xref: /freebsd/sys/netinet/ip_output.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
34  * $FreeBSD$
35  */
36 
37 #include "opt_ipfw.h"
38 #include "opt_ipdn.h"
39 #include "opt_ipdivert.h"
40 #include "opt_ipfilter.h"
41 #include "opt_ipsec.h"
42 #include "opt_mac.h"
43 #include "opt_pfil_hooks.h"
44 #include "opt_random_ip_id.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/mac.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 
56 #include <net/if.h>
57 #include <net/route.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip_var.h>
65 
66 #include <machine/in_cksum.h>
67 
68 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
69 
70 #ifdef IPSEC
71 #include <netinet6/ipsec.h>
72 #include <netkey/key.h>
73 #ifdef IPSEC_DEBUG
74 #include <netkey/key_debug.h>
75 #else
76 #define	KEYDEBUG(lev,arg)
77 #endif
78 #endif /*IPSEC*/
79 
80 #ifdef FAST_IPSEC
81 #include <netipsec/ipsec.h>
82 #include <netipsec/xform.h>
83 #include <netipsec/key.h>
84 #endif /*FAST_IPSEC*/
85 
86 #include <netinet/ip_fw.h>
87 #include <netinet/ip_dummynet.h>
88 
89 #define print_ip(x, a, y)	 printf("%s %d.%d.%d.%d%s",\
90 				x, (ntohl(a.s_addr)>>24)&0xFF,\
91 				  (ntohl(a.s_addr)>>16)&0xFF,\
92 				  (ntohl(a.s_addr)>>8)&0xFF,\
93 				  (ntohl(a.s_addr))&0xFF, y);
94 
95 u_short ip_id;
96 
97 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
98 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
99 static void	ip_mloopback
100 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
101 static int	ip_getmoptions
102 	(struct sockopt *, struct ip_moptions *);
103 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
104 static int	ip_setmoptions
105 	(struct sockopt *, struct ip_moptions **);
106 
107 int	ip_optcopy(struct ip *, struct ip *);
108 
109 
110 extern	struct protosw inetsw[];
111 
112 /*
113  * IP output.  The packet in mbuf chain m contains a skeletal IP
114  * header (with len, off, ttl, proto, tos, src, dst).
115  * The mbuf chain containing the packet will be freed.
116  * The mbuf opt, if present, will not be freed.
117  */
118 int
119 ip_output(m0, opt, ro, flags, imo, inp)
120 	struct mbuf *m0;
121 	struct mbuf *opt;
122 	struct route *ro;
123 	int flags;
124 	struct ip_moptions *imo;
125 	struct inpcb *inp;
126 {
127 	struct ip *ip, *mhip;
128 	struct ifnet *ifp = NULL;	/* keep compiler happy */
129 	struct mbuf *m;
130 	int hlen = sizeof (struct ip);
131 	int len, off, error = 0;
132 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
133 	struct in_ifaddr *ia = NULL;
134 	int isbroadcast, sw_csum;
135 	struct in_addr pkt_dst;
136 #ifdef IPSEC
137 	struct route iproute;
138 	struct secpolicy *sp = NULL;
139 #endif
140 #ifdef FAST_IPSEC
141 	struct route iproute;
142 	struct m_tag *mtag;
143 	struct secpolicy *sp = NULL;
144 	struct tdb_ident *tdbi;
145 	int s;
146 #endif /* FAST_IPSEC */
147 	struct ip_fw_args args;
148 	int src_was_INADDR_ANY = 0;	/* as the name says... */
149 #ifdef PFIL_HOOKS
150 	struct packet_filter_hook *pfh;
151 	struct mbuf *m1;
152 	int rv;
153 #endif /* PFIL_HOOKS */
154 
155 	args.eh = NULL;
156 	args.rule = NULL;
157 	args.next_hop = NULL;
158 	args.divert_rule = 0;			/* divert cookie */
159 
160 	/* Grab info from MT_TAG mbufs prepended to the chain. */
161 	for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
162 		switch(m0->_m_tag_id) {
163 		default:
164 			printf("ip_output: unrecognised MT_TAG tag %d\n",
165 			    m0->_m_tag_id);
166 			break;
167 
168 		case PACKET_TAG_DUMMYNET:
169 			/*
170 			 * the packet was already tagged, so part of the
171 			 * processing was already done, and we need to go down.
172 			 * Get parameters from the header.
173 			 */
174 			args.rule = ((struct dn_pkt *)m0)->rule;
175 			opt = NULL ;
176 			ro = & ( ((struct dn_pkt *)m0)->ro ) ;
177 			imo = NULL ;
178 			dst = ((struct dn_pkt *)m0)->dn_dst ;
179 			ifp = ((struct dn_pkt *)m0)->ifp ;
180 			flags = ((struct dn_pkt *)m0)->flags ;
181 			break;
182 
183 		case PACKET_TAG_DIVERT:
184 			args.divert_rule = (intptr_t)m0->m_data & 0xffff;
185 			break;
186 
187 		case PACKET_TAG_IPFORWARD:
188 			args.next_hop = (struct sockaddr_in *)m0->m_data;
189 			break;
190 		}
191 	}
192 	m = m0;
193 
194 	KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
195 #ifndef FAST_IPSEC
196 	KASSERT(ro != NULL, ("ip_output: no route, proto %d",
197 	    mtod(m, struct ip *)->ip_p));
198 #endif
199 
200 	if (args.rule != NULL) {	/* dummynet already saw us */
201 		ip = mtod(m, struct ip *);
202 		hlen = ip->ip_hl << 2 ;
203 		if (ro->ro_rt)
204 			ia = ifatoia(ro->ro_rt->rt_ifa);
205 		goto sendit;
206 	}
207 
208 	if (opt) {
209 		len = 0;
210 		m = ip_insertoptions(m, opt, &len);
211 		if (len != 0)
212 			hlen = len;
213 	}
214 	ip = mtod(m, struct ip *);
215 	pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
216 
217 	/*
218 	 * Fill in IP header.
219 	 */
220 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
221 		ip->ip_v = IPVERSION;
222 		ip->ip_hl = hlen >> 2;
223 		ip->ip_off &= IP_DF;
224 #ifdef RANDOM_IP_ID
225 		ip->ip_id = ip_randomid();
226 #else
227 		ip->ip_id = htons(ip_id++);
228 #endif
229 		ipstat.ips_localout++;
230 	} else {
231 		hlen = ip->ip_hl << 2;
232 	}
233 
234 #ifdef FAST_IPSEC
235 	if (ro == NULL) {
236 		ro = &iproute;
237 		bzero(ro, sizeof (*ro));
238 	}
239 #endif /* FAST_IPSEC */
240 	dst = (struct sockaddr_in *)&ro->ro_dst;
241 	/*
242 	 * If there is a cached route,
243 	 * check that it is to the same destination
244 	 * and is still up.  If not, free it and try again.
245 	 * The address family should also be checked in case of sharing the
246 	 * cache with IPv6.
247 	 */
248 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
249 			  dst->sin_family != AF_INET ||
250 			  dst->sin_addr.s_addr != pkt_dst.s_addr)) {
251 		RTFREE(ro->ro_rt);
252 		ro->ro_rt = (struct rtentry *)0;
253 	}
254 	if (ro->ro_rt == 0) {
255 		bzero(dst, sizeof(*dst));
256 		dst->sin_family = AF_INET;
257 		dst->sin_len = sizeof(*dst);
258 		dst->sin_addr = pkt_dst;
259 	}
260 	/*
261 	 * If routing to interface only,
262 	 * short circuit routing lookup.
263 	 */
264 	if (flags & IP_ROUTETOIF) {
265 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
266 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
267 			ipstat.ips_noroute++;
268 			error = ENETUNREACH;
269 			goto bad;
270 		}
271 		ifp = ia->ia_ifp;
272 		ip->ip_ttl = 1;
273 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
274 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
275 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
276 		/*
277 		 * Bypass the normal routing lookup for multicast
278 		 * packets if the interface is specified.
279 		 */
280 		ifp = imo->imo_multicast_ifp;
281 		IFP_TO_IA(ifp, ia);
282 		isbroadcast = 0;	/* fool gcc */
283 	} else {
284 		/*
285 		 * If this is the case, we probably don't want to allocate
286 		 * a protocol-cloned route since we didn't get one from the
287 		 * ULP.  This lets TCP do its thing, while not burdening
288 		 * forwarding or ICMP with the overhead of cloning a route.
289 		 * Of course, we still want to do any cloning requested by
290 		 * the link layer, as this is probably required in all cases
291 		 * for correct operation (as it is for ARP).
