xref: /dflybsd-src/sys/netinet/ip_output.c (revision 6b7bde44184a7a4502ef3ec6f061b4268b50ea82)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.56 2008/09/07 08:15:25 sephe Exp $
32  */
33 
34 #define _IP_VHL
35 
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipfilter.h"
40 #include "opt_ipsec.h"
41 #include "opt_mbuf_stress_test.h"
42 #include "opt_mpls.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/thread2.h>
55 #include <sys/in_cksum.h>
56 
57 #include <net/if.h>
58 #include <net/netisr.h>
59 #include <net/pfil.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip_var.h>
68 #ifdef IPDIVERT
69 #include <netinet/ip_divert.h>
70 #endif
71 
72 #include <netproto/mpls/mpls_var.h>
73 
74 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
75 
76 #ifdef IPSEC
77 #include <netinet6/ipsec.h>
78 #include <netproto/key/key.h>
79 #ifdef IPSEC_DEBUG
80 #include <netproto/key/key_debug.h>
81 #else
82 #define	KEYDEBUG(lev,arg)
83 #endif
84 #endif /*IPSEC*/
85 
86 #ifdef FAST_IPSEC
87 #include <netproto/ipsec/ipsec.h>
88 #include <netproto/ipsec/xform.h>
89 #include <netproto/ipsec/key.h>
90 #endif /*FAST_IPSEC*/
91 
92 #include <net/ipfw/ip_fw.h>
93 #include <net/dummynet/ip_dummynet.h>
94 
95 #define print_ip(x, a, y)	 kprintf("%s %d.%d.%d.%d%s",\
96 				x, (ntohl(a.s_addr)>>24)&0xFF,\
97 				  (ntohl(a.s_addr)>>16)&0xFF,\
98 				  (ntohl(a.s_addr)>>8)&0xFF,\
99 				  (ntohl(a.s_addr))&0xFF, y);
100 
101 u_short ip_id;
102 
103 #ifdef MBUF_STRESS_TEST
104 int mbuf_frag_size = 0;
105 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
106 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
107 #endif
108 
109 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
110 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
111 static void	ip_mloopback
112 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
113 static int	ip_getmoptions
114 	(struct sockopt *, struct ip_moptions *);
115 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
116 static int	ip_setmoptions
117 	(struct sockopt *, struct ip_moptions **);
118 
119 int	ip_optcopy(struct ip *, struct ip *);
120 
121 
122 extern	struct protosw inetsw[];
123 
124 /*
125  * IP output.  The packet in mbuf chain m contains a skeletal IP
126  * header (with len, off, ttl, proto, tos, src, dst).
127  * The mbuf chain containing the packet will be freed.
128  * The mbuf opt, if present, will not be freed.
129  */
130 int
131 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
132 	  int flags, struct ip_moptions *imo, struct inpcb *inp)
133 {
134 	struct ip *ip;
135 	struct ifnet *ifp = NULL;	/* keep compiler happy */
136 	struct mbuf *m;
137 	int hlen = sizeof(struct ip);
138 	int len, off, error = 0;
139 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
140 	struct in_ifaddr *ia = NULL;
141 	int isbroadcast, sw_csum;
142 	struct in_addr pkt_dst;
143 	struct route iproute;
144 	struct m_tag *mtag;
145 #ifdef IPSEC
146 	struct secpolicy *sp = NULL;
147 	struct socket *so = inp ? inp->inp_socket : NULL;
148 #endif
149 #ifdef FAST_IPSEC
150 	struct secpolicy *sp = NULL;
151 	struct tdb_ident *tdbi;
152 #endif /* FAST_IPSEC */
153 	struct sockaddr_in *next_hop = NULL;
154 	int src_was_INADDR_ANY = 0;	/* as the name says... */
155 
156 	m = m0;
157 	M_ASSERTPKTHDR(m);
158 
159 	if (ro == NULL) {
160 		ro = &iproute;
161 		bzero(ro, sizeof *ro);
162 	}
163 
164 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
165 		/* Next hop */
166 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
167 		KKASSERT(mtag != NULL);
168 		next_hop = m_tag_data(mtag);
169 	}
170 
171 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
172 		struct dn_pkt *dn_pkt;
173 
174 		/* Extract info from dummynet tag */
175 		mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
176 		KKASSERT(mtag != NULL);
177 		dn_pkt = m_tag_data(mtag);
178 
179 		/*
180 		 * The packet was already tagged, so part of the
181 		 * processing was already done, and we need to go down.
182 		 * Get the calculated parameters from the tag.
183 		 */
184 		ifp = dn_pkt->ifp;
185 
186 		KKASSERT(ro == &iproute);
187 		*ro = dn_pkt->ro; /* structure copy */
188 
189 		dst = dn_pkt->dn_dst;
190 		if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
191 			/* If 'dst' points into dummynet tag, adjust it */
192 			dst = (struct sockaddr_in *)&(ro->ro_dst);
193 		}
194 
195 		ip = mtod(m, struct ip *);
196 		hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
197 		if (ro->ro_rt)
198 			ia = ifatoia(ro->ro_rt->rt_ifa);
199 		goto sendit;
200 	}
201 
202 	if (opt) {
203 		len = 0;
204 		m = ip_insertoptions(m, opt, &len);
205 		if (len != 0)
206 			hlen = len;
207 	}
208 	ip = mtod(m, struct ip *);
209 	pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
210 
211 	/*
212 	 * Fill in IP header.
213 	 */
214 	if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
215 		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
216 		ip->ip_off &= IP_DF;
217 		ip->ip_id = ip_newid();
218 		ipstat.ips_localout++;
219 	} else {
220 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
221 	}
222 
223 	dst = (struct sockaddr_in *)&ro->ro_dst;
224 	/*
225 	 * If there is a cached route,
226 	 * check that it is to the same destination
227 	 * and is still up.  If not, free it and try again.
228 	 * The address family should also be checked in case of sharing the
229 	 * cache with IPv6.
230 	 */
231 	if (ro->ro_rt &&
232 	    (!(ro->ro_rt->rt_flags & RTF_UP) ||
233 	     dst->sin_family != AF_INET ||
234 	     dst->sin_addr.s_addr != pkt_dst.s_addr)) {
235 		rtfree(ro->ro_rt);
236 		ro->ro_rt = (struct rtentry *)NULL;
237 	}
238 	if (ro->ro_rt == NULL) {
239 		bzero(dst, sizeof *dst);
240 		dst->sin_family = AF_INET;
241 		dst->sin_len = sizeof *dst;
242 		dst->sin_addr = pkt_dst;
243 	}
244 	/*
245 	 * If routing to interface only,
246 	 * short circuit routing lookup.
247 	 */
248 	if (flags & IP_ROUTETOIF) {
249 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
250 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
251 			ipstat.ips_noroute++;
252 			error = ENETUNREACH;
253 			goto bad;
254 		}
255 		ifp = ia->ia_ifp;
256 		ip->ip_ttl = 1;
257 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
258 	} else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
259 		   imo != NULL && imo->imo_multicast_ifp != NULL) {
260 		/*
261 		 * Bypass the normal routing lookup for multicast
262 		 * packets if the interface is specified.
263 		 */
264 		ifp = imo->imo_multicast_ifp;
265 		ia = IFP_TO_IA(ifp);
266 		isbroadcast = 0;	/* fool gcc */
267 	} else {
268 		/*
269 		 * If this is the case, we probably don't want to allocate
270 		 * a protocol-cloned route since we didn't get one from the
271 		 * ULP.  This lets TCP do its thing, while not burdening
272 		 * forwarding or ICMP with the overhead of cloning a route.
273 		 * Of course, we still want to do any cloning requested by
274 		 * the link layer, as this is probably required in all cases
275 		 * for correct operation (as it is for ARP).
276 		 */
277 		if (ro->ro_rt == NULL)
278 			rtalloc_ign(ro, RTF_PRCLONING);
279 		if (ro->ro_rt == NULL) {
280 			ipstat.ips_noroute++;
281 			error = EHOSTUNREACH;
282 			goto bad;
283 		}
284 		ia = ifatoia(ro->ro_rt->rt_ifa);
285 		ifp = ro->ro_rt->rt_ifp;
286 		ro->ro_rt->rt_use++;
287 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
288 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
289 		if (ro->ro_rt->rt_flags & RTF_HOST)
290 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
291 		else
292 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
293 	}
294 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
295 		struct in_multi *inm;
296 
297 		m->m_flags |= M_MCAST;
298 		/*
299 		 * IP destination address is multicast.  Make sure "dst"
300 		 * still points to the address in "ro".  (It may have been
301 		 * changed to point to a gateway address, above.)
302 		 */
303 		dst = (struct sockaddr_in *)&ro->ro_dst;
304 		/*
305 		 * See if the caller provided any multicast options
306 		 */
307 		if (imo != NULL) {
308 			ip->ip_ttl = imo->imo_multicast_ttl;
309 			if (imo->imo_multicast_vif != -1)
310 				ip->ip_src.s_addr =
311 				    ip_mcast_src ?
