xref: /freebsd/sys/netinet/ip_output.c (revision c37420b0d5b3b6ef875fbf0b84a13f6f09be56d6)
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  * 4. 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  * $FreeBSD$
31  */
32 
33 #include "opt_ipfw.h"
34 #include "opt_ipdn.h"
35 #include "opt_ipdivert.h"
36 #include "opt_ipfilter.h"
37 #include "opt_ipsec.h"
38 #include "opt_mac.h"
39 #include "opt_pfil_hooks.h"
40 #include "opt_mbuf_stress_test.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/mac.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 
53 #include <net/if.h>
54 #include <net/route.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/ip.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_var.h>
61 #include <netinet/ip_var.h>
62 
63 #ifdef PFIL_HOOKS
64 #include <net/pfil.h>
65 #endif
66 
67 #include <machine/in_cksum.h>
68 
69 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
70 
71 #ifdef IPSEC
72 #include <netinet6/ipsec.h>
73 #include <netkey/key.h>
74 #ifdef IPSEC_DEBUG
75 #include <netkey/key_debug.h>
76 #else
77 #define	KEYDEBUG(lev,arg)
78 #endif
79 #endif /*IPSEC*/
80 
81 #ifdef FAST_IPSEC
82 #include <netipsec/ipsec.h>
83 #include <netipsec/xform.h>
84 #include <netipsec/key.h>
85 #endif /*FAST_IPSEC*/
86 
87 #include <netinet/ip_fw.h>
88 #include <netinet/ip_divert.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_STRESS_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  * In the IP forwarding case, the packet will arrive with options already
126  * inserted, so must have a NULL opt pointer.
127  */
128 int
129 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro,
130 	int flags, struct ip_moptions *imo, struct inpcb *inp)
131 {
132 	struct ip *ip;
133 	struct ifnet *ifp = NULL;	/* keep compiler happy */
134 	struct mbuf *m0;
135 	int hlen = sizeof (struct ip);
136 	int len, off, error = 0;
137 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
138 	struct in_ifaddr *ia = NULL;
139 	int isbroadcast, sw_csum;
140 	struct in_addr pkt_dst;
141 	struct route iproute;
142 	struct m_tag *mtag, *dummytag;
143 #ifdef IPSEC
144 	struct secpolicy *sp = NULL;
145 #endif
146 #ifdef FAST_IPSEC
147 	struct secpolicy *sp = NULL;
148 	struct tdb_ident *tdbi;
149 	int s;
150 #endif /* FAST_IPSEC */
151 	struct ip_fw_args args;
152 	int src_was_INADDR_ANY = 0;	/* as the name says... */
153 
154 	args.eh = NULL;
155 	args.rule = NULL;
156 
157 	M_ASSERTPKTHDR(m);
158 
159 	args.next_hop = m_claim_next(m, PACKET_TAG_IPFORWARD);
160 	dummytag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
161 	if (dummytag != NULL) {
162 		struct dn_pkt_tag *dt = (struct dn_pkt_tag *)(dummytag+1);
163 		/*
164 		 * Prevent lower layers from finding the tag
165 		 * Cleanup and free is done below
166 		 */
167 		m_tag_unlink(m, dummytag);
168 		/*
169 		 * the packet was already tagged, so part of the
170 		 * processing was already done, and we need to go down.
171 		 * Get parameters from the header.
172 		 */
173 		args.rule = dt->rule;
174 		ro = &(dt->ro);
175 		dst = dt->dn_dst;
176 		ifp = dt->ifp;
177 	}
178 
179 	if (ro == NULL) {
180 		ro = &iproute;
181 		bzero(ro, sizeof (*ro));
182 	}
183 
184 	if (inp != NULL)
185 		INP_LOCK_ASSERT(inp);
186 
187 	if (args.rule != NULL) {	/* dummynet already saw us */
188 		ip = mtod(m, struct ip *);
189 		hlen = ip->ip_hl << 2 ;
190 		if (ro->ro_rt)
191 			ia = ifatoia(ro->ro_rt->rt_ifa);
192 		goto sendit;
193 	}
194 
195 	if (opt) {
196 		len = 0;
197 		m = ip_insertoptions(m, opt, &len);
198 		if (len != 0)
199 			hlen = len;
200 	}
201 	ip = mtod(m, struct ip *);
202 	pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
203 
204 	/*
205 	 * Fill in IP header.  If we are not allowing fragmentation,
206 	 * then the ip_id field is meaningless, but we don't set it
207 	 * to zero.  Doing so causes various problems when devices along
208 	 * the path (routers, load balancers, firewalls, etc.) illegally
209 	 * disable DF on our packet.  Note that a 16-bit counter
210 	 * will wrap around in less than 10 seconds at 100 Mbit/s on a
211 	 * medium with MTU 1500.  See Steven M. Bellovin, "A Technique
212 	 * for Counting NATted Hosts", Proc. IMW'02, available at
213 	 * <http://www.research.att.com/~smb/papers/fnat.pdf>.
214 	 */
215 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
216 		ip->ip_v = IPVERSION;
217 		ip->ip_hl = hlen >> 2;
218 		ip->ip_id = ip_newid();
219 		ipstat.ips_localout++;
220 	} else {
221 		hlen = ip->ip_hl << 2;
222 	}
223 
224 	dst = (struct sockaddr_in *)&ro->ro_dst;
225 	/*
226 	 * If there is a cached route,
227 	 * check that it is to the same destination
228 	 * and is still up.  If not, free it and try again.
229 	 * The address family should also be checked in case of sharing the
230 	 * cache with IPv6.
231 	 */
232 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
233 			  dst->sin_family != AF_INET ||
234 			  dst->sin_addr.s_addr != pkt_dst.s_addr)) {
235 		RTFREE(ro->ro_rt);
236 		ro->ro_rt = (struct rtentry *)0;
237 	}
238 	if (ro->ro_rt == NULL) {
239 		bzero(dst, sizeof(*dst));
240 		dst->sin_family = AF_INET;
241 		dst->sin_len = sizeof(*dst);
242 		dst->sin_addr = pkt_dst;
243 	}
244 	/*
245 	 * If routing to interface only,
246 	 * short circuit routing lookup.
247 	 */
248 	if (flags & IP_ROUTETOIF) {
249 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
250 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
251 			ipstat.ips_noroute++;
252 			error = ENETUNREACH;
253 			goto bad;
254 		}
255 		ifp = ia->ia_ifp;
256 		ip->ip_ttl = 1;
257 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
258 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
259 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
260 		/*
261 		 * Bypass the normal routing lookup for multicast
262 		 * packets if the interface is specified.
263 		 */
264 		ifp = imo->imo_multicast_ifp;
265 		IFP_TO_IA(ifp, ia);
266 		isbroadcast = 0;	/* fool gcc */
267 	} else {
268 		/*
269 		 * We want to do any cloning requested by the link layer,
270 		 * as this is probably required in all cases for correct
271 		 * operation (as it is for ARP).
272 		 */
273 		if (ro->ro_rt == NULL)
274 			rtalloc_ign(ro, 0);
275 		if (ro->ro_rt == NULL) {
276 			ipstat.ips_noroute++;
277 			error = EHOSTUNREACH;
278 			goto bad;
279 		}
280 		ia = ifatoia(ro->ro_rt->rt_ifa);
281 		ifp = ro->ro_rt->rt_ifp;
282 		ro->ro_rt->rt_rmx.rmx_pksent++;
283 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
284 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
285 		if (ro->ro_rt->rt_flags & RTF_HOST)
286 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
287 		else
288 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
289 	}
290 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
291 		struct in_multi *inm;
292 
293 		m->m_flags |= M_MCAST;
294 		/*
295 		 * IP destination address is multicast.  Make sure "dst"
296 		 * still points to the address in "ro".  (It may have been
297 		 * changed to point to a gateway address, above.)
298 		 */
299 		dst = (struct sockaddr_in *)&ro->ro_dst;
300 		/*
301 		 * See if the caller provided any multicast options
302 		 */
303 		if (imo != NULL) {
304 			ip->ip_ttl = imo->imo_multicast_ttl;
305 			if (imo->imo_multicast_vif != -1)
306 				ip->ip_src.s_addr =
307 				    ip_mcast_src ?
308 				    ip_mcast_src(imo->imo_multicast_vif) :
309 				    INADDR_ANY;
310 		} else
311 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
312 		/*
313 		 * Confirm that the outgoing interface supports multicast.
314 		 */
315 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
316 			if ((ifp->if_flags & IFF_MULTICAST) == 0) {
317 				ipstat.ips_noroute++;
318 				error = ENETUNREACH;
319 				goto bad;
320 			}
321 		}
322 		/*
323 		 * If source address not specified yet, use address
324 		 * of outgoing interface.