292 		 */
293 		if (ro->ro_rt == 0)
294 			rtalloc_ign(ro, RTF_PRCLONING);
295 		if (ro->ro_rt == 0) {
296 			ipstat.ips_noroute++;
297 			error = EHOSTUNREACH;
298 			goto bad;
299 		}
300 		ia = ifatoia(ro->ro_rt->rt_ifa);
301 		ifp = ro->ro_rt->rt_ifp;
302 		ro->ro_rt->rt_use++;
303 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
304 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
305 		if (ro->ro_rt->rt_flags & RTF_HOST)
306 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
307 		else
308 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
309 	}
310 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
311 		struct in_multi *inm;
312 
313 		m->m_flags |= M_MCAST;
314 		/*
315 		 * IP destination address is multicast.  Make sure "dst"
316 		 * still points to the address in "ro".  (It may have been
317 		 * changed to point to a gateway address, above.)
318 		 */
319 		dst = (struct sockaddr_in *)&ro->ro_dst;
320 		/*
321 		 * See if the caller provided any multicast options
322 		 */
323 		if (imo != NULL) {
324 			ip->ip_ttl = imo->imo_multicast_ttl;
325 			if (imo->imo_multicast_vif != -1)
326 				ip->ip_src.s_addr =
327 				    ip_mcast_src ?
328 				    ip_mcast_src(imo->imo_multicast_vif) :
329 				    INADDR_ANY;
330 		} else
331 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
332 		/*
333 		 * Confirm that the outgoing interface supports multicast.
334 		 */
335 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
336 			if ((ifp->if_flags & IFF_MULTICAST) == 0) {
337 				ipstat.ips_noroute++;
338 				error = ENETUNREACH;
339 				goto bad;
340 			}
341 		}
342 		/*
343 		 * If source address not specified yet, use address
344 		 * of outgoing interface.
345 		 */
346 		if (ip->ip_src.s_addr == INADDR_ANY) {
347 			/* Interface may have no addresses. */
348 			if (ia != NULL)
349 				ip->ip_src = IA_SIN(ia)->sin_addr;
350 		}
351 
352 		if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
353 			/*
354 			 * XXX
355 			 * delayed checksums are not currently
356 			 * compatible with IP multicast routing
357 			 */
358 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
359 				in_delayed_cksum(m);
360 				m->m_pkthdr.csum_flags &=
361 					~CSUM_DELAY_DATA;
362 			}
363 		}
364 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
365 		if (inm != NULL &&
366 		   (imo == NULL || imo->imo_multicast_loop)) {
367 			/*
368 			 * If we belong to the destination multicast group
369 			 * on the outgoing interface, and the caller did not
370 			 * forbid loopback, loop back a copy.
371 			 */
372 			ip_mloopback(ifp, m, dst, hlen);
373 		}
374 		else {
375 			/*
376 			 * If we are acting as a multicast router, perform
377 			 * multicast forwarding as if the packet had just
378 			 * arrived on the interface to which we are about
379 			 * to send.  The multicast forwarding function
380 			 * recursively calls this function, using the
381 			 * IP_FORWARDING flag to prevent infinite recursion.
382 			 *
383 			 * Multicasts that are looped back by ip_mloopback(),
384 			 * above, will be forwarded by the ip_input() routine,
385 			 * if necessary.
386 			 */
387 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
388 				/*
389 				 * If rsvp daemon is not running, do not
390 				 * set ip_moptions. This ensures that the packet
391 				 * is multicast and not just sent down one link
392 				 * as prescribed by rsvpd.
393 				 */
394 				if (!rsvp_on)
395 					imo = NULL;
396 				if (ip_mforward &&
397 				    ip_mforward(ip, ifp, m, imo) != 0) {
398 					m_freem(m);
399 					goto done;
400 				}
401 			}
402 		}
403 
404 		/*
405 		 * Multicasts with a time-to-live of zero may be looped-
406 		 * back, above, but must not be transmitted on a network.
407 		 * Also, multicasts addressed to the loopback interface
408 		 * are not sent -- the above call to ip_mloopback() will
409 		 * loop back a copy if this host actually belongs to the
410 		 * destination group on the loopback interface.
411 		 */
412 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
413 			m_freem(m);
414 			goto done;
415 		}
416 
417 		goto sendit;
418 	}
419 #ifndef notdef
420 	/*
421 	 * If the source address is not specified yet, use the address
422 	 * of the outoing interface. In case, keep note we did that, so
423 	 * if the the firewall changes the next-hop causing the output
424 	 * interface to change, we can fix that.
425 	 */
426 	if (ip->ip_src.s_addr == INADDR_ANY) {
427 		/* Interface may have no addresses. */
428 		if (ia != NULL) {
429 			ip->ip_src = IA_SIN(ia)->sin_addr;
430 			src_was_INADDR_ANY = 1;
431 		}
432 	}
433 #endif /* notdef */
434 	/*
435 	 * Verify that we have any chance at all of being able to queue
436 	 *      the packet or packet fragments
437 	 */
438 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
439 		ifp->if_snd.ifq_maxlen) {
440 			error = ENOBUFS;
441 			ipstat.ips_odropped++;
442 			goto bad;
443 	}
444 
445 	/*
446 	 * Look for broadcast address and
447 	 * verify user is allowed to send
448 	 * such a packet.
449 	 */
450 	if (isbroadcast) {
451 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
452 			error = EADDRNOTAVAIL;
453 			goto bad;
454 		}
455 		if ((flags & IP_ALLOWBROADCAST) == 0) {
456 			error = EACCES;
457 			goto bad;
458 		}
459 		/* don't allow broadcast messages to be fragmented */
460 		if ((u_short)ip->ip_len > ifp->if_mtu) {
461 			error = EMSGSIZE;
462 			goto bad;
463 		}
464 		m->m_flags |= M_BCAST;
465 	} else {
466 		m->m_flags &= ~M_BCAST;
467 	}
468 
469 sendit:
470 #ifdef IPSEC
471 	/* get SP for this packet */
472 	if (inp == NULL)
473 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
474 	else
475 		sp = ipsec4_getpolicybypcb(m, IPSEC_DIR_OUTBOUND, inp, &error);
476 
477 	if (sp == NULL) {
478 		ipsecstat.out_inval++;
479 		goto bad;
480 	}
481 
482 	error = 0;
483 
484 	/* check policy */
485 	switch (sp->policy) {
486 	case IPSEC_POLICY_DISCARD:
487 		/*
488 		 * This packet is just discarded.
489 		 */
490 		ipsecstat.out_polvio++;
491 		goto bad;
492 
493 	case IPSEC_POLICY_BYPASS:
494 	case IPSEC_POLICY_NONE:
495 		/* no need to do IPsec. */
496 		goto skip_ipsec;
497 
498 	case IPSEC_POLICY_IPSEC:
499 		if (sp->req == NULL) {
500 			/* acquire a policy */
501 			error = key_spdacquire(sp);
502 			goto bad;
503 		}
504 		break;
505 
506 	case IPSEC_POLICY_ENTRUST:
507 	default:
508 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
509 	}
510     {
511 	struct ipsec_output_state state;
512 	bzero(&state, sizeof(state));
513 	state.m = m;
514 	if (flags & IP_ROUTETOIF) {
515 		state.ro = &iproute;
516 		bzero(&iproute, sizeof(iproute));
517 	} else
518 		state.ro = ro;
519 	state.dst = (struct sockaddr *)dst;
520 
521 	ip->ip_sum = 0;
522 
523 	/*
524 	 * XXX
525 	 * delayed checksums are not currently compatible with IPsec
526 	 */
527 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
528 		in_delayed_cksum(m);
529 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
530 	}
531 
532 	ip->ip_len = htons(ip->ip_len);
533 	ip->ip_off = htons(ip->ip_off);
534 
535 	error = ipsec4_output(&state, sp, flags);
536 
537 	m = state.m;
538 	if (flags & IP_ROUTETOIF) {
539 		/*
540 		 * if we have tunnel mode SA, we may need to ignore
541 		 * IP_ROUTETOIF.
