xref: /dflybsd-src/sys/netinet/ip_output.c (revision f7e25d559127833cbbb89b74fcf0cc036406459b)
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 1.23 2004/12/21 02:54:15 hsu Exp $
32  */
33 
34 #define _IP_VHL
35 
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipfilter.h"
40 #include "opt_ipsec.h"
41 #include "opt_random_ip_id.h"
42 #include "opt_mbuf_stress_test.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/in_cksum.h>
55 
56 #include <net/if.h>
57 #include <net/netisr.h>
58 #include <net/pfil.h>
59 #include <net/route.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 
68 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
69 
70 #ifdef IPSEC
71 #include <netinet6/ipsec.h>
72 #include <netproto/key/key.h>
73 #ifdef IPSEC_DEBUG
74 #include <netproto/key/key_debug.h>
75 #else
76 #define	KEYDEBUG(lev,arg)
77 #endif
78 #endif /*IPSEC*/
79 
80 #ifdef FAST_IPSEC
81 #include <netproto/ipsec/ipsec.h>
82 #include <netproto/ipsec/xform.h>
83 #include <netproto/ipsec/key.h>
84 #endif /*FAST_IPSEC*/
85 
86 #include <net/ipfw/ip_fw.h>
87 #include <net/dummynet/ip_dummynet.h>
88 
89 #define print_ip(x, a, y)	 printf("%s %d.%d.%d.%d%s",\
90 				x, (ntohl(a.s_addr)>>24)&0xFF,\
91 				  (ntohl(a.s_addr)>>16)&0xFF,\
92 				  (ntohl(a.s_addr)>>8)&0xFF,\
93 				  (ntohl(a.s_addr))&0xFF, y);
94 
95 u_short ip_id;
96 
97 #ifdef MBUF_STRESS_TEST
98 int mbuf_frag_size = 0;
99 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
100 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
101 #endif
102 
103 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
104 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
105 static void	ip_mloopback
106 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
107 static int	ip_getmoptions
108 	(struct sockopt *, struct ip_moptions *);
109 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
110 static int	ip_setmoptions
111 	(struct sockopt *, struct ip_moptions **);
112 
113 int	ip_optcopy(struct ip *, struct ip *);
114 
115 
116 extern	struct protosw inetsw[];
117 
118 /*
119  * IP output.  The packet in mbuf chain m contains a skeletal IP
120  * header (with len, off, ttl, proto, tos, src, dst).
121  * The mbuf chain containing the packet will be freed.
122  * The mbuf opt, if present, will not be freed.
123  */
124 int
125 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
126 	int flags, struct ip_moptions *imo, struct inpcb *inp)
127 {
128 	struct ip *ip;
129 	struct ifnet *ifp = NULL;	/* keep compiler happy */
130 	struct mbuf *m;
131 	int hlen = sizeof (struct ip);
132 	int len, off, error = 0;
133 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
134 	struct in_ifaddr *ia = NULL;
135 	int isbroadcast, sw_csum;
136 	struct in_addr pkt_dst;
137 	struct route iproute;
138 #ifdef IPSEC
139 	struct secpolicy *sp = NULL;
140 	struct socket *so = inp ? inp->inp_socket : NULL;
141 #endif
142 #ifdef FAST_IPSEC
143 	struct m_tag *mtag;
144 	struct secpolicy *sp = NULL;
145 	struct tdb_ident *tdbi;
146 	int s;
147 #endif /* FAST_IPSEC */
148 	struct ip_fw_args args;
149 	int src_was_INADDR_ANY = 0;	/* as the name says... */
150 
151 	args.eh = NULL;
152 	args.rule = NULL;
153 	args.next_hop = NULL;
154 	args.divert_rule = 0;			/* divert cookie */
155 
156 	/* Grab info from MT_TAG mbufs prepended to the chain. */
157 	for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
158 		switch(m0->_m_tag_id) {
159 		default:
160 			printf("ip_output: unrecognised MT_TAG tag %d\n",
161 			    m0->_m_tag_id);
162 			break;
163 
164 		case PACKET_TAG_DUMMYNET:
165 			/*
166 			 * the packet was already tagged, so part of the
167 			 * processing was already done, and we need to go down.
168 			 * Get parameters from the header.
169 			 */
170 			args.rule = ((struct dn_pkt *)m0)->rule;
171 			opt = NULL ;
172 			ro = & ( ((struct dn_pkt *)m0)->ro ) ;
173 			imo = NULL ;
174 			dst = ((struct dn_pkt *)m0)->dn_dst ;
175 			ifp = ((struct dn_pkt *)m0)->ifp ;
176 			flags = ((struct dn_pkt *)m0)->flags ;
177 			break;
178 
179 		case PACKET_TAG_DIVERT:
180 			args.divert_rule = (int)m0->m_data & 0xffff;
181 			break;
182 
183 		case PACKET_TAG_IPFORWARD:
184 			args.next_hop = (struct sockaddr_in *)m0->m_data;
185 			break;
186 		}
187 	}
188 	m = m0;
189 
190 	KASSERT(!m || (m->m_flags & M_PKTHDR), ("ip_output: no HDR"));
191 
192 	if (ro == NULL) {
193 		ro = &iproute;
194 		bzero(ro, sizeof(*ro));
195 	}
196 
197 	if (args.rule != NULL) {	/* dummynet already saw us */
198 		ip = mtod(m, struct ip *);
199 		hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
200 		if (ro->ro_rt)
201 			ia = ifatoia(ro->ro_rt->rt_ifa);
202 		goto sendit;
203 	}
204 
205 	if (opt) {
206 		len = 0;
207 		m = ip_insertoptions(m, opt, &len);
208 		if (len != 0)
209 			hlen = len;
210 	}
211 	ip = mtod(m, struct ip *);
212 	pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
213 
214 	/*
215 	 * Fill in IP header.
216 	 */
217 	if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
218 		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
219 		ip->ip_off &= IP_DF;
220 #ifdef RANDOM_IP_ID
221 		ip->ip_id = ip_randomid();
222 #else
223 		ip->ip_id = htons(ip_id++);
224 #endif
225 		ipstat.ips_localout++;
226 	} else {
227 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
228 	}
229 
230 	dst = (struct sockaddr_in *)&ro->ro_dst;
231 	/*
232 	 * If there is a cached route,
233 	 * check that it is to the same destination
234 	 * and is still up.  If not, free it and try again.
235 	 * The address family should also be checked in case of sharing the
236 	 * cache with IPv6.
237 	 */
238 	if (ro->ro_rt &&
239 	    (!(ro->ro_rt->rt_flags & RTF_UP) ||
240 	     dst->sin_family != AF_INET ||
241 	     dst->sin_addr.s_addr != pkt_dst.s_addr)) {
242 		RTFREE(ro->ro_rt);
243 		ro->ro_rt = (struct rtentry *)NULL;
244 	}
245 	if (ro->ro_rt == NULL) {
246 		bzero(dst, sizeof(*dst));
247 		dst->sin_family = AF_INET;
248 		dst->sin_len = sizeof(*dst);
249 		dst->sin_addr = pkt_dst;
250 	}
251 	/*
252 	 * If routing to interface only,
253 	 * short circuit routing lookup.
