xref: /dflybsd-src/sys/netinet/ip_output.c (revision b5b0912b1891e95ccc48cad83f09239ccb7ffc16)
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 1.29 2005/06/02 23:52:42 dillon 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/thread2.h>
55 #include <sys/in_cksum.h>
56 
57 #include <net/if.h>
58 #include <net/netisr.h>
59 #include <net/pfil.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip_var.h>
68 
69 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
70 
71 #ifdef IPSEC
72 #include <netinet6/ipsec.h>
73 #include <netproto/key/key.h>
74 #ifdef IPSEC_DEBUG
75 #include <netproto/key/key_debug.h>
76 #else
77 #define	KEYDEBUG(lev,arg)
78 #endif
79 #endif /*IPSEC*/
80 
81 #ifdef FAST_IPSEC
82 #include <netproto/ipsec/ipsec.h>
83 #include <netproto/ipsec/xform.h>
84 #include <netproto/ipsec/key.h>
85 #endif /*FAST_IPSEC*/
86 
87 #include <net/ipfw/ip_fw.h>
88 #include <net/dummynet/ip_dummynet.h>
89 
90 #define print_ip(x, a, y)	 printf("%s %d.%d.%d.%d%s",\
91 				x, (ntohl(a.s_addr)>>24)&0xFF,\
92 				  (ntohl(a.s_addr)>>16)&0xFF,\
93 				  (ntohl(a.s_addr)>>8)&0xFF,\
94 				  (ntohl(a.s_addr))&0xFF, y);
95 
96 u_short ip_id;
97 
98 #ifdef MBUF_STRESS_TEST
99 int mbuf_frag_size = 0;
100 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
101 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
102 #endif
103 
104 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
105 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
106 static void	ip_mloopback
107 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
108 static int	ip_getmoptions
109 	(struct sockopt *, struct ip_moptions *);
110 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
111 static int	ip_setmoptions
112 	(struct sockopt *, struct ip_moptions **);
113 
114 int	ip_optcopy(struct ip *, struct ip *);
115 
116 
117 extern	struct protosw inetsw[];
118 
119 /*
120  * IP output.  The packet in mbuf chain m contains a skeletal IP
121  * header (with len, off, ttl, proto, tos, src, dst).
122  * The mbuf chain containing the packet will be freed.
123  * The mbuf opt, if present, will not be freed.
124  */
125 int
126 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
127 	  int flags, struct ip_moptions *imo, struct inpcb *inp)
128 {
129 	struct ip *ip;
130 	struct ifnet *ifp = NULL;	/* keep compiler happy */
131 	struct mbuf *m;
132 	int hlen = sizeof(struct ip);
133 	int len, off, error = 0;
134 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
135 	struct in_ifaddr *ia = NULL;
136 	int isbroadcast, sw_csum;
137 	struct in_addr pkt_dst;
138 	struct route iproute;
139 #ifdef IPSEC
140 	struct secpolicy *sp = NULL;
141 	struct socket *so = inp ? inp->inp_socket : NULL;
142 #endif
143 #ifdef FAST_IPSEC
144 	struct m_tag *mtag;
145 	struct secpolicy *sp = NULL;
146 	struct tdb_ident *tdbi;
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 
155 	/* Grab info from MT_TAG mbufs prepended to the chain. */
156 	while (m0 != NULL && m0->m_type == MT_TAG) {
157 		switch(m0->_m_tag_id) {
158 		case PACKET_TAG_DUMMYNET:
159 			/*
160 			 * the packet was already tagged, so part of the
161 			 * processing was already done, and we need to go down.
162 			 * Get parameters from the header.
163 			 */
164 			args.rule = ((struct dn_pkt *)m0)->rule;
165 			opt = NULL ;
166 			ro = &((struct dn_pkt *)m0)->ro;
167 			imo = NULL ;
168 			dst = ((struct dn_pkt *)m0)->dn_dst ;
169 			ifp = ((struct dn_pkt *)m0)->ifp ;
170 			flags = ((struct dn_pkt *)m0)->flags ;
171 			break;
172 		case PACKET_TAG_IPFORWARD:
173 			args.next_hop = (struct sockaddr_in *)m0->m_data;
174 			break;
175 		default:
176 			printf("ip_output: unrecognised MT_TAG tag %d\n",
177 			    m0->_m_tag_id);
178 			break;
179 		}
180 		m0 = m0->m_next;
181 	}
182 	m = m0;
183 	KASSERT(m != NULL && (m->m_flags & M_PKTHDR), ("ip_output: no HDR"));
184 
185 	if (ro == NULL) {
186 		ro = &iproute;
187 		bzero(ro, sizeof *ro);
188 	}
189 
190 	if (args.rule != NULL) {	/* dummynet already saw us */
191 		ip = mtod(m, struct ip *);
192 		hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
193 		if (ro->ro_rt)
194 			ia = ifatoia(ro->ro_rt->rt_ifa);
195 		goto sendit;
196 	}
197 
198 	if (opt) {
199 		len = 0;
200 		m = ip_insertoptions(m, opt, &len);
201 		if (len != 0)
202 			hlen = len;
203 	}
204 	ip = mtod(m, struct ip *);
205 	pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
206 
207 	/*
208 	 * Fill in IP header.
209 	 */
210 	if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
211 		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
212 		ip->ip_off &= IP_DF;
213 #ifdef RANDOM_IP_ID
214 		ip->ip_id = ip_randomid();
215 #else
216 		ip->ip_id = htons(ip_id++);
217 #endif
218 		ipstat.ips_localout++;
219 	} else {
220 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
221 	}
222 
223 	dst = (struct sockaddr_in *)&ro->ro_dst;
224 	/*
225 	 * If there is a cached route,
226 	 * check that it is to the same destination
227 	 * and is still up.  If not, free it and try again.
228 	 * The address family should also be checked in case of sharing the
229 	 * cache with IPv6.
230 	 */
231 	if (ro->ro_rt &&
232 	    (!(ro->ro_rt->rt_flags & RTF_UP) ||
233 	     dst->sin_family != AF_INET ||
234 	     dst->sin_addr.s_addr != pkt_dst.s_addr)) {
235 		rtfree(ro->ro_rt);
236 		ro->ro_rt = (struct rtentry *)NULL;
237 	}
238 	if (ro->ro_rt == NULL) {
239 		bzero(dst, sizeof *dst);
240 		dst->sin_family = AF_INET;
241 		dst->sin_len = sizeof *dst;
242 		dst->sin_addr = pkt_dst;
243 	}
244 	/*
245 	 * If routing to interface only,
246 	 * short circuit routing lookup.
247 	 */
248 	if (flags & IP_ROUTETOIF) {
249 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
250 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
251 			ipstat.ips_noroute++;
252 			error = ENETUNREACH;
253 			goto bad;
254 		}
255 		ifp = ia->ia_ifp;
256 		ip->ip_ttl = 1;
257 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
258 	} else if (IN_MULTICAST(ntohl(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 		 * If this is the case, we probably don't want to allocate
270 		 * a protocol-cloned route since we didn't get one from the
271 		 * ULP.  This lets TCP do its thing, while not burdening
272 		 * forwarding or ICMP with the overhead of cloning a route.
273 		 * Of course, we still want to do any cloning requested by
274 		 * the link layer, as this is probably required in all cases
275 		 * for correct operation (as it is for ARP).
276 		 */
277 		if (ro->ro_rt == NULL)
278 			rtalloc_ign(ro, RTF_PRCLONING);
279 		if (ro->ro_rt == NULL) {
280 			ipstat.ips_noroute++;
281 			error = EHOSTUNREACH;
282 			goto bad;
283 		}
284 		ia = ifatoia(ro->ro_rt->rt_ifa);
285 		ifp = ro->ro_rt->rt_ifp;
286 		ro->ro_rt->rt_use++;
287 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
288 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
289 		if (ro->ro_rt->rt_flags & RTF_HOST)
290 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
291 		else
292 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
293 	}
294 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
295 		struct in_multi *inm;
296 
297 		m->m_flags |= M_MCAST;
298 		/*
299 		 * IP destination address is multicast.  Make sure "dst"
300 		 * still points to the address in "ro".  (It may have been
301 		 * changed to point to a gateway address, above.)
302 		 */
303 		dst = (struct sockaddr_in *)&ro->ro_dst;
304 		/*
305 		 * See if the caller provided any multicast options
306 		 */
307 		if (imo != NULL) {
308 			ip->ip_ttl = imo->imo_multicast_ttl;
309 			if (imo->imo_multicast_vif != -1)
310 				ip->ip_src.s_addr =
311 				    ip_mcast_src ?