542 		 */
543 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
544 			flags &= ~IP_ROUTETOIF;
545 			ro = state.ro;
546 		}
547 	} else
548 		ro = state.ro;
549 	dst = (struct sockaddr_in *)state.dst;
550 	if (error) {
551 		/* mbuf is already reclaimed in ipsec4_output. */
552 		m0 = NULL;
553 		switch (error) {
554 		case EHOSTUNREACH:
555 		case ENETUNREACH:
556 		case EMSGSIZE:
557 		case ENOBUFS:
558 		case ENOMEM:
559 			break;
560 		default:
561 			printf("ip4_output (ipsec): error code %d\n", error);
562 			/*fall through*/
563 		case ENOENT:
564 			/* don't show these error codes to the user */
565 			error = 0;
566 			break;
567 		}
568 		goto bad;
569 	}
570     }
571 
572 	/* be sure to update variables that are affected by ipsec4_output() */
573 	ip = mtod(m, struct ip *);
574 	hlen = ip->ip_hl << 2;
575 	if (ro->ro_rt == NULL) {
576 		if ((flags & IP_ROUTETOIF) == 0) {
577 			printf("ip_output: "
578 				"can't update route after IPsec processing\n");
579 			error = EHOSTUNREACH;	/*XXX*/
580 			goto bad;
581 		}
582 	} else {
583 		ia = ifatoia(ro->ro_rt->rt_ifa);
584 		ifp = ro->ro_rt->rt_ifp;
585 	}
586 
587 	/* make it flipped, again. */
588 	ip->ip_len = ntohs(ip->ip_len);
589 	ip->ip_off = ntohs(ip->ip_off);
590 skip_ipsec:
591 #endif /*IPSEC*/
592 #ifdef FAST_IPSEC
593 	/*
594 	 * Check the security policy (SP) for the packet and, if
595 	 * required, do IPsec-related processing.  There are two
596 	 * cases here; the first time a packet is sent through
597 	 * it will be untagged and handled by ipsec4_checkpolicy.
598 	 * If the packet is resubmitted to ip_output (e.g. after
599 	 * AH, ESP, etc. processing), there will be a tag to bypass
600 	 * the lookup and related policy checking.
601 	 */
602 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
603 	s = splnet();
604 	if (mtag != NULL) {
605 		tdbi = (struct tdb_ident *)(mtag + 1);
606 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
607 		if (sp == NULL)
608 			error = -EINVAL;	/* force silent drop */
609 		m_tag_delete(m, mtag);
610 	} else {
611 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
612 					&error, inp);
613 	}
614 	/*
615 	 * There are four return cases:
616 	 *    sp != NULL	 	    apply IPsec policy
617 	 *    sp == NULL, error == 0	    no IPsec handling needed
618 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
619 	 *    sp == NULL, error != 0	    discard packet, report error
620 	 */
621 	if (sp != NULL) {
622 		/* Loop detection, check if ipsec processing already done */
623 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
624 		for (mtag = m_tag_first(m); mtag != NULL;
625 		     mtag = m_tag_next(m, mtag)) {
626 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
627 				continue;
628 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
629 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
630 				continue;
631 			/*
632 			 * Check if policy has an SA associated with it.
633 			 * This can happen when an SP has yet to acquire
634 			 * an SA; e.g. on first reference.  If it occurs,
635 			 * then we let ipsec4_process_packet do its thing.
636 			 */
637 			if (sp->req->sav == NULL)
638 				break;
639 			tdbi = (struct tdb_ident *)(mtag + 1);
640 			if (tdbi->spi == sp->req->sav->spi &&
641 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
642 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
643 				 sizeof (union sockaddr_union)) == 0) {
644 				/*
645 				 * No IPsec processing is needed, free
646 				 * reference to SP.
647 				 *
648 				 * NB: null pointer to avoid free at
649 				 *     done: below.
650 				 */
651 				KEY_FREESP(&sp), sp = NULL;
652 				splx(s);
653 				goto spd_done;
654 			}
655 		}
656 
657 		/*
658 		 * Do delayed checksums now because we send before
659 		 * this is done in the normal processing path.
660 		 */
661 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
662 			in_delayed_cksum(m);
663 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
664 		}
665 
666 		ip->ip_len = htons(ip->ip_len);
667 		ip->ip_off = htons(ip->ip_off);
668 
669 		/* NB: callee frees mbuf */
670 		error = ipsec4_process_packet(m, sp->req, flags, 0);
671 		/*
672 		 * Preserve KAME behaviour: ENOENT can be returned
673 		 * when an SA acquire is in progress.  Don't propagate
674 		 * this to user-level; it confuses applications.
675 		 *
676 		 * XXX this will go away when the SADB is redone.
677 		 */
678 		if (error == ENOENT)
679 			error = 0;
680 		splx(s);
681 		goto done;
682 	} else {
683 		splx(s);
684 
685 		if (error != 0) {
686 			/*
687 			 * Hack: -EINVAL is used to signal that a packet
688 			 * should be silently discarded.  This is typically
689 			 * because we asked key management for an SA and
690 			 * it was delayed (e.g. kicked up to IKE).
691 			 */
692 			if (error == -EINVAL)
693 				error = 0;
694 			goto bad;
695 		} else {
696 			/* No IPsec processing for this packet. */
697 		}
698 #ifdef notyet
699 		/*
700 		 * If deferred crypto processing is needed, check that
701 		 * the interface supports it.
702 		 */
703 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
704 		if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
705 			/* notify IPsec to do its own crypto */
706 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
707 			error = EHOSTUNREACH;
708 			goto bad;
709 		}
710 #endif
711 	}
712 spd_done:
713 #endif /* FAST_IPSEC */
714 
715 	/*
716 	 * IpHack's section.
717 	 * - Xlate: translate packet's addr/port (NAT).
718 	 * - Firewall: deny/allow/etc.
719 	 * - Wrap: fake packet's addr/port <unimpl.>
720 	 * - Encapsulate: put it in another IP and send out. <unimp.>
721 	 */
722 #ifdef PFIL_HOOKS
723 	/*
724 	 * Run through list of hooks for output packets.
725 	 */
726 	m1 = m;
727 	pfh = pfil_hook_get(PFIL_OUT, &inetsw[ip_protox[IPPROTO_IP]].pr_pfh);
728 	for (; pfh; pfh = TAILQ_NEXT(pfh, pfil_link))
729 		if (pfh->pfil_func) {
730 			rv = pfh->pfil_func(ip, hlen, ifp, 1, &m1);
731 			if (rv) {
732 				error = EHOSTUNREACH;
733 				goto done;
734 			}
735 			m = m1;
736 			if (m == NULL)
737 				goto done;
738 			ip = mtod(m, struct ip *);
739 		}
740 #endif /* PFIL_HOOKS */
741 
742 	/*
743 	 * Check with the firewall...
744 	 * but not if we are already being fwd'd from a firewall.
745 	 */
746 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
747 		struct sockaddr_in *old = dst;
748 
749 		args.m = m;
750 		args.next_hop = dst;
751 		args.oif = ifp;
752 		off = ip_fw_chk_ptr(&args);
753 		m = args.m;
754 		dst = args.next_hop;
755 
756                 /*
757 		 * On return we must do the following:
758 		 * m == NULL	-> drop the pkt (old interface, deprecated)
759 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
760 		 * 1<=off<= 0xffff		-> DIVERT
761 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
762 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
763 		 * dst != old			-> IPFIREWALL_FORWARD
764 		 * off==0, dst==old		-> accept
765 		 * If some of the above modules are not compiled in, then
766 		 * we should't have to check the corresponding condition
767 		 * (because the ipfw control socket should not accept
768 		 * unsupported rules), but better play safe and drop
769 		 * packets in case of doubt.
770 		 */
771 		if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
772 			if (m)
773 				m_freem(m);
774 			error = EACCES;
775 			goto done;
776 		}
777 		ip = mtod(m, struct ip *);
778 		if (off == 0 && dst == old)		/* common case */
779 			goto pass;
780                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
781 			/*
782 			 * pass the pkt to dummynet. Need to include
783 			 * pipe number, m, ifp, ro, dst because these are
784 			 * not recomputed in the next pass.
785 			 * All other parameters have been already used and
786 			 * so they are not needed anymore.
787 			 * XXX note: if the ifp or ro entry are deleted
788 			 * while a pkt is in dummynet, we are in trouble!
789 			 */
790 			args.ro = ro;
791 			args.dst = dst;
792 			args.flags = flags;
793 
794 			error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
795 				&args);
796 			goto done;
797 		}
798 #ifdef IPDIVERT
799 		if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
800 			struct mbuf *clone = NULL;
801 
802 			/* Clone packet if we're doing a 'tee' */
803 			if ((off & IP_FW_PORT_TEE_FLAG) != 0)
804 				clone = m_dup(m, M_DONTWAIT);
805 
806 			/*
807 			 * XXX
808 			 * delayed checksums are not currently compatible
809 			 * with divert sockets.
810 			 */
811 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
812 				in_delayed_cksum(m);
813 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
814 			}
815 
816 			/* Restore packet header fields to original values */
817 			ip->ip_len = htons(ip->ip_len);
818 			ip->ip_off = htons(ip->ip_off);
819 
820 			/* Deliver packet to divert input routine */
821 			divert_packet(m, 0, off & 0xffff, args.divert_rule);
822 
823 			/* If 'tee', continue with original packet */
824 			if (clone != NULL) {
825 				m = clone;
826 				ip = mtod(m, struct ip *);
827 				goto pass;
828 			}
829 			goto done;
830 		}
831 #endif
832 
833 		/* IPFIREWALL_FORWARD */
834 		/*
835 		 * Check dst to make sure it is directly reachable on the
836 		 * interface we previously thought it was.
