xref: /netbsd-src/sys/netinet/ip_output.c (revision fdd524d4ccd2bb0c6f67401e938dabf773eb0372)
1 /*	$NetBSD: ip_output.c,v 1.259 2016/07/08 04:33:30 ozaki-r Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1998 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Public Access Networks Corporation ("Panix").  It was developed under
38  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
50  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
51  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
52  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
53  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
54  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
55  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
56  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
57  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
58  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
59  * POSSIBILITY OF SUCH DAMAGE.
60  */
61 
62 /*
63  * Copyright (c) 1982, 1986, 1988, 1990, 1993
64  *	The Regents of the University of California.  All rights reserved.
65  *
66  * Redistribution and use in source and binary forms, with or without
67  * modification, are permitted provided that the following conditions
68  * are met:
69  * 1. Redistributions of source code must retain the above copyright
70  *    notice, this list of conditions and the following disclaimer.
71  * 2. Redistributions in binary form must reproduce the above copyright
72  *    notice, this list of conditions and the following disclaimer in the
73  *    documentation and/or other materials provided with the distribution.
74  * 3. Neither the name of the University nor the names of its contributors
75  *    may be used to endorse or promote products derived from this software
76  *    without specific prior written permission.
77  *
78  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88  * SUCH DAMAGE.
89  *
90  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: ip_output.c,v 1.259 2016/07/08 04:33:30 ozaki-r Exp $");
95 
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_mrouting.h"
100 #include "opt_net_mpsafe.h"
101 #include "opt_mpls.h"
102 #endif
103 
104 #include <sys/param.h>
105 #include <sys/kmem.h>
106 #include <sys/mbuf.h>
107 #include <sys/protosw.h>
108 #include <sys/socket.h>
109 #include <sys/socketvar.h>
110 #include <sys/kauth.h>
111 #ifdef IPSEC
112 #include <sys/domain.h>
113 #endif
114 #include <sys/systm.h>
115 
116 #include <net/if.h>
117 #include <net/if_types.h>
118 #include <net/route.h>
119 #include <net/pfil.h>
120 
121 #include <netinet/in.h>
122 #include <netinet/in_systm.h>
123 #include <netinet/ip.h>
124 #include <netinet/in_pcb.h>
125 #include <netinet/in_var.h>
126 #include <netinet/ip_var.h>
127 #include <netinet/ip_private.h>
128 #include <netinet/in_offload.h>
129 #include <netinet/portalgo.h>
130 #include <netinet/udp.h>
131 
132 #ifdef INET6
133 #include <netinet6/ip6_var.h>
134 #endif
135 
136 #ifdef MROUTING
137 #include <netinet/ip_mroute.h>
138 #endif
139 
140 #ifdef IPSEC
141 #include <netipsec/ipsec.h>
142 #include <netipsec/key.h>
143 #endif
144 
145 #ifdef MPLS
146 #include <netmpls/mpls.h>
147 #include <netmpls/mpls_var.h>
148 #endif
149 
150 static int ip_pcbopts(struct inpcb *, const struct sockopt *);
151 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
152 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
153 static void ip_mloopback(struct ifnet *, struct mbuf *,
154     const struct sockaddr_in *);
155 
156 extern pfil_head_t *inet_pfil_hook;			/* XXX */
157 
158 int	ip_do_loopback_cksum = 0;
159 
160 static 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;
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;
255 	int bound;
256 
257 	len = 0;
258 
259 	MCLAIM(m, &ip_tx_mowner);
260 
261 	KASSERT((m->m_flags & M_PKTHDR) != 0);
262 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
263 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
264 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
265 
266 	if (opt) {
267 		m = ip_insertoptions(m, opt, &len);
268 		if (len >= sizeof(struct ip))
269 			hlen = len;
270 	}
271 	ip = mtod(m, struct ip *);
272 
273 	/*
274 	 * Fill in IP header.
275 	 */
276 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
277 		ip->ip_v = IPVERSION;
278 		ip->ip_off = htons(0);
279 		/* ip->ip_id filled in after we find out source ia */
280 		ip->ip_hl = hlen >> 2;
281 		IP_STATINC(IP_STAT_LOCALOUT);
282 	} else {
283 		hlen = ip->ip_hl << 2;
284 	}
285 
286 	/*
287 	 * Route packet.
288 	 */
289 	if (ro == NULL) {
290 		memset(&iproute, 0, sizeof(iproute));
291 		ro = &iproute;
292 	}
293 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
294 	dst = satocsin(rtcache_getdst(ro));
295 
296 	/*
297 	 * If there is a cached route, check that it is to the same
298 	 * destination and is still up.  If not, free it and try again.
299 	 * The address family should also be checked in case of sharing
300 	 * the cache with IPv6.
301 	 */
302 	if (dst && (dst->sin_family != AF_INET ||
303 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
304 		rtcache_free(ro);
305 
306 	if ((rt = rtcache_validate(ro)) == NULL &&
307 	    (rt = rtcache_update(ro, 1)) == NULL) {
308 		dst = &u.dst4;
309 		error = rtcache_setdst(ro, &u.dst);
310 		if (error != 0)
311 			goto bad;
312 	}
313 
314 	/*
315 	 * If routing to interface only, short circuit routing lookup.
