xref: /dflybsd-src/sys/netinet/ip_output.c (revision c6cf4f8f1ebc9e3fe2a8c566f08adfc86122c7bf)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.27 2005/03/04 04:38:47 hsu Exp $
32  */
33 
34 #define _IP_VHL
35 
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipfilter.h"
40 #include "opt_ipsec.h"
41 #include "opt_random_ip_id.h"
42 #include "opt_mbuf_stress_test.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/in_cksum.h>
55 
56 #include <net/if.h>
57 #include <net/netisr.h>
58 #include <net/pfil.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 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
69 
70 #ifdef IPSEC
71 #include <netinet6/ipsec.h>
72 #include <netproto/key/key.h>
73 #ifdef IPSEC_DEBUG
74 #include <netproto/key/key_debug.h>
75 #else
76 #define	KEYDEBUG(lev,arg)
77 #endif
78 #endif /*IPSEC*/
79 
80 #ifdef FAST_IPSEC
81 #include <netproto/ipsec/ipsec.h>
82 #include <netproto/ipsec/xform.h>
83 #include <netproto/ipsec/key.h>
84 #endif /*FAST_IPSEC*/
85 
86 #include <net/ipfw/ip_fw.h>
87 #include <net/dummynet/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 #ifdef MBUF_STRESS_TEST
98 int mbuf_frag_size = 0;
99 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
100 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
101 #endif
102 
103 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
104 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
105 static void	ip_mloopback
106 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
107 static int	ip_getmoptions
108 	(struct sockopt *, struct ip_moptions *);
109 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
110 static int	ip_setmoptions
111 	(struct sockopt *, struct ip_moptions **);
112 
113 int	ip_optcopy(struct ip *, struct ip *);
114 
115 
116 extern	struct protosw inetsw[];
117 
118 /*
119  * IP output.  The packet in mbuf chain m contains a skeletal IP
120  * header (with len, off, ttl, proto, tos, src, dst).
121  * The mbuf chain containing the packet will be freed.
122  * The mbuf opt, if present, will not be freed.
123  */
124 int
125 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
126 	  int flags, struct ip_moptions *imo, struct inpcb *inp)
127 {
128 	struct ip *ip;
129 	struct ifnet *ifp = NULL;	/* keep compiler happy */
130 	struct mbuf *m;
131 	int hlen = sizeof(struct ip);
132 	int len, off, error = 0;
133 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
134 	struct in_ifaddr *ia = NULL;
135 	int isbroadcast, sw_csum;
136 	struct in_addr pkt_dst;
137 	struct route iproute;
138 #ifdef IPSEC
139 	struct secpolicy *sp = NULL;
140 	struct socket *so = inp ? inp->inp_socket : NULL;
141 #endif
142 #ifdef FAST_IPSEC
143 	struct m_tag *mtag;
144 	struct secpolicy *sp = NULL;
145 	struct tdb_ident *tdbi;
146 	int s;
147 #endif /* FAST_IPSEC */
148 	struct ip_fw_args args;
149 	int src_was_INADDR_ANY = 0;	/* as the name says... */
150 
151 	args.eh = NULL;
152 	args.rule = NULL;
153 	args.next_hop = NULL;
154 	args.divert_rule = 0;			/* divert cookie */
155 
156 	/* Grab info from MT_TAG mbufs prepended to the chain. */
157 	while (m0 != NULL && m0->m_type == MT_TAG) {
158 		switch(m0->_m_tag_id) {
159 		case PACKET_TAG_DUMMYNET:
160 			/*
161 			 * the packet was already tagged, so part of the
162 			 * processing was already done, and we need to go down.
163 			 * Get parameters from the header.
164 			 */
165 			args.rule = ((struct dn_pkt *)m0)->rule;
166 			opt = NULL ;
167 			ro = &((struct dn_pkt *)m0)->ro;
168 			imo = NULL ;
169 			dst = ((struct dn_pkt *)m0)->dn_dst ;
170 			ifp = ((struct dn_pkt *)m0)->ifp ;
171 			flags = ((struct dn_pkt *)m0)->flags ;
172 			break;
173 		case PACKET_TAG_DIVERT:
174 			args.divert_rule = (int)m0->m_data & 0xffff;
175 			break;
176 		case PACKET_TAG_IPFORWARD:
177 			args.next_hop = (struct sockaddr_in *)m0->m_data;
178 			break;
179 		default:
180 			printf("ip_output: unrecognised MT_TAG tag %d\n",
181 			    m0->_m_tag_id);
182 			break;
183 		}
184 		m0 = m0->m_next;
185 	}
186 	m = m0;
187 	KASSERT(m != NULL && (m->m_flags & M_PKTHDR), ("ip_output: no HDR"));
188 
189 	if (ro == NULL) {
190 		ro = &iproute;
191 		bzero(ro, sizeof *ro);
192 	}
193 
194 	if (args.rule != NULL) {	/* dummynet already saw us */
195 		ip = mtod(m, struct ip *);
196 		hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
197 		if (ro->ro_rt)
198 			ia = ifatoia(ro->ro_rt->rt_ifa);
199 		goto sendit;
200 	}
201 
202 	if (opt) {
203 		len = 0;
204 		m = ip_insertoptions(m, opt, &len);
205 		if (len != 0)
206 			hlen = len;
207 	}
208 	ip = mtod(m, struct ip *);
209 	pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
210 
211 	/*
212 	 * Fill in IP header.
213 	 */
214 	if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
215 		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
216 		ip->ip_off &= IP_DF;
217 #ifdef RANDOM_IP_ID
218 		ip->ip_id = ip_randomid();
219 #else
220 		ip->ip_id = htons(ip_id++);
221 #endif
222 		ipstat.ips_localout++;
223 	} else {
224 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
225 	}
226 
227 	dst = (struct sockaddr_in *)&ro->ro_dst;
228 	/*
229 	 * If there is a cached route,
230 	 * check that it is to the same destination
231 	 * and is still up.  If not, free it and try again.
232 	 * The address family should also be checked in case of sharing the
233 	 * cache with IPv6.
234 	 */
235 	if (ro->ro_rt &&
236 	    (!(ro->ro_rt->rt_flags & RTF_UP) ||
237 	     dst->sin_family != AF_INET ||
238 	     dst->sin_addr.s_addr != pkt_dst.s_addr)) {
239 		rtfree(ro->ro_rt);
240 		ro->ro_rt = (struct rtentry *)NULL;
241 	}
242 	if (ro->ro_rt == NULL) {
243 		bzero(dst, sizeof *dst);
244 		dst->sin_family = AF_INET;
245 		dst->sin_len = sizeof *dst;
246 		dst->sin_addr = pkt_dst;
247 	}
248 	/*
249 	 * If routing to interface only,
250 	 * short circuit routing lookup.
251 	 */
252 	if (flags & IP_ROUTETOIF) {
253 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
254 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
255 			ipstat.ips_noroute++;
256 			error = ENETUNREACH;
257 			goto bad;
258 		}
259 		ifp = ia->ia_ifp;
260 		ip->ip_ttl = 1;
261 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
262 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
263 		   imo != NULL && imo->imo_multicast_ifp != NULL) {
264 		/*
265 		 * Bypass the normal routing lookup for multicast
266 		 * packets if the interface is specified.
267 		 */
268 		ifp = imo->imo_multicast_ifp;
269 		IFP_TO_IA(ifp, ia);
270 		isbroadcast = 0;	/* fool gcc */
271 	} else {
272 		/*
273 		 * If this is the case, we probably don't want to allocate
274 		 * a protocol-cloned route since we didn't get one from the
275 		 * ULP.  This lets TCP do its thing, while not burdening
276 		 * forwarding or ICMP with the overhead of cloning a route.
277 		 * Of course, we still want to do any cloning requested by
278 		 * the link layer, as this is probably required in all cases
279 		 * for correct operation (as it is for ARP).
280 		 */
281 		if (ro->ro_rt == NULL)
282 			rtalloc_ign(ro, RTF_PRCLONING);
283 		if (ro->ro_rt == NULL) {
284 			ipstat.ips_noroute++;
285 			error = EHOSTUNREACH;
286 			goto bad;
287 		}
288 		ia = ifatoia(ro->ro_rt->rt_ifa);
289 		ifp = ro->ro_rt->rt_ifp;
290 		ro->ro_rt->rt_use++;
291 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
292 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
293 		if (ro->ro_rt->rt_flags & RTF_HOST)
294 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
295 		else
296 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
297 	}
298 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
299 		struct in_multi *inm;
300 
301 		m->m_flags |= M_MCAST;
302 		/*
303 		 * IP destination address is multicast.  Make sure "dst"
304 		 * still points to the address in "ro".  (It may have been
305 		 * changed to point to a gateway address, above.)