312 				    ip_mcast_src(imo->imo_multicast_vif) :
313 				    INADDR_ANY;
314 		} else
315 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
316 		/*
317 		 * Confirm that the outgoing interface supports multicast.
318 		 */
319 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
320 			if (!(ifp->if_flags & IFF_MULTICAST)) {
321 				ipstat.ips_noroute++;
322 				error = ENETUNREACH;
323 				goto bad;
324 			}
325 		}
326 		/*
327 		 * If source address not specified yet, use address
328 		 * of outgoing interface.
329 		 */
330 		if (ip->ip_src.s_addr == INADDR_ANY) {
331 			/* Interface may have no addresses. */
332 			if (ia != NULL)
333 				ip->ip_src = IA_SIN(ia)->sin_addr;
334 		}
335 
336 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
337 		if (inm != NULL &&
338 		   (imo == NULL || imo->imo_multicast_loop)) {
339 			/*
340 			 * If we belong to the destination multicast group
341 			 * on the outgoing interface, and the caller did not
342 			 * forbid loopback, loop back a copy.
343 			 */
344 			ip_mloopback(ifp, m, dst, hlen);
345 		}
346 		else {
347 			/*
348 			 * If we are acting as a multicast router, perform
349 			 * multicast forwarding as if the packet had just
350 			 * arrived on the interface to which we are about
351 			 * to send.  The multicast forwarding function
352 			 * recursively calls this function, using the
353 			 * IP_FORWARDING flag to prevent infinite recursion.
354 			 *
355 			 * Multicasts that are looped back by ip_mloopback(),
356 			 * above, will be forwarded by the ip_input() routine,
357 			 * if necessary.
358 			 */
359 			if (ip_mrouter && !(flags & IP_FORWARDING)) {
360 				/*
361 				 * If rsvp daemon is not running, do not
362 				 * set ip_moptions. This ensures that the packet
363 				 * is multicast and not just sent down one link
364 				 * as prescribed by rsvpd.
365 				 */
366 				if (!rsvp_on)
367 					imo = NULL;
368 				if (ip_mforward &&
369 				    ip_mforward(ip, ifp, m, imo) != 0) {
370 					m_freem(m);
371 					goto done;
372 				}
373 			}
374 		}
375 
376 		/*
377 		 * Multicasts with a time-to-live of zero may be looped-
378 		 * back, above, but must not be transmitted on a network.
379 		 * Also, multicasts addressed to the loopback interface
380 		 * are not sent -- the above call to ip_mloopback() will
381 		 * loop back a copy if this host actually belongs to the
382 		 * destination group on the loopback interface.
383 		 */
384 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
385 			m_freem(m);
386 			goto done;
387 		}
388 
389 		goto sendit;
390 	} else {
391 		m->m_flags &= ~M_MCAST;
392 	}
393 #ifndef notdef
394 	/*
395 	 * If the source address is not specified yet, use the address
396 	 * of the outoing interface. In case, keep note we did that, so
397 	 * if the the firewall changes the next-hop causing the output
398 	 * interface to change, we can fix that.
399 	 */
400 	if (ip->ip_src.s_addr == INADDR_ANY) {
401 		/* Interface may have no addresses. */
402 		if (ia != NULL) {
403 			ip->ip_src = IA_SIN(ia)->sin_addr;
404 			src_was_INADDR_ANY = 1;
405 		}
406 	}
407 #endif /* notdef */
408 #ifdef ALTQ
409 	/*
410 	 * Disable packet drop hack.
411 	 * Packetdrop should be done by queueing.
412 	 */
413 #else /* !ALTQ */
414 	/*
415 	 * Verify that we have any chance at all of being able to queue
416 	 *      the packet or packet fragments
417 	 */
418 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
419 		ifp->if_snd.ifq_maxlen) {
420 			error = ENOBUFS;
421 			ipstat.ips_odropped++;
422 			goto bad;
423 	}
424 #endif /* !ALTQ */
425 
426 	/*
427 	 * Look for broadcast address and
428 	 * verify user is allowed to send
429 	 * such a packet.
430 	 */
431 	if (isbroadcast) {
432 		if (!(ifp->if_flags & IFF_BROADCAST)) {
433 			error = EADDRNOTAVAIL;
434 			goto bad;
435 		}
436 		if (!(flags & IP_ALLOWBROADCAST)) {
437 			error = EACCES;
438 			goto bad;
439 		}
440 		/* don't allow broadcast messages to be fragmented */
441 		if (ip->ip_len > ifp->if_mtu) {
442 			error = EMSGSIZE;
443 			goto bad;
444 		}
445 		m->m_flags |= M_BCAST;
446 	} else {
447 		m->m_flags &= ~M_BCAST;
448 	}
449 
450 sendit:
451 #ifdef IPSEC
452 	/* get SP for this packet */
453 	if (so == NULL)
454 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
455 	else
456 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
457 
458 	if (sp == NULL) {
459 		ipsecstat.out_inval++;
460 		goto bad;
461 	}
462 
463 	error = 0;
464 
465 	/* check policy */
466 	switch (sp->policy) {
467 	case IPSEC_POLICY_DISCARD:
468 		/*
469 		 * This packet is just discarded.
470 		 */
471 		ipsecstat.out_polvio++;
472 		goto bad;
473 
474 	case IPSEC_POLICY_BYPASS:
475 	case IPSEC_POLICY_NONE:
476 		/* no need to do IPsec. */
477 		goto skip_ipsec;
478 
479 	case IPSEC_POLICY_IPSEC:
480 		if (sp->req == NULL) {
481 			/* acquire a policy */
482 			error = key_spdacquire(sp);
483 			goto bad;
484 		}
485 		break;
486 
487 	case IPSEC_POLICY_ENTRUST:
488 	default:
489 		kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
490 	}
491     {
492 	struct ipsec_output_state state;
493 	bzero(&state, sizeof state);
494 	state.m = m;
495 	if (flags & IP_ROUTETOIF) {
496 		state.ro = &iproute;
497 		bzero(&iproute, sizeof iproute);
498 	} else
499 		state.ro = ro;
500 	state.dst = (struct sockaddr *)dst;
501 
502 	ip->ip_sum = 0;
503 
504 	/*
505 	 * XXX
506 	 * delayed checksums are not currently compatible with IPsec
507 	 */
508 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
509 		in_delayed_cksum(m);
510 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
511 	}
512 
513 	ip->ip_len = htons(ip->ip_len);
514 	ip->ip_off = htons(ip->ip_off);
515 
516 	error = ipsec4_output(&state, sp, flags);
517 
518 	m = state.m;
519 	if (flags & IP_ROUTETOIF) {
520 		/*
521 		 * if we have tunnel mode SA, we may need to ignore
522 		 * IP_ROUTETOIF.
523 		 */
524 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
525 			flags &= ~IP_ROUTETOIF;
526 			ro = state.ro;
527 		}
528 	} else
529 		ro = state.ro;
530 	dst = (struct sockaddr_in *)state.dst;
531 	if (error) {
532 		/* mbuf is already reclaimed in ipsec4_output. */
533 		m0 = NULL;
534 		switch (error) {
535 		case EHOSTUNREACH:
536 		case ENETUNREACH:
537 		case EMSGSIZE:
538 		case ENOBUFS:
539 		case ENOMEM:
540 			break;
541 		default:
542 			kprintf("ip4_output (ipsec): error code %d\n", error);
543 			/*fall through*/
544 		case ENOENT:
545 			/* don't show these error codes to the user */
546 			error = 0;
547 			break;
548 		}
549 		goto bad;
550 	}
551     }
552 
553 	/* be sure to update variables that are affected by ipsec4_output() */
554 	ip = mtod(m, struct ip *);
555 #ifdef _IP_VHL
556 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
557 #else
558 	hlen = ip->ip_hl << 2;
559 #endif
560 	if (ro->ro_rt == NULL) {
561 		if (!(flags & IP_ROUTETOIF)) {
562 			kprintf("ip_output: "
563 				"can't update route after IPsec processing\n");
564 			error = EHOSTUNREACH;	/*XXX*/
565 			goto bad;
566 		}
567 	} else {
568 		ia = ifatoia(ro->ro_rt->rt_ifa);
569 		ifp = ro->ro_rt->rt_ifp;
570 	}
571 
572 	/* make it flipped, again. */
573 	ip->ip_len = ntohs(ip->ip_len);
574 	ip->ip_off = ntohs(ip->ip_off);
575 skip_ipsec:
576 #endif /*IPSEC*/
577 #ifdef FAST_IPSEC
578 	/*
579 	 * Check the security policy (SP) for the packet and, if
580 	 * required, do IPsec-related processing.  There are two
581 	 * cases here; the first time a packet is sent through
582 	 * it will be untagged and handled by ipsec4_checkpolicy.