316 	 */
317 	if (flags & IP_ROUTETOIF) {
318 		if ((ia = ifatoia(ifa_ifwithladdr(sintocsa(dst)))) == NULL) {
319 			IP_STATINC(IP_STAT_NOROUTE);
320 			error = ENETUNREACH;
321 			goto bad;
322 		}
323 		ifp = ia->ia_ifp;
324 		mtu = ifp->if_mtu;
325 		ip->ip_ttl = 1;
326 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
327 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
328 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
329 	    imo != NULL && imo->imo_multicast_if_index != 0) {
330 		bound = curlwp_bind();
331 		ifp = mifp = if_get_byindex(imo->imo_multicast_if_index, &psref);
332 		if (ifp == NULL) {
333 			curlwp_bindx(bound);
334 			IP_STATINC(IP_STAT_NOROUTE);
335 			error = ENETUNREACH;
336 			goto bad;
337 		}
338 		mtu = ifp->if_mtu;
339 		ia = in_get_ia_from_ifp(ifp);
340 		isbroadcast = 0;
341 	} else {
342 		if (rt == NULL)
343 			rt = rtcache_init(ro);
344 		if (rt == NULL) {
345 			IP_STATINC(IP_STAT_NOROUTE);
346 			error = EHOSTUNREACH;
347 			goto bad;
348 		}
349 		ia = ifatoia(rt->rt_ifa);
350 		ifp = rt->rt_ifp;
351 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
352 			mtu = ifp->if_mtu;
353 		rt->rt_use++;
354 		if (rt->rt_flags & RTF_GATEWAY)
355 			dst = satosin(rt->rt_gateway);
356 		if (rt->rt_flags & RTF_HOST)
357 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
358 		else
359 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
360 	}
361 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
362 
363 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
364 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
365 		bool inmgroup;
366 
367 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
368 		    M_BCAST : M_MCAST;
369 		/*
370 		 * See if the caller provided any multicast options
371 		 */
372 		if (imo != NULL)
373 			ip->ip_ttl = imo->imo_multicast_ttl;
374 		else
375 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
376 
377 		/*
378 		 * if we don't know the outgoing ifp yet, we can't generate
379 		 * output
380 		 */
381 		if (!ifp) {
382 			IP_STATINC(IP_STAT_NOROUTE);
383 			error = ENETUNREACH;
384 			goto bad;
385 		}
386 
387 		/*
388 		 * If the packet is multicast or broadcast, confirm that
389 		 * the outgoing interface can transmit it.
390 		 */
391 		if (((m->m_flags & M_MCAST) &&
392 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
393 		    ((m->m_flags & M_BCAST) &&
394 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
395 			IP_STATINC(IP_STAT_NOROUTE);
396 			error = ENETUNREACH;
397 			goto bad;
398 		}
399 		/*
400 		 * If source address not specified yet, use an address
401 		 * of outgoing interface.
402 		 */
403 		if (in_nullhost(ip->ip_src)) {
404 			struct in_ifaddr *xia;
405 			struct ifaddr *xifa;
406 
407 			xia = in_get_ia_from_ifp(ifp);
408 			if (!xia) {
409 				error = EADDRNOTAVAIL;
410 				goto bad;
411 			}
412 			xifa = &xia->ia_ifa;
413 			if (xifa->ifa_getifa != NULL) {
414 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
415 				if (xia == NULL) {
416 					error = EADDRNOTAVAIL;
417 					goto bad;
418 				}
419 			}
420 			ip->ip_src = xia->ia_addr.sin_addr;
421 		}
422 
423 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
424 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
425 			/*
426 			 * If we belong to the destination multicast group
427 			 * on the outgoing interface, and the caller did not
428 			 * forbid loopback, loop back a copy.
429 			 */
430 			ip_mloopback(ifp, m, &u.dst4);
431 		}
432 #ifdef MROUTING
433 		else {
434 			/*
435 			 * If we are acting as a multicast router, perform
436 			 * multicast forwarding as if the packet had just
437 			 * arrived on the interface to which we are about
438 			 * to send.  The multicast forwarding function
439 			 * recursively calls this function, using the
440 			 * IP_FORWARDING flag to prevent infinite recursion.
441 			 *
442 			 * Multicasts that are looped back by ip_mloopback(),
443 			 * above, will be forwarded by the ip_input() routine,
444 			 * if necessary.
445 			 */
446 			extern struct socket *ip_mrouter;
447 
448 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
449 				if (ip_mforward(m, ifp) != 0) {
450 					m_freem(m);
451 					goto done;
452 				}
453 			}
454 		}
455 #endif
456 		/*
457 		 * Multicasts with a time-to-live of zero may be looped-
458 		 * back, above, but must not be transmitted on a network.
459 		 * Also, multicasts addressed to the loopback interface
460 		 * are not sent -- the above call to ip_mloopback() will
461 		 * loop back a copy if this host actually belongs to the
462 		 * destination group on the loopback interface.
463 		 */
464 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
465 			m_freem(m);
466 			goto done;
467 		}
468 		goto sendit;
469 	}
470 
471 	/*
472 	 * If source address not specified yet, use address
473 	 * of outgoing interface.
474 	 */
475 	if (in_nullhost(ip->ip_src)) {
476 		struct ifaddr *xifa;
477 
478 		xifa = &ia->ia_ifa;
479 		if (xifa->ifa_getifa != NULL) {
480 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
481 			if (ia == NULL) {
482 				error = EADDRNOTAVAIL;
483 				goto bad;
484 			}
485 		}
486 		ip->ip_src = ia->ia_addr.sin_addr;
487 	}
488 
489 	/*
490 	 * packets with Class-D address as source are not valid per
491 	 * RFC 1112
492 	 */
493 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
494 		IP_STATINC(IP_STAT_ODROPPED);
495 		error = EADDRNOTAVAIL;
496 		goto bad;
497 	}
498 
499 	/*
500 	 * Look for broadcast address and and verify user is allowed to
501 	 * send such a packet.
502 	 */
503 	if (isbroadcast) {
504 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
505 			error = EADDRNOTAVAIL;
506 			goto bad;
507 		}
508 		if ((flags & IP_ALLOWBROADCAST) == 0) {
509 			error = EACCES;
510 			goto bad;
511 		}
512 		/* don't allow broadcast messages to be fragmented */
513 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
514 			error = EMSGSIZE;
515 			goto bad;
516 		}
517 		m->m_flags |= M_BCAST;
518 	} else
519 		m->m_flags &= ~M_BCAST;
520 
521 sendit:
522 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
523 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
524 			ip->ip_id = 0;
525 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
526 			ip->ip_id = ip_newid(ia);
527 		} else {
528 
529 			/*
530 			 * TSO capable interfaces (typically?) increment
531 			 * ip_id for each segment.
532 			 * "allocate" enough ids here to increase the chance
533 			 * for them to be unique.
534 			 *
535 			 * note that the following calculation is not
536 			 * needed to be precise.  wasting some ip_id is fine.
537 			 */
538 
539 			unsigned int segsz = m->m_pkthdr.segsz;
540 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
541 			unsigned int num = howmany(datasz, segsz);
542 
543 			ip->ip_id = ip_newid_range(ia, num);
544 		}
545 	}
546 
547 	/*
548 	 * If we're doing Path MTU Discovery, we need to set DF unless
549 	 * the route's MTU is locked.