837 		 * If it isn't (which may be likely in some situations) we have
838 		 * to re-route it (ie, find a route for the next-hop and the
839 		 * associated interface) and set them here. This is nested
840 		 * forwarding which in most cases is undesirable, except where
841 		 * such control is nigh impossible. So we do it here.
842 		 * And I'm babbling.
843 		 */
844 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
845 #if 0
846 			/*
847 			 * XXX To improve readability, this block should be
848 			 * changed into a function call as below:
849 			 */
850 			error = ip_ipforward(&m, &dst, &ifp);
851 			if (error)
852 				goto bad;
853 			if (m == NULL) /* ip_input consumed the mbuf */
854 				goto done;
855 #else
856 			struct in_ifaddr *ia;
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 an ISR.
886 			 */
887 			LIST_FOREACH(ia,
888 			    INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
889 				/*
890 				 * If the addr to forward to is one
891 				 * of ours, we pretend to
892 				 * be the destination for this packet.
893 				 */
894 				if (IA_SIN(ia)->sin_addr.s_addr ==
895 						 dst->sin_addr.s_addr)
896 					break;
897 			}
898 			if (ia) {	/* tell ip_input "dont filter" */
899 				struct m_hdr tag;
900 
901 				tag.mh_type = MT_TAG;
902 				tag.mh_flags = PACKET_TAG_IPFORWARD;
903 				tag.mh_data = (caddr_t)args.next_hop;
904 				tag.mh_next = m;
905 
906 				if (m->m_pkthdr.rcvif == NULL)
907 					m->m_pkthdr.rcvif = ifunit("lo0");
908 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
909 					m->m_pkthdr.csum_flags |=
910 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
911 					m0->m_pkthdr.csum_data = 0xffff;
912 				}
913 				m->m_pkthdr.csum_flags |=
914 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
915 				ip->ip_len = htons(ip->ip_len);
916 				ip->ip_off = htons(ip->ip_off);
917 				ip_input((struct mbuf *)&tag);
918 				goto done;
919 			}
920 			/* Some of the logic for this was
921 			 * nicked from above.
922 			 *
923 			 * This rewrites the cached route in a local PCB.
924 			 * Is this what we want to do?
925 			 */
926 			bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
927 
928 			ro_fwd->ro_rt = 0;
929 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
930 
931 			if (ro_fwd->ro_rt == 0) {
932 				ipstat.ips_noroute++;
933 				error = EHOSTUNREACH;
934 				goto bad;
935 			}
936 
937 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
938 			ifp = ro_fwd->ro_rt->rt_ifp;
939 			ro_fwd->ro_rt->rt_use++;
940 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
941 				dst = (struct sockaddr_in *)
942 					ro_fwd->ro_rt->rt_gateway;
943 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
944 				isbroadcast =
945 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
946 			else
947 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
948 			if (ro->ro_rt)
949 				RTFREE(ro->ro_rt);
950 			ro->ro_rt = ro_fwd->ro_rt;
951 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
952 
953 #endif	/* ... block to be put into a function */
954 			/*
955 			 * If we added a default src ip earlier,
956 			 * which would have been gotten from the-then
957 			 * interface, do it again, from the new one.
958 			 */
959 			if (src_was_INADDR_ANY)
960 				ip->ip_src = IA_SIN(ia)->sin_addr;
961 			goto pass ;
962 		}
963 
964                 /*
965                  * if we get here, none of the above matches, and
966                  * we have to drop the pkt
967                  */
968 		m_freem(m);
969                 error = EACCES; /* not sure this is the right error msg */
970                 goto done;
971 	}
972 
973 pass:
974 	/* 127/8 must not appear on wire - RFC1122. */
975 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
976 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
977 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
978 			ipstat.ips_badaddr++;
979 			error = EADDRNOTAVAIL;
980 			goto bad;
981 		}
982 	}
983 
984 	m->m_pkthdr.csum_flags |= CSUM_IP;
985 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
986 	if (sw_csum & CSUM_DELAY_DATA) {
987 		in_delayed_cksum(m);
988 		sw_csum &= ~CSUM_DELAY_DATA;
989 	}
990 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
991 
992 	/*
993 	 * If small enough for interface, or the interface will take
994 	 * care of the fragmentation for us, can just send directly.
995 	 */
996 	if ((u_short)ip->ip_len <= ifp->if_mtu ||
997 	    ifp->if_hwassist & CSUM_FRAGMENT) {
998 		ip->ip_len = htons(ip->ip_len);
999 		ip->ip_off = htons(ip->ip_off);
1000 		ip->ip_sum = 0;
1001 		if (sw_csum & CSUM_DELAY_IP)
1002 			ip->ip_sum = in_cksum(m, hlen);
1003 
1004 		/* Record statistics for this interface address. */
1005 		if (!(flags & IP_FORWARDING) && ia) {
1006 			ia->ia_ifa.if_opackets++;
1007 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1008 		}
1009 
1010 #ifdef IPSEC
1011 		/* clean ipsec history once it goes out of the node */
1012 		ipsec_delaux(m);
1013 #endif
1014 
1015 		error = (*ifp->if_output)(ifp, m,
1016 				(struct sockaddr *)dst, ro->ro_rt);
1017 		goto done;
1018 	}
1019 	/*
1020 	 * Too large for interface; fragment if possible.
1021 	 * Must be able to put at least 8 bytes per fragment.
1022 	 */
1023 	if (ip->ip_off & IP_DF) {
1024 		error = EMSGSIZE;
1025 		/*
1026 		 * This case can happen if the user changed the MTU
1027 		 * of an interface after enabling IP on it.  Because
1028 		 * most netifs don't keep track of routes pointing to
1029 		 * them, there is no way for one to update all its
1030 		 * routes when the MTU is changed.
1031 		 */
1032 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST))
1033 		    && !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU)
1034 		    && (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1035 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1036 		}
1037 		ipstat.ips_cantfrag++;
1038 		goto bad;
1039 	}
1040 	len = (ifp->if_mtu - hlen) &~ 7;
1041 	if (len < 8) {
1042 		error = EMSGSIZE;
1043 		goto bad;
1044 	}
1045 
1046 	/*
1047 	 * if the interface will not calculate checksums on
1048 	 * fragmented packets, then do it here.
1049 	 */
1050 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1051 	    (ifp->if_hwassist & CSUM_IP_FRAGS) == 0) {
1052 		in_delayed_cksum(m);
1053 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1054 	}
1055 
1056 	if (len > PAGE_SIZE) {
1057 		/*
1058 		 * Fragement large datagrams such that each segment
1059 		 * contains a multiple of PAGE_SIZE amount of data,
1060 		 * plus headers. This enables a receiver to perform
1061 		 * page-flipping zero-copy optimizations.
1062 		 */
1063 
1064 		int newlen;
1065 		struct mbuf *mtmp;
1066 
1067 		for (mtmp = m, off = 0;
1068 		     mtmp && ((off + mtmp->m_len) <= ifp->if_mtu);
1069 		     mtmp = mtmp->m_next) {
1070 			off += mtmp->m_len;
1071 		}
1072 		/*
1073 		 * firstlen (off - hlen) must be aligned on an
1074 		 * 8-byte boundary
1075 		 */
1076 		if (off < hlen)
1077 			goto smart_frag_failure;
1078 		off = ((off - hlen) & ~7) + hlen;
1079 		newlen = (~PAGE_MASK) & ifp->if_mtu;
1080 		if ((newlen + sizeof (struct ip)) > ifp->if_mtu) {
1081 			/* we failed, go back the default */
1082 smart_frag_failure:
1083 			newlen = len;
1084 			off = hlen + len;
1085 		}
1086 
1087 /*		printf("ipfrag: len = %d, hlen = %d, mhlen = %d, newlen = %d, off = %d\n",
1088 		len, hlen, sizeof (struct ip), newlen, off);*/
1089 
1090 		len = newlen;
1091 
1092 	} else {
1093 		off = hlen + len;
1094 	}
1095 
1096 
1097 
1098     {
1099 	int mhlen, firstlen = off - hlen;
1100 	struct mbuf **mnext = &m->m_nextpkt;
1101 	int nfrags = 1;
1102 
1103 	/*
1104 	 * Loop through length of segment after first fragment,
1105 	 * make new header and copy data of each part and link onto chain.
