xref: /freebsd/sys/netinet/ip_output.c (revision bfe691b2f75de2224c7ceb304ebcdef2b42d4179)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
30  * $FreeBSD$
31  */
32 
33 #include "opt_ipfw.h"
34 #include "opt_ipsec.h"
35 #include "opt_mac.h"
36 #include "opt_mbuf_stress_test.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/priv.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/sysctl.h>
48 
49 #include <net/if.h>
50 #include <net/netisr.h>
51 #include <net/pfil.h>
52 #include <net/route.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/ip_options.h>
61 
62 #if defined(IPSEC) || defined(FAST_IPSEC)
63 #include <netinet/ip_ipsec.h>
64 #ifdef IPSEC
65 #include <netinet6/ipsec.h>
66 #endif
67 #ifdef FAST_IPSEC
68 #include <netipsec/ipsec.h>
69 #endif
70 #endif /*IPSEC*/
71 
72 #include <machine/in_cksum.h>
73 
74 #include <security/mac/mac_framework.h>
75 
76 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
77 
78 #define print_ip(x, a, y)	 printf("%s %d.%d.%d.%d%s",\
79 				x, (ntohl(a.s_addr)>>24)&0xFF,\
80 				  (ntohl(a.s_addr)>>16)&0xFF,\
81 				  (ntohl(a.s_addr)>>8)&0xFF,\
82 				  (ntohl(a.s_addr))&0xFF, y);
83 
84 u_short ip_id;
85 
86 #ifdef MBUF_STRESS_TEST
87 int mbuf_frag_size = 0;
88 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
89 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
90 #endif
91 
92 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
93 static void	ip_mloopback
94 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
95 static int	ip_getmoptions(struct inpcb *, struct sockopt *);
96 static int	ip_setmoptions(struct inpcb *, struct sockopt *);
97 
98 
99 extern	struct protosw inetsw[];
100 
101 /*
102  * IP output.  The packet in mbuf chain m contains a skeletal IP
103  * header (with len, off, ttl, proto, tos, src, dst).
104  * The mbuf chain containing the packet will be freed.
105  * The mbuf opt, if present, will not be freed.
106  * In the IP forwarding case, the packet will arrive with options already
107  * inserted, so must have a NULL opt pointer.
108  */
109 int
110 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro,
111 	int flags, struct ip_moptions *imo, struct inpcb *inp)
112 {
113 	struct ip *ip;
114 	struct ifnet *ifp = NULL;	/* keep compiler happy */
115 	struct mbuf *m0;
116 	int hlen = sizeof (struct ip);
117 	int mtu;
118 	int len, error = 0;
119 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
120 	struct in_ifaddr *ia = NULL;
121 	int isbroadcast, sw_csum;
122 	struct route iproute;
123 	struct in_addr odst;
124 #ifdef IPFIREWALL_FORWARD
125 	struct m_tag *fwd_tag = NULL;
126 #endif
127 	M_ASSERTPKTHDR(m);
128 
129 	if (ro == NULL) {
130 		ro = &iproute;
131 		bzero(ro, sizeof (*ro));
132 	}
133 
134 	if (inp != NULL)
135 		INP_LOCK_ASSERT(inp);
136 
137 	if (opt) {
138 		len = 0;
139 		m = ip_insertoptions(m, opt, &len);
140 		if (len != 0)
141 			hlen = len;
142 	}
143 	ip = mtod(m, struct ip *);
144 
145 	/*
146 	 * Fill in IP header.  If we are not allowing fragmentation,
147 	 * then the ip_id field is meaningless, but we don't set it
148 	 * to zero.  Doing so causes various problems when devices along
149 	 * the path (routers, load balancers, firewalls, etc.) illegally
150 	 * disable DF on our packet.  Note that a 16-bit counter
151 	 * will wrap around in less than 10 seconds at 100 Mbit/s on a
152 	 * medium with MTU 1500.  See Steven M. Bellovin, "A Technique
153 	 * for Counting NATted Hosts", Proc. IMW'02, available at
154 	 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>.
155 	 */
156 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
157 		ip->ip_v = IPVERSION;
158 		ip->ip_hl = hlen >> 2;
159 		ip->ip_id = ip_newid();
160 		ipstat.ips_localout++;
161 	} else {
162 		hlen = ip->ip_hl << 2;
163 	}
164 
165 	dst = (struct sockaddr_in *)&ro->ro_dst;
166 again:
167 	/*
168 	 * If there is a cached route,
169 	 * check that it is to the same destination
170 	 * and is still up.  If not, free it and try again.
171 	 * The address family should also be checked in case of sharing the
172 	 * cache with IPv6.
173 	 */
174 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
175 			  dst->sin_family != AF_INET ||
176 			  dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
177 		RTFREE(ro->ro_rt);
178 		ro->ro_rt = (struct rtentry *)NULL;
179 	}
180 #ifdef IPFIREWALL_FORWARD
181 	if (ro->ro_rt == NULL && fwd_tag == NULL) {
182 #else
183 	if (ro->ro_rt == NULL) {
184 #endif
185 		bzero(dst, sizeof(*dst));
186 		dst->sin_family = AF_INET;
187 		dst->sin_len = sizeof(*dst);
188 		dst->sin_addr = ip->ip_dst;
189 	}
190 	/*
191 	 * If routing to interface only, short circuit routing lookup.
192 	 * The use of an all-ones broadcast address implies this; an
193 	 * interface is specified by the broadcast address of an interface,
194 	 * or the destination address of a ptp interface.
195 	 */
196 	if (flags & IP_SENDONES) {
197 		if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL &&
198 		    (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) {
199 			ipstat.ips_noroute++;
200 			error = ENETUNREACH;
201 			goto bad;
202 		}
203 		ip->ip_dst.s_addr = INADDR_BROADCAST;
204 		dst->sin_addr = ip->ip_dst;
205 		ifp = ia->ia_ifp;
206 		ip->ip_ttl = 1;
207 		isbroadcast = 1;
208 	} else if (flags & IP_ROUTETOIF) {
209 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
210 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
211 			ipstat.ips_noroute++;
212 			error = ENETUNREACH;
213 			goto bad;
214 		}
215 		ifp = ia->ia_ifp;
216 		ip->ip_ttl = 1;
217 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
218 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
219 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
220 		/*
221 		 * Bypass the normal routing lookup for multicast
222 		 * packets if the interface is specified.
223 		 */
224 		ifp = imo->imo_multicast_ifp;
225 		IFP_TO_IA(ifp, ia);
226 		isbroadcast = 0;	/* fool gcc */
227 	} else {
228 		/*
229 		 * We want to do any cloning requested by the link layer,
230 		 * as this is probably required in all cases for correct
231 		 * operation (as it is for ARP).
