The header type of a mbuf doesn't change when appended onto a chain.
[dragonfly.git] / sys / netinet / ip_output.c
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.30 2005/06/09 02:03:38 hsu Exp $
32  */
33
34 #define _IP_VHL
35
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipfilter.h"
40 #include "opt_ipsec.h"
41 #include "opt_random_ip_id.h"
42 #include "opt_mbuf_stress_test.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/thread2.h>
55 #include <sys/in_cksum.h>
56
57 #include <net/if.h>
58 #include <net/netisr.h>
59 #include <net/pfil.h>
60 #include <net/route.h>
61
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip_var.h>
68
69 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
70
71 #ifdef IPSEC
72 #include <netinet6/ipsec.h>
73 #include <netproto/key/key.h>
74 #ifdef IPSEC_DEBUG
75 #include <netproto/key/key_debug.h>
76 #else
77 #define KEYDEBUG(lev,arg)
78 #endif
79 #endif /*IPSEC*/
80
81 #ifdef FAST_IPSEC
82 #include <netproto/ipsec/ipsec.h>
83 #include <netproto/ipsec/xform.h>
84 #include <netproto/ipsec/key.h>
85 #endif /*FAST_IPSEC*/
86
87 #include <net/ipfw/ip_fw.h>
88 #include <net/dummynet/ip_dummynet.h>
89
90 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
91                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
92                                   (ntohl(a.s_addr)>>16)&0xFF,\
93                                   (ntohl(a.s_addr)>>8)&0xFF,\
94                                   (ntohl(a.s_addr))&0xFF, y);
95
96 u_short ip_id;
97
98 #ifdef MBUF_STRESS_TEST
99 int mbuf_frag_size = 0;
100 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
101         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
102 #endif
103
104 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
105 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
106 static void     ip_mloopback
107         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
108 static int      ip_getmoptions
109         (struct sockopt *, struct ip_moptions *);
110 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
111 static int      ip_setmoptions
112         (struct sockopt *, struct ip_moptions **);
113
114 int     ip_optcopy(struct ip *, struct ip *);
115
116
117 extern  struct protosw inetsw[];
118
119 /*
120  * IP output.  The packet in mbuf chain m contains a skeletal IP
121  * header (with len, off, ttl, proto, tos, src, dst).
122  * The mbuf chain containing the packet will be freed.
123  * The mbuf opt, if present, will not be freed.
124  */
125 int
126 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
127           int flags, struct ip_moptions *imo, struct inpcb *inp)
128 {
129         struct ip *ip;
130         struct ifnet *ifp = NULL;       /* keep compiler happy */
131         struct mbuf *m;
132         int hlen = sizeof(struct ip);
133         int len, off, error = 0;
134         struct sockaddr_in *dst = NULL; /* keep compiler happy */
135         struct in_ifaddr *ia = NULL;
136         int isbroadcast, sw_csum;
137         struct in_addr pkt_dst;
138         struct route iproute;
139 #ifdef IPSEC
140         struct secpolicy *sp = NULL;
141         struct socket *so = inp ? inp->inp_socket : NULL;
142 #endif
143 #ifdef FAST_IPSEC
144         struct m_tag *mtag;
145         struct secpolicy *sp = NULL;
146         struct tdb_ident *tdbi;
147 #endif /* FAST_IPSEC */
148         struct ip_fw_args args;
149         int src_was_INADDR_ANY = 0;     /* as the name says... */
150
151         args.eh = NULL;
152         args.rule = NULL;
153         args.next_hop = NULL;
154
155         /* Grab info from MT_TAG mbufs prepended to the chain. */
156         while (m0 != NULL && m0->m_type == MT_TAG) {
157                 switch(m0->_m_tag_id) {
158                 case PACKET_TAG_DUMMYNET:
159                         /*
160                          * the packet was already tagged, so part of the
161                          * processing was already done, and we need to go down.
162                          * Get parameters from the header.
163                          */
164                         args.rule = ((struct dn_pkt *)m0)->rule;
165                         opt = NULL ;
166                         ro = &((struct dn_pkt *)m0)->ro;
167                         imo = NULL ;
168                         dst = ((struct dn_pkt *)m0)->dn_dst ;
169                         ifp = ((struct dn_pkt *)m0)->ifp ;
170                         flags = ((struct dn_pkt *)m0)->flags ;
171                         break;
172                 case PACKET_TAG_IPFORWARD:
173                         args.next_hop = (struct sockaddr_in *)m0->m_data;
174                         break;
175                 default:
176                         printf("ip_output: unrecognised MT_TAG tag %d\n",
177                             m0->_m_tag_id);
178                         break;
179                 }
180                 m0 = m0->m_next;
181         }
182         m = m0;
183         KASSERT(m != NULL && (m->m_flags & M_PKTHDR), ("ip_output: no HDR"));
184
185         if (ro == NULL) {
186                 ro = &iproute;
187                 bzero(ro, sizeof *ro);
188         }
189
190         if (args.rule != NULL) {        /* dummynet already saw us */
191                 ip = mtod(m, struct ip *);
192                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
193                 if (ro->ro_rt)
194                         ia = ifatoia(ro->ro_rt->rt_ifa);
195                 goto sendit;
196         }
197
198         if (opt) {
199                 len = 0;
200                 m = ip_insertoptions(m, opt, &len);
201                 if (len != 0)
202                         hlen = len;
203         }
204         ip = mtod(m, struct ip *);
205         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
206
207         /*
208          * Fill in IP header.
209          */
210         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
211                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
212                 ip->ip_off &= IP_DF;
213 #ifdef RANDOM_IP_ID
214                 ip->ip_id = ip_randomid();
215 #else
216                 ip->ip_id = htons(ip_id++);
217 #endif
218                 ipstat.ips_localout++;
219         } else {
220                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
221         }
222
223         dst = (struct sockaddr_in *)&ro->ro_dst;
224         /*
225          * If there is a cached route,
226          * check that it is to the same destination
227          * and is still up.  If not, free it and try again.
228          * The address family should also be checked in case of sharing the
229          * cache with IPv6.
230          */
231         if (ro->ro_rt &&
232             (!(ro->ro_rt->rt_flags & RTF_UP) ||
233              dst->sin_family != AF_INET ||
234              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
235                 rtfree(ro->ro_rt);
236                 ro->ro_rt = (struct rtentry *)NULL;
237         }
238         if (ro->ro_rt == NULL) {
239                 bzero(dst, sizeof *dst);
240                 dst->sin_family = AF_INET;
241                 dst->sin_len = sizeof *dst;
242                 dst->sin_addr = pkt_dst;
243         }
244         /*
245          * If routing to interface only,
246          * short circuit routing lookup.
247          */
248         if (flags & IP_ROUTETOIF) {
249                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
250                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
251                         ipstat.ips_noroute++;
252                         error = ENETUNREACH;
253                         goto bad;
254                 }
255                 ifp = ia->ia_ifp;
256                 ip->ip_ttl = 1;
257                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
258         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
259                    imo != NULL && imo->imo_multicast_ifp != NULL) {
260                 /*
261                  * Bypass the normal routing lookup for multicast
262                  * packets if the interface is specified.
263                  */
264                 ifp = imo->imo_multicast_ifp;
265                 IFP_TO_IA(ifp, ia);
266                 isbroadcast = 0;        /* fool gcc */
267         } else {
268                 /*
269                  * If this is the case, we probably don't want to allocate
270                  * a protocol-cloned route since we didn't get one from the
271                  * ULP.  This lets TCP do its thing, while not burdening
272                  * forwarding or ICMP with the overhead of cloning a route.