254 	 */
255 	if (flags & IP_ROUTETOIF) {
256 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
257 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
258 			ipstat.ips_noroute++;
259 			error = ENETUNREACH;
260 			goto bad;
261 		}
262 		ifp = ia->ia_ifp;
263 		ip->ip_ttl = 1;
264 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
265 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
266 		   imo != NULL && imo->imo_multicast_ifp != NULL) {
267 		/*
268 		 * Bypass the normal routing lookup for multicast
269 		 * packets if the interface is specified.
270 		 */
271 		ifp = imo->imo_multicast_ifp;
272 		IFP_TO_IA(ifp, ia);
273 		isbroadcast = 0;	/* fool gcc */
274 	} else {
275 		/*
276 		 * If this is the case, we probably don't want to allocate
277 		 * a protocol-cloned route since we didn't get one from the
278 		 * ULP.  This lets TCP do its thing, while not burdening
279 		 * forwarding or ICMP with the overhead of cloning a route.
280 		 * Of course, we still want to do any cloning requested by
281 		 * the link layer, as this is probably required in all cases
282 		 * for correct operation (as it is for ARP).
283 		 */
284 		if (ro->ro_rt == NULL)
285 			rtalloc_ign(ro, RTF_PRCLONING);
286 		if (ro->ro_rt == NULL) {
287 			ipstat.ips_noroute++;
288 			error = EHOSTUNREACH;
289 			goto bad;
290 		}
291 		ia = ifatoia(ro->ro_rt->rt_ifa);
292 		ifp = ro->ro_rt->rt_ifp;
293 		ro->ro_rt->rt_use++;
294 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
295 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
296 		if (ro->ro_rt->rt_flags & RTF_HOST)
297 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
298 		else
299 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
300 	}
301 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
302 		struct in_multi *inm;
303 
304 		m->m_flags |= M_MCAST;
305 		/*
306 		 * IP destination address is multicast.  Make sure "dst"
307 		 * still points to the address in "ro".  (It may have been
308 		 * changed to point to a gateway address, above.)
309 		 */
310 		dst = (struct sockaddr_in *)&ro->ro_dst;
311 		/*
312 		 * See if the caller provided any multicast options
313 		 */
314 		if (imo != NULL) {
315 			ip->ip_ttl = imo->imo_multicast_ttl;
316 			if (imo->imo_multicast_vif != -1)
317 				ip->ip_src.s_addr =
318 				    ip_mcast_src ?
319 				    ip_mcast_src(imo->imo_multicast_vif) :
320 				    INADDR_ANY;
321 		} else
322 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
323 		/*
324 		 * Confirm that the outgoing interface supports multicast.
325 		 */
326 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
327 			if (!(ifp->if_flags & IFF_MULTICAST)) {
328 				ipstat.ips_noroute++;
329 				error = ENETUNREACH;
330 				goto bad;
331 			}
332 		}
333 		/*
334 		 * If source address not specified yet, use address
335 		 * of outgoing interface.
336 		 */
337 		if (ip->ip_src.s_addr == INADDR_ANY) {
338 			/* Interface may have no addresses. */
339 			if (ia != NULL)
340 				ip->ip_src = IA_SIN(ia)->sin_addr;
341 		}
342 
343 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
344 		if (inm != NULL &&
345 		   (imo == NULL || imo->imo_multicast_loop)) {
346 			/*
347 			 * If we belong to the destination multicast group
348 			 * on the outgoing interface, and the caller did not
349 			 * forbid loopback, loop back a copy.
350 			 */
351 			ip_mloopback(ifp, m, dst, hlen);
352 		}
353 		else {
354 			/*
355 			 * If we are acting as a multicast router, perform
356 			 * multicast forwarding as if the packet had just
357 			 * arrived on the interface to which we are about
358 			 * to send.  The multicast forwarding function
359 			 * recursively calls this function, using the
360 			 * IP_FORWARDING flag to prevent infinite recursion.
361 			 *
362 			 * Multicasts that are looped back by ip_mloopback(),
363 			 * above, will be forwarded by the ip_input() routine,
364 			 * if necessary.
365 			 */
366 			if (ip_mrouter && !(flags & IP_FORWARDING)) {
367 				/*
368 				 * If rsvp daemon is not running, do not
369 				 * set ip_moptions. This ensures that the packet
370 				 * is multicast and not just sent down one link
371 				 * as prescribed by rsvpd.
372 				 */
373 				if (!rsvp_on)
374 					imo = NULL;
375 				if (ip_mforward &&
376 				    ip_mforward(ip, ifp, m, imo) != 0) {
377 					m_freem(m);
378 					goto done;
379 				}
380 			}
381 		}
382 
383 		/*
384 		 * Multicasts with a time-to-live of zero may be looped-
385 		 * back, above, but must not be transmitted on a network.
386 		 * Also, multicasts addressed to the loopback interface
387 		 * are not sent -- the above call to ip_mloopback() will
388 		 * loop back a copy if this host actually belongs to the
389 		 * destination group on the loopback interface.
390 		 */
391 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
392 			m_freem(m);
393 			goto done;
394 		}
395 
396 		goto sendit;
397 	}
398 #ifndef notdef
399 	/*
400 	 * If the source address is not specified yet, use the address
401 	 * of the outoing interface. In case, keep note we did that, so
402 	 * if the the firewall changes the next-hop causing the output
403 	 * interface to change, we can fix that.
404 	 */
405 	if (ip->ip_src.s_addr == INADDR_ANY) {
406 		/* Interface may have no addresses. */
407 		if (ia != NULL) {
408 			ip->ip_src = IA_SIN(ia)->sin_addr;
409 			src_was_INADDR_ANY = 1;
410 		}
411 	}
412 #endif /* notdef */
413 	/*
414 	 * Verify that we have any chance at all of being able to queue
415 	 *      the packet or packet fragments
416 	 */
417 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
418 		ifp->if_snd.ifq_maxlen) {
419 			error = ENOBUFS;
420 			ipstat.ips_odropped++;
421 			goto bad;
422 	}
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 	s = splnet();
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 				splx(s);
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 		splx(s);
664 		goto done;
665 	} else {
666 		splx(s);
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, args.divert_rule);
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 
900 			ro_fwd->ro_rt = NULL;
901 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
902 
903 			if (ro_fwd->ro_rt == NULL) {
904 				ipstat.ips_noroute++;
905 				error = EHOSTUNREACH;
906 				goto bad;
907 			}
908 
909 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
910 			ifp = ro_fwd->ro_rt->rt_ifp;
911 			ro_fwd->ro_rt->rt_use++;
912 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
913 				dst = (struct sockaddr_in *)
914 					ro_fwd->ro_rt->rt_gateway;
915 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
916 				isbroadcast =
917 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
918 			else
919 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
920 			if (ro->ro_rt)
921 				RTFREE(ro->ro_rt);
922 			ro->ro_rt = ro_fwd->ro_rt;
923 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
924 
925 #endif	/* ... block to be put into a function */
926 			/*
927 			 * If we added a default src ip earlier,
928 			 * which would have been gotten from the-then
929 			 * interface, do it again, from the new one.