232 		 */
233 		if (ro->ro_rt == NULL)
234 			rtalloc_ign(ro, 0);
235 		if (ro->ro_rt == NULL) {
236 			ipstat.ips_noroute++;
237 			error = EHOSTUNREACH;
238 			goto bad;
239 		}
240 		ia = ifatoia(ro->ro_rt->rt_ifa);
241 		ifp = ro->ro_rt->rt_ifp;
242 		ro->ro_rt->rt_rmx.rmx_pksent++;
243 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
244 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
245 		if (ro->ro_rt->rt_flags & RTF_HOST)
246 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
247 		else
248 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
249 	}
250 	/*
251 	 * Calculate MTU.  If we have a route that is up, use that,
252 	 * otherwise use the interface's MTU.
253 	 */
254 	if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) {
255 		/*
256 		 * This case can happen if the user changed the MTU
257 		 * of an interface after enabling IP on it.  Because
258 		 * most netifs don't keep track of routes pointing to
259 		 * them, there is no way for one to update all its
260 		 * routes when the MTU is changed.
261 		 */
262 		if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)
263 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
264 		mtu = ro->ro_rt->rt_rmx.rmx_mtu;
265 	} else {
266 		mtu = ifp->if_mtu;
267 	}
268 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
269 		struct in_multi *inm;
270 
271 		m->m_flags |= M_MCAST;
272 		/*
273 		 * IP destination address is multicast.  Make sure "dst"
274 		 * still points to the address in "ro".  (It may have been
275 		 * changed to point to a gateway address, above.)
276 		 */
277 		dst = (struct sockaddr_in *)&ro->ro_dst;
278 		/*
279 		 * See if the caller provided any multicast options
280 		 */
281 		if (imo != NULL) {
282 			ip->ip_ttl = imo->imo_multicast_ttl;
283 			if (imo->imo_multicast_vif != -1)
284 				ip->ip_src.s_addr =
285 				    ip_mcast_src ?
286 				    ip_mcast_src(imo->imo_multicast_vif) :
287 				    INADDR_ANY;
288 		} else
289 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
290 		/*
291 		 * Confirm that the outgoing interface supports multicast.
292 		 */
293 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
294 			if ((ifp->if_flags & IFF_MULTICAST) == 0) {
295 				ipstat.ips_noroute++;
296 				error = ENETUNREACH;
297 				goto bad;
298 			}
299 		}
300 		/*
301 		 * If source address not specified yet, use address
302 		 * of outgoing interface.
303 		 */
304 		if (ip->ip_src.s_addr == INADDR_ANY) {
305 			/* Interface may have no addresses. */
306 			if (ia != NULL)
307 				ip->ip_src = IA_SIN(ia)->sin_addr;
308 		}
309 
310 		IN_MULTI_LOCK();
311 		IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm);
312 		if (inm != NULL &&
313 		   (imo == NULL || imo->imo_multicast_loop)) {
314 			IN_MULTI_UNLOCK();
315 			/*
316 			 * If we belong to the destination multicast group
317 			 * on the outgoing interface, and the caller did not
318 			 * forbid loopback, loop back a copy.
319 			 */
320 			ip_mloopback(ifp, m, dst, hlen);
321 		}
322 		else {
323 			IN_MULTI_UNLOCK();
324 			/*
325 			 * If we are acting as a multicast router, perform
326 			 * multicast forwarding as if the packet had just
327 			 * arrived on the interface to which we are about
328 			 * to send.  The multicast forwarding function
329 			 * recursively calls this function, using the
330 			 * IP_FORWARDING flag to prevent infinite recursion.
331 			 *
332 			 * Multicasts that are looped back by ip_mloopback(),
333 			 * above, will be forwarded by the ip_input() routine,
334 			 * if necessary.
335 			 */
336 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
337 				/*
338 				 * If rsvp daemon is not running, do not
339 				 * set ip_moptions. This ensures that the packet
340 				 * is multicast and not just sent down one link
341 				 * as prescribed by rsvpd.
342 				 */
343 				if (!rsvp_on)
344 					imo = NULL;
345 				if (ip_mforward &&
346 				    ip_mforward(ip, ifp, m, imo) != 0) {
347 					m_freem(m);
348 					goto done;
349 				}
350 			}
351 		}
352 
353 		/*
354 		 * Multicasts with a time-to-live of zero may be looped-
355 		 * back, above, but must not be transmitted on a network.
356 		 * Also, multicasts addressed to the loopback interface
357 		 * are not sent -- the above call to ip_mloopback() will
358 		 * loop back a copy if this host actually belongs to the
359 		 * destination group on the loopback interface.
360 		 */
361 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
362 			m_freem(m);
363 			goto done;
364 		}
365 
366 		goto sendit;
367 	}
368 
369 	/*
370 	 * If the source address is not specified yet, use the address
371 	 * of the outoing interface.
372 	 */
373 	if (ip->ip_src.s_addr == INADDR_ANY) {
374 		/* Interface may have no addresses. */
375 		if (ia != NULL) {
376 			ip->ip_src = IA_SIN(ia)->sin_addr;
377 		}
378 	}
379 
380 	/*
381 	 * Verify that we have any chance at all of being able to queue the
382 	 * packet or packet fragments, unless ALTQ is enabled on the given
383 	 * interface in which case packetdrop should be done by queueing.
384 	 */
385 #ifdef ALTQ
386 	if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) &&
387 	    ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >=
388 	    ifp->if_snd.ifq_maxlen))
389 #else
390 	if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >=
391 	    ifp->if_snd.ifq_maxlen)
392 #endif /* ALTQ */
393 	{
394 		error = ENOBUFS;
395 		ipstat.ips_odropped++;
396 		ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1);
397 		goto bad;
398 	}
399 
400 	/*
401 	 * Look for broadcast address and
402 	 * verify user is allowed to send
403 	 * such a packet.