312 				    ip_mcast_src(imo->imo_multicast_vif) :
313 				    INADDR_ANY;
314 		} else
315 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
316 		/*
317 		 * Confirm that the outgoing interface supports multicast.
318 		 */
319 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
320 			if (!(ifp->if_flags & IFF_MULTICAST)) {
321 				ipstat.ips_noroute++;
322 				error = ENETUNREACH;
323 				goto bad;
324 			}
325 		}
326 		/*
327 		 * If source address not specified yet, use address
328 		 * of outgoing interface.
329 		 */
330 		if (ip->ip_src.s_addr == INADDR_ANY) {
331 			/* Interface may have no addresses. */
332 			if (ia != NULL)
333 				ip->ip_src = IA_SIN(ia)->sin_addr;
334 		}
335 
336 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
337 		if (inm != NULL &&
338 		   (imo == NULL || imo->imo_multicast_loop)) {
339 			/*
340 			 * If we belong to the destination multicast group
341 			 * on the outgoing interface, and the caller did not
342 			 * forbid loopback, loop back a copy.
343 			 */
344 			ip_mloopback(ifp, m, dst, hlen);
345 		}
346 		else {
347 			/*
348 			 * If we are acting as a multicast router, perform
349 			 * multicast forwarding as if the packet had just
350 			 * arrived on the interface to which we are about
351 			 * to send.  The multicast forwarding function
352 			 * recursively calls this function, using the
353 			 * IP_FORWARDING flag to prevent infinite recursion.
354 			 *
355 			 * Multicasts that are looped back by ip_mloopback(),
356 			 * above, will be forwarded by the ip_input() routine,
357 			 * if necessary.
358 			 */
359 			if (ip_mrouter && !(flags & IP_FORWARDING)) {
360 				/*
361 				 * If rsvp daemon is not running, do not
362 				 * set ip_moptions. This ensures that the packet
363 				 * is multicast and not just sent down one link
364 				 * as prescribed by rsvpd.
365 				 */
366 				if (!rsvp_on)
367 					imo = NULL;
368 				if (ip_mforward &&
369 				    ip_mforward(ip, ifp, m, imo) != 0) {
370 					m_freem(m);
371 					goto done;
372 				}
373 			}
374 		}
375 
376 		/*
377 		 * Multicasts with a time-to-live of zero may be looped-
378 		 * back, above, but must not be transmitted on a network.
379 		 * Also, multicasts addressed to the loopback interface
380 		 * are not sent -- the above call to ip_mloopback() will
381 		 * loop back a copy if this host actually belongs to the
382 		 * destination group on the loopback interface.
383 		 */
384 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
385 			m_freem(m);
386 			goto done;
387 		}
388 
389 		goto sendit;
390 	}
391 #ifndef notdef
392 	/*
393 	 * If the source address is not specified yet, use the address
394 	 * of the outoing interface. In case, keep note we did that, so
395 	 * if the the firewall changes the next-hop causing the output
396 	 * interface to change, we can fix that.
397 	 */
398 	if (ip->ip_src.s_addr == INADDR_ANY) {
399 		/* Interface may have no addresses. */
400 		if (ia != NULL) {
401 			ip->ip_src = IA_SIN(ia)->sin_addr;
402 			src_was_INADDR_ANY = 1;
403 		}
404 	}
405 #endif /* notdef */
406 #ifdef ALTQ
407 	/*
408 	 * Disable packet drop hack.
409 	 * Packetdrop should be done by queueing.
410 	 */
411 #else /* !ALTQ */
412 	/*
413 	 * Verify that we have any chance at all of being able to queue
414 	 *      the packet or packet fragments
415 	 */
416 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
417 		ifp->if_snd.ifq_maxlen) {
418 			error = ENOBUFS;
419 			ipstat.ips_odropped++;
420 			goto bad;
421 	}
422 #endif /* !ALTQ */
423 
424 	/*
425 	 * Look for broadcast address and
426 	 * verify user is allowed to send
427 	 * such a packet.
428 	 */
429 	if (isbroadcast) {
430 		if (!(ifp->if_flags & IFF_BROADCAST)) {
431 			error = EADDRNOTAVAIL;
432 			goto bad;
433 		}
434 		if (!(flags & IP_ALLOWBROADCAST)) {
435 			error = EACCES;
436 			goto bad;
437 		}
438 		/* don't allow broadcast messages to be fragmented */
439 		if (ip->ip_len > ifp->if_mtu) {
440 			error = EMSGSIZE;
441 			goto bad;
442 		}
443 		m->m_flags |= M_BCAST;
444 	} else {
445 		m->m_flags &= ~M_BCAST;
446 	}
447 
448 sendit:
449 #ifdef IPSEC
450 	/* get SP for this packet */
451 	if (so == NULL)
452 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
453 	else
454 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
455 
456 	if (sp == NULL) {
457 		ipsecstat.out_inval++;
458 		goto bad;
459 	}
460 
461 	error = 0;
462 
463 	/* check policy */
464 	switch (sp->policy) {
465 	case IPSEC_POLICY_DISCARD:
466 		/*
467 		 * This packet is just discarded.
468 		 */
469 		ipsecstat.out_polvio++;
470 		goto bad;
471 
472 	case IPSEC_POLICY_BYPASS:
473 	case IPSEC_POLICY_NONE:
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 		m0 = 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 
551 	/* be sure to update variables that are affected by ipsec4_output() */
552 	ip = mtod(m, struct ip *);
553 #ifdef _IP_VHL
554 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
555 #else
556 	hlen = ip->ip_hl << 2;
557 #endif
558 	if (ro->ro_rt == NULL) {
559 		if (!(flags & IP_ROUTETOIF)) {
560 			printf("ip_output: "
561 				"can't update route after IPsec processing\n");
562 			error = EHOSTUNREACH;	/*XXX*/
563 			goto bad;
564 		}
565 	} else {
566 		ia = ifatoia(ro->ro_rt->rt_ifa);
567 		ifp = ro->ro_rt->rt_ifp;
568 	}
569 
570 	/* make it flipped, again. */
571 	ip->ip_len = ntohs(ip->ip_len);
572 	ip->ip_off = ntohs(ip->ip_off);
573 skip_ipsec:
574 #endif /*IPSEC*/
575 #ifdef FAST_IPSEC
576 	/*
577 	 * Check the security policy (SP) for the packet and, if
578 	 * required, do IPsec-related processing.  There are two
579 	 * cases here; the first time a packet is sent through
580 	 * it will be untagged and handled by ipsec4_checkpolicy.
581 	 * If the packet is resubmitted to ip_output (e.g. after
582 	 * AH, ESP, etc. processing), there will be a tag to bypass
583 	 * the lookup and related policy checking.
584 	 */
585 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
586 	crit_enter();
587 	if (mtag != NULL) {
588 		tdbi = (struct tdb_ident *)(mtag + 1);
589 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
590 		if (sp == NULL)
591 			error = -EINVAL;	/* force silent drop */
592 		m_tag_delete(m, mtag);
593 	} else {
594 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
595 					&error, inp);
596 	}
597 	/*
598 	 * There are four return cases:
599 	 *    sp != NULL		    apply IPsec policy
600 	 *    sp == NULL, error == 0	    no IPsec handling needed
601 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
602 	 *    sp == NULL, error != 0	    discard packet, report error
603 	 */
604 	if (sp != NULL) {
605 		/* Loop detection, check if ipsec processing already done */
606 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
607 		for (mtag = m_tag_first(m); mtag != NULL;
608 		     mtag = m_tag_next(m, mtag)) {
609 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
610 				continue;
611 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
612 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
613 				continue;
614 			/*
615 			 * Check if policy has an SA associated with it.
616 			 * This can happen when an SP has yet to acquire
617 			 * an SA; e.g. on first reference.  If it occurs,
618 			 * then we let ipsec4_process_packet do its thing.
619 			 */
620 			if (sp->req->sav == NULL)
621 				break;
622 			tdbi = (struct tdb_ident *)(mtag + 1);
623 			if (tdbi->spi == sp->req->sav->spi &&
624 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
625 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
626 				 sizeof(union sockaddr_union)) == 0) {
627 				/*
628 				 * No IPsec processing is needed, free
629 				 * reference to SP.
630 				 *
631 				 * NB: null pointer to avoid free at
632 				 *     done: below.
