xref: /freebsd/sys/netinet/ip_output.c (revision 729362425c09cf6b362366aabc6fb547eee8035a)
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 #include "opt_mbuf_frag_test.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/mac.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/protosw.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/sysctl.h>
57 
58 #include <net/if.h>
59 #include <net/route.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 
68 #include <machine/in_cksum.h>
69 
70 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
71 
72 #ifdef IPSEC
73 #include <netinet6/ipsec.h>
74 #include <netkey/key.h>
75 #ifdef IPSEC_DEBUG
76 #include <netkey/key_debug.h>
77 #else
78 #define	KEYDEBUG(lev,arg)
79 #endif
80 #endif /*IPSEC*/
81 
82 #ifdef FAST_IPSEC
83 #include <netipsec/ipsec.h>
84 #include <netipsec/xform.h>
85 #include <netipsec/key.h>
86 #endif /*FAST_IPSEC*/
87 
88 #include <netinet/ip_fw.h>
89 #include <netinet/ip_dummynet.h>
90 
91 #define print_ip(x, a, y)	 printf("%s %d.%d.%d.%d%s",\
92 				x, (ntohl(a.s_addr)>>24)&0xFF,\
93 				  (ntohl(a.s_addr)>>16)&0xFF,\
94 				  (ntohl(a.s_addr)>>8)&0xFF,\
95 				  (ntohl(a.s_addr))&0xFF, y);
96 
97 u_short ip_id;
98 
99 #ifdef MBUF_FRAG_TEST
100 int mbuf_frag_size = 0;
101 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
102 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
103 #endif
104 
105 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
106 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
107 static void	ip_mloopback
108 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
109 static int	ip_getmoptions
110 	(struct sockopt *, struct ip_moptions *);
111 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
112 static int	ip_setmoptions
113 	(struct sockopt *, struct ip_moptions **);
114 
115 int	ip_optcopy(struct ip *, struct ip *);
116 
117 
118 extern	struct protosw inetsw[];
119 
120 /*
121  * IP output.  The packet in mbuf chain m contains a skeletal IP
122  * header (with len, off, ttl, proto, tos, src, dst).
123  * The mbuf chain containing the packet will be freed.
124  * The mbuf opt, if present, will not be freed.
125  */
126 int
127 ip_output(m0, opt, ro, flags, imo, inp)
128 	struct mbuf *m0;
129 	struct mbuf *opt;
130 	struct route *ro;
131 	int flags;
132 	struct ip_moptions *imo;
133 	struct inpcb *inp;
134 {
135 	struct ip *ip, *mhip;
136 	struct ifnet *ifp = NULL;	/* keep compiler happy */
137 	struct mbuf *m;
138 	int hlen = sizeof (struct ip);
139 	int len, off, error = 0;
140 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
141 	struct in_ifaddr *ia = NULL;
142 	int isbroadcast, sw_csum;
143 	struct in_addr pkt_dst;
144 #ifdef IPSEC
145 	struct route iproute;
146 	struct secpolicy *sp = NULL;
147 #endif
148 #ifdef FAST_IPSEC
149 	struct route iproute;
150 	struct m_tag *mtag;
151 	struct secpolicy *sp = NULL;
152 	struct tdb_ident *tdbi;
153 	int s;
154 #endif /* FAST_IPSEC */
155 	struct ip_fw_args args;
156 	int src_was_INADDR_ANY = 0;	/* as the name says... */
157 #ifdef PFIL_HOOKS
158 	struct packet_filter_hook *pfh;
159 	struct mbuf *m1;
160 	int rv;
161 #endif /* PFIL_HOOKS */
162 
163 	args.eh = NULL;
164 	args.rule = NULL;
165 	args.next_hop = NULL;
166 	args.divert_rule = 0;			/* divert cookie */
167 
168 	/* Grab info from MT_TAG mbufs prepended to the chain. */
169 	for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
170 		switch(m0->_m_tag_id) {
171 		default:
172 			printf("ip_output: unrecognised MT_TAG tag %d\n",
173 			    m0->_m_tag_id);
174 			break;
175 
176 		case PACKET_TAG_DUMMYNET:
177 			/*
178 			 * the packet was already tagged, so part of the
179 			 * processing was already done, and we need to go down.
180 			 * Get parameters from the header.
181 			 */
182 			args.rule = ((struct dn_pkt *)m0)->rule;
183 			opt = NULL ;
184 			ro = & ( ((struct dn_pkt *)m0)->ro ) ;
185 			imo = NULL ;
186 			dst = ((struct dn_pkt *)m0)->dn_dst ;
187 			ifp = ((struct dn_pkt *)m0)->ifp ;
188 			flags = ((struct dn_pkt *)m0)->flags ;
189 			break;
190 
191 		case PACKET_TAG_DIVERT:
192 			args.divert_rule = (intptr_t)m0->m_data & 0xffff;
193 			break;
194 
195 		case PACKET_TAG_IPFORWARD:
196 			args.next_hop = (struct sockaddr_in *)m0->m_data;
197 			break;
198 		}
199 	}
200 	m = m0;
201 
202 	KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
203 #ifndef FAST_IPSEC
204 	KASSERT(ro != NULL, ("ip_output: no route, proto %d",
205 	    mtod(m, struct ip *)->ip_p));
206 #endif
207 
208 	if (args.rule != NULL) {	/* dummynet already saw us */
209 		ip = mtod(m, struct ip *);
210 		hlen = ip->ip_hl << 2 ;
211 		if (ro->ro_rt)
212 			ia = ifatoia(ro->ro_rt->rt_ifa);
213 		goto sendit;
214 	}
215 
216 	if (opt) {
217 		len = 0;
218 		m = ip_insertoptions(m, opt, &len);
219 		if (len != 0)
220 			hlen = len;
221 	}
222 	ip = mtod(m, struct ip *);
223 	pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
224 
225 	/*
226 	 * Fill in IP header.
227 	 */
228 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
229 		ip->ip_v = IPVERSION;
230 		ip->ip_hl = hlen >> 2;
231 		ip->ip_off &= IP_DF;
232 #ifdef RANDOM_IP_ID
233 		ip->ip_id = ip_randomid();
234 #else
235 		ip->ip_id = htons(ip_id++);
236 #endif
237 		ipstat.ips_localout++;
238 	} else {
239 		hlen = ip->ip_hl << 2;
240 	}
241 
242 #ifdef FAST_IPSEC
243 	if (ro == NULL) {
244 		ro = &iproute;
245 		bzero(ro, sizeof (*ro));
246 	}
247 #endif /* FAST_IPSEC */
248 	dst = (struct sockaddr_in *)&ro->ro_dst;
249 	/*
250 	 * If there is a cached route,
251 	 * check that it is to the same destination
252 	 * and is still up.  If not, free it and try again.
253 	 * The address family should also be checked in case of sharing the
254 	 * cache with IPv6.
255 	 */
256 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
257 			  dst->sin_family != AF_INET ||
258 			  dst->sin_addr.s_addr != pkt_dst.s_addr)) {
259 		RTFREE(ro->ro_rt);
260 		ro->ro_rt = (struct rtentry *)0;
261 	}
262 	if (ro->ro_rt == 0) {
263 		bzero(dst, sizeof(*dst));
264 		dst->sin_family = AF_INET;
265 		dst->sin_len = sizeof(*dst);
266 		dst->sin_addr = pkt_dst;
267 	}
268 	/*
269 	 * If routing to interface only,
270 	 * short circuit routing lookup.
271 	 */
272 	if (flags & IP_ROUTETOIF) {
273 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
274 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
275 			ipstat.ips_noroute++;
276 			error = ENETUNREACH;
277 			goto bad;
278 		}
279 		ifp = ia->ia_ifp;
280 		ip->ip_ttl = 1;
281 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
282 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
283 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
284 		/*
285 		 * Bypass the normal routing lookup for multicast
286 		 * packets if the interface is specified.
287 		 */
288 		ifp = imo->imo_multicast_ifp;
289 		IFP_TO_IA(ifp, ia);
290 		isbroadcast = 0;	/* fool gcc */
291 	} else {
292 		/*
293 		 * If this is the case, we probably don't want to allocate
294 		 * a protocol-cloned route since we didn't get one from the
295 		 * ULP.  This lets TCP do its thing, while not burdening
296 		 * forwarding or ICMP with the overhead of cloning a route.
297 		 * Of course, we still want to do any cloning requested by
298 		 * the link layer, as this is probably required in all cases
299 		 * for correct operation (as it is for ARP).
300 		 */
301 		if (ro->ro_rt == 0)
302 			rtalloc_ign(ro, RTF_PRCLONING);
303 		if (ro->ro_rt == 0) {
304 			ipstat.ips_noroute++;
305 			error = EHOSTUNREACH;
306 			goto bad;
307 		}
308 		ia = ifatoia(ro->ro_rt->rt_ifa);
309 		ifp = ro->ro_rt->rt_ifp;
310 		ro->ro_rt->rt_use++;
311 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
312 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
313 		if (ro->ro_rt->rt_flags & RTF_HOST)
314 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
315 		else
316 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
317 	}
318 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
319 		struct in_multi *inm;
320 
321 		m->m_flags |= M_MCAST;
322 		/*
323 		 * IP destination address is multicast.  Make sure "dst"
324 		 * still points to the address in "ro".  (It may have been
325 		 * changed to point to a gateway address, above.)