404 	 */
405 	if (isbroadcast) {
406 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
407 			error = EADDRNOTAVAIL;
408 			goto bad;
409 		}
410 		if ((flags & IP_ALLOWBROADCAST) == 0) {
411 			error = EACCES;
412 			goto bad;
413 		}
414 		/* don't allow broadcast messages to be fragmented */
415 		if (ip->ip_len > mtu) {
416 			error = EMSGSIZE;
417 			goto bad;
418 		}
419 		m->m_flags |= M_BCAST;
420 	} else {
421 		m->m_flags &= ~M_BCAST;
422 	}
423 
424 sendit:
425 #if defined(IPSEC) || defined(FAST_IPSEC)
426 	switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) {
427 	case 1:
428 		goto bad;
429 	case -1:
430 		goto done;
431 	case 0:
432 	default:
433 		break;	/* Continue with packet processing. */
434 	}
435 	/* Update variables that are affected by ipsec4_output(). */
436 	ip = mtod(m, struct ip *);
437 	hlen = ip->ip_hl << 2;
438 #endif /* IPSEC */
439 
440 	/* Jump over all PFIL processing if hooks are not active. */
441 	if (!PFIL_HOOKED(&inet_pfil_hook))
442 		goto passout;
443 
444 	/* Run through list of hooks for output packets. */
445 	odst.s_addr = ip->ip_dst.s_addr;
446 	error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp);
447 	if (error != 0 || m == NULL)
448 		goto done;
449 
450 	ip = mtod(m, struct ip *);
451 
452 	/* See if destination IP address was changed by packet filter. */
453 	if (odst.s_addr != ip->ip_dst.s_addr) {
454 		m->m_flags |= M_SKIP_FIREWALL;
455 		/* If destination is now ourself drop to ip_input(). */
456 		if (in_localip(ip->ip_dst)) {
457 			m->m_flags |= M_FASTFWD_OURS;
458 			if (m->m_pkthdr.rcvif == NULL)
459 				m->m_pkthdr.rcvif = loif;
460 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
461 				m->m_pkthdr.csum_flags |=
462 				    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
463 				m->m_pkthdr.csum_data = 0xffff;
464 			}
465 			m->m_pkthdr.csum_flags |=
466 			    CSUM_IP_CHECKED | CSUM_IP_VALID;
467 
468 			error = netisr_queue(NETISR_IP, m);
469 			goto done;
470 		} else
471 			goto again;	/* Redo the routing table lookup. */
472 	}
473 
474 #ifdef IPFIREWALL_FORWARD
475 	/* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
476 	if (m->m_flags & M_FASTFWD_OURS) {
477 		if (m->m_pkthdr.rcvif == NULL)
478 			m->m_pkthdr.rcvif = loif;
479 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
480 			m->m_pkthdr.csum_flags |=
481 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
482 			m->m_pkthdr.csum_data = 0xffff;
483 		}
484 		m->m_pkthdr.csum_flags |=
485 			    CSUM_IP_CHECKED | CSUM_IP_VALID;
486 
487 		error = netisr_queue(NETISR_IP, m);
488 		goto done;
489 	}
490 	/* Or forward to some other address? */
491 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
492 	if (fwd_tag) {
493 		dst = (struct sockaddr_in *)&ro->ro_dst;
494 		bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
495 		m->m_flags |= M_SKIP_FIREWALL;
496 		m_tag_delete(m, fwd_tag);
497 		goto again;
498 	}
499 #endif /* IPFIREWALL_FORWARD */
500 
501 passout:
502 	/* 127/8 must not appear on wire - RFC1122. */
503 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
504 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
505 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
506 			ipstat.ips_badaddr++;
507 			error = EADDRNOTAVAIL;
508 			goto bad;
509 		}
510 	}
511 
512 	m->m_pkthdr.csum_flags |= CSUM_IP;
513 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
514 	if (sw_csum & CSUM_DELAY_DATA) {
515 		in_delayed_cksum(m);
516 		sw_csum &= ~CSUM_DELAY_DATA;
517 	}
518 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
519 
520 	/*
521 	 * If small enough for interface, or the interface will take
522 	 * care of the fragmentation for us, we can just send directly.
523 	 */
524 	if (ip->ip_len <= mtu ||
525 	    (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
526 	    ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
527 		ip->ip_len = htons(ip->ip_len);
528 		ip->ip_off = htons(ip->ip_off);
529 		ip->ip_sum = 0;
530 		if (sw_csum & CSUM_DELAY_IP)
531 			ip->ip_sum = in_cksum(m, hlen);
532 
533 		/*
534 		 * Record statistics for this interface address.
535 		 * With CSUM_TSO the byte/packet count will be slightly
536 		 * incorrect because we count the IP+TCP headers only
537 		 * once instead of for every generated packet.
538 		 */
539 		if (!(flags & IP_FORWARDING) && ia) {
540 			if (m->m_pkthdr.csum_flags & CSUM_TSO)
541 				ia->ia_ifa.if_opackets +=
542 				    m->m_pkthdr.len / m->m_pkthdr.tso_segsz;
543 			else
544 				ia->ia_ifa.if_opackets++;
545 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
546 		}
547 #ifdef IPSEC
548 		/* clean ipsec history once it goes out of the node */
549 		ipsec_delaux(m);
550 #endif
551 #ifdef MBUF_STRESS_TEST
552 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
553 			m = m_fragment(m, M_DONTWAIT, mbuf_frag_size);
554 #endif
555 		/*
556 		 * Reset layer specific mbuf flags
557 		 * to avoid confusing lower layers.
558 		 */
559 		m->m_flags &= ~(M_PROTOFLAGS);
560 
561 		error = (*ifp->if_output)(ifp, m,
562 				(struct sockaddr *)dst, ro->ro_rt);
563 		goto done;
564 	}
565 
566 	/* Balk when DF bit is set or the interface didn't support TSO. */
567 	if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) {
568 		error = EMSGSIZE;
569 		ipstat.ips_cantfrag++;
570 		goto bad;
571 	}
572 
573 	/*
574 	 * Too large for interface; fragment if possible. If successful,
575 	 * on return, m will point to a list of packets to be sent.
576 	 */
577 	error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum);
578 	if (error)
579 		goto bad;
580 	for (; m; m = m0) {
581 		m0 = m->m_nextpkt;
582 		m->m_nextpkt = 0;
583 #ifdef IPSEC
584 		/* clean ipsec history once it goes out of the node */
585 		ipsec_delaux(m);
586 #endif
587 		if (error == 0) {
588 			/* Record statistics for this interface address. */
589 			if (ia != NULL) {
590 				ia->ia_ifa.if_opackets++;
591 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
592 			}
593 			/*
594 			 * Reset layer specific mbuf flags
595 			 * to avoid confusing upper layers.
596 			 */
597 			m->m_flags &= ~(M_PROTOFLAGS);
598 
599 			error = (*ifp->if_output)(ifp, m,
600 			    (struct sockaddr *)dst, ro->ro_rt);
601 		} else
602 			m_freem(m);
603 	}
604 
605 	if (error == 0)
606 		ipstat.ips_fragmented++;
607 
608 done:
609 	if (ro == &iproute && ro->ro_rt) {
610 		RTFREE(ro->ro_rt);
611 	}
612 	return (error);
613 bad:
614 	m_freem(m);
615 	goto done;
616 }
617 
618 /*
619  * Create a chain of fragments which fit the given mtu. m_frag points to the
620  * mbuf to be fragmented; on return it points to the chain with the fragments.
621  * Return 0 if no error. If error, m_frag may contain a partially built
622  * chain of fragments that should be freed by the caller.
623  *
624  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
625  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
626  */
627 int
628 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
629 	    u_long if_hwassist_flags, int sw_csum)
630 {
631 	int error = 0;
632 	int hlen = ip->ip_hl << 2;
633 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
634 	int off;
635 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
636 	int firstlen;
637 	struct mbuf **mnext;
638 	int nfrags;
639 
640 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
641 		ipstat.ips_cantfrag++;
642 		return EMSGSIZE;
643 	}
644 
645 	/*
646 	 * Must be able to put at least 8 bytes per fragment.