930 			 */
931 			if (src_was_INADDR_ANY)
932 				ip->ip_src = IA_SIN(ia)->sin_addr;
933 			goto pass ;
934 		}
935 
936 		/*
937 		 * if we get here, none of the above matches, and
938 		 * we have to drop the pkt
939 		 */
940 		m_freem(m);
941 		error = EACCES; /* not sure this is the right error msg */
942 		goto done;
943 	}
944 
945 pass:
946 	/* 127/8 must not appear on wire - RFC1122. */
947 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
948 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
949 		if (!(ifp->if_flags & IFF_LOOPBACK)) {
950 			ipstat.ips_badaddr++;
951 			error = EADDRNOTAVAIL;
952 			goto bad;
953 		}
954 	}
955 
956 	m->m_pkthdr.csum_flags |= CSUM_IP;
957 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
958 	if (sw_csum & CSUM_DELAY_DATA) {
959 		in_delayed_cksum(m);
960 		sw_csum &= ~CSUM_DELAY_DATA;
961 	}
962 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
963 
964 	/*
965 	 * If small enough for interface, or the interface will take
966 	 * care of the fragmentation for us, can just send directly.
967 	 */
968 	if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
969 	    !(ip->ip_off & IP_DF))) {
970 		ip->ip_len = htons(ip->ip_len);
971 		ip->ip_off = htons(ip->ip_off);
972 		ip->ip_sum = 0;
973 		if (sw_csum & CSUM_DELAY_IP) {
974 			if (ip->ip_vhl == IP_VHL_BORING) {
975 				ip->ip_sum = in_cksum_hdr(ip);
976 			} else {
977 				ip->ip_sum = in_cksum(m, hlen);
978 			}
979 		}
980 
981 		/* Record statistics for this interface address. */
982 		if (!(flags & IP_FORWARDING) && ia) {
983 			ia->ia_ifa.if_opackets++;
984 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
985 		}
986 
987 #ifdef IPSEC
988 		/* clean ipsec history once it goes out of the node */
989 		ipsec_delaux(m);
990 #endif
991 
992 #ifdef MBUF_STRESS_TEST
993 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
994 			struct mbuf *m1, *m2;
995 			int length, tmp;
996 
997 			tmp = length = m->m_pkthdr.len;
998 
999 			while ((length -= mbuf_frag_size) >= 1) {
1000 				m1 = m_split(m, length, MB_DONTWAIT);
1001 				if (m1 == NULL)
1002 					break;
1003 				m1->m_flags &= ~M_PKTHDR;
1004 				m2 = m;
1005 				while (m2->m_next != NULL)
1006 					m2 = m2->m_next;
1007 				m2->m_next = m1;
1008 			}
1009 			m->m_pkthdr.len = tmp;
1010 		}
1011 #endif
1012 		error = (*ifp->if_output)(ifp, m,
1013 				(struct sockaddr *)dst, ro->ro_rt);
1014 		goto done;
1015 	}
1016 
1017 	if (ip->ip_off & IP_DF) {
1018 		error = EMSGSIZE;
1019 		/*
1020 		 * This case can happen if the user changed the MTU
1021 		 * of an interface after enabling IP on it.  Because
1022 		 * most netifs don't keep track of routes pointing to
1023 		 * them, there is no way for one to update all its
1024 		 * routes when the MTU is changed.
1025 		 */
1026 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1027 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1028 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1029 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1030 		}
1031 		ipstat.ips_cantfrag++;
1032 		goto bad;
1033 	}
1034 
1035 	/*
1036 	 * Too large for interface; fragment if possible. If successful,
1037 	 * on return, m will point to a list of packets to be sent.
1038 	 */
1039 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1040 	if (error)
1041 		goto bad;
1042 	for (; m; m = m0) {
1043 		m0 = m->m_nextpkt;
1044 		m->m_nextpkt = NULL;
1045 #ifdef IPSEC
1046 		/* clean ipsec history once it goes out of the node */
1047 		ipsec_delaux(m);
1048 #endif
1049 		if (error == 0) {
1050 			/* Record statistics for this interface address. */
1051 			if (ia != NULL) {
1052 				ia->ia_ifa.if_opackets++;
1053 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1054 			}
1055 
1056 			error = (*ifp->if_output)(ifp, m,
1057 			    (struct sockaddr *)dst, ro->ro_rt);
1058 		} else
1059 			m_freem(m);
1060 	}
1061 
1062 	if (error == 0)
1063 		ipstat.ips_fragmented++;
1064 
1065 done:
1066 	if (ro == &iproute && ro->ro_rt) {
1067 		RTFREE(ro->ro_rt);
1068 		ro->ro_rt = NULL;
1069 	}
1070 #ifdef IPSEC
1071 	if (sp != NULL) {
1072 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1073 			printf("DP ip_output call free SP:%p\n", sp));
1074 		key_freesp(sp);
1075 	}
1076 #endif
1077 #ifdef FAST_IPSEC
1078 	if (sp != NULL)
1079 		KEY_FREESP(&sp);
1080 #endif
1081 	return (error);
1082 bad:
1083 	m_freem(m);
1084 	goto done;
1085 }
1086 
1087 /*
1088  * Create a chain of fragments which fit the given mtu. m_frag points to the
1089  * mbuf to be fragmented; on return it points to the chain with the fragments.
1090  * Return 0 if no error. If error, m_frag may contain a partially built
1091  * chain of fragments that should be freed by the caller.
1092  *
1093  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1094  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1095  */
1096 int
1097 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1098 	    u_long if_hwassist_flags, int sw_csum)
1099 {
1100 	int error = 0;
1101 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1102 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1103 	int off;
1104 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1105 	int firstlen;
1106 	struct mbuf **mnext;
1107 	int nfrags;
1108 
1109 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1110 		ipstat.ips_cantfrag++;
1111 		return EMSGSIZE;
1112 	}
1113 
1114 	/*
1115 	 * Must be able to put at least 8 bytes per fragment.
1116 	 */
1117 	if (len < 8)
1118 		return EMSGSIZE;
1119 
1120 	/*
1121 	 * If the interface will not calculate checksums on
1122 	 * fragmented packets, then do it here.
1123 	 */
1124 	if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1125 	    !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1126 		in_delayed_cksum(m0);
1127 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1128 	}
1129 
1130 	if (len > PAGE_SIZE) {
1131 		/*
1132 		 * Fragment large datagrams such that each segment
1133 		 * contains a multiple of PAGE_SIZE amount of data,
1134 		 * plus headers. This enables a receiver to perform
1135 		 * page-flipping zero-copy optimizations.
