xref: /dflybsd-src/sys/netinet/ip_output.c (revision f00eae149c338528630cd778fd4de222713daa11)
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  */
32 
33 #define _IP_VHL
34 
35 #include "opt_ipdn.h"
36 #include "opt_ipdivert.h"
37 #include "opt_ipsec.h"
38 #include "opt_mbuf_stress_test.h"
39 #include "opt_mpls.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/proc.h>
50 #include <sys/priv.h>
51 #include <sys/sysctl.h>
52 #include <sys/in_cksum.h>
53 #include <sys/lock.h>
54 
55 #include <sys/thread2.h>
56 #include <sys/mplock2.h>
57 #include <sys/msgport2.h>
58 
59 #include <net/if.h>
60 #include <net/netisr.h>
61 #include <net/pfil.h>
62 #include <net/route.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip_var.h>
70 
71 #include <netproto/mpls/mpls_var.h>
72 
73 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
74 
75 #ifdef IPSEC
76 #include <netinet6/ipsec.h>
77 #include <netproto/key/key.h>
78 #ifdef IPSEC_DEBUG
79 #include <netproto/key/key_debug.h>
80 #else
81 #define	KEYDEBUG(lev,arg)
82 #endif
83 #endif /*IPSEC*/
84 
85 #ifdef FAST_IPSEC
86 #include <netproto/ipsec/ipsec.h>
87 #include <netproto/ipsec/xform.h>
88 #include <netproto/ipsec/key.h>
89 #endif /*FAST_IPSEC*/
90 
91 #include <net/ipfw/ip_fw.h>
92 #include <net/dummynet/ip_dummynet.h>
93 
94 #define print_ip(x, a, y)	 kprintf("%s %d.%d.%d.%d%s",\
95 				x, (ntohl(a.s_addr)>>24)&0xFF,\
96 				  (ntohl(a.s_addr)>>16)&0xFF,\
97 				  (ntohl(a.s_addr)>>8)&0xFF,\
98 				  (ntohl(a.s_addr))&0xFF, y);
99 
100 u_short ip_id;
101 
102 #ifdef MBUF_STRESS_TEST
103 int mbuf_frag_size = 0;
104 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
105 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
106 #endif
107 
108 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
109 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
110 static void	ip_mloopback
111 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
112 static int	ip_getmoptions
113 	(struct sockopt *, struct ip_moptions *);
114 static int	ip_pcbopts(int, struct mbuf **, struct mbuf *);
115 static int	ip_setmoptions
116 	(struct sockopt *, struct ip_moptions **);
117 
118 int	ip_optcopy(struct ip *, struct ip *);
119 
120 extern	int route_assert_owner_access;
121 
122 extern	struct protosw inetsw[];
123 
124 static int
125 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
126 {
127 	struct in_ifaddr_container *iac;
128 
129 	/*
130 	 * We need to figure out if we have been forwarded to a local
131 	 * socket.  If so, then we should somehow "loop back" to
132 	 * ip_input(), and get directed to the PCB as if we had received
133 	 * this packet.  This is because it may be difficult to identify
134 	 * the packets you want to forward until they are being output
135 	 * and have selected an interface (e.g. locally initiated
136 	 * packets).  If we used the loopback inteface, we would not be
137 	 * able to control what happens as the packet runs through
138 	 * ip_input() as it is done through a ISR.
139 	 */
140 	LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
141 		/*
142 		 * If the addr to forward to is one of ours, we pretend
143 		 * to be the destination for this packet.
144 		 */
145 		if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
146 			break;
147 	}
148 	if (iac != NULL) {
149 		struct ip *ip;
150 
151 		if (m->m_pkthdr.rcvif == NULL)
152 			m->m_pkthdr.rcvif = ifunit("lo0");
153 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
154 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
155 						  CSUM_PSEUDO_HDR;
156 			m->m_pkthdr.csum_data = 0xffff;
157 		}
158 		m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
159 
160 		/*
161 		 * Make sure that the IP header is in one mbuf,
162 		 * required by ip_input
163 		 */
164 		if (m->m_len < hlen) {
165 			m = m_pullup(m, hlen);
166 			if (m == NULL) {
167 				/* The packet was freed; we are done */
168 				return 1;
169 			}
170 		}
171 		ip = mtod(m, struct ip *);
172 
173 		ip->ip_len = htons(ip->ip_len);
174 		ip->ip_off = htons(ip->ip_off);
175 		ip_input(m);
176 
177 		return 1; /* The packet gets forwarded locally */
178 	}
179 	return 0;
180 }
181 
182 /*
183  * IP output.  The packet in mbuf chain m contains a skeletal IP
184  * header (with len, off, ttl, proto, tos, src, dst).
185  * The mbuf chain containing the packet will be freed.
186  * The mbuf opt, if present, will not be freed.
187  */
188 int
189 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
190 	  int flags, struct ip_moptions *imo, struct inpcb *inp)
191 {
192 	struct ip *ip;
193 	struct ifnet *ifp = NULL;	/* keep compiler happy */
194 	struct mbuf *m;
195 	int hlen = sizeof(struct ip);
196 	int len, error = 0;
197 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
198 	struct in_ifaddr *ia = NULL;
199 	int isbroadcast, sw_csum;
200 	struct in_addr pkt_dst;
201 	struct route iproute;
202 	struct m_tag *mtag;
203 #ifdef IPSEC
204 	struct secpolicy *sp = NULL;
205 	struct socket *so = inp ? inp->inp_socket : NULL;
206 #endif
207 #ifdef FAST_IPSEC
208 	struct secpolicy *sp = NULL;
209 	struct tdb_ident *tdbi;
210 #endif /* FAST_IPSEC */
211 	struct sockaddr_in *next_hop = NULL;
212 	int src_was_INADDR_ANY = 0;	/* as the name says... */
213 
214 	m = m0;
215 	M_ASSERTPKTHDR(m);
216 
217 	if (ro == NULL) {
218 		ro = &iproute;
219 		bzero(ro, sizeof *ro);
220 	} else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
221 		if (flags & IP_DEBUGROUTE) {
222 			if (route_assert_owner_access) {
223 				panic("ip_output: "
224 				      "rt rt_cpuid %d accessed on cpu %d\n",
225 				      ro->ro_rt->rt_cpuid, mycpuid);
226 			} else {
227 				kprintf("ip_output: "
228 					"rt rt_cpuid %d accessed on cpu %d\n",
229 					ro->ro_rt->rt_cpuid, mycpuid);
230 				print_backtrace(-1);
231 			}
232 		}
233 
234 		/*
235 		 * XXX
236 		 * If the cached rtentry's owner CPU is not the current CPU,
237 		 * then don't touch the cached rtentry (remote free is too
238 		 * expensive in this context); just relocate the route.
239 		 */
240 		ro = &iproute;
241 		bzero(ro, sizeof *ro);
242 	}
243 
244 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
245 		/* Next hop */
246 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
247 		KKASSERT(mtag != NULL);
248 		next_hop = m_tag_data(mtag);
249 	}
250 
251 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
252 		struct dn_pkt *dn_pkt;
253 
254 		/* Extract info from dummynet tag */
255 		mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
256 		KKASSERT(mtag != NULL);
257 		dn_pkt = m_tag_data(mtag);
258 
259 		/*
260 		 * The packet was already tagged, so part of the
261 		 * processing was already done, and we need to go down.
262 		 * Get the calculated parameters from the tag.
263 		 */
264 		ifp = dn_pkt->ifp;
265 
266 		KKASSERT(ro == &iproute);
267 		*ro = dn_pkt->ro; /* structure copy */
268 		KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
269 
270 		dst = dn_pkt->dn_dst;
271 		if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
272 			/* If 'dst' points into dummynet tag, adjust it */
273 			dst = (struct sockaddr_in *)&(ro->ro_dst);
274 		}
275 
276 		ip = mtod(m, struct ip *);
277 		hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
278 		if (ro->ro_rt)
279 			ia = ifatoia(ro->ro_rt->rt_ifa);
280 		goto sendit;
281 	}
282 
283 	if (opt) {
284 		len = 0;
285 		m = ip_insertoptions(m, opt, &len);
286 		if (len != 0)
287 			hlen = len;
288 	}
289 	ip = mtod(m, struct ip *);
290 
291 	/*
292 	 * Fill in IP header.
293 	 */
294 	if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
295 		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
296 		ip->ip_off &= IP_DF;
297 		ip->ip_id = ip_newid();
298 		ipstat.ips_localout++;
299 	} else {
300 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
301 	}
302 
303 reroute:
304 	pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
305 
306 	dst = (struct sockaddr_in *)&ro->ro_dst;
307 	/*
308 	 * If there is a cached route,
309 	 * check that it is to the same destination
310 	 * and is still up.  If not, free it and try again.