326 		 */
327 		dst = (struct sockaddr_in *)&ro->ro_dst;
328 		/*
329 		 * See if the caller provided any multicast options
330 		 */
331 		if (imo != NULL) {
332 			ip->ip_ttl = imo->imo_multicast_ttl;
333 			if (imo->imo_multicast_vif != -1)
334 				ip->ip_src.s_addr =
335 				    ip_mcast_src ?
336 				    ip_mcast_src(imo->imo_multicast_vif) :
337 				    INADDR_ANY;
338 		} else
339 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
340 		/*
341 		 * Confirm that the outgoing interface supports multicast.
342 		 */
343 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
344 			if ((ifp->if_flags & IFF_MULTICAST) == 0) {
345 				ipstat.ips_noroute++;
346 				error = ENETUNREACH;
347 				goto bad;
348 			}
349 		}
350 		/*
351 		 * If source address not specified yet, use address
352 		 * of outgoing interface.
353 		 */
354 		if (ip->ip_src.s_addr == INADDR_ANY) {
355 			/* Interface may have no addresses. */
356 			if (ia != NULL)
357 				ip->ip_src = IA_SIN(ia)->sin_addr;
358 		}
359 
360 		if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
361 			/*
362 			 * XXX
363 			 * delayed checksums are not currently
364 			 * compatible with IP multicast routing
365 			 */
366 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
367 				in_delayed_cksum(m);
368 				m->m_pkthdr.csum_flags &=
369 					~CSUM_DELAY_DATA;
370 			}
371 		}
372 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
373 		if (inm != NULL &&
374 		   (imo == NULL || imo->imo_multicast_loop)) {
375 			/*
376 			 * If we belong to the destination multicast group
377 			 * on the outgoing interface, and the caller did not
378 			 * forbid loopback, loop back a copy.
379 			 */
380 			ip_mloopback(ifp, m, dst, hlen);
381 		}
382 		else {
383 			/*
384 			 * If we are acting as a multicast router, perform
385 			 * multicast forwarding as if the packet had just
386 			 * arrived on the interface to which we are about
387 			 * to send.  The multicast forwarding function
388 			 * recursively calls this function, using the
389 			 * IP_FORWARDING flag to prevent infinite recursion.
390 			 *
391 			 * Multicasts that are looped back by ip_mloopback(),
392 			 * above, will be forwarded by the ip_input() routine,
393 			 * if necessary.
394 			 */
395 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
396 				/*
397 				 * If rsvp daemon is not running, do not
398 				 * set ip_moptions. This ensures that the packet
399 				 * is multicast and not just sent down one link
400 				 * as prescribed by rsvpd.
401 				 */
402 				if (!rsvp_on)
403 					imo = NULL;
404 				if (ip_mforward &&
405 				    ip_mforward(ip, ifp, m, imo) != 0) {
406 					m_freem(m);
407 					goto done;
408 				}
409 			}
410 		}
411 
412 		/*
413 		 * Multicasts with a time-to-live of zero may be looped-
414 		 * back, above, but must not be transmitted on a network.
415 		 * Also, multicasts addressed to the loopback interface
416 		 * are not sent -- the above call to ip_mloopback() will
417 		 * loop back a copy if this host actually belongs to the
418 		 * destination group on the loopback interface.
419 		 */
420 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
421 			m_freem(m);
422 			goto done;
423 		}
424 
425 		goto sendit;
426 	}
427 #ifndef notdef
428 	/*
429 	 * If the source address is not specified yet, use the address
430 	 * of the outoing interface. In case, keep note we did that, so
431 	 * if the the firewall changes the next-hop causing the output
432 	 * interface to change, we can fix that.
433 	 */
434 	if (ip->ip_src.s_addr == INADDR_ANY) {
435 		/* Interface may have no addresses. */
436 		if (ia != NULL) {
437 			ip->ip_src = IA_SIN(ia)->sin_addr;
438 			src_was_INADDR_ANY = 1;
439 		}
440 	}
441 #endif /* notdef */
442 	/*
443 	 * Verify that we have any chance at all of being able to queue
444 	 *      the packet or packet fragments
445 	 */
446 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
447 		ifp->if_snd.ifq_maxlen) {
448 			error = ENOBUFS;
449 			ipstat.ips_odropped++;
450 			goto bad;
451 	}
452 
453 	/*
454 	 * Look for broadcast address and
455 	 * verify user is allowed to send
456 	 * such a packet.
457 	 */
458 	if (isbroadcast) {
459 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
460 			error = EADDRNOTAVAIL;
461 			goto bad;
462 		}
463 		if ((flags & IP_ALLOWBROADCAST) == 0) {
464 			error = EACCES;
465 			goto bad;
466 		}
467 		/* don't allow broadcast messages to be fragmented */
468 		if ((u_short)ip->ip_len > ifp->if_mtu) {
469 			error = EMSGSIZE;
470 			goto bad;
471 		}
472 		m->m_flags |= M_BCAST;
473 	} else {
474 		m->m_flags &= ~M_BCAST;
475 	}
476 
477 sendit:
478 #ifdef IPSEC
479 	/* get SP for this packet */
480 	if (inp == NULL)
481 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
482 	else
483 		sp = ipsec4_getpolicybypcb(m, IPSEC_DIR_OUTBOUND, inp, &error);
484 
485 	if (sp == NULL) {
486 		ipsecstat.out_inval++;
487 		goto bad;
488 	}
489 
490 	error = 0;
491 
492 	/* check policy */
493 	switch (sp->policy) {
494 	case IPSEC_POLICY_DISCARD:
495 		/*
496 		 * This packet is just discarded.
497 		 */
498 		ipsecstat.out_polvio++;
499 		goto bad;
500 
501 	case IPSEC_POLICY_BYPASS:
502 	case IPSEC_POLICY_NONE:
503 		/* no need to do IPsec. */
504 		goto skip_ipsec;
505 
506 	case IPSEC_POLICY_IPSEC:
507 		if (sp->req == NULL) {
508 			/* acquire a policy */
509 			error = key_spdacquire(sp);
510 			goto bad;
511 		}
512 		break;
513 
514 	case IPSEC_POLICY_ENTRUST:
515 	default:
516 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
517 	}
518     {
519 	struct ipsec_output_state state;
520 	bzero(&state, sizeof(state));
521 	state.m = m;
522 	if (flags & IP_ROUTETOIF) {
523 		state.ro = &iproute;
524 		bzero(&iproute, sizeof(iproute));
525 	} else
526 		state.ro = ro;
527 	state.dst = (struct sockaddr *)dst;
528 
529 	ip->ip_sum = 0;
530 
531 	/*
532 	 * XXX
533 	 * delayed checksums are not currently compatible with IPsec
534 	 */
535 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
536 		in_delayed_cksum(m);
537 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
538 	}
539 
540 	ip->ip_len = htons(ip->ip_len);
541 	ip->ip_off = htons(ip->ip_off);
542 
543 	error = ipsec4_output(&state, sp, flags);
544 
545 	m = state.m;
546 	if (flags & IP_ROUTETOIF) {
547 		/*
548 		 * if we have tunnel mode SA, we may need to ignore
549 		 * IP_ROUTETOIF.
550 		 */
551 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
552 			flags &= ~IP_ROUTETOIF;
553 			ro = state.ro;
554 		}
555 	} else
556 		ro = state.ro;
557 	dst = (struct sockaddr_in *)state.dst;
558 	if (error) {
559 		/* mbuf is already reclaimed in ipsec4_output. */
560 		m0 = NULL;
561 		switch (error) {
562 		case EHOSTUNREACH:
563 		case ENETUNREACH:
564 		case EMSGSIZE:
565 		case ENOBUFS:
566 		case ENOMEM:
567 			break;
568 		default:
569 			printf("ip4_output (ipsec): error code %d\n", error);
570 			/*fall through*/
571 		case ENOENT:
572 			/* don't show these error codes to the user */
573 			error = 0;
574 			break;
575 		}
576 		goto bad;
577 	}
578     }
579 
580 	/* be sure to update variables that are affected by ipsec4_output() */
581 	ip = mtod(m, struct ip *);
582 	hlen = ip->ip_hl << 2;
583 	if (ro->ro_rt == NULL) {
584 		if ((flags & IP_ROUTETOIF) == 0) {
585 			printf("ip_output: "
586 				"can't update route after IPsec processing\n");
587 			error = EHOSTUNREACH;	/*XXX*/
588 			goto bad;
589 		}
590 	} else {
591 		ia = ifatoia(ro->ro_rt->rt_ifa);
592 		ifp = ro->ro_rt->rt_ifp;
593 	}
594 
595 	/* make it flipped, again. */
596 	ip->ip_len = ntohs(ip->ip_len);
597 	ip->ip_off = ntohs(ip->ip_off);
598 skip_ipsec:
599 #endif /*IPSEC*/
600 #ifdef FAST_IPSEC
601 	/*
602 	 * Check the security policy (SP) for the packet and, if
603 	 * required, do IPsec-related processing.  There are two
604 	 * cases here; the first time a packet is sent through
605 	 * it will be untagged and handled by ipsec4_checkpolicy.