306 		 */
307 		dst = (struct sockaddr_in *)&ro->ro_dst;
308 		/*
309 		 * See if the caller provided any multicast options
310 		 */
311 		if (imo != NULL) {
312 			ip->ip_ttl = imo->imo_multicast_ttl;
313 			if (imo->imo_multicast_vif != -1)
314 				ip->ip_src.s_addr =
315 				    ip_mcast_src ?
316 				    ip_mcast_src(imo->imo_multicast_vif) :
317 				    INADDR_ANY;
318 		} else
319 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
320 		/*
321 		 * Confirm that the outgoing interface supports multicast.
322 		 */
323 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
324 			if (!(ifp->if_flags & IFF_MULTICAST)) {
325 				ipstat.ips_noroute++;
326 				error = ENETUNREACH;
327 				goto bad;
328 			}
329 		}
330 		/*
331 		 * If source address not specified yet, use address
332 		 * of outgoing interface.
333 		 */
334 		if (ip->ip_src.s_addr == INADDR_ANY) {
335 			/* Interface may have no addresses. */
336 			if (ia != NULL)
337 				ip->ip_src = IA_SIN(ia)->sin_addr;
338 		}
339 
340 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
341 		if (inm != NULL &&
342 		   (imo == NULL || imo->imo_multicast_loop)) {
343 			/*
344 			 * If we belong to the destination multicast group
345 			 * on the outgoing interface, and the caller did not
346 			 * forbid loopback, loop back a copy.
347 			 */
348 			ip_mloopback(ifp, m, dst, hlen);
349 		}
350 		else {
351 			/*
352 			 * If we are acting as a multicast router, perform
353 			 * multicast forwarding as if the packet had just
354 			 * arrived on the interface to which we are about
355 			 * to send.  The multicast forwarding function
356 			 * recursively calls this function, using the
357 			 * IP_FORWARDING flag to prevent infinite recursion.
358 			 *
359 			 * Multicasts that are looped back by ip_mloopback(),
360 			 * above, will be forwarded by the ip_input() routine,
361 			 * if necessary.
362 			 */
363 			if (ip_mrouter && !(flags & IP_FORWARDING)) {
364 				/*
365 				 * If rsvp daemon is not running, do not
366 				 * set ip_moptions. This ensures that the packet
367 				 * is multicast and not just sent down one link
368 				 * as prescribed by rsvpd.
369 				 */
370 				if (!rsvp_on)
371 					imo = NULL;
372 				if (ip_mforward &&
373 				    ip_mforward(ip, ifp, m, imo) != 0) {
374 					m_freem(m);
375 					goto done;
376 				}
377 			}
378 		}
379 
380 		/*
381 		 * Multicasts with a time-to-live of zero may be looped-
382 		 * back, above, but must not be transmitted on a network.
383 		 * Also, multicasts addressed to the loopback interface
384 		 * are not sent -- the above call to ip_mloopback() will
385 		 * loop back a copy if this host actually belongs to the
386 		 * destination group on the loopback interface.
387 		 */
388 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
389 			m_freem(m);
390 			goto done;
391 		}
392 
393 		goto sendit;
394 	}
395 #ifndef notdef
396 	/*
397 	 * If the source address is not specified yet, use the address
398 	 * of the outoing interface. In case, keep note we did that, so
399 	 * if the the firewall changes the next-hop causing the output
400 	 * interface to change, we can fix that.
401 	 */
402 	if (ip->ip_src.s_addr == INADDR_ANY) {
403 		/* Interface may have no addresses. */
404 		if (ia != NULL) {
405 			ip->ip_src = IA_SIN(ia)->sin_addr;
406 			src_was_INADDR_ANY = 1;
407 		}
408 	}
409 #endif /* notdef */
410 #ifdef ALTQ
411 	/*
412 	 * Disable packet drop hack.
413 	 * Packetdrop should be done by queueing.
414 	 */
415 #else /* !ALTQ */
416 	/*
417 	 * Verify that we have any chance at all of being able to queue
418 	 *      the packet or packet fragments
419 	 */
420 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
421 		ifp->if_snd.ifq_maxlen) {
422 			error = ENOBUFS;
423 			ipstat.ips_odropped++;
424 			goto bad;
425 	}
426 #endif /* !ALTQ */
427 
428 	/*
429 	 * Look for broadcast address and
430 	 * verify user is allowed to send
431 	 * such a packet.
432 	 */
433 	if (isbroadcast) {
434 		if (!(ifp->if_flags & IFF_BROADCAST)) {
435 			error = EADDRNOTAVAIL;
436 			goto bad;
437 		}
438 		if (!(flags & IP_ALLOWBROADCAST)) {
439 			error = EACCES;
440 			goto bad;
441 		}
442 		/* don't allow broadcast messages to be fragmented */
443 		if (ip->ip_len > ifp->if_mtu) {
444 			error = EMSGSIZE;
445 			goto bad;
446 		}
447 		m->m_flags |= M_BCAST;
448 	} else {
449 		m->m_flags &= ~M_BCAST;
450 	}
451 
452 sendit:
453 #ifdef IPSEC
454 	/* get SP for this packet */
455 	if (so == NULL)
456 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
457 	else
458 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
459 
460 	if (sp == NULL) {
461 		ipsecstat.out_inval++;
462 		goto bad;
463 	}
464 
465 	error = 0;
466 
467 	/* check policy */
468 	switch (sp->policy) {
469 	case IPSEC_POLICY_DISCARD:
470 		/*
471 		 * This packet is just discarded.
472 		 */
473 		ipsecstat.out_polvio++;
474 		goto bad;
475 
476 	case IPSEC_POLICY_BYPASS:
477 	case IPSEC_POLICY_NONE:
478 		/* no need to do IPsec. */
479 		goto skip_ipsec;
480 
481 	case IPSEC_POLICY_IPSEC:
482 		if (sp->req == NULL) {
483 			/* acquire a policy */
484 			error = key_spdacquire(sp);
485 			goto bad;
486 		}
487 		break;
488 
489 	case IPSEC_POLICY_ENTRUST:
490 	default:
491 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
492 	}
493     {
494 	struct ipsec_output_state state;
495 	bzero(&state, sizeof state);
496 	state.m = m;
497 	if (flags & IP_ROUTETOIF) {
498 		state.ro = &iproute;
499 		bzero(&iproute, sizeof iproute);
500 	} else
501 		state.ro = ro;
502 	state.dst = (struct sockaddr *)dst;
503 
504 	ip->ip_sum = 0;
505 
506 	/*
507 	 * XXX
508 	 * delayed checksums are not currently compatible with IPsec
509 	 */
510 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
511 		in_delayed_cksum(m);
512 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
513 	}
514 
515 	ip->ip_len = htons(ip->ip_len);
516 	ip->ip_off = htons(ip->ip_off);
517 
518 	error = ipsec4_output(&state, sp, flags);
519 
520 	m = state.m;
521 	if (flags & IP_ROUTETOIF) {
522 		/*
523 		 * if we have tunnel mode SA, we may need to ignore
524 		 * IP_ROUTETOIF.
525 		 */
526 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
527 			flags &= ~IP_ROUTETOIF;
528 			ro = state.ro;
529 		}
530 	} else
531 		ro = state.ro;
532 	dst = (struct sockaddr_in *)state.dst;
533 	if (error) {
534 		/* mbuf is already reclaimed in ipsec4_output. */
535 		m0 = NULL;
536 		switch (error) {
537 		case EHOSTUNREACH:
538 		case ENETUNREACH:
539 		case EMSGSIZE:
540 		case ENOBUFS:
541 		case ENOMEM:
542 			break;
543 		default:
544 			printf("ip4_output (ipsec): error code %d\n", error);
545 			/*fall through*/
546 		case ENOENT:
547 			/* don't show these error codes to the user */
548 			error = 0;
549 			break;
550 		}
551 		goto bad;
552 	}
553     }
554 
555 	/* be sure to update variables that are affected by ipsec4_output() */
556 	ip = mtod(m, struct ip *);
557 #ifdef _IP_VHL
558 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
559 #else
560 	hlen = ip->ip_hl << 2;
561 #endif
562 	if (ro->ro_rt == NULL) {
563 		if (!(flags & IP_ROUTETOIF)) {
564 			printf("ip_output: "
565 				"can't update route after IPsec processing\n");
566 			error = EHOSTUNREACH;	/*XXX*/
567 			goto bad;
568 		}
569 	} else {
570 		ia = ifatoia(ro->ro_rt->rt_ifa);
571 		ifp = ro->ro_rt->rt_ifp;
572 	}
573 
574 	/* make it flipped, again. */
575 	ip->ip_len = ntohs(ip->ip_len);
576 	ip->ip_off = ntohs(ip->ip_off);
577 skip_ipsec:
578 #endif /*IPSEC*/
579 #ifdef FAST_IPSEC
580 	/*
581 	 * Check the security policy (SP) for the packet and, if
582 	 * required, do IPsec-related processing.  There are two
583 	 * cases here; the first time a packet is sent through
584 	 * it will be untagged and handled by ipsec4_checkpolicy.
