drop a multicast group membership.
1617  * Group must be a valid IP multicast address.
1618  */
1619 static int
1620 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1621 {
1622 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1623 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1624 	int i, error;
1625 
1626 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1627 		error = ip_get_membership(sopt, &ifp, &ia, false);
1628 	else
1629 #ifdef INET6
1630 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1631 #else
1632 		return EINVAL;
1633 #endif
1634 
1635 	if (error)
1636 		return error;
1637 
1638 	/*
1639 	 * Find the membership in the membership array.
1640 	 */
1641 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1642 		if ((ifp == NULL ||
1643 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1644 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1645 			break;
1646 	}
1647 	if (i == imo->imo_num_memberships)
1648 		return EADDRNOTAVAIL;
1649 
1650 	/*
1651 	 * Give up the multicast address record to which the
1652 	 * membership points.
1653 	 */
1654 	in_delmulti(imo->imo_membership[i]);
1655 
1656 	/*
1657 	 * Remove the gap in the membership array.
1658 	 */
1659 	for (++i; i < imo->imo_num_memberships; ++i)
1660 		imo->imo_membership[i-1] = imo->imo_membership[i];
1661 	--imo->imo_num_memberships;
1662 	return 0;
1663 }
1664 
1665 /*
1666  * Set the IP multicast options in response to user setsockopt().
1667  */
1668 int
1669 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1670 {
1671 	struct ip_moptions *imo = *pimo;
1672 	struct in_addr addr;
1673 	struct ifnet *ifp;
1674 	int ifindex, error = 0;
1675 
1676 	if (!imo) {
1677 		/*
1678 		 * No multicast option buffer attached to the pcb;
1679 		 * allocate one and initialize to default values.
1680 		 */
1681 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1682 		if (imo == NULL)
1683 			return ENOBUFS;
1684 
1685 		imo->imo_multicast_if_index = 0;
1686 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1687 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1688 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1689 		imo->imo_num_memberships = 0;
1690 		*pimo = imo;
1691 	}
1692 
1693 	switch (sopt->sopt_name) {
1694 	case IP_MULTICAST_IF:
1695 		/*
1696 		 * Select the interface for outgoing multicast packets.
1697 		 */
1698 		error = sockopt_get(sopt, &addr, sizeof(addr));
1699 		if (error)
1700 			break;
1701 
1702 		/*
1703 		 * INADDR_ANY is used to remove a previous selection.
1704 		 * When no interface is selected, a default one is
1705 		 * chosen every time a multicast packet is sent.
1706 		 */
1707 		if (in_nullhost(addr)) {
1708 			imo->imo_multicast_if_index = 0;
1709 			break;
1710 		}
1711 		/*
1712 		 * The selected interface is identified by its local
1713 		 * IP address.  Find the interface and confirm that
1714 		 * it supports multicasting.
1715 		 */
1716 		ifp = ip_multicast_if(&addr, &ifindex);
1717 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1718 			error = EADDRNOTAVAIL;
1719 			break;
1720 		}
1721 		imo->imo_multicast_if_index = ifp->if_index;
1722 		if (ifindex)
1723 			imo->imo_multicast_addr = addr;
1724 		else
1725 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1726 		break;
1727 
1728 	case IP_MULTICAST_TTL:
1729 		/*
1730 		 * Set the IP time-to-live for outgoing multicast packets.
1731 		 */
1732 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1733 		break;
1734 
1735 	case IP_MULTICAST_LOOP:
1736 		/*
1737 		 * Set the loopback flag for outgoing multicast packets.
1738 		 * Must be zero or one.
1739 		 */
1740 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1741 		break;
1742 
1743 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1744 		error = ip_add_membership(imo, sopt);
1745 		break;
1746 
1747 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1748 		error = ip_drop_membership(imo, sopt);
1749 		break;
1750 
1751 	default:
1752 		error = EOPNOTSUPP;
1753 		break;
1754 	}
1755 
1756 	/*
1757 	 * If all options have default values, no need to keep the mbuf.
1758 	 */
1759 	if (imo->imo_multicast_if_index == 0 &&
1760 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1761 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1762 	    imo->imo_num_memberships == 0) {
1763 		kmem_free(imo, sizeof(*imo));
1764 		*pimo = NULL;
1765 	}
1766 
1767 	return error;
1768 }
1769 
1770 /*
1771  * Return the IP multicast options in response to user getsockopt().