311 	 * The address family should also be checked in case of sharing the
312 	 * cache with IPv6.
313 	 */
314 	if (ro->ro_rt &&
315 	    (!(ro->ro_rt->rt_flags & RTF_UP) ||
316 	     dst->sin_family != AF_INET ||
317 	     dst->sin_addr.s_addr != pkt_dst.s_addr)) {
318 		rtfree(ro->ro_rt);
319 		ro->ro_rt = NULL;
320 	}
321 	if (ro->ro_rt == NULL) {
322 		bzero(dst, sizeof *dst);
323 		dst->sin_family = AF_INET;
324 		dst->sin_len = sizeof *dst;
325 		dst->sin_addr = pkt_dst;
326 	}
327 	/*
328 	 * If routing to interface only,
329 	 * short circuit routing lookup.
330 	 */
331 	if (flags & IP_ROUTETOIF) {
332 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
333 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
334 			ipstat.ips_noroute++;
335 			error = ENETUNREACH;
336 			goto bad;
337 		}
338 		ifp = ia->ia_ifp;
339 		ip->ip_ttl = 1;
340 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
341 	} else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
342 		   imo != NULL && imo->imo_multicast_ifp != NULL) {
343 		/*
344 		 * Bypass the normal routing lookup for multicast
345 		 * packets if the interface is specified.
346 		 */
347 		ifp = imo->imo_multicast_ifp;
348 		ia = IFP_TO_IA(ifp);
349 		isbroadcast = 0;	/* fool gcc */
350 	} else {
351 		/*
352 		 * If this is the case, we probably don't want to allocate
353 		 * a protocol-cloned route since we didn't get one from the
354 		 * ULP.  This lets TCP do its thing, while not burdening
355 		 * forwarding or ICMP with the overhead of cloning a route.
356 		 * Of course, we still want to do any cloning requested by
357 		 * the link layer, as this is probably required in all cases
358 		 * for correct operation (as it is for ARP).
359 		 */
360 		if (ro->ro_rt == NULL)
361 			rtalloc_ign(ro, RTF_PRCLONING);
362 		if (ro->ro_rt == NULL) {
363 			ipstat.ips_noroute++;
364 			error = EHOSTUNREACH;
365 			goto bad;
366 		}
367 		ia = ifatoia(ro->ro_rt->rt_ifa);
368 		ifp = ro->ro_rt->rt_ifp;
369 		ro->ro_rt->rt_use++;
370 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
371 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
372 		if (ro->ro_rt->rt_flags & RTF_HOST)
373 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
374 		else
375 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
376 	}
377 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
378 		struct in_multi *inm;
379 
380 		m->m_flags |= M_MCAST;
381 		/*
382 		 * IP destination address is multicast.  Make sure "dst"
383 		 * still points to the address in "ro".  (It may have been
384 		 * changed to point to a gateway address, above.)
385 		 */
386 		dst = (struct sockaddr_in *)&ro->ro_dst;
387 		/*
388 		 * See if the caller provided any multicast options
389 		 */
390 		if (imo != NULL) {
391 			ip->ip_ttl = imo->imo_multicast_ttl;
392 			if (imo->imo_multicast_vif != -1) {
393 				ip->ip_src.s_addr =
394 				    ip_mcast_src ?
395 				    ip_mcast_src(imo->imo_multicast_vif) :
396 				    INADDR_ANY;
397 			}
398 		} else {
399 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
400 		}
401 		/*
402 		 * Confirm that the outgoing interface supports multicast.
403 		 */
404 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
405 			if (!(ifp->if_flags & IFF_MULTICAST)) {
406 				ipstat.ips_noroute++;
407 				error = ENETUNREACH;
408 				goto bad;
409 			}
410 		}
411 		/*
412 		 * If source address not specified yet, use address
413 		 * of outgoing interface.
414 		 */
415 		if (ip->ip_src.s_addr == INADDR_ANY) {
416 			/* Interface may have no addresses. */
417 			if (ia != NULL)
418 				ip->ip_src = IA_SIN(ia)->sin_addr;
419 		}
420 
421 		IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
422 		if (inm != NULL &&
423 		    (imo == NULL || imo->imo_multicast_loop)) {
424 			/*
425 			 * If we belong to the destination multicast group
426 			 * on the outgoing interface, and the caller did not
427 			 * forbid loopback, loop back a copy.
428 			 */
429 			ip_mloopback(ifp, m, dst, hlen);
430 		} else {
431 			/*
432 			 * If we are acting as a multicast router, perform
433 			 * multicast forwarding as if the packet had just
434 			 * arrived on the interface to which we are about
435 			 * to send.  The multicast forwarding function
436 			 * recursively calls this function, using the
437 			 * IP_FORWARDING flag to prevent infinite recursion.
438 			 *
439 			 * Multicasts that are looped back by ip_mloopback(),
440 			 * above, will be forwarded by the ip_input() routine,
441 			 * if necessary.
442 			 */
443 			if (ip_mrouter && !(flags & IP_FORWARDING)) {
444 				/*
445 				 * If rsvp daemon is not running, do not
446 				 * set ip_moptions. This ensures that the packet
447 				 * is multicast and not just sent down one link
448 				 * as prescribed by rsvpd.
449 				 */
450 				if (!rsvp_on)
451 					imo = NULL;
452 				if (ip_mforward) {
453 					get_mplock();
454 					if (ip_mforward(ip, ifp, m, imo) != 0) {
455 						m_freem(m);
456 						rel_mplock();
457 						goto done;
458 					}
459 					rel_mplock();
460 				}
461 			}
462 		}
463 
464 		/*
465 		 * Multicasts with a time-to-live of zero may be looped-
466 		 * back, above, but must not be transmitted on a network.
467 		 * Also, multicasts addressed to the loopback interface
468 		 * are not sent -- the above call to ip_mloopback() will
469 		 * loop back a copy if this host actually belongs to the
470 		 * destination group on the loopback interface.
471 		 */
472 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
473 			m_freem(m);
474 			goto done;
475 		}
476 
477 		goto sendit;
478 	} else {
479 		m->m_flags &= ~M_MCAST;
480 	}
481 
482 	/*
483 	 * If the source address is not specified yet, use the address
484 	 * of the outoing interface. In case, keep note we did that, so
485 	 * if the the firewall changes the next-hop causing the output
486 	 * interface to change, we can fix that.
487 	 */
488 	if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
489 		/* Interface may have no addresses. */
490 		if (ia != NULL) {
491 			ip->ip_src = IA_SIN(ia)->sin_addr;
492 			src_was_INADDR_ANY = 1;
493 		}
494 	}
495 
496 	/*
497 	 * Look for broadcast address and
498 	 * verify user is allowed to send
499 	 * such a packet.
500 	 */
501 	if (isbroadcast) {
502 		if (!(ifp->if_flags & IFF_BROADCAST)) {
503 			error = EADDRNOTAVAIL;
504 			goto bad;
505 		}
506 		if (!(flags & IP_ALLOWBROADCAST)) {
507 			error = EACCES;
508 			goto bad;
509 		}
510 		/* don't allow broadcast messages to be fragmented */
511 		if (ip->ip_len > ifp->if_mtu) {
512 			error = EMSGSIZE;
513 			goto bad;
514 		}
515 		m->m_flags |= M_BCAST;
516 	} else {
517 		m->m_flags &= ~M_BCAST;
518 	}
519 
520 sendit:
521 #ifdef IPSEC
522 	/* get SP for this packet */
523 	if (so == NULL)
524 		sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
525 	else
526 		sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
527 
528 	if (sp == NULL) {
529 		ipsecstat.out_inval++;
530 		goto bad;
531 	}
532 
533 	error = 0;
534 
535 	/* check policy */
536 	switch (sp->policy) {
537 	case IPSEC_POLICY_DISCARD:
538 		/*
539 		 * This packet is just discarded.
540 		 */
541 		ipsecstat.out_polvio++;
542 		goto bad;
543 
544 	case IPSEC_POLICY_BYPASS:
545 	case IPSEC_POLICY_NONE:
546 	case IPSEC_POLICY_TCP:
547 		/* no need to do IPsec. */
548 		goto skip_ipsec;
549 
550 	case IPSEC_POLICY_IPSEC:
551 		if (sp->req == NULL) {
552 			/* acquire a policy */
553 			error = key_spdacquire(sp);
554 			goto bad;
555 		}
556 		break;
557 
558 	case IPSEC_POLICY_ENTRUST:
559 	default:
560 		kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
561 	}
562     {
563 	struct ipsec_output_state state;
564 	bzero(&state, sizeof state);
565 	state.m = m;
566 	if (flags & IP_ROUTETOIF) {
567 		state.ro = &iproute;
568 		bzero(&iproute, sizeof iproute);
569 	} else
570 		state.ro = ro;
571 	state.dst = (struct sockaddr *)dst;
572 
573 	ip->ip_sum = 0;
574 
575 	/*
576 	 * XXX
577 	 * delayed checksums are not currently compatible with IPsec
578 	 */
579 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
580 		in_delayed_cksum(m);
581 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
582 	}
583 
584 	ip->ip_len = htons(ip->ip_len);
585 	ip->ip_off = htons(ip->ip_off);
586 
587 	error = ipsec4_output(&state, sp, flags);
588 
589 	m = state.m;
590 	if (flags & IP_ROUTETOIF) {
591 		/*
592 		 * if we have tunnel mode SA, we may need to ignore
593 		 * IP_ROUTETOIF.