585 	 * If the packet is resubmitted to ip_output (e.g. after
586 	 * AH, ESP, etc. processing), there will be a tag to bypass
587 	 * the lookup and related policy checking.
588 	 */
589 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
590 	s = splnet();
591 	if (mtag != NULL) {
592 		tdbi = (struct tdb_ident *)(mtag + 1);
593 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
594 		if (sp == NULL)
595 			error = -EINVAL;	/* force silent drop */
596 		m_tag_delete(m, mtag);
597 	} else {
598 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
599 					&error, inp);
600 	}
601 	/*
602 	 * There are four return cases:
603 	 *    sp != NULL		    apply IPsec policy
604 	 *    sp == NULL, error == 0	    no IPsec handling needed
605 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
606 	 *    sp == NULL, error != 0	    discard packet, report error
607 	 */
608 	if (sp != NULL) {
609 		/* Loop detection, check if ipsec processing already done */
610 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
611 		for (mtag = m_tag_first(m); mtag != NULL;
612 		     mtag = m_tag_next(m, mtag)) {
613 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
614 				continue;
615 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
616 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
617 				continue;
618 			/*
619 			 * Check if policy has an SA associated with it.
620 			 * This can happen when an SP has yet to acquire
621 			 * an SA; e.g. on first reference.  If it occurs,
622 			 * then we let ipsec4_process_packet do its thing.
623 			 */
624 			if (sp->req->sav == NULL)
625 				break;
626 			tdbi = (struct tdb_ident *)(mtag + 1);
627 			if (tdbi->spi == sp->req->sav->spi &&
628 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
629 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
630 				 sizeof(union sockaddr_union)) == 0) {
631 				/*
632 				 * No IPsec processing is needed, free
633 				 * reference to SP.
634 				 *
635 				 * NB: null pointer to avoid free at
636 				 *     done: below.
637 				 */
638 				KEY_FREESP(&sp), sp = NULL;
639 				splx(s);
640 				goto spd_done;
641 			}
642 		}
643 
644 		/*
645 		 * Do delayed checksums now because we send before
646 		 * this is done in the normal processing path.
647 		 */
648 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
649 			in_delayed_cksum(m);
650 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
651 		}
652 
653 		ip->ip_len = htons(ip->ip_len);
654 		ip->ip_off = htons(ip->ip_off);
655 
656 		/* NB: callee frees mbuf */
657 		error = ipsec4_process_packet(m, sp->req, flags, 0);
658 		/*
659 		 * Preserve KAME behaviour: ENOENT can be returned
660 		 * when an SA acquire is in progress.  Don't propagate
661 		 * this to user-level; it confuses applications.
662 		 *
663 		 * XXX this will go away when the SADB is redone.
664 		 */
665 		if (error == ENOENT)
666 			error = 0;
667 		splx(s);
668 		goto done;
669 	} else {
670 		splx(s);
671 
672 		if (error != 0) {
673 			/*
674 			 * Hack: -EINVAL is used to signal that a packet
675 			 * should be silently discarded.  This is typically
676 			 * because we asked key management for an SA and
677 			 * it was delayed (e.g. kicked up to IKE).
678 			 */
679 			if (error == -EINVAL)
680 				error = 0;
681 			goto bad;
682 		} else {
683 			/* No IPsec processing for this packet. */
684 		}
685 #ifdef notyet
686 		/*
687 		 * If deferred crypto processing is needed, check that
688 		 * the interface supports it.
689 		 */
690 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
691 		if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
692 			/* notify IPsec to do its own crypto */
693 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
694 			error = EHOSTUNREACH;
695 			goto bad;
696 		}
697 #endif
698 	}
699 spd_done:
700 #endif /* FAST_IPSEC */
701 	/*
702 	 * IpHack's section.
703 	 * - Xlate: translate packet's addr/port (NAT).
704 	 * - Firewall: deny/allow/etc.
705 	 * - Wrap: fake packet's addr/port <unimpl.>
706 	 * - Encapsulate: put it in another IP and send out. <unimp.>
707 	 */
708 
709 	/*
710 	 * Run through list of hooks for output packets.
711 	 */
712 	if (pfil_has_hooks(&inet_pfil_hook)) {
713 		error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
714 		if (error != 0 || m == NULL)
715 			goto done;
716 		ip = mtod(m, struct ip *);
717 	}
718 
719 	/*
720 	 * Check with the firewall...
721 	 * but not if we are already being fwd'd from a firewall.
722 	 */
723 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
724 		struct sockaddr_in *old = dst;
725 
726 		args.m = m;
727 		args.next_hop = dst;
728 		args.oif = ifp;
729 		off = ip_fw_chk_ptr(&args);
730 		m = args.m;
731 		dst = args.next_hop;
732 
733 		/*
734 		 * On return we must do the following:
735 		 * m == NULL	-> drop the pkt (old interface, deprecated)
736 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
737 		 * 1<=off<= 0xffff		-> DIVERT
738 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
739 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
740 		 * dst != old			-> IPFIREWALL_FORWARD
741 		 * off==0, dst==old		-> accept
742 		 * If some of the above modules are not compiled in, then
743 		 * we should't have to check the corresponding condition
744 		 * (because the ipfw control socket should not accept
745 		 * unsupported rules), but better play safe and drop
746 		 * packets in case of doubt.
747 		 */
748 		if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
749 			if (m)
750 				m_freem(m);
751 			error = EACCES;
752 			goto done;
753 		}
754 		ip = mtod(m, struct ip *);
755 		if (off == 0 && dst == old)		/* common case */
756 			goto pass;
757 		if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG)) {
758 			/*
759 			 * pass the pkt to dummynet. Need to include
760 			 * pipe number, m, ifp, ro, dst because these are
761 			 * not recomputed in the next pass.
762 			 * All other parameters have been already used and
763 			 * so they are not needed anymore.
764 			 * XXX note: if the ifp or ro entry are deleted
765 			 * while a pkt is in dummynet, we are in trouble!
766 			 */
767 			args.ro = ro;
768 			args.dst = dst;
769 			args.flags = flags;
770 
771 			error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
772 				&args);
773 			goto done;
774 		}
775 #ifdef IPDIVERT
776 		if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
777 			struct mbuf *clone = NULL;
778 
779 			/* Clone packet if we're doing a 'tee' */
780 			if ((off & IP_FW_PORT_TEE_FLAG))
781 				clone = m_dup(m, MB_DONTWAIT);
782 
783 			/*
784 			 * XXX
785 			 * delayed checksums are not currently compatible
786 			 * with divert sockets.
787 			 */
788 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
789 				in_delayed_cksum(m);
790 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
791 			}
792 
793 			/* Restore packet header fields to original values */
794 			ip->ip_len = htons(ip->ip_len);
795 			ip->ip_off = htons(ip->ip_off);
796 
797 			/* Deliver packet to divert input routine */
798 			divert_packet(m, 0, off & 0xffff, args.divert_rule);
799 
800 			/* If 'tee', continue with original packet */
801 			if (clone != NULL) {
802 				m = clone;
803 				ip = mtod(m, struct ip *);
804 				goto pass;
805 			}
806 			goto done;
807 		}
808 #endif
809 
810 		/* IPFIREWALL_FORWARD */
811 		/*
812 		 * Check dst to make sure it is directly reachable on the
813 		 * interface we previously thought it was.
814 		 * If it isn't (which may be likely in some situations) we have
815 		 * to re-route it (ie, find a route for the next-hop and the
816 		 * associated interface) and set them here. This is nested
817 		 * forwarding which in most cases is undesirable, except where
818 		 * such control is nigh impossible. So we do it here.
819 		 * And I'm babbling.
820 		 */
821 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
822 #if 0
823 			/*
824 			 * XXX To improve readability, this block should be
825 			 * changed into a function call as below:
826 			 */
827 			error = ip_ipforward(&m, &dst, &ifp);
828 			if (error)
829 				goto bad;
830 			if (m == NULL) /* ip_input consumed the mbuf */
831 				goto done;
832 #else
833 			struct in_ifaddr *ia;
834 
835 			/*
836 			 * XXX sro_fwd below is static, and a pointer
837 			 * to it gets passed to routines downstream.
838 			 * This could have surprisingly bad results in
839 			 * practice, because its content is overwritten
840 			 * by subsequent packets.