1106 	 */
1107 	m0 = m;
1108 	mhlen = sizeof (struct ip);
1109 	for (; off < (u_short)ip->ip_len; off += len) {
1110 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
1111 		if (m == 0) {
1112 			error = ENOBUFS;
1113 			ipstat.ips_odropped++;
1114 			goto sendorfree;
1115 		}
1116 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1117 		m->m_data += max_linkhdr;
1118 		mhip = mtod(m, struct ip *);
1119 		*mhip = *ip;
1120 		if (hlen > sizeof (struct ip)) {
1121 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1122 			mhip->ip_v = IPVERSION;
1123 			mhip->ip_hl = mhlen >> 2;
1124 		}
1125 		m->m_len = mhlen;
1126 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1127 		if (off + len >= (u_short)ip->ip_len)
1128 			len = (u_short)ip->ip_len - off;
1129 		else
1130 			mhip->ip_off |= IP_MF;
1131 		mhip->ip_len = htons((u_short)(len + mhlen));
1132 		m->m_next = m_copy(m0, off, len);
1133 		if (m->m_next == 0) {
1134 			(void) m_free(m);
1135 			error = ENOBUFS;	/* ??? */
1136 			ipstat.ips_odropped++;
1137 			goto sendorfree;
1138 		}
1139 		m->m_pkthdr.len = mhlen + len;
1140 		m->m_pkthdr.rcvif = (struct ifnet *)0;
1141 #ifdef MAC
1142 		mac_create_fragment(m0, m);
1143 #endif
1144 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1145 		mhip->ip_off = htons(mhip->ip_off);
1146 		mhip->ip_sum = 0;
1147 		if (sw_csum & CSUM_DELAY_IP)
1148 			mhip->ip_sum = in_cksum(m, mhlen);
1149 		*mnext = m;
1150 		mnext = &m->m_nextpkt;
1151 		nfrags++;
1152 	}
1153 	ipstat.ips_ofragments += nfrags;
1154 
1155 	/* set first/last markers for fragment chain */
1156 	m->m_flags |= M_LASTFRAG;
1157 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1158 	m0->m_pkthdr.csum_data = nfrags;
1159 
1160 	/*
1161 	 * Update first fragment by trimming what's been copied out
1162 	 * and updating header, then send each fragment (in order).
1163 	 */
1164 	m = m0;
1165 	m_adj(m, hlen + firstlen - (u_short)ip->ip_len);
1166 	m->m_pkthdr.len = hlen + firstlen;
1167 	ip->ip_len = htons((u_short)m->m_pkthdr.len);
1168 	ip->ip_off |= IP_MF;
1169 	ip->ip_off = htons(ip->ip_off);
1170 	ip->ip_sum = 0;
1171 	if (sw_csum & CSUM_DELAY_IP)
1172 		ip->ip_sum = in_cksum(m, hlen);
1173 sendorfree:
1174 	for (m = m0; m; m = m0) {
1175 		m0 = m->m_nextpkt;
1176 		m->m_nextpkt = 0;
1177 #ifdef IPSEC
1178 		/* clean ipsec history once it goes out of the node */
1179 		ipsec_delaux(m);
1180 #endif
1181 		if (error == 0) {
1182 			/* Record statistics for this interface address. */
1183 			if (ia != NULL) {
1184 				ia->ia_ifa.if_opackets++;
1185 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1186 			}
1187 
1188 			error = (*ifp->if_output)(ifp, m,
1189 			    (struct sockaddr *)dst, ro->ro_rt);
1190 		} else
1191 			m_freem(m);
1192 	}
1193 
1194 	if (error == 0)
1195 		ipstat.ips_fragmented++;
1196     }
1197 done:
1198 #ifdef IPSEC
1199 	if (ro == &iproute && ro->ro_rt) {
1200 		RTFREE(ro->ro_rt);
1201 		ro->ro_rt = NULL;
1202 	}
1203 	if (sp != NULL) {
1204 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1205 			printf("DP ip_output call free SP:%p\n", sp));
1206 		key_freesp(sp);
1207 	}
1208 #endif /* IPSEC */
1209 #ifdef FAST_IPSEC
1210 	if (ro == &iproute && ro->ro_rt) {
1211 		RTFREE(ro->ro_rt);
1212 		ro->ro_rt = NULL;
1213 	}
1214 	if (sp != NULL)
1215 		KEY_FREESP(&sp);
1216 #endif /* FAST_IPSEC */
1217 	return (error);
1218 bad:
1219 	m_freem(m);
1220 	goto done;
1221 }
1222 
1223 void
1224 in_delayed_cksum(struct mbuf *m)
1225 {
1226 	struct ip *ip;
1227 	u_short csum, offset;
1228 
1229 	ip = mtod(m, struct ip *);
1230 	offset = ip->ip_hl << 2 ;
1231 	csum = in_cksum_skip(m, ip->ip_len, offset);
1232 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1233 		csum = 0xffff;
1234 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1235 
1236 	if (offset + sizeof(u_short) > m->m_len) {
1237 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1238 		    m->m_len, offset, ip->ip_p);
1239 		/*
1240 		 * XXX
1241 		 * this shouldn't happen, but if it does, the
1242 		 * correct behavior may be to insert the checksum
1243 		 * in the existing chain instead of rearranging it.
1244 		 */
1245 		m = m_pullup(m, offset + sizeof(u_short));
1246 	}
1247 	*(u_short *)(m->m_data + offset) = csum;
1248 }
1249 
1250 /*
1251  * Insert IP options into preformed packet.
1252  * Adjust IP destination as required for IP source routing,
1253  * as indicated by a non-zero in_addr at the start of the options.
1254  *
1255  * XXX This routine assumes that the packet has no options in place.
1256  */
1257 static struct mbuf *
1258 ip_insertoptions(m, opt, phlen)
1259 	register struct mbuf *m;
1260 	struct mbuf *opt;
1261 	int *phlen;
1262 {
1263 	register struct ipoption *p = mtod(opt, struct ipoption *);
1264 	struct mbuf *n;
1265 	register struct ip *ip = mtod(m, struct ip *);
1266 	unsigned optlen;
1267 
1268 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1269 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1270 		*phlen = 0;
1271 		return (m);		/* XXX should fail */
1272 	}
1273 	if (p->ipopt_dst.s_addr)
1274 		ip->ip_dst = p->ipopt_dst;
1275 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1276 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1277 		if (n == 0) {
1278 			*phlen = 0;
1279 			return (m);
1280 		}
1281 		n->m_pkthdr.rcvif = (struct ifnet *)0;
1282 #ifdef MAC
1283 		mac_create_mbuf_from_mbuf(m, n);
1284 #endif
1285 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1286 		m->m_len -= sizeof(struct ip);
1287 		m->m_data += sizeof(struct ip);
1288 		n->m_next = m;
1289 		m = n;
1290 		m->m_len = optlen + sizeof(struct ip);
1291 		m->m_data += max_linkhdr;
1292 		(void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1293 	} else {
1294 		m->m_data -= optlen;
1295 		m->m_len += optlen;
1296 		m->m_pkthdr.len += optlen;
1297 		ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1298 	}
1299 	ip = mtod(m, struct ip *);
1300 	bcopy(p->ipopt_list, ip + 1, optlen);
1301 	*phlen = sizeof(struct ip) + optlen;
1302 	ip->ip_v = IPVERSION;
1303 	ip->ip_hl = *phlen >> 2;
1304 	ip->ip_len += optlen;
1305 	return (m);
1306 }
1307 
1308 /*
1309  * Copy options from ip to jp,
1310  * omitting those not copied during fragmentation.
1311  */
1312 int
1313 ip_optcopy(ip, jp)
1314 	struct ip *ip, *jp;
1315 {
1316 	register u_char *cp, *dp;
1317 	int opt, optlen, cnt;
1318 
1319 	cp = (u_char *)(ip + 1);
1320 	dp = (u_char *)(jp + 1);
1321 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1322 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1323 		opt = cp[0];
1324 		if (opt == IPOPT_EOL)
1325 			break;
1326 		if (opt == IPOPT_NOP) {
1327 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1328 			*dp++ = IPOPT_NOP;
1329 			optlen = 1;
1330 			continue;
1331 		}
1332 
1333 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1334 		    ("ip_optcopy: malformed ipv4 option"));
1335 		optlen = cp[IPOPT_OLEN];
1336 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1337 		    ("ip_optcopy: malformed ipv4 option"));
1338 
1339 		/* bogus lengths should have been caught by ip_dooptions */
1340 		if (optlen > cnt)
1341 			optlen = cnt;
1342 		if (IPOPT_COPIED(opt)) {
1343 			bcopy(cp, dp, optlen);
1344 			dp += optlen;
1345 		}
1346 	}
1347 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1348 		*dp++ = IPOPT_EOL;
1349 	return (optlen);
1350 }
1351 
1352 /*
1353  * IP socket option processing.