583 	 * If the packet is resubmitted to ip_output (e.g. after
584 	 * AH, ESP, etc. processing), there will be a tag to bypass
585 	 * the lookup and related policy checking.
586 	 */
587 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
588 	crit_enter();
589 	if (mtag != NULL) {
590 		tdbi = (struct tdb_ident *)m_tag_data(mtag);
591 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
592 		if (sp == NULL)
593 			error = -EINVAL;	/* force silent drop */
594 		m_tag_delete(m, mtag);
595 	} else {
596 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
597 					&error, inp);
598 	}
599 	/*
600 	 * There are four return cases:
601 	 *    sp != NULL		    apply IPsec policy
602 	 *    sp == NULL, error == 0	    no IPsec handling needed
603 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
604 	 *    sp == NULL, error != 0	    discard packet, report error
605 	 */
606 	if (sp != NULL) {
607 		/* Loop detection, check if ipsec processing already done */
608 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
609 		for (mtag = m_tag_first(m); mtag != NULL;
610 		     mtag = m_tag_next(m, mtag)) {
611 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
612 				continue;
613 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
614 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
615 				continue;
616 			/*
617 			 * Check if policy has an SA associated with it.
618 			 * This can happen when an SP has yet to acquire
619 			 * an SA; e.g. on first reference.  If it occurs,
620 			 * then we let ipsec4_process_packet do its thing.
621 			 */
622 			if (sp->req->sav == NULL)
623 				break;
624 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
625 			if (tdbi->spi == sp->req->sav->spi &&
626 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
627 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
628 				 sizeof(union sockaddr_union)) == 0) {
629 				/*
630 				 * No IPsec processing is needed, free
631 				 * reference to SP.
632 				 *
633 				 * NB: null pointer to avoid free at
634 				 *     done: below.
635 				 */
636 				KEY_FREESP(&sp), sp = NULL;
637 				crit_exit();
638 				goto spd_done;
639 			}
640 		}
641 
642 		/*
643 		 * Do delayed checksums now because we send before
644 		 * this is done in the normal processing path.
645 		 */
646 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
647 			in_delayed_cksum(m);
648 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
649 		}
650 
651 		ip->ip_len = htons(ip->ip_len);
652 		ip->ip_off = htons(ip->ip_off);
653 
654 		/* NB: callee frees mbuf */
655 		error = ipsec4_process_packet(m, sp->req, flags, 0);
656 		/*
657 		 * Preserve KAME behaviour: ENOENT can be returned
658 		 * when an SA acquire is in progress.  Don't propagate
659 		 * this to user-level; it confuses applications.
660 		 *
661 		 * XXX this will go away when the SADB is redone.
662 		 */
663 		if (error == ENOENT)
664 			error = 0;
665 		crit_exit();
666 		goto done;
667 	} else {
668 		crit_exit();
669 
670 		if (error != 0) {
671 			/*
672 			 * Hack: -EINVAL is used to signal that a packet
673 			 * should be silently discarded.  This is typically
674 			 * because we asked key management for an SA and
675 			 * it was delayed (e.g. kicked up to IKE).
676 			 */
677 			if (error == -EINVAL)
678 				error = 0;
679 			goto bad;
680 		} else {
681 			/* No IPsec processing for this packet. */
682 		}
683 #ifdef notyet
684 		/*
685 		 * If deferred crypto processing is needed, check that
686 		 * the interface supports it.
687 		 */
688 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
689 		if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
690 			/* notify IPsec to do its own crypto */
691 			ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
692 			error = EHOSTUNREACH;
693 			goto bad;
694 		}
695 #endif
696 	}
697 spd_done:
698 #endif /* FAST_IPSEC */
699 	/*
700 	 * IpHack's section.
701 	 * - Xlate: translate packet's addr/port (NAT).
702 	 * - Firewall: deny/allow/etc.
703 	 * - Wrap: fake packet's addr/port <unimpl.>
704 	 * - Encapsulate: put it in another IP and send out. <unimp.>
705 	 */
706 
707 	/*
708 	 * Run through list of hooks for output packets.
709 	 */
710 	if (pfil_has_hooks(&inet_pfil_hook)) {
711 		error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
712 		if (error != 0 || m == NULL)
713 			goto done;
714 		ip = mtod(m, struct ip *);
715 	}
716 
717 	/*
718 	 * Check with the firewall...
719 	 * but not if we are already being fwd'd from a firewall.
720 	 */
721 	if (fw_enable && IPFW_LOADED && !next_hop) {
722 		struct sockaddr_in *old = dst;
723 		struct ip_fw_args args;
724 
725 		if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
726 			/* Extract info from dummynet tag */
727 			mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
728 			KKASSERT(mtag != NULL);
729 			args.rule =
730 			((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
731 			KKASSERT(args.rule != NULL);
732 
733 			m_tag_delete(m, mtag);
734 			m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
735 		} else {
736 			args.rule = NULL;
737 		}
738 
739 		args.eh = NULL;
740 		args.m = m;
741 		args.oif = ifp;
742 		off = ip_fw_chk_ptr(&args);
743 		m = args.m;
744 
745 		if (m == NULL) {
746 			error = EACCES;
747 			goto done;
748 		}
749 		ip = mtod(m, struct ip *);
750 
751 		if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
752 			mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
753 			KKASSERT(mtag != NULL);
754 			next_hop = m_tag_data(mtag);
755 			dst = next_hop;
756 		}
757 
758 		/*
759 		 * On return we must do the following:
760 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
761 		 * 1<=off<= 0xffff		-> DIVERT
762 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
763 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
764 		 * dst != old			-> IPFIREWALL_FORWARD
765 		 * off==0, dst==old		-> accept
766 		 * If some of the above modules are not compiled in, then
767 		 * we should't have to check the corresponding condition
768 		 * (because the ipfw control socket should not accept
769 		 * unsupported rules), but better play safe and drop
770 		 * packets in case of doubt.
771 		 */
772 		if (off & IP_FW_PORT_DENY_FLAG) {
773 			m_freem(m);
774 			error = EACCES;
775 			goto done;
776 		}
777 		if (off == 0 && dst == old)		/* common case */
778 			goto pass;
779 		if (off & IP_FW_PORT_DYNT_FLAG) {
780 			/*
781 			 * pass the pkt to dummynet. Need to include
782 			 * pipe number, m, ifp, ro, dst because these are
783 			 * not recomputed in the next pass.
784 			 * All other parameters have been already used and
785 			 * so they are not needed anymore.
786 			 * XXX note: if the ifp or ro entry are deleted
787 			 * while a pkt is in dummynet, we are in trouble!
788 			 */
789 			args.ro = ro;
790 			args.dst = dst;
791 			args.flags = flags;
792 
793 			error = 0;
794 			ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
795 			goto done;
796 		}
797 #ifdef IPDIVERT
798 		if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
799 			struct mbuf *clone = NULL;
800 
801 			/* Clone packet if we're doing a 'tee' */
802 			if ((off & IP_FW_PORT_TEE_FLAG))
803 				clone = m_dup(m, MB_DONTWAIT);
804 
805 			/*
806 			 * XXX
807 			 * delayed checksums are not currently compatible
808 			 * with divert sockets.
809 			 */
810 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
811 				in_delayed_cksum(m);
812 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
813 			}
814 
815 			/* Restore packet header fields to original values */
816 			ip->ip_len = htons(ip->ip_len);
817 			ip->ip_off = htons(ip->ip_off);
818 
819 			/* Deliver packet to divert input routine */
820 			divert_packet(m, 0);
821 
822 			/* If 'tee', continue with original packet */
823 			if (clone != NULL) {
824 				m = clone;
825 				ip = mtod(m, struct ip *);
826 				goto pass;
827 			}
828 			goto done;
829 		}
830 #endif
831 
832 		/* IPFIREWALL_FORWARD */
833 		/*
834 		 * Check dst to make sure it is directly reachable on the
835 		 * interface we previously thought it was.
836 		 * If it isn't (which may be likely in some situations) we have
837 		 * to re-route it (ie, find a route for the next-hop and the
838 		 * associated interface) and set them here. This is nested
839 		 * forwarding which in most cases is undesirable, except where
840 		 * such control is nigh impossible. So we do it here.
841 		 * And I'm babbling.
842 		 */
843 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
844 #if 0
845 			/*
846 			 * XXX To improve readability, this block should be
847 			 * changed into a function call as below:
848 			 */
849 			error = ip_ipforward(&m, &dst, &ifp);
850 			if (error)
851 				goto bad;
852 			if (m == NULL) /* ip_input consumed the mbuf */
853 				goto done;
854 #else
855 			struct in_ifaddr *ia;
856 			struct in_ifaddr_container *iac;
857 
858 			/*
859 			 * XXX sro_fwd below is static, and a pointer
860 			 * to it gets passed to routines downstream.
861 			 * This could have surprisingly bad results in
862 			 * practice, because its content is overwritten
863 			 * by subsequent packets.