841 			 */
842 			/* There must be a better way to do this next line... */
843 			static struct route sro_fwd;
844 			struct route *ro_fwd = &sro_fwd;
845 
846 #if 0
847 			print_ip("IPFIREWALL_FORWARD: New dst ip: ",
848 			    dst->sin_addr, "\n");
849 #endif
850 
851 			/*
852 			 * We need to figure out if we have been forwarded
853 			 * to a local socket. If so, then we should somehow
854 			 * "loop back" to ip_input, and get directed to the
855 			 * PCB as if we had received this packet. This is
856 			 * because it may be dificult to identify the packets
857 			 * you want to forward until they are being output
858 			 * and have selected an interface. (e.g. locally
859 			 * initiated packets) If we used the loopback inteface,
860 			 * we would not be able to control what happens
861 			 * as the packet runs through ip_input() as
862 			 * it is done through a ISR.
863 			 */
864 			LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
865 				     ia_hash) {
866 				/*
867 				 * If the addr to forward to is one
868 				 * of ours, we pretend to
869 				 * be the destination for this packet.
870 				 */
871 				if (IA_SIN(ia)->sin_addr.s_addr ==
872 						 dst->sin_addr.s_addr)
873 					break;
874 			}
875 			if (ia != NULL) {    /* tell ip_input "dont filter" */
876 				struct m_hdr tag;
877 
878 				tag.mh_type = MT_TAG;
879 				tag.mh_flags = PACKET_TAG_IPFORWARD;
880 				tag.mh_data = (caddr_t)args.next_hop;
881 				tag.mh_next = m;
882 
883 				if (m->m_pkthdr.rcvif == NULL)
884 					m->m_pkthdr.rcvif = ifunit("lo0");
885 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
886 					m->m_pkthdr.csum_flags |=
887 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
888 					m0->m_pkthdr.csum_data = 0xffff;
889 				}
890 				m->m_pkthdr.csum_flags |=
891 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
892 				ip->ip_len = htons(ip->ip_len);
893 				ip->ip_off = htons(ip->ip_off);
894 				ip_input((struct mbuf *)&tag);
895 				goto done;
896 			}
897 			/* Some of the logic for this was nicked from above.
898 			 *
899 			 * This rewrites the cached route in a local PCB.
900 			 * Is this what we want to do?
901 			 */
902 			bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
903 			ro_fwd->ro_rt = NULL;
904 
905 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
906 			if (ro_fwd->ro_rt == NULL) {
907 				ipstat.ips_noroute++;
908 				error = EHOSTUNREACH;
909 				goto bad;
910 			}
911 
912 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
913 			ifp = ro_fwd->ro_rt->rt_ifp;
914 			ro_fwd->ro_rt->rt_use++;
915 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
916 				dst = (struct sockaddr_in *)
917 				    ro_fwd->ro_rt->rt_gateway;
918 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
919 				isbroadcast =
920 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
921 			else
922 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
923 			if (ro->ro_rt != NULL)
924 				rtfree(ro->ro_rt);
925 			ro->ro_rt = ro_fwd->ro_rt;
926 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
927 
928 #endif	/* ... block to be put into a function */
929 			/*
930 			 * If we added a default src ip earlier,
931 			 * which would have been gotten from the-then
932 			 * interface, do it again, from the new one.
933 			 */
934 			if (src_was_INADDR_ANY)
935 				ip->ip_src = IA_SIN(ia)->sin_addr;
936 			goto pass ;
937 		}
938 
939 		/*
940 		 * if we get here, none of the above matches, and
941 		 * we have to drop the pkt
942 		 */
943 		m_freem(m);
944 		error = EACCES; /* not sure this is the right error msg */
945 		goto done;
946 	}
947 
948 pass:
949 	/* 127/8 must not appear on wire - RFC1122. */
950 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
951 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
952 		if (!(ifp->if_flags & IFF_LOOPBACK)) {
953 			ipstat.ips_badaddr++;
954 			error = EADDRNOTAVAIL;
955 			goto bad;
956 		}
957 	}
958 
959 	m->m_pkthdr.csum_flags |= CSUM_IP;
960 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
961 	if (sw_csum & CSUM_DELAY_DATA) {
962 		in_delayed_cksum(m);
963 		sw_csum &= ~CSUM_DELAY_DATA;
964 	}
965 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
966 
967 	/*
968 	 * If small enough for interface, or the interface will take
969 	 * care of the fragmentation for us, can just send directly.
970 	 */
971 	if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
972 	    !(ip->ip_off & IP_DF))) {
973 		ip->ip_len = htons(ip->ip_len);
974 		ip->ip_off = htons(ip->ip_off);
975 		ip->ip_sum = 0;
976 		if (sw_csum & CSUM_DELAY_IP) {
977 			if (ip->ip_vhl == IP_VHL_BORING) {
978 				ip->ip_sum = in_cksum_hdr(ip);
979 			} else {
980 				ip->ip_sum = in_cksum(m, hlen);
981 			}
982 		}
983 
984 		/* Record statistics for this interface address. */
985 		if (!(flags & IP_FORWARDING) && ia) {
986 			ia->ia_ifa.if_opackets++;
987 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
988 		}
989 
990 #ifdef IPSEC
991 		/* clean ipsec history once it goes out of the node */
992 		ipsec_delaux(m);
993 #endif
994 
995 #ifdef MBUF_STRESS_TEST
996 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
997 			struct mbuf *m1, *m2;
998 			int length, tmp;
999 
1000 			tmp = length = m->m_pkthdr.len;
1001 
1002 			while ((length -= mbuf_frag_size) >= 1) {
1003 				m1 = m_split(m, length, MB_DONTWAIT);
1004 				if (m1 == NULL)
1005 					break;
1006 				m1->m_flags &= ~M_PKTHDR;
1007 				m2 = m;
1008 				while (m2->m_next != NULL)
1009 					m2 = m2->m_next;
1010 				m2->m_next = m1;
1011 			}
1012 			m->m_pkthdr.len = tmp;
1013 		}
1014 #endif
1015 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1016 					  ro->ro_rt);
1017 		goto done;
1018 	}
1019 
1020 	if (ip->ip_off & IP_DF) {
1021 		error = EMSGSIZE;
1022 		/*
1023 		 * This case can happen if the user changed the MTU
1024 		 * of an interface after enabling IP on it.  Because
1025 		 * most netifs don't keep track of routes pointing to
1026 		 * them, there is no way for one to update all its
1027 		 * routes when the MTU is changed.
1028 		 */
1029 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1030 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1031 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1032 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1033 		}
1034 		ipstat.ips_cantfrag++;
1035 		goto bad;
1036 	}
1037 
1038 	/*
1039 	 * Too large for interface; fragment if possible. If successful,
1040 	 * on return, m will point to a list of packets to be sent.
1041 	 */
1042 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1043 	if (error)
1044 		goto bad;
1045 	for (; m; m = m0) {
1046 		m0 = m->m_nextpkt;
1047 		m->m_nextpkt = NULL;
1048 #ifdef IPSEC
1049 		/* clean ipsec history once it goes out of the node */
1050 		ipsec_delaux(m);
1051 #endif
1052 		if (error == 0) {
1053 			/* Record statistics for this interface address. */
1054 			if (ia != NULL) {
1055 				ia->ia_ifa.if_opackets++;
1056 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1057 			}
1058 			error = (*ifp->if_output)(ifp, m,
1059 						  (struct sockaddr *)dst,
1060 						  ro->ro_rt);
1061 		} else
1062 			m_freem(m);
1063 	}
1064 
1065 	if (error == 0)
1066 		ipstat.ips_fragmented++;
1067 
1068 done:
1069 	if (ro == &iproute && ro->ro_rt != NULL) {
1070 		RTFREE(ro->ro_rt);
1071 		ro->ro_rt = NULL;
1072 	}
1073 #ifdef IPSEC
1074 	if (sp != NULL) {
1075 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1076 			printf("DP ip_output call free SP:%p\n", sp));
1077 		key_freesp(sp);
1078 	}
1079 #endif
1080 #ifdef FAST_IPSEC
1081 	if (sp != NULL)
1082 		KEY_FREESP(&sp);
1083 #endif
1084 	return (error);
1085 bad:
1086 	m_freem(m);
1087 	goto done;
1088 }
1089 
1090 /*
1091  * Create a chain of fragments which fit the given mtu. m_frag points to the
1092  * mbuf to be fragmented; on return it points to the chain with the fragments.
1093  * Return 0 if no error. If error, m_frag may contain a partially built
1094  * chain of fragments that should be freed by the caller.
1095  *
1096  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1097  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1098  */
1099 int
1100 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1101 	    u_long if_hwassist_flags, int sw_csum)
1102 {
1103 	int error = 0;
1104 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1105 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1106 	int off;
1107 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1108 	int firstlen;
1109 	struct mbuf **mnext;
1110 	int nfrags;
1111 
1112 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1113 		ipstat.ips_cantfrag++;
1114 		return EMSGSIZE;
1115 	}
1116 
1117 	/*
1118 	 * Must be able to put at least 8 bytes per fragment.