594 		 */
595 		if (state.ro != &iproute || state.ro->ro_rt != NULL) {
596 			flags &= ~IP_ROUTETOIF;
597 			ro = state.ro;
598 		}
599 	} else
600 		ro = state.ro;
601 	dst = (struct sockaddr_in *)state.dst;
602 	if (error) {
603 		/* mbuf is already reclaimed in ipsec4_output. */
604 		m0 = NULL;
605 		switch (error) {
606 		case EHOSTUNREACH:
607 		case ENETUNREACH:
608 		case EMSGSIZE:
609 		case ENOBUFS:
610 		case ENOMEM:
611 			break;
612 		default:
613 			kprintf("ip4_output (ipsec): error code %d\n", error);
614 			/*fall through*/
615 		case ENOENT:
616 			/* don't show these error codes to the user */
617 			error = 0;
618 			break;
619 		}
620 		goto bad;
621 	}
622     }
623 
624 	/* be sure to update variables that are affected by ipsec4_output() */
625 	ip = mtod(m, struct ip *);
626 #ifdef _IP_VHL
627 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
628 #else
629 	hlen = ip->ip_hl << 2;
630 #endif
631 	if (ro->ro_rt == NULL) {
632 		if (!(flags & IP_ROUTETOIF)) {
633 			kprintf("ip_output: "
634 				"can't update route after IPsec processing\n");
635 			error = EHOSTUNREACH;	/*XXX*/
636 			goto bad;
637 		}
638 	} else {
639 		ia = ifatoia(ro->ro_rt->rt_ifa);
640 		ifp = ro->ro_rt->rt_ifp;
641 	}
642 
643 	/* make it flipped, again. */
644 	ip->ip_len = ntohs(ip->ip_len);
645 	ip->ip_off = ntohs(ip->ip_off);
646 skip_ipsec:
647 #endif /*IPSEC*/
648 #ifdef FAST_IPSEC
649 	/*
650 	 * Check the security policy (SP) for the packet and, if
651 	 * required, do IPsec-related processing.  There are two
652 	 * cases here; the first time a packet is sent through
653 	 * it will be untagged and handled by ipsec4_checkpolicy.
654 	 * If the packet is resubmitted to ip_output (e.g. after
655 	 * AH, ESP, etc. processing), there will be a tag to bypass
656 	 * the lookup and related policy checking.
657 	 */
658 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
659 	crit_enter();
660 	if (mtag != NULL) {
661 		tdbi = (struct tdb_ident *)m_tag_data(mtag);
662 		sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
663 		if (sp == NULL)
664 			error = -EINVAL;	/* force silent drop */
665 		m_tag_delete(m, mtag);
666 	} else {
667 		sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
668 					&error, inp);
669 	}
670 	/*
671 	 * There are four return cases:
672 	 *    sp != NULL		    apply IPsec policy
673 	 *    sp == NULL, error == 0	    no IPsec handling needed
674 	 *    sp == NULL, error == -EINVAL  discard packet w/o error
675 	 *    sp == NULL, error != 0	    discard packet, report error
676 	 */
677 	if (sp != NULL) {
678 		/* Loop detection, check if ipsec processing already done */
679 		KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
680 		for (mtag = m_tag_first(m); mtag != NULL;
681 		     mtag = m_tag_next(m, mtag)) {
682 			if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
683 				continue;
684 			if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
685 			    mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
686 				continue;
687 			/*
688 			 * Check if policy has an SA associated with it.
689 			 * This can happen when an SP has yet to acquire
690 			 * an SA; e.g. on first reference.  If it occurs,
691 			 * then we let ipsec4_process_packet do its thing.
692 			 */
693 			if (sp->req->sav == NULL)
694 				break;
695 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
696 			if (tdbi->spi == sp->req->sav->spi &&
697 			    tdbi->proto == sp->req->sav->sah->saidx.proto &&
698 			    bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
699 				 sizeof(union sockaddr_union)) == 0) {
700 				/*
701 				 * No IPsec processing is needed, free
702 				 * reference to SP.
703 				 *
704 				 * NB: null pointer to avoid free at
705 				 *     done: below.
706 				 */
707 				KEY_FREESP(&sp), sp = NULL;
708 				crit_exit();
709 				goto spd_done;
710 			}
711 		}
712 
713 		/*
714 		 * Do delayed checksums now because we send before
715 		 * this is done in the normal processing path.
716 		 */
717 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
718 			in_delayed_cksum(m);
719 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
720 		}
721 
722 		ip->ip_len = htons(ip->ip_len);
723 		ip->ip_off = htons(ip->ip_off);
724 
725 		/* NB: callee frees mbuf */
726 		error = ipsec4_process_packet(m, sp->req, flags, 0);
727 		/*
728 		 * Preserve KAME behaviour: ENOENT can be returned
729 		 * when an SA acquire is in progress.  Don't propagate
730 		 * this to user-level; it confuses applications.
731 		 *
732 		 * XXX this will go away when the SADB is redone.
733 		 */
734 		if (error == ENOENT)
735 			error = 0;
736 		crit_exit();
737 		goto done;
738 	} else {
739 		crit_exit();
740 
741 		if (error != 0) {
742 			/*
743 			 * Hack: -EINVAL is used to signal that a packet
744 			 * should be silently discarded.  This is typically
745 			 * because we asked key management for an SA and
746 			 * it was delayed (e.g. kicked up to IKE).
747 			 */
748 			if (error == -EINVAL)
749 				error = 0;
750 			goto bad;
751 		} else {
752 			/* No IPsec processing for this packet. */
753 		}
754 #ifdef notyet
755 		/*
756 		 * If deferred crypto processing is needed, check that
757 		 * the interface supports it.
758 		 */
759 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
760 		if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
761 			/* notify IPsec to do its own crypto */
762 			ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
763 			error = EHOSTUNREACH;
764 			goto bad;
765 		}
766 #endif
767 	}
768 spd_done:
769 #endif /* FAST_IPSEC */
770 
771 	/* We are already being fwd'd from a firewall. */
772 	if (next_hop != NULL)
773 		goto pass;
774 
775 	/* No pfil hooks */
776 	if (!pfil_has_hooks(&inet_pfil_hook)) {
777 		if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
778 			/*
779 			 * Strip dummynet tags from stranded packets
780 			 */
781 			mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
782 			KKASSERT(mtag != NULL);
783 			m_tag_delete(m, mtag);
784 			m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
785 		}
786 		goto pass;
787 	}
788 
789 	/*
790 	 * IpHack's section.
791 	 * - Xlate: translate packet's addr/port (NAT).
792 	 * - Firewall: deny/allow/etc.
793 	 * - Wrap: fake packet's addr/port <unimpl.>
794 	 * - Encapsulate: put it in another IP and send out. <unimp.>
795 	 */
796 
797 	/*
798 	 * Run through list of hooks for output packets.
799 	 */
800 	error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
801 	if (error != 0 || m == NULL)
802 		goto done;
803 	ip = mtod(m, struct ip *);
804 
805 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
806 		/*
807 		 * Check dst to make sure it is directly reachable on the
808 		 * interface we previously thought it was.
809 		 * If it isn't (which may be likely in some situations) we have
810 		 * to re-route it (ie, find a route for the next-hop and the
811 		 * associated interface) and set them here. This is nested
812 		 * forwarding which in most cases is undesirable, except where
813 		 * such control is nigh impossible. So we do it here.
814 		 * And I'm babbling.
815 		 */
816 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
817 		KKASSERT(mtag != NULL);
818 		next_hop = m_tag_data(mtag);
819 
820 		/*
821 		 * Try local forwarding first
822 		 */
823 		if (ip_localforward(m, next_hop, hlen))
824 			goto done;
825 
826 		/*
827 		 * Relocate the route based on next_hop.
828 		 * If the current route is inp's cache, keep it untouched.
829 		 */
830 		if (ro == &iproute && ro->ro_rt != NULL) {
831 			RTFREE(ro->ro_rt);
832 			ro->ro_rt = NULL;
833 		}
834 		ro = &iproute;
835 		bzero(ro, sizeof *ro);
836 
837 		/*
838 		 * Forwarding to broadcast address is not allowed.
