xref: /netbsd-src/sys/netinet/ip_output.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: ip_output.c,v 1.236 2015/04/03 07:55:18 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 1.236 2015/04/03 07:55:18 ozaki-r Exp $");
95 
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 #include "opt_mrouting.h"
99 #include "opt_net_mpsafe.h"
100 
101 #include <sys/param.h>
102 #include <sys/kmem.h>
103 #include <sys/mbuf.h>
104 #include <sys/protosw.h>
105 #include <sys/socket.h>
106 #include <sys/socketvar.h>
107 #include <sys/kauth.h>
108 #ifdef IPSEC
109 #include <sys/domain.h>
110 #endif
111 #include <sys/systm.h>
112 
113 #include <net/if.h>
114 #include <net/route.h>
115 #include <net/pfil.h>
116 
117 #include <netinet/in.h>
118 #include <netinet/in_systm.h>
119 #include <netinet/ip.h>
120 #include <netinet/in_pcb.h>
121 #include <netinet/in_var.h>
122 #include <netinet/ip_var.h>
123 #include <netinet/ip_private.h>
124 #include <netinet/in_offload.h>
125 #include <netinet/portalgo.h>
126 #include <netinet/udp.h>
127 
128 #ifdef INET6
129 #include <netinet6/ip6_var.h>
130 #endif
131 
132 #ifdef MROUTING
133 #include <netinet/ip_mroute.h>
134 #endif
135 
136 #ifdef IPSEC
137 #include <netipsec/ipsec.h>
138 #include <netipsec/key.h>
139 #endif
140 
141 static int ip_pcbopts(struct inpcb *, const struct sockopt *);
142 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
143 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
144 static void ip_mloopback(struct ifnet *, struct mbuf *,
145     const struct sockaddr_in *);
146 
147 extern pfil_head_t *inet_pfil_hook;			/* XXX */
148 
149 int	ip_do_loopback_cksum = 0;
150 
151 /*
152  * IP output.  The packet in mbuf chain m contains a skeletal IP
153  * header (with len, off, ttl, proto, tos, src, dst).
154  * The mbuf chain containing the packet will be freed.
155  * The mbuf opt, if present, will not be freed.
156  */
157 int
158 ip_output(struct mbuf *m0, ...)
159 {
160 	struct rtentry *rt;
161 	struct ip *ip;
162 	struct ifnet *ifp;
163 	struct mbuf *m = m0;
164 	int hlen = sizeof (struct ip);
165 	int len, error = 0;
166 	struct route iproute;
167 	const struct sockaddr_in *dst;
168 	struct in_ifaddr *ia;
169 	int isbroadcast;
170 	struct mbuf *opt;
171 	struct route *ro;
172 	int flags, sw_csum;
173 	u_long mtu;
174 	struct ip_moptions *imo;
175 	struct socket *so;
176 	va_list ap;
177 #ifdef IPSEC
178 	struct secpolicy *sp = NULL;
179 #endif
180 	bool natt_frag = false;
181 	bool rtmtu_nolock;
182 	union {
183 		struct sockaddr		dst;
184 		struct sockaddr_in	dst4;
185 	} u;
186 	struct sockaddr *rdst = &u.dst;	/* real IP destination, as opposed
187 					 * to the nexthop
188 					 */
189 
190 	len = 0;
191 	va_start(ap, m0);
192 	opt = va_arg(ap, struct mbuf *);
193 	ro = va_arg(ap, struct route *);
194 	flags = va_arg(ap, int);
195 	imo = va_arg(ap, struct ip_moptions *);
196 	so = va_arg(ap, struct socket *);
197 	va_end(ap);
198 
199 	MCLAIM(m, &ip_tx_mowner);
200 
201 	KASSERT((m->m_flags & M_PKTHDR) != 0);
202 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
203 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) !=
204 	    (M_CSUM_TCPv4|M_CSUM_UDPv4));
205 
206 	if (opt) {
207 		m = ip_insertoptions(m, opt, &len);
208 		if (len >= sizeof(struct ip))
209 			hlen = len;
210 	}
211 	ip = mtod(m, struct ip *);
212 
213 	/*
214 	 * Fill in IP header.
215 	 */
216 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
217 		ip->ip_v = IPVERSION;
218 		ip->ip_off = htons(0);
219 		/* ip->ip_id filled in after we find out source ia */
220 		ip->ip_hl = hlen >> 2;
221 		IP_STATINC(IP_STAT_LOCALOUT);
222 	} else {
223 		hlen = ip->ip_hl << 2;
224 	}
225 
226 	/*
227 	 * Route packet.
228 	 */
229 	if (ro == NULL) {
230 		memset(&iproute, 0, sizeof(iproute));
231 		ro = &iproute;
232 	}
233 	sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
234 	dst = satocsin(rtcache_getdst(ro));
235 
236 	/*
237 	 * If there is a cached route, check that it is to the same
238 	 * destination and is still up.  If not, free it and try again.
239 	 * The address family should also be checked in case of sharing
240 	 * the cache with IPv6.
241 	 */
242 	if (dst && (dst->sin_family != AF_INET ||
243 	    !in_hosteq(dst->sin_addr, ip->ip_dst)))
244 		rtcache_free(ro);
245 
246 	if ((rt = rtcache_validate(ro)) == NULL &&
247 	    (rt = rtcache_update(ro, 1)) == NULL) {
248 		dst = &u.dst4;
249 		rtcache_setdst(ro, &u.dst);
250 	}
251 
252 	/*
253 	 * If routing to interface only, short circuit routing lookup.
254 	 */
255 	if (flags & IP_ROUTETOIF) {
256 		if ((ia = ifatoia(ifa_ifwithladdr(sintocsa(dst)))) == NULL) {
257 			IP_STATINC(IP_STAT_NOROUTE);
258 			error = ENETUNREACH;
259 			goto bad;
260 		}
261 		ifp = ia->ia_ifp;
262 		mtu = ifp->if_mtu;
263 		ip->ip_ttl = 1;
264 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
265 	} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
266 	    ip->ip_dst.s_addr == INADDR_BROADCAST) &&
267 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
268 		ifp = imo->imo_multicast_ifp;
269 		mtu = ifp->if_mtu;
270 		IFP_TO_IA(ifp, ia);
271 		isbroadcast = 0;
272 	} else {
273 		if (rt == NULL)
274 			rt = rtcache_init(ro);
275 		if (rt == NULL) {
276 			IP_STATINC(IP_STAT_NOROUTE);
277 			error = EHOSTUNREACH;
278 			goto bad;
279 		}
280 		ia = ifatoia(rt->rt_ifa);
281 		ifp = rt->rt_ifp;
282 		if ((mtu = rt->rt_rmx.rmx_mtu) == 0)
283 			mtu = ifp->if_mtu;
284 		rt->rt_use++;
285 		if (rt->rt_flags & RTF_GATEWAY)
286 			dst = satosin(rt->rt_gateway);
287 		if (rt->rt_flags & RTF_HOST)
288 			isbroadcast = rt->rt_flags & RTF_BROADCAST;
289 		else
290 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
291 	}
292 	rtmtu_nolock = rt && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0;
293 
294 	if (IN_MULTICAST(ip->ip_dst.s_addr) ||
295 	    (ip->ip_dst.s_addr == INADDR_BROADCAST)) {
296 		bool inmgroup;
297 
298 		m->m_flags |= (ip->ip_dst.s_addr == INADDR_BROADCAST) ?