1119 	 */
1120 	if (len < 8)
1121 		return EMSGSIZE;
1122 
1123 	/*
1124 	 * If the interface will not calculate checksums on
1125 	 * fragmented packets, then do it here.
1126 	 */
1127 	if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1128 	    !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1129 		in_delayed_cksum(m0);
1130 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1131 	}
1132 
1133 	if (len > PAGE_SIZE) {
1134 		/*
1135 		 * Fragment large datagrams such that each segment
1136 		 * contains a multiple of PAGE_SIZE amount of data,
1137 		 * plus headers. This enables a receiver to perform
1138 		 * page-flipping zero-copy optimizations.
1139 		 *
1140 		 * XXX When does this help given that sender and receiver
1141 		 * could have different page sizes, and also mtu could
1142 		 * be less than the receiver's page size ?
1143 		 */
1144 		int newlen;
1145 		struct mbuf *m;
1146 
1147 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1148 			off += m->m_len;
1149 
1150 		/*
1151 		 * firstlen (off - hlen) must be aligned on an
1152 		 * 8-byte boundary
1153 		 */
1154 		if (off < hlen)
1155 			goto smart_frag_failure;
1156 		off = ((off - hlen) & ~7) + hlen;
1157 		newlen = (~PAGE_MASK) & mtu;
1158 		if ((newlen + sizeof(struct ip)) > mtu) {
1159 			/* we failed, go back the default */
1160 smart_frag_failure:
1161 			newlen = len;
1162 			off = hlen + len;
1163 		}
1164 		len = newlen;
1165 
1166 	} else {
1167 		off = hlen + len;
1168 	}
1169 
1170 	firstlen = off - hlen;
1171 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1172 
1173 	/*
1174 	 * Loop through length of segment after first fragment,
1175 	 * make new header and copy data of each part and link onto chain.
1176 	 * Here, m0 is the original packet, m is the fragment being created.
1177 	 * The fragments are linked off the m_nextpkt of the original
1178 	 * packet, which after processing serves as the first fragment.
1179 	 */
1180 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1181 		struct ip *mhip;	/* ip header on the fragment */
1182 		struct mbuf *m;
1183 		int mhlen = sizeof(struct ip);
1184 
1185 		MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1186 		if (m == NULL) {
1187 			error = ENOBUFS;
1188 			ipstat.ips_odropped++;
1189 			goto done;
1190 		}
1191 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1192 		/*
1193 		 * In the first mbuf, leave room for the link header, then
1194 		 * copy the original IP header including options. The payload
1195 		 * goes into an additional mbuf chain returned by m_copy().
1196 		 */
1197 		m->m_data += max_linkhdr;
1198 		mhip = mtod(m, struct ip *);
1199 		*mhip = *ip;
1200 		if (hlen > sizeof(struct ip)) {
1201 			mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1202 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1203 		}
1204 		m->m_len = mhlen;
1205 		/* XXX do we need to add ip->ip_off below ? */
1206 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1207 		if (off + len >= ip->ip_len) {	/* last fragment */
1208 			len = ip->ip_len - off;
1209 			m->m_flags |= M_LASTFRAG;
1210 		} else
1211 			mhip->ip_off |= IP_MF;
1212 		mhip->ip_len = htons((u_short)(len + mhlen));
1213 		m->m_next = m_copy(m0, off, len);
1214 		if (m->m_next == NULL) {		/* copy failed */
1215 			m_free(m);
1216 			error = ENOBUFS;	/* ??? */
1217 			ipstat.ips_odropped++;
1218 			goto done;
1219 		}
1220 		m->m_pkthdr.len = mhlen + len;
1221 		m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1222 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1223 		mhip->ip_off = htons(mhip->ip_off);
1224 		mhip->ip_sum = 0;
1225 		if (sw_csum & CSUM_DELAY_IP)
1226 			mhip->ip_sum = in_cksum(m, mhlen);
1227 		*mnext = m;
1228 		mnext = &m->m_nextpkt;
1229 	}
1230 	ipstat.ips_ofragments += nfrags;
1231 
1232 	/* set first marker for fragment chain */
1233 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1234 	m0->m_pkthdr.csum_data = nfrags;
1235 
1236 	/*
1237 	 * Update first fragment by trimming what's been copied out
1238 	 * and updating header.
1239 	 */
1240 	m_adj(m0, hlen + firstlen - ip->ip_len);
1241 	m0->m_pkthdr.len = hlen + firstlen;
1242 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1243 	ip->ip_off |= IP_MF;
1244 	ip->ip_off = htons(ip->ip_off);
1245 	ip->ip_sum = 0;
1246 	if (sw_csum & CSUM_DELAY_IP)
1247 		ip->ip_sum = in_cksum(m0, hlen);
1248 
1249 done:
1250 	*m_frag = m0;
1251 	return error;
1252 }
1253 
1254 void
1255 in_delayed_cksum(struct mbuf *m)
1256 {
1257 	struct ip *ip;
1258 	u_short csum, offset;
1259 
1260 	ip = mtod(m, struct ip *);
1261 	offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1262 	csum = in_cksum_skip(m, ip->ip_len, offset);
1263 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1264 		csum = 0xffff;
1265 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1266 
1267 	if (offset + sizeof(u_short) > m->m_len) {
1268 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1269 		    m->m_len, offset, ip->ip_p);
1270 		/*
1271 		 * XXX
1272 		 * this shouldn't happen, but if it does, the
1273 		 * correct behavior may be to insert the checksum
1274 		 * in the existing chain instead of rearranging it.
1275 		 */
1276 		m = m_pullup(m, offset + sizeof(u_short));
1277 	}
1278 	*(u_short *)(m->m_data + offset) = csum;
1279 }
1280 
1281 /*
1282  * Insert IP options into preformed packet.
1283  * Adjust IP destination as required for IP source routing,
1284  * as indicated by a non-zero in_addr at the start of the options.
1285  *
1286  * XXX This routine assumes that the packet has no options in place.
1287  */
1288 static struct mbuf *
1289 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1290 {
1291 	struct ipoption *p = mtod(opt, struct ipoption *);
1292 	struct mbuf *n;
1293 	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, MB_DONTWAIT, MT_HEADER);
1305 		if (n == NULL) {
1306 			*phlen = 0;
1307 			return (m);
1308 		}
1309 		n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1310 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1311 		m->m_len -= sizeof(struct ip);
1312 		m->m_data += sizeof(struct ip);
1313 		n->m_next = m;
1314 		m = n;
1315 		m->m_len = optlen + sizeof(struct ip);
1316 		m->m_data += max_linkhdr;
1317 		memcpy(mtod(m, void *), ip, sizeof(struct ip));
1318 	} else {
1319 		m->m_data -= optlen;
1320 		m->m_len += optlen;
1321 		m->m_pkthdr.len += optlen;
1322 		ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1323 	}
1324 	ip = mtod(m, struct ip *);
1325 	bcopy(p->ipopt_list, ip + 1, optlen);
1326 	*phlen = sizeof(struct ip) + optlen;
1327 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1328 	ip->ip_len += optlen;
1329 	return (m);
1330 }
1331 
1332 /*
1333  * Copy options from ip to jp,
1334  * omitting those not copied during fragmentation.
1335  */
1336 int
1337 ip_optcopy(struct ip *ip, struct ip *jp)
1338 {
1339 	u_char *cp, *dp;
1340 	int opt, optlen, cnt;
1341 
1342 	cp = (u_char *)(ip + 1);
1343 	dp = (u_char *)(jp + 1);
1344 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1345 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1346 		opt = cp[0];
1347 		if (opt == IPOPT_EOL)
1348 			break;
1349 		if (opt == IPOPT_NOP) {
1350 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1351 			*dp++ = IPOPT_NOP;
1352 			optlen = 1;
1353 			continue;
1354 		}
1355 
1356 		KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1357 		    ("ip_optcopy: malformed ipv4 option"));
1358 		optlen = cp[IPOPT_OLEN];
1359 		KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1360 		    ("ip_optcopy: malformed ipv4 option"));
1361 
1362 		/* bogus lengths should have been caught by ip_dooptions */
1363 		if (optlen > cnt)
1364 			optlen = cnt;
1365 		if (IPOPT_COPIED(opt)) {
1366 			bcopy(cp, dp, optlen);
1367 			dp += optlen;
1368 		}
1369 	}
1370 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1371 		*dp++ = IPOPT_EOL;
1372 	return (optlen);
1373 }
1374 
1375 /*
1376  * IP socket option processing.