864 			 */
865 			/* There must be a better way to do this next line... */
866 			static struct route sro_fwd;
867 			struct route *ro_fwd = &sro_fwd;
868 
869 #if 0
870 			print_ip("IPFIREWALL_FORWARD: New dst ip: ",
871 			    dst->sin_addr, "\n");
872 #endif
873 
874 			/*
875 			 * We need to figure out if we have been forwarded
876 			 * to a local socket. If so, then we should somehow
877 			 * "loop back" to ip_input, and get directed to the
878 			 * PCB as if we had received this packet. This is
879 			 * because it may be dificult to identify the packets
880 			 * you want to forward until they are being output
881 			 * and have selected an interface. (e.g. locally
882 			 * initiated packets) If we used the loopback inteface,
883 			 * we would not be able to control what happens
884 			 * as the packet runs through ip_input() as
885 			 * it is done through a ISR.
886 			 */
887 			ia = NULL;
888 			LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr),
889 				     ia_hash) {
890 				/*
891 				 * If the addr to forward to is one
892 				 * of ours, we pretend to
893 				 * be the destination for this packet.
894 				 */
895 				if (IA_SIN(iac->ia)->sin_addr.s_addr ==
896 				    dst->sin_addr.s_addr) {
897 					ia = iac->ia;
898 					break;
899 				}
900 			}
901 			if (ia != NULL) {    /* tell ip_input "dont filter" */
902 				if (m->m_pkthdr.rcvif == NULL)
903 					m->m_pkthdr.rcvif = ifunit("lo0");
904 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
905 					m->m_pkthdr.csum_flags |=
906 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
907 					m->m_pkthdr.csum_data = 0xffff;
908 				}
909 				m->m_pkthdr.csum_flags |=
910 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
911 				ip->ip_len = htons(ip->ip_len);
912 				ip->ip_off = htons(ip->ip_off);
913 				ip_input(m);
914 				goto done;
915 			}
916 			/* Some of the logic for this was nicked from above.
917 			 *
918 			 * This rewrites the cached route in a local PCB.
919 			 * Is this what we want to do?
920 			 */
921 			bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
922 			ro_fwd->ro_rt = NULL;
923 
924 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
925 			if (ro_fwd->ro_rt == NULL) {
926 				ipstat.ips_noroute++;
927 				error = EHOSTUNREACH;
928 				goto bad;
929 			}
930 
931 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
932 			ifp = ro_fwd->ro_rt->rt_ifp;
933 			ro_fwd->ro_rt->rt_use++;
934 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
935 				dst = (struct sockaddr_in *)
936 				    ro_fwd->ro_rt->rt_gateway;
937 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
938 				isbroadcast =
939 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
940 			else
941 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
942 			if (ro->ro_rt != NULL)
943 				rtfree(ro->ro_rt);
944 			ro->ro_rt = ro_fwd->ro_rt;
945 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
946 
947 #endif	/* ... block to be put into a function */
948 			/*
949 			 * If we added a default src ip earlier,
950 			 * which would have been gotten from the-then
951 			 * interface, do it again, from the new one.
952 			 */
953 			if (src_was_INADDR_ANY)
954 				ip->ip_src = IA_SIN(ia)->sin_addr;
955 			goto pass ;
956 		}
957 
958 		/*
959 		 * if we get here, none of the above matches, and
960 		 * we have to drop the pkt
961 		 */
962 		m_freem(m);
963 		error = EACCES; /* not sure this is the right error msg */
964 		goto done;
965 	}
966 
967 pass:
968 	/* 127/8 must not appear on wire - RFC1122. */
969 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
970 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
971 		if (!(ifp->if_flags & IFF_LOOPBACK)) {
972 			ipstat.ips_badaddr++;
973 			error = EADDRNOTAVAIL;
974 			goto bad;
975 		}
976 	}
977 
978 	m->m_pkthdr.csum_flags |= CSUM_IP;
979 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
980 	if (sw_csum & CSUM_DELAY_DATA) {
981 		in_delayed_cksum(m);
982 		sw_csum &= ~CSUM_DELAY_DATA;
983 	}
984 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
985 
986 	/*
987 	 * If small enough for interface, or the interface will take
988 	 * care of the fragmentation for us, can just send directly.
989 	 */
990 	if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
991 	    !(ip->ip_off & IP_DF))) {
992 		ip->ip_len = htons(ip->ip_len);
993 		ip->ip_off = htons(ip->ip_off);
994 		ip->ip_sum = 0;
995 		if (sw_csum & CSUM_DELAY_IP) {
996 			if (ip->ip_vhl == IP_VHL_BORING) {
997 				ip->ip_sum = in_cksum_hdr(ip);
998 			} else {
999 				ip->ip_sum = in_cksum(m, hlen);
1000 			}
1001 		}
1002 
1003 		/* Record statistics for this interface address. */
1004 		if (!(flags & IP_FORWARDING) && ia) {
1005 			ia->ia_ifa.if_opackets++;
1006 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1007 		}
1008 
1009 #ifdef IPSEC
1010 		/* clean ipsec history once it goes out of the node */
1011 		ipsec_delaux(m);
1012 #endif
1013 
1014 #ifdef MBUF_STRESS_TEST
1015 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1016 			struct mbuf *m1, *m2;
1017 			int length, tmp;
1018 
1019 			tmp = length = m->m_pkthdr.len;
1020 
1021 			while ((length -= mbuf_frag_size) >= 1) {
1022 				m1 = m_split(m, length, MB_DONTWAIT);
1023 				if (m1 == NULL)
1024 					break;
1025 				m2 = m;
1026 				while (m2->m_next != NULL)
1027 					m2 = m2->m_next;
1028 				m2->m_next = m1;
1029 			}
1030 			m->m_pkthdr.len = tmp;
1031 		}
1032 #endif
1033 
1034 #ifdef MPLS
1035 		if (!mpls_output_process(m, ro->ro_rt))
1036 			goto done;
1037 #endif
1038 		error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1039 				       ro->ro_rt);
1040 		goto done;
1041 	}
1042 
1043 	if (ip->ip_off & IP_DF) {
1044 		error = EMSGSIZE;
1045 		/*
1046 		 * This case can happen if the user changed the MTU
1047 		 * of an interface after enabling IP on it.  Because
1048 		 * most netifs don't keep track of routes pointing to
1049 		 * them, there is no way for one to update all its
1050 		 * routes when the MTU is changed.
1051 		 */
1052 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1053 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1054 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1055 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1056 		}
1057 		ipstat.ips_cantfrag++;
1058 		goto bad;
1059 	}
1060 
1061 	/*
1062 	 * Too large for interface; fragment if possible. If successful,
1063 	 * on return, m will point to a list of packets to be sent.
1064 	 */
1065 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1066 	if (error)
1067 		goto bad;
1068 	for (; m; m = m0) {
1069 		m0 = m->m_nextpkt;
1070 		m->m_nextpkt = NULL;
1071 #ifdef IPSEC
1072 		/* clean ipsec history once it goes out of the node */
1073 		ipsec_delaux(m);
1074 #endif
1075 		if (error == 0) {
1076 			/* Record statistics for this interface address. */
1077 			if (ia != NULL) {
1078 				ia->ia_ifa.if_opackets++;
1079 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1080 			}
1081 #ifdef MPLS
1082 			if (!mpls_output_process(m, ro->ro_rt))
1083 				continue;
1084 #endif
1085 			error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1086 					       ro->ro_rt);
1087 		} else {
1088 			m_freem(m);
1089 		}
1090 	}
1091 
1092 	if (error == 0)
1093 		ipstat.ips_fragmented++;
1094 
1095 done:
1096 	if (ro == &iproute && ro->ro_rt != NULL) {
1097 		RTFREE(ro->ro_rt);
1098 		ro->ro_rt = NULL;
1099 	}
1100 #ifdef IPSEC
1101 	if (sp != NULL) {
1102 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1103 			kprintf("DP ip_output call free SP:%p\n", sp));
1104 		key_freesp(sp);
1105 	}
1106 #endif
1107 #ifdef FAST_IPSEC
1108 	if (sp != NULL)
1109 		KEY_FREESP(&sp);
1110 #endif
1111 	return (error);
1112 bad:
1113 	m_freem(m);
1114 	goto done;
1115 }
1116 
1117 /*
1118  * Create a chain of fragments which fit the given mtu. m_frag points to the
1119  * mbuf to be fragmented; on return it points to the chain with the fragments.
1120  * Return 0 if no error. If error, m_frag may contain a partially built
1121  * chain of fragments that should be freed by the caller.
1122  *
1123  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1124  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1125  */
1126 int
1127 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1128 	    u_long if_hwassist_flags, int sw_csum)
1129 {
1130 	int error = 0;
1131 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1132 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1133 	int off;
1134 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1135 	int firstlen;
1136 	struct mbuf **mnext;
1137 	int nfrags;
1138 
1139 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1140 		ipstat.ips_cantfrag++;
1141 		return EMSGSIZE;
1142 	}
1143 
1144 	/*
1145 	 * Must be able to put at least 8 bytes per fragment.