839 		 * XXX Should we follow IP_ROUTETOIF?
840 		 */
841 		flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
842 
843 		/* We are doing forwarding now */
844 		flags |= IP_FORWARDING;
845 
846 		goto reroute;
847 	}
848 
849 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
850 		struct dn_pkt *dn_pkt;
851 
852 		mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
853 		KKASSERT(mtag != NULL);
854 		dn_pkt = m_tag_data(mtag);
855 
856 		/*
857 		 * Under certain cases it is not possible to recalculate
858 		 * 'ro' and 'dst', let alone 'flags', so just save them in
859 		 * dummynet tag and avoid the possible wrong reculcalation
860 		 * when we come back to ip_output() again.
861 		 *
862 		 * All other parameters have been already used and so they
863 		 * are not needed anymore.
864 		 * XXX if the ifp is deleted while a pkt is in dummynet,
865 		 * we are in trouble! (TODO use ifnet_detach_event)
866 		 *
867 		 * We need to copy *ro because for ICMP pkts (and maybe
868 		 * others) the caller passed a pointer into the stack;
869 		 * dst might also be a pointer into *ro so it needs to
870 		 * be updated.
871 		 */
872 		dn_pkt->ro = *ro;
873 		if (ro->ro_rt)
874 			ro->ro_rt->rt_refcnt++;
875 		if (dst == (struct sockaddr_in *)&ro->ro_dst) {
876 			/* 'dst' points into 'ro' */
877 			dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
878 		}
879 		dn_pkt->dn_dst = dst;
880 		dn_pkt->flags = flags;
881 
882 		ip_dn_queue(m);
883 		goto done;
884 	}
885 pass:
886 	/* 127/8 must not appear on wire - RFC1122. */
887 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
888 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
889 		if (!(ifp->if_flags & IFF_LOOPBACK)) {
890 			ipstat.ips_badaddr++;
891 			error = EADDRNOTAVAIL;
892 			goto bad;
893 		}
894 	}
895 
896 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
897 		m->m_pkthdr.csum_flags |= CSUM_IP;
898 		sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
899 		if (sw_csum & CSUM_DELAY_DATA) {
900 			in_delayed_cksum(m);
901 			sw_csum &= ~CSUM_DELAY_DATA;
902 		}
903 		m->m_pkthdr.csum_flags &= ifp->if_hwassist;
904 	} else {
905 		sw_csum = 0;
906 	}
907 	m->m_pkthdr.csum_iphlen = hlen;
908 
909 	/*
910 	 * If small enough for interface, or the interface will take
911 	 * care of the fragmentation or segmentation for us, can just
912 	 * send directly.
913 	 */
914 	if (ip->ip_len <= ifp->if_mtu ||
915 	    ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
916 	    (m->m_pkthdr.csum_flags & CSUM_TSO)) {
917 		ip->ip_len = htons(ip->ip_len);
918 		ip->ip_off = htons(ip->ip_off);
919 		ip->ip_sum = 0;
920 		if (sw_csum & CSUM_DELAY_IP) {
921 			if (ip->ip_vhl == IP_VHL_BORING)
922 				ip->ip_sum = in_cksum_hdr(ip);
923 			else
924 				ip->ip_sum = in_cksum(m, hlen);
925 		}
926 
927 		/* Record statistics for this interface address. */
928 		if (!(flags & IP_FORWARDING) && ia) {
929 			IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
930 			IFA_STAT_INC(&ia->ia_ifa, obytes, m->m_pkthdr.len);
931 		}
932 
933 #ifdef IPSEC
934 		/* clean ipsec history once it goes out of the node */
935 		ipsec_delaux(m);
936 #endif
937 
938 #ifdef MBUF_STRESS_TEST
939 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
940 			struct mbuf *m1, *m2;
941 			int length, tmp;
942 
943 			tmp = length = m->m_pkthdr.len;
944 
945 			while ((length -= mbuf_frag_size) >= 1) {
946 				m1 = m_split(m, length, MB_DONTWAIT);
947 				if (m1 == NULL)
948 					break;
949 				m2 = m;
950 				while (m2->m_next != NULL)
951 					m2 = m2->m_next;
952 				m2->m_next = m1;
953 			}
954 			m->m_pkthdr.len = tmp;
955 		}
956 #endif
957 
958 #ifdef MPLS
959 		if (!mpls_output_process(m, ro->ro_rt))
960 			goto done;
961 #endif
962 		error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
963 				       ro->ro_rt);
964 		goto done;
965 	}
966 
967 	if (ip->ip_off & IP_DF) {
968 		error = EMSGSIZE;
969 		/*
970 		 * This case can happen if the user changed the MTU
971 		 * of an interface after enabling IP on it.  Because
972 		 * most netifs don't keep track of routes pointing to
973 		 * them, there is no way for one to update all its
974 		 * routes when the MTU is changed.
975 		 */
976 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
977 		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
978 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
979 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
980 		}
981 		ipstat.ips_cantfrag++;
982 		goto bad;
983 	}
984 
985 	/*
986 	 * Too large for interface; fragment if possible. If successful,
987 	 * on return, m will point to a list of packets to be sent.
988 	 */
989 	error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
990 	if (error)
991 		goto bad;
992 	for (; m; m = m0) {
993 		m0 = m->m_nextpkt;
994 		m->m_nextpkt = NULL;
995 #ifdef IPSEC
996 		/* clean ipsec history once it goes out of the node */
997 		ipsec_delaux(m);
998 #endif
999 		if (error == 0) {
1000 			/* Record statistics for this interface address. */
1001 			if (ia != NULL) {
1002 				IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
1003 				IFA_STAT_INC(&ia->ia_ifa, obytes,
1004 				    m->m_pkthdr.len);
1005 			}
1006 #ifdef MPLS
1007 			if (!mpls_output_process(m, ro->ro_rt))
1008 				continue;
1009 #endif
1010 			error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1011 					       ro->ro_rt);
1012 		} else {
1013 			m_freem(m);
1014 		}
1015 	}
1016 
1017 	if (error == 0)
1018 		ipstat.ips_fragmented++;
1019 
1020 done:
1021 	if (ro == &iproute && ro->ro_rt != NULL) {
1022 		RTFREE(ro->ro_rt);
1023 		ro->ro_rt = NULL;
1024 	}
1025 #ifdef IPSEC
1026 	if (sp != NULL) {
1027 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1028 			kprintf("DP ip_output call free SP:%p\n", sp));
1029 		key_freesp(sp);
1030 	}
1031 #endif
1032 #ifdef FAST_IPSEC
1033 	if (sp != NULL)
1034 		KEY_FREESP(&sp);
1035 #endif
1036 	return (error);
1037 bad:
1038 	m_freem(m);
1039 	goto done;
1040 }
1041 
1042 /*
1043  * Create a chain of fragments which fit the given mtu. m_frag points to the
1044  * mbuf to be fragmented; on return it points to the chain with the fragments.
1045  * Return 0 if no error. If error, m_frag may contain a partially built
1046  * chain of fragments that should be freed by the caller.
1047  *
1048  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1049  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1050  */
1051 int
1052 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1053 	    u_long if_hwassist_flags, int sw_csum)
1054 {
1055 	int error = 0;
1056 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1057 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
1058 	int off;
1059 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
1060 	int firstlen;
1061 	struct mbuf **mnext;
1062 	int nfrags;
1063 
1064 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
1065 		ipstat.ips_cantfrag++;
1066 		return EMSGSIZE;
1067 	}
1068 
1069 	/*
1070 	 * Must be able to put at least 8 bytes per fragment.
1071 	 */
1072 	if (len < 8)
1073 		return EMSGSIZE;
1074 
1075 	/*
1076 	 * If the interface will not calculate checksums on
1077 	 * fragmented packets, then do it here.
1078 	 */
1079 	if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1080 	    !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1081 		in_delayed_cksum(m0);
1082 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1083 	}
1084 
1085 	if (len > PAGE_SIZE) {
1086 		/*
1087 		 * Fragment large datagrams such that each segment
1088 		 * contains a multiple of PAGE_SIZE amount of data,
1089 		 * plus headers. This enables a receiver to perform
1090 		 * page-flipping zero-copy optimizations.
1091 		 *
1092 		 * XXX When does this help given that sender and receiver
1093 		 * could have different page sizes, and also mtu could
1094 		 * be less than the receiver's page size ?