273                  * Of course, we still want to do any cloning requested by
274                  * the link layer, as this is probably required in all cases
275                  * for correct operation (as it is for ARP).
276                  */
277                 if (ro->ro_rt == NULL)
278                         rtalloc_ign(ro, RTF_PRCLONING);
279                 if (ro->ro_rt == NULL) {
280                         ipstat.ips_noroute++;
281                         error = EHOSTUNREACH;
282                         goto bad;
283                 }
284                 ia = ifatoia(ro->ro_rt->rt_ifa);
285                 ifp = ro->ro_rt->rt_ifp;
286                 ro->ro_rt->rt_use++;
287                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
288                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
289                 if (ro->ro_rt->rt_flags & RTF_HOST)
290                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
291                 else
292                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
293         }
294         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
295                 struct in_multi *inm;
296
297                 m->m_flags |= M_MCAST;
298                 /*
299                  * IP destination address is multicast.  Make sure "dst"
300                  * still points to the address in "ro".  (It may have been
301                  * changed to point to a gateway address, above.)
302                  */
303                 dst = (struct sockaddr_in *)&ro->ro_dst;
304                 /*
305                  * See if the caller provided any multicast options
306                  */
307                 if (imo != NULL) {
308                         ip->ip_ttl = imo->imo_multicast_ttl;
309                         if (imo->imo_multicast_vif != -1)
310                                 ip->ip_src.s_addr =
311                                     ip_mcast_src ?
312                                     ip_mcast_src(imo->imo_multicast_vif) :
313                                     INADDR_ANY;
314                 } else
315                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
316                 /*
317                  * Confirm that the outgoing interface supports multicast.
318                  */
319                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
320                         if (!(ifp->if_flags & IFF_MULTICAST)) {
321                                 ipstat.ips_noroute++;
322                                 error = ENETUNREACH;
323                                 goto bad;
324                         }
325                 }
326                 /*
327                  * If source address not specified yet, use address
328                  * of outgoing interface.
329                  */
330                 if (ip->ip_src.s_addr == INADDR_ANY) {
331                         /* Interface may have no addresses. */
332                         if (ia != NULL)
333                                 ip->ip_src = IA_SIN(ia)->sin_addr;
334                 }
335
336                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
337                 if (inm != NULL &&
338                    (imo == NULL || imo->imo_multicast_loop)) {
339                         /*
340                          * If we belong to the destination multicast group
341                          * on the outgoing interface, and the caller did not
342                          * forbid loopback, loop back a copy.
343                          */
344                         ip_mloopback(ifp, m, dst, hlen);
345                 }
346                 else {
347                         /*
348                          * If we are acting as a multicast router, perform
349                          * multicast forwarding as if the packet had just
350                          * arrived on the interface to which we are about
351                          * to send.  The multicast forwarding function
352                          * recursively calls this function, using the
353                          * IP_FORWARDING flag to prevent infinite recursion.
354                          *
355                          * Multicasts that are looped back by ip_mloopback(),
356                          * above, will be forwarded by the ip_input() routine,
357                          * if necessary.
358                          */
359                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
360                                 /*
361                                  * If rsvp daemon is not running, do not
362                                  * set ip_moptions. This ensures that the packet
363                                  * is multicast and not just sent down one link
364                                  * as prescribed by rsvpd.
365                                  */
366                                 if (!rsvp_on)
367                                         imo = NULL;
368                                 if (ip_mforward &&
369                                     ip_mforward(ip, ifp, m, imo) != 0) {
370                                         m_freem(m);
371                                         goto done;
372                                 }
373                         }
374                 }
375
376                 /*
377                  * Multicasts with a time-to-live of zero may be looped-
378                  * back, above, but must not be transmitted on a network.
379                  * Also, multicasts addressed to the loopback interface
380                  * are not sent -- the above call to ip_mloopback() will
381                  * loop back a copy if this host actually belongs to the
382                  * destination group on the loopback interface.
383                  */
384                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
385                         m_freem(m);
386                         goto done;
387                 }
388
389                 goto sendit;
390         }
391 #ifndef notdef
392         /*
393          * If the source address is not specified yet, use the address
394          * of the outoing interface. In case, keep note we did that, so
395          * if the the firewall changes the next-hop causing the output
396          * interface to change, we can fix that.
397          */
398         if (ip->ip_src.s_addr == INADDR_ANY) {
399                 /* Interface may have no addresses. */
400                 if (ia != NULL) {
401                         ip->ip_src = IA_SIN(ia)->sin_addr;
402                         src_was_INADDR_ANY = 1;
403                 }
404         }
405 #endif /* notdef */
406 #ifdef ALTQ
407         /*
408          * Disable packet drop hack.
409          * Packetdrop should be done by queueing.
410          */
411 #else /* !ALTQ */
412         /*
413          * Verify that we have any chance at all of being able to queue
414          *      the packet or packet fragments
415          */
416         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
417                 ifp->if_snd.ifq_maxlen) {
418                         error = ENOBUFS;
419                         ipstat.ips_odropped++;
420                         goto bad;
421         }
422 #endif /* !ALTQ */
423
424         /*
425          * Look for broadcast address and
426          * verify user is allowed to send
427          * such a packet.
428          */
429         if (isbroadcast) {
430                 if (!(ifp->if_flags & IFF_BROADCAST)) {
431                         error = EADDRNOTAVAIL;
432                         goto bad;
433                 }
434                 if (!(flags & IP_ALLOWBROADCAST)) {
435                         error = EACCES;
436                         goto bad;
437                 }
438                 /* don't allow broadcast messages to be fragmented */
439                 if (ip->ip_len > ifp->if_mtu) {
440                         error = EMSGSIZE;
441                         goto bad;
442                 }
443                 m->m_flags |= M_BCAST;
444         } else {
445                 m->m_flags &= ~M_BCAST;
446         }
447
448 sendit:
449 #ifdef IPSEC
450         /* get SP for this packet */
451         if (so == NULL)
452                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
453         else
454                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
455
456         if (sp == NULL) {
457                 ipsecstat.out_inval++;
458                 goto bad;
459         }
460
461         error = 0;
462
463         /* check policy */
464         switch (sp->policy) {
465         case IPSEC_POLICY_DISCARD:
466                 /*
467                  * This packet is just discarded.
468                  */
469                 ipsecstat.out_polvio++;
470                 goto bad;
471
472         case IPSEC_POLICY_BYPASS:
473         case IPSEC_POLICY_NONE:
474                 /* no need to do IPsec. */
475                 goto skip_ipsec;
476
477         case IPSEC_POLICY_IPSEC:
478                 if (sp->req == NULL) {
479                         /* acquire a policy */
480                         error = key_spdacquire(sp);
481                         goto bad;
482                 }
483                 break;
484
485         case IPSEC_POLICY_ENTRUST:
486         default:
487                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
488         }
489     {
490         struct ipsec_output_state state;
491         bzero(&state, sizeof state);
492         state.m = m;
493         if (flags & IP_ROUTETOIF) {
494                 state.ro = &iproute;
495                 bzero(&iproute, sizeof iproute);
496         } else
497                 state.ro = ro;
498         state.dst = (struct sockaddr *)dst;
499
500         ip->ip_sum = 0;
501
502         /*
503          * XXX
504          * delayed checksums are not currently compatible with IPsec
505          */
506         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
507                 in_delayed_cksum(m);
508                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
509         }
510
511         ip->ip_len = htons(ip->ip_len);
512         ip->ip_off = htons(ip->ip_off);
513
514         error = ipsec4_output(&state, sp, flags);
515
516         m = state.m;
517         if (flags & IP_ROUTETOIF) {
518                 /*
519                  * if we have tunnel mode SA, we may need to ignore
520                  * IP_ROUTETOIF.