550 	 */
551 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
552 		ip->ip_off |= htons(IP_DF);
553 	}
554 
555 #ifdef IPSEC
556 	if (ipsec_used) {
557 		bool ipsec_done = false;
558 
559 		/* Perform IPsec processing, if any. */
560 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
561 		    &ipsec_done);
562 		if (error || ipsec_done)
563 			goto done;
564 	}
565 #endif
566 
567 	/*
568 	 * Run through list of hooks for output packets.
569 	 */
570 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
571 	if (error)
572 		goto done;
573 	if (m == NULL)
574 		goto done;
575 
576 	ip = mtod(m, struct ip *);
577 	hlen = ip->ip_hl << 2;
578 
579 	m->m_pkthdr.csum_data |= hlen << 16;
580 
581 #if IFA_STATS
582 	/*
583 	 * search for the source address structure to
584 	 * maintain output statistics.
585 	 */
586 	ia = in_get_ia(ip->ip_src);
587 #endif
588 
589 	/* Maybe skip checksums on loopback interfaces. */
590 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
591 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
592 	}
593 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
594 	/*
595 	 * If small enough for mtu of path, or if using TCP segmentation
596 	 * offload, can just send directly.
597 	 */
598 	if (ntohs(ip->ip_len) <= mtu ||
599 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
600 		const struct sockaddr *sa;
601 
602 #if IFA_STATS
603 		if (ia)
604 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
605 #endif
606 		/*
607 		 * Always initialize the sum to 0!  Some HW assisted
608 		 * checksumming requires this.
609 		 */
610 		ip->ip_sum = 0;
611 
612 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
613 			/*
614 			 * Perform any checksums that the hardware can't do
615 			 * for us.
616 			 *
617 			 * XXX Does any hardware require the {th,uh}_sum
618 			 * XXX fields to be 0?
619 			 */
620 			if (sw_csum & M_CSUM_IPv4) {
621 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
622 				ip->ip_sum = in_cksum(m, hlen);
623 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
624 			}
625 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
626 				if (IN_NEED_CHECKSUM(ifp,
627 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
628 					in_delayed_cksum(m);
629 				}
630 				m->m_pkthdr.csum_flags &=
631 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
632 			}
633 		}
634 
635 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
636 		if (__predict_true(
637 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
638 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
639 			error = ip_if_output(ifp, m, sa, rt);
640 		} else {
641 			error = ip_tso_output(ifp, m, sa, rt);
642 		}
643 		goto done;
644 	}
645 
646 	/*
647 	 * We can't use HW checksumming if we're about to
648 	 * to fragment the packet.
649 	 *
650 	 * XXX Some hardware can do this.
651 	 */
652 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
653 		if (IN_NEED_CHECKSUM(ifp,
654 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
655 			in_delayed_cksum(m);
656 		}
657 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
658 	}
659 
660 	/*
661 	 * Too large for interface; fragment if possible.
662 	 * Must be able to put at least 8 bytes per fragment.
663 	 */
664 	if (ntohs(ip->ip_off) & IP_DF) {
665 		if (flags & IP_RETURNMTU) {
666 			struct inpcb *inp;
667 
668 			KASSERT(so && solocked(so));
669 			inp = sotoinpcb(so);
670 			inp->inp_errormtu = mtu;
671 		}
672 		error = EMSGSIZE;
673 		IP_STATINC(IP_STAT_CANTFRAG);
674 		goto bad;
675 	}
676 
677 	error = ip_fragment(m, ifp, mtu);
678 	if (error) {
679 		m = NULL;
680 		goto bad;
681 	}
682 
683 	for (; m; m = m0) {
684 		m0 = m->m_nextpkt;
685 		m->m_nextpkt = 0;
686 		if (error) {
687 			m_freem(m);
688 			continue;
689 		}
690 #if IFA_STATS
691 		if (ia)
692 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
693 #endif
694 		/*
695 		 * If we get there, the packet has not been handled by
696 		 * IPsec whereas it should have. Now that it has been
697 		 * fragmented, re-inject it in ip_output so that IPsec
698 		 * processing can occur.
699 		 */
700 		if (natt_frag) {
701 			error = ip_output(m, opt, ro,
702 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
703 			    imo, so);
704 		} else {
705 			KASSERT((m->m_pkthdr.csum_flags &
706 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
707 			error = ip_if_output(ifp, m,
708 			    (m->m_flags & M_MCAST) ?
709 			    sintocsa(rdst) : sintocsa(dst), rt);
710 		}
711 	}
712 	if (error == 0) {
713 		IP_STATINC(IP_STAT_FRAGMENTED);
714 	}
715 done:
716 	if (ro == &iproute) {
717 		rtcache_free(&iproute);
718 	}
719 #ifdef IPSEC
720 	if (sp) {
721 		KEY_FREESP(&sp);
722 	}
723 #endif
724 	if (mifp != NULL) {
725 		if_put(mifp, &psref);
726 		curlwp_bindx(bound);
727 	}
728 	return error;
729 bad:
730 	m_freem(m);
731 	goto done;
732 }
733 
734 int
735 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
736 {
737 	struct ip *ip, *mhip;
738 	struct mbuf *m0;
739 	int len, hlen, off;
740 	int mhlen, firstlen;
741 	struct mbuf **mnext;
742 	int sw_csum = m->m_pkthdr.csum_flags;
743 	int fragments = 0;
744 	int s;
745 	int error = 0;
746 
747 	ip = mtod(m, struct ip *);
748 	hlen = ip->ip_hl << 2;
749 	if (ifp != NULL)
750 		sw_csum &= ~ifp->if_csum_flags_tx;
751 
752 	len = (mtu - hlen) &~ 7;
753 	if (len < 8) {
754 		m_freem(m);
755 		return (EMSGSIZE);
756 	}
757 
758 	firstlen = len;
759 	mnext = &m->m_nextpkt;
760 
761 	/*
762 	 * Loop through length of segment after first fragment,
763 	 * make new header and copy data of each part and link onto chain.