299 			M_BCAST : M_MCAST;
300 		/*
301 		 * See if the caller provided any multicast options
302 		 */
303 		if (imo != NULL)
304 			ip->ip_ttl = imo->imo_multicast_ttl;
305 		else
306 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
307 
308 		/*
309 		 * if we don't know the outgoing ifp yet, we can't generate
310 		 * output
311 		 */
312 		if (!ifp) {
313 			IP_STATINC(IP_STAT_NOROUTE);
314 			error = ENETUNREACH;
315 			goto bad;
316 		}
317 
318 		/*
319 		 * If the packet is multicast or broadcast, confirm that
320 		 * the outgoing interface can transmit it.
321 		 */
322 		if (((m->m_flags & M_MCAST) &&
323 		     (ifp->if_flags & IFF_MULTICAST) == 0) ||
324 		    ((m->m_flags & M_BCAST) &&
325 		     (ifp->if_flags & (IFF_BROADCAST|IFF_POINTOPOINT)) == 0))  {
326 			IP_STATINC(IP_STAT_NOROUTE);
327 			error = ENETUNREACH;
328 			goto bad;
329 		}
330 		/*
331 		 * If source address not specified yet, use an address
332 		 * of outgoing interface.
333 		 */
334 		if (in_nullhost(ip->ip_src)) {
335 			struct in_ifaddr *xia;
336 			struct ifaddr *xifa;
337 
338 			IFP_TO_IA(ifp, xia);
339 			if (!xia) {
340 				error = EADDRNOTAVAIL;
341 				goto bad;
342 			}
343 			xifa = &xia->ia_ifa;
344 			if (xifa->ifa_getifa != NULL) {
345 				xia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
346 			}
347 			ip->ip_src = xia->ia_addr.sin_addr;
348 		}
349 
350 		inmgroup = in_multi_group(ip->ip_dst, ifp, flags);
351 		if (inmgroup && (imo == NULL || imo->imo_multicast_loop)) {
352 			/*
353 			 * If we belong to the destination multicast group
354 			 * on the outgoing interface, and the caller did not
355 			 * forbid loopback, loop back a copy.
356 			 */
357 			ip_mloopback(ifp, m, &u.dst4);
358 		}
359 #ifdef MROUTING
360 		else {
361 			/*
362 			 * If we are acting as a multicast router, perform
363 			 * multicast forwarding as if the packet had just
364 			 * arrived on the interface to which we are about
365 			 * to send.  The multicast forwarding function
366 			 * recursively calls this function, using the
367 			 * IP_FORWARDING flag to prevent infinite recursion.
368 			 *
369 			 * Multicasts that are looped back by ip_mloopback(),
370 			 * above, will be forwarded by the ip_input() routine,
371 			 * if necessary.
372 			 */
373 			extern struct socket *ip_mrouter;
374 
375 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
376 				if (ip_mforward(m, ifp) != 0) {
377 					m_freem(m);
378 					goto done;
379 				}
380 			}
381 		}
382 #endif
383 		/*
384 		 * Multicasts with a time-to-live of zero may be looped-
385 		 * back, above, but must not be transmitted on a network.
386 		 * Also, multicasts addressed to the loopback interface
387 		 * are not sent -- the above call to ip_mloopback() will
388 		 * loop back a copy if this host actually belongs to the
389 		 * destination group on the loopback interface.
390 		 */
391 		if (ip->ip_ttl == 0 || (ifp->if_flags & IFF_LOOPBACK) != 0) {
392 			m_freem(m);
393 			goto done;
394 		}
395 		goto sendit;
396 	}
397 
398 	/*
399 	 * If source address not specified yet, use address
400 	 * of outgoing interface.
401 	 */
402 	if (in_nullhost(ip->ip_src)) {
403 		struct ifaddr *xifa;
404 
405 		xifa = &ia->ia_ifa;
406 		if (xifa->ifa_getifa != NULL)
407 			ia = ifatoia((*xifa->ifa_getifa)(xifa, rdst));
408 		ip->ip_src = ia->ia_addr.sin_addr;
409 	}
410 
411 	/*
412 	 * packets with Class-D address as source are not valid per
413 	 * RFC 1112
414 	 */
415 	if (IN_MULTICAST(ip->ip_src.s_addr)) {
416 		IP_STATINC(IP_STAT_ODROPPED);
417 		error = EADDRNOTAVAIL;
418 		goto bad;
419 	}
420 
421 	/*
422 	 * Look for broadcast address and and verify user is allowed to
423 	 * send such a packet.
424 	 */
425 	if (isbroadcast) {
426 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
427 			error = EADDRNOTAVAIL;
428 			goto bad;
429 		}
430 		if ((flags & IP_ALLOWBROADCAST) == 0) {
431 			error = EACCES;
432 			goto bad;
433 		}
434 		/* don't allow broadcast messages to be fragmented */
435 		if (ntohs(ip->ip_len) > ifp->if_mtu) {
436 			error = EMSGSIZE;
437 			goto bad;
438 		}
439 		m->m_flags |= M_BCAST;
440 	} else
441 		m->m_flags &= ~M_BCAST;
442 
443 sendit:
444 	if ((flags & (IP_FORWARDING|IP_NOIPNEWID)) == 0) {
445 		if (m->m_pkthdr.len < IP_MINFRAGSIZE) {
446 			ip->ip_id = 0;
447 		} else if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
448 			ip->ip_id = ip_newid(ia);
449 		} else {
450 
451 			/*
452 			 * TSO capable interfaces (typically?) increment
453 			 * ip_id for each segment.
454 			 * "allocate" enough ids here to increase the chance
455 			 * for them to be unique.
456 			 *
457 			 * note that the following calculation is not
458 			 * needed to be precise.  wasting some ip_id is fine.
459 			 */
460 
461 			unsigned int segsz = m->m_pkthdr.segsz;
462 			unsigned int datasz = ntohs(ip->ip_len) - hlen;
463 			unsigned int num = howmany(datasz, segsz);
464 
465 			ip->ip_id = ip_newid_range(ia, num);
466 		}
467 	}
468 
469 	/*
470 	 * If we're doing Path MTU Discovery, we need to set DF unless
471 	 * the route's MTU is locked.
472 	 */
473 	if ((flags & IP_MTUDISC) != 0 && rtmtu_nolock) {
474 		ip->ip_off |= htons(IP_DF);
475 	}
476 
477 #ifdef IPSEC
478 	if (ipsec_used) {
479 		bool ipsec_done = false;
480 
481 		/* Perform IPsec processing, if any. */
482 		error = ipsec4_output(m, so, flags, &sp, &mtu, &natt_frag,
483 		    &ipsec_done);
484 		if (error || ipsec_done)
485 			goto done;
486 	}
487 #endif
488 
489 	/*
490 	 * Run through list of hooks for output packets.