606 	 * If the packet is resubmitted to ip_output (e.g. after
607 	 * AH, ESP, etc. processing), there will be a tag to bypass
608 	 * the lookup and related policy checking.
609 	 */
610 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
611 	s = splnet();
612 	if (mtag != NULL) {
613 		tdbi = (struct tdb_ident *)(mtag + 1);
614 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
615 		if (sp == NULL)
616 			error = -EINVAL;	/* force silent drop */
617 		m_tag_delete(m, mtag);
618 	} else {
619 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
620 					&error, inp);
621 	}
622 	/*
623 	 * There are four return cases:
624 	 *    sp != NULL	 	    apply IPsec policy
625 	 *    sp == NULL, error == 0	    no IPsec handling needed
626 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
627 	 *    sp == NULL, error != 0	    discard packet, report error
628 	 */
629 	if (sp != NULL) {
630 		/* Loop detection, check if ipsec processing already done */
631 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
632 		for (mtag = m_tag_first(m); mtag != NULL;
633 		     mtag = m_tag_next(m, mtag)) {
634 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
635 				continue;
636 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
637 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
638 				continue;
639 			/*
640 			 * Check if policy has an SA associated with it.
641 			 * This can happen when an SP has yet to acquire
642 			 * an SA; e.g. on first reference.  If it occurs,
643 			 * then we let ipsec4_process_packet do its thing.
644 			 */
645 			if (sp->req->sav == NULL)
646 				break;
647 			tdbi = (struct tdb_ident *)(mtag + 1);
648 			if (tdbi->spi == sp->req->sav->spi &&
649 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
650 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
651 				 sizeof (union sockaddr_union)) == 0) {
652 				/*
653 				 * No IPsec processing is needed, free
654 				 * reference to SP.
655 				 *
656 				 * NB: null pointer to avoid free at
657 				 *     done: below.
658 				 */
659 				KEY_FREESP(&sp), sp = NULL;
660 				splx(s);
661 				goto spd_done;
662 			}
663 		}
664 
665 		/*
666 		 * Do delayed checksums now because we send before
667 		 * this is done in the normal processing path.
668 		 */
669 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
670 			in_delayed_cksum(m);
671 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
672 		}
673 
674 		ip->ip_len = htons(ip->ip_len);
675 		ip->ip_off = htons(ip->ip_off);
676 
677 		/* NB: callee frees mbuf */
678 		error = ipsec4_process_packet(m, sp->req, flags, 0);
679 		/*
680 		 * Preserve KAME behaviour: ENOENT can be returned
681 		 * when an SA acquire is in progress.  Don't propagate
682 		 * this to user-level; it confuses applications.
683 		 *
684 		 * XXX this will go away when the SADB is redone.
685 		 */
686 		if (error == ENOENT)
687 			error = 0;
688 		splx(s);
689 		goto done;
690 	} else {
691 		splx(s);
692 
693 		if (error != 0) {
694 			/*
695 			 * Hack: -EINVAL is used to signal that a packet
696 			 * should be silently discarded.  This is typically
697 			 * because we asked key management for an SA and
698 			 * it was delayed (e.g. kicked up to IKE).
699 			 */
700 			if (error == -EINVAL)
701 				error = 0;
702 			goto bad;
703 		} else {
704 			/* No IPsec processing for this packet. */
705 		}
706 #ifdef notyet
707 		/*
708 		 * If deferred crypto processing is needed, check that
709 		 * the interface supports it.
710 		 */
711 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
712 		if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
713 			/* notify IPsec to do its own crypto */
714 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
715 			error = EHOSTUNREACH;
716 			goto bad;
717 		}
718 #endif
719 	}
720 spd_done:
721 #endif /* FAST_IPSEC */
722 
723 	/*
724 	 * IpHack's section.
725 	 * - Xlate: translate packet's addr/port (NAT).
726 	 * - Firewall: deny/allow/etc.
727 	 * - Wrap: fake packet's addr/port <unimpl.>
728 	 * - Encapsulate: put it in another IP and send out. <unimp.>
729 	 */
730 #ifdef PFIL_HOOKS
731 	/*
732 	 * Run through list of hooks for output packets.
733 	 */
734 	m1 = m;
735 	pfh = pfil_hook_get(PFIL_OUT, &inetsw[ip_protox[IPPROTO_IP]].pr_pfh);
736 	for (; pfh; pfh = TAILQ_NEXT(pfh, pfil_link))
737 		if (pfh->pfil_func) {
738 			rv = pfh->pfil_func(ip, hlen, ifp, 1, &m1);
739 			if (rv) {
740 				error = EHOSTUNREACH;
741 				goto done;
742 			}
743 			m = m1;
744 			if (m == NULL)
745 				goto done;
746 			ip = mtod(m, struct ip *);
747 		}
748 #endif /* PFIL_HOOKS */
749 
750 	/*
751 	 * Check with the firewall...
752 	 * but not if we are already being fwd'd from a firewall.
753 	 */
754 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
755 		struct sockaddr_in *old = dst;
756 
757 		args.m = m;
758 		args.next_hop = dst;
759 		args.oif = ifp;
760 		off = ip_fw_chk_ptr(&args);
761 		m = args.m;
762 		dst = args.next_hop;
763 
764                 /*
765 		 * On return we must do the following:
766 		 * m == NULL	-> drop the pkt (old interface, deprecated)
767 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
768 		 * 1<=off<= 0xffff		-> DIVERT
769 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
770 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
771 		 * dst != old			-> IPFIREWALL_FORWARD
772 		 * off==0, dst==old		-> accept
773 		 * If some of the above modules are not compiled in, then
774 		 * we should't have to check the corresponding condition
775 		 * (because the ipfw control socket should not accept
776 		 * unsupported rules), but better play safe and drop
777 		 * packets in case of doubt.
778 		 */
779 		if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
780 			if (m)
781 				m_freem(m);
782 			error = EACCES;
783 			goto done;
784 		}
785 		ip = mtod(m, struct ip *);
786 		if (off == 0 && dst == old)		/* common case */
787 			goto pass;
788                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
789 			/*
790 			 * pass the pkt to dummynet. Need to include
791 			 * pipe number, m, ifp, ro, dst because these are
792 			 * not recomputed in the next pass.
793 			 * All other parameters have been already used and
794 			 * so they are not needed anymore.
795 			 * XXX note: if the ifp or ro entry are deleted
796 			 * while a pkt is in dummynet, we are in trouble!
797 			 */
798 			args.ro = ro;
799 			args.dst = dst;
800 			args.flags = flags;
801 
802 			error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
803 				&args);
804 			goto done;
805 		}
806 #ifdef IPDIVERT
807 		if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
808 			struct mbuf *clone = NULL;
809 
810 			/* Clone packet if we're doing a 'tee' */
811 			if ((off & IP_FW_PORT_TEE_FLAG) != 0)
812 				clone = m_dup(m, M_DONTWAIT);
813 
814 			/*
815 			 * XXX
816 			 * delayed checksums are not currently compatible
817 			 * with divert sockets.
818 			 */
819 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
820 				in_delayed_cksum(m);
821 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
822 			}
823 
824 			/* Restore packet header fields to original values */
825 			ip->ip_len = htons(ip->ip_len);
826 			ip->ip_off = htons(ip->ip_off);
827 
828 			/* Deliver packet to divert input routine */
829 			divert_packet(m, 0, off & 0xffff, args.divert_rule);
830 
831 			/* If 'tee', continue with original packet */
832 			if (clone != NULL) {
833 				m = clone;
834 				ip = mtod(m, struct ip *);
835 				goto pass;
836 			}
837 			goto done;
838 		}
839 #endif
840 
841 		/* IPFIREWALL_FORWARD */
842 		/*
843 		 * Check dst to make sure it is directly reachable on the
844 		 * interface we previously thought it was.
845 		 * If it isn't (which may be likely in some situations) we have
846 		 * to re-route it (ie, find a route for the next-hop and the
847 		 * associated interface) and set them here. This is nested
848 		 * forwarding which in most cases is undesirable, except where
849 		 * such control is nigh impossible. So we do it here.
850 		 * And I'm babbling.
851 		 */
852 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
853 #if 0
854 			/*
855 			 * XXX To improve readability, this block should be
856 			 * changed into a function call as below:
857 			 */
858 			error = ip_ipforward(&m, &dst, &ifp);
859 			if (error)
860 				goto bad;
861 			if (m == NULL) /* ip_input consumed the mbuf */
862 				goto done;
863 #else
864 			struct in_ifaddr *ia;
865 
866 			/*
867 			 * XXX sro_fwd below is static, and a pointer
868 			 * to it gets passed to routines downstream.