325 		 */
326 		if (ip->ip_src.s_addr == INADDR_ANY) {
327 			/* Interface may have no addresses. */
328 			if (ia != NULL)
329 				ip->ip_src = IA_SIN(ia)->sin_addr;
330 		}
331 
332 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
333 		if (inm != NULL &&
334 		   (imo == NULL || imo->imo_multicast_loop)) {
335 			/*
336 			 * If we belong to the destination multicast group
337 			 * on the outgoing interface, and the caller did not
338 			 * forbid loopback, loop back a copy.
339 			 */
340 			ip_mloopback(ifp, m, dst, hlen);
341 		}
342 		else {
343 			/*
344 			 * If we are acting as a multicast router, perform
345 			 * multicast forwarding as if the packet had just
346 			 * arrived on the interface to which we are about
347 			 * to send.  The multicast forwarding function
348 			 * recursively calls this function, using the
349 			 * IP_FORWARDING flag to prevent infinite recursion.
350 			 *
351 			 * Multicasts that are looped back by ip_mloopback(),
352 			 * above, will be forwarded by the ip_input() routine,
353 			 * if necessary.
354 			 */
355 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
356 				/*
357 				 * If rsvp daemon is not running, do not
358 				 * set ip_moptions. This ensures that the packet
359 				 * is multicast and not just sent down one link
360 				 * as prescribed by rsvpd.
361 				 */
362 				if (!rsvp_on)
363 					imo = NULL;
364 				if (ip_mforward &&
365 				    ip_mforward(ip, ifp, m, imo) != 0) {
366 					m_freem(m);
367 					goto done;
368 				}
369 			}
370 		}
371 
372 		/*
373 		 * Multicasts with a time-to-live of zero may be looped-
374 		 * back, above, but must not be transmitted on a network.
375 		 * Also, multicasts addressed to the loopback interface
376 		 * are not sent -- the above call to ip_mloopback() will
377 		 * loop back a copy if this host actually belongs to the
378 		 * destination group on the loopback interface.
379 		 */
380 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
381 			m_freem(m);
382 			goto done;
383 		}
384 
385 		goto sendit;
386 	}
387 #ifndef notdef
388 	/*
389 	 * If the source address is not specified yet, use the address
390 	 * of the outoing interface. In case, keep note we did that, so
391 	 * if the the firewall changes the next-hop causing the output
392 	 * interface to change, we can fix that.
393 	 */
394 	if (ip->ip_src.s_addr == INADDR_ANY) {
395 		/* Interface may have no addresses. */
396 		if (ia != NULL) {
397 			ip->ip_src = IA_SIN(ia)->sin_addr;
398 			src_was_INADDR_ANY = 1;
399 		}
400 	}
401 #endif /* notdef */
402 #ifdef ALTQ
403 	/*
404 	 * disable packet drop hack.
405 	 * packetdrop should be done by queueing.
406 	 */
407 #else /* !ALTQ */
408 	/*
409 	 * Verify that we have any chance at all of being able to queue
410 	 *      the packet or packet fragments
411 	 */
412 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
413 		ifp->if_snd.ifq_maxlen) {
414 			error = ENOBUFS;
415 			ipstat.ips_odropped++;
416 			goto bad;
417 	}
418 #endif /* !ALTQ */
419 
420 	/*
421 	 * Look for broadcast address and
422 	 * verify user is allowed to send
423 	 * such a packet.
424 	 */
425 	if (isbroadcast) {
426 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
427 			error = EADDRNOTAVAIL;
428 			goto bad;
429 		}
430 		if ((flags & IP_ALLOWBROADCAST) == 0) {
431 			error = EACCES;
432 			goto bad;
433 		}
434 		/* don't allow broadcast messages to be fragmented */
435 		if (ip->ip_len > ifp->if_mtu) {
436 			error = EMSGSIZE;
437 			goto bad;
438 		}
439 		if (flags & IP_SENDONES)
440 			ip->ip_dst.s_addr = INADDR_BROADCAST;
441 		m->m_flags |= M_BCAST;
442 	} else {
443 		m->m_flags &= ~M_BCAST;
444 	}
445 
446 sendit:
447 #ifdef IPSEC
448 	/* get SP for this packet */
449 	if (inp == NULL)
450 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND,
451 		    flags, &error);
452 	else
453 		sp = ipsec4_getpolicybypcb(m, IPSEC_DIR_OUTBOUND, inp, &error);
454 
455 	if (sp == NULL) {
456 		ipsecstat.out_inval++;
457 		goto bad;
458 	}
459 
460 	error = 0;
461 
462 	/* check policy */
463 	switch (sp->policy) {
464 	case IPSEC_POLICY_DISCARD:
465 		/*
466 		 * This packet is just discarded.
467 		 */
468 		ipsecstat.out_polvio++;
469 		goto bad;
470 
471 	case IPSEC_POLICY_BYPASS:
472 	case IPSEC_POLICY_NONE:
473 	case IPSEC_POLICY_TCP:
474 		/* no need to do IPsec. */
475 		goto skip_ipsec;
476 
477 	case IPSEC_POLICY_IPSEC:
478 		if (sp->req == NULL) {
479 			/* acquire a policy */
480 			error = key_spdacquire(sp);
481 			goto bad;
482 		}
483 		break;
484 
485 	case IPSEC_POLICY_ENTRUST:
486 	default:
487 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
488 	}
489     {
490 	struct ipsec_output_state state;
491 	bzero(&state, sizeof(state));
492 	state.m = m;
493 	if (flags & IP_ROUTETOIF) {
494 		state.ro = &iproute;
495 		bzero(&iproute, sizeof(iproute));
496 	} else
497 		state.ro = ro;
498 	state.dst = (struct sockaddr *)dst;
499 
500 	ip->ip_sum = 0;
501 
502 	/*
503 	 * XXX
504 	 * delayed checksums are not currently compatible with IPsec
505 	 */
506 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
507 		in_delayed_cksum(m);
508 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
509 	}
510 
511 	ip->ip_len = htons(ip->ip_len);
512 	ip->ip_off = htons(ip->ip_off);
513 
514 	error = ipsec4_output(&state, sp, flags);
515 
516 	m = state.m;
517 	if (flags & IP_ROUTETOIF) {
518 		/*
519 		 * if we have tunnel mode SA, we may need to ignore
520 		 * IP_ROUTETOIF.
521 		 */
522 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
523 			flags &= ~IP_ROUTETOIF;
524 			ro = state.ro;
525 		}
526 	} else
527 		ro = state.ro;
528 	dst = (struct sockaddr_in *)state.dst;
529 	if (error) {
530 		/* mbuf is already reclaimed in ipsec4_output. */
531 		m = NULL;
532 		switch (error) {
533 		case EHOSTUNREACH:
534 		case ENETUNREACH:
535 		case EMSGSIZE:
536 		case ENOBUFS:
537 		case ENOMEM:
538 			break;
539 		default:
540 			printf("ip4_output (ipsec): error code %d\n", error);
541 			/*fall through*/
542 		case ENOENT:
543 			/* don't show these error codes to the user */
544 			error = 0;
545 			break;
546 		}
547 		goto bad;
548 	}
549 
550 	/* be sure to update variables that are affected by ipsec4_output() */
551 	ip = mtod(m, struct ip *);
552 	hlen = ip->ip_hl << 2;
553 	if (ro->ro_rt == NULL) {
554 		if ((flags & IP_ROUTETOIF) == 0) {
555 			printf("ip_output: "
556 				"can't update route after IPsec processing\n");
557 			error = EHOSTUNREACH;	/*XXX*/
558 			goto bad;
559 		}
560 	} else {
561 		if (state.encap) {
562 			ia = ifatoia(ro->ro_rt->rt_ifa);
563 			ifp = ro->ro_rt->rt_ifp;
564 		}
565 	}
566     }
567 
568 	/* make it flipped, again. */
569 	ip->ip_len = ntohs(ip->ip_len);
570 	ip->ip_off = ntohs(ip->ip_off);
571 skip_ipsec:
572 #endif /*IPSEC*/
573 #ifdef FAST_IPSEC
574 	/*
575 	 * Check the security policy (SP) for the packet and, if
576 	 * required, do IPsec-related processing.  There are two
577 	 * cases here; the first time a packet is sent through
578 	 * it will be untagged and handled by ipsec4_checkpolicy.
579 	 * If the packet is resubmitted to ip_output (e.g. after
580 	 * AH, ESP, etc. processing), there will be a tag to bypass
581 	 * the lookup and related policy checking.
582 	 */
583 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
584 	s = splnet();
585 	if (mtag != NULL) {
586 		tdbi = (struct tdb_ident *)(mtag + 1);
587 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
588 		if (sp == NULL)
589 			error = -EINVAL;	/* force silent drop */
590 		m_tag_delete(m, mtag);
591 	} else {
592 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
593 					&error, inp);
594 	}
595 	/*
596 	 * There are four return cases:
597 	 *    sp != NULL	 	    apply IPsec policy
598 	 *    sp == NULL, error == 0	    no IPsec handling needed
599 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
600 	 *    sp == NULL, error != 0	    discard packet, report error
601 	 */
602 	if (sp != NULL) {
603 		/* Loop detection, check if ipsec processing already done */
604 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
605 		for (mtag = m_tag_first(m); mtag != NULL;
606 		     mtag = m_tag_next(m, mtag)) {
607 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
608 				continue;
609 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
610 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
611 				continue;
612 			/*
613 			 * Check if policy has an SA associated with it.