491 	 */
492 	error = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_OUT);
493 	if (error)
494 		goto done;
495 	if (m == NULL)
496 		goto done;
497 
498 	ip = mtod(m, struct ip *);
499 	hlen = ip->ip_hl << 2;
500 
501 	m->m_pkthdr.csum_data |= hlen << 16;
502 
503 #if IFA_STATS
504 	/*
505 	 * search for the source address structure to
506 	 * maintain output statistics.
507 	 */
508 	INADDR_TO_IA(ip->ip_src, ia);
509 #endif
510 
511 	/* Maybe skip checksums on loopback interfaces. */
512 	if (IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4)) {
513 		m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
514 	}
515 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx;
516 	/*
517 	 * If small enough for mtu of path, or if using TCP segmentation
518 	 * offload, can just send directly.
519 	 */
520 	if (ntohs(ip->ip_len) <= mtu ||
521 	    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
522 		const struct sockaddr *sa;
523 
524 #if IFA_STATS
525 		if (ia)
526 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
527 #endif
528 		/*
529 		 * Always initialize the sum to 0!  Some HW assisted
530 		 * checksumming requires this.
531 		 */
532 		ip->ip_sum = 0;
533 
534 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0) {
535 			/*
536 			 * Perform any checksums that the hardware can't do
537 			 * for us.
538 			 *
539 			 * XXX Does any hardware require the {th,uh}_sum
540 			 * XXX fields to be 0?
541 			 */
542 			if (sw_csum & M_CSUM_IPv4) {
543 				KASSERT(IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4));
544 				ip->ip_sum = in_cksum(m, hlen);
545 				m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
546 			}
547 			if (sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
548 				if (IN_NEED_CHECKSUM(ifp,
549 				    sw_csum & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
550 					in_delayed_cksum(m);
551 				}
552 				m->m_pkthdr.csum_flags &=
553 				    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
554 			}
555 		}
556 
557 		sa = (m->m_flags & M_MCAST) ? sintocsa(rdst) : sintocsa(dst);
558 		if (__predict_true(
559 		    (m->m_pkthdr.csum_flags & M_CSUM_TSOv4) == 0 ||
560 		    (ifp->if_capenable & IFCAP_TSOv4) != 0)) {
561 #ifndef NET_MPSAFE
562 			KERNEL_LOCK(1, NULL);
563 #endif
564 			error = (*ifp->if_output)(ifp, m, sa, rt);
565 #ifndef NET_MPSAFE
566 			KERNEL_UNLOCK_ONE(NULL);
567 #endif
568 		} else {
569 			error = ip_tso_output(ifp, m, sa, rt);
570 		}
571 		goto done;
572 	}
573 
574 	/*
575 	 * We can't use HW checksumming if we're about to
576 	 * to fragment the packet.
577 	 *
578 	 * XXX Some hardware can do this.
579 	 */
580 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
581 		if (IN_NEED_CHECKSUM(ifp,
582 		    m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))) {
583 			in_delayed_cksum(m);
584 		}
585 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
586 	}
587 
588 	/*
589 	 * Too large for interface; fragment if possible.
590 	 * Must be able to put at least 8 bytes per fragment.
591 	 */
592 	if (ntohs(ip->ip_off) & IP_DF) {
593 		if (flags & IP_RETURNMTU) {
594 			struct inpcb *inp;
595 
596 			KASSERT(so && solocked(so));
597 			inp = sotoinpcb(so);
598 			inp->inp_errormtu = mtu;
599 		}
600 		error = EMSGSIZE;
601 		IP_STATINC(IP_STAT_CANTFRAG);
602 		goto bad;
603 	}
604 
605 	error = ip_fragment(m, ifp, mtu);
606 	if (error) {
607 		m = NULL;
608 		goto bad;
609 	}
610 
611 	for (; m; m = m0) {
612 		m0 = m->m_nextpkt;
613 		m->m_nextpkt = 0;
614 		if (error) {
615 			m_freem(m);
616 			continue;
617 		}
618 #if IFA_STATS
619 		if (ia)
620 			ia->ia_ifa.ifa_data.ifad_outbytes += ntohs(ip->ip_len);
621 #endif
622 		/*
623 		 * If we get there, the packet has not been handled by
624 		 * IPsec whereas it should have. Now that it has been
625 		 * fragmented, re-inject it in ip_output so that IPsec
626 		 * processing can occur.
627 		 */
628 		if (natt_frag) {
629 			error = ip_output(m, opt, ro,
630 			    flags | IP_RAWOUTPUT | IP_NOIPNEWID,
631 			    imo, so);
632 		} else {
633 			KASSERT((m->m_pkthdr.csum_flags &
634 			    (M_CSUM_UDPv4 | M_CSUM_TCPv4)) == 0);
635 #ifndef NET_MPSAFE
636 			KERNEL_LOCK(1, NULL);
637 #endif
638 			error = (*ifp->if_output)(ifp, m,
639 			    (m->m_flags & M_MCAST) ?
640 			    sintocsa(rdst) : sintocsa(dst), rt);
641 #ifndef NET_MPSAFE
642 			KERNEL_UNLOCK_ONE(NULL);
643 #endif
644 		}
645 	}
646 	if (error == 0) {
647 		IP_STATINC(IP_STAT_FRAGMENTED);
648 	}
649 done:
650 	if (ro == &iproute) {
651 		rtcache_free(&iproute);
652 	}
653 #ifdef IPSEC
654 	if (sp) {
655 		KEY_FREESP(&sp);
656 	}
657 #endif
658 	return error;
659 bad:
660 	m_freem(m);
661 	goto done;
662 }
663 
664 int
665 ip_fragment(struct mbuf *m, struct ifnet *ifp, u_long mtu)
666 {
667 	struct ip *ip, *mhip;
668 	struct mbuf *m0;
669 	int len, hlen, off;
670 	int mhlen, firstlen;
671 	struct mbuf **mnext;
672 	int sw_csum = m->m_pkthdr.csum_flags;
673 	int fragments = 0;
674 	int s;
675 	int error = 0;
676 
677 	ip = mtod(m, struct ip *);
678 	hlen = ip->ip_hl << 2;
679 	if (ifp != NULL)
680 		sw_csum &= ~ifp->if_csum_flags_tx;
681 
682 	len = (mtu - hlen) &~ 7;
683 	if (len < 8) {
684 		m_freem(m);
685 		return (EMSGSIZE);
686 	}
687 
688 	firstlen = len;
689 	mnext = &m->m_nextpkt;
690 
691 	/*
692 	 * Loop through length of segment after first fragment,
693 	 * make new header and copy data of each part and link onto chain.