764 	 */
765 	m0 = m;
766 	mhlen = sizeof (struct ip);
767 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
768 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
769 		if (m == 0) {
770 			error = ENOBUFS;
771 			IP_STATINC(IP_STAT_ODROPPED);
772 			goto sendorfree;
773 		}
774 		MCLAIM(m, m0->m_owner);
775 		*mnext = m;
776 		mnext = &m->m_nextpkt;
777 		m->m_data += max_linkhdr;
778 		mhip = mtod(m, struct ip *);
779 		*mhip = *ip;
780 		/* we must inherit MCAST and BCAST flags */
781 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
782 		if (hlen > sizeof (struct ip)) {
783 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
784 			mhip->ip_hl = mhlen >> 2;
785 		}
786 		m->m_len = mhlen;
787 		mhip->ip_off = ((off - hlen) >> 3) +
788 		    (ntohs(ip->ip_off) & ~IP_MF);
789 		if (ip->ip_off & htons(IP_MF))
790 			mhip->ip_off |= IP_MF;
791 		if (off + len >= ntohs(ip->ip_len))
792 			len = ntohs(ip->ip_len) - off;
793 		else
794 			mhip->ip_off |= IP_MF;
795 		HTONS(mhip->ip_off);
796 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
797 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
798 		if (m->m_next == 0) {
799 			error = ENOBUFS;	/* ??? */
800 			IP_STATINC(IP_STAT_ODROPPED);
801 			goto sendorfree;
802 		}
803 		m->m_pkthdr.len = mhlen + len;
804 		m_reset_rcvif(m);
805 		mhip->ip_sum = 0;
806 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
807 		if (sw_csum & M_CSUM_IPv4) {
808 			mhip->ip_sum = in_cksum(m, mhlen);
809 		} else {
810 			/*
811 			 * checksum is hw-offloaded or not necessary.
812 			 */
813 			m->m_pkthdr.csum_flags |=
814 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
815 			m->m_pkthdr.csum_data |= mhlen << 16;
816 			KASSERT(!(ifp != NULL &&
817 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
818 			    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
819 		}
820 		IP_STATINC(IP_STAT_OFRAGMENTS);
821 		fragments++;
822 	}
823 	/*
824 	 * Update first fragment by trimming what's been copied out
825 	 * and updating header, then send each fragment (in order).
826 	 */
827 	m = m0;
828 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
829 	m->m_pkthdr.len = hlen + firstlen;
830 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
831 	ip->ip_off |= htons(IP_MF);
832 	ip->ip_sum = 0;
833 	if (sw_csum & M_CSUM_IPv4) {
834 		ip->ip_sum = in_cksum(m, hlen);
835 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
836 	} else {
837 		/*
838 		 * checksum is hw-offloaded or not necessary.
839 		 */
840 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) ||
841 		    (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
842 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
843 		    sizeof(struct ip));
844 	}
845 sendorfree:
846 	/*
847 	 * If there is no room for all the fragments, don't queue
848 	 * any of them.
849 	 */
850 	if (ifp != NULL) {
851 		s = splnet();
852 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
853 		    error == 0) {
854 			error = ENOBUFS;
855 			IP_STATINC(IP_STAT_ODROPPED);
856 			IFQ_INC_DROPS(&ifp->if_snd);
857 		}
858 		splx(s);
859 	}
860 	if (error) {
861 		for (m = m0; m; m = m0) {
862 			m0 = m->m_nextpkt;
863 			m->m_nextpkt = NULL;
864 			m_freem(m);
865 		}
866 	}
867 	return (error);
868 }
869 
870 /*
871  * Process a delayed payload checksum calculation.
872  */
873 void
874 in_delayed_cksum(struct mbuf *m)
875 {
876 	struct ip *ip;
877 	u_int16_t csum, offset;
878 
879 	ip = mtod(m, struct ip *);
880 	offset = ip->ip_hl << 2;
881 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
882 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
883 		csum = 0xffff;
884 
885 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
886 
887 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
888 		/* This happen when ip options were inserted
889 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
890 		    m->m_len, offset, ip->ip_p);
891 		 */
892 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
893 	} else
894 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
895 }
896 
897 /*
898  * Determine the maximum length of the options to be inserted;
899  * we would far rather allocate too much space rather than too little.
900  */
901 
902 u_int
903 ip_optlen(struct inpcb *inp)
904 {
905 	struct mbuf *m = inp->inp_options;
906 
907 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
908 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
909 	}
910 	return 0;
911 }
912 
913 /*
914  * Insert IP options into preformed packet.
915  * Adjust IP destination as required for IP source routing,
916  * as indicated by a non-zero in_addr at the start of the options.
917  */
918 static struct mbuf *
919 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
920 {
921 	struct ipoption *p = mtod(opt, struct ipoption *);
922 	struct mbuf *n;
923 	struct ip *ip = mtod(m, struct ip *);
924 	unsigned optlen;
925 
926 	optlen = opt->m_len - sizeof(p->ipopt_dst);
927 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
928 		return (m);		/* XXX should fail */
929 	if (!in_nullhost(p->ipopt_dst))
930 		ip->ip_dst = p->ipopt_dst;
931 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
932 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
933 		if (n == 0)
934 			return (m);
935 		MCLAIM(n, m->m_owner);
936 		M_MOVE_PKTHDR(n, m);
937 		m->m_len -= sizeof(struct ip);
938 		m->m_data += sizeof(struct ip);
939 		n->m_next = m;
940 		m = n;
941 		m->m_len = optlen + sizeof(struct ip);
942 		m->m_data += max_linkhdr;
943 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
944 	} else {
945 		m->m_data -= optlen;
946 		m->m_len += optlen;
947 		memmove(mtod(m, void *), ip, sizeof(struct ip));
948 	}
949 	m->m_pkthdr.len += optlen;
950 	ip = mtod(m, struct ip *);
951 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
952 	*phlen = sizeof(struct ip) + optlen;
953 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
954 	return (m);
955 }
956 
957 /*
958  * Copy options from ip to jp,
959  * omitting those not copied during fragmentation.