1095 		 */
1096 		int newlen;
1097 		struct mbuf *m;
1098 
1099 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1100 			off += m->m_len;
1101 
1102 		/*
1103 		 * firstlen (off - hlen) must be aligned on an
1104 		 * 8-byte boundary
1105 		 */
1106 		if (off < hlen)
1107 			goto smart_frag_failure;
1108 		off = ((off - hlen) & ~7) + hlen;
1109 		newlen = (~PAGE_MASK) & mtu;
1110 		if ((newlen + sizeof(struct ip)) > mtu) {
1111 			/* we failed, go back the default */
1112 smart_frag_failure:
1113 			newlen = len;
1114 			off = hlen + len;
1115 		}
1116 		len = newlen;
1117 
1118 	} else {
1119 		off = hlen + len;
1120 	}
1121 
1122 	firstlen = off - hlen;
1123 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
1124 
1125 	/*
1126 	 * Loop through length of segment after first fragment,
1127 	 * make new header and copy data of each part and link onto chain.
1128 	 * Here, m0 is the original packet, m is the fragment being created.
1129 	 * The fragments are linked off the m_nextpkt of the original
1130 	 * packet, which after processing serves as the first fragment.
1131 	 */
1132 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1133 		struct ip *mhip;	/* ip header on the fragment */
1134 		struct mbuf *m;
1135 		int mhlen = sizeof(struct ip);
1136 
1137 		MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1138 		if (m == NULL) {
1139 			error = ENOBUFS;
1140 			ipstat.ips_odropped++;
1141 			goto done;
1142 		}
1143 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1144 		/*
1145 		 * In the first mbuf, leave room for the link header, then
1146 		 * copy the original IP header including options. The payload
1147 		 * goes into an additional mbuf chain returned by m_copy().
1148 		 */
1149 		m->m_data += max_linkhdr;
1150 		mhip = mtod(m, struct ip *);
1151 		*mhip = *ip;
1152 		if (hlen > sizeof(struct ip)) {
1153 			mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1154 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1155 		}
1156 		m->m_len = mhlen;
1157 		/* XXX do we need to add ip->ip_off below ? */
1158 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1159 		if (off + len >= ip->ip_len) {	/* last fragment */
1160 			len = ip->ip_len - off;
1161 			m->m_flags |= M_LASTFRAG;
1162 		} else
1163 			mhip->ip_off |= IP_MF;
1164 		mhip->ip_len = htons((u_short)(len + mhlen));
1165 		m->m_next = m_copy(m0, off, len);
1166 		if (m->m_next == NULL) {		/* copy failed */
1167 			m_free(m);
1168 			error = ENOBUFS;	/* ??? */
1169 			ipstat.ips_odropped++;
1170 			goto done;
1171 		}
1172 		m->m_pkthdr.len = mhlen + len;
1173 		m->m_pkthdr.rcvif = NULL;
1174 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1175 		m->m_pkthdr.csum_iphlen = mhlen;
1176 		mhip->ip_off = htons(mhip->ip_off);
1177 		mhip->ip_sum = 0;
1178 		if (sw_csum & CSUM_DELAY_IP)
1179 			mhip->ip_sum = in_cksum(m, mhlen);
1180 		*mnext = m;
1181 		mnext = &m->m_nextpkt;
1182 	}
1183 	ipstat.ips_ofragments += nfrags;
1184 
1185 	/* set first marker for fragment chain */
1186 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1187 	m0->m_pkthdr.csum_data = nfrags;
1188 
1189 	/*
1190 	 * Update first fragment by trimming what's been copied out
1191 	 * and updating header.
1192 	 */
1193 	m_adj(m0, hlen + firstlen - ip->ip_len);
1194 	m0->m_pkthdr.len = hlen + firstlen;
1195 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1196 	ip->ip_off |= IP_MF;
1197 	ip->ip_off = htons(ip->ip_off);
1198 	ip->ip_sum = 0;
1199 	if (sw_csum & CSUM_DELAY_IP)
1200 		ip->ip_sum = in_cksum(m0, hlen);
1201 
1202 done:
1203 	*m_frag = m0;
1204 	return error;
1205 }
1206 
1207 void
1208 in_delayed_cksum(struct mbuf *m)
1209 {
1210 	struct ip *ip;
1211 	u_short csum, offset;
1212 
1213 	ip = mtod(m, struct ip *);
1214 	offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1215 	csum = in_cksum_skip(m, ip->ip_len, offset);
1216 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1217 		csum = 0xffff;
1218 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
1219 
1220 	if (offset + sizeof(u_short) > m->m_len) {
1221 		kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1222 		    m->m_len, offset, ip->ip_p);
1223 		/*
1224 		 * XXX
1225 		 * this shouldn't happen, but if it does, the
1226 		 * correct behavior may be to insert the checksum
1227 		 * in the existing chain instead of rearranging it.
1228 		 */
1229 		m = m_pullup(m, offset + sizeof(u_short));
1230 	}
1231 	*(u_short *)(m->m_data + offset) = csum;
1232 }
1233 
1234 /*
1235  * Insert IP options into preformed packet.
1236  * Adjust IP destination as required for IP source routing,
1237  * as indicated by a non-zero in_addr at the start of the options.
1238  *
1239  * XXX This routine assumes that the packet has no options in place.
1240  */
1241 static struct mbuf *
1242 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1243 {
1244 	struct ipoption *p = mtod(opt, struct ipoption *);
1245 	struct mbuf *n;
1246 	struct ip *ip = mtod(m, struct ip *);
1247 	unsigned optlen;
1248 
1249 	optlen = opt->m_len - sizeof p->ipopt_dst;
1250 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1251 		*phlen = 0;
1252 		return (m);		/* XXX should fail */
1253 	}
1254 	if (p->ipopt_dst.s_addr)
1255 		ip->ip_dst = p->ipopt_dst;
1256 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1257 		MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1258 		if (n == NULL) {
1259 			*phlen = 0;
1260 			return (m);
1261 		}
1262 		n->m_pkthdr.rcvif = NULL;
1263 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1264 		m->m_len -= sizeof(struct ip);
1265 		m->m_data += sizeof(struct ip);
1266 		n->m_next = m;
1267 		m = n;
1268 		m->m_len = optlen + sizeof(struct ip);
1269 		m->m_data += max_linkhdr;
1270 		memcpy(mtod(m, void *), ip, sizeof(struct ip));
1271 	} else {
1272 		m->m_data -= optlen;
1273 		m->m_len += optlen;
1274 		m->m_pkthdr.len += optlen;
1275 		ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1276 	}
1277 	ip = mtod(m, struct ip *);
1278 	bcopy(p->ipopt_list, ip + 1, optlen);
1279 	*phlen = sizeof(struct ip) + optlen;
1280 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1281 	ip->ip_len += optlen;
1282 	return (m);
1283 }
1284 
1285 /*
1286  * Copy options from ip to jp,
1287  * omitting those not copied during fragmentation.
1288  */
1289 int
1290 ip_optcopy(struct ip *ip, struct ip *jp)
1291 {
1292 	u_char *cp, *dp;
1293 	int opt, optlen, cnt;
1294 
1295 	cp = (u_char *)(ip + 1);
1296 	dp = (u_char *)(jp + 1);
1297 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1298 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1299 		opt = cp[0];
1300 		if (opt == IPOPT_EOL)
1301 			break;
1302 		if (opt == IPOPT_NOP) {
1303 			/* Preserve for IP mcast tunnel's LSRR alignment. */
1304 			*dp++ = IPOPT_NOP;
1305 			optlen = 1;
1306 			continue;
1307 		}
1308 
1309 		KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1310 		    ("ip_optcopy: malformed ipv4 option"));
1311 		optlen = cp[IPOPT_OLEN];
1312 		KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1313 		    ("ip_optcopy: malformed ipv4 option"));
1314 
1315 		/* bogus lengths should have been caught by ip_dooptions */
1316 		if (optlen > cnt)
1317 			optlen = cnt;
1318 		if (IPOPT_COPIED(opt)) {
1319 			bcopy(cp, dp, optlen);
1320 			dp += optlen;
1321 		}
1322 	}
1323 	for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1324 		*dp++ = IPOPT_EOL;
1325 	return (optlen);
1326 }
1327 
1328 /*
1329  * IP socket option processing.