521                  */
522                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
523                         flags &= ~IP_ROUTETOIF;
524                         ro = state.ro;
525                 }
526         } else
527                 ro = state.ro;
528         dst = (struct sockaddr_in *)state.dst;
529         if (error) {
530                 /* mbuf is already reclaimed in ipsec4_output. */
531                 m0 = NULL;
532                 switch (error) {
533                 case EHOSTUNREACH:
534                 case ENETUNREACH:
535                 case EMSGSIZE:
536                 case ENOBUFS:
537                 case ENOMEM:
538                         break;
539                 default:
540                         printf("ip4_output (ipsec): error code %d\n", error);
541                         /*fall through*/
542                 case ENOENT:
543                         /* don't show these error codes to the user */
544                         error = 0;
545                         break;
546                 }
547                 goto bad;
548         }
549     }
550
551         /* be sure to update variables that are affected by ipsec4_output() */
552         ip = mtod(m, struct ip *);
553 #ifdef _IP_VHL
554         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
555 #else
556         hlen = ip->ip_hl << 2;
557 #endif
558         if (ro->ro_rt == NULL) {
559                 if (!(flags & IP_ROUTETOIF)) {
560                         printf("ip_output: "
561                                 "can't update route after IPsec processing\n");
562                         error = EHOSTUNREACH;   /*XXX*/
563                         goto bad;
564                 }
565         } else {
566                 ia = ifatoia(ro->ro_rt->rt_ifa);
567                 ifp = ro->ro_rt->rt_ifp;
568         }
569
570         /* make it flipped, again. */
571         ip->ip_len = ntohs(ip->ip_len);
572         ip->ip_off = ntohs(ip->ip_off);
573 skip_ipsec:
574 #endif /*IPSEC*/
575 #ifdef FAST_IPSEC
576         /*
577          * Check the security policy (SP) for the packet and, if
578          * required, do IPsec-related processing.  There are two
579          * cases here; the first time a packet is sent through
580          * it will be untagged and handled by ipsec4_checkpolicy.
581          * If the packet is resubmitted to ip_output (e.g. after
582          * AH, ESP, etc. processing), there will be a tag to bypass
583          * the lookup and related policy checking.
584          */
585         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
586         crit_enter();
587         if (mtag != NULL) {
588                 tdbi = (struct tdb_ident *)(mtag + 1);
589                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
590                 if (sp == NULL)
591                         error = -EINVAL;        /* force silent drop */
592                 m_tag_delete(m, mtag);
593         } else {
594                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
595                                         &error, inp);
596         }
597         /*
598          * There are four return cases:
599          *    sp != NULL                    apply IPsec policy
600          *    sp == NULL, error == 0        no IPsec handling needed
601          *    sp == NULL, error == -EINVAL  discard packet w/o error
602          *    sp == NULL, error != 0        discard packet, report error
603          */
604         if (sp != NULL) {
605                 /* Loop detection, check if ipsec processing already done */
606                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
607                 for (mtag = m_tag_first(m); mtag != NULL;
608                      mtag = m_tag_next(m, mtag)) {
609                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
610                                 continue;
611                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
612                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
613                                 continue;
614                         /*
615                          * Check if policy has an SA associated with it.
616                          * This can happen when an SP has yet to acquire
617                          * an SA; e.g. on first reference.  If it occurs,
618                          * then we let ipsec4_process_packet do its thing.
619                          */
620                         if (sp->req->sav == NULL)
621                                 break;
622                         tdbi = (struct tdb_ident *)(mtag + 1);
623                         if (tdbi->spi == sp->req->sav->spi &&
624                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
625                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
626                                  sizeof(union sockaddr_union)) == 0) {
627                                 /*
628                                  * No IPsec processing is needed, free
629                                  * reference to SP.
630                                  *
631                                  * NB: null pointer to avoid free at
632                                  *     done: below.
633                                  */
634                                 KEY_FREESP(&sp), sp = NULL;
635                                 crit_exit();
636                                 goto spd_done;
637                         }
638                 }
639
640                 /*
641                  * Do delayed checksums now because we send before
642                  * this is done in the normal processing path.
643                  */
644                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
645                         in_delayed_cksum(m);
646                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
647                 }
648
649                 ip->ip_len = htons(ip->ip_len);
650                 ip->ip_off = htons(ip->ip_off);
651
652                 /* NB: callee frees mbuf */
653                 error = ipsec4_process_packet(m, sp->req, flags, 0);
654                 /*
655                  * Preserve KAME behaviour: ENOENT can be returned
656                  * when an SA acquire is in progress.  Don't propagate
657                  * this to user-level; it confuses applications.
658                  *
659                  * XXX this will go away when the SADB is redone.
660                  */
661                 if (error == ENOENT)
662                         error = 0;
663                 crit_exit();
664                 goto done;
665         } else {
666                 crit_exit();
667
668                 if (error != 0) {
669                         /*
670                          * Hack: -EINVAL is used to signal that a packet
671                          * should be silently discarded.  This is typically
672                          * because we asked key management for an SA and
673                          * it was delayed (e.g. kicked up to IKE).
674                          */
675                         if (error == -EINVAL)
676                                 error = 0;
677                         goto bad;
678                 } else {
679                         /* No IPsec processing for this packet. */
680                 }
681 #ifdef notyet
682                 /*
683                  * If deferred crypto processing is needed, check that
684                  * the interface supports it.
685                  */
686                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
687                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
688                         /* notify IPsec to do its own crypto */
689                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
690                         error = EHOSTUNREACH;
691                         goto bad;
692                 }
693 #endif
694         }
695 spd_done:
696 #endif /* FAST_IPSEC */
697         /*
698          * IpHack's section.
699          * - Xlate: translate packet's addr/port (NAT).
700          * - Firewall: deny/allow/etc.
701          * - Wrap: fake packet's addr/port <unimpl.>
702          * - Encapsulate: put it in another IP and send out. <unimp.>
703          */
704
705         /*
706          * Run through list of hooks for output packets.
707          */
708         if (pfil_has_hooks(&inet_pfil_hook)) {
709                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
710                 if (error != 0 || m == NULL)
711                         goto done;
712                 ip = mtod(m, struct ip *);
713         }
714
715         /*
716          * Check with the firewall...
717          * but not if we are already being fwd'd from a firewall.
718          */
719         if (fw_enable && IPFW_LOADED && !args.next_hop) {
720                 struct sockaddr_in *old = dst;
721
722                 args.m = m;
723                 args.next_hop = dst;
724                 args.oif = ifp;
725                 off = ip_fw_chk_ptr(&args);
726                 m = args.m;
727                 dst = args.next_hop;
728
729                 /*
730                  * On return we must do the following:
731                  * m == NULL    -> drop the pkt (old interface, deprecated)
732                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
733                  * 1<=off<= 0xffff              -> DIVERT
734                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
735                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
736                  * dst != old                   -> IPFIREWALL_FORWARD
737                  * off==0, dst==old             -> accept
738                  * If some of the above modules are not compiled in, then
739                  * we should't have to check the corresponding condition
740                  * (because the ipfw control socket should not accept
741                  * unsupported rules), but better play safe and drop
742                  * packets in case of doubt.