647 	 */
648 	if (len < 8)
649 		return EMSGSIZE;
650 
651 	/*
652 	 * If the interface will not calculate checksums on
653 	 * fragmented packets, then do it here.
654 	 */
655 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
656 	    (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
657 		in_delayed_cksum(m0);
658 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
659 	}
660 
661 	if (len > PAGE_SIZE) {
662 		/*
663 		 * Fragment large datagrams such that each segment
664 		 * contains a multiple of PAGE_SIZE amount of data,
665 		 * plus headers. This enables a receiver to perform
666 		 * page-flipping zero-copy optimizations.
667 		 *
668 		 * XXX When does this help given that sender and receiver
669 		 * could have different page sizes, and also mtu could
670 		 * be less than the receiver's page size ?
671 		 */
672 		int newlen;
673 		struct mbuf *m;
674 
675 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
676 			off += m->m_len;
677 
678 		/*
679 		 * firstlen (off - hlen) must be aligned on an
680 		 * 8-byte boundary
681 		 */
682 		if (off < hlen)
683 			goto smart_frag_failure;
684 		off = ((off - hlen) & ~7) + hlen;
685 		newlen = (~PAGE_MASK) & mtu;
686 		if ((newlen + sizeof (struct ip)) > mtu) {
687 			/* we failed, go back the default */
688 smart_frag_failure:
689 			newlen = len;
690 			off = hlen + len;
691 		}
692 		len = newlen;
693 
694 	} else {
695 		off = hlen + len;
696 	}
697 
698 	firstlen = off - hlen;
699 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
700 
701 	/*
702 	 * Loop through length of segment after first fragment,
703 	 * make new header and copy data of each part and link onto chain.
704 	 * Here, m0 is the original packet, m is the fragment being created.
705 	 * The fragments are linked off the m_nextpkt of the original
706 	 * packet, which after processing serves as the first fragment.
707 	 */
708 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
709 		struct ip *mhip;	/* ip header on the fragment */
710 		struct mbuf *m;
711 		int mhlen = sizeof (struct ip);
712 
713 		MGETHDR(m, M_DONTWAIT, MT_DATA);
714 		if (m == NULL) {
715 			error = ENOBUFS;
716 			ipstat.ips_odropped++;
717 			goto done;
718 		}
719 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
720 		/*
721 		 * In the first mbuf, leave room for the link header, then
722 		 * copy the original IP header including options. The payload
723 		 * goes into an additional mbuf chain returned by m_copy().
724 		 */
725 		m->m_data += max_linkhdr;
726 		mhip = mtod(m, struct ip *);
727 		*mhip = *ip;
728 		if (hlen > sizeof (struct ip)) {
729 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
730 			mhip->ip_v = IPVERSION;
731 			mhip->ip_hl = mhlen >> 2;
732 		}
733 		m->m_len = mhlen;
734 		/* XXX do we need to add ip->ip_off below ? */
735 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
736 		if (off + len >= ip->ip_len) {	/* last fragment */
737 			len = ip->ip_len - off;
738 			m->m_flags |= M_LASTFRAG;
739 		} else
740 			mhip->ip_off |= IP_MF;
741 		mhip->ip_len = htons((u_short)(len + mhlen));
742 		m->m_next = m_copy(m0, off, len);
743 		if (m->m_next == NULL) {	/* copy failed */
744 			m_free(m);
745 			error = ENOBUFS;	/* ??? */
746 			ipstat.ips_odropped++;
747 			goto done;
748 		}
749 		m->m_pkthdr.len = mhlen + len;
750 		m->m_pkthdr.rcvif = NULL;
751 #ifdef MAC
752 		mac_create_fragment(m0, m);
753 #endif
754 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
755 		mhip->ip_off = htons(mhip->ip_off);
756 		mhip->ip_sum = 0;
757 		if (sw_csum & CSUM_DELAY_IP)
758 			mhip->ip_sum = in_cksum(m, mhlen);
759 		*mnext = m;
760 		mnext = &m->m_nextpkt;
761 	}
762 	ipstat.ips_ofragments += nfrags;
763 
764 	/* set first marker for fragment chain */
765 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
766 	m0->m_pkthdr.csum_data = nfrags;
767 
768 	/*
769 	 * Update first fragment by trimming what's been copied out
770 	 * and updating header.
771 	 */
772 	m_adj(m0, hlen + firstlen - ip->ip_len);
773 	m0->m_pkthdr.len = hlen + firstlen;
774 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
775 	ip->ip_off |= IP_MF;
776 	ip->ip_off = htons(ip->ip_off);
777 	ip->ip_sum = 0;
778 	if (sw_csum & CSUM_DELAY_IP)
779 		ip->ip_sum = in_cksum(m0, hlen);
780 
781 done:
782 	*m_frag = m0;
783 	return error;
784 }
785 
786 void
787 in_delayed_cksum(struct mbuf *m)
788 {
789 	struct ip *ip;
790 	u_short csum, offset;
791 
792 	ip = mtod(m, struct ip *);
793 	offset = ip->ip_hl << 2 ;
794 	csum = in_cksum_skip(m, ip->ip_len, offset);
795 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
796 		csum = 0xffff;
797 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
798 
799 	if (offset + sizeof(u_short) > m->m_len) {
800 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
801 		    m->m_len, offset, ip->ip_p);
802 		/*
803 		 * XXX
804 		 * this shouldn't happen, but if it does, the
805 		 * correct behavior may be to insert the checksum
806 		 * in the appropriate next mbuf in the chain.
807 		 */
808 		return;
809 	}
810 	*(u_short *)(m->m_data + offset) = csum;
811 }
812 
813 /*
814  * IP socket option processing.