1330  */
1331 void
1332 ip_ctloutput(netmsg_t msg)
1333 {
1334 	struct socket *so = msg->base.nm_so;
1335 	struct sockopt *sopt = msg->ctloutput.nm_sopt;
1336 	struct	inpcb *inp = so->so_pcb;
1337 	int	error, optval;
1338 
1339 	error = optval = 0;
1340 	if (sopt->sopt_level != IPPROTO_IP) {
1341 		error = EINVAL;
1342 		goto done;
1343 	}
1344 
1345 	switch (sopt->sopt_dir) {
1346 	case SOPT_SET:
1347 		switch (sopt->sopt_name) {
1348 		case IP_OPTIONS:
1349 #ifdef notyet
1350 		case IP_RETOPTS:
1351 #endif
1352 		{
1353 			struct mbuf *m;
1354 			if (sopt->sopt_valsize > MLEN) {
1355 				error = EMSGSIZE;
1356 				break;
1357 			}
1358 			MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1359 			if (m == NULL) {
1360 				error = ENOBUFS;
1361 				break;
1362 			}
1363 			m->m_len = sopt->sopt_valsize;
1364 			error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1365 					      m->m_len);
1366 			error = ip_pcbopts(sopt->sopt_name,
1367 					   &inp->inp_options, m);
1368 			goto done;
1369 		}
1370 
1371 		case IP_TOS:
1372 		case IP_TTL:
1373 		case IP_MINTTL:
1374 		case IP_RECVOPTS:
1375 		case IP_RECVRETOPTS:
1376 		case IP_RECVDSTADDR:
1377 		case IP_RECVIF:
1378 		case IP_RECVTTL:
1379 		case IP_FAITH:
1380 			error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1381 					     sizeof optval);
1382 			if (error)
1383 				break;
1384 			switch (sopt->sopt_name) {
1385 			case IP_TOS:
1386 				inp->inp_ip_tos = optval;
1387 				break;
1388 
1389 			case IP_TTL:
1390 				inp->inp_ip_ttl = optval;
1391 				break;
1392 			case IP_MINTTL:
1393 				if (optval >= 0 && optval <= MAXTTL)
1394 					inp->inp_ip_minttl = optval;
1395 				else
1396 					error = EINVAL;
1397 				break;
1398 #define	OPTSET(bit) \
1399 	if (optval) \
1400 		inp->inp_flags |= bit; \
1401 	else \
1402 		inp->inp_flags &= ~bit;
1403 
1404 			case IP_RECVOPTS:
1405 				OPTSET(INP_RECVOPTS);
1406 				break;
1407 
1408 			case IP_RECVRETOPTS:
1409 				OPTSET(INP_RECVRETOPTS);
1410 				break;
1411 
1412 			case IP_RECVDSTADDR:
1413 				OPTSET(INP_RECVDSTADDR);
1414 				break;
1415 
1416 			case IP_RECVIF:
1417 				OPTSET(INP_RECVIF);
1418 				break;
1419 
1420 			case IP_RECVTTL:
1421 				OPTSET(INP_RECVTTL);
1422 				break;
1423 
1424 			case IP_FAITH:
1425 				OPTSET(INP_FAITH);
1426 				break;
1427 			}
1428 			break;
1429 #undef OPTSET
1430 
1431 		case IP_MULTICAST_IF:
1432 		case IP_MULTICAST_VIF:
1433 		case IP_MULTICAST_TTL:
1434 		case IP_MULTICAST_LOOP:
1435 		case IP_ADD_MEMBERSHIP:
1436 		case IP_DROP_MEMBERSHIP:
1437 			error = ip_setmoptions(sopt, &inp->inp_moptions);
1438 			break;
1439 
1440 		case IP_PORTRANGE:
1441 			error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1442 					    sizeof optval);
1443 			if (error)
1444 				break;
1445 
1446 			switch (optval) {
1447 			case IP_PORTRANGE_DEFAULT:
1448 				inp->inp_flags &= ~(INP_LOWPORT);
1449 				inp->inp_flags &= ~(INP_HIGHPORT);
1450 				break;
1451 
1452 			case IP_PORTRANGE_HIGH:
1453 				inp->inp_flags &= ~(INP_LOWPORT);
1454 				inp->inp_flags |= INP_HIGHPORT;
1455 				break;
1456 
1457 			case IP_PORTRANGE_LOW:
1458 				inp->inp_flags &= ~(INP_HIGHPORT);
1459 				inp->inp_flags |= INP_LOWPORT;
1460 				break;
1461 
1462 			default:
1463 				error = EINVAL;
1464 				break;
1465 			}
1466 			break;
1467 
1468 #if defined(IPSEC) || defined(FAST_IPSEC)
1469 		case IP_IPSEC_POLICY:
1470 		{
1471 			caddr_t req;
1472 			size_t len = 0;
1473 			int priv;
1474 			struct mbuf *m;
1475 			int optname;
1476 
1477 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1478 				break;
1479 			soopt_to_mbuf(sopt, m);
1480 			priv = (sopt->sopt_td != NULL &&
1481 				priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1482 			req = mtod(m, caddr_t);
1483 			len = m->m_len;
1484 			optname = sopt->sopt_name;
1485 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1486 			m_freem(m);
1487 			break;
1488 		}
1489 #endif /*IPSEC*/
1490 
1491 		default:
1492 			error = ENOPROTOOPT;
1493 			break;
1494 		}
1495 		break;
1496 
1497 	case SOPT_GET:
1498 		switch (sopt->sopt_name) {
1499 		case IP_OPTIONS:
1500 		case IP_RETOPTS:
1501 			if (inp->inp_options)
1502 				soopt_from_kbuf(sopt, mtod(inp->inp_options,
1503 							   char *),
1504 						inp->inp_options->m_len);
1505 			else
1506 				sopt->sopt_valsize = 0;
1507 			break;
1508 
1509 		case IP_TOS:
1510 		case IP_TTL:
1511 		case IP_MINTTL:
1512 		case IP_RECVOPTS:
1513 		case IP_RECVRETOPTS:
1514 		case IP_RECVDSTADDR:
1515 		case IP_RECVTTL:
1516 		case IP_RECVIF:
1517 		case IP_PORTRANGE:
1518 		case IP_FAITH:
1519 			switch (sopt->sopt_name) {
1520 
1521 			case IP_TOS:
1522 				optval = inp->inp_ip_tos;
1523 				break;
1524 
1525 			case IP_TTL:
1526 				optval = inp->inp_ip_ttl;
1527 				break;
1528 			case IP_MINTTL:
1529 				optval = inp->inp_ip_minttl;
1530 				break;
1531 
1532 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1533 
1534 			case IP_RECVOPTS:
1535 				optval = OPTBIT(INP_RECVOPTS);
1536 				break;
1537 
1538 			case IP_RECVRETOPTS:
1539 				optval = OPTBIT(INP_RECVRETOPTS);
1540 				break;
1541 
1542 			case IP_RECVDSTADDR:
1543 				optval = OPTBIT(INP_RECVDSTADDR);
1544 				break;
1545 
1546 			case IP_RECVTTL:
1547 				optval = OPTBIT(INP_RECVTTL);
1548 				break;
1549 
1550 			case IP_RECVIF:
1551 				optval = OPTBIT(INP_RECVIF);
1552 				break;
1553 
1554 			case IP_PORTRANGE:
1555 				if (inp->inp_flags & INP_HIGHPORT)
1556 					optval = IP_PORTRANGE_HIGH;
1557 				else if (inp->inp_flags & INP_LOWPORT)
1558 					optval = IP_PORTRANGE_LOW;
1559 				else
1560 					optval = 0;
1561 				break;
1562 
1563 			case IP_FAITH:
1564 				optval = OPTBIT(INP_FAITH);
1565 				break;
1566 			}
1567 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1568 			break;
1569 
1570 		case IP_MULTICAST_IF:
1571 		case IP_MULTICAST_VIF:
1572 		case IP_MULTICAST_TTL:
1573 		case IP_MULTICAST_LOOP:
1574 		case IP_ADD_MEMBERSHIP:
1575 		case IP_DROP_MEMBERSHIP:
1576 			error = ip_getmoptions(sopt, inp->inp_moptions);
1577 			break;
1578 
1579 #if defined(IPSEC) || defined(FAST_IPSEC)
1580 		case IP_IPSEC_POLICY:
1581 		{
1582 			struct mbuf *m = NULL;
1583 			caddr_t req = NULL;
1584 			size_t len = 0;
1585 
1586 			if (m != NULL) {
1587 				req = mtod(m, caddr_t);
1588 				len = m->m_len;
1589 			}
1590 			error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1591 			if (error == 0)
1592 				error = soopt_from_mbuf(sopt, m); /* XXX */
1593 			if (error == 0)
1594 				m_freem(m);
1595 			break;
1596 		}
1597 #endif /*IPSEC*/
1598 
1599 		default:
1600 			error = ENOPROTOOPT;
1601 			break;
1602 		}
1603 		break;
1604 	}
1605 done:
1606 	lwkt_replymsg(&msg->lmsg, error);
1607 }
1608 
1609 /*
1610  * Set up IP options in pcb for insertion in output packets.
1611  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1612  * with destination address if source routed.
1613  */
1614 static int
1615 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1616 {
1617 	int cnt, optlen;
1618 	u_char *cp;
1619 	u_char opt;
1620 
1621 	/* turn off any old options */
1622 	if (*pcbopt)
1623 		m_free(*pcbopt);
1624 	*pcbopt = NULL;
1625 	if (m == NULL || m->m_len == 0) {
1626 		/*
1627 		 * Only turning off any previous options.