633 				 */
634 				KEY_FREESP(&sp), sp = NULL;
635 				crit_exit();
636 				goto spd_done;
637 			}
638 		}
639 
640 		/*
641 		 * Do delayed checksums now because we send before
642 		 * this is done in the normal processing path.
643 		 */
644 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
645 			in_delayed_cksum(m);
646 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
647 		}
648 
649 		ip->ip_len = htons(ip->ip_len);
650 		ip->ip_off = htons(ip->ip_off);
651 
652 		/* NB: callee frees mbuf */
653 		error = ipsec4_process_packet(m, sp->req, flags, 0);
654 		/*
655 		 * Preserve KAME behaviour: ENOENT can be returned
656 		 * when an SA acquire is in progress.  Don't propagate
657 		 * this to user-level; it confuses applications.
658 		 *
659 		 * XXX this will go away when the SADB is redone.
660 		 */
661 		if (error == ENOENT)
662 			error = 0;
663 		crit_exit();
664 		goto done;
665 	} else {
666 		crit_exit();
667 
668 		if (error != 0) {
669 			/*
670 			 * Hack: -EINVAL is used to signal that a packet
671 			 * should be silently discarded.  This is typically
672 			 * because we asked key management for an SA and
673 			 * it was delayed (e.g. kicked up to IKE).
674 			 */
675 			if (error == -EINVAL)
676 				error = 0;
677 			goto bad;
678 		} else {
679 			/* No IPsec processing for this packet. */
680 		}
681 #ifdef notyet
682 		/*
683 		 * If deferred crypto processing is needed, check that
684 		 * the interface supports it.
685 		 */
686 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
687 		if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
688 			/* notify IPsec to do its own crypto */
689 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
690 			error = EHOSTUNREACH;
691 			goto bad;
692 		}
693 #endif
694 	}
695 spd_done:
696 #endif /* FAST_IPSEC */
697 	/*
698 	 * IpHack's section.
699 	 * - Xlate: translate packet's addr/port (NAT).
700 	 * - Firewall: deny/allow/etc.
701 	 * - Wrap: fake packet's addr/port <unimpl.>
702 	 * - Encapsulate: put it in another IP and send out. <unimp.>
703 	 */
704 
705 	/*
706 	 * Run through list of hooks for output packets.
707 	 */
708 	if (pfil_has_hooks(&inet_pfil_hook)) {
709 		error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
710 		if (error != 0 || m == NULL)
711 			goto done;
712 		ip = mtod(m, struct ip *);
713 	}
714 
715 	/*
716 	 * Check with the firewall...
717 	 * but not if we are already being fwd'd from a firewall.
718 	 */
719 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
720 		struct sockaddr_in *old = dst;
721 
722 		args.m = m;
723 		args.next_hop = dst;
724 		args.oif = ifp;
725 		off = ip_fw_chk_ptr(&args);
726 		m = args.m;
727 		dst = args.next_hop;
728 
729 		/*
730 		 * On return we must do the following:
731 		 * m == NULL	-> drop the pkt (old interface, deprecated)
732 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
733 		 * 1<=off<= 0xffff		-> DIVERT
734 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
735 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
736 		 * dst != old			-> IPFIREWALL_FORWARD
737 		 * off==0, dst==old		-> accept
738 		 * If some of the above modules are not compiled in, then
739 		 * we should't have to check the corresponding condition
740 		 * (because the ipfw control socket should not accept
741 		 * unsupported rules), but better play safe and drop
742 		 * packets in case of doubt.
743 		 */
744 		if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
745 			if (m)
746 				m_freem(m);
747 			error = EACCES;
748 			goto done;
749 		}
750 		ip = mtod(m, struct ip *);
751 		if (off == 0 && dst == old)		/* common case */
752 			goto pass;
753 		if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG)) {
754 			/*
755 			 * pass the pkt to dummynet. Need to include
756 			 * pipe number, m, ifp, ro, dst because these are
757 			 * not recomputed in the next pass.
758 			 * All other parameters have been already used and
759 			 * so they are not needed anymore.
760 			 * XXX note: if the ifp or ro entry are deleted
761 			 * while a pkt is in dummynet, we are in trouble!
762 			 */
763 			args.ro = ro;
764 			args.dst = dst;
765 			args.flags = flags;
766 
767 			error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
768 				&args);
769 			goto done;
770 		}
771 #ifdef IPDIVERT
772 		if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
773 			struct mbuf *clone = NULL;
774 
775 			/* Clone packet if we're doing a 'tee' */
776 			if ((off & IP_FW_PORT_TEE_FLAG))
777 				clone = m_dup(m, MB_DONTWAIT);
778 
779 			/*
780 			 * XXX
781 			 * delayed checksums are not currently compatible
782 			 * with divert sockets.
783 			 */
784 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
785 				in_delayed_cksum(m);
786 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
787 			}
788 
789 			/* Restore packet header fields to original values */
790 			ip->ip_len = htons(ip->ip_len);
791 			ip->ip_off = htons(ip->ip_off);
792 
793 			/* Deliver packet to divert input routine */
794 			divert_packet(m, 0, off & 0xffff);
795 
796 			/* If 'tee', continue with original packet */
797 			if (clone != NULL) {
798 				m = clone;
799 				ip = mtod(m, struct ip *);
800 				goto pass;
801 			}
802 			goto done;
803 		}
804 #endif
805 
806 		/* IPFIREWALL_FORWARD */
807 		/*
808 		 * Check dst to make sure it is directly reachable on the
809 		 * interface we previously thought it was.
810 		 * If it isn't (which may be likely in some situations) we have
811 		 * to re-route it (ie, find a route for the next-hop and the
812 		 * associated interface) and set them here. This is nested
813 		 * forwarding which in most cases is undesirable, except where
814 		 * such control is nigh impossible. So we do it here.
815 		 * And I'm babbling.
816 		 */
817 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
818 #if 0
819 			/*
820 			 * XXX To improve readability, this block should be
821 			 * changed into a function call as below:
822 			 */
823 			error = ip_ipforward(&m, &dst, &ifp);
824 			if (error)
825 				goto bad;
826 			if (m == NULL) /* ip_input consumed the mbuf */
827 				goto done;
828 #else
829 			struct in_ifaddr *ia;
830 
831 			/*
832 			 * XXX sro_fwd below is static, and a pointer
833 			 * to it gets passed to routines downstream.
834 			 * This could have surprisingly bad results in
835 			 * practice, because its content is overwritten
836 			 * by subsequent packets.
837 			 */
838 			/* There must be a better way to do this next line... */
839 			static struct route sro_fwd;
840 			struct route *ro_fwd = &sro_fwd;
841 
842 #if 0
843 			print_ip("IPFIREWALL_FORWARD: New dst ip: ",
844 			    dst->sin_addr, "\n");
845 #endif
846 
847 			/*
848 			 * We need to figure out if we have been forwarded
849 			 * to a local socket. If so, then we should somehow
850 			 * "loop back" to ip_input, and get directed to the
851 			 * PCB as if we had received this packet. This is
852 			 * because it may be dificult to identify the packets
853 			 * you want to forward until they are being output
854 			 * and have selected an interface. (e.g. locally
855 			 * initiated packets) If we used the loopback inteface,
856 			 * we would not be able to control what happens
857 			 * as the packet runs through ip_input() as
858 			 * it is done through a ISR.
859 			 */
860 			LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
861 				     ia_hash) {
862 				/*
863 				 * If the addr to forward to is one
864 				 * of ours, we pretend to
865 				 * be the destination for this packet.
866 				 */
867 				if (IA_SIN(ia)->sin_addr.s_addr ==
868 						 dst->sin_addr.s_addr)
869 					break;
870 			}
871 			if (ia != NULL) {    /* tell ip_input "dont filter" */
872 				struct m_hdr tag;
873 
874 				tag.mh_type = MT_TAG;
875 				tag.mh_flags = PACKET_TAG_IPFORWARD;
876 				tag.mh_data = (caddr_t)args.next_hop;
877 				tag.mh_next = m;
878 
879 				if (m->m_pkthdr.rcvif == NULL)
880 					m->m_pkthdr.rcvif = ifunit("lo0");
881 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
882 					m->m_pkthdr.csum_flags |=
883 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
884 					m0->m_pkthdr.csum_data = 0xffff;
885 				}
886 				m->m_pkthdr.csum_flags |=
887 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
888 				ip->ip_len = htons(ip->ip_len);
889 				ip->ip_off = htons(ip->ip_off);
890 				ip_input((struct mbuf *)&tag);
891 				goto done;
892 			}
893 			/* Some of the logic for this was nicked from above.