743                  */
744                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
745                         if (m)
746                                 m_freem(m);
747                         error = EACCES;
748                         goto done;
749                 }
750                 ip = mtod(m, struct ip *);
751                 if (off == 0 && dst == old)             /* common case */
752                         goto pass;
753                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG)) {
754                         /*
755                          * pass the pkt to dummynet. Need to include
756                          * pipe number, m, ifp, ro, dst because these are
757                          * not recomputed in the next pass.
758                          * All other parameters have been already used and
759                          * so they are not needed anymore.
760                          * XXX note: if the ifp or ro entry are deleted
761                          * while a pkt is in dummynet, we are in trouble!
762                          */
763                         args.ro = ro;
764                         args.dst = dst;
765                         args.flags = flags;
766
767                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
768                                 &args);
769                         goto done;
770                 }
771 #ifdef IPDIVERT
772                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
773                         struct mbuf *clone = NULL;
774
775                         /* Clone packet if we're doing a 'tee' */
776                         if ((off & IP_FW_PORT_TEE_FLAG))
777                                 clone = m_dup(m, MB_DONTWAIT);
778
779                         /*
780                          * XXX
781                          * delayed checksums are not currently compatible
782                          * with divert sockets.
783                          */
784                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
785                                 in_delayed_cksum(m);
786                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
787                         }
788
789                         /* Restore packet header fields to original values */
790                         ip->ip_len = htons(ip->ip_len);
791                         ip->ip_off = htons(ip->ip_off);
792
793                         /* Deliver packet to divert input routine */
794                         divert_packet(m, 0, off & 0xffff);
795
796                         /* If 'tee', continue with original packet */
797                         if (clone != NULL) {
798                                 m = clone;
799                                 ip = mtod(m, struct ip *);
800                                 goto pass;
801                         }
802                         goto done;
803                 }
804 #endif
805
806                 /* IPFIREWALL_FORWARD */
807                 /*
808                  * Check dst to make sure it is directly reachable on the
809                  * interface we previously thought it was.
810                  * If it isn't (which may be likely in some situations) we have
811                  * to re-route it (ie, find a route for the next-hop and the
812                  * associated interface) and set them here. This is nested
813                  * forwarding which in most cases is undesirable, except where
814                  * such control is nigh impossible. So we do it here.
815                  * And I'm babbling.
816                  */
817                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
818 #if 0
819                         /*
820                          * XXX To improve readability, this block should be
821                          * changed into a function call as below:
822                          */
823                         error = ip_ipforward(&m, &dst, &ifp);
824                         if (error)
825                                 goto bad;
826                         if (m == NULL) /* ip_input consumed the mbuf */
827                                 goto done;
828 #else
829                         struct in_ifaddr *ia;
830
831                         /*
832                          * XXX sro_fwd below is static, and a pointer
833                          * to it gets passed to routines downstream.
834                          * This could have surprisingly bad results in
835                          * practice, because its content is overwritten
836                          * by subsequent packets.
837                          */
838                         /* There must be a better way to do this next line... */
839                         static struct route sro_fwd;
840                         struct route *ro_fwd = &sro_fwd;
841
842 #if 0
843                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
844                             dst->sin_addr, "\n");
845 #endif
846
847                         /*
848                          * We need to figure out if we have been forwarded
849                          * to a local socket. If so, then we should somehow
850                          * "loop back" to ip_input, and get directed to the
851                          * PCB as if we had received this packet. This is
852                          * because it may be dificult to identify the packets
853                          * you want to forward until they are being output
854                          * and have selected an interface. (e.g. locally
855                          * initiated packets) If we used the loopback inteface,
856                          * we would not be able to control what happens
857                          * as the packet runs through ip_input() as
858                          * it is done through a ISR.
859                          */
860                         LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
861                                      ia_hash) {
862                                 /*
863                                  * If the addr to forward to is one
864                                  * of ours, we pretend to
865                                  * be the destination for this packet.
866                                  */
867                                 if (IA_SIN(ia)->sin_addr.s_addr ==
868                                                  dst->sin_addr.s_addr)
869                                         break;
870                         }
871                         if (ia != NULL) {    /* tell ip_input "dont filter" */
872                                 struct m_hdr tag;
873
874                                 tag.mh_type = MT_TAG;
875                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
876                                 tag.mh_data = (caddr_t)args.next_hop;
877                                 tag.mh_next = m;
878
879                                 if (m->m_pkthdr.rcvif == NULL)
880                                         m->m_pkthdr.rcvif = ifunit("lo0");
881                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
882                                         m->m_pkthdr.csum_flags |=
883                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
884                                         m0->m_pkthdr.csum_data = 0xffff;
885                                 }
886                                 m->m_pkthdr.csum_flags |=
887                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
888                                 ip->ip_len = htons(ip->ip_len);
889                                 ip->ip_off = htons(ip->ip_off);
890                                 ip_input((struct mbuf *)&tag);
891                                 goto done;
892                         }
893                         /* Some of the logic for this was nicked from above.
894                          *
895                          * This rewrites the cached route in a local PCB.
896                          * Is this what we want to do?
897                          */
898                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
899                         ro_fwd->ro_rt = NULL;
900
901                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
902                         if (ro_fwd->ro_rt == NULL) {
903                                 ipstat.ips_noroute++;
904                                 error = EHOSTUNREACH;
905                                 goto bad;
906                         }
907
908                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
909                         ifp = ro_fwd->ro_rt->rt_ifp;
910                         ro_fwd->ro_rt->rt_use++;
911                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
912                                 dst = (struct sockaddr_in *)
913                                     ro_fwd->ro_rt->rt_gateway;
914                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
915                                 isbroadcast =
916                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
917                         else
918                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
919                         if (ro->ro_rt != NULL)
920                                 rtfree(ro->ro_rt);
921                         ro->ro_rt = ro_fwd->ro_rt;
922                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
923
924 #endif  /* ... block to be put into a function */
925                         /*
926                          * If we added a default src ip earlier,
927                          * which would have been gotten from the-then
928                          * interface, do it again, from the new one.
929                          */
930                         if (src_was_INADDR_ANY)
931                                 ip->ip_src = IA_SIN(ia)->sin_addr;
932                         goto pass ;
933                 }
934
935                 /*
936                  * if we get here, none of the above matches, and
937                  * we have to drop the pkt
938                  */
939                 m_freem(m);
940                 error = EACCES; /* not sure this is the right error msg */
941                 goto done;
942         }
943
944 pass:
945         /* 127/8 must not appear on wire - RFC1122. */
946         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
947             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
948                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
949                         ipstat.ips_badaddr++;
950                         error = EADDRNOTAVAIL;
951                         goto bad;
952                 }
953         }
954
955         m->m_pkthdr.csum_flags |= CSUM_IP;
956         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
957         if (sw_csum & CSUM_DELAY_DATA) {
958                 in_delayed_cksum(m);
959                 sw_csum &= ~CSUM_DELAY_DATA;
960         }
961         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
962
963         /*
964          * If small enough for interface, or the interface will take
965          * care of the fragmentation for us, can just send directly.