869 			 * This could have surprisingly bad results in
870 			 * practice, because its content is overwritten
871 			 * by subsequent packets.
872 			 */
873 			/* There must be a better way to do this next line... */
874 			static struct route sro_fwd;
875 			struct route *ro_fwd = &sro_fwd;
876 
877 #if 0
878 			print_ip("IPFIREWALL_FORWARD: New dst ip: ",
879 			    dst->sin_addr, "\n");
880 #endif
881 
882 			/*
883 			 * We need to figure out if we have been forwarded
884 			 * to a local socket. If so, then we should somehow
885 			 * "loop back" to ip_input, and get directed to the
886 			 * PCB as if we had received this packet. This is
887 			 * because it may be dificult to identify the packets
888 			 * you want to forward until they are being output
889 			 * and have selected an interface. (e.g. locally
890 			 * initiated packets) If we used the loopback inteface,
891 			 * we would not be able to control what happens
892 			 * as the packet runs through ip_input() as
893 			 * it is done through an ISR.
894 			 */
895 			LIST_FOREACH(ia,
896 			    INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
897 				/*
898 				 * If the addr to forward to is one
899 				 * of ours, we pretend to
900 				 * be the destination for this packet.
901 				 */
902 				if (IA_SIN(ia)->sin_addr.s_addr ==
903 						 dst->sin_addr.s_addr)
904 					break;
905 			}
906 			if (ia) {	/* tell ip_input "dont filter" */
907 				struct m_hdr tag;
908 
909 				tag.mh_type = MT_TAG;
910 				tag.mh_flags = PACKET_TAG_IPFORWARD;
911 				tag.mh_data = (caddr_t)args.next_hop;
912 				tag.mh_next = m;
913 
914 				if (m->m_pkthdr.rcvif == NULL)
915 					m->m_pkthdr.rcvif = ifunit("lo0");
916 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
917 					m->m_pkthdr.csum_flags |=
918 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
919 					m0->m_pkthdr.csum_data = 0xffff;
920 				}
921 				m->m_pkthdr.csum_flags |=
922 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
923 				ip->ip_len = htons(ip->ip_len);
924 				ip->ip_off = htons(ip->ip_off);
925 				ip_input((struct mbuf *)&tag);
926 				goto done;
927 			}
928 			/* Some of the logic for this was
929 			 * nicked from above.
930 			 *
931 			 * This rewrites the cached route in a local PCB.
932 			 * Is this what we want to do?
933 			 */
934 			bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
935 
936 			ro_fwd->ro_rt = 0;
937 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
938 
939 			if (ro_fwd->ro_rt == 0) {
940 				ipstat.ips_noroute++;
941 				error = EHOSTUNREACH;
942 				goto bad;
943 			}
944 
945 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
946 			ifp = ro_fwd->ro_rt->rt_ifp;
947 			ro_fwd->ro_rt->rt_use++;
948 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
949 				dst = (struct sockaddr_in *)
950 					ro_fwd->ro_rt->rt_gateway;
951 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
952 				isbroadcast =
953 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
954 			else
955 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
956 			if (ro->ro_rt)
957 				RTFREE(ro->ro_rt);
958 			ro->ro_rt = ro_fwd->ro_rt;
959 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
960 
961 #endif	/* ... block to be put into a function */
962 			/*
963 			 * If we added a default src ip earlier,
964 			 * which would have been gotten from the-then
965 			 * interface, do it again, from the new one.
966 			 */
967 			if (src_was_INADDR_ANY)
968 				ip->ip_src = IA_SIN(ia)->sin_addr;
969 			goto pass ;
970 		}
971 
972                 /*
973                  * if we get here, none of the above matches, and
974                  * we have to drop the pkt
975                  */
976 		m_freem(m);
977                 error = EACCES; /* not sure this is the right error msg */
978                 goto done;
979 	}
980 
981 pass:
982 	/* 127/8 must not appear on wire - RFC1122. */
983 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
984 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
985 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
986 			ipstat.ips_badaddr++;
987 			error = EADDRNOTAVAIL;
988 			goto bad;
989 		}
990 	}
991 
992 	m->m_pkthdr.csum_flags |= CSUM_IP;
993 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
994 	if (sw_csum & CSUM_DELAY_DATA) {
995 		in_delayed_cksum(m);
996 		sw_csum &= ~CSUM_DELAY_DATA;
997 	}
998 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
999 
1000 	/*
1001 	 * If small enough for interface, or the interface will take
1002 	 * care of the fragmentation for us, can just send directly.
1003 	 */
1004 	if ((u_short)ip->ip_len <= ifp->if_mtu ||
1005 	    ifp->if_hwassist & CSUM_FRAGMENT) {
1006 		ip->ip_len = htons(ip->ip_len);
1007 		ip->ip_off = htons(ip->ip_off);
1008 		ip->ip_sum = 0;
1009 		if (sw_csum & CSUM_DELAY_IP)
1010 			ip->ip_sum = in_cksum(m, hlen);
1011 
1012 		/* Record statistics for this interface address. */
1013 		if (!(flags & IP_FORWARDING) && ia) {
1014 			ia->ia_ifa.if_opackets++;
1015 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1016 		}
1017 
1018 #ifdef IPSEC
1019 		/* clean ipsec history once it goes out of the node */
1020 		ipsec_delaux(m);
1021 #endif
1022 
1023 #ifdef MBUF_FRAG_TEST
1024 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1025 			struct mbuf *m1, *m2;
1026 			int length, tmp;
1027 
1028 			tmp = length = m->m_pkthdr.len;
1029 
1030 			while ((length -= mbuf_frag_size) >= 1) {
1031 				m1 = m_split(m, length, M_DONTWAIT);
1032 				if (m1 == NULL)
1033 					break;
1034 				m1->m_flags &= ~M_PKTHDR;
1035 				m2 = m;
1036 				while (m2->m_next != NULL)
1037 					m2 = m2->m_next;
1038 				m2->m_next = m1;
1039 			}
1040 			m->m_pkthdr.len = tmp;
1041 		}
1042 #endif
1043 		error = (*ifp->if_output)(ifp, m,
1044 				(struct sockaddr *)dst, ro->ro_rt);
1045 		goto done;
1046 	}
1047 	/*
1048 	 * Too large for interface; fragment if possible.
1049 	 * Must be able to put at least 8 bytes per fragment.
1050 	 */
1051 	if (ip->ip_off & IP_DF) {
1052 		error = EMSGSIZE;
1053 		/*
1054 		 * This case can happen if the user changed the MTU
1055 		 * of an interface after enabling IP on it.  Because
1056 		 * most netifs don't keep track of routes pointing to
1057 		 * them, there is no way for one to update all its
1058 		 * routes when the MTU is changed.
1059 		 */
1060 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST))
1061 		    && !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU)
1062 		    && (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1063 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1064 		}
1065 		ipstat.ips_cantfrag++;
1066 		goto bad;
1067 	}
1068 	len = (ifp->if_mtu - hlen) &~ 7;
1069 	if (len < 8) {
1070 		error = EMSGSIZE;
1071 		goto bad;
1072 	}
1073 
1074 	/*
1075 	 * if the interface will not calculate checksums on
1076 	 * fragmented packets, then do it here.
1077 	 */
1078 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1079 	    (ifp->if_hwassist & CSUM_IP_FRAGS) == 0) {
1080 		in_delayed_cksum(m);
1081 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1082 	}
1083 
1084 	if (len > PAGE_SIZE) {
1085 		/*
1086 		 * Fragement large datagrams such that each segment
1087 		 * contains a multiple of PAGE_SIZE amount of data,
1088 		 * plus headers. This enables a receiver to perform
1089 		 * page-flipping zero-copy optimizations.
1090 		 */
1091 
1092 		int newlen;
1093 		struct mbuf *mtmp;
1094 
1095 		for (mtmp = m, off = 0;
1096 		     mtmp && ((off + mtmp->m_len) <= ifp->if_mtu);
1097 		     mtmp = mtmp->m_next) {
1098 			off += mtmp->m_len;
1099 		}
1100 		/*
1101 		 * firstlen (off - hlen) must be aligned on an
1102 		 * 8-byte boundary
1103 		 */
1104 		if (off < hlen)
1105 			goto smart_frag_failure;
1106 		off = ((off - hlen) & ~7) + hlen;
1107 		newlen = (~PAGE_MASK) & ifp->if_mtu;
1108 		if ((newlen + sizeof (struct ip)) > ifp->if_mtu) {
1109 			/* we failed, go back the default */
1110 smart_frag_failure:
1111 			newlen = len;
1112 			off = hlen + len;
1113 		}
1114 
1115 /*		printf("ipfrag: len = %d, hlen = %d, mhlen = %d, newlen = %d, off = %d\n",
1116 		len, hlen, sizeof (struct ip), newlen, off);*/
1117 
1118 		len = newlen;
1119 
1120 	} else {
1121 		off = hlen + len;
1122 	}
1123 
1124 
1125 
1126     {
1127 	int mhlen, firstlen = off - hlen;
1128 	struct mbuf **mnext = &m->m_nextpkt;
1129 	int nfrags = 1;
1130 
1131 	/*
1132 	 * Loop through length of segment after first fragment,
1133 	 * make new header and copy data of each part and link onto chain.