894 			 *
895 			 * This rewrites the cached route in a local PCB.
896 			 * Is this what we want to do?
897 			 */
898 			bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
899 			ro_fwd->ro_rt = NULL;
900 
901 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
902 			if (ro_fwd->ro_rt == NULL) {
903 				ipstat.ips_noroute++;
904 				error = EHOSTUNREACH;
905 				goto bad;
906 			}
907 
908 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
909 			ifp = ro_fwd->ro_rt->rt_ifp;
910 			ro_fwd->ro_rt->rt_use++;
911 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
912 				dst = (struct sockaddr_in *)
913 				    ro_fwd->ro_rt->rt_gateway;
914 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
915 				isbroadcast =
916 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
917 			else
918 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
919 			if (ro->ro_rt != NULL)
920 				rtfree(ro->ro_rt);
921 			ro->ro_rt = ro_fwd->ro_rt;
922 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
923 
924 #endif	/* ... block to be put into a function */
925 			/*
926 			 * If we added a default src ip earlier,
927 			 * which would have been gotten from the-then
928 			 * interface, do it again, from the new one.
929 			 */
930 			if (src_was_INADDR_ANY)
931 				ip->ip_src = IA_SIN(ia)->sin_addr;
932 			goto pass ;
933 		}
934 
935 		/*
936 		 * if we get here, none of the above matches, and
937 		 * we have to drop the pkt
938 		 */
939 		m_freem(m);
940 		error = EACCES; /* not sure this is the right error msg */
941 		goto done;
942 	}
943 
944 pass:
945 	/* 127/8 must not appear on wire - RFC1122. */
946 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
947 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
948 		if (!(ifp->if_flags & IFF_LOOPBACK)) {
949 			ipstat.ips_badaddr++;
950 			error = EADDRNOTAVAIL;
951 			goto bad;
952 		}
953 	}
954 
955 	m->m_pkthdr.csum_flags |= CSUM_IP;
956 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
957 	if (sw_csum & CSUM_DELAY_DATA) {
958 		in_delayed_cksum(m);
959 		sw_csum &= ~CSUM_DELAY_DATA;
960 	}
961 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
962 
963 	/*
964 	 * If small enough for interface, or the interface will take
965 	 * care of the fragmentation for us, can just send directly.
966 	 */
967 	if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
968 	    !(ip->ip_off & IP_DF))) {
969 		ip->ip_len = htons(ip->ip_len);
970 		ip->ip_off = htons(ip->ip_off);
971 		ip->ip_sum = 0;
972 		if (sw_csum & CSUM_DELAY_IP) {
973 			if (ip->ip_vhl == IP_VHL_BORING) {
974 				ip->ip_sum = in_cksum_hdr(ip);
975 			} else {
976 				ip->ip_sum = in_cksum(m, hlen);
977 			}
978 		}
979 
980 		/* Record statistics for this interface address. */
981 		if (!(flags & IP_FORWARDING) && ia) {
982 			ia->ia_ifa.if_opackets++;
983 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
984 		}
985 
986 #ifdef IPSEC
987 		/* clean ipsec history once it goes out of the node */
988 		ipsec_delaux(m);
989 #endif
990 
991 #ifdef MBUF_STRESS_TEST
992 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
993 			struct mbuf *m1, *m2;
994 			int length, tmp;
995 
996 			tmp = length = m->m_pkthdr.len;
997 
998 			while ((length -= mbuf_frag_size) >= 1) {
999 				m1 = m_split(m, length, MB_DONTWAIT);
1000 				if (m1 == NULL)
1001 					break;
1002 				m1->m_flags &= ~M_PKTHDR;
1003 				m2 = m;
1004 				while (m2->m_next != NULL)
1005 					m2 = m2->m_next;
1006 				m2->m_next = m1;
1007 			}
1008 			m->m_pkthdr.len = tmp;
1009 		}
1010 #endif
1011 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1012 					  ro->ro_rt);
1013 		goto done;
1014 	}
1015 
1016 	if (ip->ip_off & IP_DF) {
1017 		error = EMSGSIZE;
1018 		/*
1019 		 * This case can happen if the user changed the MTU
1020 		 * of an interface after enabling IP on it.  Because
1021 		 * most netifs don't keep track of routes pointing to
1022 		 * them, there is no way for one to update all its
1023 		 * routes when the MTU is changed.
1024 		 */
1025 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1026 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1027 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1028 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1029 		}
1030 		ipstat.ips_cantfrag++;
1031 		goto bad;
1032 	}
1033 
1034 	/*
1035 	 * Too large for interface; fragment if possible. If successful,
1036 	 * on return, m will point to a list of packets to be sent.
1037 	 */
1038 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1039 	if (error)
1040 		goto bad;
1041 	for (; m; m = m0) {
1042 		m0 = m->m_nextpkt;
1043 		m->m_nextpkt = NULL;
1044 #ifdef IPSEC
1045 		/* clean ipsec history once it goes out of the node */
1046 		ipsec_delaux(m);
1047 #endif
1048 		if (error == 0) {
1049 			/* Record statistics for this interface address. */
1050 			if (ia != NULL) {
1051 				ia->ia_ifa.if_opackets++;
1052 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1053 			}
1054 			error = (*ifp->if_output)(ifp, m,
1055 						  (struct sockaddr *)dst,
1056 						  ro->ro_rt);
1057 		} else
1058 			m_freem(m);
1059 	}
1060 
1061 	if (error == 0)
1062 		ipstat.ips_fragmented++;
1063 
1064 done:
1065 	if (ro == &iproute && ro->ro_rt != NULL) {
1066 		RTFREE(ro->ro_rt);
1067 		ro->ro_rt = NULL;
1068 	}
1069 #ifdef IPSEC
1070 	if (sp != NULL) {
1071 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1072 			printf("DP ip_output call free SP:%p\n", sp));
1073 		key_freesp(sp);
1074 	}
1075 #endif
1076 #ifdef FAST_IPSEC
1077 	if (sp != NULL)
1078 		KEY_FREESP(&sp);
1079 #endif
1080 	return (error);
1081 bad:
1082 	m_freem(m);
1083 	goto done;
1084 }
1085 
1086 /*
1087  * Create a chain of fragments which fit the given mtu. m_frag points to the
1088  * mbuf to be fragmented; on return it points to the chain with the fragments.
1089  * Return 0 if no error. If error, m_frag may contain a partially built
1090  * chain of fragments that should be freed by the caller.
1091  *
1092  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1093  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1094  */
1095 int
1096 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1097 	    u_long if_hwassist_flags, int sw_csum)
1098 {
1099 	int error = 0;
1100 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1101 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1102 	int off;
1103 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1104 	int firstlen;
1105 	struct mbuf **mnext;
1106 	int nfrags;
1107 
1108 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1109 		ipstat.ips_cantfrag++;
1110 		return EMSGSIZE;
1111 	}
1112 
1113 	/*
1114 	 * Must be able to put at least 8 bytes per fragment.
1115 	 */
1116 	if (len < 8)
1117 		return EMSGSIZE;
1118 
1119 	/*
1120 	 * If the interface will not calculate checksums on
1121 	 * fragmented packets, then do it here.
1122 	 */
1123 	if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1124 	    !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1125 		in_delayed_cksum(m0);
1126 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1127 	}
1128 
1129 	if (len > PAGE_SIZE) {
1130 		/*
1131 		 * Fragment large datagrams such that each segment
1132 		 * contains a multiple of PAGE_SIZE amount of data,
1133 		 * plus headers. This enables a receiver to perform
1134 		 * page-flipping zero-copy optimizations.
1135 		 *
1136 		 * XXX When does this help given that sender and receiver
1137 		 * could have different page sizes, and also mtu could
1138 		 * be less than the receiver's page size ?
1139 		 */
1140 		int newlen;
1141 		struct mbuf *m;
1142 
1143 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1144 			off += m->m_len;
1145 
1146 		/*
1147 		 * firstlen (off - hlen) must be aligned on an
1148 		 * 8-byte boundary
1149 		 */
1150 		if (off < hlen)
1151 			goto smart_frag_failure;
1152 		off = ((off - hlen) & ~7) + hlen;
1153 		newlen = (~PAGE_MASK) & mtu;
1154 		if ((newlen + sizeof(struct ip)) > mtu) {
1155 			/* we failed, go back the default */
1156 smart_frag_failure:
1157 			newlen = len;
1158 			off = hlen + len;
1159 		}
1160 		len = newlen;
1161 
1162 	} else {
1163 		off = hlen + len;
1164 	}
1165 
1166 	firstlen = off - hlen;
1167 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1168 
1169 	/*
1170 	 * Loop through length of segment after first fragment,
1171 	 * make new header and copy data of each part and link onto chain.