966          */
967         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
968             !(ip->ip_off & IP_DF))) {
969                 ip->ip_len = htons(ip->ip_len);
970                 ip->ip_off = htons(ip->ip_off);
971                 ip->ip_sum = 0;
972                 if (sw_csum & CSUM_DELAY_IP) {
973                         if (ip->ip_vhl == IP_VHL_BORING) {
974                                 ip->ip_sum = in_cksum_hdr(ip);
975                         } else {
976                                 ip->ip_sum = in_cksum(m, hlen);
977                         }
978                 }
979
980                 /* Record statistics for this interface address. */
981                 if (!(flags & IP_FORWARDING) && ia) {
982                         ia->ia_ifa.if_opackets++;
983                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
984                 }
985
986 #ifdef IPSEC
987                 /* clean ipsec history once it goes out of the node */
988                 ipsec_delaux(m);
989 #endif
990
991 #ifdef MBUF_STRESS_TEST
992                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
993                         struct mbuf *m1, *m2;
994                         int length, tmp;
995
996                         tmp = length = m->m_pkthdr.len;
997
998                         while ((length -= mbuf_frag_size) >= 1) {
999                                 m1 = m_split(m, length, MB_DONTWAIT);
1000                                 if (m1 == NULL)
1001                                         break;
1002                                 m2 = m;
1003                                 while (m2->m_next != NULL)
1004                                         m2 = m2->m_next;
1005                                 m2->m_next = m1;
1006                         }
1007                         m->m_pkthdr.len = tmp;
1008                 }
1009 #endif
1010                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1011                                           ro->ro_rt);
1012                 goto done;
1013         }
1014
1015         if (ip->ip_off & IP_DF) {
1016                 error = EMSGSIZE;
1017                 /*
1018                  * This case can happen if the user changed the MTU
1019                  * of an interface after enabling IP on it.  Because
1020                  * most netifs don't keep track of routes pointing to
1021                  * them, there is no way for one to update all its
1022                  * routes when the MTU is changed.
1023                  */
1024                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1025                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1026                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1027                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1028                 }
1029                 ipstat.ips_cantfrag++;
1030                 goto bad;
1031         }
1032
1033         /*
1034          * Too large for interface; fragment if possible. If successful,
1035          * on return, m will point to a list of packets to be sent.
1036          */
1037         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1038         if (error)
1039                 goto bad;
1040         for (; m; m = m0) {
1041                 m0 = m->m_nextpkt;
1042                 m->m_nextpkt = NULL;
1043 #ifdef IPSEC
1044                 /* clean ipsec history once it goes out of the node */
1045                 ipsec_delaux(m);
1046 #endif
1047                 if (error == 0) {
1048                         /* Record statistics for this interface address. */
1049                         if (ia != NULL) {
1050                                 ia->ia_ifa.if_opackets++;
1051                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1052                         }
1053                         error = (*ifp->if_output)(ifp, m,
1054                                                   (struct sockaddr *)dst,
1055                                                   ro->ro_rt);
1056                 } else
1057                         m_freem(m);
1058         }
1059
1060         if (error == 0)
1061                 ipstat.ips_fragmented++;
1062
1063 done:
1064         if (ro == &iproute && ro->ro_rt != NULL) {
1065                 RTFREE(ro->ro_rt);
1066                 ro->ro_rt = NULL;
1067         }
1068 #ifdef IPSEC
1069         if (sp != NULL) {
1070                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1071                         printf("DP ip_output call free SP:%p\n", sp));
1072                 key_freesp(sp);
1073         }
1074 #endif
1075 #ifdef FAST_IPSEC
1076         if (sp != NULL)
1077                 KEY_FREESP(&sp);
1078 #endif
1079         return (error);
1080 bad:
1081         m_freem(m);
1082         goto done;
1083 }
1084
1085 /*
1086  * Create a chain of fragments which fit the given mtu. m_frag points to the
1087  * mbuf to be fragmented; on return it points to the chain with the fragments.
1088  * Return 0 if no error. If error, m_frag may contain a partially built
1089  * chain of fragments that should be freed by the caller.
1090  *
1091  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1092  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1093  */
1094 int
1095 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1096             u_long if_hwassist_flags, int sw_csum)
1097 {
1098         int error = 0;
1099         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1100         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1101         int off;
1102         struct mbuf *m0 = *m_frag;      /* the original packet          */
1103         int firstlen;
1104         struct mbuf **mnext;
1105         int nfrags;
1106
1107         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1108                 ipstat.ips_cantfrag++;
1109                 return EMSGSIZE;
1110         }
1111
1112         /*
1113          * Must be able to put at least 8 bytes per fragment.
1114          */
1115         if (len < 8)
1116                 return EMSGSIZE;
1117
1118         /*
1119          * If the interface will not calculate checksums on
1120          * fragmented packets, then do it here.
1121          */
1122         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1123             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1124                 in_delayed_cksum(m0);
1125                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1126         }
1127
1128         if (len > PAGE_SIZE) {
1129                 /*
1130                  * Fragment large datagrams such that each segment
1131                  * contains a multiple of PAGE_SIZE amount of data,
1132                  * plus headers. This enables a receiver to perform
1133                  * page-flipping zero-copy optimizations.
1134                  *
1135                  * XXX When does this help given that sender and receiver
1136                  * could have different page sizes, and also mtu could
1137                  * be less than the receiver's page size ?
1138                  */
1139                 int newlen;
1140                 struct mbuf *m;
1141
1142                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1143                         off += m->m_len;
1144
1145                 /*
1146                  * firstlen (off - hlen) must be aligned on an
1147                  * 8-byte boundary
1148                  */
1149                 if (off < hlen)
1150                         goto smart_frag_failure;
1151                 off = ((off - hlen) & ~7) + hlen;
1152                 newlen = (~PAGE_MASK) & mtu;
1153                 if ((newlen + sizeof(struct ip)) > mtu) {
1154                         /* we failed, go back the default */
1155 smart_frag_failure:
1156                         newlen = len;
1157                         off = hlen + len;
1158                 }
1159                 len = newlen;
1160
1161         } else {
1162                 off = hlen + len;
1163         }
1164
1165         firstlen = off - hlen;
1166         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1167
1168         /*
1169          * Loop through length of segment after first fragment,
1170          * make new header and copy data of each part and link onto chain.
1171          * Here, m0 is the original packet, m is the fragment being created.
1172          * The fragments are linked off the m_nextpkt of the original
1173          * packet, which after processing serves as the first fragment.
1174          */
1175         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1176                 struct ip *mhip;        /* ip header on the fragment */
1177                 struct mbuf *m;
1178                 int mhlen = sizeof(struct ip);
1179
1180                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1181                 if (m == NULL) {
1182                         error = ENOBUFS;
1183                         ipstat.ips_odropped++;
1184                         goto done;
1185                 }
1186                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1187                 /*
1188                  * In the first mbuf, leave room for the link header, then
1189                  * copy the original IP header including options. The payload
1190                  * goes into an additional mbuf chain returned by m_copy().