694 	 */
695 	m0 = m;
696 	mhlen = sizeof (struct ip);
697 	for (off = hlen + len; off < ntohs(ip->ip_len); off += len) {
698 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
699 		if (m == 0) {
700 			error = ENOBUFS;
701 			IP_STATINC(IP_STAT_ODROPPED);
702 			goto sendorfree;
703 		}
704 		MCLAIM(m, m0->m_owner);
705 		*mnext = m;
706 		mnext = &m->m_nextpkt;
707 		m->m_data += max_linkhdr;
708 		mhip = mtod(m, struct ip *);
709 		*mhip = *ip;
710 		/* we must inherit MCAST and BCAST flags */
711 		m->m_flags |= m0->m_flags & (M_MCAST|M_BCAST);
712 		if (hlen > sizeof (struct ip)) {
713 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
714 			mhip->ip_hl = mhlen >> 2;
715 		}
716 		m->m_len = mhlen;
717 		mhip->ip_off = ((off - hlen) >> 3) +
718 		    (ntohs(ip->ip_off) & ~IP_MF);
719 		if (ip->ip_off & htons(IP_MF))
720 			mhip->ip_off |= IP_MF;
721 		if (off + len >= ntohs(ip->ip_len))
722 			len = ntohs(ip->ip_len) - off;
723 		else
724 			mhip->ip_off |= IP_MF;
725 		HTONS(mhip->ip_off);
726 		mhip->ip_len = htons((u_int16_t)(len + mhlen));
727 		m->m_next = m_copym(m0, off, len, M_DONTWAIT);
728 		if (m->m_next == 0) {
729 			error = ENOBUFS;	/* ??? */
730 			IP_STATINC(IP_STAT_ODROPPED);
731 			goto sendorfree;
732 		}
733 		m->m_pkthdr.len = mhlen + len;
734 		m->m_pkthdr.rcvif = NULL;
735 		mhip->ip_sum = 0;
736 		KASSERT((m->m_pkthdr.csum_flags & M_CSUM_IPv4) == 0);
737 		if (sw_csum & M_CSUM_IPv4) {
738 			mhip->ip_sum = in_cksum(m, mhlen);
739 		} else {
740 			/*
741 			 * checksum is hw-offloaded or not necessary.
742 			 */
743 			m->m_pkthdr.csum_flags |=
744 			    m0->m_pkthdr.csum_flags & M_CSUM_IPv4;
745 			m->m_pkthdr.csum_data |= mhlen << 16;
746 			KASSERT(!(ifp != NULL &&
747 			    IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4))
748 			    || (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
749 		}
750 		IP_STATINC(IP_STAT_OFRAGMENTS);
751 		fragments++;
752 	}
753 	/*
754 	 * Update first fragment by trimming what's been copied out
755 	 * and updating header, then send each fragment (in order).
756 	 */
757 	m = m0;
758 	m_adj(m, hlen + firstlen - ntohs(ip->ip_len));
759 	m->m_pkthdr.len = hlen + firstlen;
760 	ip->ip_len = htons((u_int16_t)m->m_pkthdr.len);
761 	ip->ip_off |= htons(IP_MF);
762 	ip->ip_sum = 0;
763 	if (sw_csum & M_CSUM_IPv4) {
764 		ip->ip_sum = in_cksum(m, hlen);
765 		m->m_pkthdr.csum_flags &= ~M_CSUM_IPv4;
766 	} else {
767 		/*
768 		 * checksum is hw-offloaded or not necessary.
769 		 */
770 		KASSERT(!(ifp != NULL && IN_NEED_CHECKSUM(ifp, M_CSUM_IPv4))
771 		   || (m->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0);
772 		KASSERT(M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data) >=
773 			sizeof(struct ip));
774 	}
775 sendorfree:
776 	/*
777 	 * If there is no room for all the fragments, don't queue
778 	 * any of them.
779 	 */
780 	if (ifp != NULL) {
781 		s = splnet();
782 		if (ifp->if_snd.ifq_maxlen - ifp->if_snd.ifq_len < fragments &&
783 		    error == 0) {
784 			error = ENOBUFS;
785 			IP_STATINC(IP_STAT_ODROPPED);
786 			IFQ_INC_DROPS(&ifp->if_snd);
787 		}
788 		splx(s);
789 	}
790 	if (error) {
791 		for (m = m0; m; m = m0) {
792 			m0 = m->m_nextpkt;
793 			m->m_nextpkt = NULL;
794 			m_freem(m);
795 		}
796 	}
797 	return (error);
798 }
799 
800 /*
801  * Process a delayed payload checksum calculation.
802  */
803 void
804 in_delayed_cksum(struct mbuf *m)
805 {
806 	struct ip *ip;
807 	u_int16_t csum, offset;
808 
809 	ip = mtod(m, struct ip *);
810 	offset = ip->ip_hl << 2;
811 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
812 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
813 		csum = 0xffff;
814 
815 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
816 
817 	if ((offset + sizeof(u_int16_t)) > m->m_len) {
818 		/* This happen when ip options were inserted
819 		printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
820 		    m->m_len, offset, ip->ip_p);
821 		 */
822 		m_copyback(m, offset, sizeof(csum), (void *) &csum);
823 	} else
824 		*(u_int16_t *)(mtod(m, char *) + offset) = csum;
825 }
826 
827 /*
828  * Determine the maximum length of the options to be inserted;
829  * we would far rather allocate too much space rather than too little.
830  */
831 
832 u_int
833 ip_optlen(struct inpcb *inp)
834 {
835 	struct mbuf *m = inp->inp_options;
836 
837 	if (m && m->m_len > offsetof(struct ipoption, ipopt_dst)) {
838 		return (m->m_len - offsetof(struct ipoption, ipopt_dst));
839 	}
840 	return 0;
841 }
842 
843 /*
844  * Insert IP options into preformed packet.
845  * Adjust IP destination as required for IP source routing,
846  * as indicated by a non-zero in_addr at the start of the options.
847  */
848 static struct mbuf *
849 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
850 {
851 	struct ipoption *p = mtod(opt, struct ipoption *);
852 	struct mbuf *n;
853 	struct ip *ip = mtod(m, struct ip *);
854 	unsigned optlen;
855 
856 	optlen = opt->m_len - sizeof(p->ipopt_dst);
857 	if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET)
858 		return (m);		/* XXX should fail */
859 	if (!in_nullhost(p->ipopt_dst))
860 		ip->ip_dst = p->ipopt_dst;
861 	if (M_READONLY(m) || M_LEADINGSPACE(m) < optlen) {
862 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
863 		if (n == 0)
864 			return (m);
865 		MCLAIM(n, m->m_owner);
866 		M_MOVE_PKTHDR(n, m);
867 		m->m_len -= sizeof(struct ip);
868 		m->m_data += sizeof(struct ip);
869 		n->m_next = m;
870 		m = n;
871 		m->m_len = optlen + sizeof(struct ip);
872 		m->m_data += max_linkhdr;
873 		bcopy((void *)ip, mtod(m, void *), sizeof(struct ip));
874 	} else {
875 		m->m_data -= optlen;
876 		m->m_len += optlen;
877 		memmove(mtod(m, void *), ip, sizeof(struct ip));
878 	}
879 	m->m_pkthdr.len += optlen;
880 	ip = mtod(m, struct ip *);
881 	bcopy((void *)p->ipopt_list, (void *)(ip + 1), (unsigned)optlen);
882 	*phlen = sizeof(struct ip) + optlen;
883 	ip->ip_len = htons(ntohs(ip->ip_len) + optlen);
884 	return (m);
885 }
886 
887 /*
888  * Copy options from ip to jp,
889  * omitting those not copied during fragmentation.