1628 		 */
1629 		if (m != NULL)
1630 			m_free(m);
1631 		return (0);
1632 	}
1633 
1634 	if (m->m_len % sizeof(int32_t))
1635 		goto bad;
1636 	/*
1637 	 * IP first-hop destination address will be stored before
1638 	 * actual options; move other options back
1639 	 * and clear it when none present.
1640 	 */
1641 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1642 		goto bad;
1643 	cnt = m->m_len;
1644 	m->m_len += sizeof(struct in_addr);
1645 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
1646 	ovbcopy(mtod(m, caddr_t), cp, cnt);
1647 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1648 
1649 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1650 		opt = cp[IPOPT_OPTVAL];
1651 		if (opt == IPOPT_EOL)
1652 			break;
1653 		if (opt == IPOPT_NOP)
1654 			optlen = 1;
1655 		else {
1656 			if (cnt < IPOPT_OLEN + sizeof *cp)
1657 				goto bad;
1658 			optlen = cp[IPOPT_OLEN];
1659 			if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1660 				goto bad;
1661 		}
1662 		switch (opt) {
1663 
1664 		default:
1665 			break;
1666 
1667 		case IPOPT_LSRR:
1668 		case IPOPT_SSRR:
1669 			/*
1670 			 * user process specifies route as:
1671 			 *	->A->B->C->D
1672 			 * D must be our final destination (but we can't
1673 			 * check that since we may not have connected yet).
1674 			 * A is first hop destination, which doesn't appear in
1675 			 * actual IP option, but is stored before the options.
1676 			 */
1677 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1678 				goto bad;
1679 			m->m_len -= sizeof(struct in_addr);
1680 			cnt -= sizeof(struct in_addr);
1681 			optlen -= sizeof(struct in_addr);
1682 			cp[IPOPT_OLEN] = optlen;
1683 			/*
1684 			 * Move first hop before start of options.
1685 			 */
1686 			bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1687 			      sizeof(struct in_addr));
1688 			/*
1689 			 * Then copy rest of options back
1690 			 * to close up the deleted entry.
1691 			 */
1692 			ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1693 				&cp[IPOPT_OFFSET+1],
1694 				cnt - (IPOPT_MINOFF - 1));
1695 			break;
1696 		}
1697 	}
1698 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1699 		goto bad;
1700 	*pcbopt = m;
1701 	return (0);
1702 
1703 bad:
1704 	m_free(m);
1705 	return (EINVAL);
1706 }
1707 
1708 /*
1709  * XXX
1710  * The whole multicast option thing needs to be re-thought.
1711  * Several of these options are equally applicable to non-multicast
1712  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1713  * standard option (IP_TTL).
1714  */
1715 
1716 /*
1717  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1718  */
1719 static struct ifnet *
1720 ip_multicast_if(struct in_addr *a, int *ifindexp)
1721 {
1722 	int ifindex;
1723 	struct ifnet *ifp;
1724 
1725 	if (ifindexp)
1726 		*ifindexp = 0;
1727 	if (ntohl(a->s_addr) >> 24 == 0) {
1728 		ifindex = ntohl(a->s_addr) & 0xffffff;
1729 		if (ifindex < 0 || if_index < ifindex)
1730 			return NULL;
1731 		ifp = ifindex2ifnet[ifindex];
1732 		if (ifindexp)
1733 			*ifindexp = ifindex;
1734 	} else {
1735 		ifp = INADDR_TO_IFP(a);
1736 	}
1737 	return ifp;
1738 }
1739 
1740 /*
1741  * Set the IP multicast options in response to user setsockopt().
1742  */
1743 static int
1744 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1745 {
1746 	int error = 0;
1747 	int i;
1748 	struct in_addr addr;
1749 	struct ip_mreq mreq;
1750 	struct ifnet *ifp;
1751 	struct ip_moptions *imo = *imop;
1752 	int ifindex;
1753 
1754 	if (imo == NULL) {
1755 		/*
1756 		 * No multicast option buffer attached to the pcb;
1757 		 * allocate one and initialize to default values.
1758 		 */
1759 		imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1760 
1761 		*imop = imo;
1762 		imo->imo_multicast_ifp = NULL;
1763 		imo->imo_multicast_addr.s_addr = INADDR_ANY;
1764 		imo->imo_multicast_vif = -1;
1765 		imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1766 		imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1767 		imo->imo_num_memberships = 0;
1768 	}
1769 	switch (sopt->sopt_name) {
1770 	/* store an index number for the vif you wanna use in the send */
1771 	case IP_MULTICAST_VIF:
1772 		if (legal_vif_num == 0) {
1773 			error = EOPNOTSUPP;
1774 			break;
1775 		}
1776 		error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1777 		if (error)
1778 			break;
1779 		if (!legal_vif_num(i) && (i != -1)) {
1780 			error = EINVAL;
1781 			break;
1782 		}
1783 		imo->imo_multicast_vif = i;
1784 		break;
1785 
1786 	case IP_MULTICAST_IF:
1787 		/*
1788 		 * Select the interface for outgoing multicast packets.
1789 		 */
1790 		error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1791 		if (error)
1792 			break;
1793 
1794 		/*
1795 		 * INADDR_ANY is used to remove a previous selection.
1796 		 * When no interface is selected, a default one is
1797 		 * chosen every time a multicast packet is sent.
1798 		 */
1799 		if (addr.s_addr == INADDR_ANY) {
1800 			imo->imo_multicast_ifp = NULL;
1801 			break;
1802 		}
1803 		/*
1804 		 * The selected interface is identified by its local
1805 		 * IP address.  Find the interface and confirm that
1806 		 * it supports multicasting.
1807 		 */
1808 		crit_enter();
1809 		ifp = ip_multicast_if(&addr, &ifindex);
1810 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1811 			crit_exit();
1812 			error = EADDRNOTAVAIL;
1813 			break;
1814 		}
1815 		imo->imo_multicast_ifp = ifp;
1816 		if (ifindex)
1817 			imo->imo_multicast_addr = addr;
1818 		else
1819 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1820 		crit_exit();
1821 		break;
1822 
1823 	case IP_MULTICAST_TTL:
1824 		/*
1825 		 * Set the IP time-to-live for outgoing multicast packets.
1826 		 * The original multicast API required a char argument,
1827 		 * which is inconsistent with the rest of the socket API.
1828 		 * We allow either a char or an int.
1829 		 */
1830 		if (sopt->sopt_valsize == 1) {
1831 			u_char ttl;
1832 			error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1833 			if (error)
1834 				break;
1835 			imo->imo_multicast_ttl = ttl;
1836 		} else {
1837 			u_int ttl;
1838 			error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1839 			if (error)
1840 				break;
1841 			if (ttl > 255)
1842 				error = EINVAL;
1843 			else
1844 				imo->imo_multicast_ttl = ttl;
1845 		}
1846 		break;
1847 
1848 	case IP_MULTICAST_LOOP:
1849 		/*
1850 		 * Set the loopback flag for outgoing multicast packets.
1851 		 * Must be zero or one.  The original multicast API required a
1852 		 * char argument, which is inconsistent with the rest
1853 		 * of the socket API.  We allow either a char or an int.
1854 		 */
1855 		if (sopt->sopt_valsize == 1) {
1856 			u_char loop;
1857 
1858 			error = soopt_to_kbuf(sopt, &loop, 1, 1);
1859 			if (error)
1860 				break;
1861 			imo->imo_multicast_loop = !!loop;
1862 		} else {
1863 			u_int loop;
1864 
1865 			error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1866 					    sizeof loop);
1867 			if (error)
1868 				break;
1869 			imo->imo_multicast_loop = !!loop;
1870 		}
1871 		break;
1872 
1873 	case IP_ADD_MEMBERSHIP:
1874 		/*
1875 		 * Add a multicast group membership.
1876 		 * Group must be a valid IP multicast address.
1877 		 */
1878 		error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1879 		if (error)
1880 			break;
1881 
1882 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1883 			error = EINVAL;
1884 			break;
1885 		}
1886 		crit_enter();
1887 		/*
1888 		 * If no interface address was provided, use the interface of
1889 		 * the route to the given multicast address.
1890 		 */
1891 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1892 			struct sockaddr_in dst;
1893 			struct rtentry *rt;
1894 
1895 			bzero(&dst, sizeof(struct sockaddr_in));
1896 			dst.sin_len = sizeof(struct sockaddr_in);
1897 			dst.sin_family = AF_INET;
1898 			dst.sin_addr = mreq.imr_multiaddr;
1899 			rt = rtlookup((struct sockaddr *)&dst);
1900 			if (rt == NULL) {
1901 				error = EADDRNOTAVAIL;
1902 				crit_exit();
1903 				break;
1904 			}
1905 			--rt->rt_refcnt;
1906 			ifp = rt->rt_ifp;
1907 		} else {
1908 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1909 		}
1910 
1911 		/*
1912 		 * See if we found an interface, and confirm that it
1913 		 * supports multicast.