960  */
961 int
962 ip_optcopy(struct ip *ip, struct ip *jp)
963 {
964 	u_char *cp, *dp;
965 	int opt, optlen, cnt;
966 
967 	cp = (u_char *)(ip + 1);
968 	dp = (u_char *)(jp + 1);
969 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
970 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
971 		opt = cp[0];
972 		if (opt == IPOPT_EOL)
973 			break;
974 		if (opt == IPOPT_NOP) {
975 			/* Preserve for IP mcast tunnel's LSRR alignment. */
976 			*dp++ = IPOPT_NOP;
977 			optlen = 1;
978 			continue;
979 		}
980 
981 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
982 		optlen = cp[IPOPT_OLEN];
983 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
984 
985 		/* Invalid lengths should have been caught by ip_dooptions. */
986 		if (optlen > cnt)
987 			optlen = cnt;
988 		if (IPOPT_COPIED(opt)) {
989 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
990 			dp += optlen;
991 		}
992 	}
993 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
994 		*dp++ = IPOPT_EOL;
995 	return (optlen);
996 }
997 
998 /*
999  * IP socket option processing.
1000  */
1001 int
1002 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
1003 {
1004 	struct inpcb *inp = sotoinpcb(so);
1005 	struct ip *ip = &inp->inp_ip;
1006 	int inpflags = inp->inp_flags;
1007 	int optval = 0, error = 0;
1008 
1009 	if (sopt->sopt_level != IPPROTO_IP) {
1010 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
1011 			return 0;
1012 		return ENOPROTOOPT;
1013 	}
1014 
1015 	switch (op) {
1016 	case PRCO_SETOPT:
1017 		switch (sopt->sopt_name) {
1018 		case IP_OPTIONS:
1019 #ifdef notyet
1020 		case IP_RETOPTS:
1021 #endif
1022 			error = ip_pcbopts(inp, sopt);
1023 			break;
1024 
1025 		case IP_TOS:
1026 		case IP_TTL:
1027 		case IP_MINTTL:
1028 		case IP_PKTINFO:
1029 		case IP_RECVOPTS:
1030 		case IP_RECVRETOPTS:
1031 		case IP_RECVDSTADDR:
1032 		case IP_RECVIF:
1033 		case IP_RECVPKTINFO:
1034 		case IP_RECVTTL:
1035 			error = sockopt_getint(sopt, &optval);
1036 			if (error)
1037 				break;
1038 
1039 			switch (sopt->sopt_name) {
1040 			case IP_TOS:
1041 				ip->ip_tos = optval;
1042 				break;
1043 
1044 			case IP_TTL:
1045 				ip->ip_ttl = optval;
1046 				break;
1047 
1048 			case IP_MINTTL:
1049 				if (optval > 0 && optval <= MAXTTL)
1050 					inp->inp_ip_minttl = optval;
1051 				else
1052 					error = EINVAL;
1053 				break;
1054 #define	OPTSET(bit) \
1055 	if (optval) \
1056 		inpflags |= bit; \
1057 	else \
1058 		inpflags &= ~bit;
1059 
1060 			case IP_PKTINFO:
1061 				OPTSET(INP_PKTINFO);
1062 				break;
1063 
1064 			case IP_RECVOPTS:
1065 				OPTSET(INP_RECVOPTS);
1066 				break;
1067 
1068 			case IP_RECVPKTINFO:
1069 				OPTSET(INP_RECVPKTINFO);
1070 				break;
1071 
1072 			case IP_RECVRETOPTS:
1073 				OPTSET(INP_RECVRETOPTS);
1074 				break;
1075 
1076 			case IP_RECVDSTADDR:
1077 				OPTSET(INP_RECVDSTADDR);
1078 				break;
1079 
1080 			case IP_RECVIF:
1081 				OPTSET(INP_RECVIF);
1082 				break;
1083 
1084 			case IP_RECVTTL:
1085 				OPTSET(INP_RECVTTL);
1086 				break;
1087 			}
1088 		break;
1089 #undef OPTSET
1090 
1091 		case IP_MULTICAST_IF:
1092 		case IP_MULTICAST_TTL:
1093 		case IP_MULTICAST_LOOP:
1094 		case IP_ADD_MEMBERSHIP:
1095 		case IP_DROP_MEMBERSHIP:
1096 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1097 			break;
1098 
1099 		case IP_PORTRANGE:
1100 			error = sockopt_getint(sopt, &optval);
1101 			if (error)
1102 				break;
1103 
1104 			switch (optval) {
1105 			case IP_PORTRANGE_DEFAULT:
1106 			case IP_PORTRANGE_HIGH:
1107 				inpflags &= ~(INP_LOWPORT);
1108 				break;
1109 
1110 			case IP_PORTRANGE_LOW:
1111 				inpflags |= INP_LOWPORT;
1112 				break;
1113 
1114 			default:
1115 				error = EINVAL;
1116 				break;
1117 			}
1118 			break;
1119 
1120 		case IP_PORTALGO:
1121 			error = sockopt_getint(sopt, &optval);
1122 			if (error)
1123 				break;
1124 
1125 			error = portalgo_algo_index_select(
1126 			    (struct inpcb_hdr *)inp, optval);
1127 			break;
1128 
1129 #if defined(IPSEC)
1130 		case IP_IPSEC_POLICY:
1131 			if (ipsec_enabled) {
1132 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1133 				    sopt->sopt_data, sopt->sopt_size,
1134 				    curlwp->l_cred);
1135 				break;
1136 			}
1137 			/*FALLTHROUGH*/
1138 #endif /* IPSEC */
1139 
1140 		default:
1141 			error = ENOPROTOOPT;
1142 			break;
1143 		}
1144 		break;
1145 
1146 	case PRCO_GETOPT:
1147 		switch (sopt->sopt_name) {
1148 		case IP_OPTIONS:
1149 		case IP_RETOPTS: {
1150 			struct mbuf *mopts = inp->inp_options;
1151 
1152 			if (mopts) {
1153 				struct mbuf *m;
1154 
1155 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1156 				if (m == NULL) {
1157 					error = ENOBUFS;
1158 					break;
1159 				}
