xref: /dflybsd-src/sys/netinet/ip_output.c (revision 48d201a5a8c1dab4aa7166b0812594c101fc43c3)
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 1.20 2004/08/26 21:21:46 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/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 <netipsec/ipsec.h>
82 #include <netipsec/xform.h>
83 #include <netipsec/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 #ifdef IPSEC
138 	struct route iproute;
139 	struct secpolicy *sp = NULL;
140 	struct socket *so = inp ? inp->inp_socket : NULL;
141 #endif
142 #ifdef FAST_IPSEC
143 	struct route iproute;
144 	struct m_tag *mtag;
145 	struct secpolicy *sp = NULL;
146 	struct tdb_ident *tdbi;
147 	int s;
148 #endif /* FAST_IPSEC */
149 	struct ip_fw_args args;
150 	int src_was_INADDR_ANY = 0;	/* as the name says... */
151 
152 	args.eh = NULL;
153 	args.rule = NULL;
154 	args.next_hop = NULL;
155 	args.divert_rule = 0;			/* divert cookie */
156 
157 	/* Grab info from MT_TAG mbufs prepended to the chain. */
158 	for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
159 		switch(m0->_m_tag_id) {
160 		default:
161 			printf("ip_output: unrecognised MT_TAG tag %d\n",
162 			    m0->_m_tag_id);
163 			break;
164 
165 		case PACKET_TAG_DUMMYNET:
166 			/*
167 			 * the packet was already tagged, so part of the
168 			 * processing was already done, and we need to go down.
169 			 * Get parameters from the header.
170 			 */
171 			args.rule = ((struct dn_pkt *)m0)->rule;
172 			opt = NULL ;
173 			ro = & ( ((struct dn_pkt *)m0)->ro ) ;
174 			imo = NULL ;
175 			dst = ((struct dn_pkt *)m0)->dn_dst ;
176 			ifp = ((struct dn_pkt *)m0)->ifp ;
177 			flags = ((struct dn_pkt *)m0)->flags ;
178 			break;
179 
180 		case PACKET_TAG_DIVERT:
181 			args.divert_rule = (int)m0->m_data & 0xffff;
182 			break;
183 
184 		case PACKET_TAG_IPFORWARD:
185 			args.next_hop = (struct sockaddr_in *)m0->m_data;
186 			break;
187 		}
188 	}
189 	m = m0;
190 
191 	KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
192 #ifndef FAST_IPSEC
193 	KASSERT(ro != NULL, ("ip_output: no route, proto %d",
194 	    mtod(m, struct ip *)->ip_p));
195 #endif
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)) == 0) {
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 #ifdef FAST_IPSEC
231 	if (ro == NULL) {
232 		ro = &iproute;
233 		bzero(ro, sizeof (*ro));
234 	}
235 #endif /* FAST_IPSEC */
236 	dst = (struct sockaddr_in *)&ro->ro_dst;
237 	/*
238 	 * If there is a cached route,
239 	 * check that it is to the same destination
240 	 * and is still up.  If not, free it and try again.
241 	 * The address family should also be checked in case of sharing the
242 	 * cache with IPv6.
243 	 */
244 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
245 			  dst->sin_family != AF_INET ||
246 			  dst->sin_addr.s_addr != pkt_dst.s_addr)) {
247 		RTFREE(ro->ro_rt);
248 		ro->ro_rt = (struct rtentry *)0;
249 	}
250 	if (ro->ro_rt == 0) {
251 		bzero(dst, sizeof(*dst));
252 		dst->sin_family = AF_INET;
253 		dst->sin_len = sizeof(*dst);
254 		dst->sin_addr = pkt_dst;
255 	}
256 	/*
257 	 * If routing to interface only,
258 	 * short circuit routing lookup.
259 	 */
260 	if (flags & IP_ROUTETOIF) {
261 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
262 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
263 			ipstat.ips_noroute++;
264 			error = ENETUNREACH;
265 			goto bad;
266 		}
267 		ifp = ia->ia_ifp;
268 		ip->ip_ttl = 1;
269 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
270 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
271 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
272 		/*
273 		 * Bypass the normal routing lookup for multicast
274 		 * packets if the interface is specified.
275 		 */
276 		ifp = imo->imo_multicast_ifp;
277 		IFP_TO_IA(ifp, ia);
278 		isbroadcast = 0;	/* fool gcc */
279 	} else {
280 		/*
281 		 * If this is the case, we probably don't want to allocate
282 		 * a protocol-cloned route since we didn't get one from the
283 		 * ULP.  This lets TCP do its thing, while not burdening
284 		 * forwarding or ICMP with the overhead of cloning a route.
285 		 * Of course, we still want to do any cloning requested by
286 		 * the link layer, as this is probably required in all cases
287 		 * for correct operation (as it is for ARP).
288 		 */
289 		if (ro->ro_rt == 0)
290 			rtalloc_ign(ro, RTF_PRCLONING);
291 		if (ro->ro_rt == 0) {
292 			ipstat.ips_noroute++;
293 			error = EHOSTUNREACH;
294 			goto bad;
295 		}
296 		ia = ifatoia(ro->ro_rt->rt_ifa);
297 		ifp = ro->ro_rt->rt_ifp;
298 		ro->ro_rt->rt_use++;
299 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
300 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
301 		if (ro->ro_rt->rt_flags & RTF_HOST)
302 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
303 		else
304 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
305 	}
306 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
307 		struct in_multi *inm;
308 
309 		m->m_flags |= M_MCAST;
310 		/*
311 		 * IP destination address is multicast.  Make sure "dst"
312 		 * still points to the address in "ro".  (It may have been
313 		 * changed to point to a gateway address, above.)
314 		 */
315 		dst = (struct sockaddr_in *)&ro->ro_dst;
316 		/*
317 		 * See if the caller provided any multicast options
318 		 */
319 		if (imo != NULL) {
320 			ip->ip_ttl = imo->imo_multicast_ttl;
321 			if (imo->imo_multicast_vif != -1)
322 				ip->ip_src.s_addr =
323 				    ip_mcast_src ?
324 				    ip_mcast_src(imo->imo_multicast_vif) :
325 				    INADDR_ANY;
326 		} else
327 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
328 		/*
329 		 * Confirm that the outgoing interface supports multicast.
330 		 */
331 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
332 			if ((ifp->if_flags & IFF_MULTICAST) == 0) {
333 				ipstat.ips_noroute++;
334 				error = ENETUNREACH;
335 				goto bad;
336 			}
337 		}
338 		/*
339 		 * If source address not specified yet, use address
340 		 * of outgoing interface.
341 		 */
342 		if (ip->ip_src.s_addr == INADDR_ANY) {
343 			/* Interface may have no addresses. */
344 			if (ia != NULL)
345 				ip->ip_src = IA_SIN(ia)->sin_addr;
346 		}
347 
348 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
349 		if (inm != NULL &&
350 		   (imo == NULL || imo->imo_multicast_loop)) {
351 			/*
352 			 * If we belong to the destination multicast group
353 			 * on the outgoing interface, and the caller did not
354 			 * forbid loopback, loop back a copy.
355 			 */
356 			ip_mloopback(ifp, m, dst, hlen);
357 		}
358 		else {
359 			/*
360 			 * If we are acting as a multicast router, perform
361 			 * multicast forwarding as if the packet had just
362 			 * arrived on the interface to which we are about
363 			 * to send.  The multicast forwarding function
364 			 * recursively calls this function, using the
365 			 * IP_FORWARDING flag to prevent infinite recursion.
366 			 *
367 			 * Multicasts that are looped back by ip_mloopback(),
368 			 * above, will be forwarded by the ip_input() routine,
369 			 * if necessary.
370 			 */
371 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
372 				/*
373 				 * If rsvp daemon is not running, do not
374 				 * set ip_moptions. This ensures that the packet
375 				 * is multicast and not just sent down one link
376 				 * as prescribed by rsvpd.
377 				 */
378 				if (!rsvp_on)
379 					imo = NULL;
380 				if (ip_mforward &&
381 				    ip_mforward(ip, ifp, m, imo) != 0) {
382 					m_freem(m);
383 					goto done;
384 				}
385 			}
386 		}
387 
388 		/*
389 		 * Multicasts with a time-to-live of zero may be looped-
390 		 * back, above, but must not be transmitted on a network.
391 		 * Also, multicasts addressed to the loopback interface
392 		 * are not sent -- the above call to ip_mloopback() will
393 		 * loop back a copy if this host actually belongs to the
394 		 * destination group on the loopback interface.