1136 		 *
1137 		 * XXX When does this help given that sender and receiver
1138 		 * could have different page sizes, and also mtu could
1139 		 * be less than the receiver's page size ?
1140 		 */
1141 		int newlen;
1142 		struct mbuf *m;
1143 
1144 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1145 			off += m->m_len;
1146 
1147 		/*
1148 		 * firstlen (off - hlen) must be aligned on an
1149 		 * 8-byte boundary
1150 		 */
1151 		if (off < hlen)
1152 			goto smart_frag_failure;
1153 		off = ((off - hlen) & ~7) + hlen;
1154 		newlen = (~PAGE_MASK) & mtu;
1155 		if ((newlen + sizeof (struct ip)) > mtu) {
1156 			/* we failed, go back the default */
1157 smart_frag_failure:
1158 			newlen = len;
1159 			off = hlen + len;
1160 		}
1161 		len = newlen;
1162 
1163 	} else {
1164 		off = hlen + len;
1165 	}
1166 
1167 	firstlen = off - hlen;
1168 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1169 
1170 	/*
1171 	 * Loop through length of segment after first fragment,
1172 	 * make new header and copy data of each part and link onto chain.
1173 	 * Here, m0 is the original packet, m is the fragment being created.
1174 	 * The fragments are linked off the m_nextpkt of the original
1175 	 * packet, which after processing serves as the first fragment.
1176 	 */
1177 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1178 		struct ip *mhip;	/* ip header on the fragment */
1179 		struct mbuf *m;
1180 		int mhlen = sizeof (struct ip);
1181 
1182 		MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1183 		if (m == NULL) {
1184 			error = ENOBUFS;
1185 			ipstat.ips_odropped++;
1186 			goto done;
1187 		}
1188 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1189 		/*
1190 		 * In the first mbuf, leave room for the link header, then
1191 		 * copy the original IP header including options. The payload
1192 		 * goes into an additional mbuf chain returned by m_copy().
1193 		 */
1194 		m->m_data += max_linkhdr;
1195 		mhip = mtod(m, struct ip *);
1196 		*mhip = *ip;
1197 		if (hlen > sizeof (struct ip)) {
1198 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1199 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1200 		}
1201 		m->m_len = mhlen;
1202 		/* XXX do we need to add ip->ip_off below ? */
1203 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1204 		if (off + len >= ip->ip_len) {	/* last fragment */
1205 			len = ip->ip_len - off;
1206 			m->m_flags |= M_LASTFRAG;
1207 		} else
1208 			mhip->ip_off |= IP_MF;
1209 		mhip->ip_len = htons((u_short)(len + mhlen));
1210 		m->m_next = m_copy(m0, off, len);
1211 		if (m->m_next == NULL) {		/* copy failed */
1212 			m_free(m);
1213 			error = ENOBUFS;	/* ??? */
1214 			ipstat.ips_odropped++;
1215 			goto done;
1216 		}
1217 		m->m_pkthdr.len = mhlen + len;
1218 		m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1219 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1220 		mhip->ip_off = htons(mhip->ip_off);
1221 		mhip->ip_sum = 0;
1222 		if (sw_csum & CSUM_DELAY_IP)
1223 			mhip->ip_sum = in_cksum(m, mhlen);
1224 		*mnext = m;
1225 		mnext = &m->m_nextpkt;
1226 	}
1227 	ipstat.ips_ofragments += nfrags;
1228 
1229 	/* set first marker for fragment chain */
1230 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1231 	m0->m_pkthdr.csum_data = nfrags;
1232 
1233 	/*
1234 	 * Update first fragment by trimming what's been copied out
1235 	 * and updating header.
1236 	 */
1237 	m_adj(m0, hlen + firstlen - ip->ip_len);
1238 	m0->m_pkthdr.len = hlen + firstlen;
1239 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1240 	ip->ip_off |= IP_MF;
1241 	ip->ip_off = htons(ip->ip_off);
1242 	ip->ip_sum = 0;
1243 	if (sw_csum & CSUM_DELAY_IP)
1244 		ip->ip_sum = in_cksum(m0, hlen);
1245 
1246 done:
1247 	*m_frag = m0;
1248 	return error;
1249 }
1250 
1251 void
1252 in_delayed_cksum(struct mbuf *m)
1253 {
1254 	struct ip *ip;
1255 	u_short csum, offset;
1256 
1257 	ip = mtod(m, struct ip *);
1258 	offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1259 	csum = in_cksum_skip(m, ip->ip_len, offset);
1260 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1261 		csum = 0xffff;
1262 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1263 
1264 	if (offset + sizeof(u_short) > m->m_len) {
1265 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1266 		    m->m_len, offset, ip->ip_p);
1267 		/*
1268 		 * XXX
1269 		 * this shouldn't happen, but if it does, the
1270 		 * correct behavior may be to insert the checksum
1271 		 * in the existing chain instead of rearranging it.
1272 		 */
1273 		m = m_pullup(m, offset + sizeof(u_short));
1274 	}
1275 	*(u_short *)(m->m_data + offset) = csum;
1276 }
1277 
1278 /*
1279  * Insert IP options into preformed packet.
1280  * Adjust IP destination as required for IP source routing,
1281  * as indicated by a non-zero in_addr at the start of the options.
1282  *
1283  * XXX This routine assumes that the packet has no options in place.
1284  */
1285 static struct mbuf *
1286 ip_insertoptions(m, opt, phlen)
1287 	struct mbuf *m;
1288 	struct mbuf *opt;
1289 	int *phlen;
1290 {
1291 	struct ipoption *p = mtod(opt, struct ipoption *);
1292 	struct mbuf *n;
1293 	struct ip *ip = mtod(m, struct ip *);
1294 	unsigned optlen;
1295 
1296 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1297 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1298 		*phlen = 0;
1299 		return (m);		/* XXX should fail */
1300 	}
1301 	if (p->ipopt_dst.s_addr)
1302 		ip->ip_dst = p->ipopt_dst;
1303 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1304 		MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1305 		if (n == NULL) {
1306 			*phlen = 0;
1307 			return (m);
1308 		}
1309 		n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1310 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1311 		m->m_len -= sizeof(struct ip);
1312 		m->m_data += sizeof(struct ip);
1313 		n->m_next = m;
1314 		m = n;
1315 		m->m_len = optlen + sizeof(struct ip);
1316 		m->m_data += max_linkhdr;
1317 		memcpy(mtod(m, void *), ip, sizeof(struct ip));
1318 	} else {
1319 		m->m_data -= optlen;
1320 		m->m_len += optlen;
1321 		m->m_pkthdr.len += optlen;
1322 		ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1323 	}
1324 	ip = mtod(m, struct ip *);
1325 	bcopy(p->ipopt_list, ip + 1, optlen);
1326 	*phlen = sizeof(struct ip) + optlen;
1327 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1328 	ip->ip_len += optlen;
1329 	return (m);
1330 }
1331 
1332 /*
1333  * Copy options from ip to jp,
1334  * omitting those not copied during fragmentation.