1377  */
1378 int
1379 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1380 {
1381 	struct	inpcb *inp = so->so_pcb;
1382 	int	error, optval;
1383 
1384 	error = optval = 0;
1385 	if (sopt->sopt_level != IPPROTO_IP) {
1386 		return (EINVAL);
1387 	}
1388 
1389 	switch (sopt->sopt_dir) {
1390 	case SOPT_SET:
1391 		switch (sopt->sopt_name) {
1392 		case IP_OPTIONS:
1393 #ifdef notyet
1394 		case IP_RETOPTS:
1395 #endif
1396 		{
1397 			struct mbuf *m;
1398 			if (sopt->sopt_valsize > MLEN) {
1399 				error = EMSGSIZE;
1400 				break;
1401 			}
1402 			MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1403 			if (m == NULL) {
1404 				error = ENOBUFS;
1405 				break;
1406 			}
1407 			m->m_len = sopt->sopt_valsize;
1408 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1409 					    m->m_len);
1410 
1411 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1412 					   m));
1413 		}
1414 
1415 		case IP_TOS:
1416 		case IP_TTL:
1417 		case IP_RECVOPTS:
1418 		case IP_RECVRETOPTS:
1419 		case IP_RECVDSTADDR:
1420 		case IP_RECVIF:
1421 		case IP_FAITH:
1422 			error = sooptcopyin(sopt, &optval, sizeof optval,
1423 					    sizeof optval);
1424 			if (error)
1425 				break;
1426 
1427 			switch (sopt->sopt_name) {
1428 			case IP_TOS:
1429 				inp->inp_ip_tos = optval;
1430 				break;
1431 
1432 			case IP_TTL:
1433 				inp->inp_ip_ttl = optval;
1434 				break;
1435 #define	OPTSET(bit) \
1436 	if (optval) \
1437 		inp->inp_flags |= bit; \
1438 	else \
1439 		inp->inp_flags &= ~bit;
1440 
1441 			case IP_RECVOPTS:
1442 				OPTSET(INP_RECVOPTS);
1443 				break;
1444 
1445 			case IP_RECVRETOPTS:
1446 				OPTSET(INP_RECVRETOPTS);
1447 				break;
1448 
1449 			case IP_RECVDSTADDR:
1450 				OPTSET(INP_RECVDSTADDR);
1451 				break;
1452 
1453 			case IP_RECVIF:
1454 				OPTSET(INP_RECVIF);
1455 				break;
1456 
1457 			case IP_FAITH:
1458 				OPTSET(INP_FAITH);
1459 				break;
1460 			}
1461 			break;
1462 #undef OPTSET
1463 
1464 		case IP_MULTICAST_IF:
1465 		case IP_MULTICAST_VIF:
1466 		case IP_MULTICAST_TTL:
1467 		case IP_MULTICAST_LOOP:
1468 		case IP_ADD_MEMBERSHIP:
1469 		case IP_DROP_MEMBERSHIP:
1470 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1471 			break;
1472 
1473 		case IP_PORTRANGE:
1474 			error = sooptcopyin(sopt, &optval, sizeof optval,
1475 					    sizeof optval);
1476 			if (error)
1477 				break;
1478 
1479 			switch (optval) {
1480 			case IP_PORTRANGE_DEFAULT:
1481 				inp->inp_flags &= ~(INP_LOWPORT);
1482 				inp->inp_flags &= ~(INP_HIGHPORT);
1483 				break;
1484 
1485 			case IP_PORTRANGE_HIGH:
1486 				inp->inp_flags &= ~(INP_LOWPORT);
1487 				inp->inp_flags |= INP_HIGHPORT;
1488 				break;
1489 
1490 			case IP_PORTRANGE_LOW:
1491 				inp->inp_flags &= ~(INP_HIGHPORT);
1492 				inp->inp_flags |= INP_LOWPORT;
1493 				break;
1494 
1495 			default:
1496 				error = EINVAL;
1497 				break;
1498 			}
1499 			break;
1500 
1501 #if defined(IPSEC) || defined(FAST_IPSEC)
1502 		case IP_IPSEC_POLICY:
1503 		{
1504 			caddr_t req;
1505 			size_t len = 0;
1506 			int priv;
1507 			struct mbuf *m;
1508 			int optname;
1509 
1510 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1511 				break;
1512 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1513 				break;
1514 			priv = (sopt->sopt_td != NULL &&
1515 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1516 			req = mtod(m, caddr_t);
1517 			len = m->m_len;
1518 			optname = sopt->sopt_name;
1519 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1520 			m_freem(m);
1521 			break;
1522 		}
1523 #endif /*IPSEC*/
1524 
1525 		default:
1526 			error = ENOPROTOOPT;
1527 			break;
1528 		}
1529 		break;
1530 
1531 	case SOPT_GET:
1532 		switch (sopt->sopt_name) {
1533 		case IP_OPTIONS:
1534 		case IP_RETOPTS:
1535 			if (inp->inp_options)
1536 				error = sooptcopyout(sopt,
1537 						     mtod(inp->inp_options,
1538 							  char *),
1539 						     inp->inp_options->m_len);
1540 			else
1541 				sopt->sopt_valsize = 0;
1542 			break;
1543 
1544 		case IP_TOS:
1545 		case IP_TTL:
1546 		case IP_RECVOPTS:
1547 		case IP_RECVRETOPTS:
1548 		case IP_RECVDSTADDR:
1549 		case IP_RECVIF:
1550 		case IP_PORTRANGE:
1551 		case IP_FAITH:
1552 			switch (sopt->sopt_name) {
1553 
1554 			case IP_TOS:
1555 				optval = inp->inp_ip_tos;
1556 				break;
1557 
1558 			case IP_TTL:
1559 				optval = inp->inp_ip_ttl;
1560 				break;
1561 
1562 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1563 
1564 			case IP_RECVOPTS:
1565 				optval = OPTBIT(INP_RECVOPTS);
1566 				break;
1567 
1568 			case IP_RECVRETOPTS:
1569 				optval = OPTBIT(INP_RECVRETOPTS);
1570 				break;
1571 
1572 			case IP_RECVDSTADDR:
1573 				optval = OPTBIT(INP_RECVDSTADDR);
1574 				break;
1575 
1576 			case IP_RECVIF:
1577 				optval = OPTBIT(INP_RECVIF);
1578 				break;
1579 
1580 			case IP_PORTRANGE:
1581 				if (inp->inp_flags & INP_HIGHPORT)
1582 					optval = IP_PORTRANGE_HIGH;
1583 				else if (inp->inp_flags & INP_LOWPORT)
1584 					optval = IP_PORTRANGE_LOW;
1585 				else
1586 					optval = 0;
1587 				break;
1588 
1589 			case IP_FAITH:
1590 				optval = OPTBIT(INP_FAITH);
1591 				break;
1592 			}
1593 			error = sooptcopyout(sopt, &optval, sizeof optval);
1594 			break;
1595 
1596 		case IP_MULTICAST_IF:
1597 		case IP_MULTICAST_VIF:
1598 		case IP_MULTICAST_TTL:
1599 		case IP_MULTICAST_LOOP:
1600 		case IP_ADD_MEMBERSHIP:
1601 		case IP_DROP_MEMBERSHIP:
1602 			error = ip_getmoptions(sopt, inp->inp_moptions);
1603 			break;
1604 
1605 #if defined(IPSEC) || defined(FAST_IPSEC)
1606 		case IP_IPSEC_POLICY:
1607 		{
1608 			struct mbuf *m = NULL;
1609 			caddr_t req = NULL;
1610 			size_t len = 0;
1611 
1612 			if (m != NULL) {
1613 				req = mtod(m, caddr_t);
1614 				len = m->m_len;
1615 			}
1616 			error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1617 			if (error == 0)
1618 				error = soopt_mcopyout(sopt, m); /* XXX */
1619 			if (error == 0)
1620 				m_freem(m);
1621 			break;
1622 		}
1623 #endif /*IPSEC*/
1624 
1625 		default:
1626 			error = ENOPROTOOPT;
1627 			break;
1628 		}
1629 		break;
1630 	}
1631 	return (error);
1632 }
1633 
1634 /*
1635  * Set up IP options in pcb for insertion in output packets.
1636  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1637  * with destination address if source routed.
1638  */
1639 static int
1640 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1641 {
1642 	int cnt, optlen;
1643 	u_char *cp;
1644 	u_char opt;
1645 
1646 	/* turn off any old options */
1647 	if (*pcbopt)
1648 		m_free(*pcbopt);
1649 	*pcbopt = 0;
1650 	if (m == NULL || m->m_len == 0) {
1651 		/*
1652 		 * Only turning off any previous options.