1134 	 */
1135 	m0 = m;
1136 	mhlen = sizeof (struct ip);
1137 	for (; off < (u_short)ip->ip_len; off += len) {
1138 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
1139 		if (m == 0) {
1140 			error = ENOBUFS;
1141 			ipstat.ips_odropped++;
1142 			goto sendorfree;
1143 		}
1144 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1145 		m->m_data += max_linkhdr;
1146 		mhip = mtod(m, struct ip *);
1147 		*mhip = *ip;
1148 		if (hlen > sizeof (struct ip)) {
1149 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1150 			mhip->ip_v = IPVERSION;
1151 			mhip->ip_hl = mhlen >> 2;
1152 		}
1153 		m->m_len = mhlen;
1154 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1155 		if (off + len >= (u_short)ip->ip_len)
1156 			len = (u_short)ip->ip_len - off;
1157 		else
1158 			mhip->ip_off |= IP_MF;
1159 		mhip->ip_len = htons((u_short)(len + mhlen));
1160 		m->m_next = m_copy(m0, off, len);
1161 		if (m->m_next == 0) {
1162 			(void) m_free(m);
1163 			error = ENOBUFS;	/* ??? */
1164 			ipstat.ips_odropped++;
1165 			goto sendorfree;
1166 		}
1167 		m->m_pkthdr.len = mhlen + len;
1168 		m->m_pkthdr.rcvif = (struct ifnet *)0;
1169 #ifdef MAC
1170 		mac_create_fragment(m0, m);
1171 #endif
1172 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1173 		mhip->ip_off = htons(mhip->ip_off);
1174 		mhip->ip_sum = 0;
1175 		if (sw_csum & CSUM_DELAY_IP)
1176 			mhip->ip_sum = in_cksum(m, mhlen);
1177 		*mnext = m;
1178 		mnext = &m->m_nextpkt;
1179 		nfrags++;
1180 	}
1181 	ipstat.ips_ofragments += nfrags;
1182 
1183 	/* set first/last markers for fragment chain */
1184 	m->m_flags |= M_LASTFRAG;
1185 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1186 	m0->m_pkthdr.csum_data = nfrags;
1187 
1188 	/*
1189 	 * Update first fragment by trimming what's been copied out
1190 	 * and updating header, then send each fragment (in order).
1191 	 */
1192 	m = m0;
1193 	m_adj(m, hlen + firstlen - (u_short)ip->ip_len);
1194 	m->m_pkthdr.len = hlen + firstlen;
1195 	ip->ip_len = htons((u_short)m->m_pkthdr.len);
1196 	ip->ip_off |= IP_MF;
1197 	ip->ip_off = htons(ip->ip_off);
1198 	ip->ip_sum = 0;
1199 	if (sw_csum & CSUM_DELAY_IP)
1200 		ip->ip_sum = in_cksum(m, hlen);
1201 sendorfree:
1202 	for (m = m0; m; m = m0) {
1203 		m0 = m->m_nextpkt;
1204 		m->m_nextpkt = 0;
1205 #ifdef IPSEC
1206 		/* clean ipsec history once it goes out of the node */
1207 		ipsec_delaux(m);
1208 #endif
1209 		if (error == 0) {
1210 			/* Record statistics for this interface address. */
1211 			if (ia != NULL) {
1212 				ia->ia_ifa.if_opackets++;
1213 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1214 			}
1215 
1216 			error = (*ifp->if_output)(ifp, m,
1217 			    (struct sockaddr *)dst, ro->ro_rt);
1218 		} else
1219 			m_freem(m);
1220 	}
1221 
1222 	if (error == 0)
1223 		ipstat.ips_fragmented++;
1224     }
1225 done:
1226 #ifdef IPSEC
1227 	if (ro == &iproute && ro->ro_rt) {
1228 		RTFREE(ro->ro_rt);
1229 		ro->ro_rt = NULL;
1230 	}
1231 	if (sp != NULL) {
1232 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1233 			printf("DP ip_output call free SP:%p\n", sp));
1234 		key_freesp(sp);
1235 	}
1236 #endif /* IPSEC */
1237 #ifdef FAST_IPSEC
1238 	if (ro == &iproute && ro->ro_rt) {
1239 		RTFREE(ro->ro_rt);
1240 		ro->ro_rt = NULL;
1241 	}
1242 	if (sp != NULL)
1243 		KEY_FREESP(&sp);
1244 #endif /* FAST_IPSEC */
1245 	return (error);
1246 bad:
1247 	m_freem(m);
1248 	goto done;
1249 }
1250 
1251 void
1252 in_delayed_cksum(struct mbuf *m)
1253 {
1254 	struct ip *ip;
1255 	u_short csum, offset;
1256 
1257 	ip = mtod(m, struct ip *);
1258 	offset = ip->ip_hl << 2 ;
1259 	csum = in_cksum_skip(m, ip->ip_len, offset);
1260 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1261 		csum = 0xffff;
1262 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1263 
1264 	if (offset + sizeof(u_short) > m->m_len) {
1265 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1266 		    m->m_len, offset, ip->ip_p);
1267 		/*
1268 		 * XXX
1269 		 * this shouldn't happen, but if it does, the
1270 		 * correct behavior may be to insert the checksum
1271 		 * in the existing chain instead of rearranging it.
1272 		 */
1273 		m = m_pullup(m, offset + sizeof(u_short));
1274 	}
1275 	*(u_short *)(m->m_data + offset) = csum;
1276 }
1277 
1278 /*
1279  * Insert IP options into preformed packet.
1280  * Adjust IP destination as required for IP source routing,
1281  * as indicated by a non-zero in_addr at the start of the options.
1282  *
1283  * XXX This routine assumes that the packet has no options in place.
1284  */
1285 static struct mbuf *
1286 ip_insertoptions(m, opt, phlen)
1287 	register struct mbuf *m;
1288 	struct mbuf *opt;
1289 	int *phlen;
1290 {
1291 	register struct ipoption *p = mtod(opt, struct ipoption *);
1292 	struct mbuf *n;
1293 	register struct ip *ip = mtod(m, struct ip *);
1294 	unsigned optlen;
1295 
1296 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1297 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1298 		*phlen = 0;
1299 		return (m);		/* XXX should fail */
1300 	}
1301 	if (p->ipopt_dst.s_addr)
1302 		ip->ip_dst = p->ipopt_dst;
1303 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1304 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1305 		if (n == 0) {
1306 			*phlen = 0;
1307 			return (m);
1308 		}
1309 		n->m_pkthdr.rcvif = (struct ifnet *)0;
1310 #ifdef MAC
1311 		mac_create_mbuf_from_mbuf(m, n);
1312 #endif
1313 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1314 		m->m_len -= sizeof(struct ip);
1315 		m->m_data += sizeof(struct ip);
1316 		n->m_next = m;
1317 		m = n;
1318 		m->m_len = optlen + sizeof(struct ip);
1319 		m->m_data += max_linkhdr;
1320 		(void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1321 	} else {
1322 		m->m_data -= optlen;
1323 		m->m_len += optlen;
1324 		m->m_pkthdr.len += optlen;
1325 		ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1326 	}
1327 	ip = mtod(m, struct ip *);
1328 	bcopy(p->ipopt_list, ip + 1, optlen);
1329 	*phlen = sizeof(struct ip) + optlen;
1330 	ip->ip_v = IPVERSION;
1331 	ip->ip_hl = *phlen >> 2;
1332 	ip->ip_len += optlen;
1333 	return (m);
1334 }
1335 
1336 /*
1337  * Copy options from ip to jp,
1338  * omitting those not copied during fragmentation.
1339  */
1340 int
1341 ip_optcopy(ip, jp)
1342 	struct ip *ip, *jp;
1343 {
1344 	register u_char *cp, *dp;
1345 	int opt, optlen, cnt;
1346 
1347 	cp = (u_char *)(ip + 1);
1348 	dp = (u_char *)(jp + 1);
1349 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1350 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1351 		opt = cp[0];
1352 		if (opt == IPOPT_EOL)
1353 			break;
1354 		if (opt == IPOPT_NOP) {
1355 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1356 			*dp++ = IPOPT_NOP;
1357 			optlen = 1;
1358 			continue;
1359 		}
1360 
1361 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1362 		    ("ip_optcopy: malformed ipv4 option"));
1363 		optlen = cp[IPOPT_OLEN];
1364 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1365 		    ("ip_optcopy: malformed ipv4 option"));
1366 
1367 		/* bogus lengths should have been caught by ip_dooptions */
1368 		if (optlen > cnt)
1369 			optlen = cnt;
1370 		if (IPOPT_COPIED(opt)) {
1371 			bcopy(cp, dp, optlen);
1372 			dp += optlen;
1373 		}
1374 	}
1375 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1376 		*dp++ = IPOPT_EOL;
1377 	return (optlen);
1378 }
1379 
1380 /*
1381  * IP socket option processing.