1354  */
1355 int
1356 ip_ctloutput(so, sopt)
1357 	struct socket *so;
1358 	struct sockopt *sopt;
1359 {
1360 	struct	inpcb *inp = sotoinpcb(so);
1361 	int	error, optval;
1362 
1363 	error = optval = 0;
1364 	if (sopt->sopt_level != IPPROTO_IP) {
1365 		return (EINVAL);
1366 	}
1367 
1368 	switch (sopt->sopt_dir) {
1369 	case SOPT_SET:
1370 		switch (sopt->sopt_name) {
1371 		case IP_OPTIONS:
1372 #ifdef notyet
1373 		case IP_RETOPTS:
1374 #endif
1375 		{
1376 			struct mbuf *m;
1377 			if (sopt->sopt_valsize > MLEN) {
1378 				error = EMSGSIZE;
1379 				break;
1380 			}
1381 			MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_HEADER);
1382 			if (m == 0) {
1383 				error = ENOBUFS;
1384 				break;
1385 			}
1386 			m->m_len = sopt->sopt_valsize;
1387 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1388 					    m->m_len);
1389 
1390 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1391 					   m));
1392 		}
1393 
1394 		case IP_TOS:
1395 		case IP_TTL:
1396 		case IP_RECVOPTS:
1397 		case IP_RECVRETOPTS:
1398 		case IP_RECVDSTADDR:
1399 		case IP_RECVIF:
1400 		case IP_FAITH:
1401 			error = sooptcopyin(sopt, &optval, sizeof optval,
1402 					    sizeof optval);
1403 			if (error)
1404 				break;
1405 
1406 			switch (sopt->sopt_name) {
1407 			case IP_TOS:
1408 				inp->inp_ip_tos = optval;
1409 				break;
1410 
1411 			case IP_TTL:
1412 				inp->inp_ip_ttl = optval;
1413 				break;
1414 #define	OPTSET(bit) \
1415 	if (optval) \
1416 		inp->inp_flags |= bit; \
1417 	else \
1418 		inp->inp_flags &= ~bit;
1419 
1420 			case IP_RECVOPTS:
1421 				OPTSET(INP_RECVOPTS);
1422 				break;
1423 
1424 			case IP_RECVRETOPTS:
1425 				OPTSET(INP_RECVRETOPTS);
1426 				break;
1427 
1428 			case IP_RECVDSTADDR:
1429 				OPTSET(INP_RECVDSTADDR);
1430 				break;
1431 
1432 			case IP_RECVIF:
1433 				OPTSET(INP_RECVIF);
1434 				break;
1435 
1436 			case IP_FAITH:
1437 				OPTSET(INP_FAITH);
1438 				break;
1439 			}
1440 			break;
1441 #undef OPTSET
1442 
1443 		case IP_MULTICAST_IF:
1444 		case IP_MULTICAST_VIF:
1445 		case IP_MULTICAST_TTL:
1446 		case IP_MULTICAST_LOOP:
1447 		case IP_ADD_MEMBERSHIP:
1448 		case IP_DROP_MEMBERSHIP:
1449 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1450 			break;
1451 
1452 		case IP_PORTRANGE:
1453 			error = sooptcopyin(sopt, &optval, sizeof optval,
1454 					    sizeof optval);
1455 			if (error)
1456 				break;
1457 
1458 			switch (optval) {
1459 			case IP_PORTRANGE_DEFAULT:
1460 				inp->inp_flags &= ~(INP_LOWPORT);
1461 				inp->inp_flags &= ~(INP_HIGHPORT);
1462 				break;
1463 
1464 			case IP_PORTRANGE_HIGH:
1465 				inp->inp_flags &= ~(INP_LOWPORT);
1466 				inp->inp_flags |= INP_HIGHPORT;
1467 				break;
1468 
1469 			case IP_PORTRANGE_LOW:
1470 				inp->inp_flags &= ~(INP_HIGHPORT);
1471 				inp->inp_flags |= INP_LOWPORT;
1472 				break;
1473 
1474 			default:
1475 				error = EINVAL;
1476 				break;
1477 			}
1478 			break;
1479 
1480 #if defined(IPSEC) || defined(FAST_IPSEC)
1481 		case IP_IPSEC_POLICY:
1482 		{
1483 			caddr_t req;
1484 			size_t len = 0;
1485 			int priv;
1486 			struct mbuf *m;
1487 			int optname;
1488 
1489 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1490 				break;
1491 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1492 				break;
1493 			priv = (sopt->sopt_td != NULL &&
1494 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1495 			req = mtod(m, caddr_t);
1496 			len = m->m_len;
1497 			optname = sopt->sopt_name;
1498 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1499 			m_freem(m);
1500 			break;
1501 		}
1502 #endif /*IPSEC*/
1503 
1504 		default:
1505 			error = ENOPROTOOPT;
1506 			break;
1507 		}
1508 		break;
1509 
1510 	case SOPT_GET:
1511 		switch (sopt->sopt_name) {
1512 		case IP_OPTIONS:
1513 		case IP_RETOPTS:
1514 			if (inp->inp_options)
1515 				error = sooptcopyout(sopt,
1516 						     mtod(inp->inp_options,
1517 							  char *),
1518 						     inp->inp_options->m_len);
1519 			else
1520 				sopt->sopt_valsize = 0;
1521 			break;
1522 
1523 		case IP_TOS:
1524 		case IP_TTL:
1525 		case IP_RECVOPTS:
1526 		case IP_RECVRETOPTS:
1527 		case IP_RECVDSTADDR:
1528 		case IP_RECVIF:
1529 		case IP_PORTRANGE:
1530 		case IP_FAITH:
1531 			switch (sopt->sopt_name) {
1532 
1533 			case IP_TOS:
1534 				optval = inp->inp_ip_tos;
1535 				break;
1536 
1537 			case IP_TTL:
1538 				optval = inp->inp_ip_ttl;
1539 				break;
1540 
1541 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1542 
1543 			case IP_RECVOPTS:
1544 				optval = OPTBIT(INP_RECVOPTS);
1545 				break;
1546 
1547 			case IP_RECVRETOPTS:
1548 				optval = OPTBIT(INP_RECVRETOPTS);
1549 				break;
1550 
1551 			case IP_RECVDSTADDR:
1552 				optval = OPTBIT(INP_RECVDSTADDR);
1553 				break;
1554 
1555 			case IP_RECVIF:
1556 				optval = OPTBIT(INP_RECVIF);
1557 				break;
1558 
1559 			case IP_PORTRANGE:
1560 				if (inp->inp_flags & INP_HIGHPORT)
1561 					optval = IP_PORTRANGE_HIGH;
1562 				else if (inp->inp_flags & INP_LOWPORT)
1563 					optval = IP_PORTRANGE_LOW;
1564 				else
1565 					optval = 0;
1566 				break;
1567 
1568 			case IP_FAITH:
1569 				optval = OPTBIT(INP_FAITH);
1570 				break;
1571 			}
1572 			error = sooptcopyout(sopt, &optval, sizeof optval);
1573 			break;
1574 
1575 		case IP_MULTICAST_IF:
1576 		case IP_MULTICAST_VIF:
1577 		case IP_MULTICAST_TTL:
1578 		case IP_MULTICAST_LOOP:
1579 		case IP_ADD_MEMBERSHIP:
1580 		case IP_DROP_MEMBERSHIP:
1581 			error = ip_getmoptions(sopt, inp->inp_moptions);
1582 			break;
1583 
1584 #if defined(IPSEC) || defined(FAST_IPSEC)
1585 		case IP_IPSEC_POLICY:
1586 		{
1587 			struct mbuf *m = NULL;
1588 			caddr_t req = NULL;
1589 			size_t len = 0;
1590 
1591 			if (m != 0) {
1592 				req = mtod(m, caddr_t);
1593 				len = m->m_len;
1594 			}
1595 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1596 			if (error == 0)
1597 				error = soopt_mcopyout(sopt, m); /* XXX */
1598 			if (error == 0)
1599 				m_freem(m);
1600 			break;
1601 		}
1602 #endif /*IPSEC*/
1603 
1604 		default:
1605 			error = ENOPROTOOPT;
1606 			break;
1607 		}
1608 		break;
1609 	}
1610 	return (error);
1611 }
1612 
1613 /*
1614  * Set up IP options in pcb for insertion in output packets.
1615  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1616  * with destination address if source routed.
1617  */
1618 static int
1619 ip_pcbopts(optname, pcbopt, m)
1620 	int optname;
1621 	struct mbuf **pcbopt;
1622 	register struct mbuf *m;
1623 {
1624 	register int cnt, optlen;
1625 	register u_char *cp;
1626 	u_char opt;
1627 
1628 	/* turn off any old options */
1629 	if (*pcbopt)
1630 		(void)m_free(*pcbopt);
1631 	*pcbopt = 0;
1632 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1633 		/*
1634 		 * Only turning off any previous options.
1635 		 */
1636 		if (m)
1637 			(void)m_free(m);
1638 		return (0);
1639 	}
1640 
1641 	if (m->m_len % sizeof(int32_t))
1642 		goto bad;
1643 	/*
1644 	 * IP first-hop destination address will be stored before
1645 	 * actual options; move other options back
1646 	 * and clear it when none present.