890  */
891 int
892 ip_optcopy(struct ip *ip, struct ip *jp)
893 {
894 	u_char *cp, *dp;
895 	int opt, optlen, cnt;
896 
897 	cp = (u_char *)(ip + 1);
898 	dp = (u_char *)(jp + 1);
899 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
900 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
901 		opt = cp[0];
902 		if (opt == IPOPT_EOL)
903 			break;
904 		if (opt == IPOPT_NOP) {
905 			/* Preserve for IP mcast tunnel's LSRR alignment. */
906 			*dp++ = IPOPT_NOP;
907 			optlen = 1;
908 			continue;
909 		}
910 
911 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp));
912 		optlen = cp[IPOPT_OLEN];
913 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen < cnt);
914 
915 		/* Invalid lengths should have been caught by ip_dooptions. */
916 		if (optlen > cnt)
917 			optlen = cnt;
918 		if (IPOPT_COPIED(opt)) {
919 			bcopy((void *)cp, (void *)dp, (unsigned)optlen);
920 			dp += optlen;
921 		}
922 	}
923 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
924 		*dp++ = IPOPT_EOL;
925 	return (optlen);
926 }
927 
928 /*
929  * IP socket option processing.
930  */
931 int
932 ip_ctloutput(int op, struct socket *so, struct sockopt *sopt)
933 {
934 	struct inpcb *inp = sotoinpcb(so);
935 	struct ip *ip = &inp->inp_ip;
936 	int inpflags = inp->inp_flags;
937 	int optval = 0, error = 0;
938 
939 	if (sopt->sopt_level != IPPROTO_IP) {
940 		if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_NOHEADER)
941 			return 0;
942 		return ENOPROTOOPT;
943 	}
944 
945 	switch (op) {
946 	case PRCO_SETOPT:
947 		switch (sopt->sopt_name) {
948 		case IP_OPTIONS:
949 #ifdef notyet
950 		case IP_RETOPTS:
951 #endif
952 			error = ip_pcbopts(inp, sopt);
953 			break;
954 
955 		case IP_TOS:
956 		case IP_TTL:
957 		case IP_MINTTL:
958 		case IP_PKTINFO:
959 		case IP_RECVOPTS:
960 		case IP_RECVRETOPTS:
961 		case IP_RECVDSTADDR:
962 		case IP_RECVIF:
963 		case IP_RECVPKTINFO:
964 		case IP_RECVTTL:
965 			error = sockopt_getint(sopt, &optval);
966 			if (error)
967 				break;
968 
969 			switch (sopt->sopt_name) {
970 			case IP_TOS:
971 				ip->ip_tos = optval;
972 				break;
973 
974 			case IP_TTL:
975 				ip->ip_ttl = optval;
976 				break;
977 
978 			case IP_MINTTL:
979 				if (optval > 0 && optval <= MAXTTL)
980 					inp->inp_ip_minttl = optval;
981 				else
982 					error = EINVAL;
983 				break;
984 #define	OPTSET(bit) \
985 	if (optval) \
986 		inpflags |= bit; \
987 	else \
988 		inpflags &= ~bit;
989 
990 			case IP_PKTINFO:
991 				OPTSET(INP_PKTINFO);
992 				break;
993 
994 			case IP_RECVOPTS:
995 				OPTSET(INP_RECVOPTS);
996 				break;
997 
998 			case IP_RECVPKTINFO:
999 				OPTSET(INP_RECVPKTINFO);
1000 				break;
1001 
1002 			case IP_RECVRETOPTS:
1003 				OPTSET(INP_RECVRETOPTS);
1004 				break;
1005 
1006 			case IP_RECVDSTADDR:
1007 				OPTSET(INP_RECVDSTADDR);
1008 				break;
1009 
1010 			case IP_RECVIF:
1011 				OPTSET(INP_RECVIF);
1012 				break;
1013 
1014 			case IP_RECVTTL:
1015 				OPTSET(INP_RECVTTL);
1016 				break;
1017 			}
1018 		break;
1019 #undef OPTSET
1020 
1021 		case IP_MULTICAST_IF:
1022 		case IP_MULTICAST_TTL:
1023 		case IP_MULTICAST_LOOP:
1024 		case IP_ADD_MEMBERSHIP:
1025 		case IP_DROP_MEMBERSHIP:
1026 			error = ip_setmoptions(&inp->inp_moptions, sopt);
1027 			break;
1028 
1029 		case IP_PORTRANGE:
1030 			error = sockopt_getint(sopt, &optval);
1031 			if (error)
1032 				break;
1033 
1034 			switch (optval) {
1035 			case IP_PORTRANGE_DEFAULT:
1036 			case IP_PORTRANGE_HIGH:
1037 				inpflags &= ~(INP_LOWPORT);
1038 				break;
1039 
1040 			case IP_PORTRANGE_LOW:
1041 				inpflags |= INP_LOWPORT;
1042 				break;
1043 
1044 			default:
1045 				error = EINVAL;
1046 				break;
1047 			}
1048 			break;
1049 
1050 		case IP_PORTALGO:
1051 			error = sockopt_getint(sopt, &optval);
1052 			if (error)
1053 				break;
1054 
1055 			error = portalgo_algo_index_select(
1056 			    (struct inpcb_hdr *)inp, optval);
1057 			break;
1058 
1059 #if defined(IPSEC)
1060 		case IP_IPSEC_POLICY:
1061 			if (ipsec_enabled) {
1062 				error = ipsec4_set_policy(inp, sopt->sopt_name,
1063 				    sopt->sopt_data, sopt->sopt_size,
1064 				    curlwp->l_cred);
1065 				break;
1066 			}
1067 			/*FALLTHROUGH*/
1068 #endif /* IPSEC */
1069 
1070 		default:
1071 			error = ENOPROTOOPT;
1072 			break;
1073 		}
1074 		break;
1075 
1076 	case PRCO_GETOPT:
1077 		switch (sopt->sopt_name) {
1078 		case IP_OPTIONS:
1079 		case IP_RETOPTS: {
1080 			struct mbuf *mopts = inp->inp_options;
1081 
1082 			if (mopts) {
1083 				struct mbuf *m;
1084 
1085 				m = m_copym(mopts, 0, M_COPYALL, M_DONTWAIT);
1086 				if (m == NULL) {
1087 					error = ENOBUFS;
1088 					break;
1089 				}