1335  */
1336 int
1337 ip_optcopy(ip, jp)
1338 	struct ip *ip, *jp;
1339 {
1340 	u_char *cp, *dp;
1341 	int opt, optlen, cnt;
1342 
1343 	cp = (u_char *)(ip + 1);
1344 	dp = (u_char *)(jp + 1);
1345 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1346 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1347 		opt = cp[0];
1348 		if (opt == IPOPT_EOL)
1349 			break;
1350 		if (opt == IPOPT_NOP) {
1351 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1352 			*dp++ = IPOPT_NOP;
1353 			optlen = 1;
1354 			continue;
1355 		}
1356 
1357 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1358 		    ("ip_optcopy: malformed ipv4 option"));
1359 		optlen = cp[IPOPT_OLEN];
1360 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1361 		    ("ip_optcopy: malformed ipv4 option"));
1362 
1363 		/* bogus lengths should have been caught by ip_dooptions */
1364 		if (optlen > cnt)
1365 			optlen = cnt;
1366 		if (IPOPT_COPIED(opt)) {
1367 			bcopy(cp, dp, optlen);
1368 			dp += optlen;
1369 		}
1370 	}
1371 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1372 		*dp++ = IPOPT_EOL;
1373 	return (optlen);
1374 }
1375 
1376 /*
1377  * IP socket option processing.
1378  */
1379 int
1380 ip_ctloutput(so, sopt)
1381 	struct socket *so;
1382 	struct sockopt *sopt;
1383 {
1384 	struct	inpcb *inp = sotoinpcb(so);
1385 	int	error, optval;
1386 
1387 	error = optval = 0;
1388 	if (sopt->sopt_level != IPPROTO_IP) {
1389 		return (EINVAL);
1390 	}
1391 
1392 	switch (sopt->sopt_dir) {
1393 	case SOPT_SET:
1394 		switch (sopt->sopt_name) {
1395 		case IP_OPTIONS:
1396 #ifdef notyet
1397 		case IP_RETOPTS:
1398 #endif
1399 		{
1400 			struct mbuf *m;
1401 			if (sopt->sopt_valsize > MLEN) {
1402 				error = EMSGSIZE;
1403 				break;
1404 			}
1405 			MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1406 			if (m == NULL) {
1407 				error = ENOBUFS;
1408 				break;
1409 			}
1410 			m->m_len = sopt->sopt_valsize;
1411 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1412 					    m->m_len);
1413 
1414 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1415 					   m));
1416 		}
1417 
1418 		case IP_TOS:
1419 		case IP_TTL:
1420 		case IP_RECVOPTS:
1421 		case IP_RECVRETOPTS:
1422 		case IP_RECVDSTADDR:
1423 		case IP_RECVIF:
1424 		case IP_FAITH:
1425 			error = sooptcopyin(sopt, &optval, sizeof optval,
1426 					    sizeof optval);
1427 			if (error)
1428 				break;
1429 
1430 			switch (sopt->sopt_name) {
1431 			case IP_TOS:
1432 				inp->inp_ip_tos = optval;
1433 				break;
1434 
1435 			case IP_TTL:
1436 				inp->inp_ip_ttl = optval;
1437 				break;
1438 #define	OPTSET(bit) \
1439 	if (optval) \
1440 		inp->inp_flags |= bit; \
1441 	else \
1442 		inp->inp_flags &= ~bit;
1443 
1444 			case IP_RECVOPTS:
1445 				OPTSET(INP_RECVOPTS);
1446 				break;
1447 
1448 			case IP_RECVRETOPTS:
1449 				OPTSET(INP_RECVRETOPTS);
1450 				break;
1451 
1452 			case IP_RECVDSTADDR:
1453 				OPTSET(INP_RECVDSTADDR);
1454 				break;
1455 
1456 			case IP_RECVIF:
1457 				OPTSET(INP_RECVIF);
1458 				break;
1459 
1460 			case IP_FAITH:
1461 				OPTSET(INP_FAITH);
1462 				break;
1463 			}
1464 			break;
1465 #undef OPTSET
1466 
1467 		case IP_MULTICAST_IF:
1468 		case IP_MULTICAST_VIF:
1469 		case IP_MULTICAST_TTL:
1470 		case IP_MULTICAST_LOOP:
1471 		case IP_ADD_MEMBERSHIP:
1472 		case IP_DROP_MEMBERSHIP:
1473 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1474 			break;
1475 
1476 		case IP_PORTRANGE:
1477 			error = sooptcopyin(sopt, &optval, sizeof optval,
1478 					    sizeof optval);
1479 			if (error)
1480 				break;
1481 
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 			break;
1503 
1504 #if defined(IPSEC) || defined(FAST_IPSEC)
1505 		case IP_IPSEC_POLICY:
1506 		{
1507 			caddr_t req;
1508 			size_t len = 0;
1509 			int priv;
1510 			struct mbuf *m;
1511 			int optname;
1512 
1513 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1514 				break;
1515 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1516 				break;
1517 			priv = (sopt->sopt_td != NULL &&
1518 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1519 			req = mtod(m, caddr_t);
1520 			len = m->m_len;
1521 			optname = sopt->sopt_name;
1522 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1523 			m_freem(m);
1524 			break;
1525 		}
1526 #endif /*IPSEC*/
1527 
1528 		default:
1529 			error = ENOPROTOOPT;
1530 			break;
1531 		}
1532 		break;
1533 
1534 	case SOPT_GET:
1535 		switch (sopt->sopt_name) {
1536 		case IP_OPTIONS:
1537 		case IP_RETOPTS:
1538 			if (inp->inp_options)
1539 				error = sooptcopyout(sopt,
1540 						     mtod(inp->inp_options,
1541 							  char *),
1542 						     inp->inp_options->m_len);
1543 			else
1544 				sopt->sopt_valsize = 0;
1545 			break;
1546 
1547 		case IP_TOS:
1548 		case IP_TTL:
1549 		case IP_RECVOPTS:
1550 		case IP_RECVRETOPTS:
1551 		case IP_RECVDSTADDR:
1552 		case IP_RECVIF:
1553 		case IP_PORTRANGE:
1554 		case IP_FAITH:
1555 			switch (sopt->sopt_name) {
1556 
1557 			case IP_TOS:
1558 				optval = inp->inp_ip_tos;
1559 				break;
1560 
1561 			case IP_TTL:
1562 				optval = inp->inp_ip_ttl;
1563 				break;
1564 
1565 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1566 
1567 			case IP_RECVOPTS:
1568 				optval = OPTBIT(INP_RECVOPTS);
1569 				break;
1570 
1571 			case IP_RECVRETOPTS:
1572 				optval = OPTBIT(INP_RECVRETOPTS);
1573 				break;
1574 
1575 			case IP_RECVDSTADDR:
1576 				optval = OPTBIT(INP_RECVDSTADDR);
1577 				break;
1578 
1579 			case IP_RECVIF:
1580 				optval = OPTBIT(INP_RECVIF);
1581 				break;
1582 