614 			 * This can happen when an SP has yet to acquire
615 			 * an SA; e.g. on first reference.  If it occurs,
616 			 * then we let ipsec4_process_packet do its thing.
617 			 */
618 			if (sp->req->sav == NULL)
619 				break;
620 			tdbi = (struct tdb_ident *)(mtag + 1);
621 			if (tdbi->spi == sp->req->sav->spi &&
622 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
623 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
624 				 sizeof (union sockaddr_union)) == 0) {
625 				/*
626 				 * No IPsec processing is needed, free
627 				 * reference to SP.
628 				 *
629 				 * NB: null pointer to avoid free at
630 				 *     done: below.
631 				 */
632 				KEY_FREESP(&sp), sp = NULL;
633 				splx(s);
634 				goto spd_done;
635 			}
636 		}
637 
638 		/*
639 		 * Do delayed checksums now because we send before
640 		 * this is done in the normal processing path.
641 		 */
642 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
643 			in_delayed_cksum(m);
644 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
645 		}
646 
647 		ip->ip_len = htons(ip->ip_len);
648 		ip->ip_off = htons(ip->ip_off);
649 
650 		/* NB: callee frees mbuf */
651 		error = ipsec4_process_packet(m, sp->req, flags, 0);
652 		/*
653 		 * Preserve KAME behaviour: ENOENT can be returned
654 		 * when an SA acquire is in progress.  Don't propagate
655 		 * this to user-level; it confuses applications.
656 		 *
657 		 * XXX this will go away when the SADB is redone.
658 		 */
659 		if (error == ENOENT)
660 			error = 0;
661 		splx(s);
662 		goto done;
663 	} else {
664 		splx(s);
665 
666 		if (error != 0) {
667 			/*
668 			 * Hack: -EINVAL is used to signal that a packet
669 			 * should be silently discarded.  This is typically
670 			 * because we asked key management for an SA and
671 			 * it was delayed (e.g. kicked up to IKE).
672 			 */
673 			if (error == -EINVAL)
674 				error = 0;
675 			goto bad;
676 		} else {
677 			/* No IPsec processing for this packet. */
678 		}
679 #ifdef notyet
680 		/*
681 		 * If deferred crypto processing is needed, check that
682 		 * the interface supports it.
683 		 */
684 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
685 		if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
686 			/* notify IPsec to do its own crypto */
687 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
688 			error = EHOSTUNREACH;
689 			goto bad;
690 		}
691 #endif
692 	}
693 spd_done:
694 #endif /* FAST_IPSEC */
695 
696 	/*
697 	 * IpHack's section.
698 	 * - Xlate: translate packet's addr/port (NAT).
699 	 * - Firewall: deny/allow/etc.
700 	 * - Wrap: fake packet's addr/port <unimpl.>
701 	 * - Encapsulate: put it in another IP and send out. <unimp.>
702 	 */
703 #ifdef PFIL_HOOKS
704 	/*
705 	 * Run through list of hooks for output packets.
706 	 */
707 	error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
708 	if (error != 0 || m == NULL)
709 		goto done;
710 	ip = mtod(m, struct ip *);
711 #endif /* PFIL_HOOKS */
712 
713 	/*
714 	 * Check with the firewall...
715 	 * but not if we are already being fwd'd from a firewall.
716 	 */
717 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
718 		struct sockaddr_in *old = dst;
719 
720 		args.m = m;
721 		args.next_hop = dst;
722 		args.oif = ifp;
723 		off = ip_fw_chk_ptr(&args);
724 		m = args.m;
725 		dst = args.next_hop;
726 
727                 /*
728 		 * On return we must do the following:
729 		 * m == NULL	-> drop the pkt (old interface, deprecated)
730 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
731 		 * 1<=off<= 0xffff		-> DIVERT
732 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
733 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
734 		 * dst != old			-> IPFIREWALL_FORWARD
735 		 * off==0, dst==old		-> accept
736 		 * If some of the above modules are not compiled in, then
737 		 * we should't have to check the corresponding condition
738 		 * (because the ipfw control socket should not accept
739 		 * unsupported rules), but better play safe and drop
740 		 * packets in case of doubt.
741 		 */
742 		if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
743 			if (m)
744 				m_freem(m);
745 			error = EACCES;
746 			goto done;
747 		}
748 		ip = mtod(m, struct ip *);
749 		if (off == 0 && dst == old)		/* common case */
750 			goto pass;
751                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
752 			/*
753 			 * pass the pkt to dummynet. Need to include
754 			 * pipe number, m, ifp, ro, dst because these are
755 			 * not recomputed in the next pass.
756 			 * All other parameters have been already used and
757 			 * so they are not needed anymore.
758 			 * XXX note: if the ifp or ro entry are deleted
759 			 * while a pkt is in dummynet, we are in trouble!
760 			 */
761 			args.ro = ro;
762 			args.dst = dst;
763 			args.flags = flags;
764 
765 			error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
766 				&args);
767 			goto done;
768 		}
769 #ifdef IPDIVERT
770 		if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
771 			struct mbuf *clone;
772 
773 			/* Clone packet if we're doing a 'tee' */
774 			if ((off & IP_FW_PORT_TEE_FLAG) != 0)
775 				clone = divert_clone(m);
776 			else
777 				clone = NULL;
778 
779 			/* Restore packet header fields to original values */
780 			ip->ip_len = htons(ip->ip_len);
781 			ip->ip_off = htons(ip->ip_off);
782 
783 			/* Deliver packet to divert input routine */
784 			divert_packet(m, 0);
785 
786 			/* If 'tee', continue with original packet */
787 			if (clone != NULL) {
788 				m = clone;
789 				ip = mtod(m, struct ip *);
790 				goto pass;
791 			}
792 			goto done;
793 		}
794 #endif
795 
796 		/* IPFIREWALL_FORWARD */
797 		/*
798 		 * Check dst to make sure it is directly reachable on the
799 		 * interface we previously thought it was.
800 		 * If it isn't (which may be likely in some situations) we have
801 		 * to re-route it (ie, find a route for the next-hop and the
802 		 * associated interface) and set them here. This is nested
803 		 * forwarding which in most cases is undesirable, except where
804 		 * such control is nigh impossible. So we do it here.
805 		 * And I'm babbling.
806 		 */
807 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
808 #if 0
809 			/*
810 			 * XXX To improve readability, this block should be
811 			 * changed into a function call as below:
812 			 */
813 			error = ip_ipforward(&m, &dst, &ifp);
814 			if (error)
815 				goto bad;
816 			if (m == NULL) /* ip_input consumed the mbuf */
817 				goto done;
818 #else
819 			struct in_ifaddr *ia;
820 
821 			/*
822 			 * XXX sro_fwd below is static, and a pointer
823 			 * to it gets passed to routines downstream.
824 			 * This could have surprisingly bad results in
825 			 * practice, because its content is overwritten
826 			 * by subsequent packets.
827 			 * XXX: Breaks on SMP and possibly preemption!
828 			 */
829 			/* There must be a better way to do this next line... */
830 			static struct route sro_fwd;
831 			struct route *ro_fwd = &sro_fwd;
832 
833 #if 0
834 			print_ip("IPFIREWALL_FORWARD: New dst ip: ",
835 			    dst->sin_addr, "\n");
836 #endif
837 
838 			/*
839 			 * We need to figure out if we have been forwarded
840 			 * to a local socket. If so, then we should somehow
841 			 * "loop back" to ip_input, and get directed to the
842 			 * PCB as if we had received this packet. This is
843 			 * because it may be dificult to identify the packets
844 			 * you want to forward until they are being output
845 			 * and have selected an interface. (e.g. locally
846 			 * initiated packets) If we used the loopback inteface,
847 			 * we would not be able to control what happens
848 			 * as the packet runs through ip_input() as
849 			 * it is done through an ISR.
850 			 */
851 			LIST_FOREACH(ia,
852 			    INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
853 				/*
854 				 * If the addr to forward to is one
855 				 * of ours, we pretend to
856 				 * be the destination for this packet.
857 				 */
858 				if (IA_SIN(ia)->sin_addr.s_addr ==
859 						 dst->sin_addr.s_addr)
860 					break;
861 			}
862 			if (ia) {	/* tell ip_input "dont filter" */
863 				mtag = m_tag_get(
864 				    PACKET_TAG_IPFORWARD,
865 				    sizeof(struct sockaddr_in *), M_NOWAIT);
866 				if (mtag == NULL) {
867 					error = ENOBUFS;
868 					goto bad;
869 				}
870 				*(struct sockaddr_in **)(mtag+1) =
871 				    args.next_hop;
872 				m_tag_prepend(m, mtag);
873 
874 				if (m->m_pkthdr.rcvif == NULL)
875 					m->m_pkthdr.rcvif = ifunit("lo0");
876 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
877 					m->m_pkthdr.csum_flags |=
878 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
879 					m->m_pkthdr.csum_data = 0xffff;
880 				}
881 				m->m_pkthdr.csum_flags |=
882 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
883 				ip->ip_len = htons(ip->ip_len);
884 				ip->ip_off = htons(ip->ip_off);
885 				ip_input(m);
886 				goto done;
887 			}
888 			/*
889 			 * Some of the logic for this was
890 			 * nicked from above.