395 		 */
396 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
397 			m_freem(m);
398 			goto done;
399 		}
400 
401 		goto sendit;
402 	}
403 #ifndef notdef
404 	/*
405 	 * If the source address is not specified yet, use the address
406 	 * of the outoing interface. In case, keep note we did that, so
407 	 * if the the firewall changes the next-hop causing the output
408 	 * interface to change, we can fix that.
409 	 */
410 	if (ip->ip_src.s_addr == INADDR_ANY) {
411 		/* Interface may have no addresses. */
412 		if (ia != NULL) {
413 			ip->ip_src = IA_SIN(ia)->sin_addr;
414 			src_was_INADDR_ANY = 1;
415 		}
416 	}
417 #endif /* notdef */
418 	/*
419 	 * Verify that we have any chance at all of being able to queue
420 	 *      the packet or packet fragments
421 	 */
422 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
423 		ifp->if_snd.ifq_maxlen) {
424 			error = ENOBUFS;
425 			ipstat.ips_odropped++;
426 			goto bad;
427 	}
428 
429 	/*
430 	 * Look for broadcast address and
431 	 * verify user is allowed to send
432 	 * such a packet.
433 	 */
434 	if (isbroadcast) {
435 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
436 			error = EADDRNOTAVAIL;
437 			goto bad;
438 		}
439 		if ((flags & IP_ALLOWBROADCAST) == 0) {
440 			error = EACCES;
441 			goto bad;
442 		}
443 		/* don't allow broadcast messages to be fragmented */
444 		if (ip->ip_len > ifp->if_mtu) {
445 			error = EMSGSIZE;
446 			goto bad;
447 		}
448 		m->m_flags |= M_BCAST;
449 	} else {
450 		m->m_flags &= ~M_BCAST;
451 	}
452 
453 sendit:
454 #ifdef IPSEC
455 	/* get SP for this packet */
456 	if (so == NULL)
457 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
458 	else
459 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
460 
461 	if (sp == NULL) {
462 		ipsecstat.out_inval++;
463 		goto bad;
464 	}
465 
466 	error = 0;
467 
468 	/* check policy */
469 	switch (sp->policy) {
470 	case IPSEC_POLICY_DISCARD:
471 		/*
472 		 * This packet is just discarded.
473 		 */
474 		ipsecstat.out_polvio++;
475 		goto bad;
476 
477 	case IPSEC_POLICY_BYPASS:
478 	case IPSEC_POLICY_NONE:
479 		/* no need to do IPsec. */
480 		goto skip_ipsec;
481 
482 	case IPSEC_POLICY_IPSEC:
483 		if (sp->req == NULL) {
484 			/* acquire a policy */
485 			error = key_spdacquire(sp);
486 			goto bad;
487 		}
488 		break;
489 
490 	case IPSEC_POLICY_ENTRUST:
491 	default:
492 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
493 	}
494     {
495 	struct ipsec_output_state state;
496 	bzero(&state, sizeof(state));
497 	state.m = m;
498 	if (flags & IP_ROUTETOIF) {
499 		state.ro = &iproute;
500 		bzero(&iproute, sizeof(iproute));
501 	} else
502 		state.ro = ro;
503 	state.dst = (struct sockaddr *)dst;
504 
505 	ip->ip_sum = 0;
506 
507 	/*
508 	 * XXX
509 	 * delayed checksums are not currently compatible with IPsec
510 	 */
511 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
512 		in_delayed_cksum(m);
513 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
514 	}
515 
516 	ip->ip_len = htons(ip->ip_len);
517 	ip->ip_off = htons(ip->ip_off);
518 
519 	error = ipsec4_output(&state, sp, flags);
520 
521 	m = state.m;
522 	if (flags & IP_ROUTETOIF) {
523 		/*
524 		 * if we have tunnel mode SA, we may need to ignore
525 		 * IP_ROUTETOIF.
526 		 */
527 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
528 			flags &= ~IP_ROUTETOIF;
529 			ro = state.ro;
530 		}
531 	} else
532 		ro = state.ro;
533 	dst = (struct sockaddr_in *)state.dst;
534 	if (error) {
535 		/* mbuf is already reclaimed in ipsec4_output. */
536 		m0 = NULL;
537 		switch (error) {
538 		case EHOSTUNREACH:
539 		case ENETUNREACH:
540 		case EMSGSIZE:
541 		case ENOBUFS:
542 		case ENOMEM:
543 			break;
544 		default:
545 			printf("ip4_output (ipsec): error code %d\n", error);
546 			/*fall through*/
547 		case ENOENT:
548 			/* don't show these error codes to the user */
549 			error = 0;
550 			break;
551 		}
552 		goto bad;
553 	}
554     }
555 
556 	/* be sure to update variables that are affected by ipsec4_output() */
557 	ip = mtod(m, struct ip *);
558 #ifdef _IP_VHL
559 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
560 #else
561 	hlen = ip->ip_hl << 2;
562 #endif
563 	if (ro->ro_rt == NULL) {
564 		if ((flags & IP_ROUTETOIF) == 0) {
565 			printf("ip_output: "
566 				"can't update route after IPsec processing\n");
567 			error = EHOSTUNREACH;	/*XXX*/
568 			goto bad;
569 		}
570 	} else {
571 		ia = ifatoia(ro->ro_rt->rt_ifa);
572 		ifp = ro->ro_rt->rt_ifp;
573 	}
574 
575 	/* make it flipped, again. */
576 	ip->ip_len = ntohs(ip->ip_len);
577 	ip->ip_off = ntohs(ip->ip_off);
578 skip_ipsec:
579 #endif /*IPSEC*/
580 #ifdef FAST_IPSEC
581 	/*
582 	 * Check the security policy (SP) for the packet and, if
583 	 * required, do IPsec-related processing.  There are two
584 	 * cases here; the first time a packet is sent through
585 	 * it will be untagged and handled by ipsec4_checkpolicy.
586 	 * If the packet is resubmitted to ip_output (e.g. after
587 	 * AH, ESP, etc. processing), there will be a tag to bypass
588 	 * the lookup and related policy checking.
589 	 */
590 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
591 	s = splnet();
592 	if (mtag != NULL) {
593 		tdbi = (struct tdb_ident *)(mtag + 1);
594 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
595 		if (sp == NULL)
596 			error = -EINVAL;	/* force silent drop */
597 		m_tag_delete(m, mtag);
598 	} else {
599 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
600 					&error, inp);
601 	}
602 	/*
603 	 * There are four return cases:
604 	 *    sp != NULL	 	    apply IPsec policy
605 	 *    sp == NULL, error == 0	    no IPsec handling needed
606 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
607 	 *    sp == NULL, error != 0	    discard packet, report error
608 	 */
609 	if (sp != NULL) {
610 		/* Loop detection, check if ipsec processing already done */
611 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
612 		for (mtag = m_tag_first(m); mtag != NULL;
613 		     mtag = m_tag_next(m, mtag)) {
614 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
615 				continue;
616 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
617 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
618 				continue;
619 			/*
620 			 * Check if policy has an SA associated with it.
621 			 * This can happen when an SP has yet to acquire
622 			 * an SA; e.g. on first reference.  If it occurs,
623 			 * then we let ipsec4_process_packet do its thing.
624 			 */
625 			if (sp->req->sav == NULL)
626 				break;
627 			tdbi = (struct tdb_ident *)(mtag + 1);
628 			if (tdbi->spi == sp->req->sav->spi &&
629 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
630 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
631 				 sizeof (union sockaddr_union)) == 0) {
632 				/*
633 				 * No IPsec processing is needed, free
634 				 * reference to SP.
635 				 *
636 				 * NB: null pointer to avoid free at
637 				 *     done: below.
638 				 */
639 				KEY_FREESP(&sp), sp = NULL;
640 				splx(s);
641 				goto spd_done;
642 			}
643 		}
644 
645 		/*
646 		 * Do delayed checksums now because we send before
647 		 * this is done in the normal processing path.
648 		 */
649 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
650 			in_delayed_cksum(m);
651 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
652 		}
653 
654 		ip->ip_len = htons(ip->ip_len);
655 		ip->ip_off = htons(ip->ip_off);
656 
657 		/* NB: callee frees mbuf */
658 		error = ipsec4_process_packet(m, sp->req, flags, 0);
659 		/*
660 		 * Preserve KAME behaviour: ENOENT can be returned
661 		 * when an SA acquire is in progress.  Don't propagate
662 		 * this to user-level; it confuses applications.
663 		 *
664 		 * XXX this will go away when the SADB is redone.