1382  */
1383 int
1384 ip_ctloutput(so, sopt)
1385 	struct socket *so;
1386 	struct sockopt *sopt;
1387 {
1388 	struct	inpcb *inp = sotoinpcb(so);
1389 	int	error, optval;
1390 
1391 	error = optval = 0;
1392 	if (sopt->sopt_level != IPPROTO_IP) {
1393 		return (EINVAL);
1394 	}
1395 
1396 	switch (sopt->sopt_dir) {
1397 	case SOPT_SET:
1398 		switch (sopt->sopt_name) {
1399 		case IP_OPTIONS:
1400 #ifdef notyet
1401 		case IP_RETOPTS:
1402 #endif
1403 		{
1404 			struct mbuf *m;
1405 			if (sopt->sopt_valsize > MLEN) {
1406 				error = EMSGSIZE;
1407 				break;
1408 			}
1409 			MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_HEADER);
1410 			if (m == 0) {
1411 				error = ENOBUFS;
1412 				break;
1413 			}
1414 			m->m_len = sopt->sopt_valsize;
1415 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1416 					    m->m_len);
1417 
1418 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1419 					   m));
1420 		}
1421 
1422 		case IP_TOS:
1423 		case IP_TTL:
1424 		case IP_RECVOPTS:
1425 		case IP_RECVRETOPTS:
1426 		case IP_RECVDSTADDR:
1427 		case IP_RECVIF:
1428 		case IP_FAITH:
1429 			error = sooptcopyin(sopt, &optval, sizeof optval,
1430 					    sizeof optval);
1431 			if (error)
1432 				break;
1433 
1434 			switch (sopt->sopt_name) {
1435 			case IP_TOS:
1436 				inp->inp_ip_tos = optval;
1437 				break;
1438 
1439 			case IP_TTL:
1440 				inp->inp_ip_ttl = optval;
1441 				break;
1442 #define	OPTSET(bit) \
1443 	if (optval) \
1444 		inp->inp_flags |= bit; \
1445 	else \
1446 		inp->inp_flags &= ~bit;
1447 
1448 			case IP_RECVOPTS:
1449 				OPTSET(INP_RECVOPTS);
1450 				break;
1451 
1452 			case IP_RECVRETOPTS:
1453 				OPTSET(INP_RECVRETOPTS);
1454 				break;
1455 
1456 			case IP_RECVDSTADDR:
1457 				OPTSET(INP_RECVDSTADDR);
1458 				break;
1459 
1460 			case IP_RECVIF:
1461 				OPTSET(INP_RECVIF);
1462 				break;
1463 
1464 			case IP_FAITH:
1465 				OPTSET(INP_FAITH);
1466 				break;
1467 			}
1468 			break;
1469 #undef OPTSET
1470 
1471 		case IP_MULTICAST_IF:
1472 		case IP_MULTICAST_VIF:
1473 		case IP_MULTICAST_TTL:
1474 		case IP_MULTICAST_LOOP:
1475 		case IP_ADD_MEMBERSHIP:
1476 		case IP_DROP_MEMBERSHIP:
1477 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1478 			break;
1479 
1480 		case IP_PORTRANGE:
1481 			error = sooptcopyin(sopt, &optval, sizeof optval,
1482 					    sizeof optval);
1483 			if (error)
1484 				break;
1485 
1486 			switch (optval) {
1487 			case IP_PORTRANGE_DEFAULT:
1488 				inp->inp_flags &= ~(INP_LOWPORT);
1489 				inp->inp_flags &= ~(INP_HIGHPORT);
1490 				break;
1491 
1492 			case IP_PORTRANGE_HIGH:
1493 				inp->inp_flags &= ~(INP_LOWPORT);
1494 				inp->inp_flags |= INP_HIGHPORT;
1495 				break;
1496 
1497 			case IP_PORTRANGE_LOW:
1498 				inp->inp_flags &= ~(INP_HIGHPORT);
1499 				inp->inp_flags |= INP_LOWPORT;
1500 				break;
1501 
1502 			default:
1503 				error = EINVAL;
1504 				break;
1505 			}
1506 			break;
1507 
1508 #if defined(IPSEC) || defined(FAST_IPSEC)
1509 		case IP_IPSEC_POLICY:
1510 		{
1511 			caddr_t req;
1512 			size_t len = 0;
1513 			int priv;
1514 			struct mbuf *m;
1515 			int optname;
1516 
1517 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1518 				break;
1519 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1520 				break;
1521 			priv = (sopt->sopt_td != NULL &&
1522 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1523 			req = mtod(m, caddr_t);
1524 			len = m->m_len;
1525 			optname = sopt->sopt_name;
1526 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1527 			m_freem(m);
1528 			break;
1529 		}
1530 #endif /*IPSEC*/
1531 
1532 		default:
1533 			error = ENOPROTOOPT;
1534 			break;
1535 		}
1536 		break;
1537 
1538 	case SOPT_GET:
1539 		switch (sopt->sopt_name) {
1540 		case IP_OPTIONS:
1541 		case IP_RETOPTS:
1542 			if (inp->inp_options)
1543 				error = sooptcopyout(sopt,
1544 						     mtod(inp->inp_options,
1545 							  char *),
1546 						     inp->inp_options->m_len);
1547 			else
1548 				sopt->sopt_valsize = 0;
1549 			break;
1550 
1551 		case IP_TOS:
1552 		case IP_TTL:
1553 		case IP_RECVOPTS:
1554 		case IP_RECVRETOPTS:
1555 		case IP_RECVDSTADDR:
1556 		case IP_RECVIF:
1557 		case IP_PORTRANGE:
1558 		case IP_FAITH:
1559 			switch (sopt->sopt_name) {
1560 
1561 			case IP_TOS:
1562 				optval = inp->inp_ip_tos;
1563 				break;
1564 
1565 			case IP_TTL:
1566 				optval = inp->inp_ip_ttl;
1567 				break;
1568 
1569 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1570 
1571 			case IP_RECVOPTS:
1572 				optval = OPTBIT(INP_RECVOPTS);
1573 				break;
1574 
1575 			case IP_RECVRETOPTS:
1576 				optval = OPTBIT(INP_RECVRETOPTS);
1577 				break;
1578 
1579 			case IP_RECVDSTADDR:
1580 				optval = OPTBIT(INP_RECVDSTADDR);
1581 				break;
1582 
1583 			case IP_RECVIF:
1584 				optval = OPTBIT(INP_RECVIF);
1585 				break;
1586 
1587 			case IP_PORTRANGE:
1588 				if (inp->inp_flags & INP_HIGHPORT)
1589 					optval = IP_PORTRANGE_HIGH;
1590 				else if (inp->inp_flags & INP_LOWPORT)
1591 					optval = IP_PORTRANGE_LOW;
1592 				else
1593 					optval = 0;
1594 				break;
1595 
1596 			case IP_FAITH:
1597 				optval = OPTBIT(INP_FAITH);
1598 				break;
1599 			}
1600 			error = sooptcopyout(sopt, &optval, sizeof optval);
1601 			break;
1602 
1603 		case IP_MULTICAST_IF:
1604 		case IP_MULTICAST_VIF:
1605 		case IP_MULTICAST_TTL:
1606 		case IP_MULTICAST_LOOP:
1607 		case IP_ADD_MEMBERSHIP:
1608 		case IP_DROP_MEMBERSHIP:
1609 			error = ip_getmoptions(sopt, inp->inp_moptions);
1610 			break;
1611 
1612 #if defined(IPSEC) || defined(FAST_IPSEC)
1613 		case IP_IPSEC_POLICY:
1614 		{
1615 			struct mbuf *m = NULL;
1616 			caddr_t req = NULL;
1617 			size_t len = 0;
1618 
1619 			if (m != 0) {
1620 				req = mtod(m, caddr_t);
1621 				len = m->m_len;
1622 			}
1623 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1624 			if (error == 0)
1625 				error = soopt_mcopyout(sopt, m); /* XXX */
1626 			if (error == 0)
1627 				m_freem(m);
1628 			break;
1629 		}
1630 #endif /*IPSEC*/
1631 
1632 		default:
1633 			error = ENOPROTOOPT;
1634 			break;
1635 		}
1636 		break;
1637 	}
1638 	return (error);
1639 }
1640 
1641 /*
1642  * Set up IP options in pcb for insertion in output packets.
1643  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1644  * with destination address if source routed.
1645  */
1646 static int
1647 ip_pcbopts(optname, pcbopt, m)
1648 	int optname;
1649 	struct mbuf **pcbopt;
1650 	register struct mbuf *m;
1651 {
1652 	register int cnt, optlen;
1653 	register u_char *cp;
1654 	u_char opt;
1655 
1656 	/* turn off any old options */
1657 	if (*pcbopt)
1658 		(void)m_free(*pcbopt);
1659 	*pcbopt = 0;
1660 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1661 		/*
1662 		 * Only turning off any previous options.