891 			 */
892 			bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
893 
894 			ro_fwd->ro_rt = 0;
895 			rtalloc_ign(ro_fwd, RTF_CLONING);
896 
897 			if (ro_fwd->ro_rt == NULL) {
898 				ipstat.ips_noroute++;
899 				error = EHOSTUNREACH;
900 				goto bad;
901 			}
902 
903 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
904 			ifp = ro_fwd->ro_rt->rt_ifp;
905 			ro_fwd->ro_rt->rt_rmx.rmx_pksent++;
906 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
907 				dst = (struct sockaddr_in *)
908 					ro_fwd->ro_rt->rt_gateway;
909 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
910 				isbroadcast =
911 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
912 			else
913 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
914 			if (ro->ro_rt)
915 				RTFREE(ro->ro_rt);
916 			ro->ro_rt = ro_fwd->ro_rt;
917 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
918 
919 #endif	/* ... block to be put into a function */
920 			/*
921 			 * If we added a default src ip earlier,
922 			 * which would have been gotten from the-then
923 			 * interface, do it again, from the new one.
924 			 */
925 			if (src_was_INADDR_ANY)
926 				ip->ip_src = IA_SIN(ia)->sin_addr;
927 			goto pass ;
928 		}
929 
930                 /*
931                  * if we get here, none of the above matches, and
932                  * we have to drop the pkt
933                  */
934 		m_freem(m);
935                 error = EACCES; /* not sure this is the right error msg */
936                 goto done;
937 	}
938 
939 pass:
940 	/* 127/8 must not appear on wire - RFC1122. */
941 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
942 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
943 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
944 			ipstat.ips_badaddr++;
945 			error = EADDRNOTAVAIL;
946 			goto bad;
947 		}
948 	}
949 
950 	m->m_pkthdr.csum_flags |= CSUM_IP;
951 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
952 	if (sw_csum & CSUM_DELAY_DATA) {
953 		in_delayed_cksum(m);
954 		sw_csum &= ~CSUM_DELAY_DATA;
955 	}
956 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
957 
958 	/*
959 	 * If small enough for interface, or the interface will take
960 	 * care of the fragmentation for us, can just send directly.
961 	 */
962 	if (ip->ip_len <= ifp->if_mtu || (ifp->if_hwassist & CSUM_FRAGMENT &&
963 	    ((ip->ip_off & IP_DF) == 0))) {
964 		ip->ip_len = htons(ip->ip_len);
965 		ip->ip_off = htons(ip->ip_off);
966 		ip->ip_sum = 0;
967 		if (sw_csum & CSUM_DELAY_IP)
968 			ip->ip_sum = in_cksum(m, hlen);
969 
970 		/* Record statistics for this interface address. */
971 		if (!(flags & IP_FORWARDING) && ia) {
972 			ia->ia_ifa.if_opackets++;
973 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
974 		}
975 
976 #ifdef IPSEC
977 		/* clean ipsec history once it goes out of the node */
978 		ipsec_delaux(m);
979 #endif
980 
981 #ifdef MBUF_STRESS_TEST
982 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
983 			m = m_fragment(m, M_DONTWAIT, mbuf_frag_size);
984 #endif
985 		error = (*ifp->if_output)(ifp, m,
986 				(struct sockaddr *)dst, ro->ro_rt);
987 		goto done;
988 	}
989 
990 	if (ip->ip_off & IP_DF) {
991 		error = EMSGSIZE;
992 		/*
993 		 * This case can happen if the user changed the MTU
994 		 * of an interface after enabling IP on it.  Because
995 		 * most netifs don't keep track of routes pointing to
996 		 * them, there is no way for one to update all its
997 		 * routes when the MTU is changed.
998 		 */
999 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1000 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1001 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1002 		}
1003 		ipstat.ips_cantfrag++;
1004 		goto bad;
1005 	}
1006 
1007 	/*
1008 	 * Too large for interface; fragment if possible. If successful,
1009 	 * on return, m will point to a list of packets to be sent.
1010 	 */
1011 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1012 	if (error)
1013 		goto bad;
1014 	for (; m; m = m0) {
1015 		m0 = m->m_nextpkt;
1016 		m->m_nextpkt = 0;
1017 #ifdef IPSEC
1018 		/* clean ipsec history once it goes out of the node */
1019 		ipsec_delaux(m);
1020 #endif
1021 		if (error == 0) {
1022 			/* Record statistics for this interface address. */
1023 			if (ia != NULL) {
1024 				ia->ia_ifa.if_opackets++;
1025 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1026 			}
1027 
1028 			error = (*ifp->if_output)(ifp, m,
1029 			    (struct sockaddr *)dst, ro->ro_rt);
1030 		} else
1031 			m_freem(m);
1032 	}
1033 
1034 	if (error == 0)
1035 		ipstat.ips_fragmented++;
1036 
1037 done:
1038 	if (ro == &iproute && ro->ro_rt) {
1039 		RTFREE(ro->ro_rt);
1040 		ro->ro_rt = NULL;
1041 	}
1042 	if (dummytag) {
1043 		struct dn_pkt_tag *dt = (struct dn_pkt_tag *)(dummytag+1);
1044 		if (dt->ro.ro_rt)
1045 			RTFREE(dt->ro.ro_rt);
1046 		m_tag_free(dummytag);
1047 	}
1048 #ifdef IPSEC
1049 	if (sp != NULL) {
1050 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1051 			printf("DP ip_output call free SP:%p\n", sp));
1052 		key_freesp(sp);
1053 	}
1054 #endif
1055 #ifdef FAST_IPSEC
1056 	if (sp != NULL)
1057 		KEY_FREESP(&sp);
1058 #endif
1059 	return (error);
1060 bad:
1061 	m_freem(m);
1062 	goto done;
1063 }
1064 
1065 /*
1066  * Create a chain of fragments which fit the given mtu. m_frag points to the
1067  * mbuf to be fragmented; on return it points to the chain with the fragments.
1068  * Return 0 if no error. If error, m_frag may contain a partially built
1069  * chain of fragments that should be freed by the caller.
1070  *
1071  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1072  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1073  */
1074 int
1075 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1076 	    u_long if_hwassist_flags, int sw_csum)
1077 {
1078 	int error = 0;
1079 	int hlen = ip->ip_hl << 2;
1080 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1081 	int off;
1082 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1083 	int firstlen;
1084 	struct mbuf **mnext;
1085 	int nfrags;
1086 
1087 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1088 		ipstat.ips_cantfrag++;
1089 		return EMSGSIZE;
1090 	}
1091 
1092 	/*
1093 	 * Must be able to put at least 8 bytes per fragment.
1094 	 */
1095 	if (len < 8)
1096 		return EMSGSIZE;
1097 
1098 	/*
1099 	 * If the interface will not calculate checksums on
1100 	 * fragmented packets, then do it here.
1101 	 */
1102 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1103 	    (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1104 		in_delayed_cksum(m0);
1105 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1106 	}
1107 
1108 	if (len > PAGE_SIZE) {
1109 		/*
1110 		 * Fragment large datagrams such that each segment
1111 		 * contains a multiple of PAGE_SIZE amount of data,
1112 		 * plus headers. This enables a receiver to perform
1113 		 * page-flipping zero-copy optimizations.
1114 		 *
1115 		 * XXX When does this help given that sender and receiver
1116 		 * could have different page sizes, and also mtu could
1117 		 * be less than the receiver's page size ?
1118 		 */
1119 		int newlen;
1120 		struct mbuf *m;
1121 
1122 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1123 			off += m->m_len;
1124 
1125 		/*
1126 		 * firstlen (off - hlen) must be aligned on an
1127 		 * 8-byte boundary
1128 		 */
1129 		if (off < hlen)
1130 			goto smart_frag_failure;
1131 		off = ((off - hlen) & ~7) + hlen;
1132 		newlen = (~PAGE_MASK) & mtu;
1133 		if ((newlen + sizeof (struct ip)) > mtu) {
1134 			/* we failed, go back the default */
1135 smart_frag_failure:
1136 			newlen = len;
1137 			off = hlen + len;
1138 		}
1139 		len = newlen;
1140 
1141 	} else {
1142 		off = hlen + len;
1143 	}
1144 
1145 	firstlen = off - hlen;
1146 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1147 
1148 	/*
1149 	 * Loop through length of segment after first fragment,
1150 	 * make new header and copy data of each part and link onto chain.