1090 				error = sockopt_setmbuf(sopt, m);
1091 			}
1092 			break;
1093 		}
1094 		case IP_PKTINFO:
1095 		case IP_TOS:
1096 		case IP_TTL:
1097 		case IP_MINTTL:
1098 		case IP_RECVOPTS:
1099 		case IP_RECVRETOPTS:
1100 		case IP_RECVDSTADDR:
1101 		case IP_RECVIF:
1102 		case IP_RECVPKTINFO:
1103 		case IP_RECVTTL:
1104 		case IP_ERRORMTU:
1105 			switch (sopt->sopt_name) {
1106 			case IP_TOS:
1107 				optval = ip->ip_tos;
1108 				break;
1109 
1110 			case IP_TTL:
1111 				optval = ip->ip_ttl;
1112 				break;
1113 
1114 			case IP_MINTTL:
1115 				optval = inp->inp_ip_minttl;
1116 				break;
1117 
1118 			case IP_ERRORMTU:
1119 				optval = inp->inp_errormtu;
1120 				break;
1121 
1122 #define	OPTBIT(bit)	(inpflags & bit ? 1 : 0)
1123 
1124 			case IP_PKTINFO:
1125 				optval = OPTBIT(INP_PKTINFO);
1126 				break;
1127 
1128 			case IP_RECVOPTS:
1129 				optval = OPTBIT(INP_RECVOPTS);
1130 				break;
1131 
1132 			case IP_RECVPKTINFO:
1133 				optval = OPTBIT(INP_RECVPKTINFO);
1134 				break;
1135 
1136 			case IP_RECVRETOPTS:
1137 				optval = OPTBIT(INP_RECVRETOPTS);
1138 				break;
1139 
1140 			case IP_RECVDSTADDR:
1141 				optval = OPTBIT(INP_RECVDSTADDR);
1142 				break;
1143 
1144 			case IP_RECVIF:
1145 				optval = OPTBIT(INP_RECVIF);
1146 				break;
1147 
1148 			case IP_RECVTTL:
1149 				optval = OPTBIT(INP_RECVTTL);
1150 				break;
1151 			}
1152 			error = sockopt_setint(sopt, optval);
1153 			break;
1154 
1155 #if 0	/* defined(IPSEC) */
1156 		case IP_IPSEC_POLICY:
1157 		{
1158 			struct mbuf *m = NULL;
1159 
1160 			/* XXX this will return EINVAL as sopt is empty */
1161 			error = ipsec4_get_policy(inp, sopt->sopt_data,
1162 			    sopt->sopt_size, &m);
1163 			if (error == 0)
1164 				error = sockopt_setmbuf(sopt, m);
1165 			break;
1166 		}
1167 #endif /*IPSEC*/
1168 
1169 		case IP_MULTICAST_IF:
1170 		case IP_MULTICAST_TTL:
1171 		case IP_MULTICAST_LOOP:
1172 		case IP_ADD_MEMBERSHIP:
1173 		case IP_DROP_MEMBERSHIP:
1174 			error = ip_getmoptions(inp->inp_moptions, sopt);
1175 			break;
1176 
1177 		case IP_PORTRANGE:
1178 			if (inpflags & INP_LOWPORT)
1179 				optval = IP_PORTRANGE_LOW;
1180 			else
1181 				optval = IP_PORTRANGE_DEFAULT;
1182 			error = sockopt_setint(sopt, optval);
1183 			break;
1184 
1185 		case IP_PORTALGO:
1186 			optval = inp->inp_portalgo;
1187 			error = sockopt_setint(sopt, optval);
1188 			break;
1189 
1190 		default:
1191 			error = ENOPROTOOPT;
1192 			break;
1193 		}
1194 		break;
1195 	}
1196 
1197 	if (!error) {
1198 		inp->inp_flags = inpflags;
1199 	}
1200 	return error;
1201 }
1202 
1203 /*
1204  * Set up IP options in pcb for insertion in output packets.
1205  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1206  * with destination address if source routed.
1207  */
1208 static int
1209 ip_pcbopts(struct inpcb *inp, const struct sockopt *sopt)
1210 {
1211 	struct mbuf *m;
1212 	const u_char *cp;
1213 	u_char *dp;
1214 	int cnt;
1215 
1216 	/* Turn off any old options. */
1217 	if (inp->inp_options) {
1218 		m_free(inp->inp_options);
1219 	}
1220 	inp->inp_options = NULL;
1221 	if ((cnt = sopt->sopt_size) == 0) {
1222 		/* Only turning off any previous options. */
1223 		return 0;
1224 	}
1225 	cp = sopt->sopt_data;
1226 
1227 #ifndef	__vax__
1228 	if (cnt % sizeof(int32_t))
1229 		return (EINVAL);
1230 #endif
1231 
1232 	m = m_get(M_DONTWAIT, MT_SOOPTS);
1233 	if (m == NULL)
1234 		return (ENOBUFS);
1235 
1236 	dp = mtod(m, u_char *);
1237 	memset(dp, 0, sizeof(struct in_addr));
1238 	dp += sizeof(struct in_addr);
1239 	m->m_len = sizeof(struct in_addr);
1240 
1241 	/*
1242 	 * IP option list according to RFC791. Each option is of the form
1243 	 *
1244 	 *	[optval] [olen] [(olen - 2) data bytes]
1245 	 *
1246 	 * We validate the list and copy options to an mbuf for prepending
1247 	 * to data packets. The IP first-hop destination address will be
1248 	 * stored before actual options and is zero if unset.
1249 	 */
1250 	while (cnt > 0) {
1251 		uint8_t optval, olen, offset;
1252 
1253 		optval = cp[IPOPT_OPTVAL];
1254 
1255 		if (optval == IPOPT_EOL || optval == IPOPT_NOP) {
1256 			olen = 1;
1257 		} else {
1258 			if (cnt < IPOPT_OLEN + 1)
1259 				goto bad;
1260 
1261 			olen = cp[IPOPT_OLEN];
1262 			if (olen < IPOPT_OLEN + 1 || olen > cnt)
1263 				goto bad;
1264 		}
1265 
1266 		if (optval == IPOPT_LSRR || optval == IPOPT_SSRR) {
1267 			/*
1268 			 * user process specifies route as:
1269 			 *	->A->B->C->D
1270 			 * D must be our final destination (but we can't
1271 			 * check that since we may not have connected yet).
1272 			 * A is first hop destination, which doesn't appear in
1273 			 * actual IP option, but is stored before the options.