1172 	 * Here, m0 is the original packet, m is the fragment being created.
1173 	 * The fragments are linked off the m_nextpkt of the original
1174 	 * packet, which after processing serves as the first fragment.
1175 	 */
1176 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1177 		struct ip *mhip;	/* ip header on the fragment */
1178 		struct mbuf *m;
1179 		int mhlen = sizeof(struct ip);
1180 
1181 		MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1182 		if (m == NULL) {
1183 			error = ENOBUFS;
1184 			ipstat.ips_odropped++;
1185 			goto done;
1186 		}
1187 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1188 		/*
1189 		 * In the first mbuf, leave room for the link header, then
1190 		 * copy the original IP header including options. The payload
1191 		 * goes into an additional mbuf chain returned by m_copy().
1192 		 */
1193 		m->m_data += max_linkhdr;
1194 		mhip = mtod(m, struct ip *);
1195 		*mhip = *ip;
1196 		if (hlen > sizeof(struct ip)) {
1197 			mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1198 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1199 		}
1200 		m->m_len = mhlen;
1201 		/* XXX do we need to add ip->ip_off below ? */
1202 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1203 		if (off + len >= ip->ip_len) {	/* last fragment */
1204 			len = ip->ip_len - off;
1205 			m->m_flags |= M_LASTFRAG;
1206 		} else
1207 			mhip->ip_off |= IP_MF;
1208 		mhip->ip_len = htons((u_short)(len + mhlen));
1209 		m->m_next = m_copy(m0, off, len);
1210 		if (m->m_next == NULL) {		/* copy failed */
1211 			m_free(m);
1212 			error = ENOBUFS;	/* ??? */
1213 			ipstat.ips_odropped++;
1214 			goto done;
1215 		}
1216 		m->m_pkthdr.len = mhlen + len;
1217 		m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1218 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1219 		mhip->ip_off = htons(mhip->ip_off);
1220 		mhip->ip_sum = 0;
1221 		if (sw_csum & CSUM_DELAY_IP)
1222 			mhip->ip_sum = in_cksum(m, mhlen);
1223 		*mnext = m;
1224 		mnext = &m->m_nextpkt;
1225 	}
1226 	ipstat.ips_ofragments += nfrags;
1227 
1228 	/* set first marker for fragment chain */
1229 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1230 	m0->m_pkthdr.csum_data = nfrags;
1231 
1232 	/*
1233 	 * Update first fragment by trimming what's been copied out
1234 	 * and updating header.
1235 	 */
1236 	m_adj(m0, hlen + firstlen - ip->ip_len);
1237 	m0->m_pkthdr.len = hlen + firstlen;
1238 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1239 	ip->ip_off |= IP_MF;
1240 	ip->ip_off = htons(ip->ip_off);
1241 	ip->ip_sum = 0;
1242 	if (sw_csum & CSUM_DELAY_IP)
1243 		ip->ip_sum = in_cksum(m0, hlen);
1244 
1245 done:
1246 	*m_frag = m0;
1247 	return error;
1248 }
1249 
1250 void
1251 in_delayed_cksum(struct mbuf *m)
1252 {
1253 	struct ip *ip;
1254 	u_short csum, offset;
1255 
1256 	ip = mtod(m, struct ip *);
1257 	offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1258 	csum = in_cksum_skip(m, ip->ip_len, offset);
1259 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1260 		csum = 0xffff;
1261 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1262 
1263 	if (offset + sizeof(u_short) > m->m_len) {
1264 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1265 		    m->m_len, offset, ip->ip_p);
1266 		/*
1267 		 * XXX
1268 		 * this shouldn't happen, but if it does, the
1269 		 * correct behavior may be to insert the checksum
1270 		 * in the existing chain instead of rearranging it.
1271 		 */
1272 		m = m_pullup(m, offset + sizeof(u_short));
1273 	}
1274 	*(u_short *)(m->m_data + offset) = csum;
1275 }
1276 
1277 /*
1278  * Insert IP options into preformed packet.
1279  * Adjust IP destination as required for IP source routing,
1280  * as indicated by a non-zero in_addr at the start of the options.
1281  *
1282  * XXX This routine assumes that the packet has no options in place.
1283  */
1284 static struct mbuf *
1285 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1286 {
1287 	struct ipoption *p = mtod(opt, struct ipoption *);
1288 	struct mbuf *n;
1289 	struct ip *ip = mtod(m, struct ip *);
1290 	unsigned optlen;
1291 
1292 	optlen = opt->m_len - sizeof p->ipopt_dst;
1293 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1294 		*phlen = 0;
1295 		return (m);		/* XXX should fail */
1296 	}
1297 	if (p->ipopt_dst.s_addr)
1298 		ip->ip_dst = p->ipopt_dst;
1299 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1300 		MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1301 		if (n == NULL) {
1302 			*phlen = 0;
1303 			return (m);
1304 		}
1305 		n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1306 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1307 		m->m_len -= sizeof(struct ip);
1308 		m->m_data += sizeof(struct ip);
1309 		n->m_next = m;
1310 		m = n;
1311 		m->m_len = optlen + sizeof(struct ip);
1312 		m->m_data += max_linkhdr;
1313 		memcpy(mtod(m, void *), ip, sizeof(struct ip));
1314 	} else {
1315 		m->m_data -= optlen;
1316 		m->m_len += optlen;
1317 		m->m_pkthdr.len += optlen;
1318 		ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1319 	}
1320 	ip = mtod(m, struct ip *);
1321 	bcopy(p->ipopt_list, ip + 1, optlen);
1322 	*phlen = sizeof(struct ip) + optlen;
1323 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1324 	ip->ip_len += optlen;
1325 	return (m);
1326 }
1327 
1328 /*
1329  * Copy options from ip to jp,
1330  * omitting those not copied during fragmentation.
1331  */
1332 int
1333 ip_optcopy(struct ip *ip, struct ip *jp)
1334 {
1335 	u_char *cp, *dp;
1336 	int opt, optlen, cnt;
1337 
1338 	cp = (u_char *)(ip + 1);
1339 	dp = (u_char *)(jp + 1);
1340 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1341 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1342 		opt = cp[0];
1343 		if (opt == IPOPT_EOL)
1344 			break;
1345 		if (opt == IPOPT_NOP) {
1346 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1347 			*dp++ = IPOPT_NOP;
1348 			optlen = 1;
1349 			continue;
1350 		}
1351 
1352 		KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1353 		    ("ip_optcopy: malformed ipv4 option"));
1354 		optlen = cp[IPOPT_OLEN];
1355 		KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1356 		    ("ip_optcopy: malformed ipv4 option"));
1357 
1358 		/* bogus lengths should have been caught by ip_dooptions */
1359 		if (optlen > cnt)
1360 			optlen = cnt;
1361 		if (IPOPT_COPIED(opt)) {
1362 			bcopy(cp, dp, optlen);
1363 			dp += optlen;
1364 		}
1365 	}
1366 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1367 		*dp++ = IPOPT_EOL;
1368 	return (optlen);
1369 }
1370 
1371 /*
1372  * IP socket option processing.