1146 	 */
1147 	if (len < 8)
1148 		return EMSGSIZE;
1149 
1150 	/*
1151 	 * If the interface will not calculate checksums on
1152 	 * fragmented packets, then do it here.
1153 	 */
1154 	if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1155 	    !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1156 		in_delayed_cksum(m0);
1157 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1158 	}
1159 
1160 	if (len > PAGE_SIZE) {
1161 		/*
1162 		 * Fragment large datagrams such that each segment
1163 		 * contains a multiple of PAGE_SIZE amount of data,
1164 		 * plus headers. This enables a receiver to perform
1165 		 * page-flipping zero-copy optimizations.
1166 		 *
1167 		 * XXX When does this help given that sender and receiver
1168 		 * could have different page sizes, and also mtu could
1169 		 * be less than the receiver's page size ?
1170 		 */
1171 		int newlen;
1172 		struct mbuf *m;
1173 
1174 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1175 			off += m->m_len;
1176 
1177 		/*
1178 		 * firstlen (off - hlen) must be aligned on an
1179 		 * 8-byte boundary
1180 		 */
1181 		if (off < hlen)
1182 			goto smart_frag_failure;
1183 		off = ((off - hlen) & ~7) + hlen;
1184 		newlen = (~PAGE_MASK) & mtu;
1185 		if ((newlen + sizeof(struct ip)) > mtu) {
1186 			/* we failed, go back the default */
1187 smart_frag_failure:
1188 			newlen = len;
1189 			off = hlen + len;
1190 		}
1191 		len = newlen;
1192 
1193 	} else {
1194 		off = hlen + len;
1195 	}
1196 
1197 	firstlen = off - hlen;
1198 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1199 
1200 	/*
1201 	 * Loop through length of segment after first fragment,
1202 	 * make new header and copy data of each part and link onto chain.
1203 	 * Here, m0 is the original packet, m is the fragment being created.
1204 	 * The fragments are linked off the m_nextpkt of the original
1205 	 * packet, which after processing serves as the first fragment.
1206 	 */
1207 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1208 		struct ip *mhip;	/* ip header on the fragment */
1209 		struct mbuf *m;
1210 		int mhlen = sizeof(struct ip);
1211 
1212 		MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1213 		if (m == NULL) {
1214 			error = ENOBUFS;
1215 			ipstat.ips_odropped++;
1216 			goto done;
1217 		}
1218 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1219 		/*
1220 		 * In the first mbuf, leave room for the link header, then
1221 		 * copy the original IP header including options. The payload
1222 		 * goes into an additional mbuf chain returned by m_copy().
1223 		 */
1224 		m->m_data += max_linkhdr;
1225 		mhip = mtod(m, struct ip *);
1226 		*mhip = *ip;
1227 		if (hlen > sizeof(struct ip)) {
1228 			mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1229 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1230 		}
1231 		m->m_len = mhlen;
1232 		/* XXX do we need to add ip->ip_off below ? */
1233 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1234 		if (off + len >= ip->ip_len) {	/* last fragment */
1235 			len = ip->ip_len - off;
1236 			m->m_flags |= M_LASTFRAG;
1237 		} else
1238 			mhip->ip_off |= IP_MF;
1239 		mhip->ip_len = htons((u_short)(len + mhlen));
1240 		m->m_next = m_copy(m0, off, len);
1241 		if (m->m_next == NULL) {		/* copy failed */
1242 			m_free(m);
1243 			error = ENOBUFS;	/* ??? */
1244 			ipstat.ips_odropped++;
1245 			goto done;
1246 		}
1247 		m->m_pkthdr.len = mhlen + len;
1248 		m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1249 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1250 		mhip->ip_off = htons(mhip->ip_off);
1251 		mhip->ip_sum = 0;
1252 		if (sw_csum & CSUM_DELAY_IP)
1253 			mhip->ip_sum = in_cksum(m, mhlen);
1254 		*mnext = m;
1255 		mnext = &m->m_nextpkt;
1256 	}
1257 	ipstat.ips_ofragments += nfrags;
1258 
1259 	/* set first marker for fragment chain */
1260 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1261 	m0->m_pkthdr.csum_data = nfrags;
1262 
1263 	/*
1264 	 * Update first fragment by trimming what's been copied out
1265 	 * and updating header.
1266 	 */
1267 	m_adj(m0, hlen + firstlen - ip->ip_len);
1268 	m0->m_pkthdr.len = hlen + firstlen;
1269 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1270 	ip->ip_off |= IP_MF;
1271 	ip->ip_off = htons(ip->ip_off);
1272 	ip->ip_sum = 0;
1273 	if (sw_csum & CSUM_DELAY_IP)
1274 		ip->ip_sum = in_cksum(m0, hlen);
1275 
1276 done:
1277 	*m_frag = m0;
1278 	return error;
1279 }
1280 
1281 void
1282 in_delayed_cksum(struct mbuf *m)
1283 {
1284 	struct ip *ip;
1285 	u_short csum, offset;
1286 
1287 	ip = mtod(m, struct ip *);
1288 	offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1289 	csum = in_cksum_skip(m, ip->ip_len, offset);
1290 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1291 		csum = 0xffff;
1292 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1293 
1294 	if (offset + sizeof(u_short) > m->m_len) {
1295 		kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1296 		    m->m_len, offset, ip->ip_p);
1297 		/*
1298 		 * XXX
1299 		 * this shouldn't happen, but if it does, the
1300 		 * correct behavior may be to insert the checksum
1301 		 * in the existing chain instead of rearranging it.
1302 		 */
1303 		m = m_pullup(m, offset + sizeof(u_short));
1304 	}
1305 	*(u_short *)(m->m_data + offset) = csum;
1306 }
1307 
1308 /*
1309  * Insert IP options into preformed packet.
1310  * Adjust IP destination as required for IP source routing,
1311  * as indicated by a non-zero in_addr at the start of the options.
1312  *
1313  * XXX This routine assumes that the packet has no options in place.
1314  */
1315 static struct mbuf *
1316 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1317 {
1318 	struct ipoption *p = mtod(opt, struct ipoption *);
1319 	struct mbuf *n;
1320 	struct ip *ip = mtod(m, struct ip *);
1321 	unsigned optlen;
1322 
1323 	optlen = opt->m_len - sizeof p->ipopt_dst;
1324 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1325 		*phlen = 0;
1326 		return (m);		/* XXX should fail */
1327 	}
1328 	if (p->ipopt_dst.s_addr)
1329 		ip->ip_dst = p->ipopt_dst;
1330 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1331 		MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1332 		if (n == NULL) {
1333 			*phlen = 0;
1334 			return (m);
1335 		}
1336 		n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1337 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1338 		m->m_len -= sizeof(struct ip);
1339 		m->m_data += sizeof(struct ip);
1340 		n->m_next = m;
1341 		m = n;
1342 		m->m_len = optlen + sizeof(struct ip);
1343 		m->m_data += max_linkhdr;
1344 		memcpy(mtod(m, void *), ip, sizeof(struct ip));
1345 	} else {
1346 		m->m_data -= optlen;
1347 		m->m_len += optlen;
1348 		m->m_pkthdr.len += optlen;
1349 		ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1350 	}
1351 	ip = mtod(m, struct ip *);
1352 	bcopy(p->ipopt_list, ip + 1, optlen);
1353 	*phlen = sizeof(struct ip) + optlen;
1354 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1355 	ip->ip_len += optlen;
1356 	return (m);
1357 }
1358 
1359 /*
1360  * Copy options from ip to jp,
1361  * omitting those not copied during fragmentation.
1362  */
1363 int
1364 ip_optcopy(struct ip *ip, struct ip *jp)
1365 {
1366 	u_char *cp, *dp;
1367 	int opt, optlen, cnt;
1368 
1369 	cp = (u_char *)(ip + 1);
1370 	dp = (u_char *)(jp + 1);
1371 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1372 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1373 		opt = cp[0];
1374 		if (opt == IPOPT_EOL)
1375 			break;
1376 		if (opt == IPOPT_NOP) {
1377 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1378 			*dp++ = IPOPT_NOP;
1379 			optlen = 1;
1380 			continue;
1381 		}
1382 
1383 		KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1384 		    ("ip_optcopy: malformed ipv4 option"));
1385 		optlen = cp[IPOPT_OLEN];
1386 		KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1387 		    ("ip_optcopy: malformed ipv4 option"));
1388 
1389 		/* bogus lengths should have been caught by ip_dooptions */
1390 		if (optlen > cnt)
1391 			optlen = cnt;
1392 		if (IPOPT_COPIED(opt)) {
1393 			bcopy(cp, dp, optlen);
1394 			dp += optlen;
1395 		}
1396 	}
1397 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1398 		*dp++ = IPOPT_EOL;
1399 	return (optlen);
1400 }
1401 
1402 /*
1403  * IP socket option processing.