665 		 */
666 		if (error == ENOENT)
667 			error = 0;
668 		splx(s);
669 		goto done;
670 	} else {
671 		splx(s);
672 
673 		if (error != 0) {
674 			/*
675 			 * Hack: -EINVAL is used to signal that a packet
676 			 * should be silently discarded.  This is typically
677 			 * because we asked key management for an SA and
678 			 * it was delayed (e.g. kicked up to IKE).
679 			 */
680 			if (error == -EINVAL)
681 				error = 0;
682 			goto bad;
683 		} else {
684 			/* No IPsec processing for this packet. */
685 		}
686 #ifdef notyet
687 		/*
688 		 * If deferred crypto processing is needed, check that
689 		 * the interface supports it.
690 		 */
691 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
692 		if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
693 			/* notify IPsec to do its own crypto */
694 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
695 			error = EHOSTUNREACH;
696 			goto bad;
697 		}
698 #endif
699 	}
700 spd_done:
701 #endif /* FAST_IPSEC */
702 	/*
703 	 * IpHack's section.
704 	 * - Xlate: translate packet's addr/port (NAT).
705 	 * - Firewall: deny/allow/etc.
706 	 * - Wrap: fake packet's addr/port <unimpl.>
707 	 * - Encapsulate: put it in another IP and send out. <unimp.>
708 	 */
709 
710 	/*
711 	 * Run through list of hooks for output packets.
712 	 */
713 	if (pfil_has_hooks(&inet_pfil_hook)) {
714 		error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
715 		if (error != 0 || m == NULL)
716 			goto done;
717 		ip = mtod(m, struct ip *);
718 	}
719 
720 	/*
721 	 * Check with the firewall...
722 	 * but not if we are already being fwd'd from a firewall.
723 	 */
724 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
725 		struct sockaddr_in *old = dst;
726 
727 		args.m = m;
728 		args.next_hop = dst;
729 		args.oif = ifp;
730 		off = ip_fw_chk_ptr(&args);
731 		m = args.m;
732 		dst = args.next_hop;
733 
734                 /*
735 		 * On return we must do the following:
736 		 * m == NULL	-> drop the pkt (old interface, deprecated)
737 		 * (off & IP_FW_PORT_DENY_FLAG)	-> drop the pkt (new interface)
738 		 * 1<=off<= 0xffff		-> DIVERT
739 		 * (off & IP_FW_PORT_DYNT_FLAG)	-> send to a DUMMYNET pipe
740 		 * (off & IP_FW_PORT_TEE_FLAG)	-> TEE the packet
741 		 * dst != old			-> IPFIREWALL_FORWARD
742 		 * off==0, dst==old		-> accept
743 		 * If some of the above modules are not compiled in, then
744 		 * we should't have to check the corresponding condition
745 		 * (because the ipfw control socket should not accept
746 		 * unsupported rules), but better play safe and drop
747 		 * packets in case of doubt.
748 		 */
749 		if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
750 			if (m)
751 				m_freem(m);
752 			error = EACCES;
753 			goto done;
754 		}
755 		ip = mtod(m, struct ip *);
756 		if (off == 0 && dst == old)		/* common case */
757 			goto pass;
758                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
759 			/*
760 			 * pass the pkt to dummynet. Need to include
761 			 * pipe number, m, ifp, ro, dst because these are
762 			 * not recomputed in the next pass.
763 			 * All other parameters have been already used and
764 			 * so they are not needed anymore.
765 			 * XXX note: if the ifp or ro entry are deleted
766 			 * while a pkt is in dummynet, we are in trouble!
767 			 */
768 			args.ro = ro;
769 			args.dst = dst;
770 			args.flags = flags;
771 
772 			error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
773 				&args);
774 			goto done;
775 		}
776 #ifdef IPDIVERT
777 		if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
778 			struct mbuf *clone = NULL;
779 
780 			/* Clone packet if we're doing a 'tee' */
781 			if ((off & IP_FW_PORT_TEE_FLAG) != 0)
782 				clone = m_dup(m, MB_DONTWAIT);
783 
784 			/*
785 			 * XXX
786 			 * delayed checksums are not currently compatible
787 			 * with divert sockets.
788 			 */
789 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
790 				in_delayed_cksum(m);
791 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
792 			}
793 
794 			/* Restore packet header fields to original values */
795 			ip->ip_len = htons(ip->ip_len);
796 			ip->ip_off = htons(ip->ip_off);
797 
798 			/* Deliver packet to divert input routine */
799 			divert_packet(m, 0, off & 0xffff, args.divert_rule);
800 
801 			/* If 'tee', continue with original packet */
802 			if (clone != NULL) {
803 				m = clone;
804 				ip = mtod(m, struct ip *);
805 				goto pass;
806 			}
807 			goto done;
808 		}
809 #endif
810 
811 		/* IPFIREWALL_FORWARD */
812 		/*
813 		 * Check dst to make sure it is directly reachable on the
814 		 * interface we previously thought it was.
815 		 * If it isn't (which may be likely in some situations) we have
816 		 * to re-route it (ie, find a route for the next-hop and the
817 		 * associated interface) and set them here. This is nested
818 		 * forwarding which in most cases is undesirable, except where
819 		 * such control is nigh impossible. So we do it here.
820 		 * And I'm babbling.
821 		 */
822 		if (off == 0 && old != dst) { /* FORWARD, dst has changed */
823 #if 0
824 			/*
825 			 * XXX To improve readability, this block should be
826 			 * changed into a function call as below:
827 			 */
828 			error = ip_ipforward(&m, &dst, &ifp);
829 			if (error)
830 				goto bad;
831 			if (m == NULL) /* ip_input consumed the mbuf */
832 				goto done;
833 #else
834 			struct in_ifaddr *ia;
835 
836 			/*
837 			 * XXX sro_fwd below is static, and a pointer
838 			 * to it gets passed to routines downstream.
839 			 * This could have surprisingly bad results in
840 			 * practice, because its content is overwritten
841 			 * by subsequent packets.
842 			 */
843 			/* There must be a better way to do this next line... */
844 			static struct route sro_fwd;
845 			struct route *ro_fwd = &sro_fwd;
846 
847 #if 0
848 			print_ip("IPFIREWALL_FORWARD: New dst ip: ",
849 			    dst->sin_addr, "\n");
850 #endif
851 
852 			/*
853 			 * We need to figure out if we have been forwarded
854 			 * to a local socket. If so, then we should somehow
855 			 * "loop back" to ip_input, and get directed to the
856 			 * PCB as if we had received this packet. This is
857 			 * because it may be dificult to identify the packets
858 			 * you want to forward until they are being output
859 			 * and have selected an interface. (e.g. locally
860 			 * initiated packets) If we used the loopback inteface,
861 			 * we would not be able to control what happens
862 			 * as the packet runs through ip_input() as
863 			 * it is done through a ISR.
864 			 */
865 			LIST_FOREACH(ia,
866 			    INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
867 				/*
868 				 * If the addr to forward to is one
869 				 * of ours, we pretend to
870 				 * be the destination for this packet.
871 				 */
872 				if (IA_SIN(ia)->sin_addr.s_addr ==
873 						 dst->sin_addr.s_addr)
874 					break;
875 			}
876 			if (ia) {	/* tell ip_input "dont filter" */
877 				struct m_hdr tag;
878 
879 				tag.mh_type = MT_TAG;
880 				tag.mh_flags = PACKET_TAG_IPFORWARD;
881 				tag.mh_data = (caddr_t)args.next_hop;
882 				tag.mh_next = m;
883 
884 				if (m->m_pkthdr.rcvif == NULL)
885 					m->m_pkthdr.rcvif = ifunit("lo0");
886 				if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
887 					m->m_pkthdr.csum_flags |=
888 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
889 					m0->m_pkthdr.csum_data = 0xffff;
890 				}
891 				m->m_pkthdr.csum_flags |=
892 				    CSUM_IP_CHECKED | CSUM_IP_VALID;
893 				ip->ip_len = htons(ip->ip_len);
894 				ip->ip_off = htons(ip->ip_off);
895 				ip_input((struct mbuf *)&tag);
896 				goto done;
897 			}
898 			/* Some of the logic for this was
899 			 * nicked from above.
900 			 *
901 			 * This rewrites the cached route in a local PCB.
902 			 * Is this what we want to do?