1373  */
1374 int
1375 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1376 {
1377 	struct	inpcb *inp = so->so_pcb;
1378 	int	error, optval;
1379 
1380 	error = optval = 0;
1381 	if (sopt->sopt_level != IPPROTO_IP) {
1382 		return (EINVAL);
1383 	}
1384 
1385 	switch (sopt->sopt_dir) {
1386 	case SOPT_SET:
1387 		switch (sopt->sopt_name) {
1388 		case IP_OPTIONS:
1389 #ifdef notyet
1390 		case IP_RETOPTS:
1391 #endif
1392 		{
1393 			struct mbuf *m;
1394 			if (sopt->sopt_valsize > MLEN) {
1395 				error = EMSGSIZE;
1396 				break;
1397 			}
1398 			MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1399 			if (m == NULL) {
1400 				error = ENOBUFS;
1401 				break;
1402 			}
1403 			m->m_len = sopt->sopt_valsize;
1404 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1405 					    m->m_len);
1406 
1407 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1408 					   m));
1409 		}
1410 
1411 		case IP_TOS:
1412 		case IP_TTL:
1413 		case IP_RECVOPTS:
1414 		case IP_RECVRETOPTS:
1415 		case IP_RECVDSTADDR:
1416 		case IP_RECVIF:
1417 		case IP_FAITH:
1418 			error = sooptcopyin(sopt, &optval, sizeof optval,
1419 					    sizeof optval);
1420 			if (error)
1421 				break;
1422 
1423 			switch (sopt->sopt_name) {
1424 			case IP_TOS:
1425 				inp->inp_ip_tos = optval;
1426 				break;
1427 
1428 			case IP_TTL:
1429 				inp->inp_ip_ttl = optval;
1430 				break;
1431 #define	OPTSET(bit) \
1432 	if (optval) \
1433 		inp->inp_flags |= bit; \
1434 	else \
1435 		inp->inp_flags &= ~bit;
1436 
1437 			case IP_RECVOPTS:
1438 				OPTSET(INP_RECVOPTS);
1439 				break;
1440 
1441 			case IP_RECVRETOPTS:
1442 				OPTSET(INP_RECVRETOPTS);
1443 				break;
1444 
1445 			case IP_RECVDSTADDR:
1446 				OPTSET(INP_RECVDSTADDR);
1447 				break;
1448 
1449 			case IP_RECVIF:
1450 				OPTSET(INP_RECVIF);
1451 				break;
1452 
1453 			case IP_FAITH:
1454 				OPTSET(INP_FAITH);
1455 				break;
1456 			}
1457 			break;
1458 #undef OPTSET
1459 
1460 		case IP_MULTICAST_IF:
1461 		case IP_MULTICAST_VIF:
1462 		case IP_MULTICAST_TTL:
1463 		case IP_MULTICAST_LOOP:
1464 		case IP_ADD_MEMBERSHIP:
1465 		case IP_DROP_MEMBERSHIP:
1466 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1467 			break;
1468 
1469 		case IP_PORTRANGE:
1470 			error = sooptcopyin(sopt, &optval, sizeof optval,
1471 					    sizeof optval);
1472 			if (error)
1473 				break;
1474 
1475 			switch (optval) {
1476 			case IP_PORTRANGE_DEFAULT:
1477 				inp->inp_flags &= ~(INP_LOWPORT);
1478 				inp->inp_flags &= ~(INP_HIGHPORT);
1479 				break;
1480 
1481 			case IP_PORTRANGE_HIGH:
1482 				inp->inp_flags &= ~(INP_LOWPORT);
1483 				inp->inp_flags |= INP_HIGHPORT;
1484 				break;
1485 
1486 			case IP_PORTRANGE_LOW:
1487 				inp->inp_flags &= ~(INP_HIGHPORT);
1488 				inp->inp_flags |= INP_LOWPORT;
1489 				break;
1490 
1491 			default:
1492 				error = EINVAL;
1493 				break;
1494 			}
1495 			break;
1496 
1497 #if defined(IPSEC) || defined(FAST_IPSEC)
1498 		case IP_IPSEC_POLICY:
1499 		{
1500 			caddr_t req;
1501 			size_t len = 0;
1502 			int priv;
1503 			struct mbuf *m;
1504 			int optname;
1505 
1506 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1507 				break;
1508 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1509 				break;
1510 			priv = (sopt->sopt_td != NULL &&
1511 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1512 			req = mtod(m, caddr_t);
1513 			len = m->m_len;
1514 			optname = sopt->sopt_name;
1515 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1516 			m_freem(m);
1517 			break;
1518 		}
1519 #endif /*IPSEC*/
1520 
1521 		default:
1522 			error = ENOPROTOOPT;
1523 			break;
1524 		}
1525 		break;
1526 
1527 	case SOPT_GET:
1528 		switch (sopt->sopt_name) {
1529 		case IP_OPTIONS:
1530 		case IP_RETOPTS:
1531 			if (inp->inp_options)
1532 				error = sooptcopyout(sopt,
1533 						     mtod(inp->inp_options,
1534 							  char *),
1535 						     inp->inp_options->m_len);
1536 			else
1537 				sopt->sopt_valsize = 0;
1538 			break;
1539 
1540 		case IP_TOS:
1541 		case IP_TTL:
1542 		case IP_RECVOPTS:
1543 		case IP_RECVRETOPTS:
1544 		case IP_RECVDSTADDR:
1545 		case IP_RECVIF:
1546 		case IP_PORTRANGE:
1547 		case IP_FAITH:
1548 			switch (sopt->sopt_name) {
1549 
1550 			case IP_TOS:
1551 				optval = inp->inp_ip_tos;
1552 				break;
1553 
1554 			case IP_TTL:
1555 				optval = inp->inp_ip_ttl;
1556 				break;
1557 
1558 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1559 
1560 			case IP_RECVOPTS:
1561 				optval = OPTBIT(INP_RECVOPTS);
1562 				break;
1563 
1564 			case IP_RECVRETOPTS:
1565 				optval = OPTBIT(INP_RECVRETOPTS);
1566 				break;
1567 
1568 			case IP_RECVDSTADDR:
1569 				optval = OPTBIT(INP_RECVDSTADDR);
1570 				break;
1571 
1572 			case IP_RECVIF:
1573 				optval = OPTBIT(INP_RECVIF);
1574 				break;
1575 
1576 			case IP_PORTRANGE:
1577 				if (inp->inp_flags & INP_HIGHPORT)
1578 					optval = IP_PORTRANGE_HIGH;
1579 				else if (inp->inp_flags & INP_LOWPORT)
1580 					optval = IP_PORTRANGE_LOW;
1581 				else
1582 					optval = 0;
1583 				break;
1584 
1585 			case IP_FAITH:
1586 				optval = OPTBIT(INP_FAITH);
1587 				break;
1588 			}
1589 			error = sooptcopyout(sopt, &optval, sizeof optval);
1590 			break;
1591 
1592 		case IP_MULTICAST_IF:
1593 		case IP_MULTICAST_VIF:
1594 		case IP_MULTICAST_TTL:
1595 		case IP_MULTICAST_LOOP:
1596 		case IP_ADD_MEMBERSHIP:
1597 		case IP_DROP_MEMBERSHIP:
1598 			error = ip_getmoptions(sopt, inp->inp_moptions);
1599 			break;
1600 
1601 #if defined(IPSEC) || defined(FAST_IPSEC)
1602 		case IP_IPSEC_POLICY:
1603 		{
1604 			struct mbuf *m = NULL;
1605 			caddr_t req = NULL;
1606 			size_t len = 0;
1607 
1608 			if (m != NULL) {
1609 				req = mtod(m, caddr_t);
1610 				len = m->m_len;
1611 			}
1612 			error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1613 			if (error == 0)
1614 				error = soopt_mcopyout(sopt, m); /* XXX */
1615 			if (error == 0)
1616 				m_freem(m);
1617 			break;
1618 		}
1619 #endif /*IPSEC*/
1620 
1621 		default:
1622 			error = ENOPROTOOPT;
1623 			break;
1624 		}
1625 		break;
1626 	}
1627 	return (error);
1628 }
1629 
1630 /*
1631  * Set up IP options in pcb for insertion in output packets.
1632  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1633  * with destination address if source routed.
1634  */
1635 static int
1636 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1637 {
1638 	int cnt, optlen;
1639 	u_char *cp;
1640 	u_char opt;
1641 
1642 	/* turn off any old options */
1643 	if (*pcbopt)
1644 		m_free(*pcbopt);
1645 	*pcbopt = 0;
1646 	if (m == NULL || m->m_len == 0) {
1647 		/*
1648 		 * Only turning off any previous options.
1649 		 */
1650 		if (m != NULL)
1651 			m_free(m);
1652 		return (0);
1653 	}
1654 
1655 	if (m->m_len % sizeof(int32_t))
1656 		goto bad;
1657 	/*
1658 	 * IP first-hop destination address will be stored before
1659 	 * actual options; move other options back
1660 	 * and clear it when none present.