1647 	 */
1648 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1649 		goto bad;
1650 	cnt = m->m_len;
1651 	m->m_len += sizeof(struct in_addr);
1652 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1653 	ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1654 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1655 
1656 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1657 		opt = cp[IPOPT_OPTVAL];
1658 		if (opt == IPOPT_EOL)
1659 			break;
1660 		if (opt == IPOPT_NOP)
1661 			optlen = 1;
1662 		else {
1663 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1664 				goto bad;
1665 			optlen = cp[IPOPT_OLEN];
1666 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1667 				goto bad;
1668 		}
1669 		switch (opt) {
1670 
1671 		default:
1672 			break;
1673 
1674 		case IPOPT_LSRR:
1675 		case IPOPT_SSRR:
1676 			/*
1677 			 * user process specifies route as:
1678 			 *	->A->B->C->D
1679 			 * D must be our final destination (but we can't
1680 			 * check that since we may not have connected yet).
1681 			 * A is first hop destination, which doesn't appear in
1682 			 * actual IP option, but is stored before the options.
1683 			 */
1684 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1685 				goto bad;
1686 			m->m_len -= sizeof(struct in_addr);
1687 			cnt -= sizeof(struct in_addr);
1688 			optlen -= sizeof(struct in_addr);
1689 			cp[IPOPT_OLEN] = optlen;
1690 			/*
1691 			 * Move first hop before start of options.
1692 			 */
1693 			bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1694 			    sizeof(struct in_addr));
1695 			/*
1696 			 * Then copy rest of options back
1697 			 * to close up the deleted entry.
1698 			 */
1699 			ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1700 			    sizeof(struct in_addr)),
1701 			    (caddr_t)&cp[IPOPT_OFFSET+1],
1702 			    (unsigned)cnt + sizeof(struct in_addr));
1703 			break;
1704 		}
1705 	}
1706 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1707 		goto bad;
1708 	*pcbopt = m;
1709 	return (0);
1710 
1711 bad:
1712 	(void)m_free(m);
1713 	return (EINVAL);
1714 }
1715 
1716 /*
1717  * XXX
1718  * The whole multicast option thing needs to be re-thought.
1719  * Several of these options are equally applicable to non-multicast
1720  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1721  * standard option (IP_TTL).
1722  */
1723 
1724 /*
1725  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1726  */
1727 static struct ifnet *
1728 ip_multicast_if(a, ifindexp)
1729 	struct in_addr *a;
1730 	int *ifindexp;
1731 {
1732 	int ifindex;
1733 	struct ifnet *ifp;
1734 
1735 	if (ifindexp)
1736 		*ifindexp = 0;
1737 	if (ntohl(a->s_addr) >> 24 == 0) {
1738 		ifindex = ntohl(a->s_addr) & 0xffffff;
1739 		if (ifindex < 0 || if_index < ifindex)
1740 			return NULL;
1741 		ifp = ifnet_byindex(ifindex);
1742 		if (ifindexp)
1743 			*ifindexp = ifindex;
1744 	} else {
1745 		INADDR_TO_IFP(*a, ifp);
1746 	}
1747 	return ifp;
1748 }
1749 
1750 /*
1751  * Set the IP multicast options in response to user setsockopt().
1752  */
1753 static int
1754 ip_setmoptions(sopt, imop)
1755 	struct sockopt *sopt;
1756 	struct ip_moptions **imop;
1757 {
1758 	int error = 0;
1759 	int i;
1760 	struct in_addr addr;
1761 	struct ip_mreq mreq;
1762 	struct ifnet *ifp;
1763 	struct ip_moptions *imo = *imop;
1764 	struct route ro;
1765 	struct sockaddr_in *dst;
1766 	int ifindex;
1767 	int s;
1768 
1769 	if (imo == NULL) {
1770 		/*
1771 		 * No multicast option buffer attached to the pcb;
1772 		 * allocate one and initialize to default values.
1773 		 */
1774 		imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1775 		    M_WAITOK);
1776 
1777 		if (imo == NULL)
1778 			return (ENOBUFS);
1779 		*imop = imo;
1780 		imo->imo_multicast_ifp = NULL;
1781 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1782 		imo->imo_multicast_vif = -1;
1783 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1784 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1785 		imo->imo_num_memberships = 0;
1786 	}
1787 
1788 	switch (sopt->sopt_name) {
1789 	/* store an index number for the vif you wanna use in the send */
1790 	case IP_MULTICAST_VIF:
1791 		if (legal_vif_num == 0) {
1792 			error = EOPNOTSUPP;
1793 			break;
1794 		}
1795 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1796 		if (error)
1797 			break;
1798 		if (!legal_vif_num(i) && (i != -1)) {
1799 			error = EINVAL;
1800 			break;
1801 		}
1802 		imo->imo_multicast_vif = i;
1803 		break;
1804 
1805 	case IP_MULTICAST_IF:
1806 		/*
1807 		 * Select the interface for outgoing multicast packets.
1808 		 */
1809 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1810 		if (error)
1811 			break;
1812 		/*
1813 		 * INADDR_ANY is used to remove a previous selection.
1814 		 * When no interface is selected, a default one is
1815 		 * chosen every time a multicast packet is sent.
1816 		 */
1817 		if (addr.s_addr == INADDR_ANY) {
1818 			imo->imo_multicast_ifp = NULL;
1819 			break;
1820 		}
1821 		/*
1822 		 * The selected interface is identified by its local
1823 		 * IP address.  Find the interface and confirm that
1824 		 * it supports multicasting.
1825 		 */
1826 		s = splimp();
1827 		ifp = ip_multicast_if(&addr, &ifindex);
1828 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1829 			splx(s);
1830 			error = EADDRNOTAVAIL;
1831 			break;
1832 		}
1833 		imo->imo_multicast_ifp = ifp;
1834 		if (ifindex)
1835 			imo->imo_multicast_addr = addr;
1836 		else
1837 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1838 		splx(s);
1839 		break;
1840 
1841 	case IP_MULTICAST_TTL:
1842 		/*
1843 		 * Set the IP time-to-live for outgoing multicast packets.
1844 		 * The original multicast API required a char argument,
1845 		 * which is inconsistent with the rest of the socket API.
1846 		 * We allow either a char or an int.
1847 		 */
1848 		if (sopt->sopt_valsize == 1) {
1849 			u_char ttl;
1850 			error = sooptcopyin(sopt, &ttl, 1, 1);
1851 			if (error)
1852 				break;
1853 			imo->imo_multicast_ttl = ttl;
1854 		} else {
1855 			u_int ttl;
1856 			error = sooptcopyin(sopt, &ttl, sizeof ttl,
1857 					    sizeof ttl);
1858 			if (error)
1859 				break;
1860 			if (ttl > 255)
1861 				error = EINVAL;
1862 			else
1863 				imo->imo_multicast_ttl = ttl;
1864 		}
1865 		break;
1866 
1867 	case IP_MULTICAST_LOOP:
1868 		/*
1869 		 * Set the loopback flag for outgoing multicast packets.
1870 		 * Must be zero or one.  The original multicast API required a
1871 		 * char argument, which is inconsistent with the rest
1872 		 * of the socket API.  We allow either a char or an int.
1873 		 */
1874 		if (sopt->sopt_valsize == 1) {
1875 			u_char loop;
1876 			error = sooptcopyin(sopt, &loop, 1, 1);
1877 			if (error)
1878 				break;
1879 			imo->imo_multicast_loop = !!loop;
1880 		} else {
1881 			u_int loop;
1882 			error = sooptcopyin(sopt, &loop, sizeof loop,
1883 					    sizeof loop);
1884 			if (error)
1885 				break;
1886 			imo->imo_multicast_loop = !!loop;
1887 		}
1888 		break;
1889 
1890 	case IP_ADD_MEMBERSHIP:
1891 		/*
1892 		 * Add a multicast group membership.
1893 		 * Group must be a valid IP multicast address.
1894 		 */
1895 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1896 		if (error)
1897 			break;
1898 
1899 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1900 			error = EINVAL;
1901 			break;
1902 		}
1903 		s = splimp();
1904 		/*
1905 		 * If no interface address was provided, use the interface of
1906 		 * the route to the given multicast address.
1907 		 */
1908 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1909 			bzero((caddr_t)&ro, sizeof(ro));
1910 			dst = (struct sockaddr_in *)&ro.ro_dst;
1911 			dst->sin_len = sizeof(*dst);
1912 			dst->sin_family = AF_INET;
1913 			dst->sin_addr = mreq.imr_multiaddr;
1914 			rtalloc(&ro);
1915 			if (ro.ro_rt == NULL) {
1916 				error = EADDRNOTAVAIL;
1917 				splx(s);
1918 				break;
1919 			}
1920 			ifp = ro.ro_rt->rt_ifp;
1921 			rtfree(ro.ro_rt);
1922 		}
1923 		else {
1924 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1925 		}
1926 
1927 		/*
1928 		 * See if we found an interface, and confirm that it
1929 		 * supports multicast.