1653 		 */
1654 		if (m != NULL)
1655 			m_free(m);
1656 		return (0);
1657 	}
1658 
1659 	if (m->m_len % sizeof(int32_t))
1660 		goto bad;
1661 	/*
1662 	 * IP first-hop destination address will be stored before
1663 	 * actual options; move other options back
1664 	 * and clear it when none present.
1665 	 */
1666 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1667 		goto bad;
1668 	cnt = m->m_len;
1669 	m->m_len += sizeof(struct in_addr);
1670 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1671 	ovbcopy(mtod(m, caddr_t), cp, cnt);
1672 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1673 
1674 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1675 		opt = cp[IPOPT_OPTVAL];
1676 		if (opt == IPOPT_EOL)
1677 			break;
1678 		if (opt == IPOPT_NOP)
1679 			optlen = 1;
1680 		else {
1681 			if (cnt < IPOPT_OLEN + sizeof *cp)
1682 				goto bad;
1683 			optlen = cp[IPOPT_OLEN];
1684 			if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1685 				goto bad;
1686 		}
1687 		switch (opt) {
1688 
1689 		default:
1690 			break;
1691 
1692 		case IPOPT_LSRR:
1693 		case IPOPT_SSRR:
1694 			/*
1695 			 * user process specifies route as:
1696 			 *	->A->B->C->D
1697 			 * D must be our final destination (but we can't
1698 			 * check that since we may not have connected yet).
1699 			 * A is first hop destination, which doesn't appear in
1700 			 * actual IP option, but is stored before the options.
1701 			 */
1702 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1703 				goto bad;
1704 			m->m_len -= sizeof(struct in_addr);
1705 			cnt -= sizeof(struct in_addr);
1706 			optlen -= sizeof(struct in_addr);
1707 			cp[IPOPT_OLEN] = optlen;
1708 			/*
1709 			 * Move first hop before start of options.
1710 			 */
1711 			bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1712 			      sizeof(struct in_addr));
1713 			/*
1714 			 * Then copy rest of options back
1715 			 * to close up the deleted entry.
1716 			 */
1717 			ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1718 				&cp[IPOPT_OFFSET+1],
1719 				cnt - (IPOPT_MINOFF - 1));
1720 			break;
1721 		}
1722 	}
1723 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1724 		goto bad;
1725 	*pcbopt = m;
1726 	return (0);
1727 
1728 bad:
1729 	m_free(m);
1730 	return (EINVAL);
1731 }
1732 
1733 /*
1734  * XXX
1735  * The whole multicast option thing needs to be re-thought.
1736  * Several of these options are equally applicable to non-multicast
1737  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1738  * standard option (IP_TTL).
1739  */
1740 
1741 /*
1742  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1743  */
1744 static struct ifnet *
1745 ip_multicast_if(struct in_addr *a, int *ifindexp)
1746 {
1747 	int ifindex;
1748 	struct ifnet *ifp;
1749 
1750 	if (ifindexp)
1751 		*ifindexp = 0;
1752 	if (ntohl(a->s_addr) >> 24 == 0) {
1753 		ifindex = ntohl(a->s_addr) & 0xffffff;
1754 		if (ifindex < 0 || if_index < ifindex)
1755 			return NULL;
1756 		ifp = ifindex2ifnet[ifindex];
1757 		if (ifindexp)
1758 			*ifindexp = ifindex;
1759 	} else {
1760 		INADDR_TO_IFP(*a, ifp);
1761 	}
1762 	return ifp;
1763 }
1764 
1765 /*
1766  * Set the IP multicast options in response to user setsockopt().
1767  */
1768 static int
1769 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1770 {
1771 	int error = 0;
1772 	int i;
1773 	struct in_addr addr;
1774 	struct ip_mreq mreq;
1775 	struct ifnet *ifp;
1776 	struct ip_moptions *imo = *imop;
1777 	int ifindex;
1778 	int s;
1779 
1780 	if (imo == NULL) {
1781 		/*
1782 		 * No multicast option buffer attached to the pcb;
1783 		 * allocate one and initialize to default values.
1784 		 */
1785 		imo = malloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1786 
1787 		if (imo == NULL)
1788 			return (ENOBUFS);
1789 		*imop = imo;
1790 		imo->imo_multicast_ifp = NULL;
1791 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1792 		imo->imo_multicast_vif = -1;
1793 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1794 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1795 		imo->imo_num_memberships = 0;
1796 	}
1797 
1798 	switch (sopt->sopt_name) {
1799 	/* store an index number for the vif you wanna use in the send */
1800 	case IP_MULTICAST_VIF:
1801 		if (legal_vif_num == 0) {
1802 			error = EOPNOTSUPP;
1803 			break;
1804 		}
1805 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1806 		if (error)
1807 			break;
1808 		if (!legal_vif_num(i) && (i != -1)) {
1809 			error = EINVAL;
1810 			break;
1811 		}
1812 		imo->imo_multicast_vif = i;
1813 		break;
1814 
1815 	case IP_MULTICAST_IF:
1816 		/*
1817 		 * Select the interface for outgoing multicast packets.
1818 		 */
1819 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1820 		if (error)
1821 			break;
1822 		/*
1823 		 * INADDR_ANY is used to remove a previous selection.
1824 		 * When no interface is selected, a default one is
1825 		 * chosen every time a multicast packet is sent.
1826 		 */
1827 		if (addr.s_addr == INADDR_ANY) {
1828 			imo->imo_multicast_ifp = NULL;
1829 			break;
1830 		}
1831 		/*
1832 		 * The selected interface is identified by its local
1833 		 * IP address.  Find the interface and confirm that
1834 		 * it supports multicasting.
1835 		 */
1836 		s = splimp();
1837 		ifp = ip_multicast_if(&addr, &ifindex);
1838 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1839 			splx(s);
1840 			error = EADDRNOTAVAIL;
1841 			break;
1842 		}
1843 		imo->imo_multicast_ifp = ifp;
1844 		if (ifindex)
1845 			imo->imo_multicast_addr = addr;
1846 		else
1847 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1848 		splx(s);
1849 		break;
1850 
1851 	case IP_MULTICAST_TTL:
1852 		/*
1853 		 * Set the IP time-to-live for outgoing multicast packets.
1854 		 * The original multicast API required a char argument,
1855 		 * which is inconsistent with the rest of the socket API.
1856 		 * We allow either a char or an int.
1857 		 */
1858 		if (sopt->sopt_valsize == 1) {
1859 			u_char ttl;
1860 			error = sooptcopyin(sopt, &ttl, 1, 1);
1861 			if (error)
1862 				break;
1863 			imo->imo_multicast_ttl = ttl;
1864 		} else {
1865 			u_int ttl;
1866 			error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1867 			if (error)
1868 				break;
1869 			if (ttl > 255)
1870 				error = EINVAL;
1871 			else
1872 				imo->imo_multicast_ttl = ttl;
1873 		}
1874 		break;
1875 
1876 	case IP_MULTICAST_LOOP:
1877 		/*
1878 		 * Set the loopback flag for outgoing multicast packets.
1879 		 * Must be zero or one.  The original multicast API required a
1880 		 * char argument, which is inconsistent with the rest
1881 		 * of the socket API.  We allow either a char or an int.
1882 		 */
1883 		if (sopt->sopt_valsize == 1) {
1884 			u_char loop;
1885 
1886 			error = sooptcopyin(sopt, &loop, 1, 1);
1887 			if (error)
1888 				break;
1889 			imo->imo_multicast_loop = !!loop;
1890 		} else {
1891 			u_int loop;
1892 
1893 			error = sooptcopyin(sopt, &loop, sizeof loop,
1894 					    sizeof loop);
1895 			if (error)
1896 				break;
1897 			imo->imo_multicast_loop = !!loop;
1898 		}
1899 		break;
1900 
1901 	case IP_ADD_MEMBERSHIP:
1902 		/*
1903 		 * Add a multicast group membership.
1904 		 * Group must be a valid IP multicast address.
1905 		 */
1906 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1907 		if (error)
1908 			break;
1909 
1910 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1911 			error = EINVAL;
1912 			break;
1913 		}
1914 		s = splimp();
1915 		/*
1916 		 * If no interface address was provided, use the interface of
1917 		 * the route to the given multicast address.
1918 		 */
1919 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1920 			struct sockaddr_in dst;
1921 			struct rtentry *rt;
1922 
1923 			bzero(&dst, sizeof(struct sockaddr_in));
1924 			dst.sin_len = sizeof(struct sockaddr_in);
1925 			dst.sin_family = AF_INET;
1926 			dst.sin_addr = mreq.imr_multiaddr;
1927 			rt = rtlookup((struct sockaddr *)&dst);
1928 			if (rt == NULL) {
1929 				error = EADDRNOTAVAIL;
1930 				splx(s);
1931 				break;
1932 			}
1933 			--rt->rt_refcnt;
1934 			ifp = rt->rt_ifp;
1935 		} else {
1936 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1937 		}
1938 
1939 		/*
1940 		 * See if we found an interface, and confirm that it
1941 		 * supports multicast.