903 			 */
904 			bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
905 
906 			ro_fwd->ro_rt = 0;
907 			rtalloc_ign(ro_fwd, RTF_PRCLONING);
908 
909 			if (ro_fwd->ro_rt == 0) {
910 				ipstat.ips_noroute++;
911 				error = EHOSTUNREACH;
912 				goto bad;
913 			}
914 
915 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
916 			ifp = ro_fwd->ro_rt->rt_ifp;
917 			ro_fwd->ro_rt->rt_use++;
918 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
919 				dst = (struct sockaddr_in *)
920 					ro_fwd->ro_rt->rt_gateway;
921 			if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
922 				isbroadcast =
923 				    (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
924 			else
925 				isbroadcast = in_broadcast(dst->sin_addr, ifp);
926 			if (ro->ro_rt)
927 				RTFREE(ro->ro_rt);
928 			ro->ro_rt = ro_fwd->ro_rt;
929 			dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
930 
931 #endif	/* ... block to be put into a function */
932 			/*
933 			 * If we added a default src ip earlier,
934 			 * which would have been gotten from the-then
935 			 * interface, do it again, from the new one.
936 			 */
937 			if (src_was_INADDR_ANY)
938 				ip->ip_src = IA_SIN(ia)->sin_addr;
939 			goto pass ;
940 		}
941 
942                 /*
943                  * if we get here, none of the above matches, and
944                  * we have to drop the pkt
945                  */
946 		m_freem(m);
947                 error = EACCES; /* not sure this is the right error msg */
948                 goto done;
949 	}
950 
951 pass:
952 	/* 127/8 must not appear on wire - RFC1122. */
953 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
954 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
955 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
956 			ipstat.ips_badaddr++;
957 			error = EADDRNOTAVAIL;
958 			goto bad;
959 		}
960 	}
961 
962 	m->m_pkthdr.csum_flags |= CSUM_IP;
963 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
964 	if (sw_csum & CSUM_DELAY_DATA) {
965 		in_delayed_cksum(m);
966 		sw_csum &= ~CSUM_DELAY_DATA;
967 	}
968 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
969 
970 	/*
971 	 * If small enough for interface, or the interface will take
972 	 * care of the fragmentation for us, can just send directly.
973 	 */
974 	if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
975 	    (ip->ip_off & IP_DF) == 0)) {
976 		ip->ip_len = htons(ip->ip_len);
977 		ip->ip_off = htons(ip->ip_off);
978 		ip->ip_sum = 0;
979 		if (sw_csum & CSUM_DELAY_IP) {
980 			if (ip->ip_vhl == IP_VHL_BORING) {
981 				ip->ip_sum = in_cksum_hdr(ip);
982 			} else {
983 				ip->ip_sum = in_cksum(m, hlen);
984 			}
985 		}
986 
987 		/* Record statistics for this interface address. */
988 		if (!(flags & IP_FORWARDING) && ia) {
989 			ia->ia_ifa.if_opackets++;
990 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
991 		}
992 
993 #ifdef IPSEC
994 		/* clean ipsec history once it goes out of the node */
995 		ipsec_delaux(m);
996 #endif
997 
998 #ifdef MBUF_STRESS_TEST
999 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1000 			struct mbuf *m1, *m2;
1001 			int length, tmp;
1002 
1003 			tmp = length = m->m_pkthdr.len;
1004 
1005 			while ((length -= mbuf_frag_size) >= 1) {
1006 				m1 = m_split(m, length, MB_DONTWAIT);
1007 				if (m1 == NULL)
1008 					break;
1009 				m1->m_flags &= ~M_PKTHDR;
1010 				m2 = m;
1011 				while (m2->m_next != NULL)
1012 					m2 = m2->m_next;
1013 				m2->m_next = m1;
1014 			}
1015 			m->m_pkthdr.len = tmp;
1016 		}
1017 #endif
1018 		error = (*ifp->if_output)(ifp, m,
1019 				(struct sockaddr *)dst, ro->ro_rt);
1020 		goto done;
1021 	}
1022 
1023 	if (ip->ip_off & IP_DF) {
1024 		error = EMSGSIZE;
1025 		/*
1026 		 * This case can happen if the user changed the MTU
1027 		 * of an interface after enabling IP on it.  Because
1028 		 * most netifs don't keep track of routes pointing to
1029 		 * them, there is no way for one to update all its
1030 		 * routes when the MTU is changed.
1031 		 */
1032 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1033 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1034 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1035 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1036 		}
1037 		ipstat.ips_cantfrag++;
1038 		goto bad;
1039 	}
1040 
1041 	/*
1042 	 * Too large for interface; fragment if possible. If successful,
1043 	 * on return, m will point to a list of packets to be sent.
1044 	 */
1045 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1046 	if (error)
1047 		goto bad;
1048 	for (; m; m = m0) {
1049 		m0 = m->m_nextpkt;
1050 		m->m_nextpkt = 0;
1051 #ifdef IPSEC
1052 		/* clean ipsec history once it goes out of the node */
1053 		ipsec_delaux(m);
1054 #endif
1055 		if (error == 0) {
1056 			/* Record statistics for this interface address. */
1057 			if (ia != NULL) {
1058 				ia->ia_ifa.if_opackets++;
1059 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1060 			}
1061 
1062 			error = (*ifp->if_output)(ifp, m,
1063 			    (struct sockaddr *)dst, ro->ro_rt);
1064 		} else
1065 			m_freem(m);
1066 	}
1067 
1068 	if (error == 0)
1069 		ipstat.ips_fragmented++;
1070 
1071 done:
1072 #ifdef IPSEC
1073 	if (ro == &iproute && ro->ro_rt) {
1074 		RTFREE(ro->ro_rt);
1075 		ro->ro_rt = NULL;
1076 	}
1077 	if (sp != NULL) {
1078 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1079 			printf("DP ip_output call free SP:%p\n", sp));
1080 		key_freesp(sp);
1081 	}
1082 #endif
1083 #ifdef FAST_IPSEC
1084 	if (ro == &iproute && ro->ro_rt) {
1085 		RTFREE(ro->ro_rt);
1086 		ro->ro_rt = NULL;
1087 	}
1088 	if (sp != NULL)
1089 		KEY_FREESP(&sp);
1090 #endif
1091 	return (error);
1092 bad:
1093 	m_freem(m);
1094 	goto done;
1095 }
1096 
1097 /*
1098  * Create a chain of fragments which fit the given mtu. m_frag points to the
1099  * mbuf to be fragmented; on return it points to the chain with the fragments.
1100  * Return 0 if no error. If error, m_frag may contain a partially built
1101  * chain of fragments that should be freed by the caller.
1102  *
1103  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1104  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1105  */
1106 int
1107 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1108 	    u_long if_hwassist_flags, int sw_csum)
1109 {
1110 	int error = 0;
1111 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1112 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1113 	int off;
1114 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1115 	int firstlen;
1116 	struct mbuf **mnext;
1117 	int nfrags;
1118 
1119 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1120 		ipstat.ips_cantfrag++;
1121 		return EMSGSIZE;
1122 	}
1123 
1124 	/*
1125 	 * Must be able to put at least 8 bytes per fragment.
1126 	 */
1127 	if (len < 8)
1128 		return EMSGSIZE;
1129 
1130 	/*
1131 	 * If the interface will not calculate checksums on
1132 	 * fragmented packets, then do it here.
1133 	 */
1134 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1135 	    (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1136 		in_delayed_cksum(m0);
1137 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1138 	}
1139 
1140 	if (len > PAGE_SIZE) {
1141 		/*
1142 		 * Fragment large datagrams such that each segment
1143 		 * contains a multiple of PAGE_SIZE amount of data,
1144 		 * plus headers. This enables a receiver to perform
1145 		 * page-flipping zero-copy optimizations.
1146 		 *
1147 		 * XXX When does this help given that sender and receiver
1148 		 * could have different page sizes, and also mtu could
1149 		 * be less than the receiver's page size ?
1150 		 */
1151 		int newlen;
1152 		struct mbuf *m;
1153 
1154 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1155 			off += m->m_len;
1156 
1157 		/*
1158 		 * firstlen (off - hlen) must be aligned on an
1159 		 * 8-byte boundary
1160 		 */
1161 		if (off < hlen)
1162 			goto smart_frag_failure;
1163 		off = ((off - hlen) & ~7) + hlen;
1164 		newlen = (~PAGE_MASK) & mtu;
1165 		if ((newlen + sizeof (struct ip)) > mtu) {
1166 			/* we failed, go back the default */
1167 smart_frag_failure:
1168 			newlen = len;
1169 			off = hlen + len;
1170 		}
1171 		len = newlen;
1172 
1173 	} else {
1174 		off = hlen + len;
1175 	}
1176 
1177 	firstlen = off - hlen;
1178 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1179 
1180 	/*
1181 	 * Loop through length of segment after first fragment,
1182 	 * make new header and copy data of each part and link onto chain.
1183 	 * Here, m0 is the original packet, m is the fragment being created.
1184 	 * The fragments are linked off the m_nextpkt of the original
1185 	 * packet, which after processing serves as the first fragment.