1583 			case IP_PORTRANGE:
1584 				if (inp->inp_flags & INP_HIGHPORT)
1585 					optval = IP_PORTRANGE_HIGH;
1586 				else if (inp->inp_flags & INP_LOWPORT)
1587 					optval = IP_PORTRANGE_LOW;
1588 				else
1589 					optval = 0;
1590 				break;
1591 
1592 			case IP_FAITH:
1593 				optval = OPTBIT(INP_FAITH);
1594 				break;
1595 			}
1596 			error = sooptcopyout(sopt, &optval, sizeof optval);
1597 			break;
1598 
1599 		case IP_MULTICAST_IF:
1600 		case IP_MULTICAST_VIF:
1601 		case IP_MULTICAST_TTL:
1602 		case IP_MULTICAST_LOOP:
1603 		case IP_ADD_MEMBERSHIP:
1604 		case IP_DROP_MEMBERSHIP:
1605 			error = ip_getmoptions(sopt, inp->inp_moptions);
1606 			break;
1607 
1608 #if defined(IPSEC) || defined(FAST_IPSEC)
1609 		case IP_IPSEC_POLICY:
1610 		{
1611 			struct mbuf *m = NULL;
1612 			caddr_t req = NULL;
1613 			size_t len = 0;
1614 
1615 			if (m != NULL) {
1616 				req = mtod(m, caddr_t);
1617 				len = m->m_len;
1618 			}
1619 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1620 			if (error == 0)
1621 				error = soopt_mcopyout(sopt, m); /* XXX */
1622 			if (error == 0)
1623 				m_freem(m);
1624 			break;
1625 		}
1626 #endif /*IPSEC*/
1627 
1628 		default:
1629 			error = ENOPROTOOPT;
1630 			break;
1631 		}
1632 		break;
1633 	}
1634 	return (error);
1635 }
1636 
1637 /*
1638  * Set up IP options in pcb for insertion in output packets.
1639  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1640  * with destination address if source routed.
1641  */
1642 static int
1643 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1644 {
1645 	int cnt, optlen;
1646 	u_char *cp;
1647 	u_char opt;
1648 
1649 	/* turn off any old options */
1650 	if (*pcbopt)
1651 		m_free(*pcbopt);
1652 	*pcbopt = 0;
1653 	if (m == NULL || m->m_len == 0) {
1654 		/*
1655 		 * Only turning off any previous options.
1656 		 */
1657 		if (m != NULL)
1658 			m_free(m);
1659 		return (0);
1660 	}
1661 
1662 	if (m->m_len % sizeof(int32_t))
1663 		goto bad;
1664 	/*
1665 	 * IP first-hop destination address will be stored before
1666 	 * actual options; move other options back
1667 	 * and clear it when none present.
1668 	 */
1669 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1670 		goto bad;
1671 	cnt = m->m_len;
1672 	m->m_len += sizeof(struct in_addr);
1673 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1674 	ovbcopy(mtod(m, caddr_t), cp, cnt);
1675 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1676 
1677 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1678 		opt = cp[IPOPT_OPTVAL];
1679 		if (opt == IPOPT_EOL)
1680 			break;
1681 		if (opt == IPOPT_NOP)
1682 			optlen = 1;
1683 		else {
1684 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1685 				goto bad;
1686 			optlen = cp[IPOPT_OLEN];
1687 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1688 				goto bad;
1689 		}
1690 		switch (opt) {
1691 
1692 		default:
1693 			break;
1694 
1695 		case IPOPT_LSRR:
1696 		case IPOPT_SSRR:
1697 			/*
1698 			 * user process specifies route as:
1699 			 *	->A->B->C->D
1700 			 * D must be our final destination (but we can't
1701 			 * check that since we may not have connected yet).
1702 			 * A is first hop destination, which doesn't appear in
1703 			 * actual IP option, but is stored before the options.
1704 			 */
1705 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1706 				goto bad;
1707 			m->m_len -= sizeof(struct in_addr);
1708 			cnt -= sizeof(struct in_addr);
1709 			optlen -= sizeof(struct in_addr);
1710 			cp[IPOPT_OLEN] = optlen;
1711 			/*
1712 			 * Move first hop before start of options.
1713 			 */
1714 			bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1715 			    sizeof(struct in_addr));
1716 			/*
1717 			 * Then copy rest of options back
1718 			 * to close up the deleted entry.
1719 			 */
1720 			ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1721 				&cp[IPOPT_OFFSET+1],
1722 				cnt - (IPOPT_MINOFF - 1));
1723 			break;
1724 		}
1725 	}
1726 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1727 		goto bad;
1728 	*pcbopt = m;
1729 	return (0);
1730 
1731 bad:
1732 	m_free(m);
1733 	return (EINVAL);
1734 }
1735 
1736 /*
1737  * XXX
1738  * The whole multicast option thing needs to be re-thought.
1739  * Several of these options are equally applicable to non-multicast
1740  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1741  * standard option (IP_TTL).
1742  */
1743 
1744 /*
1745  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1746  */
1747 static struct ifnet *
1748 ip_multicast_if(struct in_addr *a, int *ifindexp)
1749 {
1750 	int ifindex;
1751 	struct ifnet *ifp;
1752 
1753 	if (ifindexp)
1754 		*ifindexp = 0;
1755 	if (ntohl(a->s_addr) >> 24 == 0) {
1756 		ifindex = ntohl(a->s_addr) & 0xffffff;
1757 		if (ifindex < 0 || if_index < ifindex)
1758 			return NULL;
1759 		ifp = ifindex2ifnet[ifindex];
1760 		if (ifindexp)
1761 			*ifindexp = ifindex;
1762 	} else {
1763 		INADDR_TO_IFP(*a, ifp);
1764 	}
1765 	return ifp;
1766 }
1767 
1768 /*
1769  * Set the IP multicast options in response to user setsockopt().
1770  */
1771 static int
1772 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1773 {
1774 	int error = 0;
1775 	int i;
1776 	struct in_addr addr;
1777 	struct ip_mreq mreq;
1778 	struct ifnet *ifp;
1779 	struct ip_moptions *imo = *imop;
1780 	struct route ro;
1781 	struct sockaddr_in *dst;
1782 	int ifindex;
1783 	int s;
1784 
1785 	if (imo == NULL) {
1786 		/*
1787 		 * No multicast option buffer attached to the pcb;
1788 		 * allocate one and initialize to default values.