1404  */
1405 int
1406 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1407 {
1408 	struct	inpcb *inp = so->so_pcb;
1409 	int	error, optval;
1410 
1411 	error = optval = 0;
1412 	if (sopt->sopt_level != IPPROTO_IP) {
1413 		return (EINVAL);
1414 	}
1415 
1416 	switch (sopt->sopt_dir) {
1417 	case SOPT_SET:
1418 		switch (sopt->sopt_name) {
1419 		case IP_OPTIONS:
1420 #ifdef notyet
1421 		case IP_RETOPTS:
1422 #endif
1423 		{
1424 			struct mbuf *m;
1425 			if (sopt->sopt_valsize > MLEN) {
1426 				error = EMSGSIZE;
1427 				break;
1428 			}
1429 			MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1430 			if (m == NULL) {
1431 				error = ENOBUFS;
1432 				break;
1433 			}
1434 			m->m_len = sopt->sopt_valsize;
1435 			error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1436 					      m->m_len);
1437 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1438 					   m));
1439 		}
1440 
1441 		case IP_TOS:
1442 		case IP_TTL:
1443 		case IP_MINTTL:
1444 		case IP_RECVOPTS:
1445 		case IP_RECVRETOPTS:
1446 		case IP_RECVDSTADDR:
1447 		case IP_RECVIF:
1448 		case IP_RECVTTL:
1449 		case IP_FAITH:
1450 			error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1451 					     sizeof optval);
1452 			if (error)
1453 				break;
1454 			switch (sopt->sopt_name) {
1455 			case IP_TOS:
1456 				inp->inp_ip_tos = optval;
1457 				break;
1458 
1459 			case IP_TTL:
1460 				inp->inp_ip_ttl = optval;
1461 				break;
1462 			case IP_MINTTL:
1463 				if (optval > 0 && optval <= MAXTTL)
1464 					inp->inp_ip_minttl = optval;
1465 				else
1466 					error = EINVAL;
1467 				break;
1468 #define	OPTSET(bit) \
1469 	if (optval) \
1470 		inp->inp_flags |= bit; \
1471 	else \
1472 		inp->inp_flags &= ~bit;
1473 
1474 			case IP_RECVOPTS:
1475 				OPTSET(INP_RECVOPTS);
1476 				break;
1477 
1478 			case IP_RECVRETOPTS:
1479 				OPTSET(INP_RECVRETOPTS);
1480 				break;
1481 
1482 			case IP_RECVDSTADDR:
1483 				OPTSET(INP_RECVDSTADDR);
1484 				break;
1485 
1486 			case IP_RECVIF:
1487 				OPTSET(INP_RECVIF);
1488 				break;
1489 
1490 			case IP_RECVTTL:
1491 				OPTSET(INP_RECVTTL);
1492 				break;
1493 
1494 			case IP_FAITH:
1495 				OPTSET(INP_FAITH);
1496 				break;
1497 			}
1498 			break;
1499 #undef OPTSET
1500 
1501 		case IP_MULTICAST_IF:
1502 		case IP_MULTICAST_VIF:
1503 		case IP_MULTICAST_TTL:
1504 		case IP_MULTICAST_LOOP:
1505 		case IP_ADD_MEMBERSHIP:
1506 		case IP_DROP_MEMBERSHIP:
1507 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1508 			break;
1509 
1510 		case IP_PORTRANGE:
1511 			error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1512 					    sizeof optval);
1513 			if (error)
1514 				break;
1515 
1516 			switch (optval) {
1517 			case IP_PORTRANGE_DEFAULT:
1518 				inp->inp_flags &= ~(INP_LOWPORT);
1519 				inp->inp_flags &= ~(INP_HIGHPORT);
1520 				break;
1521 
1522 			case IP_PORTRANGE_HIGH:
1523 				inp->inp_flags &= ~(INP_LOWPORT);
1524 				inp->inp_flags |= INP_HIGHPORT;
1525 				break;
1526 
1527 			case IP_PORTRANGE_LOW:
1528 				inp->inp_flags &= ~(INP_HIGHPORT);
1529 				inp->inp_flags |= INP_LOWPORT;
1530 				break;
1531 
1532 			default:
1533 				error = EINVAL;
1534 				break;
1535 			}
1536 			break;
1537 
1538 #if defined(IPSEC) || defined(FAST_IPSEC)
1539 		case IP_IPSEC_POLICY:
1540 		{
1541 			caddr_t req;
1542 			size_t len = 0;
1543 			int priv;
1544 			struct mbuf *m;
1545 			int optname;
1546 
1547 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1548 				break;
1549 			soopt_to_mbuf(sopt, m);
1550 			priv = (sopt->sopt_td != NULL &&
1551 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1552 			req = mtod(m, caddr_t);
1553 			len = m->m_len;
1554 			optname = sopt->sopt_name;
1555 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1556 			m_freem(m);
1557 			break;
1558 		}
1559 #endif /*IPSEC*/
1560 
1561 		default:
1562 			error = ENOPROTOOPT;
1563 			break;
1564 		}
1565 		break;
1566 
1567 	case SOPT_GET:
1568 		switch (sopt->sopt_name) {
1569 		case IP_OPTIONS:
1570 		case IP_RETOPTS:
1571 			if (inp->inp_options)
1572 				soopt_from_kbuf(sopt, mtod(inp->inp_options,
1573 							   char *),
1574 						inp->inp_options->m_len);
1575 			else
1576 				sopt->sopt_valsize = 0;
1577 			break;
1578 
1579 		case IP_TOS:
1580 		case IP_TTL:
1581 		case IP_MINTTL:
1582 		case IP_RECVOPTS:
1583 		case IP_RECVRETOPTS:
1584 		case IP_RECVDSTADDR:
1585 		case IP_RECVTTL:
1586 		case IP_RECVIF:
1587 		case IP_PORTRANGE:
1588 		case IP_FAITH:
1589 			switch (sopt->sopt_name) {
1590 
1591 			case IP_TOS:
1592 				optval = inp->inp_ip_tos;
1593 				break;
1594 
1595 			case IP_TTL:
1596 				optval = inp->inp_ip_ttl;
1597 				break;
1598 			case IP_MINTTL:
1599 				optval = inp->inp_ip_minttl;
1600 				break;
1601 
1602 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1603 
1604 			case IP_RECVOPTS:
1605 				optval = OPTBIT(INP_RECVOPTS);
1606 				break;
1607 
1608 			case IP_RECVRETOPTS:
1609 				optval = OPTBIT(INP_RECVRETOPTS);
1610 				break;
1611 
1612 			case IP_RECVDSTADDR:
1613 				optval = OPTBIT(INP_RECVDSTADDR);
1614 				break;
1615 
1616 			case IP_RECVTTL:
1617 				optval = OPTBIT(INP_RECVTTL);
1618 				break;
1619 
1620 			case IP_RECVIF:
1621 				optval = OPTBIT(INP_RECVIF);
1622 				break;
1623 
1624 			case IP_PORTRANGE:
1625 				if (inp->inp_flags & INP_HIGHPORT)
1626 					optval = IP_PORTRANGE_HIGH;
1627 				else if (inp->inp_flags & INP_LOWPORT)
1628 					optval = IP_PORTRANGE_LOW;
1629 				else
1630 					optval = 0;
1631 				break;
1632 
1633 			case IP_FAITH:
1634 				optval = OPTBIT(INP_FAITH);
1635 				break;
1636 			}
1637 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1638 			break;
1639 
1640 		case IP_MULTICAST_IF:
1641 		case IP_MULTICAST_VIF:
1642 		case IP_MULTICAST_TTL:
1643 		case IP_MULTICAST_LOOP:
1644 		case IP_ADD_MEMBERSHIP:
1645 		case IP_DROP_MEMBERSHIP:
1646 			error = ip_getmoptions(sopt, inp->inp_moptions);
1647 			break;
1648 
1649 #if defined(IPSEC) || defined(FAST_IPSEC)
1650 		case IP_IPSEC_POLICY:
1651 		{
1652 			struct mbuf *m = NULL;
1653 			caddr_t req = NULL;
1654 			size_t len = 0;
1655 
1656 			if (m != NULL) {
1657 				req = mtod(m, caddr_t);
1658 				len = m->m_len;
1659 			}
1660 			error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1661 			if (error == 0)
1662 				error = soopt_from_mbuf(sopt, m); /* XXX */
1663 			if (error == 0)
1664 				m_freem(m);
1665 			break;
1666 		}
1667 #endif /*IPSEC*/
1668 
1669 		default:
1670 			error = ENOPROTOOPT;
1671 			break;
1672 		}
1673 		break;
1674 	}
1675 	return (error);
1676 }
1677 
1678 /*
1679  * Set up IP options in pcb for insertion in output packets.
1680  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1681  * with destination address if source routed.