1191                  */
1192                 m->m_data += max_linkhdr;
1193                 mhip = mtod(m, struct ip *);
1194                 *mhip = *ip;
1195                 if (hlen > sizeof(struct ip)) {
1196                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1197                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1198                 }
1199                 m->m_len = mhlen;
1200                 /* XXX do we need to add ip->ip_off below ? */
1201                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1202                 if (off + len >= ip->ip_len) {  /* last fragment */
1203                         len = ip->ip_len - off;
1204                         m->m_flags |= M_LASTFRAG;
1205                 } else
1206                         mhip->ip_off |= IP_MF;
1207                 mhip->ip_len = htons((u_short)(len + mhlen));
1208                 m->m_next = m_copy(m0, off, len);
1209                 if (m->m_next == NULL) {                /* copy failed */
1210                         m_free(m);
1211                         error = ENOBUFS;        /* ??? */
1212                         ipstat.ips_odropped++;
1213                         goto done;
1214                 }
1215                 m->m_pkthdr.len = mhlen + len;
1216                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1217                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1218                 mhip->ip_off = htons(mhip->ip_off);
1219                 mhip->ip_sum = 0;
1220                 if (sw_csum & CSUM_DELAY_IP)
1221                         mhip->ip_sum = in_cksum(m, mhlen);
1222                 *mnext = m;
1223                 mnext = &m->m_nextpkt;
1224         }
1225         ipstat.ips_ofragments += nfrags;
1226
1227         /* set first marker for fragment chain */
1228         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1229         m0->m_pkthdr.csum_data = nfrags;
1230
1231         /*
1232          * Update first fragment by trimming what's been copied out
1233          * and updating header.
1234          */
1235         m_adj(m0, hlen + firstlen - ip->ip_len);
1236         m0->m_pkthdr.len = hlen + firstlen;
1237         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1238         ip->ip_off |= IP_MF;
1239         ip->ip_off = htons(ip->ip_off);
1240         ip->ip_sum = 0;
1241         if (sw_csum & CSUM_DELAY_IP)
1242                 ip->ip_sum = in_cksum(m0, hlen);
1243
1244 done:
1245         *m_frag = m0;
1246         return error;
1247 }
1248
1249 void
1250 in_delayed_cksum(struct mbuf *m)
1251 {
1252         struct ip *ip;
1253         u_short csum, offset;
1254
1255         ip = mtod(m, struct ip *);
1256         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1257         csum = in_cksum_skip(m, ip->ip_len, offset);
1258         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1259                 csum = 0xffff;
1260         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1261
1262         if (offset + sizeof(u_short) > m->m_len) {
1263                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1264                     m->m_len, offset, ip->ip_p);
1265                 /*
1266                  * XXX
1267                  * this shouldn't happen, but if it does, the
1268                  * correct behavior may be to insert the checksum
1269                  * in the existing chain instead of rearranging it.
1270                  */
1271                 m = m_pullup(m, offset + sizeof(u_short));
1272         }
1273         *(u_short *)(m->m_data + offset) = csum;
1274 }
1275
1276 /*
1277  * Insert IP options into preformed packet.
1278  * Adjust IP destination as required for IP source routing,
1279  * as indicated by a non-zero in_addr at the start of the options.
1280  *
1281  * XXX This routine assumes that the packet has no options in place.
1282  */
1283 static struct mbuf *
1284 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1285 {
1286         struct ipoption *p = mtod(opt, struct ipoption *);
1287         struct mbuf *n;
1288         struct ip *ip = mtod(m, struct ip *);
1289         unsigned optlen;
1290
1291         optlen = opt->m_len - sizeof p->ipopt_dst;
1292         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1293                 *phlen = 0;
1294                 return (m);             /* XXX should fail */
1295         }
1296         if (p->ipopt_dst.s_addr)
1297                 ip->ip_dst = p->ipopt_dst;
1298         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1299                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1300                 if (n == NULL) {
1301                         *phlen = 0;
1302                         return (m);
1303                 }
1304                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1305                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1306                 m->m_len -= sizeof(struct ip);
1307                 m->m_data += sizeof(struct ip);
1308                 n->m_next = m;
1309                 m = n;
1310                 m->m_len = optlen + sizeof(struct ip);
1311                 m->m_data += max_linkhdr;
1312                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1313         } else {
1314                 m->m_data -= optlen;
1315                 m->m_len += optlen;
1316                 m->m_pkthdr.len += optlen;
1317                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1318         }
1319         ip = mtod(m, struct ip *);
1320         bcopy(p->ipopt_list, ip + 1, optlen);
1321         *phlen = sizeof(struct ip) + optlen;
1322         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1323         ip->ip_len += optlen;
1324         return (m);
1325 }
1326
1327 /*
1328  * Copy options from ip to jp,
1329  * omitting those not copied during fragmentation.
1330  */
1331 int
1332 ip_optcopy(struct ip *ip, struct ip *jp)
1333 {
1334         u_char *cp, *dp;
1335         int opt, optlen, cnt;
1336
1337         cp = (u_char *)(ip + 1);
1338         dp = (u_char *)(jp + 1);
1339         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1340         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1341                 opt = cp[0];
1342                 if (opt == IPOPT_EOL)
1343                         break;
1344                 if (opt == IPOPT_NOP) {
1345                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1346                         *dp++ = IPOPT_NOP;
1347                         optlen = 1;
1348                         continue;
1349                 }
1350
1351                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1352                     ("ip_optcopy: malformed ipv4 option"));
1353                 optlen = cp[IPOPT_OLEN];
1354                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1355                     ("ip_optcopy: malformed ipv4 option"));
1356
1357                 /* bogus lengths should have been caught by ip_dooptions */
1358                 if (optlen > cnt)
1359                         optlen = cnt;
1360                 if (IPOPT_COPIED(opt)) {
1361                         bcopy(cp, dp, optlen);
1362                         dp += optlen;
1363                 }
1364         }
1365         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1366                 *dp++ = IPOPT_EOL;
1367         return (optlen);
1368 }
1369
1370 /*
1371  * IP socket option processing.