1160 				error = sockopt_setmbuf(sopt, m);
1161 			}
1162 			break;
1163 		}
1164 		case IP_PKTINFO:
1165 		case IP_TOS:
1166 		case IP_TTL:
1167 		case IP_MINTTL:
1168 		case IP_RECVOPTS:
1169 		case IP_RECVRETOPTS:
1170 		case IP_RECVDSTADDR:
1171 		case IP_RECVIF:
1172 		case IP_RECVPKTINFO:
1173 		case IP_RECVTTL:
1174 		case IP_ERRORMTU:
1175 			switch (sopt->sopt_name) {
1176 			case IP_TOS:
1177 				optval = ip->ip_tos;
1178 				break;
1179 
1180 			case IP_TTL:
1181 				optval = ip->ip_ttl;
1182 				break;
1183 
1184 			case IP_MINTTL:
1185 				optval = inp->inp_ip_minttl;
1186 				break;
1187 
1188 			case IP_ERRORMTU:
1189 				optval = inp->inp_errormtu;
1190 				break;
1191 
1192 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1193 
1194 			case IP_PKTINFO:
1195 				optval = OPTBIT(INP_PKTINFO);
1196 				break;
1197 
1198 			case IP_RECVOPTS:
1199 				optval = OPTBIT(INP_RECVOPTS);
1200 				break;
1201 
1202 			case IP_RECVPKTINFO:
1203 				optval = OPTBIT(INP_RECVPKTINFO);
1204 				break;
1205 
1206 			case IP_RECVRETOPTS:
1207 				optval = OPTBIT(INP_RECVRETOPTS);
1208 				break;
1209 
1210 			case IP_RECVDSTADDR:
1211 				optval = OPTBIT(INP_RECVDSTADDR);
1212 				break;
1213 
1214 			case IP_RECVIF:
1215 				optval = OPTBIT(INP_RECVIF);
1216 				break;
1217 
1218 			case IP_RECVTTL:
1219 				optval = OPTBIT(INP_RECVTTL);
1220 				break;
1221 			}
1222 			error = sockopt_setint(sopt, optval);
1223 			break;
1224 
1225 #if 0	/* defined(IPSEC) */
1226 		case IP_IPSEC_POLICY:
1227 		{
1228 			struct mbuf *m = NULL;
1229 
1230 			/* XXX this will return EINVAL as sopt is empty */
1231 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1232 			    sopt->sopt_size, &m);
1233 			if (error == 0)
1234 				error = sockopt_setmbuf(sopt, m);
1235 			break;
1236 		}
1237 #endif /*IPSEC*/
1238 
1239 		case IP_MULTICAST_IF:
1240 		case IP_MULTICAST_TTL:
1241 		case IP_MULTICAST_LOOP:
1242 		case IP_ADD_MEMBERSHIP:
1243 		case IP_DROP_MEMBERSHIP:
1244 			error = ip_getmoptions(inp->inp_moptions, sopt);
1245 			break;
1246 
1247 		case IP_PORTRANGE:
1248 			if (inpflags & INP_LOWPORT)
1249 				optval = IP_PORTRANGE_LOW;
1250 			else
1251 				optval = IP_PORTRANGE_DEFAULT;
1252 			error = sockopt_setint(sopt, optval);
1253 			break;
1254 
1255 		case IP_PORTALGO:
1256 			optval = inp->inp_portalgo;
1257 			error = sockopt_setint(sopt, optval);
1258 			break;
1259 
1260 		default:
1261 			error = ENOPROTOOPT;
1262 			break;
1263 		}
1264 		break;
1265 	}
1266 
1267 	if (!error) {
1268 		inp->inp_flags = inpflags;
1269 	}
1270 	return error;
1271 }
1272 
1273 /*
1274  * Set up IP options in pcb for insertion in output packets.
1275  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1276  * with destination address if source routed.
1277  */
1278 static int
1279 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1280 {
1281 	struct mbuf *m;
1282 	const u_char *cp;
1283 	u_char *dp;
1284 	int cnt;
1285 
1286 	/* Turn off any old options. */
1287 	if (inp->inp_options) {
1288 		m_free(inp->inp_options);
1289 	}
1290 	inp->inp_options = NULL;
1291 	if ((cnt = sopt->sopt_size) == 0) {
1292 		/* Only turning off any previous options. */
1293 		return 0;
1294 	}
1295 	cp = sopt->sopt_data;
1296 
1297 #ifndef	__vax__
1298 	if (cnt % sizeof(int32_t))
1299 		return (EINVAL);
1300 #endif
1301 
1302 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1303 	if (m == NULL)
1304 		return (ENOBUFS);
1305 
1306 	dp = mtod(m, u_char *);
1307 	memset(dp, 0, sizeof(struct in_addr));
1308 	dp += sizeof(struct in_addr);
1309 	m->m_len = sizeof(struct in_addr);
1310 
1311 	/*
1312 	 * IP option list according to RFC791. Each option is of the form
1313 	 *
1314 	 *	[optval] [olen] [(olen - 2) data bytes]
1315 	 *
1316 	 * We validate the list and copy options to an mbuf for prepending
1317 	 * to data packets. The IP first-hop destination address will be
1318 	 * stored before actual options and is zero if unset.
1319 	 */
1320 	while (cnt > 0) {
1321 		uint8_t optval, olen, offset;
1322 
1323 		optval = cp[IPOPT_OPTVAL];
1324 
1325 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1326 			olen = 1;
1327 		} else {
1328 			if (cnt < IPOPT_OLEN + 1)
1329 				goto bad;
1330 
1331 			olen = cp[IPOPT_OLEN];
1332 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1333 				goto bad;
1334 		}
1335 
1336 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1337 			/*
1338 			 * user process specifies route as:
1339 			 *	->A->B->C->D
1340 			 * D must be our final destination (but we can't
1341 			 * check that since we may not have connected yet).
1342 			 * A is first hop destination, which doesn't appear in
1343 			 * actual IP option, but is stored before the options.