1914 		 */
1915 		if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1916 			error = EADDRNOTAVAIL;
1917 			crit_exit();
1918 			break;
1919 		}
1920 		/*
1921 		 * See if the membership already exists or if all the
1922 		 * membership slots are full.
1923 		 */
1924 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1925 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1926 			    imo->imo_membership[i]->inm_addr.s_addr
1927 						== mreq.imr_multiaddr.s_addr)
1928 				break;
1929 		}
1930 		if (i < imo->imo_num_memberships) {
1931 			error = EADDRINUSE;
1932 			crit_exit();
1933 			break;
1934 		}
1935 		if (i == IP_MAX_MEMBERSHIPS) {
1936 			error = ETOOMANYREFS;
1937 			crit_exit();
1938 			break;
1939 		}
1940 		/*
1941 		 * Everything looks good; add a new record to the multicast
1942 		 * address list for the given interface.
1943 		 */
1944 		if ((imo->imo_membership[i] =
1945 		     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1946 			error = ENOBUFS;
1947 			crit_exit();
1948 			break;
1949 		}
1950 		++imo->imo_num_memberships;
1951 		crit_exit();
1952 		break;
1953 
1954 	case IP_DROP_MEMBERSHIP:
1955 		/*
1956 		 * Drop a multicast group membership.
1957 		 * Group must be a valid IP multicast address.
1958 		 */
1959 		error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1960 		if (error)
1961 			break;
1962 
1963 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1964 			error = EINVAL;
1965 			break;
1966 		}
1967 
1968 		crit_enter();
1969 		/*
1970 		 * If an interface address was specified, get a pointer
1971 		 * to its ifnet structure.
1972 		 */
1973 		if (mreq.imr_interface.s_addr == INADDR_ANY)
1974 			ifp = NULL;
1975 		else {
1976 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1977 			if (ifp == NULL) {
1978 				error = EADDRNOTAVAIL;
1979 				crit_exit();
1980 				break;
1981 			}
1982 		}
1983 		/*
1984 		 * Find the membership in the membership array.
1985 		 */
1986 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1987 			if ((ifp == NULL ||
1988 			     imo->imo_membership[i]->inm_ifp == ifp) &&
1989 			    imo->imo_membership[i]->inm_addr.s_addr ==
1990 			    mreq.imr_multiaddr.s_addr)
1991 				break;
1992 		}
1993 		if (i == imo->imo_num_memberships) {
1994 			error = EADDRNOTAVAIL;
1995 			crit_exit();
1996 			break;
1997 		}
1998 		/*
1999 		 * Give up the multicast address record to which the
2000 		 * membership points.
2001 		 */
2002 		in_delmulti(imo->imo_membership[i]);
2003 		/*
2004 		 * Remove the gap in the membership array.
2005 		 */
2006 		for (++i; i < imo->imo_num_memberships; ++i)
2007 			imo->imo_membership[i-1] = imo->imo_membership[i];
2008 		--imo->imo_num_memberships;
2009 		crit_exit();
2010 		break;
2011 
2012 	default:
2013 		error = EOPNOTSUPP;
2014 		break;
2015 	}
2016 
2017 	/*
2018 	 * If all options have default values, no need to keep the mbuf.
2019 	 */
2020 	if (imo->imo_multicast_ifp == NULL &&
2021 	    imo->imo_multicast_vif == -1 &&
2022 	    imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2023 	    imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2024 	    imo->imo_num_memberships == 0) {
2025 		kfree(*imop, M_IPMOPTS);
2026 		*imop = NULL;
2027 	}
2028 
2029 	return (error);
2030 }
2031 
2032 /*
2033  * Return the IP multicast options in response to user getsockopt().
2034  */
2035 static int
2036 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2037 {
2038 	struct in_addr addr;
2039 	struct in_ifaddr *ia;
2040 	int error, optval;
2041 	u_char coptval;
2042 
2043 	error = 0;
2044 	switch (sopt->sopt_name) {
2045 	case IP_MULTICAST_VIF:
2046 		if (imo != NULL)
2047 			optval = imo->imo_multicast_vif;
2048 		else
2049 			optval = -1;
2050 		soopt_from_kbuf(sopt, &optval, sizeof optval);
2051 		break;
2052 
2053 	case IP_MULTICAST_IF:
2054 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
2055 			addr.s_addr = INADDR_ANY;
2056 		else if (imo->imo_multicast_addr.s_addr) {
2057 			/* return the value user has set */
2058 			addr = imo->imo_multicast_addr;
2059 		} else {
2060 			ia = IFP_TO_IA(imo->imo_multicast_ifp);
2061 			addr.s_addr = (ia == NULL) ? INADDR_ANY
2062 				: IA_SIN(ia)->sin_addr.s_addr;
2063 		}
2064 		soopt_from_kbuf(sopt, &addr, sizeof addr);
2065 		break;
2066 
2067 	case IP_MULTICAST_TTL:
2068 		if (imo == NULL)
2069 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2070 		else
2071 			optval = coptval = imo->imo_multicast_ttl;
2072 		if (sopt->sopt_valsize == 1)
2073 			soopt_from_kbuf(sopt, &coptval, 1);
2074 		else
2075 			soopt_from_kbuf(sopt, &optval, sizeof optval);
2076 		break;
2077 
2078 	case IP_MULTICAST_LOOP:
2079 		if (imo == NULL)
2080 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2081 		else
2082 			optval = coptval = imo->imo_multicast_loop;
2083 		if (sopt->sopt_valsize == 1)
2084 			soopt_from_kbuf(sopt, &coptval, 1);
2085 		else
2086 			soopt_from_kbuf(sopt, &optval, sizeof optval);
2087 		break;
2088 
2089 	default:
2090 		error = ENOPROTOOPT;
2091 		break;
2092 	}
2093 	return (error);
2094 }
2095 
2096 /*
2097  * Discard the IP multicast options.
2098  */
2099 void
2100 ip_freemoptions(struct ip_moptions *imo)
2101 {
2102 	int i;
2103 
2104 	if (imo != NULL) {
2105 		for (i = 0; i < imo->imo_num_memberships; ++i)
2106 			in_delmulti(imo->imo_membership[i]);
2107 		kfree(imo, M_IPMOPTS);
2108 	}
2109 }
2110 
2111 /*
2112  * Routine called from ip_output() to loop back a copy of an IP multicast
2113  * packet to the input queue of a specified interface.  Note that this
2114  * calls the output routine of the loopback "driver", but with an interface
2115  * pointer that might NOT be a loopback interface -- evil, but easier than
2116  * replicating that code here.
2117  */
2118 static void
2119 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2120 	     int hlen)
2121 {
2122 	struct ip *ip;
2123 	struct mbuf *copym;
2124 
2125 	copym = m_copypacket(m, MB_DONTWAIT);
2126 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2127 		copym = m_pullup(copym, hlen);
2128 	if (copym != NULL) {
2129 		/*
2130 		 * if the checksum hasn't been computed, mark it as valid
2131 		 */
2132 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2133 			in_delayed_cksum(copym);
2134 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2135 			copym->m_pkthdr.csum_flags |=
2136 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2137 			copym->m_pkthdr.csum_data = 0xffff;
2138 		}
2139 		/*
2140 		 * We don't bother to fragment if the IP length is greater
2141 		 * than the interface's MTU.  Can this possibly matter?
2142 		 */
2143 		ip = mtod(copym, struct ip *);
2144 		ip->ip_len = htons(ip->ip_len);
2145 		ip->ip_off = htons(ip->ip_off);
2146 		ip->ip_sum = 0;
2147 		if (ip->ip_vhl == IP_VHL_BORING) {
2148 			ip->ip_sum = in_cksum_hdr(ip);
2149 		} else {
2150 			ip->ip_sum = in_cksum(copym, hlen);
2151 		}
2152 		/*
2153 		 * NB:
2154 		 * It's not clear whether there are any lingering
2155 		 * reentrancy problems in other areas which might
2156 		 * be exposed by using ip_input directly (in
2157 		 * particular, everything which modifies the packet
2158 		 * in-place).  Yet another option is using the
2159 		 * protosw directly to deliver the looped back
2160 		 * packet.  For the moment, we'll err on the side
2161 		 * of safety by using if_simloop().
2162 		 */
2163 #if 1 /* XXX */
2164 		if (dst->sin_family != AF_INET) {
2165 			kprintf("ip_mloopback: bad address family %d\n",
2166 						dst->sin_family);
2167 			dst->sin_family = AF_INET;
2168 		}
2169 #endif
2170 		get_mplock();	/* is if_simloop() mpsafe yet? */
2171 		if_simloop(ifp, copym, dst->sin_family, 0);
2172 		rel_mplock();
2173 	}
2174 }
2175