815  */
816 int
817 ip_ctloutput(so, sopt)
818 	struct socket *so;
819 	struct sockopt *sopt;
820 {
821 	struct	inpcb *inp = sotoinpcb(so);
822 	int	error, optval;
823 
824 	error = optval = 0;
825 	if (sopt->sopt_level != IPPROTO_IP) {
826 		return (EINVAL);
827 	}
828 
829 	switch (sopt->sopt_dir) {
830 	case SOPT_SET:
831 		switch (sopt->sopt_name) {
832 		case IP_OPTIONS:
833 #ifdef notyet
834 		case IP_RETOPTS:
835 #endif
836 		{
837 			struct mbuf *m;
838 			if (sopt->sopt_valsize > MLEN) {
839 				error = EMSGSIZE;
840 				break;
841 			}
842 			MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
843 			if (m == NULL) {
844 				error = ENOBUFS;
845 				break;
846 			}
847 			m->m_len = sopt->sopt_valsize;
848 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
849 					    m->m_len);
850 			if (error) {
851 				m_free(m);
852 				break;
853 			}
854 			INP_LOCK(inp);
855 			error = ip_pcbopts(inp, sopt->sopt_name, m);
856 			INP_UNLOCK(inp);
857 			return (error);
858 		}
859 
860 		case IP_TOS:
861 		case IP_TTL:
862 		case IP_MINTTL:
863 		case IP_RECVOPTS:
864 		case IP_RECVRETOPTS:
865 		case IP_RECVDSTADDR:
866 		case IP_RECVTTL:
867 		case IP_RECVIF:
868 		case IP_FAITH:
869 		case IP_ONESBCAST:
870 		case IP_DONTFRAG:
871 			error = sooptcopyin(sopt, &optval, sizeof optval,
872 					    sizeof optval);
873 			if (error)
874 				break;
875 
876 			switch (sopt->sopt_name) {
877 			case IP_TOS:
878 				inp->inp_ip_tos = optval;
879 				break;
880 
881 			case IP_TTL:
882 				inp->inp_ip_ttl = optval;
883 				break;
884 
885 			case IP_MINTTL:
886 				if (optval > 0 && optval <= MAXTTL)
887 					inp->inp_ip_minttl = optval;
888 				else
889 					error = EINVAL;
890 				break;
891 
892 #define	OPTSET(bit) do {						\
893 	INP_LOCK(inp);							\
894 	if (optval)							\
895 		inp->inp_flags |= bit;					\
896 	else								\
897 		inp->inp_flags &= ~bit;					\
898 	INP_UNLOCK(inp);						\
899 } while (0)
900 
901 			case IP_RECVOPTS:
902 				OPTSET(INP_RECVOPTS);
903 				break;
904 
905 			case IP_RECVRETOPTS:
906 				OPTSET(INP_RECVRETOPTS);
907 				break;
908 
909 			case IP_RECVDSTADDR:
910 				OPTSET(INP_RECVDSTADDR);
911 				break;
912 
913 			case IP_RECVTTL:
914 				OPTSET(INP_RECVTTL);
915 				break;
916 
917 			case IP_RECVIF:
918 				OPTSET(INP_RECVIF);
919 				break;
920 
921 			case IP_FAITH:
922 				OPTSET(INP_FAITH);
923 				break;
924 
925 			case IP_ONESBCAST:
926 				OPTSET(INP_ONESBCAST);
927 				break;
928 			case IP_DONTFRAG:
929 				OPTSET(INP_DONTFRAG);
930 				break;
931 			}
932 			break;
933 #undef OPTSET
934 
935 		case IP_MULTICAST_IF:
936 		case IP_MULTICAST_VIF:
937 		case IP_MULTICAST_TTL:
938 		case IP_MULTICAST_LOOP:
939 		case IP_ADD_MEMBERSHIP:
940 		case IP_DROP_MEMBERSHIP:
941 			error = ip_setmoptions(inp, sopt);
942 			break;
943 
944 		case IP_PORTRANGE:
945 			error = sooptcopyin(sopt, &optval, sizeof optval,
946 					    sizeof optval);
947 			if (error)
948 				break;
949 
950 			INP_LOCK(inp);
951 			switch (optval) {
952 			case IP_PORTRANGE_DEFAULT:
953 				inp->inp_flags &= ~(INP_LOWPORT);
954 				inp->inp_flags &= ~(INP_HIGHPORT);
955 				break;
956 
957 			case IP_PORTRANGE_HIGH:
958 				inp->inp_flags &= ~(INP_LOWPORT);
959 				inp->inp_flags |= INP_HIGHPORT;
960 				break;
961 
962 			case IP_PORTRANGE_LOW:
963 				inp->inp_flags &= ~(INP_HIGHPORT);
964 				inp->inp_flags |= INP_LOWPORT;
965 				break;
966 
967 			default:
968 				error = EINVAL;
969 				break;
970 			}
971 			INP_UNLOCK(inp);
972 			break;
973 
974 #if defined(IPSEC) || defined(FAST_IPSEC)
975 		case IP_IPSEC_POLICY:
976 		{
977 			caddr_t req;
978 			size_t len = 0;
979 			int priv;
980 			struct mbuf *m;
981 			int optname;
982 
983 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
984 				break;
985 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
986 				break;
987 			if (sopt->sopt_td != NULL) {
988 				/*
989 				 * XXXRW: Would be more desirable to do this
990 				 * one layer down so that we only exercise
991 				 * privilege if it is needed.
992 				 */
993 				error = priv_check(sopt->sopt_td,
994 				    PRIV_NETINET_IPSEC);
995 				if (error)
996 					priv = 0;
997 				else
998 					priv = 1;
999 			} else
1000 				priv = 1;
1001 			req = mtod(m, caddr_t);
1002 			len = m->m_len;
1003 			optname = sopt->sopt_name;
1004 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1005 			m_freem(m);
1006 			break;
1007 		}
1008 #endif /*IPSEC*/
1009 
1010 		default:
1011 			error = ENOPROTOOPT;
1012 			break;
1013 		}
1014 		break;
1015 
1016 	case SOPT_GET:
1017 		switch (sopt->sopt_name) {
1018 		case IP_OPTIONS:
1019 		case IP_RETOPTS:
1020 			if (inp->inp_options)
1021 				error = sooptcopyout(sopt,
1022 						     mtod(inp->inp_options,
1023 							  char *),
1024 						     inp->inp_options->m_len);
1025 			else
1026 				sopt->sopt_valsize = 0;
1027 			break;
1028 
1029 		case IP_TOS:
1030 		case IP_TTL:
1031 		case IP_MINTTL:
1032 		case IP_RECVOPTS:
1033 		case IP_RECVRETOPTS:
1034 		case IP_RECVDSTADDR:
1035 		case IP_RECVTTL:
1036 		case IP_RECVIF:
1037 		case IP_PORTRANGE:
1038 		case IP_FAITH:
1039 		case IP_ONESBCAST:
1040 		case IP_DONTFRAG:
1041 			switch (sopt->sopt_name) {
1042 
1043 			case IP_TOS:
1044 				optval = inp->inp_ip_tos;
1045 				break;
1046 
1047 			case IP_TTL:
1048 				optval = inp->inp_ip_ttl;
1049 				break;
1050 
1051 			case IP_MINTTL:
1052 				optval = inp->inp_ip_minttl;
1053 				break;
1054 
1055 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1056 
1057 			case IP_RECVOPTS:
1058 				optval = OPTBIT(INP_RECVOPTS);
1059 				break;
1060 
1061 			case IP_RECVRETOPTS:
1062 				optval = OPTBIT(INP_RECVRETOPTS);
1063 				break;
1064 
1065 			case IP_RECVDSTADDR:
1066 				optval = OPTBIT(INP_RECVDSTADDR);
1067 				break;
1068 
1069 			case IP_RECVTTL:
1070 				optval = OPTBIT(INP_RECVTTL);
1071 				break;
1072 
1073 			case IP_RECVIF:
1074 				optval = OPTBIT(INP_RECVIF);
1075 				break;
1076 
1077 			case IP_PORTRANGE:
1078 				if (inp->inp_flags & INP_HIGHPORT)
1079 					optval = IP_PORTRANGE_HIGH;
1080 				else if (inp->inp_flags & INP_LOWPORT)
1081 					optval = IP_PORTRANGE_LOW;
1082 				else
1083 					optval = 0;
1084 				break;
1085 
1086 			case IP_FAITH:
1087 				optval = OPTBIT(INP_FAITH);
1088 				break;
1089 
1090 			case IP_ONESBCAST:
1091 				optval = OPTBIT(INP_ONESBCAST);
1092 				break;
1093 			case IP_DONTFRAG:
1094 				optval = OPTBIT(INP_DONTFRAG);
1095 				break;
1096 			}
1097 			error = sooptcopyout(sopt, &optval, sizeof optval);
1098 			break;
1099 
1100 		case IP_MULTICAST_IF:
1101 		case IP_MULTICAST_VIF:
1102 		case IP_MULTICAST_TTL:
1103 		case IP_MULTICAST_LOOP:
1104 			error = ip_getmoptions(inp, sopt);
1105 			break;
1106 
1107 #if defined(IPSEC) || defined(FAST_IPSEC)
1108 		case IP_IPSEC_POLICY:
1109 		{
1110 			struct mbuf *m = NULL;
1111 			caddr_t req = NULL;
1112 			size_t len = 0;
1113 
1114 			if (m != 0) {
1115 				req = mtod(m, caddr_t);
1116 				len = m->m_len;
1117 			}
1118 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1119 			if (error == 0)
1120 				error = soopt_mcopyout(sopt, m); /* XXX */
1121 			if (error == 0)
1122 				m_freem(m);
1123 			break;
1124 		}
1125 #endif /*IPSEC*/
1126 
1127 		default:
1128 			error = ENOPROTOOPT;
1129 			break;
1130 		}
1131 		break;
1132 	}
1133 	return (error);
1134 }
1135 
1136 /*
1137  * XXX
1138  * The whole multicast option thing needs to be re-thought.