1661 	 */
1662 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1663 		goto bad;
1664 	cnt = m->m_len;
1665 	m->m_len += sizeof(struct in_addr);
1666 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1667 	ovbcopy(mtod(m, caddr_t), cp, cnt);
1668 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1669 
1670 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1671 		opt = cp[IPOPT_OPTVAL];
1672 		if (opt == IPOPT_EOL)
1673 			break;
1674 		if (opt == IPOPT_NOP)
1675 			optlen = 1;
1676 		else {
1677 			if (cnt < IPOPT_OLEN + sizeof *cp)
1678 				goto bad;
1679 			optlen = cp[IPOPT_OLEN];
1680 			if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1681 				goto bad;
1682 		}
1683 		switch (opt) {
1684 
1685 		default:
1686 			break;
1687 
1688 		case IPOPT_LSRR:
1689 		case IPOPT_SSRR:
1690 			/*
1691 			 * user process specifies route as:
1692 			 *	->A->B->C->D
1693 			 * D must be our final destination (but we can't
1694 			 * check that since we may not have connected yet).
1695 			 * A is first hop destination, which doesn't appear in
1696 			 * actual IP option, but is stored before the options.
1697 			 */
1698 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1699 				goto bad;
1700 			m->m_len -= sizeof(struct in_addr);
1701 			cnt -= sizeof(struct in_addr);
1702 			optlen -= sizeof(struct in_addr);
1703 			cp[IPOPT_OLEN] = optlen;
1704 			/*
1705 			 * Move first hop before start of options.
1706 			 */
1707 			bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1708 			      sizeof(struct in_addr));
1709 			/*
1710 			 * Then copy rest of options back
1711 			 * to close up the deleted entry.
1712 			 */
1713 			ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1714 				&cp[IPOPT_OFFSET+1],
1715 				cnt - (IPOPT_MINOFF - 1));
1716 			break;
1717 		}
1718 	}
1719 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1720 		goto bad;
1721 	*pcbopt = m;
1722 	return (0);
1723 
1724 bad:
1725 	m_free(m);
1726 	return (EINVAL);
1727 }
1728 
1729 /*
1730  * XXX
1731  * The whole multicast option thing needs to be re-thought.
1732  * Several of these options are equally applicable to non-multicast
1733  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1734  * standard option (IP_TTL).
1735  */
1736 
1737 /*
1738  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1739  */
1740 static struct ifnet *
1741 ip_multicast_if(struct in_addr *a, int *ifindexp)
1742 {
1743 	int ifindex;
1744 	struct ifnet *ifp;
1745 
1746 	if (ifindexp)
1747 		*ifindexp = 0;
1748 	if (ntohl(a->s_addr) >> 24 == 0) {
1749 		ifindex = ntohl(a->s_addr) & 0xffffff;
1750 		if (ifindex < 0 || if_index < ifindex)
1751 			return NULL;
1752 		ifp = ifindex2ifnet[ifindex];
1753 		if (ifindexp)
1754 			*ifindexp = ifindex;
1755 	} else {
1756 		INADDR_TO_IFP(*a, ifp);
1757 	}
1758 	return ifp;
1759 }
1760 
1761 /*
1762  * Set the IP multicast options in response to user setsockopt().
1763  */
1764 static int
1765 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1766 {
1767 	int error = 0;
1768 	int i;
1769 	struct in_addr addr;
1770 	struct ip_mreq mreq;
1771 	struct ifnet *ifp;
1772 	struct ip_moptions *imo = *imop;
1773 	int ifindex;
1774 
1775 	if (imo == NULL) {
1776 		/*
1777 		 * No multicast option buffer attached to the pcb;
1778 		 * allocate one and initialize to default values.
1779 		 */
1780 		imo = malloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1781 
1782 		if (imo == NULL)
1783 			return (ENOBUFS);
1784 		*imop = imo;
1785 		imo->imo_multicast_ifp = NULL;
1786 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1787 		imo->imo_multicast_vif = -1;
1788 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1789 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1790 		imo->imo_num_memberships = 0;
1791 	}
1792 
1793 	switch (sopt->sopt_name) {
1794 	/* store an index number for the vif you wanna use in the send */
1795 	case IP_MULTICAST_VIF:
1796 		if (legal_vif_num == 0) {
1797 			error = EOPNOTSUPP;
1798 			break;
1799 		}
1800 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1801 		if (error)
1802 			break;
1803 		if (!legal_vif_num(i) && (i != -1)) {
1804 			error = EINVAL;
1805 			break;
1806 		}
1807 		imo->imo_multicast_vif = i;
1808 		break;
1809 
1810 	case IP_MULTICAST_IF:
1811 		/*
1812 		 * Select the interface for outgoing multicast packets.
1813 		 */
1814 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1815 		if (error)
1816 			break;
1817 		/*
1818 		 * INADDR_ANY is used to remove a previous selection.
1819 		 * When no interface is selected, a default one is
1820 		 * chosen every time a multicast packet is sent.
1821 		 */
1822 		if (addr.s_addr == INADDR_ANY) {
1823 			imo->imo_multicast_ifp = NULL;
1824 			break;
1825 		}
1826 		/*
1827 		 * The selected interface is identified by its local
1828 		 * IP address.  Find the interface and confirm that
1829 		 * it supports multicasting.
1830 		 */
1831 		crit_enter();
1832 		ifp = ip_multicast_if(&addr, &ifindex);
1833 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1834 			crit_exit();
1835 			error = EADDRNOTAVAIL;
1836 			break;
1837 		}
1838 		imo->imo_multicast_ifp = ifp;
1839 		if (ifindex)
1840 			imo->imo_multicast_addr = addr;
1841 		else
1842 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1843 		crit_exit();
1844 		break;
1845 
1846 	case IP_MULTICAST_TTL:
1847 		/*
1848 		 * Set the IP time-to-live for outgoing multicast packets.
1849 		 * The original multicast API required a char argument,
1850 		 * which is inconsistent with the rest of the socket API.
1851 		 * We allow either a char or an int.
1852 		 */
1853 		if (sopt->sopt_valsize == 1) {
1854 			u_char ttl;
1855 			error = sooptcopyin(sopt, &ttl, 1, 1);
1856 			if (error)
1857 				break;
1858 			imo->imo_multicast_ttl = ttl;
1859 		} else {
1860 			u_int ttl;
1861 			error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1862 			if (error)
1863 				break;
1864 			if (ttl > 255)
1865 				error = EINVAL;
1866 			else
1867 				imo->imo_multicast_ttl = ttl;
1868 		}
1869 		break;
1870 
1871 	case IP_MULTICAST_LOOP:
1872 		/*
1873 		 * Set the loopback flag for outgoing multicast packets.
1874 		 * Must be zero or one.  The original multicast API required a
1875 		 * char argument, which is inconsistent with the rest
1876 		 * of the socket API.  We allow either a char or an int.
1877 		 */
1878 		if (sopt->sopt_valsize == 1) {
1879 			u_char loop;
1880 
1881 			error = sooptcopyin(sopt, &loop, 1, 1);
1882 			if (error)
1883 				break;
1884 			imo->imo_multicast_loop = !!loop;
1885 		} else {
1886 			u_int loop;
1887 
1888 			error = sooptcopyin(sopt, &loop, sizeof loop,
1889 					    sizeof loop);
1890 			if (error)
1891 				break;
1892 			imo->imo_multicast_loop = !!loop;
1893 		}
1894 		break;
1895 
1896 	case IP_ADD_MEMBERSHIP:
1897 		/*
1898 		 * Add a multicast group membership.
1899 		 * Group must be a valid IP multicast address.
1900 		 */
1901 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1902 		if (error)
1903 			break;
1904 
1905 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1906 			error = EINVAL;
1907 			break;
1908 		}
1909 		crit_enter();
1910 		/*
1911 		 * If no interface address was provided, use the interface of
1912 		 * the route to the given multicast address.
1913 		 */
1914 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1915 			struct sockaddr_in dst;
1916 			struct rtentry *rt;
1917 
1918 			bzero(&dst, sizeof(struct sockaddr_in));
1919 			dst.sin_len = sizeof(struct sockaddr_in);
1920 			dst.sin_family = AF_INET;
1921 			dst.sin_addr = mreq.imr_multiaddr;
1922 			rt = rtlookup((struct sockaddr *)&dst);
1923 			if (rt == NULL) {
1924 				error = EADDRNOTAVAIL;
1925 				crit_exit();
1926 				break;
1927 			}
1928 			--rt->rt_refcnt;
1929 			ifp = rt->rt_ifp;
1930 		} else {
1931 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1932 		}
1933 
1934 		/*
1935 		 * See if we found an interface, and confirm that it
1936 		 * supports multicast.