1186 	 */
1187 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1188 		struct ip *mhip;	/* ip header on the fragment */
1189 		struct mbuf *m;
1190 		int mhlen = sizeof (struct ip);
1191 
1192 		MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1193 		if (m == 0) {
1194 			error = ENOBUFS;
1195 			ipstat.ips_odropped++;
1196 			goto done;
1197 		}
1198 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1199 		/*
1200 		 * In the first mbuf, leave room for the link header, then
1201 		 * copy the original IP header including options. The payload
1202 		 * goes into an additional mbuf chain returned by m_copy().
1203 		 */
1204 		m->m_data += max_linkhdr;
1205 		mhip = mtod(m, struct ip *);
1206 		*mhip = *ip;
1207 		if (hlen > sizeof (struct ip)) {
1208 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1209 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1210 		}
1211 		m->m_len = mhlen;
1212 		/* XXX do we need to add ip->ip_off below ? */
1213 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1214 		if (off + len >= ip->ip_len) {	/* last fragment */
1215 			len = ip->ip_len - off;
1216 			m->m_flags |= M_LASTFRAG;
1217 		} else
1218 			mhip->ip_off |= IP_MF;
1219 		mhip->ip_len = htons((u_short)(len + mhlen));
1220 		m->m_next = m_copy(m0, off, len);
1221 		if (m->m_next == 0) {		/* copy failed */
1222 			m_free(m);
1223 			error = ENOBUFS;	/* ??? */
1224 			ipstat.ips_odropped++;
1225 			goto done;
1226 		}
1227 		m->m_pkthdr.len = mhlen + len;
1228 		m->m_pkthdr.rcvif = (struct ifnet *)0;
1229 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1230 		mhip->ip_off = htons(mhip->ip_off);
1231 		mhip->ip_sum = 0;
1232 		if (sw_csum & CSUM_DELAY_IP)
1233 			mhip->ip_sum = in_cksum(m, mhlen);
1234 		*mnext = m;
1235 		mnext = &m->m_nextpkt;
1236 	}
1237 	ipstat.ips_ofragments += nfrags;
1238 
1239 	/* set first marker for fragment chain */
1240 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1241 	m0->m_pkthdr.csum_data = nfrags;
1242 
1243 	/*
1244 	 * Update first fragment by trimming what's been copied out
1245 	 * and updating header.
1246 	 */
1247 	m_adj(m0, hlen + firstlen - ip->ip_len);
1248 	m0->m_pkthdr.len = hlen + firstlen;
1249 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1250 	ip->ip_off |= IP_MF;
1251 	ip->ip_off = htons(ip->ip_off);
1252 	ip->ip_sum = 0;
1253 	if (sw_csum & CSUM_DELAY_IP)
1254 		ip->ip_sum = in_cksum(m0, hlen);
1255 
1256 done:
1257 	*m_frag = m0;
1258 	return error;
1259 }
1260 
1261 void
1262 in_delayed_cksum(struct mbuf *m)
1263 {
1264 	struct ip *ip;
1265 	u_short csum, offset;
1266 
1267 	ip = mtod(m, struct ip *);
1268 	offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1269 	csum = in_cksum_skip(m, ip->ip_len, offset);
1270 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1271 		csum = 0xffff;
1272 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1273 
1274 	if (offset + sizeof(u_short) > m->m_len) {
1275 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1276 		    m->m_len, offset, ip->ip_p);
1277 		/*
1278 		 * XXX
1279 		 * this shouldn't happen, but if it does, the
1280 		 * correct behavior may be to insert the checksum
1281 		 * in the existing chain instead of rearranging it.
1282 		 */
1283 		m = m_pullup(m, offset + sizeof(u_short));
1284 	}
1285 	*(u_short *)(m->m_data + offset) = csum;
1286 }
1287 
1288 /*
1289  * Insert IP options into preformed packet.
1290  * Adjust IP destination as required for IP source routing,
1291  * as indicated by a non-zero in_addr at the start of the options.
1292  *
1293  * XXX This routine assumes that the packet has no options in place.
1294  */
1295 static struct mbuf *
1296 ip_insertoptions(m, opt, phlen)
1297 	struct mbuf *m;
1298 	struct mbuf *opt;
1299 	int *phlen;
1300 {
1301 	struct ipoption *p = mtod(opt, struct ipoption *);
1302 	struct mbuf *n;
1303 	struct ip *ip = mtod(m, struct ip *);
1304 	unsigned optlen;
1305 
1306 	optlen = opt->m_len - sizeof(p->ipopt_dst);
1307 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1308 		*phlen = 0;
1309 		return (m);		/* XXX should fail */
1310 	}
1311 	if (p->ipopt_dst.s_addr)
1312 		ip->ip_dst = p->ipopt_dst;
1313 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1314 		MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1315 		if (n == 0) {
1316 			*phlen = 0;
1317 			return (m);
1318 		}
1319 		n->m_pkthdr.rcvif = (struct ifnet *)0;
1320 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1321 		m->m_len -= sizeof(struct ip);
1322 		m->m_data += sizeof(struct ip);
1323 		n->m_next = m;
1324 		m = n;
1325 		m->m_len = optlen + sizeof(struct ip);
1326 		m->m_data += max_linkhdr;
1327 		(void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1328 	} else {
1329 		m->m_data -= optlen;
1330 		m->m_len += optlen;
1331 		m->m_pkthdr.len += optlen;
1332 		ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1333 	}
1334 	ip = mtod(m, struct ip *);
1335 	bcopy(p->ipopt_list, ip + 1, optlen);
1336 	*phlen = sizeof(struct ip) + optlen;
1337 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1338 	ip->ip_len += optlen;
1339 	return (m);
1340 }
1341 
1342 /*
1343  * Copy options from ip to jp,
1344  * omitting those not copied during fragmentation.
1345  */
1346 int
1347 ip_optcopy(ip, jp)
1348 	struct ip *ip, *jp;
1349 {
1350 	u_char *cp, *dp;
1351 	int opt, optlen, cnt;
1352 
1353 	cp = (u_char *)(ip + 1);
1354 	dp = (u_char *)(jp + 1);
1355 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1356 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1357 		opt = cp[0];
1358 		if (opt == IPOPT_EOL)
1359 			break;
1360 		if (opt == IPOPT_NOP) {
1361 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1362 			*dp++ = IPOPT_NOP;
1363 			optlen = 1;
1364 			continue;
1365 		}
1366 
1367 		KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1368 		    ("ip_optcopy: malformed ipv4 option"));
1369 		optlen = cp[IPOPT_OLEN];
1370 		KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1371 		    ("ip_optcopy: malformed ipv4 option"));
1372 
1373 		/* bogus lengths should have been caught by ip_dooptions */
1374 		if (optlen > cnt)
1375 			optlen = cnt;
1376 		if (IPOPT_COPIED(opt)) {
1377 			bcopy(cp, dp, optlen);
1378 			dp += optlen;
1379 		}
1380 	}
1381 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1382 		*dp++ = IPOPT_EOL;
1383 	return (optlen);
1384 }
1385 
1386 /*
1387  * IP socket option processing.