1663 		 */
1664 		if (m)
1665 			(void)m_free(m);
1666 		return (0);
1667 	}
1668 
1669 	if (m->m_len % sizeof(int32_t))
1670 		goto bad;
1671 	/*
1672 	 * IP first-hop destination address will be stored before
1673 	 * actual options; move other options back
1674 	 * and clear it when none present.
1675 	 */
1676 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1677 		goto bad;
1678 	cnt = m->m_len;
1679 	m->m_len += sizeof(struct in_addr);
1680 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1681 	ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1682 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1683 
1684 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1685 		opt = cp[IPOPT_OPTVAL];
1686 		if (opt == IPOPT_EOL)
1687 			break;
1688 		if (opt == IPOPT_NOP)
1689 			optlen = 1;
1690 		else {
1691 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1692 				goto bad;
1693 			optlen = cp[IPOPT_OLEN];
1694 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1695 				goto bad;
1696 		}
1697 		switch (opt) {
1698 
1699 		default:
1700 			break;
1701 
1702 		case IPOPT_LSRR:
1703 		case IPOPT_SSRR:
1704 			/*
1705 			 * user process specifies route as:
1706 			 *	->A->B->C->D
1707 			 * D must be our final destination (but we can't
1708 			 * check that since we may not have connected yet).
1709 			 * A is first hop destination, which doesn't appear in
1710 			 * actual IP option, but is stored before the options.
1711 			 */
1712 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1713 				goto bad;
1714 			m->m_len -= sizeof(struct in_addr);
1715 			cnt -= sizeof(struct in_addr);
1716 			optlen -= sizeof(struct in_addr);
1717 			cp[IPOPT_OLEN] = optlen;
1718 			/*
1719 			 * Move first hop before start of options.
1720 			 */
1721 			bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1722 			    sizeof(struct in_addr));
1723 			/*
1724 			 * Then copy rest of options back
1725 			 * to close up the deleted entry.
1726 			 */
1727 			ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1728 			    sizeof(struct in_addr)),
1729 			    (caddr_t)&cp[IPOPT_OFFSET+1],
1730 			    (unsigned)cnt + sizeof(struct in_addr));
1731 			break;
1732 		}
1733 	}
1734 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1735 		goto bad;
1736 	*pcbopt = m;
1737 	return (0);
1738 
1739 bad:
1740 	(void)m_free(m);
1741 	return (EINVAL);
1742 }
1743 
1744 /*
1745  * XXX
1746  * The whole multicast option thing needs to be re-thought.
1747  * Several of these options are equally applicable to non-multicast
1748  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1749  * standard option (IP_TTL).
1750  */
1751 
1752 /*
1753  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1754  */
1755 static struct ifnet *
1756 ip_multicast_if(a, ifindexp)
1757 	struct in_addr *a;
1758 	int *ifindexp;
1759 {
1760 	int ifindex;
1761 	struct ifnet *ifp;
1762 
1763 	if (ifindexp)
1764 		*ifindexp = 0;
1765 	if (ntohl(a->s_addr) >> 24 == 0) {
1766 		ifindex = ntohl(a->s_addr) & 0xffffff;
1767 		if (ifindex < 0 || if_index < ifindex)
1768 			return NULL;
1769 		ifp = ifnet_byindex(ifindex);
1770 		if (ifindexp)
1771 			*ifindexp = ifindex;
1772 	} else {
1773 		INADDR_TO_IFP(*a, ifp);
1774 	}
1775 	return ifp;
1776 }
1777 
1778 /*
1779  * Set the IP multicast options in response to user setsockopt().
1780  */
1781 static int
1782 ip_setmoptions(sopt, imop)
1783 	struct sockopt *sopt;
1784 	struct ip_moptions **imop;
1785 {
1786 	int error = 0;
1787 	int i;
1788 	struct in_addr addr;
1789 	struct ip_mreq mreq;
1790 	struct ifnet *ifp;
1791 	struct ip_moptions *imo = *imop;
1792 	struct route ro;
1793 	struct sockaddr_in *dst;
1794 	int ifindex;
1795 	int s;
1796 
1797 	if (imo == NULL) {
1798 		/*
1799 		 * No multicast option buffer attached to the pcb;
1800 		 * allocate one and initialize to default values.
1801 		 */
1802 		imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1803 		    M_WAITOK);
1804 
1805 		if (imo == NULL)
1806 			return (ENOBUFS);
1807 		*imop = imo;
1808 		imo->imo_multicast_ifp = NULL;
1809 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1810 		imo->imo_multicast_vif = -1;
1811 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1812 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1813 		imo->imo_num_memberships = 0;
1814 	}
1815 
1816 	switch (sopt->sopt_name) {
1817 	/* store an index number for the vif you wanna use in the send */
1818 	case IP_MULTICAST_VIF:
1819 		if (legal_vif_num == 0) {
1820 			error = EOPNOTSUPP;
1821 			break;
1822 		}
1823 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1824 		if (error)
1825 			break;
1826 		if (!legal_vif_num(i) && (i != -1)) {
1827 			error = EINVAL;
1828 			break;
1829 		}
1830 		imo->imo_multicast_vif = i;
1831 		break;
1832 
1833 	case IP_MULTICAST_IF:
1834 		/*
1835 		 * Select the interface for outgoing multicast packets.
1836 		 */
1837 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1838 		if (error)
1839 			break;
1840 		/*
1841 		 * INADDR_ANY is used to remove a previous selection.
1842 		 * When no interface is selected, a default one is
1843 		 * chosen every time a multicast packet is sent.
1844 		 */
1845 		if (addr.s_addr == INADDR_ANY) {
1846 			imo->imo_multicast_ifp = NULL;
1847 			break;
1848 		}
1849 		/*
1850 		 * The selected interface is identified by its local
1851 		 * IP address.  Find the interface and confirm that
1852 		 * it supports multicasting.
1853 		 */
1854 		s = splimp();
1855 		ifp = ip_multicast_if(&addr, &ifindex);
1856 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1857 			splx(s);
1858 			error = EADDRNOTAVAIL;
1859 			break;
1860 		}
1861 		imo->imo_multicast_ifp = ifp;
1862 		if (ifindex)
1863 			imo->imo_multicast_addr = addr;
1864 		else
1865 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1866 		splx(s);
1867 		break;
1868 
1869 	case IP_MULTICAST_TTL:
1870 		/*
1871 		 * Set the IP time-to-live for outgoing multicast packets.
1872 		 * The original multicast API required a char argument,
1873 		 * which is inconsistent with the rest of the socket API.
1874 		 * We allow either a char or an int.
1875 		 */
1876 		if (sopt->sopt_valsize == 1) {
1877 			u_char ttl;
1878 			error = sooptcopyin(sopt, &ttl, 1, 1);
1879 			if (error)
1880 				break;
1881 			imo->imo_multicast_ttl = ttl;
1882 		} else {
1883 			u_int ttl;
1884 			error = sooptcopyin(sopt, &ttl, sizeof ttl,
1885 					    sizeof ttl);
1886 			if (error)
1887 				break;
1888 			if (ttl > 255)
1889 				error = EINVAL;
1890 			else
1891 				imo->imo_multicast_ttl = ttl;
1892 		}
1893 		break;
1894 
1895 	case IP_MULTICAST_LOOP:
1896 		/*
1897 		 * Set the loopback flag for outgoing multicast packets.
1898 		 * Must be zero or one.  The original multicast API required a
1899 		 * char argument, which is inconsistent with the rest
1900 		 * of the socket API.  We allow either a char or an int.
1901 		 */
1902 		if (sopt->sopt_valsize == 1) {
1903 			u_char loop;
1904 			error = sooptcopyin(sopt, &loop, 1, 1);
1905 			if (error)
1906 				break;
1907 			imo->imo_multicast_loop = !!loop;
1908 		} else {
1909 			u_int loop;
1910 			error = sooptcopyin(sopt, &loop, sizeof loop,
1911 					    sizeof loop);
1912 			if (error)
1913 				break;
1914 			imo->imo_multicast_loop = !!loop;
1915 		}
1916 		break;
1917 
1918 	case IP_ADD_MEMBERSHIP:
1919 		/*
1920 		 * Add a multicast group membership.
1921 		 * Group must be a valid IP multicast address.
1922 		 */
1923 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1924 		if (error)
1925 			break;
1926 
1927 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1928 			error = EINVAL;
1929 			break;
1930 		}
1931 		s = splimp();
1932 		/*
1933 		 * If no interface address was provided, use the interface of
1934 		 * the route to the given multicast address.