1930 		 */
1931 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1932 			error = EADDRNOTAVAIL;
1933 			splx(s);
1934 			break;
1935 		}
1936 		/*
1937 		 * See if the membership already exists or if all the
1938 		 * membership slots are full.
1939 		 */
1940 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1941 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1942 			    imo->imo_membership[i]->inm_addr.s_addr
1943 						== mreq.imr_multiaddr.s_addr)
1944 				break;
1945 		}
1946 		if (i < imo->imo_num_memberships) {
1947 			error = EADDRINUSE;
1948 			splx(s);
1949 			break;
1950 		}
1951 		if (i == IP_MAX_MEMBERSHIPS) {
1952 			error = ETOOMANYREFS;
1953 			splx(s);
1954 			break;
1955 		}
1956 		/*
1957 		 * Everything looks good; add a new record to the multicast
1958 		 * address list for the given interface.
1959 		 */
1960 		if ((imo->imo_membership[i] =
1961 		    in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1962 			error = ENOBUFS;
1963 			splx(s);
1964 			break;
1965 		}
1966 		++imo->imo_num_memberships;
1967 		splx(s);
1968 		break;
1969 
1970 	case IP_DROP_MEMBERSHIP:
1971 		/*
1972 		 * Drop a multicast group membership.
1973 		 * Group must be a valid IP multicast address.
1974 		 */
1975 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1976 		if (error)
1977 			break;
1978 
1979 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1980 			error = EINVAL;
1981 			break;
1982 		}
1983 
1984 		s = splimp();
1985 		/*
1986 		 * If an interface address was specified, get a pointer
1987 		 * to its ifnet structure.
1988 		 */
1989 		if (mreq.imr_interface.s_addr == INADDR_ANY)
1990 			ifp = NULL;
1991 		else {
1992 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1993 			if (ifp == NULL) {
1994 				error = EADDRNOTAVAIL;
1995 				splx(s);
1996 				break;
1997 			}
1998 		}
1999 		/*
2000 		 * Find the membership in the membership array.
2001 		 */
2002 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2003 			if ((ifp == NULL ||
2004 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2005 			     imo->imo_membership[i]->inm_addr.s_addr ==
2006 			     mreq.imr_multiaddr.s_addr)
2007 				break;
2008 		}
2009 		if (i == imo->imo_num_memberships) {
2010 			error = EADDRNOTAVAIL;
2011 			splx(s);
2012 			break;
2013 		}
2014 		/*
2015 		 * Give up the multicast address record to which the
2016 		 * membership points.
2017 		 */
2018 		in_delmulti(imo->imo_membership[i]);
2019 		/*
2020 		 * Remove the gap in the membership array.
2021 		 */
2022 		for (++i; i < imo->imo_num_memberships; ++i)
2023 			imo->imo_membership[i-1] = imo->imo_membership[i];
2024 		--imo->imo_num_memberships;
2025 		splx(s);
2026 		break;
2027 
2028 	default:
2029 		error = EOPNOTSUPP;
2030 		break;
2031 	}
2032 
2033 	/*
2034 	 * If all options have default values, no need to keep the mbuf.
2035 	 */
2036 	if (imo->imo_multicast_ifp == NULL &&
2037 	    imo->imo_multicast_vif == -1 &&
2038 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2039 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2040 	    imo->imo_num_memberships == 0) {
2041 		free(*imop, M_IPMOPTS);
2042 		*imop = NULL;
2043 	}
2044 
2045 	return (error);
2046 }
2047 
2048 /*
2049  * Return the IP multicast options in response to user getsockopt().
2050  */
2051 static int
2052 ip_getmoptions(sopt, imo)
2053 	struct sockopt *sopt;
2054 	register struct ip_moptions *imo;
2055 {
2056 	struct in_addr addr;
2057 	struct in_ifaddr *ia;
2058 	int error, optval;
2059 	u_char coptval;
2060 
2061 	error = 0;
2062 	switch (sopt->sopt_name) {
2063 	case IP_MULTICAST_VIF:
2064 		if (imo != NULL)
2065 			optval = imo->imo_multicast_vif;
2066 		else
2067 			optval = -1;
2068 		error = sooptcopyout(sopt, &optval, sizeof optval);
2069 		break;
2070 
2071 	case IP_MULTICAST_IF:
2072 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2073 			addr.s_addr = INADDR_ANY;
2074 		else if (imo->imo_multicast_addr.s_addr) {
2075 			/* return the value user has set */
2076 			addr = imo->imo_multicast_addr;
2077 		} else {
2078 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2079 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2080 				: IA_SIN(ia)->sin_addr.s_addr;
2081 		}
2082 		error = sooptcopyout(sopt, &addr, sizeof addr);
2083 		break;
2084 
2085 	case IP_MULTICAST_TTL:
2086 		if (imo == 0)
2087 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2088 		else
2089 			optval = coptval = imo->imo_multicast_ttl;
2090 		if (sopt->sopt_valsize == 1)
2091 			error = sooptcopyout(sopt, &coptval, 1);
2092 		else
2093 			error = sooptcopyout(sopt, &optval, sizeof optval);
2094 		break;
2095 
2096 	case IP_MULTICAST_LOOP:
2097 		if (imo == 0)
2098 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2099 		else
2100 			optval = coptval = imo->imo_multicast_loop;
2101 		if (sopt->sopt_valsize == 1)
2102 			error = sooptcopyout(sopt, &coptval, 1);
2103 		else
2104 			error = sooptcopyout(sopt, &optval, sizeof optval);
2105 		break;
2106 
2107 	default:
2108 		error = ENOPROTOOPT;
2109 		break;
2110 	}
2111 	return (error);
2112 }
2113 
2114 /*
2115  * Discard the IP multicast options.
2116  */
2117 void
2118 ip_freemoptions(imo)
2119 	register struct ip_moptions *imo;
2120 {
2121 	register int i;
2122 
2123 	if (imo != NULL) {
2124 		for (i = 0; i < imo->imo_num_memberships; ++i)
2125 			in_delmulti(imo->imo_membership[i]);
2126 		free(imo, M_IPMOPTS);
2127 	}
2128 }
2129 
2130 /*
2131  * Routine called from ip_output() to loop back a copy of an IP multicast
2132  * packet to the input queue of a specified interface.  Note that this
2133  * calls the output routine of the loopback "driver", but with an interface
2134  * pointer that might NOT be a loopback interface -- evil, but easier than
2135  * replicating that code here.
2136  */
2137 static void
2138 ip_mloopback(ifp, m, dst, hlen)
2139 	struct ifnet *ifp;
2140 	register struct mbuf *m;
2141 	register struct sockaddr_in *dst;
2142 	int hlen;
2143 {
2144 	register struct ip *ip;
2145 	struct mbuf *copym;
2146 
2147 	copym = m_copy(m, 0, M_COPYALL);
2148 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2149 		copym = m_pullup(copym, hlen);
2150 	if (copym != NULL) {
2151 		/*
2152 		 * We don't bother to fragment if the IP length is greater
2153 		 * than the interface's MTU.  Can this possibly matter?
2154 		 */
2155 		ip = mtod(copym, struct ip *);
2156 		ip->ip_len = htons(ip->ip_len);
2157 		ip->ip_off = htons(ip->ip_off);
2158 		ip->ip_sum = 0;
2159 		ip->ip_sum = in_cksum(copym, hlen);
2160 		/*
2161 		 * NB:
2162 		 * It's not clear whether there are any lingering
2163 		 * reentrancy problems in other areas which might
2164 		 * be exposed by using ip_input directly (in
2165 		 * particular, everything which modifies the packet
2166 		 * in-place).  Yet another option is using the
2167 		 * protosw directly to deliver the looped back
2168 		 * packet.  For the moment, we'll err on the side
2169 		 * of safety by using if_simloop().
2170 		 */
2171 #if 1 /* XXX */
2172 		if (dst->sin_family != AF_INET) {
2173 			printf("ip_mloopback: bad address family %d\n",
2174 						dst->sin_family);
2175 			dst->sin_family = AF_INET;
2176 		}
2177 #endif
2178 
2179 #ifdef notdef
2180 		copym->m_pkthdr.rcvif = ifp;
2181 		ip_input(copym);
2182 #else
2183 		/* if the checksum hasn't been computed, mark it as valid */
2184 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2185 			copym->m_pkthdr.csum_flags |=
2186 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2187 			copym->m_pkthdr.csum_data = 0xffff;
2188 		}
2189 		if_simloop(ifp, copym, dst->sin_family, 0);
2190 #endif
2191 	}
2192 }
2193