1151 	 * Here, m0 is the original packet, m is the fragment being created.
1152 	 * The fragments are linked off the m_nextpkt of the original
1153 	 * packet, which after processing serves as the first fragment.
1154 	 */
1155 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1156 		struct ip *mhip;	/* ip header on the fragment */
1157 		struct mbuf *m;
1158 		int mhlen = sizeof (struct ip);
1159 
1160 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
1161 		if (m == NULL) {
1162 			error = ENOBUFS;
1163 			ipstat.ips_odropped++;
1164 			goto done;
1165 		}
1166 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1167 		/*
1168 		 * In the first mbuf, leave room for the link header, then
1169 		 * copy the original IP header including options. The payload
1170 		 * goes into an additional mbuf chain returned by m_copy().
1171 		 */
1172 		m->m_data += max_linkhdr;
1173 		mhip = mtod(m, struct ip *);
1174 		*mhip = *ip;
1175 		if (hlen > sizeof (struct ip)) {
1176 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1177 			mhip->ip_v = IPVERSION;
1178 			mhip->ip_hl = mhlen >> 2;
1179 		}
1180 		m->m_len = mhlen;
1181 		/* XXX do we need to add ip->ip_off below ? */
1182 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1183 		if (off + len >= ip->ip_len) {	/* last fragment */
1184 			len = ip->ip_len - off;
1185 			m->m_flags |= M_LASTFRAG;
1186 		} else
1187 			mhip->ip_off |= IP_MF;
1188 		mhip->ip_len = htons((u_short)(len + mhlen));
1189 		m->m_next = m_copy(m0, off, len);
1190 		if (m->m_next == NULL) {	/* copy failed */
1191 			m_free(m);
1192 			error = ENOBUFS;	/* ??? */
1193 			ipstat.ips_odropped++;
1194 			goto done;
1195 		}
1196 		m->m_pkthdr.len = mhlen + len;
1197 		m->m_pkthdr.rcvif = (struct ifnet *)0;
1198 #ifdef MAC
1199 		mac_create_fragment(m0, m);
1200 #endif
1201 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1202 		mhip->ip_off = htons(mhip->ip_off);
1203 		mhip->ip_sum = 0;
1204 		if (sw_csum & CSUM_DELAY_IP)
1205 			mhip->ip_sum = in_cksum(m, mhlen);
1206 		*mnext = m;
1207 		mnext = &m->m_nextpkt;
1208 	}
1209 	ipstat.ips_ofragments += nfrags;
1210 
1211 	/* set first marker for fragment chain */
1212 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1213 	m0->m_pkthdr.csum_data = nfrags;
1214 
1215 	/*
1216 	 * Update first fragment by trimming what's been copied out
1217 	 * and updating header.
1218 	 */
1219 	m_adj(m0, hlen + firstlen - ip->ip_len);
1220 	m0->m_pkthdr.len = hlen + firstlen;
1221 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1222 	ip->ip_off |= IP_MF;
1223 	ip->ip_off = htons(ip->ip_off);
1224 	ip->ip_sum = 0;
1225 	if (sw_csum & CSUM_DELAY_IP)
1226 		ip->ip_sum = in_cksum(m0, hlen);
1227 
1228 done:
1229 	*m_frag = m0;
1230 	return error;
1231 }
1232 
1233 void
1234 in_delayed_cksum(struct mbuf *m)
1235 {
1236 	struct ip *ip;
1237 	u_short csum, offset;
1238 
1239 	ip = mtod(m, struct ip *);
1240 	offset = ip->ip_hl << 2 ;
1241 	csum = in_cksum_skip(m, ip->ip_len, offset);
1242 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1243 		csum = 0xffff;
1244 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1245 
1246 	if (offset + sizeof(u_short) > m->m_len) {
1247 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1248 		    m->m_len, offset, ip->ip_p);
1249 		/*
1250 		 * XXX
1251 		 * this shouldn't happen, but if it does, the
1252 		 * correct behavior may be to insert the checksum
1253 		 * in the existing chain instead of rearranging it.
1254 		 */
1255 		m = m_pullup(m, offset + sizeof(u_short));
1256 	}
1257 	*(u_short *)(m->m_data + offset) = csum;
1258 }
1259 
1260 /*
1261  * Insert IP options into preformed packet.
1262  * Adjust IP destination as required for IP source routing,
1263  * as indicated by a non-zero in_addr at the start of the options.
1264  *
1265  * XXX This routine assumes that the packet has no options in place.
1266  */
1267 static struct mbuf *
1268 ip_insertoptions(m, opt, phlen)
1269 	register struct mbuf *m;
1270 	struct mbuf *opt;
1271 	int *phlen;
1272 {
1273 	register struct ipoption *p = mtod(opt, struct ipoption *);
1274 	struct mbuf *n;
1275 	register struct ip *ip = mtod(m, struct ip *);
1276 	unsigned optlen;
1277 
1278 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1279 	if (optlen + ip->ip_len > IP_MAXPACKET) {
1280 		*phlen = 0;
1281 		return (m);		/* XXX should fail */
1282 	}
1283 	if (p->ipopt_dst.s_addr)
1284 		ip->ip_dst = p->ipopt_dst;
1285 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1286 		MGETHDR(n, M_DONTWAIT, MT_HEADER);
1287 		if (n == NULL) {
1288 			*phlen = 0;
1289 			return (m);
1290 		}
1291 		n->m_pkthdr.rcvif = (struct ifnet *)0;
1292 #ifdef MAC
1293 		mac_create_mbuf_from_mbuf(m, n);
1294 #endif
1295 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1296 		m->m_len -= sizeof(struct ip);
1297 		m->m_data += sizeof(struct ip);
1298 		n->m_next = m;
1299 		m = n;
1300 		m->m_len = optlen + sizeof(struct ip);
1301 		m->m_data += max_linkhdr;
1302 		bcopy(ip, mtod(m, void *), sizeof(struct ip));
1303 	} else {
1304 		m->m_data -= optlen;
1305 		m->m_len += optlen;
1306 		m->m_pkthdr.len += optlen;
1307 		bcopy(ip, mtod(m, void *), sizeof(struct ip));
1308 	}
1309 	ip = mtod(m, struct ip *);
1310 	bcopy(p->ipopt_list, ip + 1, optlen);
1311 	*phlen = sizeof(struct ip) + optlen;
1312 	ip->ip_v = IPVERSION;
1313 	ip->ip_hl = *phlen >> 2;
1314 	ip->ip_len += optlen;
1315 	return (m);
1316 }
1317 
1318 /*
1319  * Copy options from ip to jp,
1320  * omitting those not copied during fragmentation.
1321  */
1322 int
1323 ip_optcopy(ip, jp)
1324 	struct ip *ip, *jp;
1325 {
1326 	register u_char *cp, *dp;
1327 	int opt, optlen, cnt;
1328 
1329 	cp = (u_char *)(ip + 1);
1330 	dp = (u_char *)(jp + 1);
1331 	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
1332 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1333 		opt = cp[0];
1334 		if (opt == IPOPT_EOL)
1335 			break;
1336 		if (opt == IPOPT_NOP) {
1337 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1338 			*dp++ = IPOPT_NOP;
1339 			optlen = 1;
1340 			continue;
1341 		}
1342 
1343 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1344 		    ("ip_optcopy: malformed ipv4 option"));
1345 		optlen = cp[IPOPT_OLEN];
1346 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1347 		    ("ip_optcopy: malformed ipv4 option"));
1348 
1349 		/* bogus lengths should have been caught by ip_dooptions */
1350 		if (optlen > cnt)
1351 			optlen = cnt;
1352 		if (IPOPT_COPIED(opt)) {
1353 			bcopy(cp, dp, optlen);
1354 			dp += optlen;
1355 		}
1356 	}
1357 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1358 		*dp++ = IPOPT_EOL;
1359 	return (optlen);
1360 }
1361 
1362 /*
1363  * IP socket option processing.