1274 			 */
1275 			if (olen < IPOPT_OFFSET + 1 + sizeof(struct in_addr))
1276 				goto bad;
1277 
1278 			offset = cp[IPOPT_OFFSET];
1279 			memcpy(mtod(m, u_char *), cp + IPOPT_OFFSET + 1,
1280 			    sizeof(struct in_addr));
1281 
1282 			cp += sizeof(struct in_addr);
1283 			cnt -= sizeof(struct in_addr);
1284 			olen -= sizeof(struct in_addr);
1285 
1286 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1287 				goto bad;
1288 
1289 			memcpy(dp, cp, olen);
1290 			dp[IPOPT_OPTVAL] = optval;
1291 			dp[IPOPT_OLEN] = olen;
1292 			dp[IPOPT_OFFSET] = offset;
1293 			break;
1294 		} else {
1295 			if (m->m_len + olen > MAX_IPOPTLEN + sizeof(struct in_addr))
1296 				goto bad;
1297 
1298 			memcpy(dp, cp, olen);
1299 			break;
1300 		}
1301 
1302 		dp += olen;
1303 		m->m_len += olen;
1304 
1305 		if (optval == IPOPT_EOL)
1306 			break;
1307 
1308 		cp += olen;
1309 		cnt -= olen;
1310 	}
1311 
1312 	inp->inp_options = m;
1313 	return 0;
1314 bad:
1315 	(void)m_free(m);
1316 	return EINVAL;
1317 }
1318 
1319 /*
1320  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1321  */
1322 static struct ifnet *
1323 ip_multicast_if(struct in_addr *a, int *ifindexp)
1324 {
1325 	int ifindex;
1326 	struct ifnet *ifp = NULL;
1327 	struct in_ifaddr *ia;
1328 
1329 	if (ifindexp)
1330 		*ifindexp = 0;
1331 	if (ntohl(a->s_addr) >> 24 == 0) {
1332 		ifindex = ntohl(a->s_addr) & 0xffffff;
1333 		ifp = if_byindex(ifindex);
1334 		if (!ifp)
1335 			return NULL;
1336 		if (ifindexp)
1337 			*ifindexp = ifindex;
1338 	} else {
1339 		LIST_FOREACH(ia, &IN_IFADDR_HASH(a->s_addr), ia_hash) {
1340 			if (in_hosteq(ia->ia_addr.sin_addr, *a) &&
1341 			    (ia->ia_ifp->if_flags & IFF_MULTICAST) != 0) {
1342 				ifp = ia->ia_ifp;
1343 				break;
1344 			}
1345 		}
1346 	}
1347 	return ifp;
1348 }
1349 
1350 static int
1351 ip_getoptval(const struct sockopt *sopt, u_int8_t *val, u_int maxval)
1352 {
1353 	u_int tval;
1354 	u_char cval;
1355 	int error;
1356 
1357 	if (sopt == NULL)
1358 		return EINVAL;
1359 
1360 	switch (sopt->sopt_size) {
1361 	case sizeof(u_char):
1362 		error = sockopt_get(sopt, &cval, sizeof(u_char));
1363 		tval = cval;
1364 		break;
1365 
1366 	case sizeof(u_int):
1367 		error = sockopt_get(sopt, &tval, sizeof(u_int));
1368 		break;
1369 
1370 	default:
1371 		error = EINVAL;
1372 	}
1373 
1374 	if (error)
1375 		return error;
1376 
1377 	if (tval > maxval)
1378 		return EINVAL;
1379 
1380 	*val = tval;
1381 	return 0;
1382 }
1383 
1384 static int
1385 ip_get_membership(const struct sockopt *sopt, struct ifnet **ifp,
1386     struct in_addr *ia, bool add)
1387 {
1388 	int error;
1389 	struct ip_mreq mreq;
1390 
1391 	error = sockopt_get(sopt, &mreq, sizeof(mreq));
1392 	if (error)
1393 		return error;
1394 
1395 	if (!IN_MULTICAST(mreq.imr_multiaddr.s_addr))
1396 		return EINVAL;
1397 
1398 	memcpy(ia, &mreq.imr_multiaddr, sizeof(*ia));
1399 
1400 	if (in_nullhost(mreq.imr_interface)) {
1401 		union {
1402 			struct sockaddr		dst;
1403 			struct sockaddr_in	dst4;
1404 		} u;
1405 		struct route ro;
1406 
1407 		if (!add) {
1408 			*ifp = NULL;
1409 			return 0;
1410 		}
1411 		/*
1412 		 * If no interface address was provided, use the interface of
1413 		 * the route to the given multicast address.
1414 		 */
1415 		struct rtentry *rt;
1416 		memset(&ro, 0, sizeof(ro));
1417 
1418 		sockaddr_in_init(&u.dst4, ia, 0);
1419 		rtcache_setdst(&ro, &u.dst);
1420 		*ifp = (rt = rtcache_init(&ro)) != NULL ? rt->rt_ifp : NULL;
1421 		rtcache_free(&ro);
1422 	} else {
1423 		*ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1424 		if (!add && *ifp == NULL)
1425 			return EADDRNOTAVAIL;
1426 	}
1427 	return 0;
1428 }
1429 
1430 /*
1431  * Add a multicast group membership.
1432  * Group must be a valid IP multicast address.
1433  */
1434 static int
1435 ip_add_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1436 {
1437 	struct ifnet *ifp;
1438 	struct in_addr ia;
1439 	int i, error;
1440 
1441 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1442 		error = ip_get_membership(sopt, &ifp, &ia, true);
1443 	else
1444 #ifdef INET6
1445 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1446 #else
1447 		return EINVAL;
1448 #endif
1449 
1450 	if (error)
1451 		return error;
1452 
1453 	/*
1454 	 * See if we found an interface, and confirm that it
1455 	 * supports multicast.
1456 	 */
1457 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1458 		return EADDRNOTAVAIL;
1459 
1460 	/*
1461 	 * See if the membership already exists or if all the
1462 	 * membership slots are full.
1463 	 */
1464 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1465 		if (imo->imo_membership[i]->inm_ifp == ifp &&
1466 		    in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1467 			break;
1468 	}
1469 	if (i < imo->imo_num_memberships)
1470 		return EADDRINUSE;
1471 
1472 	if (i == IP_MAX_MEMBERSHIPS)
1473 		return ETOOMANYREFS;
1474 
1475 	/*
1476 	 * Everything looks good; add a new record to the multicast
1477 	 * address list for the given interface.
1478 	 */
1479 	if ((imo->imo_membership[i] = in_addmulti(&ia, ifp)) == NULL)
1480 		return ENOBUFS;
1481 
1482 	++imo->imo_num_memberships;
1483 	return 0;
1484 }
1485 
1486 /*
1487  * Drop a multicast group membership.
1488  * Group must be a valid IP multicast address.
1489  */
1490 static int
1491 ip_drop_membership(struct ip_moptions *imo, const struct sockopt *sopt)
1492 {
1493 	struct in_addr ia;
1494 	struct ifnet *ifp;
1495 	int i, error;
1496 
1497 	if (sopt->sopt_size == sizeof(struct ip_mreq))
1498 		error = ip_get_membership(sopt, &ifp, &ia, false);
1499 	else
1500 #ifdef INET6
1501 		error = ip6_get_membership(sopt, &ifp, &ia, sizeof(ia));
1502 #else
1503 		return EINVAL;
1504 #endif
1505 
1506 	if (error)
1507 		return error;
1508 
1509 	/*
1510 	 * Find the membership in the membership array.
1511 	 */
1512 	for (i = 0; i < imo->imo_num_memberships; ++i) {
1513 		if ((ifp == NULL ||
1514 		     imo->imo_membership[i]->inm_ifp == ifp) &&
1515 		     in_hosteq(imo->imo_membership[i]->inm_addr, ia))
1516 			break;
1517 	}
1518 	if (i == imo->imo_num_memberships)
1519 		return EADDRNOTAVAIL;
1520 
1521 	/*
1522 	 * Give up the multicast address record to which the
1523 	 * membership points.
1524 	 */
1525 	in_delmulti(imo->imo_membership[i]);
1526 
1527 	/*
1528 	 * Remove the gap in the membership array.
1529 	 */
1530 	for (++i; i < imo->imo_num_memberships; ++i)
1531 		imo->imo_membership[i-1] = imo->imo_membership[i];
1532 	--imo->imo_num_memberships;
1533 	return 0;
1534 }
1535 
1536 /*
1537  * Set the IP multicast options in response to user setsockopt().