1372  */
1373 int
1374 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1375 {
1376         struct  inpcb *inp = so->so_pcb;
1377         int     error, optval;
1378
1379         error = optval = 0;
1380         if (sopt->sopt_level != IPPROTO_IP) {
1381                 return (EINVAL);
1382         }
1383
1384         switch (sopt->sopt_dir) {
1385         case SOPT_SET:
1386                 switch (sopt->sopt_name) {
1387                 case IP_OPTIONS:
1388 #ifdef notyet
1389                 case IP_RETOPTS:
1390 #endif
1391                 {
1392                         struct mbuf *m;
1393                         if (sopt->sopt_valsize > MLEN) {
1394                                 error = EMSGSIZE;
1395                                 break;
1396                         }
1397                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1398                         if (m == NULL) {
1399                                 error = ENOBUFS;
1400                                 break;
1401                         }
1402                         m->m_len = sopt->sopt_valsize;
1403                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1404                                             m->m_len);
1405
1406                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1407                                            m));
1408                 }
1409
1410                 case IP_TOS:
1411                 case IP_TTL:
1412                 case IP_RECVOPTS:
1413                 case IP_RECVRETOPTS:
1414                 case IP_RECVDSTADDR:
1415                 case IP_RECVIF:
1416                 case IP_FAITH:
1417                         error = sooptcopyin(sopt, &optval, sizeof optval,
1418                                             sizeof optval);
1419                         if (error)
1420                                 break;
1421
1422                         switch (sopt->sopt_name) {
1423                         case IP_TOS:
1424                                 inp->inp_ip_tos = optval;
1425                                 break;
1426
1427                         case IP_TTL:
1428                                 inp->inp_ip_ttl = optval;
1429                                 break;
1430 #define OPTSET(bit) \
1431         if (optval) \
1432                 inp->inp_flags |= bit; \
1433         else \
1434                 inp->inp_flags &= ~bit;
1435
1436                         case IP_RECVOPTS:
1437                                 OPTSET(INP_RECVOPTS);
1438                                 break;
1439
1440                         case IP_RECVRETOPTS:
1441                                 OPTSET(INP_RECVRETOPTS);
1442                                 break;
1443
1444                         case IP_RECVDSTADDR:
1445                                 OPTSET(INP_RECVDSTADDR);
1446                                 break;
1447
1448                         case IP_RECVIF:
1449                                 OPTSET(INP_RECVIF);
1450                                 break;
1451
1452                         case IP_FAITH:
1453                                 OPTSET(INP_FAITH);
1454                                 break;
1455                         }
1456                         break;
1457 #undef OPTSET
1458
1459                 case IP_MULTICAST_IF:
1460                 case IP_MULTICAST_VIF:
1461                 case IP_MULTICAST_TTL:
1462                 case IP_MULTICAST_LOOP:
1463                 case IP_ADD_MEMBERSHIP:
1464                 case IP_DROP_MEMBERSHIP:
1465                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1466                         break;
1467
1468                 case IP_PORTRANGE:
1469                         error = sooptcopyin(sopt, &optval, sizeof optval,
1470                                             sizeof optval);
1471                         if (error)
1472                                 break;
1473
1474                         switch (optval) {
1475                         case IP_PORTRANGE_DEFAULT:
1476                                 inp->inp_flags &= ~(INP_LOWPORT);
1477                                 inp->inp_flags &= ~(INP_HIGHPORT);
1478                                 break;
1479
1480                         case IP_PORTRANGE_HIGH:
1481                                 inp->inp_flags &= ~(INP_LOWPORT);
1482                                 inp->inp_flags |= INP_HIGHPORT;
1483                                 break;
1484
1485                         case IP_PORTRANGE_LOW:
1486                                 inp->inp_flags &= ~(INP_HIGHPORT);
1487                                 inp->inp_flags |= INP_LOWPORT;
1488                                 break;
1489
1490                         default:
1491                                 error = EINVAL;
1492                                 break;
1493                         }
1494                         break;
1495
1496 #if defined(IPSEC) || defined(FAST_IPSEC)
1497                 case IP_IPSEC_POLICY:
1498                 {
1499                         caddr_t req;
1500                         size_t len = 0;
1501                         int priv;
1502                         struct mbuf *m;
1503                         int optname;
1504
1505                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1506                                 break;
1507                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1508                                 break;
1509                         priv = (sopt->sopt_td != NULL &&
1510                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1511                         req = mtod(m, caddr_t);
1512                         len = m->m_len;
1513                         optname = sopt->sopt_name;
1514                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1515                         m_freem(m);
1516                         break;
1517                 }
1518 #endif /*IPSEC*/
1519
1520                 default:
1521                         error = ENOPROTOOPT;
1522                         break;
1523                 }
1524                 break;
1525
1526         case SOPT_GET:
1527                 switch (sopt->sopt_name) {
1528                 case IP_OPTIONS:
1529                 case IP_RETOPTS:
1530                         if (inp->inp_options)
1531                                 error = sooptcopyout(sopt,
1532                                                      mtod(inp->inp_options,
1533                                                           char *),
1534                                                      inp->inp_options->m_len);
1535                         else
1536                                 sopt->sopt_valsize = 0;
1537                         break;
1538
1539                 case IP_TOS:
1540                 case IP_TTL:
1541                 case IP_RECVOPTS:
1542                 case IP_RECVRETOPTS:
1543                 case IP_RECVDSTADDR:
1544                 case IP_RECVIF:
1545                 case IP_PORTRANGE:
1546                 case IP_FAITH:
1547                         switch (sopt->sopt_name) {
1548
1549                         case IP_TOS:
1550                                 optval = inp->inp_ip_tos;
1551                                 break;
1552
1553                         case IP_TTL:
1554                                 optval = inp->inp_ip_ttl;
1555                                 break;
1556
1557 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1558
1559                         case IP_RECVOPTS:
1560                                 optval = OPTBIT(INP_RECVOPTS);
1561                                 break;
1562
1563                         case IP_RECVRETOPTS:
1564                                 optval = OPTBIT(INP_RECVRETOPTS);
1565                                 break;
1566
1567                         case IP_RECVDSTADDR:
1568                                 optval = OPTBIT(INP_RECVDSTADDR);
1569                                 break;
1570
1571                         case IP_RECVIF:
1572                                 optval = OPTBIT(INP_RECVIF);
1573                                 break;
1574
1575                         case IP_PORTRANGE:
1576                                 if (inp->inp_flags & INP_HIGHPORT)
1577                                         optval = IP_PORTRANGE_HIGH;
1578                                 else if (inp->inp_flags & INP_LOWPORT)
1579                                         optval = IP_PORTRANGE_LOW;
1580                                 else
1581                                         optval = 0;
1582                                 break;
1583
1584                         case IP_FAITH:
1585                                 optval = OPTBIT(INP_FAITH);
1586                                 break;
1587                         }
1588                         error = sooptcopyout(sopt, &optval, sizeof optval);
1589                         break;
1590
1591                 case IP_MULTICAST_IF:
1592                 case IP_MULTICAST_VIF:
1593                 case IP_MULTICAST_TTL:
1594                 case IP_MULTICAST_LOOP:
1595                 case IP_ADD_MEMBERSHIP:
1596                 case IP_DROP_MEMBERSHIP:
1597                         error = ip_getmoptions(sopt, inp->inp_moptions);
1598                         break;
1599
1600 #if defined(IPSEC) || defined(FAST_IPSEC)
1601                 case IP_IPSEC_POLICY:
1602                 {
1603                         struct mbuf *m = NULL;
1604                         caddr_t req = NULL;
1605                         size_t len = 0;
1606
1607                         if (m != NULL) {
1608                                 req = mtod(m, caddr_t);
1609                                 len = m->m_len;
1610                         }
1611                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1612                         if (error == 0)
1613                                 error = soopt_mcopyout(sopt, m); /* XXX */
1614                         if (error == 0)
1615                                 m_freem(m);
1616                         break;
1617                 }
1618 #endif /*IPSEC*/
1619
1620                 default:
1621                         error = ENOPROTOOPT;
1622                         break;
1623                 }
1624                 break;
1625         }
1626         return (error);
1627 }
1628
1629 /*
1630  * Set up IP options in pcb for insertion in output packets.
1631  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1632  * with destination address if source routed.
1633  */
1634 static int
1635 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1636 {
1637         int cnt, optlen;
1638         u_char *cp;
1639         u_char opt;
1640
1641         /* turn off any old options */
1642         if (*pcbopt)
1643                 m_free(*pcbopt);
1644         *pcbopt = 0;
1645         if (m == NULL || m->m_len == 0) {
1646                 /*
1647                  * Only turning off any previous options.
1648                  */
1649                 if (m != NULL)
1650                         m_free(m);
1651                 return (0);
1652         }
1653
1654         if (m->m_len % sizeof(int32_t))
1655                 goto bad;
1656         /*
1657          * IP first-hop destination address will be stored before
1658          * actual options; move other options back
1659          * and clear it when none present.