1344 			 */
1345 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1346 				goto bad;
1347 
1348 			offset = cp[IPOPT_OFFSET];
1349 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1350 			    sizeof(struct in_addr));
1351 
1352 			cp += sizeof(struct in_addr);
1353 			cnt -= sizeof(struct in_addr);
1354 			olen -= sizeof(struct in_addr);
1355 
1356 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1357 				goto bad;
1358 
1359 			memcpy(dp, cp, olen);
1360 			dp[IPOPT_OPTVAL] = optval;
1361 			dp[IPOPT_OLEN] = olen;
1362 			dp[IPOPT_OFFSET] = offset;
1363 			break;
1364 		} else {
1365 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1366 				goto bad;
1367 
1368 			memcpy(dp, cp, olen);
1369 			break;
1370 		}
1371 
1372 		dp += olen;
1373 		m->m_len += olen;
1374 
1375 		if (optval == IPOPT_EOL)
1376 			break;
1377 
1378 		cp += olen;
1379 		cnt -= olen;
1380 	}
1381 
1382 	inp->inp_options = m;
1383 	return 0;
1384 bad:
1385 	(void)m_free(m);
1386 	return EINVAL;
1387 }
1388 
1389 /*
1390  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1391  */
1392 static struct ifnet *
1393 ip_multicast_if(struct in_addr *a, int *ifindexp)
1394 {
1395 	int ifindex;
1396 	struct ifnet *ifp = NULL;
1397 	struct in_ifaddr *ia;
1398 
1399 	if (ifindexp)
1400 		*ifindexp = 0;
1401 	if (ntohl(a->s_addr) >> 24 == 0) {
1402 		ifindex = ntohl(a->s_addr) & 0xffffff;
1403 		ifp = if_byindex(ifindex);
1404 		if (!ifp)
1405 			return NULL;
1406 		if (ifindexp)
1407 			*ifindexp = ifindex;
1408 	} else {
1409 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1410 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1411 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1412 				ifp = ia->ia_ifp;
1413 				break;
1414 			}
1415 		}
1416 	}
1417 	return ifp;
1418 }
1419 
1420 static int
1421 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1422 {
1423 	u_int tval;
1424 	u_char cval;
1425 	int error;
1426 
1427 	if (sopt == NULL)
1428 		return EINVAL;
1429 
1430 	switch (sopt->sopt_size) {
1431 	case sizeof(u_char):
1432 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1433 		tval = cval;
1434 		break;
1435 
1436 	case sizeof(u_int):
1437 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1438 		break;
1439 
1440 	default:
1441 		error = EINVAL;
1442 	}
1443 
1444 	if (error)
1445 		return error;
1446 
1447 	if (tval > maxval)
1448 		return EINVAL;
1449 
1450 	*val = tval;
1451 	return 0;
1452 }
1453 
1454 static int
1455 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1456     struct in_addr *ia, bool add)
1457 {
1458 	int error;
1459 	struct ip_mreq mreq;
1460 
1461 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1462 	if (error)
1463 		return error;
1464 
1465 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1466 		return EINVAL;
1467 
1468 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1469 
1470 	if (in_nullhost(mreq.imr_interface)) {
1471 		union {
1472 			struct sockaddr		dst;
1473 			struct sockaddr_in	dst4;
1474 		} u;
1475 		struct route ro;
1476 
1477 		if (!add) {
1478 			*ifp = NULL;
1479 			return 0;
1480 		}
1481 		/*
1482 		 * If no interface address was provided, use the interface of
1483 		 * the route to the given multicast address.
1484 		 */
1485 		struct rtentry *rt;
1486 		memset(&ro, 0, sizeof(ro));
1487 
1488 		sockaddr_in_init(&u.dst4, ia, 0);
1489 		error = rtcache_setdst(&ro, &u.dst);
1490 		if (error != 0)
1491 			return error;
1492 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1493 		rtcache_free(&ro);
1494 	} else {
1495 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1496 		if (!add && *ifp == NULL)
1497 			return EADDRNOTAVAIL;
1498 	}
1499 	return 0;
1500 }
1501 
1502 /*
1503  * Add a multicast group membership.
1504  * Group must be a valid IP multicast address.
1505  */
1506 static int
1507 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1508 {
1509 	struct ifnet *ifp = NULL;	// XXX: gcc [ppc]
1510 	struct in_addr ia;
1511 	int i, error;
1512 
1513 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1514 		error = ip_get_membership(sopt, &ifp, &ia, true);
1515 	else
1516 #ifdef INET6
1517 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1518 #else
1519 		return EINVAL;
1520 #endif
1521 
1522 	if (error)
1523 		return error;
1524 
1525 	/*
1526 	 * See if we found an interface, and confirm that it
1527 	 * supports multicast.
1528 	 */
1529 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1530 		return EADDRNOTAVAIL;
1531 
1532 	/*
1533 	 * See if the membership already exists or if all the
1534 	 * membership slots are full.
1535 	 */
1536 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1537 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1538 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1539 			break;
1540 	}
1541 	if (i < imo->imo_num_memberships)
1542 		return EADDRINUSE;
1543 
1544 	if (i == IP_MAX_MEMBERSHIPS)
1545 		return ETOOMANYREFS;
1546 
1547 	/*
1548 	 * Everything looks good; add a new record to the multicast
1549 	 * address list for the given interface.
1550 	 */
1551 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1552 		return ENOBUFS;
1553 
1554 	++imo->imo_num_memberships;
1555 	return 0;
1556 }
1557 
1558 /*
1559  * Drop a multicast group membership.
1560  * Group must be a valid IP multicast address.
1561  */
1562 static int
1563 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1564 {
1565 	struct in_addr ia = { .s_addr = 0 };	// XXX: gcc [ppc]
1566 	struct ifnet *ifp = NULL;		// XXX: gcc [ppc]
1567 	int i, error;
1568 
1569 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1570 		error = ip_get_membership(sopt, &ifp, &ia, false);
1571 	else
1572 #ifdef INET6
1573 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1574 #else
1575 		return EINVAL;
1576 #endif
1577 
1578 	if (error)
1579 		return error;
1580 
1581 	/*
1582 	 * Find the membership in the membership array.
1583 	 */
1584 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1585 		if ((ifp == NULL ||
1586 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1587 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1588 			break;
1589 	}
1590 	if (i == imo->imo_num_memberships)
1591 		return EADDRNOTAVAIL;
1592 
1593 	/*
1594 	 * Give up the multicast address record to which the
1595 	 * membership points.
1596 	 */
1597 	in_delmulti(imo->imo_membership[i]);
1598 
1599 	/*
1600 	 * Remove the gap in the membership array.
1601 	 */
1602 	for (++i; i < imo->imo_num_memberships; ++i)
1603 		imo->imo_membership[i-1] = imo->imo_membership[i];
1604 	--imo->imo_num_memberships;
1605 	return 0;
1606 }
1607 
1608 /*
1609  * Set the IP multicast options in response to user setsockopt().
1610  */
1611 int
1612 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1613 {
1614 	struct ip_moptions *imo = *pimo;
1615 	struct in_addr addr;
1616 	struct ifnet *ifp;
1617 	int ifindex, error = 0;
1618 
1619 	if (!imo) {
1620 		/*
1621 		 * No multicast option buffer attached to the pcb;
1622 		 * allocate one and initialize to default values.