1538  */
1539 int
1540 ip_setmoptions(struct ip_moptions **pimo, const struct sockopt *sopt)
1541 {
1542 	struct ip_moptions *imo = *pimo;
1543 	struct in_addr addr;
1544 	struct ifnet *ifp;
1545 	int ifindex, error = 0;
1546 
1547 	if (!imo) {
1548 		/*
1549 		 * No multicast option buffer attached to the pcb;
1550 		 * allocate one and initialize to default values.
1551 		 */
1552 		imo = kmem_intr_alloc(sizeof(*imo), KM_NOSLEEP);
1553 		if (imo == NULL)
1554 			return ENOBUFS;
1555 
1556 		imo->imo_multicast_ifp = NULL;
1557 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1558 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1559 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1560 		imo->imo_num_memberships = 0;
1561 		*pimo = imo;
1562 	}
1563 
1564 	switch (sopt->sopt_name) {
1565 	case IP_MULTICAST_IF:
1566 		/*
1567 		 * Select the interface for outgoing multicast packets.
1568 		 */
1569 		error = sockopt_get(sopt, &addr, sizeof(addr));
1570 		if (error)
1571 			break;
1572 
1573 		/*
1574 		 * INADDR_ANY is used to remove a previous selection.
1575 		 * When no interface is selected, a default one is
1576 		 * chosen every time a multicast packet is sent.
1577 		 */
1578 		if (in_nullhost(addr)) {
1579 			imo->imo_multicast_ifp = NULL;
1580 			break;
1581 		}
1582 		/*
1583 		 * The selected interface is identified by its local
1584 		 * IP address.  Find the interface and confirm that
1585 		 * it supports multicasting.
1586 		 */
1587 		ifp = ip_multicast_if(&addr, &ifindex);
1588 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1589 			error = EADDRNOTAVAIL;
1590 			break;
1591 		}
1592 		imo->imo_multicast_ifp = ifp;
1593 		if (ifindex)
1594 			imo->imo_multicast_addr = addr;
1595 		else
1596 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1597 		break;
1598 
1599 	case IP_MULTICAST_TTL:
1600 		/*
1601 		 * Set the IP time-to-live for outgoing multicast packets.
1602 		 */
1603 		error = ip_getoptval(sopt, &imo->imo_multicast_ttl, MAXTTL);
1604 		break;
1605 
1606 	case IP_MULTICAST_LOOP:
1607 		/*
1608 		 * Set the loopback flag for outgoing multicast packets.
1609 		 * Must be zero or one.
1610 		 */
1611 		error = ip_getoptval(sopt, &imo->imo_multicast_loop, 1);
1612 		break;
1613 
1614 	case IP_ADD_MEMBERSHIP: /* IPV6_JOIN_GROUP */
1615 		error = ip_add_membership(imo, sopt);
1616 		break;
1617 
1618 	case IP_DROP_MEMBERSHIP: /* IPV6_LEAVE_GROUP */
1619 		error = ip_drop_membership(imo, sopt);
1620 		break;
1621 
1622 	default:
1623 		error = EOPNOTSUPP;
1624 		break;
1625 	}
1626 
1627 	/*
1628 	 * If all options have default values, no need to keep the mbuf.
1629 	 */
1630 	if (imo->imo_multicast_ifp == NULL &&
1631 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1632 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1633 	    imo->imo_num_memberships == 0) {
1634 		kmem_free(imo, sizeof(*imo));
1635 		*pimo = NULL;
1636 	}
1637 
1638 	return error;
1639 }
1640 
1641 /*
1642  * Return the IP multicast options in response to user getsockopt().
1643  */
1644 int
1645 ip_getmoptions(struct ip_moptions *imo, struct sockopt *sopt)
1646 {
1647 	struct in_addr addr;
1648 	struct in_ifaddr *ia;
1649 	uint8_t optval;
1650 	int error = 0;
1651 
1652 	switch (sopt->sopt_name) {
1653 	case IP_MULTICAST_IF:
1654 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
1655 			addr = zeroin_addr;
1656 		else if (imo->imo_multicast_addr.s_addr) {
1657 			/* return the value user has set */
1658 			addr = imo->imo_multicast_addr;
1659 		} else {
1660 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
1661 			addr = ia ? ia->ia_addr.sin_addr : zeroin_addr;
1662 		}
1663 		error = sockopt_set(sopt, &addr, sizeof(addr));
1664 		break;
1665 
1666 	case IP_MULTICAST_TTL:
1667 		optval = imo ? imo->imo_multicast_ttl
1668 			     : IP_DEFAULT_MULTICAST_TTL;
1669 
1670 		error = sockopt_set(sopt, &optval, sizeof(optval));
1671 		break;
1672 
1673 	case IP_MULTICAST_LOOP:
1674 		optval = imo ? imo->imo_multicast_loop
1675 			     : IP_DEFAULT_MULTICAST_LOOP;
1676 
1677 		error = sockopt_set(sopt, &optval, sizeof(optval));
1678 		break;
1679 
1680 	default:
1681 		error = EOPNOTSUPP;
1682 	}
1683 
1684 	return error;
1685 }
1686 
1687 /*
1688  * Discard the IP multicast options.
1689  */
1690 void
1691 ip_freemoptions(struct ip_moptions *imo)
1692 {
1693 	int i;
1694 
1695 	if (imo != NULL) {
1696 		for (i = 0; i < imo->imo_num_memberships; ++i)
1697 			in_delmulti(imo->imo_membership[i]);
1698 		kmem_free(imo, sizeof(*imo));
1699 	}
1700 }
1701 
1702 /*
1703  * Routine called from ip_output() to loop back a copy of an IP multicast
1704  * packet to the input queue of a specified interface.  Note that this
1705  * calls the output routine of the loopback "driver", but with an interface
1706  * pointer that might NOT be lo0ifp -- easier than replicating that code here.
1707  */
1708 static void
1709 ip_mloopback(struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *dst)
1710 {
1711 	struct ip *ip;
1712 	struct mbuf *copym;
1713 
1714 	copym = m_copypacket(m, M_DONTWAIT);
1715 	if (copym != NULL
1716 	 && (copym->m_flags & M_EXT || copym->m_len < sizeof(struct ip)))
1717 		copym = m_pullup(copym, sizeof(struct ip));
1718 	if (copym == NULL)
1719 		return;
1720 	/*
1721 	 * We don't bother to fragment if the IP length is greater
1722 	 * than the interface's MTU.  Can this possibly matter?
1723 	 */
1724 	ip = mtod(copym, struct ip *);
1725 
1726 	if (copym->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1727 		in_delayed_cksum(copym);
1728 		copym->m_pkthdr.csum_flags &=
1729 		    ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1730 	}
1731 
1732 	ip->ip_sum = 0;
1733 	ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
1734 #ifndef NET_MPSAFE
1735 	KERNEL_LOCK(1, NULL);
1736 #endif
1737 	(void)looutput(ifp, copym, sintocsa(dst), NULL);
1738 #ifndef NET_MPSAFE
1739 	KERNEL_UNLOCK_ONE(NULL);
1740 #endif
1741 }
1742