1682  */
1683 static int
1684 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1685 {
1686 	int cnt, optlen;
1687 	u_char *cp;
1688 	u_char opt;
1689 
1690 	/* turn off any old options */
1691 	if (*pcbopt)
1692 		m_free(*pcbopt);
1693 	*pcbopt = 0;
1694 	if (m == NULL || m->m_len == 0) {
1695 		/*
1696 		 * Only turning off any previous options.
1697 		 */
1698 		if (m != NULL)
1699 			m_free(m);
1700 		return (0);
1701 	}
1702 
1703 	if (m->m_len % sizeof(int32_t))
1704 		goto bad;
1705 	/*
1706 	 * IP first-hop destination address will be stored before
1707 	 * actual options; move other options back
1708 	 * and clear it when none present.
1709 	 */
1710 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1711 		goto bad;
1712 	cnt = m->m_len;
1713 	m->m_len += sizeof(struct in_addr);
1714 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1715 	ovbcopy(mtod(m, caddr_t), cp, cnt);
1716 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1717 
1718 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1719 		opt = cp[IPOPT_OPTVAL];
1720 		if (opt == IPOPT_EOL)
1721 			break;
1722 		if (opt == IPOPT_NOP)
1723 			optlen = 1;
1724 		else {
1725 			if (cnt < IPOPT_OLEN + sizeof *cp)
1726 				goto bad;
1727 			optlen = cp[IPOPT_OLEN];
1728 			if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1729 				goto bad;
1730 		}
1731 		switch (opt) {
1732 
1733 		default:
1734 			break;
1735 
1736 		case IPOPT_LSRR:
1737 		case IPOPT_SSRR:
1738 			/*
1739 			 * user process specifies route as:
1740 			 *	->A->B->C->D
1741 			 * D must be our final destination (but we can't
1742 			 * check that since we may not have connected yet).
1743 			 * A is first hop destination, which doesn't appear in
1744 			 * actual IP option, but is stored before the options.
1745 			 */
1746 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1747 				goto bad;
1748 			m->m_len -= sizeof(struct in_addr);
1749 			cnt -= sizeof(struct in_addr);
1750 			optlen -= sizeof(struct in_addr);
1751 			cp[IPOPT_OLEN] = optlen;
1752 			/*
1753 			 * Move first hop before start of options.
1754 			 */
1755 			bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1756 			      sizeof(struct in_addr));
1757 			/*
1758 			 * Then copy rest of options back
1759 			 * to close up the deleted entry.
1760 			 */
1761 			ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1762 				&cp[IPOPT_OFFSET+1],
1763 				cnt - (IPOPT_MINOFF - 1));
1764 			break;
1765 		}
1766 	}
1767 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1768 		goto bad;
1769 	*pcbopt = m;
1770 	return (0);
1771 
1772 bad:
1773 	m_free(m);
1774 	return (EINVAL);
1775 }
1776 
1777 /*
1778  * XXX
1779  * The whole multicast option thing needs to be re-thought.
1780  * Several of these options are equally applicable to non-multicast
1781  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1782  * standard option (IP_TTL).
1783  */
1784 
1785 /*
1786  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1787  */
1788 static struct ifnet *
1789 ip_multicast_if(struct in_addr *a, int *ifindexp)
1790 {
1791 	int ifindex;
1792 	struct ifnet *ifp;
1793 
1794 	if (ifindexp)
1795 		*ifindexp = 0;
1796 	if (ntohl(a->s_addr) >> 24 == 0) {
1797 		ifindex = ntohl(a->s_addr) & 0xffffff;
1798 		if (ifindex < 0 || if_index < ifindex)
1799 			return NULL;
1800 		ifp = ifindex2ifnet[ifindex];
1801 		if (ifindexp)
1802 			*ifindexp = ifindex;
1803 	} else {
1804 		ifp = INADDR_TO_IFP(a);
1805 	}
1806 	return ifp;
1807 }
1808 
1809 /*
1810  * Set the IP multicast options in response to user setsockopt().
1811  */
1812 static int
1813 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1814 {
1815 	int error = 0;
1816 	int i;
1817 	struct in_addr addr;
1818 	struct ip_mreq mreq;
1819 	struct ifnet *ifp;
1820 	struct ip_moptions *imo = *imop;
1821 	int ifindex;
1822 
1823 	if (imo == NULL) {
1824 		/*
1825 		 * No multicast option buffer attached to the pcb;
1826 		 * allocate one and initialize to default values.
1827 		 */
1828 		imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1829 
1830 		*imop = imo;
1831 		imo->imo_multicast_ifp = NULL;
1832 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1833 		imo->imo_multicast_vif = -1;
1834 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1835 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1836 		imo->imo_num_memberships = 0;
1837 	}
1838 	switch (sopt->sopt_name) {
1839 	/* store an index number for the vif you wanna use in the send */
1840 	case IP_MULTICAST_VIF:
1841 		if (legal_vif_num == 0) {
1842 			error = EOPNOTSUPP;
1843 			break;
1844 		}
1845 		error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1846 		if (error)
1847 			break;
1848 		if (!legal_vif_num(i) && (i != -1)) {
1849 			error = EINVAL;
1850 			break;
1851 		}
1852 		imo->imo_multicast_vif = i;
1853 		break;
1854 
1855 	case IP_MULTICAST_IF:
1856 		/*
1857 		 * Select the interface for outgoing multicast packets.
1858 		 */
1859 		error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1860 		if (error)
1861 			break;
1862 
1863 		/*
1864 		 * INADDR_ANY is used to remove a previous selection.
1865 		 * When no interface is selected, a default one is
1866 		 * chosen every time a multicast packet is sent.
1867 		 */
1868 		if (addr.s_addr == INADDR_ANY) {
1869 			imo->imo_multicast_ifp = NULL;
1870 			break;
1871 		}
1872 		/*
1873 		 * The selected interface is identified by its local
1874 		 * IP address.  Find the interface and confirm that
1875 		 * it supports multicasting.
1876 		 */
1877 		crit_enter();
1878 		ifp = ip_multicast_if(&addr, &ifindex);
1879 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1880 			crit_exit();
1881 			error = EADDRNOTAVAIL;
1882 			break;
1883 		}
1884 		imo->imo_multicast_ifp = ifp;
1885 		if (ifindex)
1886 			imo->imo_multicast_addr = addr;
1887 		else
1888 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1889 		crit_exit();
1890 		break;
1891 
1892 	case IP_MULTICAST_TTL:
1893 		/*
1894 		 * Set the IP time-to-live for outgoing multicast packets.
1895 		 * The original multicast API required a char argument,
1896 		 * which is inconsistent with the rest of the socket API.
1897 		 * We allow either a char or an int.
1898 		 */
1899 		if (sopt->sopt_valsize == 1) {
1900 			u_char ttl;
1901 			error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1902 			if (error)
1903 				break;
1904 			imo->imo_multicast_ttl = ttl;
1905 		} else {
1906 			u_int ttl;
1907 			error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1908 			if (error)
1909 				break;
1910 			if (ttl > 255)
1911 				error = EINVAL;
1912 			else
1913 				imo->imo_multicast_ttl = ttl;
1914 		}
1915 		break;
1916 
1917 	case IP_MULTICAST_LOOP:
1918 		/*
1919 		 * Set the loopback flag for outgoing multicast packets.
1920 		 * Must be zero or one.  The original multicast API required a
1921 		 * char argument, which is inconsistent with the rest
1922 		 * of the socket API.  We allow either a char or an int.
1923 		 */
1924 		if (sopt->sopt_valsize == 1) {
1925 			u_char loop;
1926 
1927 			error = soopt_to_kbuf(sopt, &loop, 1, 1);
1928 			if (error)
1929 				break;
1930 			imo->imo_multicast_loop = !!loop;
1931 		} else {
1932 			u_int loop;
1933 
1934 			error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1935 					    sizeof loop);
1936 			if (error)
1937 				break;
1938 			imo->imo_multicast_loop = !!loop;
1939 		}
1940 		break;
1941 
1942 	case IP_ADD_MEMBERSHIP:
1943 		/*
1944 		 * Add a multicast group membership.
1945 		 * Group must be a valid IP multicast address.
1946 		 */
1947 		error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1948 		if (error)
1949 			break;
1950 
1951 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1952 			error = EINVAL;
1953 			break;
1954 		}
1955 		crit_enter();
1956 		/*
1957 		 * If no interface address was provided, use the interface of
1958 		 * the route to the given multicast address.
1959 		 */
1960 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1961 			struct sockaddr_in dst;
1962 			struct rtentry *rt;
1963 
1964 			bzero(&dst, sizeof(struct sockaddr_in));
1965 			dst.sin_len = sizeof(struct sockaddr_in);
1966 			dst.sin_family = AF_INET;
1967 			dst.sin_addr = mreq.imr_multiaddr;
1968 			rt = rtlookup((struct sockaddr *)&dst);
1969 			if (rt == NULL) {
1970 				error = EADDRNOTAVAIL;
1971 				crit_exit();
1972 				break;
1973 			}
1974 			--rt->rt_refcnt;
1975 			ifp = rt->rt_ifp;
1976 		} else {
1977 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1978 		}
1979 
1980 		/*
1981 		 * See if we found an interface, and confirm that it
1982 		 * supports multicast.