1623 		 */
1624 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1625 		if (imo == NULL)
1626 			return ENOBUFS;
1627 
1628 		imo->imo_multicast_if_index = 0;
1629 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1630 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1631 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1632 		imo->imo_num_memberships = 0;
1633 		*pimo = imo;
1634 	}
1635 
1636 	switch (sopt->sopt_name) {
1637 	case IP_MULTICAST_IF:
1638 		/*
1639 		 * Select the interface for outgoing multicast packets.
1640 		 */
1641 		error = sockopt_get(sopt, &addr, sizeof(addr));
1642 		if (error)
1643 			break;
1644 
1645 		/*
1646 		 * INADDR_ANY is used to remove a previous selection.
1647 		 * When no interface is selected, a default one is
1648 		 * chosen every time a multicast packet is sent.
1649 		 */
1650 		if (in_nullhost(addr)) {
1651 			imo->imo_multicast_if_index = 0;
1652 			break;
1653 		}
1654 		/*
1655 		 * The selected interface is identified by its local
1656 		 * IP address.  Find the interface and confirm that
1657 		 * it supports multicasting.
1658 		 */
1659 		ifp = ip_multicast_if(&addr, &ifindex);
1660 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1661 			error = EADDRNOTAVAIL;
1662 			break;
1663 		}
1664 		imo->imo_multicast_if_index = ifp->if_index;
1665 		if (ifindex)
1666 			imo->imo_multicast_addr = addr;
1667 		else
1668 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1669 		break;
1670 
1671 	case IP_MULTICAST_TTL:
1672 		/*
1673 		 * Set the IP time-to-live for outgoing multicast packets.
1674 		 */
1675 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1676 		break;
1677 
1678 	case IP_MULTICAST_LOOP:
1679 		/*
1680 		 * Set the loopback flag for outgoing multicast packets.
1681 		 * Must be zero or one.
1682 		 */
1683 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1684 		break;
1685 
1686 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1687 		error = ip_add_membership(imo, sopt);
1688 		break;
1689 
1690 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1691 		error = ip_drop_membership(imo, sopt);
1692 		break;
1693 
1694 	default:
1695 		error = EOPNOTSUPP;
1696 		break;
1697 	}
1698 
1699 	/*
1700 	 * If all options have default values, no need to keep the mbuf.
1701 	 */
1702 	if (imo->imo_multicast_if_index == 0 &&
1703 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1704 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1705 	    imo->imo_num_memberships == 0) {
1706 		kmem_free(imo, sizeof(*imo));
1707 		*pimo = NULL;
1708 	}
1709 
1710 	return error;
1711 }
1712 
1713 /*
1714  * Return the IP multicast options in response to user getsockopt().
1715  */
1716 int
1717 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1718 {
1719 	struct in_addr addr;
1720 	uint8_t optval;
1721 	int error = 0;
1722 
1723 	switch (sopt->sopt_name) {
1724 	case IP_MULTICAST_IF:
1725 		if (imo == NULL || imo->imo_multicast_if_index == 0)
1726 			addr = zeroin_addr;
1727 		else if (imo->imo_multicast_addr.s_addr) {
1728 			/* return the value user has set */
1729 			addr = imo->imo_multicast_addr;
1730 		} else {
1731 			struct ifnet *ifp;
1732 			struct in_ifaddr *ia = NULL;
1733 			int s = pserialize_read_enter();
1734 
1735 			ifp = if_byindex(imo->imo_multicast_if_index);
1736 			if (ifp != NULL) {
1737 				ia = in_get_ia_from_ifp(ifp);
1738 			}
1739 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1740 			pserialize_read_exit(s);
1741 		}
1742 		error = sockopt_set(sopt, &addr, sizeof(addr));
1743 		break;
1744 
1745 	case IP_MULTICAST_TTL:
1746 		optval = imo ? imo->imo_multicast_ttl
1747 		    : IP_DEFAULT_MULTICAST_TTL;
1748 
1749 		error = sockopt_set(sopt, &optval, sizeof(optval));
1750 		break;
1751 
1752 	case IP_MULTICAST_LOOP:
1753 		optval = imo ? imo->imo_multicast_loop
1754 		    : IP_DEFAULT_MULTICAST_LOOP;
1755 
1756 		error = sockopt_set(sopt, &optval, sizeof(optval));
1757 		break;
1758 
1759 	default:
1760 		error = EOPNOTSUPP;
1761 	}
1762 
1763 	return error;
1764 }
1765 
1766 /*
1767  * Discard the IP multicast options.
1768  */
1769 void
1770 ip_freemoptions(struct ip_moptions *imo)
1771 {
1772 	int i;
1773 
1774 	if (imo != NULL) {
1775 		for (i = 0; i < imo->imo_num_memberships; ++i)
1776 			in_delmulti(imo->imo_membership[i]);
1777 		kmem_free(imo, sizeof(*imo));
1778 	}
1779 }
1780 
1781 /*
1782  * Routine called from ip_output() to loop back a copy of an IP multicast
1783  * packet to the input queue of a specified interface.  Note that this
1784  * calls the output routine of the loopback "driver", but with an interface
1785  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1786  */
1787 static void
1788 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1789 {
1790 	struct ip *ip;
1791 	struct mbuf *copym;
1792 
1793 	copym = m_copypacket(m, M_DONTWAIT);
1794 	if (copym != NULL &&
1795 	    (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1796 		copym = m_pullup(copym, sizeof(struct ip));
1797 	if (copym == NULL)
1798 		return;
1799 	/*
1800 	 * We don't bother to fragment if the IP length is greater
1801 	 * than the interface's MTU.  Can this possibly matter?
1802 	 */
1803 	ip = mtod(copym, struct ip *);
1804 
1805 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1806 		in_delayed_cksum(copym);
1807 		copym->m_pkthdr.csum_flags &=
1808 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1809 	}
1810 
1811 	ip->ip_sum = 0;
1812 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1813 #ifndef NET_MPSAFE
1814 	KERNEL_LOCK(1, NULL);
1815 #endif
1816 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1817 #ifndef NET_MPSAFE
1818 	KERNEL_UNLOCK_ONE(NULL);
1819 #endif
1820 }
1821