1388  */
1389 int
1390 ip_ctloutput(so, sopt)
1391 	struct socket *so;
1392 	struct sockopt *sopt;
1393 {
1394 	struct	inpcb *inp = sotoinpcb(so);
1395 	int	error, optval;
1396 
1397 	error = optval = 0;
1398 	if (sopt->sopt_level != IPPROTO_IP) {
1399 		return (EINVAL);
1400 	}
1401 
1402 	switch (sopt->sopt_dir) {
1403 	case SOPT_SET:
1404 		switch (sopt->sopt_name) {
1405 		case IP_OPTIONS:
1406 #ifdef notyet
1407 		case IP_RETOPTS:
1408 #endif
1409 		{
1410 			struct mbuf *m;
1411 			if (sopt->sopt_valsize > MLEN) {
1412 				error = EMSGSIZE;
1413 				break;
1414 			}
1415 			MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1416 			if (m == 0) {
1417 				error = ENOBUFS;
1418 				break;
1419 			}
1420 			m->m_len = sopt->sopt_valsize;
1421 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1422 					    m->m_len);
1423 
1424 			return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1425 					   m));
1426 		}
1427 
1428 		case IP_TOS:
1429 		case IP_TTL:
1430 		case IP_RECVOPTS:
1431 		case IP_RECVRETOPTS:
1432 		case IP_RECVDSTADDR:
1433 		case IP_RECVIF:
1434 		case IP_FAITH:
1435 			error = sooptcopyin(sopt, &optval, sizeof optval,
1436 					    sizeof optval);
1437 			if (error)
1438 				break;
1439 
1440 			switch (sopt->sopt_name) {
1441 			case IP_TOS:
1442 				inp->inp_ip_tos = optval;
1443 				break;
1444 
1445 			case IP_TTL:
1446 				inp->inp_ip_ttl = optval;
1447 				break;
1448 #define	OPTSET(bit) \
1449 	if (optval) \
1450 		inp->inp_flags |= bit; \
1451 	else \
1452 		inp->inp_flags &= ~bit;
1453 
1454 			case IP_RECVOPTS:
1455 				OPTSET(INP_RECVOPTS);
1456 				break;
1457 
1458 			case IP_RECVRETOPTS:
1459 				OPTSET(INP_RECVRETOPTS);
1460 				break;
1461 
1462 			case IP_RECVDSTADDR:
1463 				OPTSET(INP_RECVDSTADDR);
1464 				break;
1465 
1466 			case IP_RECVIF:
1467 				OPTSET(INP_RECVIF);
1468 				break;
1469 
1470 			case IP_FAITH:
1471 				OPTSET(INP_FAITH);
1472 				break;
1473 			}
1474 			break;
1475 #undef OPTSET
1476 
1477 		case IP_MULTICAST_IF:
1478 		case IP_MULTICAST_VIF:
1479 		case IP_MULTICAST_TTL:
1480 		case IP_MULTICAST_LOOP:
1481 		case IP_ADD_MEMBERSHIP:
1482 		case IP_DROP_MEMBERSHIP:
1483 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1484 			break;
1485 
1486 		case IP_PORTRANGE:
1487 			error = sooptcopyin(sopt, &optval, sizeof optval,
1488 					    sizeof optval);
1489 			if (error)
1490 				break;
1491 
1492 			switch (optval) {
1493 			case IP_PORTRANGE_DEFAULT:
1494 				inp->inp_flags &= ~(INP_LOWPORT);
1495 				inp->inp_flags &= ~(INP_HIGHPORT);
1496 				break;
1497 
1498 			case IP_PORTRANGE_HIGH:
1499 				inp->inp_flags &= ~(INP_LOWPORT);
1500 				inp->inp_flags |= INP_HIGHPORT;
1501 				break;
1502 
1503 			case IP_PORTRANGE_LOW:
1504 				inp->inp_flags &= ~(INP_HIGHPORT);
1505 				inp->inp_flags |= INP_LOWPORT;
1506 				break;
1507 
1508 			default:
1509 				error = EINVAL;
1510 				break;
1511 			}
1512 			break;
1513 
1514 #if defined(IPSEC) || defined(FAST_IPSEC)
1515 		case IP_IPSEC_POLICY:
1516 		{
1517 			caddr_t req;
1518 			size_t len = 0;
1519 			int priv;
1520 			struct mbuf *m;
1521 			int optname;
1522 
1523 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1524 				break;
1525 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1526 				break;
1527 			priv = (sopt->sopt_td != NULL &&
1528 				suser(sopt->sopt_td) != 0) ? 0 : 1;
1529 			req = mtod(m, caddr_t);
1530 			len = m->m_len;
1531 			optname = sopt->sopt_name;
1532 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1533 			m_freem(m);
1534 			break;
1535 		}
1536 #endif /*IPSEC*/
1537 
1538 		default:
1539 			error = ENOPROTOOPT;
1540 			break;
1541 		}
1542 		break;
1543 
1544 	case SOPT_GET:
1545 		switch (sopt->sopt_name) {
1546 		case IP_OPTIONS:
1547 		case IP_RETOPTS:
1548 			if (inp->inp_options)
1549 				error = sooptcopyout(sopt,
1550 						     mtod(inp->inp_options,
1551 							  char *),
1552 						     inp->inp_options->m_len);
1553 			else
1554 				sopt->sopt_valsize = 0;
1555 			break;
1556 
1557 		case IP_TOS:
1558 		case IP_TTL:
1559 		case IP_RECVOPTS:
1560 		case IP_RECVRETOPTS:
1561 		case IP_RECVDSTADDR:
1562 		case IP_RECVIF:
1563 		case IP_PORTRANGE:
1564 		case IP_FAITH:
1565 			switch (sopt->sopt_name) {
1566 
1567 			case IP_TOS:
1568 				optval = inp->inp_ip_tos;
1569 				break;
1570 
1571 			case IP_TTL:
1572 				optval = inp->inp_ip_ttl;
1573 				break;
1574 
1575 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1576 
1577 			case IP_RECVOPTS:
1578 				optval = OPTBIT(INP_RECVOPTS);
1579 				break;
1580 
1581 			case IP_RECVRETOPTS:
1582 				optval = OPTBIT(INP_RECVRETOPTS);
1583 				break;
1584 
1585 			case IP_RECVDSTADDR:
1586 				optval = OPTBIT(INP_RECVDSTADDR);
1587 				break;
1588 
1589 			case IP_RECVIF:
1590 				optval = OPTBIT(INP_RECVIF);
1591 				break;
1592 
1593 			case IP_PORTRANGE:
1594 				if (inp->inp_flags & INP_HIGHPORT)
1595 					optval = IP_PORTRANGE_HIGH;
1596 				else if (inp->inp_flags & INP_LOWPORT)
1597 					optval = IP_PORTRANGE_LOW;
1598 				else
1599 					optval = 0;
1600 				break;
1601 
1602 			case IP_FAITH:
1603 				optval = OPTBIT(INP_FAITH);
1604 				break;
1605 			}
1606 			error = sooptcopyout(sopt, &optval, sizeof optval);
1607 			break;
1608 
1609 		case IP_MULTICAST_IF:
1610 		case IP_MULTICAST_VIF:
1611 		case IP_MULTICAST_TTL:
1612 		case IP_MULTICAST_LOOP:
1613 		case IP_ADD_MEMBERSHIP:
1614 		case IP_DROP_MEMBERSHIP:
1615 			error = ip_getmoptions(sopt, inp->inp_moptions);
1616 			break;
1617 
1618 #if defined(IPSEC) || defined(FAST_IPSEC)
1619 		case IP_IPSEC_POLICY:
1620 		{
1621 			struct mbuf *m = NULL;
1622 			caddr_t req = NULL;
1623 			size_t len = 0;
1624 
1625 			if (m != 0) {
1626 				req = mtod(m, caddr_t);
1627 				len = m->m_len;
1628 			}
1629 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1630 			if (error == 0)
1631 				error = soopt_mcopyout(sopt, m); /* XXX */
1632 			if (error == 0)
1633 				m_freem(m);
1634 			break;
1635 		}
1636 #endif /*IPSEC*/
1637 
1638 		default:
1639 			error = ENOPROTOOPT;
1640 			break;
1641 		}
1642 		break;
1643 	}
1644 	return (error);
1645 }
1646 
1647 /*
1648  * Set up IP options in pcb for insertion in output packets.
1649  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1650  * with destination address if source routed.
1651  */
1652 static int
1653 ip_pcbopts(optname, pcbopt, m)
1654 	int optname;
1655 	struct mbuf **pcbopt;
1656 	struct mbuf *m;
1657 {
1658 	int cnt, optlen;
1659 	u_char *cp;
1660 	u_char opt;
1661 
1662 	/* turn off any old options */
1663 	if (*pcbopt)
1664 		(void)m_free(*pcbopt);
1665 	*pcbopt = 0;
1666 	if (m == (struct mbuf *)0 || m->m_len == 0) {
1667 		/*
1668 		 * Only turning off any previous options.
1669 		 */
1670 		if (m)
1671 			(void)m_free(m);
1672 		return (0);
1673 	}
1674 
1675 	if (m->m_len % sizeof(int32_t))
1676 		goto bad;
1677 	/*
1678 	 * IP first-hop destination address will be stored before
1679 	 * actual options; move other options back
1680 	 * and clear it when none present.
1681 	 */
1682 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1683 		goto bad;
1684 	cnt = m->m_len;
1685 	m->m_len += sizeof(struct in_addr);
1686 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1687 	ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1688 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1689 
1690 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1691 		opt = cp[IPOPT_OPTVAL];
1692 		if (opt == IPOPT_EOL)
1693 			break;
1694 		if (opt == IPOPT_NOP)
1695 			optlen = 1;
1696 		else {
1697 			if (cnt < IPOPT_OLEN + sizeof(*cp))
1698 				goto bad;
1699 			optlen = cp[IPOPT_OLEN];
1700 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1701 				goto bad;
1702 		}
1703 		switch (opt) {
1704 
1705 		default:
1706 			break;
1707 
1708 		case IPOPT_LSRR:
1709 		case IPOPT_SSRR:
1710 			/*
1711 			 * user process specifies route as:
1712 			 *	->A->B->C->D
1713 			 * D must be our final destination (but we can't
1714 			 * check that since we may not have connected yet).