1139  * Several of these options are equally applicable to non-multicast
1140  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1141  * standard option (IP_TTL).
1142  */
1143 
1144 /*
1145  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1146  */
1147 static struct ifnet *
1148 ip_multicast_if(a, ifindexp)
1149 	struct in_addr *a;
1150 	int *ifindexp;
1151 {
1152 	int ifindex;
1153 	struct ifnet *ifp;
1154 
1155 	if (ifindexp)
1156 		*ifindexp = 0;
1157 	if (ntohl(a->s_addr) >> 24 == 0) {
1158 		ifindex = ntohl(a->s_addr) & 0xffffff;
1159 		if (ifindex < 0 || if_index < ifindex)
1160 			return NULL;
1161 		ifp = ifnet_byindex(ifindex);
1162 		if (ifindexp)
1163 			*ifindexp = ifindex;
1164 	} else {
1165 		INADDR_TO_IFP(*a, ifp);
1166 	}
1167 	return ifp;
1168 }
1169 
1170 /*
1171  * Given an inpcb, return its multicast options structure pointer.  Accepts
1172  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1173  */
1174 static struct ip_moptions *
1175 ip_findmoptions(struct inpcb *inp)
1176 {
1177 	struct ip_moptions *imo;
1178 	struct in_multi **immp;
1179 
1180 	INP_LOCK(inp);
1181 	if (inp->inp_moptions != NULL)
1182 		return (inp->inp_moptions);
1183 
1184 	INP_UNLOCK(inp);
1185 
1186 	imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1187 	immp = (struct in_multi **)malloc((sizeof(*immp) * IP_MIN_MEMBERSHIPS),
1188 					  M_IPMOPTS, M_WAITOK);
1189 
1190 	imo->imo_multicast_ifp = NULL;
1191 	imo->imo_multicast_addr.s_addr = INADDR_ANY;
1192 	imo->imo_multicast_vif = -1;
1193 	imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1194 	imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1195 	imo->imo_num_memberships = 0;
1196 	imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1197 	imo->imo_membership = immp;
1198 
1199 	INP_LOCK(inp);
1200 	if (inp->inp_moptions != NULL) {
1201 		free(immp, M_IPMOPTS);
1202 		free(imo, M_IPMOPTS);
1203 		return (inp->inp_moptions);
1204 	}
1205 	inp->inp_moptions = imo;
1206 	return (imo);
1207 }
1208 
1209 /*
1210  * Set the IP multicast options in response to user setsockopt().
1211  */
1212 static int
1213 ip_setmoptions(struct inpcb *inp, struct sockopt *sopt)
1214 {
1215 	int error = 0;
1216 	int i;
1217 	struct in_addr addr;
1218 	struct ip_mreq mreq;
1219 	struct ifnet *ifp;
1220 	struct ip_moptions *imo;
1221 	struct route ro;
1222 	struct sockaddr_in *dst;
1223 	int ifindex;
1224 	int s;
1225 
1226 	switch (sopt->sopt_name) {
1227 	/* store an index number for the vif you wanna use in the send */
1228 	case IP_MULTICAST_VIF:
1229 		if (legal_vif_num == 0) {
1230 			error = EOPNOTSUPP;
1231 			break;
1232 		}
1233 		error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1234 		if (error)
1235 			break;
1236 		if (!legal_vif_num(i) && (i != -1)) {
1237 			error = EINVAL;
1238 			break;
1239 		}
1240 		imo = ip_findmoptions(inp);
1241 		imo->imo_multicast_vif = i;
1242 		INP_UNLOCK(inp);
1243 		break;
1244 
1245 	case IP_MULTICAST_IF:
1246 		/*
1247 		 * Select the interface for outgoing multicast packets.
1248 		 */
1249 		error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1250 		if (error)
1251 			break;
1252 		/*
1253 		 * INADDR_ANY is used to remove a previous selection.
1254 		 * When no interface is selected, a default one is
1255 		 * chosen every time a multicast packet is sent.
1256 		 */
1257 		imo = ip_findmoptions(inp);
1258 		if (addr.s_addr == INADDR_ANY) {
1259 			imo->imo_multicast_ifp = NULL;
1260 			INP_UNLOCK(inp);
1261 			break;
1262 		}
1263 		/*
1264 		 * The selected interface is identified by its local
1265 		 * IP address.  Find the interface and confirm that
1266 		 * it supports multicasting.
1267 		 */
1268 		s = splimp();
1269 		ifp = ip_multicast_if(&addr, &ifindex);
1270 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1271 			INP_UNLOCK(inp);
1272 			splx(s);
1273 			error = EADDRNOTAVAIL;
1274 			break;
1275 		}
1276 		imo->imo_multicast_ifp = ifp;
1277 		if (ifindex)
1278 			imo->imo_multicast_addr = addr;
1279 		else
1280 			imo->imo_multicast_addr.s_addr = INADDR_ANY;
1281 		INP_UNLOCK(inp);
1282 		splx(s);
1283 		break;
1284 
1285 	case IP_MULTICAST_TTL:
1286 		/*
1287 		 * Set the IP time-to-live for outgoing multicast packets.