1364  */
1365 int
1366 ip_ctloutput(so, sopt)
1367 	struct socket *so;
1368 	struct sockopt *sopt;
1369 {
1370 	struct	inpcb *inp = sotoinpcb(so);
1371 	int	error, optval;
1372 
1373 	error = optval = 0;
1374 	if (sopt->sopt_level != IPPROTO_IP) {
1375 		return (EINVAL);
1376 	}
1377 
1378 	switch (sopt->sopt_dir) {
1379 	case SOPT_SET:
1380 		switch (sopt->sopt_name) {
1381 		case IP_OPTIONS:
1382 #ifdef notyet
1383 		case IP_RETOPTS:
1384 #endif
1385 		{
1386 			struct mbuf *m;
1387 			if (sopt->sopt_valsize > MLEN) {
1388 				error = EMSGSIZE;
1389 				break;
1390 			}
1391 			MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_HEADER);
1392 			if (m == NULL) {
1393 				error = ENOBUFS;
1394 				break;
1395 			}
1396 			m->m_len = sopt->sopt_valsize;
1397 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1398 					    m->m_len);
1399 
1400 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1401 					   m));
1402 		}
1403 
1404 		case IP_TOS:
1405 		case IP_TTL:
1406 		case IP_RECVOPTS:
1407 		case IP_RECVRETOPTS:
1408 		case IP_RECVDSTADDR:
1409 		case IP_RECVTTL:
1410 		case IP_RECVIF:
1411 		case IP_FAITH:
1412 		case IP_ONESBCAST:
1413 			error = sooptcopyin(sopt, &optval, sizeof optval,
1414 					    sizeof optval);
1415 			if (error)
1416 				break;
1417 
1418 			switch (sopt->sopt_name) {
1419 			case IP_TOS:
1420 				inp->inp_ip_tos = optval;
1421 				break;
1422 
1423 			case IP_TTL:
1424 				inp->inp_ip_ttl = optval;
1425 				break;
1426 #define	OPTSET(bit) do {						\
1427 	INP_LOCK(inp);							\
1428 	if (optval)							\
1429 		inp->inp_flags |= bit;					\
1430 	else								\
1431 		inp->inp_flags &= ~bit;					\
1432 	INP_UNLOCK(inp);						\
1433 } while (0)
1434 
1435 			case IP_RECVOPTS:
1436 				OPTSET(INP_RECVOPTS);
1437 				break;
1438 
1439 			case IP_RECVRETOPTS:
1440 				OPTSET(INP_RECVRETOPTS);
1441 				break;
1442 
1443 			case IP_RECVDSTADDR:
1444 				OPTSET(INP_RECVDSTADDR);
1445 				break;
1446 
1447 			case IP_RECVTTL:
1448 				OPTSET(INP_RECVTTL);
1449 				break;
1450 
1451 			case IP_RECVIF:
1452 				OPTSET(INP_RECVIF);
1453 				break;
1454 
1455 			case IP_FAITH:
1456 				OPTSET(INP_FAITH);
1457 				break;
1458 
1459 			case IP_ONESBCAST:
1460 				OPTSET(INP_ONESBCAST);
1461 				break;
1462 			}
1463 			break;
1464 #undef OPTSET
1465 
1466 		case IP_MULTICAST_IF:
1467 		case IP_MULTICAST_VIF:
1468 		case IP_MULTICAST_TTL:
1469 		case IP_MULTICAST_LOOP:
1470 		case IP_ADD_MEMBERSHIP:
1471 		case IP_DROP_MEMBERSHIP:
1472 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1473 			break;
1474 
1475 		case IP_PORTRANGE:
1476 			error = sooptcopyin(sopt, &optval, sizeof optval,
1477 					    sizeof optval);
1478 			if (error)
1479 				break;
1480 
1481 			INP_LOCK(inp);
1482 			switch (optval) {
1483 			case IP_PORTRANGE_DEFAULT:
1484 				inp->inp_flags &= ~(INP_LOWPORT);
1485 				inp->inp_flags &= ~(INP_HIGHPORT);
1486 				break;
1487 
1488 			case IP_PORTRANGE_HIGH:
1489 				inp->inp_flags &= ~(INP_LOWPORT);
1490 				inp->inp_flags |= INP_HIGHPORT;
1491 				break;
1492 
1493 			case IP_PORTRANGE_LOW:
1494 				inp->inp_flags &= ~(INP_HIGHPORT);
1495 				inp->inp_flags |= INP_LOWPORT;
1496 				break;
1497 
1498 			default:
1499 				error = EINVAL;
1500 				break;
1501 			}
1502 			INP_UNLOCK(inp);
1503 			break;
1504 
1505 #if defined(IPSEC) || defined(FAST_IPSEC)
1506 		case IP_IPSEC_POLICY:
1507 		{
1508 			caddr_t req;
1509 			size_t len = 0;
1510 			int priv;
1511 			struct mbuf *m;
1512 			int optname;
1513 
1514 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1515 				break;
1516 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1517 				break;
1518 			priv = (sopt->sopt_td != NULL &&
1519 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1520 			req = mtod(m, caddr_t);
1521 			len = m->m_len;
1522 			optname = sopt->sopt_name;
1523 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1524 			m_freem(m);
1525 			break;
1526 		}
1527 #endif /*IPSEC*/
1528 
1529 		default:
1530 			error = ENOPROTOOPT;
1531 			break;
1532 		}
1533 		break;
1534 
1535 	case SOPT_GET:
1536 		switch (sopt->sopt_name) {
1537 		case IP_OPTIONS:
1538 		case IP_RETOPTS:
1539 			if (inp->inp_options)
1540 				error = sooptcopyout(sopt,
1541 						     mtod(inp->inp_options,
1542 							  char *),
1543 						     inp->inp_options->m_len);
1544 			else
1545 				sopt->sopt_valsize = 0;
1546 			break;
1547 
1548 		case IP_TOS:
1549 		case IP_TTL:
1550 		case IP_RECVOPTS:
1551 		case IP_RECVRETOPTS:
1552 		case IP_RECVDSTADDR:
1553 		case IP_RECVTTL:
1554 		case IP_RECVIF:
1555 		case IP_PORTRANGE:
1556 		case IP_FAITH:
1557 		case IP_ONESBCAST:
1558 			switch (sopt->sopt_name) {
1559 
1560 			case IP_TOS:
1561 				optval = inp->inp_ip_tos;
1562 				break;
1563 
1564 			case IP_TTL:
1565 				optval = inp->inp_ip_ttl;
1566 				break;
1567 
1568 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1569 
1570 			case IP_RECVOPTS:
1571 				optval = OPTBIT(INP_RECVOPTS);
1572 				break;
1573 
1574 			case IP_RECVRETOPTS:
1575 				optval = OPTBIT(INP_RECVRETOPTS);
1576 				break;
1577 
1578 			case IP_RECVDSTADDR:
1579 				optval = OPTBIT(INP_RECVDSTADDR);
1580 				break;
1581 
1582 			case IP_RECVTTL:
1583 				optval = OPTBIT(INP_RECVTTL);
1584 				break;
1585 
1586 			case IP_RECVIF:
1587 				optval = OPTBIT(INP_RECVIF);
1588 				break;
1589 
1590 			case IP_PORTRANGE:
1591 				if (inp->inp_flags & INP_HIGHPORT)
1592 					optval = IP_PORTRANGE_HIGH;
1593 				else if (inp->inp_flags & INP_LOWPORT)
1594 					optval = IP_PORTRANGE_LOW;
1595 				else
1596 					optval = 0;
1597 				break;
1598 
1599 			case IP_FAITH:
1600 				optval = OPTBIT(INP_FAITH);
1601 				break;
1602 
1603 			case IP_ONESBCAST:
1604 				optval = OPTBIT(INP_ONESBCAST);
1605 				break;
1606 			}
1607 			error = sooptcopyout(sopt, &optval, sizeof optval);
1608 			break;
1609 
1610 		case IP_MULTICAST_IF:
1611 		case IP_MULTICAST_VIF:
1612 		case IP_MULTICAST_TTL:
1613 		case IP_MULTICAST_LOOP:
1614 		case IP_ADD_MEMBERSHIP:
1615 		case IP_DROP_MEMBERSHIP:
1616 			error = ip_getmoptions(sopt, inp->inp_moptions);
1617 			break;
1618 
1619 #if defined(IPSEC) || defined(FAST_IPSEC)
1620 		case IP_IPSEC_POLICY:
1621 		{
1622 			struct mbuf *m = NULL;
1623 			caddr_t req = NULL;
1624 			size_t len = 0;
1625 
1626 			if (m != 0) {
1627 				req = mtod(m, caddr_t);
1628 				len = m->m_len;
1629 			}
1630 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1631 			if (error == 0)
1632 				error = soopt_mcopyout(sopt, m); /* XXX */
1633 			if (error == 0)
1634 				m_freem(m);
1635 			break;
1636 		}
1637 #endif /*IPSEC*/
1638 
1639 		default:
1640 			error = ENOPROTOOPT;
1641 			break;
1642 		}
1643 		break;
1644 	}
1645 	return (error);
1646 }
1647 
1648 /*
1649  * Set up IP options in pcb for insertion in output packets.
1650  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1651  * with destination address if source routed.
1652  */
1653 static int
1654 ip_pcbopts(optname, pcbopt, m)
1655 	int optname;
1656 	struct mbuf **pcbopt;
1657 	register struct mbuf *m;
1658 {
1659 	register int cnt, optlen;
1660 	register u_char *cp;
1661 	u_char opt;
1662 
1663 	/* turn off any old options */
1664 	if (*pcbopt)
1665 		(void)m_free(*pcbopt);
1666 	*pcbopt = 0;
1667 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1668 		/*
1669 		 * Only turning off any previous options.