1789 		 */
1790 		imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1791 		    M_WAITOK);
1792 
1793 		if (imo == NULL)
1794 			return (ENOBUFS);
1795 		*imop = imo;
1796 		imo->imo_multicast_ifp = NULL;
1797 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1798 		imo->imo_multicast_vif = -1;
1799 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1800 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1801 		imo->imo_num_memberships = 0;
1802 	}
1803 
1804 	switch (sopt->sopt_name) {
1805 	/* store an index number for the vif you wanna use in the send */
1806 	case IP_MULTICAST_VIF:
1807 		if (legal_vif_num == 0) {
1808 			error = EOPNOTSUPP;
1809 			break;
1810 		}
1811 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1812 		if (error)
1813 			break;
1814 		if (!legal_vif_num(i) && (i != -1)) {
1815 			error = EINVAL;
1816 			break;
1817 		}
1818 		imo->imo_multicast_vif = i;
1819 		break;
1820 
1821 	case IP_MULTICAST_IF:
1822 		/*
1823 		 * Select the interface for outgoing multicast packets.
1824 		 */
1825 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1826 		if (error)
1827 			break;
1828 		/*
1829 		 * INADDR_ANY is used to remove a previous selection.
1830 		 * When no interface is selected, a default one is
1831 		 * chosen every time a multicast packet is sent.
1832 		 */
1833 		if (addr.s_addr == INADDR_ANY) {
1834 			imo->imo_multicast_ifp = NULL;
1835 			break;
1836 		}
1837 		/*
1838 		 * The selected interface is identified by its local
1839 		 * IP address.  Find the interface and confirm that
1840 		 * it supports multicasting.
1841 		 */
1842 		s = splimp();
1843 		ifp = ip_multicast_if(&addr, &ifindex);
1844 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1845 			splx(s);
1846 			error = EADDRNOTAVAIL;
1847 			break;
1848 		}
1849 		imo->imo_multicast_ifp = ifp;
1850 		if (ifindex)
1851 			imo->imo_multicast_addr = addr;
1852 		else
1853 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1854 		splx(s);
1855 		break;
1856 
1857 	case IP_MULTICAST_TTL:
1858 		/*
1859 		 * Set the IP time-to-live for outgoing multicast packets.
1860 		 * The original multicast API required a char argument,
1861 		 * which is inconsistent with the rest of the socket API.
1862 		 * We allow either a char or an int.
1863 		 */
1864 		if (sopt->sopt_valsize == 1) {
1865 			u_char ttl;
1866 			error = sooptcopyin(sopt, &ttl, 1, 1);
1867 			if (error)
1868 				break;
1869 			imo->imo_multicast_ttl = ttl;
1870 		} else {
1871 			u_int ttl;
1872 			error = sooptcopyin(sopt, &ttl, sizeof ttl,
1873 					    sizeof ttl);
1874 			if (error)
1875 				break;
1876 			if (ttl > 255)
1877 				error = EINVAL;
1878 			else
1879 				imo->imo_multicast_ttl = ttl;
1880 		}
1881 		break;
1882 
1883 	case IP_MULTICAST_LOOP:
1884 		/*
1885 		 * Set the loopback flag for outgoing multicast packets.
1886 		 * Must be zero or one.  The original multicast API required a
1887 		 * char argument, which is inconsistent with the rest
1888 		 * of the socket API.  We allow either a char or an int.
1889 		 */
1890 		if (sopt->sopt_valsize == 1) {
1891 			u_char loop;
1892 			error = sooptcopyin(sopt, &loop, 1, 1);
1893 			if (error)
1894 				break;
1895 			imo->imo_multicast_loop = !!loop;
1896 		} else {
1897 			u_int loop;
1898 			error = sooptcopyin(sopt, &loop, sizeof loop,
1899 					    sizeof loop);
1900 			if (error)
1901 				break;
1902 			imo->imo_multicast_loop = !!loop;
1903 		}
1904 		break;
1905 
1906 	case IP_ADD_MEMBERSHIP:
1907 		/*
1908 		 * Add a multicast group membership.
1909 		 * Group must be a valid IP multicast address.
1910 		 */
1911 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1912 		if (error)
1913 			break;
1914 
1915 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1916 			error = EINVAL;
1917 			break;
1918 		}
1919 		s = splimp();
1920 		/*
1921 		 * If no interface address was provided, use the interface of
1922 		 * the route to the given multicast address.
1923 		 */
1924 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1925 			bzero(&ro, sizeof(ro));
1926 			dst = (struct sockaddr_in *)&ro.ro_dst;
1927 			dst->sin_len = sizeof(*dst);
1928 			dst->sin_family = AF_INET;
1929 			dst->sin_addr = mreq.imr_multiaddr;
1930 			rtalloc(&ro);
1931 			if (ro.ro_rt == NULL) {
1932 				error = EADDRNOTAVAIL;
1933 				splx(s);
1934 				break;
1935 			}
1936 			ifp = ro.ro_rt->rt_ifp;
1937 			rtfree(ro.ro_rt);
1938 		}
1939 		else {
1940 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1941 		}
1942 
1943 		/*
1944 		 * See if we found an interface, and confirm that it
1945 		 * supports multicast.
1946 		 */
1947 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1948 			error = EADDRNOTAVAIL;
1949 			splx(s);
1950 			break;
1951 		}
1952 		/*
1953 		 * See if the membership already exists or if all the
1954 		 * membership slots are full.
1955 		 */
1956 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1957 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1958 			    imo->imo_membership[i]->inm_addr.s_addr
1959 						== mreq.imr_multiaddr.s_addr)
1960 				break;
1961 		}
1962 		if (i < imo->imo_num_memberships) {
1963 			error = EADDRINUSE;
1964 			splx(s);
1965 			break;
1966 		}
1967 		if (i == IP_MAX_MEMBERSHIPS) {
1968 			error = ETOOMANYREFS;
1969 			splx(s);
1970 			break;
1971 		}
1972 		/*
1973 		 * Everything looks good; add a new record to the multicast
1974 		 * address list for the given interface.
1975 		 */
1976 		if ((imo->imo_membership[i] =
1977 		    in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1978 			error = ENOBUFS;
1979 			splx(s);
1980 			break;
1981 		}
1982 		++imo->imo_num_memberships;
1983 		splx(s);
1984 		break;
1985 
1986 	case IP_DROP_MEMBERSHIP:
1987 		/*
1988 		 * Drop a multicast group membership.
1989 		 * Group must be a valid IP multicast address.
1990 		 */
1991 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1992 		if (error)
1993 			break;
1994 
1995 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1996 			error = EINVAL;
1997 			break;
1998 		}
1999 
2000 		s = splimp();
2001 		/*
2002 		 * If an interface address was specified, get a pointer
2003 		 * to its ifnet structure.
2004 		 */
2005 		if (mreq.imr_interface.s_addr == INADDR_ANY)
2006 			ifp = NULL;
2007 		else {
2008 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2009 			if (ifp == NULL) {
2010 				error = EADDRNOTAVAIL;
2011 				splx(s);
2012 				break;
2013 			}
2014 		}
2015 		/*
2016 		 * Find the membership in the membership array.