1715 			 * A is first hop destination, which doesn't appear in
1716 			 * actual IP option, but is stored before the options.
1717 			 */
1718 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1719 				goto bad;
1720 			m->m_len -= sizeof(struct in_addr);
1721 			cnt -= sizeof(struct in_addr);
1722 			optlen -= sizeof(struct in_addr);
1723 			cp[IPOPT_OLEN] = optlen;
1724 			/*
1725 			 * Move first hop before start of options.
1726 			 */
1727 			bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1728 			    sizeof(struct in_addr));
1729 			/*
1730 			 * Then copy rest of options back
1731 			 * to close up the deleted entry.
1732 			 */
1733 			ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1734 			    sizeof(struct in_addr)),
1735 			    &cp[IPOPT_OFFSET+1],
1736 			    (unsigned)cnt - (IPOPT_MINOFF - 1));
1737 			break;
1738 		}
1739 	}
1740 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1741 		goto bad;
1742 	*pcbopt = m;
1743 	return (0);
1744 
1745 bad:
1746 	(void)m_free(m);
1747 	return (EINVAL);
1748 }
1749 
1750 /*
1751  * XXX
1752  * The whole multicast option thing needs to be re-thought.
1753  * Several of these options are equally applicable to non-multicast
1754  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1755  * standard option (IP_TTL).
1756  */
1757 
1758 /*
1759  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1760  */
1761 static struct ifnet *
1762 ip_multicast_if(a, ifindexp)
1763 	struct in_addr *a;
1764 	int *ifindexp;
1765 {
1766 	int ifindex;
1767 	struct ifnet *ifp;
1768 
1769 	if (ifindexp)
1770 		*ifindexp = 0;
1771 	if (ntohl(a->s_addr) >> 24 == 0) {
1772 		ifindex = ntohl(a->s_addr) & 0xffffff;
1773 		if (ifindex < 0 || if_index < ifindex)
1774 			return NULL;
1775 		ifp = ifindex2ifnet[ifindex];
1776 		if (ifindexp)
1777 			*ifindexp = ifindex;
1778 	} else {
1779 		INADDR_TO_IFP(*a, ifp);
1780 	}
1781 	return ifp;
1782 }
1783 
1784 /*
1785  * Set the IP multicast options in response to user setsockopt().
1786  */
1787 static int
1788 ip_setmoptions(sopt, imop)
1789 	struct sockopt *sopt;
1790 	struct ip_moptions **imop;
1791 {
1792 	int error = 0;
1793 	int i;
1794 	struct in_addr addr;
1795 	struct ip_mreq mreq;
1796 	struct ifnet *ifp;
1797 	struct ip_moptions *imo = *imop;
1798 	struct route ro;
1799 	struct sockaddr_in *dst;
1800 	int ifindex;
1801 	int s;
1802 
1803 	if (imo == NULL) {
1804 		/*
1805 		 * No multicast option buffer attached to the pcb;
1806 		 * allocate one and initialize to default values.
1807 		 */
1808 		imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1809 		    M_WAITOK);
1810 
1811 		if (imo == NULL)
1812 			return (ENOBUFS);
1813 		*imop = imo;
1814 		imo->imo_multicast_ifp = NULL;
1815 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1816 		imo->imo_multicast_vif = -1;
1817 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1818 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1819 		imo->imo_num_memberships = 0;
1820 	}
1821 
1822 	switch (sopt->sopt_name) {
1823 	/* store an index number for the vif you wanna use in the send */
1824 	case IP_MULTICAST_VIF:
1825 		if (legal_vif_num == 0) {
1826 			error = EOPNOTSUPP;
1827 			break;
1828 		}
1829 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1830 		if (error)
1831 			break;
1832 		if (!legal_vif_num(i) && (i != -1)) {
1833 			error = EINVAL;
1834 			break;
1835 		}
1836 		imo->imo_multicast_vif = i;
1837 		break;
1838 
1839 	case IP_MULTICAST_IF:
1840 		/*
1841 		 * Select the interface for outgoing multicast packets.
1842 		 */
1843 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1844 		if (error)
1845 			break;
1846 		/*
1847 		 * INADDR_ANY is used to remove a previous selection.
1848 		 * When no interface is selected, a default one is
1849 		 * chosen every time a multicast packet is sent.
1850 		 */
1851 		if (addr.s_addr == INADDR_ANY) {
1852 			imo->imo_multicast_ifp = NULL;
1853 			break;
1854 		}
1855 		/*
1856 		 * The selected interface is identified by its local
1857 		 * IP address.  Find the interface and confirm that
1858 		 * it supports multicasting.
1859 		 */
1860 		s = splimp();
1861 		ifp = ip_multicast_if(&addr, &ifindex);
1862 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1863 			splx(s);
1864 			error = EADDRNOTAVAIL;
1865 			break;
1866 		}
1867 		imo->imo_multicast_ifp = ifp;
1868 		if (ifindex)
1869 			imo->imo_multicast_addr = addr;
1870 		else
1871 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1872 		splx(s);
1873 		break;
1874 
1875 	case IP_MULTICAST_TTL:
1876 		/*
1877 		 * Set the IP time-to-live for outgoing multicast packets.
1878 		 * The original multicast API required a char argument,
1879 		 * which is inconsistent with the rest of the socket API.
1880 		 * We allow either a char or an int.
1881 		 */
1882 		if (sopt->sopt_valsize == 1) {
1883 			u_char ttl;
1884 			error = sooptcopyin(sopt, &ttl, 1, 1);
1885 			if (error)
1886 				break;
1887 			imo->imo_multicast_ttl = ttl;
1888 		} else {
1889 			u_int ttl;
1890 			error = sooptcopyin(sopt, &ttl, sizeof ttl,
1891 					    sizeof ttl);
1892 			if (error)
1893 				break;
1894 			if (ttl > 255)
1895 				error = EINVAL;
1896 			else
1897 				imo->imo_multicast_ttl = ttl;
1898 		}
1899 		break;
1900 
1901 	case IP_MULTICAST_LOOP:
1902 		/*
1903 		 * Set the loopback flag for outgoing multicast packets.
1904 		 * Must be zero or one.  The original multicast API required a
1905 		 * char argument, which is inconsistent with the rest
1906 		 * of the socket API.  We allow either a char or an int.
1907 		 */
1908 		if (sopt->sopt_valsize == 1) {
1909 			u_char loop;
1910 			error = sooptcopyin(sopt, &loop, 1, 1);
1911 			if (error)
1912 				break;
1913 			imo->imo_multicast_loop = !!loop;
1914 		} else {
1915 			u_int loop;
1916 			error = sooptcopyin(sopt, &loop, sizeof loop,
1917 					    sizeof loop);
1918 			if (error)
1919 				break;
1920 			imo->imo_multicast_loop = !!loop;
1921 		}
1922 		break;
1923 
1924 	case IP_ADD_MEMBERSHIP:
1925 		/*
1926 		 * Add a multicast group membership.
1927 		 * Group must be a valid IP multicast address.
1928 		 */
1929 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1930 		if (error)
1931 			break;
1932 
1933 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1934 			error = EINVAL;
1935 			break;
1936 		}
1937 		s = splimp();
1938 		/*
1939 		 * If no interface address was provided, use the interface of
1940 		 * the route to the given multicast address.
1941 		 */
1942 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1943 			bzero((caddr_t)&ro, sizeof(ro));
1944 			dst = (struct sockaddr_in *)&ro.ro_dst;
1945 			dst->sin_len = sizeof(*dst);
1946 			dst->sin_family = AF_INET;
1947 			dst->sin_addr = mreq.imr_multiaddr;
1948 			rtalloc(&ro);
1949 			if (ro.ro_rt == NULL) {
1950 				error = EADDRNOTAVAIL;
1951 				splx(s);
1952 				break;
1953 			}
1954 			ifp = ro.ro_rt->rt_ifp;
1955 			rtfree(ro.ro_rt);
1956 		}
1957 		else {
1958 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1959 		}
1960 
1961 		/*
1962 		 * See if we found an interface, and confirm that it
1963 		 * supports multicast.
1964 		 */
1965 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1966 			error = EADDRNOTAVAIL;
1967 			splx(s);
1968 			break;
1969 		}
1970 		/*
1971 		 * See if the membership already exists or if all the
1972 		 * membership slots are full.