1288 		 * The original multicast API required a char argument,
1289 		 * which is inconsistent with the rest of the socket API.
1290 		 * We allow either a char or an int.
1291 		 */
1292 		if (sopt->sopt_valsize == 1) {
1293 			u_char ttl;
1294 			error = sooptcopyin(sopt, &ttl, 1, 1);
1295 			if (error)
1296 				break;
1297 			imo = ip_findmoptions(inp);
1298 			imo->imo_multicast_ttl = ttl;
1299 			INP_UNLOCK(inp);
1300 		} else {
1301 			u_int ttl;
1302 			error = sooptcopyin(sopt, &ttl, sizeof ttl,
1303 					    sizeof ttl);
1304 			if (error)
1305 				break;
1306 			if (ttl > 255)
1307 				error = EINVAL;
1308 			else {
1309 				imo = ip_findmoptions(inp);
1310 				imo->imo_multicast_ttl = ttl;
1311 				INP_UNLOCK(inp);
1312 			}
1313 		}
1314 		break;
1315 
1316 	case IP_MULTICAST_LOOP:
1317 		/*
1318 		 * Set the loopback flag for outgoing multicast packets.
1319 		 * Must be zero or one.  The original multicast API required a
1320 		 * char argument, which is inconsistent with the rest
1321 		 * of the socket API.  We allow either a char or an int.
1322 		 */
1323 		if (sopt->sopt_valsize == 1) {
1324 			u_char loop;
1325 			error = sooptcopyin(sopt, &loop, 1, 1);
1326 			if (error)
1327 				break;
1328 			imo = ip_findmoptions(inp);
1329 			imo->imo_multicast_loop = !!loop;
1330 			INP_UNLOCK(inp);
1331 		} else {
1332 			u_int loop;
1333 			error = sooptcopyin(sopt, &loop, sizeof loop,
1334 					    sizeof loop);
1335 			if (error)
1336 				break;
1337 			imo = ip_findmoptions(inp);
1338 			imo->imo_multicast_loop = !!loop;
1339 			INP_UNLOCK(inp);
1340 		}
1341 		break;
1342 
1343 	case IP_ADD_MEMBERSHIP:
1344 		/*
1345 		 * Add a multicast group membership.
1346 		 * Group must be a valid IP multicast address.
1347 		 */
1348 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1349 		if (error)
1350 			break;
1351 
1352 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1353 			error = EINVAL;
1354 			break;
1355 		}
1356 		s = splimp();
1357 		/*
1358 		 * If no interface address was provided, use the interface of
1359 		 * the route to the given multicast address.
1360 		 */
1361 		if (mreq.imr_interface.s_addr == INADDR_ANY) {
1362 			bzero((caddr_t)&ro, sizeof(ro));
1363 			dst = (struct sockaddr_in *)&ro.ro_dst;
1364 			dst->sin_len = sizeof(*dst);
1365 			dst->sin_family = AF_INET;
1366 			dst->sin_addr = mreq.imr_multiaddr;
1367 			rtalloc_ign(&ro, RTF_CLONING);
1368 			if (ro.ro_rt == NULL) {
1369 				error = EADDRNOTAVAIL;
1370 				splx(s);
1371 				break;
1372 			}
1373 			ifp = ro.ro_rt->rt_ifp;
1374 			RTFREE(ro.ro_rt);
1375 		}
1376 		else {
1377 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1378 		}
1379 
1380 		/*
1381 		 * See if we found an interface, and confirm that it
1382 		 * supports multicast.
1383 		 */
1384 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1385 			error = EADDRNOTAVAIL;
1386 			splx(s);
1387 			break;
1388 		}
1389 		/*
1390 		 * See if the membership already exists or if all the
1391 		 * membership slots are full.
1392 		 */
1393 		imo = ip_findmoptions(inp);
1394 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1395 			if (imo->imo_membership[i]->inm_ifp == ifp &&
1396 			    imo->imo_membership[i]->inm_addr.s_addr
1397 						== mreq.imr_multiaddr.s_addr)
1398 				break;
1399 		}
1400 		if (i < imo->imo_num_memberships) {
1401 			INP_UNLOCK(inp);
1402 			error = EADDRINUSE;
1403 			splx(s);
1404 			break;
1405 		}
1406 		if (imo->imo_num_memberships == imo->imo_max_memberships) {
1407 		    struct in_multi **nmships, **omships;
1408 		    size_t newmax;
1409 		    /*
1410 		     * Resize the vector to next power-of-two minus 1. If the
1411 		     * size would exceed the maximum then we know we've really
1412 		     * run out of entries. Otherwise, we realloc() the vector
1413 		     * with the INP lock held to avoid introducing a race.
1414 		     */
1415 		    nmships = NULL;
1416 		    omships = imo->imo_membership;
1417 		    newmax = ((imo->imo_max_memberships + 1) * 2) - 1;
1418 		    if (newmax <= IP_MAX_MEMBERSHIPS) {
1419 			nmships = (struct in_multi **)realloc(omships,
1420 sizeof(*nmships) * newmax, M_IPMOPTS, M_NOWAIT);
1421 			if (nmships != NULL) {
1422 			    imo->imo_membership = nmships;
1423 			    imo->imo_max_memberships = newmax;
1424 			}
1425 		    }
1426 		    if (nmships == NULL) {
1427 			INP_UNLOCK(inp);
1428 			error = ETOOMANYREFS;
1429 			splx(s);
1430 			break;
1431 		    }
1432 		}
1433 		/*
1434 		 * Everything looks good; add a new record to the multicast
1435 		 * address list for the given interface.
1436 		 */
1437 		if ((imo->imo_membership[i] =
1438 		    in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1439 			INP_UNLOCK(inp);
1440 			error = ENOBUFS;
1441 			splx(s);
1442 			break;
1443 		}
1444 		++imo->imo_num_memberships;
1445 		INP_UNLOCK(inp);
1446 		splx(s);
1447 		break;
1448 
1449 	case IP_DROP_MEMBERSHIP:
1450 		/*
1451 		 * Drop a multicast group membership.
1452 		 * Group must be a valid IP multicast address.
1453 		 */
1454 		error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1455 		if (error)
1456 			break;
1457 
1458 		if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1459 			error = EINVAL;
1460 			break;
1461 		}
1462 
1463 		s = splimp();
1464 		/*
1465 		 * If an interface address was specified, get a pointer
1466 		 * to its ifnet structure.
1467 		 */
1468 		if (mreq.imr_interface.s_addr == INADDR_ANY)
1469 			ifp = NULL;
1470 		else {
1471 			ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1472 			if (ifp == NULL) {
1473 				error = EADDRNOTAVAIL;
1474 				splx(s);
1475 				break;
1476 			}
1477 		}
1478 		/*
1479 		 * Find the membership in the membership array.