1772  */
1773 int
1774 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1775 {
1776 	struct in_addr addr;
1777 	uint8_t optval;
1778 	int error = 0;
1779 
1780 	switch (sopt->sopt_name) {
1781 	case IP_MULTICAST_IF:
1782 		if (imo == NULL || imo->imo_multicast_if_index == 0)
1783 			addr = zeroin_addr;
1784 		else if (imo->imo_multicast_addr.s_addr) {
1785 			/* return the value user has set */
1786 			addr = imo->imo_multicast_addr;
1787 		} else {
1788 			struct ifnet *ifp;
1789 			struct in_ifaddr *ia = NULL;
1790 			int s = pserialize_read_enter();
1791 
1792 			ifp = if_byindex(imo->imo_multicast_if_index);
1793 			if (ifp != NULL) {
1794 				ia = in_get_ia_from_ifp(ifp);
1795 			}
1796 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1797 			pserialize_read_exit(s);
1798 		}
1799 		error = sockopt_set(sopt, &addr, sizeof(addr));
1800 		break;
1801 
1802 	case IP_MULTICAST_TTL:
1803 		optval = imo ? imo->imo_multicast_ttl
1804 		    : IP_DEFAULT_MULTICAST_TTL;
1805 
1806 		error = sockopt_set(sopt, &optval, sizeof(optval));
1807 		break;
1808 
1809 	case IP_MULTICAST_LOOP:
1810 		optval = imo ? imo->imo_multicast_loop
1811 		    : IP_DEFAULT_MULTICAST_LOOP;
1812 
1813 		error = sockopt_set(sopt, &optval, sizeof(optval));
1814 		break;
1815 
1816 	default:
1817 		error = EOPNOTSUPP;
1818 	}
1819 
1820 	return error;
1821 }
1822 
1823 /*
1824  * Discard the IP multicast options.
1825  */
1826 void
1827 ip_freemoptions(struct ip_moptions *imo)
1828 {
1829 	int i;
1830 
1831 	if (imo != NULL) {
1832 		for (i = 0; i < imo->imo_num_memberships; ++i)
1833 			in_delmulti(imo->imo_membership[i]);
1834 		kmem_free(imo, sizeof(*imo));
1835 	}
1836 }
1837 
1838 /*
1839  * Routine called from ip_output() to loop back a copy of an IP multicast
1840  * packet to the input queue of a specified interface.  Note that this
1841  * calls the output routine of the loopback "driver", but with an interface
1842  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1843  */
1844 static void
1845 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1846 {
1847 	struct ip *ip;
1848 	struct mbuf *copym;
1849 
1850 	copym = m_copypacket(m, M_DONTWAIT);
1851 	if (copym != NULL &&
1852 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1853 		copym = m_pullup(copym, sizeof(struct ip));
1854 	if (copym == NULL)
1855 		return;
1856 	/*
1857 	 * We don't bother to fragment if the IP length is greater
1858 	 * than the interface's MTU.  Can this possibly matter?
1859 	 */
1860 	ip = mtod(copym, struct ip *);
1861 
1862 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1863 		in_delayed_cksum(copym);
1864 		copym->m_pkthdr.csum_flags &=
1865 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1866 	}
1867 
1868 	ip->ip_sum = 0;
1869 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1870 #ifndef NET_MPSAFE
1871 	KERNEL_LOCK(1, NULL);
1872 #endif
1873 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1874 #ifndef NET_MPSAFE
1875 	KERNEL_UNLOCK_ONE(NULL);
1876 #endif
1877 }
1878 
1879 /*
1880  * Ensure sending address is valid.
1881  * Returns 0 on success, -1 if an error should be sent back or 1
1882  * if the packet could be dropped without error (protocol dependent).
1883  */
1884 static int
1885 ip_ifaddrvalid(const struct in_ifaddr *ia)
1886 {
1887 
1888 	if (ia == NULL)
1889 		return -1;
1890 
1891 	if (ia->ia_addr.sin_addr.s_addr == INADDR_ANY)
1892 		return 0;
1893 
1894 	if (ia->ia4_flags & IN_IFF_DUPLICATED)
1895 		return -1;
1896 	else if (ia->ia4_flags & (IN_IFF_TENTATIVE | IN_IFF_DETACHED))
1897 		return 1;
1898 
1899 	return 0;
1900 }
1901