1973 		 */
1974 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1975 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1976 			    imo->imo_membership[i]->inm_addr.s_addr
1977 						== mreq.imr_multiaddr.s_addr)
1978 				break;
1979 		}
1980 		if (i < imo->imo_num_memberships) {
1981 			error = EADDRINUSE;
1982 			splx(s);
1983 			break;
1984 		}
1985 		if (i == IP_MAX_MEMBERSHIPS) {
1986 			error = ETOOMANYREFS;
1987 			splx(s);
1988 			break;
1989 		}
1990 		/*
1991 		 * Everything looks good; add a new record to the multicast
1992 		 * address list for the given interface.
1993 		 */
1994 		if ((imo->imo_membership[i] =
1995 		    in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1996 			error = ENOBUFS;
1997 			splx(s);
1998 			break;
1999 		}
2000 		++imo->imo_num_memberships;
2001 		splx(s);
2002 		break;
2003 
2004 	case IP_DROP_MEMBERSHIP:
2005 		/*
2006 		 * Drop a multicast group membership.
2007 		 * Group must be a valid IP multicast address.
2008 		 */
2009 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2010 		if (error)
2011 			break;
2012 
2013 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2014 			error = EINVAL;
2015 			break;
2016 		}
2017 
2018 		s = splimp();
2019 		/*
2020 		 * If an interface address was specified, get a pointer
2021 		 * to its ifnet structure.
2022 		 */
2023 		if (mreq.imr_interface.s_addr == INADDR_ANY)
2024 			ifp = NULL;
2025 		else {
2026 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2027 			if (ifp == NULL) {
2028 				error = EADDRNOTAVAIL;
2029 				splx(s);
2030 				break;
2031 			}
2032 		}
2033 		/*
2034 		 * Find the membership in the membership array.
2035 		 */
2036 		for (i = 0; i < imo->imo_num_memberships; ++i) {
2037 			if ((ifp == NULL ||
2038 			     imo->imo_membership[i]->inm_ifp == ifp) &&
2039 			     imo->imo_membership[i]->inm_addr.s_addr ==
2040 			     mreq.imr_multiaddr.s_addr)
2041 				break;
2042 		}
2043 		if (i == imo->imo_num_memberships) {
2044 			error = EADDRNOTAVAIL;
2045 			splx(s);
2046 			break;
2047 		}
2048 		/*
2049 		 * Give up the multicast address record to which the
2050 		 * membership points.
2051 		 */
2052 		in_delmulti(imo->imo_membership[i]);
2053 		/*
2054 		 * Remove the gap in the membership array.
2055 		 */
2056 		for (++i; i < imo->imo_num_memberships; ++i)
2057 			imo->imo_membership[i-1] = imo->imo_membership[i];
2058 		--imo->imo_num_memberships;
2059 		splx(s);
2060 		break;
2061 
2062 	default:
2063 		error = EOPNOTSUPP;
2064 		break;
2065 	}
2066 
2067 	/*
2068 	 * If all options have default values, no need to keep the mbuf.
2069 	 */
2070 	if (imo->imo_multicast_ifp == NULL &&
2071 	    imo->imo_multicast_vif == -1 &&
2072 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2073 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2074 	    imo->imo_num_memberships == 0) {
2075 		free(*imop, M_IPMOPTS);
2076 		*imop = NULL;
2077 	}
2078 
2079 	return (error);
2080 }
2081 
2082 /*
2083  * Return the IP multicast options in response to user getsockopt().
2084  */
2085 static int
2086 ip_getmoptions(sopt, imo)
2087 	struct sockopt *sopt;
2088 	struct ip_moptions *imo;
2089 {
2090 	struct in_addr addr;
2091 	struct in_ifaddr *ia;
2092 	int error, optval;
2093 	u_char coptval;
2094 
2095 	error = 0;
2096 	switch (sopt->sopt_name) {
2097 	case IP_MULTICAST_VIF:
2098 		if (imo != NULL)
2099 			optval = imo->imo_multicast_vif;
2100 		else
2101 			optval = -1;
2102 		error = sooptcopyout(sopt, &optval, sizeof optval);
2103 		break;
2104 
2105 	case IP_MULTICAST_IF:
2106 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2107 			addr.s_addr = INADDR_ANY;
2108 		else if (imo->imo_multicast_addr.s_addr) {
2109 			/* return the value user has set */
2110 			addr = imo->imo_multicast_addr;
2111 		} else {
2112 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
2113 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2114 				: IA_SIN(ia)->sin_addr.s_addr;
2115 		}
2116 		error = sooptcopyout(sopt, &addr, sizeof addr);
2117 		break;
2118 
2119 	case IP_MULTICAST_TTL:
2120 		if (imo == 0)
2121 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2122 		else
2123 			optval = coptval = imo->imo_multicast_ttl;
2124 		if (sopt->sopt_valsize == 1)
2125 			error = sooptcopyout(sopt, &coptval, 1);
2126 		else
2127 			error = sooptcopyout(sopt, &optval, sizeof optval);
2128 		break;
2129 
2130 	case IP_MULTICAST_LOOP:
2131 		if (imo == 0)
2132 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2133 		else
2134 			optval = coptval = imo->imo_multicast_loop;
2135 		if (sopt->sopt_valsize == 1)
2136 			error = sooptcopyout(sopt, &coptval, 1);
2137 		else
2138 			error = sooptcopyout(sopt, &optval, sizeof optval);
2139 		break;
2140 
2141 	default:
2142 		error = ENOPROTOOPT;
2143 		break;
2144 	}
2145 	return (error);
2146 }
2147 
2148 /*
2149  * Discard the IP multicast options.
2150  */
2151 void
2152 ip_freemoptions(imo)
2153 	struct ip_moptions *imo;
2154 {
2155 	int i;
2156 
2157 	if (imo != NULL) {
2158 		for (i = 0; i < imo->imo_num_memberships; ++i)
2159 			in_delmulti(imo->imo_membership[i]);
2160 		free(imo, M_IPMOPTS);
2161 	}
2162 }
2163 
2164 /*
2165  * Routine called from ip_output() to loop back a copy of an IP multicast
2166  * packet to the input queue of a specified interface.  Note that this
2167  * calls the output routine of the loopback "driver", but with an interface
2168  * pointer that might NOT be a loopback interface -- evil, but easier than
2169  * replicating that code here.
2170  */
2171 static void
2172 ip_mloopback(ifp, m, dst, hlen)
2173 	struct ifnet *ifp;
2174 	struct mbuf *m;
2175 	struct sockaddr_in *dst;
2176 	int hlen;
2177 {
2178 	struct ip *ip;
2179 	struct mbuf *copym;
2180 
2181 	copym = m_copy(m, 0, M_COPYALL);
2182 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2183 		copym = m_pullup(copym, hlen);
2184 	if (copym != NULL) {
2185 		/*
2186 		 * if the checksum hasn't been computed, mark it as valid
2187 		 */
2188 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2189 			in_delayed_cksum(copym);
2190 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2191 			copym->m_pkthdr.csum_flags |=
2192 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2193 			copym->m_pkthdr.csum_data = 0xffff;
2194 		}
2195 		/*
2196 		 * We don't bother to fragment if the IP length is greater
2197 		 * than the interface's MTU.  Can this possibly matter?
2198 		 */
2199 		ip = mtod(copym, struct ip *);
2200 		ip->ip_len = htons(ip->ip_len);
2201 		ip->ip_off = htons(ip->ip_off);
2202 		ip->ip_sum = 0;
2203 		if (ip->ip_vhl == IP_VHL_BORING) {
2204 			ip->ip_sum = in_cksum_hdr(ip);
2205 		} else {
2206 			ip->ip_sum = in_cksum(copym, hlen);
2207 		}
2208 		/*
2209 		 * NB:
2210 		 * It's not clear whether there are any lingering
2211 		 * reentrancy problems in other areas which might
2212 		 * be exposed by using ip_input directly (in
2213 		 * particular, everything which modifies the packet
2214 		 * in-place).  Yet another option is using the
2215 		 * protosw directly to deliver the looped back
2216 		 * packet.  For the moment, we'll err on the side
2217 		 * of safety by using if_simloop().
2218 		 */
2219 #if 1 /* XXX */
2220 		if (dst->sin_family != AF_INET) {
2221 			printf("ip_mloopback: bad address family %d\n",
2222 						dst->sin_family);
2223 			dst->sin_family = AF_INET;
2224 		}
2225 #endif
2226 
2227 #ifdef notdef
2228 		copym->m_pkthdr.rcvif = ifp;
2229 		ip_input(copym);
2230 #else
2231 		if_simloop(ifp, copym, dst->sin_family, 0);
2232 #endif
2233 	}
2234 }
2235