1935 		 */
1936 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1937 			bzero((caddr_t)&ro, sizeof(ro));
1938 			dst = (struct sockaddr_in *)&ro.ro_dst;
1939 			dst->sin_len = sizeof(*dst);
1940 			dst->sin_family = AF_INET;
1941 			dst->sin_addr = mreq.imr_multiaddr;
1942 			rtalloc(&ro);
1943 			if (ro.ro_rt == NULL) {
1944 				error = EADDRNOTAVAIL;
1945 				splx(s);
1946 				break;
1947 			}
1948 			ifp = ro.ro_rt->rt_ifp;
1949 			rtfree(ro.ro_rt);
1950 		}
1951 		else {
1952 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1953 		}
1954 
1955 		/*
1956 		 * See if we found an interface, and confirm that it
1957 		 * supports multicast.
1958 		 */
1959 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1960 			error = EADDRNOTAVAIL;
1961 			splx(s);
1962 			break;
1963 		}
1964 		/*
1965 		 * See if the membership already exists or if all the
1966 		 * membership slots are full.
1967 		 */
1968 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1969 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1970 			    imo->imo_membership[i]->inm_addr.s_addr
1971 						== mreq.imr_multiaddr.s_addr)
1972 				break;
1973 		}
1974 		if (i < imo->imo_num_memberships) {
1975 			error = EADDRINUSE;
1976 			splx(s);
1977 			break;
1978 		}
1979 		if (i == IP_MAX_MEMBERSHIPS) {
1980 			error = ETOOMANYREFS;
1981 			splx(s);
1982 			break;
1983 		}
1984 		/*
1985 		 * Everything looks good; add a new record to the multicast
1986 		 * address list for the given interface.
1987 		 */
1988 		if ((imo->imo_membership[i] =
1989 		    in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1990 			error = ENOBUFS;
1991 			splx(s);
1992 			break;
1993 		}
1994 		++imo->imo_num_memberships;
1995 		splx(s);
1996 		break;
1997 
1998 	case IP_DROP_MEMBERSHIP:
1999 		/*
2000 		 * Drop a multicast group membership.
2001 		 * Group must be a valid IP multicast address.
2002 		 */
2003 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2004 		if (error)
2005 			break;
2006 
2007 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2008 			error = EINVAL;
2009 			break;
2010 		}
2011 
2012 		s = splimp();
2013 		/*
2014 		 * If an interface address was specified, get a pointer
2015 		 * to its ifnet structure.
2016 		 */
2017 		if (mreq.imr_interface.s_addr == INADDR_ANY)
2018 			ifp = NULL;
2019 		else {
2020 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2021 			if (ifp == NULL) {
2022 				error = EADDRNOTAVAIL;
2023 				splx(s);
2024 				break;
2025 			}
2026 		}
2027 		/*
2028 		 * Find the membership in the membership array.
2029 		 */
2030 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2031 			if ((ifp == NULL ||
2032 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2033 			     imo->imo_membership[i]->inm_addr.s_addr ==
2034 			     mreq.imr_multiaddr.s_addr)
2035 				break;
2036 		}
2037 		if (i == imo->imo_num_memberships) {
2038 			error = EADDRNOTAVAIL;
2039 			splx(s);
2040 			break;
2041 		}
2042 		/*
2043 		 * Give up the multicast address record to which the
2044 		 * membership points.
2045 		 */
2046 		in_delmulti(imo->imo_membership[i]);
2047 		/*
2048 		 * Remove the gap in the membership array.
2049 		 */
2050 		for (++i; i < imo->imo_num_memberships; ++i)
2051 			imo->imo_membership[i-1] = imo->imo_membership[i];
2052 		--imo->imo_num_memberships;
2053 		splx(s);
2054 		break;
2055 
2056 	default:
2057 		error = EOPNOTSUPP;
2058 		break;
2059 	}
2060 
2061 	/*
2062 	 * If all options have default values, no need to keep the mbuf.
2063 	 */
2064 	if (imo->imo_multicast_ifp == NULL &&
2065 	    imo->imo_multicast_vif == -1 &&
2066 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2067 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2068 	    imo->imo_num_memberships == 0) {
2069 		free(*imop, M_IPMOPTS);
2070 		*imop = NULL;
2071 	}
2072 
2073 	return (error);
2074 }
2075 
2076 /*
2077  * Return the IP multicast options in response to user getsockopt().
2078  */
2079 static int
2080 ip_getmoptions(sopt, imo)
2081 	struct sockopt *sopt;
2082 	register struct ip_moptions *imo;
2083 {
2084 	struct in_addr addr;
2085 	struct in_ifaddr *ia;
2086 	int error, optval;
2087 	u_char coptval;
2088 
2089 	error = 0;
2090 	switch (sopt->sopt_name) {
2091 	case IP_MULTICAST_VIF:
2092 		if (imo != NULL)
2093 			optval = imo->imo_multicast_vif;
2094 		else
2095 			optval = -1;
2096 		error = sooptcopyout(sopt, &optval, sizeof optval);
2097 		break;
2098 
2099 	case IP_MULTICAST_IF:
2100 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2101 			addr.s_addr = INADDR_ANY;
2102 		else if (imo->imo_multicast_addr.s_addr) {
2103 			/* return the value user has set */
2104 			addr = imo->imo_multicast_addr;
2105 		} else {
2106 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2107 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2108 				: IA_SIN(ia)->sin_addr.s_addr;
2109 		}
2110 		error = sooptcopyout(sopt, &addr, sizeof addr);
2111 		break;
2112 
2113 	case IP_MULTICAST_TTL:
2114 		if (imo == 0)
2115 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2116 		else
2117 			optval = coptval = imo->imo_multicast_ttl;
2118 		if (sopt->sopt_valsize == 1)
2119 			error = sooptcopyout(sopt, &coptval, 1);
2120 		else
2121 			error = sooptcopyout(sopt, &optval, sizeof optval);
2122 		break;
2123 
2124 	case IP_MULTICAST_LOOP:
2125 		if (imo == 0)
2126 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2127 		else
2128 			optval = coptval = imo->imo_multicast_loop;
2129 		if (sopt->sopt_valsize == 1)
2130 			error = sooptcopyout(sopt, &coptval, 1);
2131 		else
2132 			error = sooptcopyout(sopt, &optval, sizeof optval);
2133 		break;
2134 
2135 	default:
2136 		error = ENOPROTOOPT;
2137 		break;
2138 	}
2139 	return (error);
2140 }
2141 
2142 /*
2143  * Discard the IP multicast options.
2144  */
2145 void
2146 ip_freemoptions(imo)
2147 	register struct ip_moptions *imo;
2148 {
2149 	register int i;
2150 
2151 	if (imo != NULL) {
2152 		for (i = 0; i < imo->imo_num_memberships; ++i)
2153 			in_delmulti(imo->imo_membership[i]);
2154 		free(imo, M_IPMOPTS);
2155 	}
2156 }
2157 
2158 /*
2159  * Routine called from ip_output() to loop back a copy of an IP multicast
2160  * packet to the input queue of a specified interface.  Note that this
2161  * calls the output routine of the loopback "driver", but with an interface
2162  * pointer that might NOT be a loopback interface -- evil, but easier than
2163  * replicating that code here.
2164  */
2165 static void
2166 ip_mloopback(ifp, m, dst, hlen)
2167 	struct ifnet *ifp;
2168 	register struct mbuf *m;
2169 	register struct sockaddr_in *dst;
2170 	int hlen;
2171 {
2172 	register struct ip *ip;
2173 	struct mbuf *copym;
2174 
2175 	copym = m_copy(m, 0, M_COPYALL);
2176 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2177 		copym = m_pullup(copym, hlen);
2178 	if (copym != NULL) {
2179 		/*
2180 		 * We don't bother to fragment if the IP length is greater
2181 		 * than the interface's MTU.  Can this possibly matter?
2182 		 */
2183 		ip = mtod(copym, struct ip *);
2184 		ip->ip_len = htons(ip->ip_len);
2185 		ip->ip_off = htons(ip->ip_off);
2186 		ip->ip_sum = 0;
2187 		ip->ip_sum = in_cksum(copym, hlen);
2188 		/*
2189 		 * NB:
2190 		 * It's not clear whether there are any lingering
2191 		 * reentrancy problems in other areas which might
2192 		 * be exposed by using ip_input directly (in
2193 		 * particular, everything which modifies the packet
2194 		 * in-place).  Yet another option is using the
2195 		 * protosw directly to deliver the looped back
2196 		 * packet.  For the moment, we'll err on the side
2197 		 * of safety by using if_simloop().
2198 		 */
2199 #if 1 /* XXX */
2200 		if (dst->sin_family != AF_INET) {
2201 			printf("ip_mloopback: bad address family %d\n",
2202 						dst->sin_family);
2203 			dst->sin_family = AF_INET;
2204 		}
2205 #endif
2206 
2207 #ifdef notdef
2208 		copym->m_pkthdr.rcvif = ifp;
2209 		ip_input(copym);
2210 #else
2211 		/* if the checksum hasn't been computed, mark it as valid */
2212 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2213 			copym->m_pkthdr.csum_flags |=
2214 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2215 			copym->m_pkthdr.csum_data = 0xffff;
2216 		}
2217 		if_simloop(ifp, copym, dst->sin_family, 0);
2218 #endif
2219 	}
2220 }
2221