1983 		 */
1984 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1985 			error = EADDRNOTAVAIL;
1986 			crit_exit();
1987 			break;
1988 		}
1989 		/*
1990 		 * See if the membership already exists or if all the
1991 		 * membership slots are full.
1992 		 */
1993 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1994 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1995 			    imo->imo_membership[i]->inm_addr.s_addr
1996 						== mreq.imr_multiaddr.s_addr)
1997 				break;
1998 		}
1999 		if (i < imo->imo_num_memberships) {
2000 			error = EADDRINUSE;
2001 			crit_exit();
2002 			break;
2003 		}
2004 		if (i == IP_MAX_MEMBERSHIPS) {
2005 			error = ETOOMANYREFS;
2006 			crit_exit();
2007 			break;
2008 		}
2009 		/*
2010 		 * Everything looks good; add a new record to the multicast
2011 		 * address list for the given interface.
2012 		 */
2013 		if ((imo->imo_membership[i] =
2014 		     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2015 			error = ENOBUFS;
2016 			crit_exit();
2017 			break;
2018 		}
2019 		++imo->imo_num_memberships;
2020 		crit_exit();
2021 		break;
2022 
2023 	case IP_DROP_MEMBERSHIP:
2024 		/*
2025 		 * Drop a multicast group membership.
2026 		 * Group must be a valid IP multicast address.
2027 		 */
2028 		error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2029 		if (error)
2030 			break;
2031 
2032 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2033 			error = EINVAL;
2034 			break;
2035 		}
2036 
2037 		crit_enter();
2038 		/*
2039 		 * If an interface address was specified, get a pointer
2040 		 * to its ifnet structure.
2041 		 */
2042 		if (mreq.imr_interface.s_addr == INADDR_ANY)
2043 			ifp = NULL;
2044 		else {
2045 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2046 			if (ifp == NULL) {
2047 				error = EADDRNOTAVAIL;
2048 				crit_exit();
2049 				break;
2050 			}
2051 		}
2052 		/*
2053 		 * Find the membership in the membership array.
2054 		 */
2055 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2056 			if ((ifp == NULL ||
2057 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2058 			    imo->imo_membership[i]->inm_addr.s_addr ==
2059 			    mreq.imr_multiaddr.s_addr)
2060 				break;
2061 		}
2062 		if (i == imo->imo_num_memberships) {
2063 			error = EADDRNOTAVAIL;
2064 			crit_exit();
2065 			break;
2066 		}
2067 		/*
2068 		 * Give up the multicast address record to which the
2069 		 * membership points.
2070 		 */
2071 		in_delmulti(imo->imo_membership[i]);
2072 		/*
2073 		 * Remove the gap in the membership array.
2074 		 */
2075 		for (++i; i < imo->imo_num_memberships; ++i)
2076 			imo->imo_membership[i-1] = imo->imo_membership[i];
2077 		--imo->imo_num_memberships;
2078 		crit_exit();
2079 		break;
2080 
2081 	default:
2082 		error = EOPNOTSUPP;
2083 		break;
2084 	}
2085 
2086 	/*
2087 	 * If all options have default values, no need to keep the mbuf.
2088 	 */
2089 	if (imo->imo_multicast_ifp == NULL &&
2090 	    imo->imo_multicast_vif == -1 &&
2091 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2092 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2093 	    imo->imo_num_memberships == 0) {
2094 		kfree(*imop, M_IPMOPTS);
2095 		*imop = NULL;
2096 	}
2097 
2098 	return (error);
2099 }
2100 
2101 /*
2102  * Return the IP multicast options in response to user getsockopt().
2103  */
2104 static int
2105 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2106 {
2107 	struct in_addr addr;
2108 	struct in_ifaddr *ia;
2109 	int error, optval;
2110 	u_char coptval;
2111 
2112 	error = 0;
2113 	switch (sopt->sopt_name) {
2114 	case IP_MULTICAST_VIF:
2115 		if (imo != NULL)
2116 			optval = imo->imo_multicast_vif;
2117 		else
2118 			optval = -1;
2119 		soopt_from_kbuf(sopt, &optval, sizeof optval);
2120 		break;
2121 
2122 	case IP_MULTICAST_IF:
2123 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2124 			addr.s_addr = INADDR_ANY;
2125 		else if (imo->imo_multicast_addr.s_addr) {
2126 			/* return the value user has set */
2127 			addr = imo->imo_multicast_addr;
2128 		} else {
2129 			ia = IFP_TO_IA(imo->imo_multicast_ifp);
2130 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2131 				: IA_SIN(ia)->sin_addr.s_addr;
2132 		}
2133 		soopt_from_kbuf(sopt, &addr, sizeof addr);
2134 		break;
2135 
2136 	case IP_MULTICAST_TTL:
2137 		if (imo == NULL)
2138 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2139 		else
2140 			optval = coptval = imo->imo_multicast_ttl;
2141 		if (sopt->sopt_valsize == 1)
2142 			soopt_from_kbuf(sopt, &coptval, 1);
2143 		else
2144 			soopt_from_kbuf(sopt, &optval, sizeof optval);
2145 		break;
2146 
2147 	case IP_MULTICAST_LOOP:
2148 		if (imo == NULL)
2149 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2150 		else
2151 			optval = coptval = imo->imo_multicast_loop;
2152 		if (sopt->sopt_valsize == 1)
2153 			soopt_from_kbuf(sopt, &coptval, 1);
2154 		else
2155 			soopt_from_kbuf(sopt, &optval, sizeof optval);
2156 		break;
2157 
2158 	default:
2159 		error = ENOPROTOOPT;
2160 		break;
2161 	}
2162 	return (error);
2163 }
2164 
2165 /*
2166  * Discard the IP multicast options.
2167  */
2168 void
2169 ip_freemoptions(struct ip_moptions *imo)
2170 {
2171 	int i;
2172 
2173 	if (imo != NULL) {
2174 		for (i = 0; i < imo->imo_num_memberships; ++i)
2175 			in_delmulti(imo->imo_membership[i]);
2176 		kfree(imo, M_IPMOPTS);
2177 	}
2178 }
2179 
2180 /*
2181  * Routine called from ip_output() to loop back a copy of an IP multicast
2182  * packet to the input queue of a specified interface.  Note that this
2183  * calls the output routine of the loopback "driver", but with an interface
2184  * pointer that might NOT be a loopback interface -- evil, but easier than
2185  * replicating that code here.
2186  */
2187 static void
2188 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2189 	     int hlen)
2190 {
2191 	struct ip *ip;
2192 	struct mbuf *copym;
2193 
2194 	copym = m_copypacket(m, MB_DONTWAIT);
2195 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2196 		copym = m_pullup(copym, hlen);
2197 	if (copym != NULL) {
2198 		/*
2199 		 * if the checksum hasn't been computed, mark it as valid
2200 		 */
2201 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2202 			in_delayed_cksum(copym);
2203 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2204 			copym->m_pkthdr.csum_flags |=
2205 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2206 			copym->m_pkthdr.csum_data = 0xffff;
2207 		}
2208 		/*
2209 		 * We don't bother to fragment if the IP length is greater
2210 		 * than the interface's MTU.  Can this possibly matter?
2211 		 */
2212 		ip = mtod(copym, struct ip *);
2213 		ip->ip_len = htons(ip->ip_len);
2214 		ip->ip_off = htons(ip->ip_off);
2215 		ip->ip_sum = 0;
2216 		if (ip->ip_vhl == IP_VHL_BORING) {
2217 			ip->ip_sum = in_cksum_hdr(ip);
2218 		} else {
2219 			ip->ip_sum = in_cksum(copym, hlen);
2220 		}
2221 		/*
2222 		 * NB:
2223 		 * It's not clear whether there are any lingering
2224 		 * reentrancy problems in other areas which might
2225 		 * be exposed by using ip_input directly (in
2226 		 * particular, everything which modifies the packet
2227 		 * in-place).  Yet another option is using the
2228 		 * protosw directly to deliver the looped back
2229 		 * packet.  For the moment, we'll err on the side
2230 		 * of safety by using if_simloop().
2231 		 */
2232 #if 1 /* XXX */
2233 		if (dst->sin_family != AF_INET) {
2234 			kprintf("ip_mloopback: bad address family %d\n",
2235 						dst->sin_family);
2236 			dst->sin_family = AF_INET;
2237 		}
2238 #endif
2239 
2240 #ifdef notdef
2241 		copym->m_pkthdr.rcvif = ifp;
2242 		ip_input(copym);
2243 #else
2244 		if_simloop(ifp, copym, dst->sin_family, 0);
2245 #endif
2246 	}
2247 }
2248