1670 		 */
1671 		if (m)
1672 			(void)m_free(m);
1673 		return (0);
1674 	}
1675 
1676 	if (m->m_len % sizeof(int32_t))
1677 		goto bad;
1678 	/*
1679 	 * IP first-hop destination address will be stored before
1680 	 * actual options; move other options back
1681 	 * and clear it when none present.
1682 	 */
1683 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1684 		goto bad;
1685 	cnt = m->m_len;
1686 	m->m_len += sizeof(struct in_addr);
1687 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1688 	bcopy(mtod(m, void *), cp, (unsigned)cnt);
1689 	bzero(mtod(m, void *), sizeof(struct in_addr));
1690 
1691 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1692 		opt = cp[IPOPT_OPTVAL];
1693 		if (opt == IPOPT_EOL)
1694 			break;
1695 		if (opt == IPOPT_NOP)
1696 			optlen = 1;
1697 		else {
1698 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1699 				goto bad;
1700 			optlen = cp[IPOPT_OLEN];
1701 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1702 				goto bad;
1703 		}
1704 		switch (opt) {
1705 
1706 		default:
1707 			break;
1708 
1709 		case IPOPT_LSRR:
1710 		case IPOPT_SSRR:
1711 			/*
1712 			 * user process specifies route as:
1713 			 *	->A->B->C->D
1714 			 * D must be our final destination (but we can't
1715 			 * check that since we may not have connected yet).
1716 			 * A is first hop destination, which doesn't appear in
1717 			 * actual IP option, but is stored before the options.
1718 			 */
1719 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1720 				goto bad;
1721 			m->m_len -= sizeof(struct in_addr);
1722 			cnt -= sizeof(struct in_addr);
1723 			optlen -= sizeof(struct in_addr);
1724 			cp[IPOPT_OLEN] = optlen;
1725 			/*
1726 			 * Move first hop before start of options.
1727 			 */
1728 			bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1729 			    sizeof(struct in_addr));
1730 			/*
1731 			 * Then copy rest of options back
1732 			 * to close up the deleted entry.
1733 			 */
1734 			bcopy((&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr)),
1735 			    &cp[IPOPT_OFFSET+1],
1736 			    (unsigned)cnt - (IPOPT_MINOFF - 1));
1737 			break;
1738 		}
1739 	}
1740 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1741 		goto bad;
1742 	*pcbopt = m;
1743 	return (0);
1744 
1745 bad:
1746 	(void)m_free(m);
1747 	return (EINVAL);
1748 }
1749 
1750 /*
1751  * XXX
1752  * The whole multicast option thing needs to be re-thought.
1753  * Several of these options are equally applicable to non-multicast
1754  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1755  * standard option (IP_TTL).
1756  */
1757 
1758 /*
1759  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1760  */
1761 static struct ifnet *
1762 ip_multicast_if(a, ifindexp)
1763 	struct in_addr *a;
1764 	int *ifindexp;
1765 {
1766 	int ifindex;
1767 	struct ifnet *ifp;
1768 
1769 	if (ifindexp)
1770 		*ifindexp = 0;
1771 	if (ntohl(a->s_addr) >> 24 == 0) {
1772 		ifindex = ntohl(a->s_addr) & 0xffffff;
1773 		if (ifindex < 0 || if_index < ifindex)
1774 			return NULL;
1775 		ifp = ifnet_byindex(ifindex);
1776 		if (ifindexp)
1777 			*ifindexp = ifindex;
1778 	} else {
1779 		INADDR_TO_IFP(*a, ifp);
1780 	}
1781 	return ifp;
1782 }
1783 
1784 /*
1785  * Set the IP multicast options in response to user setsockopt().
1786  */
1787 static int
1788 ip_setmoptions(sopt, imop)
1789 	struct sockopt *sopt;
1790 	struct ip_moptions **imop;
1791 {
1792 	int error = 0;
1793 	int i;
1794 	struct in_addr addr;
1795 	struct ip_mreq mreq;
1796 	struct ifnet *ifp;
1797 	struct ip_moptions *imo = *imop;
1798 	struct route ro;
1799 	struct sockaddr_in *dst;
1800 	int ifindex;
1801 	int s;
1802 
1803 	if (imo == NULL) {
1804 		/*
1805 		 * No multicast option buffer attached to the pcb;
1806 		 * allocate one and initialize to default values.
1807 		 */
1808 		imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1809 		    M_WAITOK);
1810 
1811 		if (imo == NULL)
1812 			return (ENOBUFS);
1813 		*imop = imo;
1814 		imo->imo_multicast_ifp = NULL;
1815 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1816 		imo->imo_multicast_vif = -1;
1817 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1818 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1819 		imo->imo_num_memberships = 0;
1820 	}
1821 
1822 	switch (sopt->sopt_name) {
1823 	/* store an index number for the vif you wanna use in the send */
1824 	case IP_MULTICAST_VIF:
1825 		if (legal_vif_num == 0) {
1826 			error = EOPNOTSUPP;
1827 			break;
1828 		}
1829 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1830 		if (error)
1831 			break;
1832 		if (!legal_vif_num(i) && (i != -1)) {
1833 			error = EINVAL;
1834 			break;
1835 		}
1836 		imo->imo_multicast_vif = i;
1837 		break;
1838 
1839 	case IP_MULTICAST_IF:
1840 		/*
1841 		 * Select the interface for outgoing multicast packets.
1842 		 */
1843 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1844 		if (error)
1845 			break;
1846 		/*
1847 		 * INADDR_ANY is used to remove a previous selection.
1848 		 * When no interface is selected, a default one is
1849 		 * chosen every time a multicast packet is sent.
1850 		 */
1851 		if (addr.s_addr == INADDR_ANY) {
1852 			imo->imo_multicast_ifp = NULL;
1853 			break;
1854 		}
1855 		/*
1856 		 * The selected interface is identified by its local
1857 		 * IP address.  Find the interface and confirm that
1858 		 * it supports multicasting.
1859 		 */
1860 		s = splimp();
1861 		ifp = ip_multicast_if(&addr, &ifindex);
1862 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1863 			splx(s);
1864 			error = EADDRNOTAVAIL;
1865 			break;
1866 		}
1867 		imo->imo_multicast_ifp = ifp;
1868 		if (ifindex)
1869 			imo->imo_multicast_addr = addr;
1870 		else
1871 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1872 		splx(s);
1873 		break;
1874 
1875 	case IP_MULTICAST_TTL:
1876 		/*
1877 		 * Set the IP time-to-live for outgoing multicast packets.
1878 		 * The original multicast API required a char argument,
1879 		 * which is inconsistent with the rest of the socket API.
1880 		 * We allow either a char or an int.
1881 		 */
1882 		if (sopt->sopt_valsize == 1) {
1883 			u_char ttl;
1884 			error = sooptcopyin(sopt, &ttl, 1, 1);
1885 			if (error)
1886 				break;
1887 			imo->imo_multicast_ttl = ttl;
1888 		} else {
1889 			u_int ttl;
1890 			error = sooptcopyin(sopt, &ttl, sizeof ttl,
1891 					    sizeof ttl);
1892 			if (error)
1893 				break;
1894 			if (ttl > 255)
1895 				error = EINVAL;
1896 			else
1897 				imo->imo_multicast_ttl = ttl;
1898 		}
1899 		break;
1900 
1901 	case IP_MULTICAST_LOOP:
1902 		/*
1903 		 * Set the loopback flag for outgoing multicast packets.
1904 		 * Must be zero or one.  The original multicast API required a
1905 		 * char argument, which is inconsistent with the rest
1906 		 * of the socket API.  We allow either a char or an int.
1907 		 */
1908 		if (sopt->sopt_valsize == 1) {
1909 			u_char loop;
1910 			error = sooptcopyin(sopt, &loop, 1, 1);
1911 			if (error)
1912 				break;
1913 			imo->imo_multicast_loop = !!loop;
1914 		} else {
1915 			u_int loop;
1916 			error = sooptcopyin(sopt, &loop, sizeof loop,
1917 					    sizeof loop);
1918 			if (error)
1919 				break;
1920 			imo->imo_multicast_loop = !!loop;
1921 		}
1922 		break;
1923 
1924 	case IP_ADD_MEMBERSHIP:
1925 		/*
1926 		 * Add a multicast group membership.
1927 		 * Group must be a valid IP multicast address.
1928 		 */
1929 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1930 		if (error)
1931 			break;
1932 
1933 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1934 			error = EINVAL;
1935 			break;
1936 		}
1937 		s = splimp();
1938 		/*
1939 		 * If no interface address was provided, use the interface of
1940 		 * the route to the given multicast address.
1941 		 */
1942 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1943 			bzero((caddr_t)&ro, sizeof(ro));
1944 			dst = (struct sockaddr_in *)&ro.ro_dst;
1945 			dst->sin_len = sizeof(*dst);
1946 			dst->sin_family = AF_INET;
1947 			dst->sin_addr = mreq.imr_multiaddr;
1948 			rtalloc_ign(&ro, RTF_CLONING);
1949 			if (ro.ro_rt == NULL) {
1950 				error = EADDRNOTAVAIL;
1951 				splx(s);
1952 				break;
1953 			}
1954 			ifp = ro.ro_rt->rt_ifp;
1955 			RTFREE(ro.ro_rt);
1956 		}
1957 		else {
1958 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1959 		}
1960 
1961 		/*
1962 		 * See if we found an interface, and confirm that it
1963 		 * supports multicast.