1942 		 */
1943 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1944 			error = EADDRNOTAVAIL;
1945 			splx(s);
1946 			break;
1947 		}
1948 		/*
1949 		 * See if the membership already exists or if all the
1950 		 * membership slots are full.
1951 		 */
1952 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1953 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1954 			    imo->imo_membership[i]->inm_addr.s_addr
1955 						== mreq.imr_multiaddr.s_addr)
1956 				break;
1957 		}
1958 		if (i < imo->imo_num_memberships) {
1959 			error = EADDRINUSE;
1960 			splx(s);
1961 			break;
1962 		}
1963 		if (i == IP_MAX_MEMBERSHIPS) {
1964 			error = ETOOMANYREFS;
1965 			splx(s);
1966 			break;
1967 		}
1968 		/*
1969 		 * Everything looks good; add a new record to the multicast
1970 		 * address list for the given interface.
1971 		 */
1972 		if ((imo->imo_membership[i] =
1973 		     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1974 			error = ENOBUFS;
1975 			splx(s);
1976 			break;
1977 		}
1978 		++imo->imo_num_memberships;
1979 		splx(s);
1980 		break;
1981 
1982 	case IP_DROP_MEMBERSHIP:
1983 		/*
1984 		 * Drop a multicast group membership.
1985 		 * Group must be a valid IP multicast address.
1986 		 */
1987 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1988 		if (error)
1989 			break;
1990 
1991 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1992 			error = EINVAL;
1993 			break;
1994 		}
1995 
1996 		s = splimp();
1997 		/*
1998 		 * If an interface address was specified, get a pointer
1999 		 * to its ifnet structure.
2000 		 */
2001 		if (mreq.imr_interface.s_addr == INADDR_ANY)
2002 			ifp = NULL;
2003 		else {
2004 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2005 			if (ifp == NULL) {
2006 				error = EADDRNOTAVAIL;
2007 				splx(s);
2008 				break;
2009 			}
2010 		}
2011 		/*
2012 		 * Find the membership in the membership array.
2013 		 */
2014 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2015 			if ((ifp == NULL ||
2016 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2017 			    imo->imo_membership[i]->inm_addr.s_addr ==
2018 			    mreq.imr_multiaddr.s_addr)
2019 				break;
2020 		}
2021 		if (i == imo->imo_num_memberships) {
2022 			error = EADDRNOTAVAIL;
2023 			splx(s);
2024 			break;
2025 		}
2026 		/*
2027 		 * Give up the multicast address record to which the
2028 		 * membership points.
2029 		 */
2030 		in_delmulti(imo->imo_membership[i]);
2031 		/*
2032 		 * Remove the gap in the membership array.
2033 		 */
2034 		for (++i; i < imo->imo_num_memberships; ++i)
2035 			imo->imo_membership[i-1] = imo->imo_membership[i];
2036 		--imo->imo_num_memberships;
2037 		splx(s);
2038 		break;
2039 
2040 	default:
2041 		error = EOPNOTSUPP;
2042 		break;
2043 	}
2044 
2045 	/*
2046 	 * If all options have default values, no need to keep the mbuf.
2047 	 */
2048 	if (imo->imo_multicast_ifp == NULL &&
2049 	    imo->imo_multicast_vif == -1 &&
2050 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2051 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2052 	    imo->imo_num_memberships == 0) {
2053 		free(*imop, M_IPMOPTS);
2054 		*imop = NULL;
2055 	}
2056 
2057 	return (error);
2058 }
2059 
2060 /*
2061  * Return the IP multicast options in response to user getsockopt().
2062  */
2063 static int
2064 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2065 {
2066 	struct in_addr addr;
2067 	struct in_ifaddr *ia;
2068 	int error, optval;
2069 	u_char coptval;
2070 
2071 	error = 0;
2072 	switch (sopt->sopt_name) {
2073 	case IP_MULTICAST_VIF:
2074 		if (imo != NULL)
2075 			optval = imo->imo_multicast_vif;
2076 		else
2077 			optval = -1;
2078 		error = sooptcopyout(sopt, &optval, sizeof optval);
2079 		break;
2080 
2081 	case IP_MULTICAST_IF:
2082 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2083 			addr.s_addr = INADDR_ANY;
2084 		else if (imo->imo_multicast_addr.s_addr) {
2085 			/* return the value user has set */
2086 			addr = imo->imo_multicast_addr;
2087 		} else {
2088 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2089 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2090 				: IA_SIN(ia)->sin_addr.s_addr;
2091 		}
2092 		error = sooptcopyout(sopt, &addr, sizeof addr);
2093 		break;
2094 
2095 	case IP_MULTICAST_TTL:
2096 		if (imo == NULL)
2097 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2098 		else
2099 			optval = coptval = imo->imo_multicast_ttl;
2100 		if (sopt->sopt_valsize == 1)
2101 			error = sooptcopyout(sopt, &coptval, 1);
2102 		else
2103 			error = sooptcopyout(sopt, &optval, sizeof optval);
2104 		break;
2105 
2106 	case IP_MULTICAST_LOOP:
2107 		if (imo == NULL)
2108 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2109 		else
2110 			optval = coptval = imo->imo_multicast_loop;
2111 		if (sopt->sopt_valsize == 1)
2112 			error = sooptcopyout(sopt, &coptval, 1);
2113 		else
2114 			error = sooptcopyout(sopt, &optval, sizeof optval);
2115 		break;
2116 
2117 	default:
2118 		error = ENOPROTOOPT;
2119 		break;
2120 	}
2121 	return (error);
2122 }
2123 
2124 /*
2125  * Discard the IP multicast options.
2126  */
2127 void
2128 ip_freemoptions(struct ip_moptions *imo)
2129 {
2130 	int i;
2131 
2132 	if (imo != NULL) {
2133 		for (i = 0; i < imo->imo_num_memberships; ++i)
2134 			in_delmulti(imo->imo_membership[i]);
2135 		free(imo, M_IPMOPTS);
2136 	}
2137 }
2138 
2139 /*
2140  * Routine called from ip_output() to loop back a copy of an IP multicast
2141  * packet to the input queue of a specified interface.  Note that this
2142  * calls the output routine of the loopback "driver", but with an interface
2143  * pointer that might NOT be a loopback interface -- evil, but easier than
2144  * replicating that code here.
2145  */
2146 static void
2147 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2148 	     int hlen)
2149 {
2150 	struct ip *ip;
2151 	struct mbuf *copym;
2152 
2153 	copym = m_copypacket(m, MB_DONTWAIT);
2154 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2155 		copym = m_pullup(copym, hlen);
2156 	if (copym != NULL) {
2157 		/*
2158 		 * if the checksum hasn't been computed, mark it as valid
2159 		 */
2160 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2161 			in_delayed_cksum(copym);
2162 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2163 			copym->m_pkthdr.csum_flags |=
2164 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2165 			copym->m_pkthdr.csum_data = 0xffff;
2166 		}
2167 		/*
2168 		 * We don't bother to fragment if the IP length is greater
2169 		 * than the interface's MTU.  Can this possibly matter?
2170 		 */
2171 		ip = mtod(copym, struct ip *);
2172 		ip->ip_len = htons(ip->ip_len);
2173 		ip->ip_off = htons(ip->ip_off);
2174 		ip->ip_sum = 0;
2175 		if (ip->ip_vhl == IP_VHL_BORING) {
2176 			ip->ip_sum = in_cksum_hdr(ip);
2177 		} else {
2178 			ip->ip_sum = in_cksum(copym, hlen);
2179 		}
2180 		/*
2181 		 * NB:
2182 		 * It's not clear whether there are any lingering
2183 		 * reentrancy problems in other areas which might
2184 		 * be exposed by using ip_input directly (in
2185 		 * particular, everything which modifies the packet
2186 		 * in-place).  Yet another option is using the
2187 		 * protosw directly to deliver the looped back
2188 		 * packet.  For the moment, we'll err on the side
2189 		 * of safety by using if_simloop().
2190 		 */
2191 #if 1 /* XXX */
2192 		if (dst->sin_family != AF_INET) {
2193 			printf("ip_mloopback: bad address family %d\n",
2194 						dst->sin_family);
2195 			dst->sin_family = AF_INET;
2196 		}
2197 #endif
2198 
2199 #ifdef notdef
2200 		copym->m_pkthdr.rcvif = ifp;
2201 		ip_input(copym);
2202 #else
2203 		if_simloop(ifp, copym, dst->sin_family, 0);
2204 #endif
2205 	}
2206 }
2207