1480 		 */
1481 		imo = ip_findmoptions(inp);
1482 		for (i = 0; i < imo->imo_num_memberships; ++i) {
1483 			if ((ifp == NULL ||
1484 			     imo->imo_membership[i]->inm_ifp == ifp) &&
1485 			     imo->imo_membership[i]->inm_addr.s_addr ==
1486 			     mreq.imr_multiaddr.s_addr)
1487 				break;
1488 		}
1489 		if (i == imo->imo_num_memberships) {
1490 			INP_UNLOCK(inp);
1491 			error = EADDRNOTAVAIL;
1492 			splx(s);
1493 			break;
1494 		}
1495 		/*
1496 		 * Give up the multicast address record to which the
1497 		 * membership points.
1498 		 */
1499 		in_delmulti(imo->imo_membership[i]);
1500 		/*
1501 		 * Remove the gap in the membership array.
1502 		 */
1503 		for (++i; i < imo->imo_num_memberships; ++i)
1504 			imo->imo_membership[i-1] = imo->imo_membership[i];
1505 		--imo->imo_num_memberships;
1506 		INP_UNLOCK(inp);
1507 		splx(s);
1508 		break;
1509 
1510 	default:
1511 		error = EOPNOTSUPP;
1512 		break;
1513 	}
1514 
1515 	return (error);
1516 }
1517 
1518 /*
1519  * Return the IP multicast options in response to user getsockopt().
1520  */
1521 static int
1522 ip_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1523 {
1524 	struct ip_moptions *imo;
1525 	struct in_addr addr;
1526 	struct in_ifaddr *ia;
1527 	int error, optval;
1528 	u_char coptval;
1529 
1530 	INP_LOCK(inp);
1531 	imo = inp->inp_moptions;
1532 
1533 	error = 0;
1534 	switch (sopt->sopt_name) {
1535 	case IP_MULTICAST_VIF:
1536 		if (imo != NULL)
1537 			optval = imo->imo_multicast_vif;
1538 		else
1539 			optval = -1;
1540 		INP_UNLOCK(inp);
1541 		error = sooptcopyout(sopt, &optval, sizeof optval);
1542 		break;
1543 
1544 	case IP_MULTICAST_IF:
1545 		if (imo == NULL || imo->imo_multicast_ifp == NULL)
1546 			addr.s_addr = INADDR_ANY;
1547 		else if (imo->imo_multicast_addr.s_addr) {
1548 			/* return the value user has set */
1549 			addr = imo->imo_multicast_addr;
1550 		} else {
1551 			IFP_TO_IA(imo->imo_multicast_ifp, ia);
1552 			addr.s_addr = (ia == NULL) ? INADDR_ANY
1553 				: IA_SIN(ia)->sin_addr.s_addr;
1554 		}
1555 		INP_UNLOCK(inp);
1556 		error = sooptcopyout(sopt, &addr, sizeof addr);
1557 		break;
1558 
1559 	case IP_MULTICAST_TTL:
1560 		if (imo == 0)
1561 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1562 		else
1563 			optval = coptval = imo->imo_multicast_ttl;
1564 		INP_UNLOCK(inp);
1565 		if (sopt->sopt_valsize == 1)
1566 			error = sooptcopyout(sopt, &coptval, 1);
1567 		else
1568 			error = sooptcopyout(sopt, &optval, sizeof optval);
1569 		break;
1570 
1571 	case IP_MULTICAST_LOOP:
1572 		if (imo == 0)
1573 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1574 		else
1575 			optval = coptval = imo->imo_multicast_loop;
1576 		INP_UNLOCK(inp);
1577 		if (sopt->sopt_valsize == 1)
1578 			error = sooptcopyout(sopt, &coptval, 1);
1579 		else
1580 			error = sooptcopyout(sopt, &optval, sizeof optval);
1581 		break;
1582 
1583 	default:
1584 		INP_UNLOCK(inp);
1585 		error = ENOPROTOOPT;
1586 		break;
1587 	}
1588 	INP_UNLOCK_ASSERT(inp);
1589 
1590 	return (error);
1591 }
1592 
1593 /*
1594  * Discard the IP multicast options.
1595  */
1596 void
1597 ip_freemoptions(imo)
1598 	register struct ip_moptions *imo;
1599 {
1600 	register int i;
1601 
1602 	if (imo != NULL) {
1603 		for (i = 0; i < imo->imo_num_memberships; ++i)
1604 			in_delmulti(imo->imo_membership[i]);
1605 		free(imo->imo_membership, M_IPMOPTS);
1606 		free(imo, M_IPMOPTS);
1607 	}
1608 }
1609 
1610 /*
1611  * Routine called from ip_output() to loop back a copy of an IP multicast
1612  * packet to the input queue of a specified interface.  Note that this
1613  * calls the output routine of the loopback "driver", but with an interface
1614  * pointer that might NOT be a loopback interface -- evil, but easier than
1615  * replicating that code here.
1616  */
1617 static void
1618 ip_mloopback(ifp, m, dst, hlen)
1619 	struct ifnet *ifp;
1620 	register struct mbuf *m;
1621 	register struct sockaddr_in *dst;
1622 	int hlen;
1623 {
1624 	register struct ip *ip;
1625 	struct mbuf *copym;
1626 
1627 	copym = m_copy(m, 0, M_COPYALL);
1628 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
1629 		copym = m_pullup(copym, hlen);
1630 	if (copym != NULL) {
1631 		/* If needed, compute the checksum and mark it as valid. */
1632 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1633 			in_delayed_cksum(copym);
1634 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1635 			copym->m_pkthdr.csum_flags |=
1636 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1637 			copym->m_pkthdr.csum_data = 0xffff;
1638 		}
1639 		/*
1640 		 * We don't bother to fragment if the IP length is greater
1641 		 * than the interface's MTU.  Can this possibly matter?
1642 		 */
1643 		ip = mtod(copym, struct ip *);
1644 		ip->ip_len = htons(ip->ip_len);
1645 		ip->ip_off = htons(ip->ip_off);
1646 		ip->ip_sum = 0;
1647 		ip->ip_sum = in_cksum(copym, hlen);
1648 		/*
1649 		 * NB:
1650 		 * It's not clear whether there are any lingering
1651 		 * reentrancy problems in other areas which might
1652 		 * be exposed by using ip_input directly (in
1653 		 * particular, everything which modifies the packet
1654 		 * in-place).  Yet another option is using the
1655 		 * protosw directly to deliver the looped back
1656 		 * packet.  For the moment, we'll err on the side
1657 		 * of safety by using if_simloop().
1658 		 */
1659 #if 1 /* XXX */
1660 		if (dst->sin_family != AF_INET) {
1661 			printf("ip_mloopback: bad address family %d\n",
1662 						dst->sin_family);
1663 			dst->sin_family = AF_INET;
1664 		}
1665 #endif
1666 
1667 #ifdef notdef
1668 		copym->m_pkthdr.rcvif = ifp;
1669 		ip_input(copym);
1670 #else
1671 		if_simloop(ifp, copym, dst->sin_family, 0);
1672 #endif
1673 	}
1674 }
1675