1937 		 */
1938 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1939 			error = EADDRNOTAVAIL;
1940 			crit_exit();
1941 			break;
1942 		}
1943 		/*
1944 		 * See if the membership already exists or if all the
1945 		 * membership slots are full.
1946 		 */
1947 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1948 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1949 			    imo->imo_membership[i]->inm_addr.s_addr
1950 						== mreq.imr_multiaddr.s_addr)
1951 				break;
1952 		}
1953 		if (i < imo->imo_num_memberships) {
1954 			error = EADDRINUSE;
1955 			crit_exit();
1956 			break;
1957 		}
1958 		if (i == IP_MAX_MEMBERSHIPS) {
1959 			error = ETOOMANYREFS;
1960 			crit_exit();
1961 			break;
1962 		}
1963 		/*
1964 		 * Everything looks good; add a new record to the multicast
1965 		 * address list for the given interface.
1966 		 */
1967 		if ((imo->imo_membership[i] =
1968 		     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1969 			error = ENOBUFS;
1970 			crit_exit();
1971 			break;
1972 		}
1973 		++imo->imo_num_memberships;
1974 		crit_exit();
1975 		break;
1976 
1977 	case IP_DROP_MEMBERSHIP:
1978 		/*
1979 		 * Drop a multicast group membership.
1980 		 * Group must be a valid IP multicast address.
1981 		 */
1982 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1983 		if (error)
1984 			break;
1985 
1986 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1987 			error = EINVAL;
1988 			break;
1989 		}
1990 
1991 		crit_enter();
1992 		/*
1993 		 * If an interface address was specified, get a pointer
1994 		 * to its ifnet structure.
1995 		 */
1996 		if (mreq.imr_interface.s_addr == INADDR_ANY)
1997 			ifp = NULL;
1998 		else {
1999 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2000 			if (ifp == NULL) {
2001 				error = EADDRNOTAVAIL;
2002 				crit_exit();
2003 				break;
2004 			}
2005 		}
2006 		/*
2007 		 * Find the membership in the membership array.
2008 		 */
2009 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2010 			if ((ifp == NULL ||
2011 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2012 			    imo->imo_membership[i]->inm_addr.s_addr ==
2013 			    mreq.imr_multiaddr.s_addr)
2014 				break;
2015 		}
2016 		if (i == imo->imo_num_memberships) {
2017 			error = EADDRNOTAVAIL;
2018 			crit_exit();
2019 			break;
2020 		}
2021 		/*
2022 		 * Give up the multicast address record to which the
2023 		 * membership points.
2024 		 */
2025 		in_delmulti(imo->imo_membership[i]);
2026 		/*
2027 		 * Remove the gap in the membership array.
2028 		 */
2029 		for (++i; i < imo->imo_num_memberships; ++i)
2030 			imo->imo_membership[i-1] = imo->imo_membership[i];
2031 		--imo->imo_num_memberships;
2032 		crit_exit();
2033 		break;
2034 
2035 	default:
2036 		error = EOPNOTSUPP;
2037 		break;
2038 	}
2039 
2040 	/*
2041 	 * If all options have default values, no need to keep the mbuf.
2042 	 */
2043 	if (imo->imo_multicast_ifp == NULL &&
2044 	    imo->imo_multicast_vif == -1 &&
2045 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2046 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2047 	    imo->imo_num_memberships == 0) {
2048 		free(*imop, M_IPMOPTS);
2049 		*imop = NULL;
2050 	}
2051 
2052 	return (error);
2053 }
2054 
2055 /*
2056  * Return the IP multicast options in response to user getsockopt().
2057  */
2058 static int
2059 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2060 {
2061 	struct in_addr addr;
2062 	struct in_ifaddr *ia;
2063 	int error, optval;
2064 	u_char coptval;
2065 
2066 	error = 0;
2067 	switch (sopt->sopt_name) {
2068 	case IP_MULTICAST_VIF:
2069 		if (imo != NULL)
2070 			optval = imo->imo_multicast_vif;
2071 		else
2072 			optval = -1;
2073 		error = sooptcopyout(sopt, &optval, sizeof optval);
2074 		break;
2075 
2076 	case IP_MULTICAST_IF:
2077 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2078 			addr.s_addr = INADDR_ANY;
2079 		else if (imo->imo_multicast_addr.s_addr) {
2080 			/* return the value user has set */
2081 			addr = imo->imo_multicast_addr;
2082 		} else {
2083 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2084 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2085 				: IA_SIN(ia)->sin_addr.s_addr;
2086 		}
2087 		error = sooptcopyout(sopt, &addr, sizeof addr);
2088 		break;
2089 
2090 	case IP_MULTICAST_TTL:
2091 		if (imo == NULL)
2092 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2093 		else
2094 			optval = coptval = imo->imo_multicast_ttl;
2095 		if (sopt->sopt_valsize == 1)
2096 			error = sooptcopyout(sopt, &coptval, 1);
2097 		else
2098 			error = sooptcopyout(sopt, &optval, sizeof optval);
2099 		break;
2100 
2101 	case IP_MULTICAST_LOOP:
2102 		if (imo == NULL)
2103 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2104 		else
2105 			optval = coptval = imo->imo_multicast_loop;
2106 		if (sopt->sopt_valsize == 1)
2107 			error = sooptcopyout(sopt, &coptval, 1);
2108 		else
2109 			error = sooptcopyout(sopt, &optval, sizeof optval);
2110 		break;
2111 
2112 	default:
2113 		error = ENOPROTOOPT;
2114 		break;
2115 	}
2116 	return (error);
2117 }
2118 
2119 /*
2120  * Discard the IP multicast options.
2121  */
2122 void
2123 ip_freemoptions(struct ip_moptions *imo)
2124 {
2125 	int i;
2126 
2127 	if (imo != NULL) {
2128 		for (i = 0; i < imo->imo_num_memberships; ++i)
2129 			in_delmulti(imo->imo_membership[i]);
2130 		free(imo, M_IPMOPTS);
2131 	}
2132 }
2133 
2134 /*
2135  * Routine called from ip_output() to loop back a copy of an IP multicast
2136  * packet to the input queue of a specified interface.  Note that this
2137  * calls the output routine of the loopback "driver", but with an interface
2138  * pointer that might NOT be a loopback interface -- evil, but easier than
2139  * replicating that code here.
2140  */
2141 static void
2142 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2143 	     int hlen)
2144 {
2145 	struct ip *ip;
2146 	struct mbuf *copym;
2147 
2148 	copym = m_copypacket(m, MB_DONTWAIT);
2149 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2150 		copym = m_pullup(copym, hlen);
2151 	if (copym != NULL) {
2152 		/*
2153 		 * if the checksum hasn't been computed, mark it as valid
2154 		 */
2155 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2156 			in_delayed_cksum(copym);
2157 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2158 			copym->m_pkthdr.csum_flags |=
2159 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2160 			copym->m_pkthdr.csum_data = 0xffff;
2161 		}
2162 		/*
2163 		 * We don't bother to fragment if the IP length is greater
2164 		 * than the interface's MTU.  Can this possibly matter?
2165 		 */
2166 		ip = mtod(copym, struct ip *);
2167 		ip->ip_len = htons(ip->ip_len);
2168 		ip->ip_off = htons(ip->ip_off);
2169 		ip->ip_sum = 0;
2170 		if (ip->ip_vhl == IP_VHL_BORING) {
2171 			ip->ip_sum = in_cksum_hdr(ip);
2172 		} else {
2173 			ip->ip_sum = in_cksum(copym, hlen);
2174 		}
2175 		/*
2176 		 * NB:
2177 		 * It's not clear whether there are any lingering
2178 		 * reentrancy problems in other areas which might
2179 		 * be exposed by using ip_input directly (in
2180 		 * particular, everything which modifies the packet
2181 		 * in-place).  Yet another option is using the
2182 		 * protosw directly to deliver the looped back
2183 		 * packet.  For the moment, we'll err on the side
2184 		 * of safety by using if_simloop().
2185 		 */
2186 #if 1 /* XXX */
2187 		if (dst->sin_family != AF_INET) {
2188 			printf("ip_mloopback: bad address family %d\n",
2189 						dst->sin_family);
2190 			dst->sin_family = AF_INET;
2191 		}
2192 #endif
2193 
2194 #ifdef notdef
2195 		copym->m_pkthdr.rcvif = ifp;
2196 		ip_input(copym);
2197 #else
2198 		if_simloop(ifp, copym, dst->sin_family, 0);
2199 #endif
2200 	}
2201 }
2202