2017 		 */
2018 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2019 			if ((ifp == NULL ||
2020 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2021 			     imo->imo_membership[i]->inm_addr.s_addr ==
2022 			     mreq.imr_multiaddr.s_addr)
2023 				break;
2024 		}
2025 		if (i == imo->imo_num_memberships) {
2026 			error = EADDRNOTAVAIL;
2027 			splx(s);
2028 			break;
2029 		}
2030 		/*
2031 		 * Give up the multicast address record to which the
2032 		 * membership points.
2033 		 */
2034 		in_delmulti(imo->imo_membership[i]);
2035 		/*
2036 		 * Remove the gap in the membership array.
2037 		 */
2038 		for (++i; i < imo->imo_num_memberships; ++i)
2039 			imo->imo_membership[i-1] = imo->imo_membership[i];
2040 		--imo->imo_num_memberships;
2041 		splx(s);
2042 		break;
2043 
2044 	default:
2045 		error = EOPNOTSUPP;
2046 		break;
2047 	}
2048 
2049 	/*
2050 	 * If all options have default values, no need to keep the mbuf.
2051 	 */
2052 	if (imo->imo_multicast_ifp == NULL &&
2053 	    imo->imo_multicast_vif == -1 &&
2054 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2055 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2056 	    imo->imo_num_memberships == 0) {
2057 		free(*imop, M_IPMOPTS);
2058 		*imop = NULL;
2059 	}
2060 
2061 	return (error);
2062 }
2063 
2064 /*
2065  * Return the IP multicast options in response to user getsockopt().
2066  */
2067 static int
2068 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2069 {
2070 	struct in_addr addr;
2071 	struct in_ifaddr *ia;
2072 	int error, optval;
2073 	u_char coptval;
2074 
2075 	error = 0;
2076 	switch (sopt->sopt_name) {
2077 	case IP_MULTICAST_VIF:
2078 		if (imo != NULL)
2079 			optval = imo->imo_multicast_vif;
2080 		else
2081 			optval = -1;
2082 		error = sooptcopyout(sopt, &optval, sizeof optval);
2083 		break;
2084 
2085 	case IP_MULTICAST_IF:
2086 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2087 			addr.s_addr = INADDR_ANY;
2088 		else if (imo->imo_multicast_addr.s_addr) {
2089 			/* return the value user has set */
2090 			addr = imo->imo_multicast_addr;
2091 		} else {
2092 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2093 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2094 				: IA_SIN(ia)->sin_addr.s_addr;
2095 		}
2096 		error = sooptcopyout(sopt, &addr, sizeof addr);
2097 		break;
2098 
2099 	case IP_MULTICAST_TTL:
2100 		if (imo == NULL)
2101 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2102 		else
2103 			optval = coptval = imo->imo_multicast_ttl;
2104 		if (sopt->sopt_valsize == 1)
2105 			error = sooptcopyout(sopt, &coptval, 1);
2106 		else
2107 			error = sooptcopyout(sopt, &optval, sizeof optval);
2108 		break;
2109 
2110 	case IP_MULTICAST_LOOP:
2111 		if (imo == NULL)
2112 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2113 		else
2114 			optval = coptval = imo->imo_multicast_loop;
2115 		if (sopt->sopt_valsize == 1)
2116 			error = sooptcopyout(sopt, &coptval, 1);
2117 		else
2118 			error = sooptcopyout(sopt, &optval, sizeof optval);
2119 		break;
2120 
2121 	default:
2122 		error = ENOPROTOOPT;
2123 		break;
2124 	}
2125 	return (error);
2126 }
2127 
2128 /*
2129  * Discard the IP multicast options.
2130  */
2131 void
2132 ip_freemoptions(struct ip_moptions *imo)
2133 {
2134 	int i;
2135 
2136 	if (imo != NULL) {
2137 		for (i = 0; i < imo->imo_num_memberships; ++i)
2138 			in_delmulti(imo->imo_membership[i]);
2139 		free(imo, M_IPMOPTS);
2140 	}
2141 }
2142 
2143 /*
2144  * Routine called from ip_output() to loop back a copy of an IP multicast
2145  * packet to the input queue of a specified interface.  Note that this
2146  * calls the output routine of the loopback "driver", but with an interface
2147  * pointer that might NOT be a loopback interface -- evil, but easier than
2148  * replicating that code here.
2149  */
2150 static void
2151 ip_mloopback(ifp, m, dst, hlen)
2152 	struct ifnet *ifp;
2153 	struct mbuf *m;
2154 	struct sockaddr_in *dst;
2155 	int hlen;
2156 {
2157 	struct ip *ip;
2158 	struct mbuf *copym;
2159 
2160 	copym = m_copypacket(m, MB_DONTWAIT);
2161 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2162 		copym = m_pullup(copym, hlen);
2163 	if (copym != NULL) {
2164 		/*
2165 		 * if the checksum hasn't been computed, mark it as valid
2166 		 */
2167 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2168 			in_delayed_cksum(copym);
2169 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2170 			copym->m_pkthdr.csum_flags |=
2171 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2172 			copym->m_pkthdr.csum_data = 0xffff;
2173 		}
2174 		/*
2175 		 * We don't bother to fragment if the IP length is greater
2176 		 * than the interface's MTU.  Can this possibly matter?
2177 		 */
2178 		ip = mtod(copym, struct ip *);
2179 		ip->ip_len = htons(ip->ip_len);
2180 		ip->ip_off = htons(ip->ip_off);
2181 		ip->ip_sum = 0;
2182 		if (ip->ip_vhl == IP_VHL_BORING) {
2183 			ip->ip_sum = in_cksum_hdr(ip);
2184 		} else {
2185 			ip->ip_sum = in_cksum(copym, hlen);
2186 		}
2187 		/*
2188 		 * NB:
2189 		 * It's not clear whether there are any lingering
2190 		 * reentrancy problems in other areas which might
2191 		 * be exposed by using ip_input directly (in
2192 		 * particular, everything which modifies the packet
2193 		 * in-place).  Yet another option is using the
2194 		 * protosw directly to deliver the looped back
2195 		 * packet.  For the moment, we'll err on the side
2196 		 * of safety by using if_simloop().
2197 		 */
2198 #if 1 /* XXX */
2199 		if (dst->sin_family != AF_INET) {
2200 			printf("ip_mloopback: bad address family %d\n",
2201 						dst->sin_family);
2202 			dst->sin_family = AF_INET;
2203 		}
2204 #endif
2205 
2206 #ifdef notdef
2207 		copym->m_pkthdr.rcvif = ifp;
2208 		ip_input(copym);
2209 #else
2210 		if_simloop(ifp, copym, dst->sin_family, 0);
2211 #endif
2212 	}
2213 }
2214