1964 		 */
1965 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1966 			error = EADDRNOTAVAIL;
1967 			splx(s);
1968 			break;
1969 		}
1970 		/*
1971 		 * See if the membership already exists or if all the
1972 		 * membership slots are full.
1973 		 */
1974 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1975 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1976 			    imo->imo_membership[i]->inm_addr.s_addr
1977 						== mreq.imr_multiaddr.s_addr)
1978 				break;
1979 		}
1980 		if (i < imo->imo_num_memberships) {
1981 			error = EADDRINUSE;
1982 			splx(s);
1983 			break;
1984 		}
1985 		if (i == IP_MAX_MEMBERSHIPS) {
1986 			error = ETOOMANYREFS;
1987 			splx(s);
1988 			break;
1989 		}
1990 		/*
1991 		 * Everything looks good; add a new record to the multicast
1992 		 * address list for the given interface.
1993 		 */
1994 		if ((imo->imo_membership[i] =
1995 		    in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1996 			error = ENOBUFS;
1997 			splx(s);
1998 			break;
1999 		}
2000 		++imo->imo_num_memberships;
2001 		splx(s);
2002 		break;
2003 
2004 	case IP_DROP_MEMBERSHIP:
2005 		/*
2006 		 * Drop a multicast group membership.
2007 		 * Group must be a valid IP multicast address.
2008 		 */
2009 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2010 		if (error)
2011 			break;
2012 
2013 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2014 			error = EINVAL;
2015 			break;
2016 		}
2017 
2018 		s = splimp();
2019 		/*
2020 		 * If an interface address was specified, get a pointer
2021 		 * to its ifnet structure.
2022 		 */
2023 		if (mreq.imr_interface.s_addr == INADDR_ANY)
2024 			ifp = NULL;
2025 		else {
2026 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2027 			if (ifp == NULL) {
2028 				error = EADDRNOTAVAIL;
2029 				splx(s);
2030 				break;
2031 			}
2032 		}
2033 		/*
2034 		 * Find the membership in the membership array.
2035 		 */
2036 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2037 			if ((ifp == NULL ||
2038 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2039 			     imo->imo_membership[i]->inm_addr.s_addr ==
2040 			     mreq.imr_multiaddr.s_addr)
2041 				break;
2042 		}
2043 		if (i == imo->imo_num_memberships) {
2044 			error = EADDRNOTAVAIL;
2045 			splx(s);
2046 			break;
2047 		}
2048 		/*
2049 		 * Give up the multicast address record to which the
2050 		 * membership points.
2051 		 */
2052 		in_delmulti(imo->imo_membership[i]);
2053 		/*
2054 		 * Remove the gap in the membership array.
2055 		 */
2056 		for (++i; i < imo->imo_num_memberships; ++i)
2057 			imo->imo_membership[i-1] = imo->imo_membership[i];
2058 		--imo->imo_num_memberships;
2059 		splx(s);
2060 		break;
2061 
2062 	default:
2063 		error = EOPNOTSUPP;
2064 		break;
2065 	}
2066 
2067 	/*
2068 	 * If all options have default values, no need to keep the mbuf.
2069 	 */
2070 	if (imo->imo_multicast_ifp == NULL &&
2071 	    imo->imo_multicast_vif == -1 &&
2072 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2073 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2074 	    imo->imo_num_memberships == 0) {
2075 		free(*imop, M_IPMOPTS);
2076 		*imop = NULL;
2077 	}
2078 
2079 	return (error);
2080 }
2081 
2082 /*
2083  * Return the IP multicast options in response to user getsockopt().
2084  */
2085 static int
2086 ip_getmoptions(sopt, imo)
2087 	struct sockopt *sopt;
2088 	register struct ip_moptions *imo;
2089 {
2090 	struct in_addr addr;
2091 	struct in_ifaddr *ia;
2092 	int error, optval;
2093 	u_char coptval;
2094 
2095 	error = 0;
2096 	switch (sopt->sopt_name) {
2097 	case IP_MULTICAST_VIF:
2098 		if (imo != NULL)
2099 			optval = imo->imo_multicast_vif;
2100 		else
2101 			optval = -1;
2102 		error = sooptcopyout(sopt, &optval, sizeof optval);
2103 		break;
2104 
2105 	case IP_MULTICAST_IF:
2106 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2107 			addr.s_addr = INADDR_ANY;
2108 		else if (imo->imo_multicast_addr.s_addr) {
2109 			/* return the value user has set */
2110 			addr = imo->imo_multicast_addr;
2111 		} else {
2112 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2113 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2114 				: IA_SIN(ia)->sin_addr.s_addr;
2115 		}
2116 		error = sooptcopyout(sopt, &addr, sizeof addr);
2117 		break;
2118 
2119 	case IP_MULTICAST_TTL:
2120 		if (imo == 0)
2121 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2122 		else
2123 			optval = coptval = imo->imo_multicast_ttl;
2124 		if (sopt->sopt_valsize == 1)
2125 			error = sooptcopyout(sopt, &coptval, 1);
2126 		else
2127 			error = sooptcopyout(sopt, &optval, sizeof optval);
2128 		break;
2129 
2130 	case IP_MULTICAST_LOOP:
2131 		if (imo == 0)
2132 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2133 		else
2134 			optval = coptval = imo->imo_multicast_loop;
2135 		if (sopt->sopt_valsize == 1)
2136 			error = sooptcopyout(sopt, &coptval, 1);
2137 		else
2138 			error = sooptcopyout(sopt, &optval, sizeof optval);
2139 		break;
2140 
2141 	default:
2142 		error = ENOPROTOOPT;
2143 		break;
2144 	}
2145 	return (error);
2146 }
2147 
2148 /*
2149  * Discard the IP multicast options.
2150  */
2151 void
2152 ip_freemoptions(imo)
2153 	register struct ip_moptions *imo;
2154 {
2155 	register int i;
2156 
2157 	if (imo != NULL) {
2158 		for (i = 0; i < imo->imo_num_memberships; ++i)
2159 			in_delmulti(imo->imo_membership[i]);
2160 		free(imo, M_IPMOPTS);
2161 	}
2162 }
2163 
2164 /*
2165  * Routine called from ip_output() to loop back a copy of an IP multicast
2166  * packet to the input queue of a specified interface.  Note that this
2167  * calls the output routine of the loopback "driver", but with an interface
2168  * pointer that might NOT be a loopback interface -- evil, but easier than
2169  * replicating that code here.
2170  */
2171 static void
2172 ip_mloopback(ifp, m, dst, hlen)
2173 	struct ifnet *ifp;
2174 	register struct mbuf *m;
2175 	register struct sockaddr_in *dst;
2176 	int hlen;
2177 {
2178 	register struct ip *ip;
2179 	struct mbuf *copym;
2180 
2181 	copym = m_copy(m, 0, M_COPYALL);
2182 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2183 		copym = m_pullup(copym, hlen);
2184 	if (copym != NULL) {
2185 		/* If needed, compute the checksum and mark it as valid. */
2186 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2187 			in_delayed_cksum(copym);
2188 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2189 			copym->m_pkthdr.csum_flags |=
2190 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2191 			copym->m_pkthdr.csum_data = 0xffff;
2192 		}
2193 		/*
2194 		 * We don't bother to fragment if the IP length is greater
2195 		 * than the interface's MTU.  Can this possibly matter?
2196 		 */
2197 		ip = mtod(copym, struct ip *);
2198 		ip->ip_len = htons(ip->ip_len);
2199 		ip->ip_off = htons(ip->ip_off);
2200 		ip->ip_sum = 0;
2201 		ip->ip_sum = in_cksum(copym, hlen);
2202 		/*
2203 		 * NB:
2204 		 * It's not clear whether there are any lingering
2205 		 * reentrancy problems in other areas which might
2206 		 * be exposed by using ip_input directly (in
2207 		 * particular, everything which modifies the packet
2208 		 * in-place).  Yet another option is using the
2209 		 * protosw directly to deliver the looped back
2210 		 * packet.  For the moment, we'll err on the side
2211 		 * of safety by using if_simloop().
2212 		 */
2213 #if 1 /* XXX */
2214 		if (dst->sin_family != AF_INET) {
2215 			printf("ip_mloopback: bad address family %d\n",
2216 						dst->sin_family);
2217 			dst->sin_family = AF_INET;
2218 		}
2219 #endif
2220 
2221 #ifdef notdef
2222 		copym->m_pkthdr.rcvif = ifp;
2223 		ip_input(copym);
2224 #else
2225 		if_simloop(ifp, copym, dst->sin_family, 0);
2226 #endif
2227 	}
2228 }
2229