1660          */
1661         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1662                 goto bad;
1663         cnt = m->m_len;
1664         m->m_len += sizeof(struct in_addr);
1665         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1666         ovbcopy(mtod(m, caddr_t), cp, cnt);
1667         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1668
1669         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1670                 opt = cp[IPOPT_OPTVAL];
1671                 if (opt == IPOPT_EOL)
1672                         break;
1673                 if (opt == IPOPT_NOP)
1674                         optlen = 1;
1675                 else {
1676                         if (cnt < IPOPT_OLEN + sizeof *cp)
1677                                 goto bad;
1678                         optlen = cp[IPOPT_OLEN];
1679                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1680                                 goto bad;
1681                 }
1682                 switch (opt) {
1683
1684                 default:
1685                         break;
1686
1687                 case IPOPT_LSRR:
1688                 case IPOPT_SSRR:
1689                         /*
1690                          * user process specifies route as:
1691                          *      ->A->B->C->D
1692                          * D must be our final destination (but we can't
1693                          * check that since we may not have connected yet).
1694                          * A is first hop destination, which doesn't appear in
1695                          * actual IP option, but is stored before the options.
1696                          */
1697                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1698                                 goto bad;
1699                         m->m_len -= sizeof(struct in_addr);
1700                         cnt -= sizeof(struct in_addr);
1701                         optlen -= sizeof(struct in_addr);
1702                         cp[IPOPT_OLEN] = optlen;
1703                         /*
1704                          * Move first hop before start of options.
1705                          */
1706                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1707                               sizeof(struct in_addr));
1708                         /*
1709                          * Then copy rest of options back
1710                          * to close up the deleted entry.
1711                          */
1712                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1713                                 &cp[IPOPT_OFFSET+1],
1714                                 cnt - (IPOPT_MINOFF - 1));
1715                         break;
1716                 }
1717         }
1718         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1719                 goto bad;
1720         *pcbopt = m;
1721         return (0);
1722
1723 bad:
1724         m_free(m);
1725         return (EINVAL);
1726 }
1727
1728 /*
1729  * XXX
1730  * The whole multicast option thing needs to be re-thought.
1731  * Several of these options are equally applicable to non-multicast
1732  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1733  * standard option (IP_TTL).
1734  */
1735
1736 /*
1737  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1738  */
1739 static struct ifnet *
1740 ip_multicast_if(struct in_addr *a, int *ifindexp)
1741 {
1742         int ifindex;
1743         struct ifnet *ifp;
1744
1745         if (ifindexp)
1746                 *ifindexp = 0;
1747         if (ntohl(a->s_addr) >> 24 == 0) {
1748                 ifindex = ntohl(a->s_addr) & 0xffffff;
1749                 if (ifindex < 0 || if_index < ifindex)
1750                         return NULL;
1751                 ifp = ifindex2ifnet[ifindex];
1752                 if (ifindexp)
1753                         *ifindexp = ifindex;
1754         } else {
1755                 INADDR_TO_IFP(*a, ifp);
1756         }
1757         return ifp;
1758 }
1759
1760 /*
1761  * Set the IP multicast options in response to user setsockopt().
1762  */
1763 static int
1764 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1765 {
1766         int error = 0;
1767         int i;
1768         struct in_addr addr;
1769         struct ip_mreq mreq;
1770         struct ifnet *ifp;
1771         struct ip_moptions *imo = *imop;
1772         int ifindex;
1773
1774         if (imo == NULL) {
1775                 /*
1776                  * No multicast option buffer attached to the pcb;
1777                  * allocate one and initialize to default values.
1778                  */
1779                 imo = malloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1780
1781                 if (imo == NULL)
1782                         return (ENOBUFS);
1783                 *imop = imo;
1784                 imo->imo_multicast_ifp = NULL;
1785                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1786                 imo->imo_multicast_vif = -1;
1787                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1788                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1789                 imo->imo_num_memberships = 0;
1790         }
1791
1792         switch (sopt->sopt_name) {
1793         /* store an index number for the vif you wanna use in the send */
1794         case IP_MULTICAST_VIF:
1795                 if (legal_vif_num == 0) {
1796                         error = EOPNOTSUPP;
1797                         break;
1798                 }
1799                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1800                 if (error)
1801                         break;
1802                 if (!legal_vif_num(i) && (i != -1)) {
1803                         error = EINVAL;
1804                         break;
1805                 }
1806                 imo->imo_multicast_vif = i;
1807                 break;
1808
1809         case IP_MULTICAST_IF:
1810                 /*
1811                  * Select the interface for outgoing multicast packets.
1812                  */
1813                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1814                 if (error)
1815                         break;
1816                 /*
1817                  * INADDR_ANY is used to remove a previous selection.
1818                  * When no interface is selected, a default one is
1819                  * chosen every time a multicast packet is sent.
1820                  */
1821                 if (addr.s_addr == INADDR_ANY) {
1822                         imo->imo_multicast_ifp = NULL;
1823                         break;
1824                 }
1825                 /*
1826                  * The selected interface is identified by its local
1827                  * IP address.  Find the interface and confirm that
1828                  * it supports multicasting.
1829                  */
1830                 crit_enter();
1831                 ifp = ip_multicast_if(&addr, &ifindex);
1832                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1833                         crit_exit();
1834                         error = EADDRNOTAVAIL;
1835                         break;
1836                 }
1837                 imo->imo_multicast_ifp = ifp;
1838                 if (ifindex)
1839                         imo->imo_multicast_addr = addr;
1840                 else
1841                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1842                 crit_exit();
1843                 break;
1844
1845         case IP_MULTICAST_TTL:
1846                 /*
1847                  * Set the IP time-to-live for outgoing multicast packets.
1848                  * The original multicast API required a char argument,
1849                  * which is inconsistent with the rest of the socket API.
1850                  * We allow either a char or an int.
1851                  */
1852                 if (sopt->sopt_valsize == 1) {
1853                         u_char ttl;
1854                         error = sooptcopyin(sopt, &ttl, 1, 1);
1855                         if (error)
1856                                 break;
1857                         imo->imo_multicast_ttl = ttl;
1858                 } else {
1859                         u_int ttl;
1860                         error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1861                         if (error)
1862                                 break;
1863                         if (ttl > 255)
1864                                 error = EINVAL;
1865                         else
1866                                 imo->imo_multicast_ttl = ttl;
1867                 }
1868                 break;
1869
1870         case IP_MULTICAST_LOOP:
1871                 /*
1872                  * Set the loopback flag for outgoing multicast packets.
1873                  * Must be zero or one.  The original multicast API required a
1874                  * char argument, which is inconsistent with the rest
1875                  * of the socket API.  We allow either a char or an int.
1876                  */
1877                 if (sopt->sopt_valsize == 1) {
1878                         u_char loop;
1879
1880                         error = sooptcopyin(sopt, &loop, 1, 1);
1881                         if (error)
1882                                 break;
1883                         imo->imo_multicast_loop = !!loop;
1884                 } else {
1885                         u_int loop;
1886
1887                         error = sooptcopyin(sopt, &loop, sizeof loop,
1888                                             sizeof loop);
1889                         if (error)
1890                                 break;
1891                         imo->imo_multicast_loop = !!loop;
1892                 }
1893                 break;
1894
1895         case IP_ADD_MEMBERSHIP:
1896                 /*
1897                  * Add a multicast group membership.
1898                  * Group must be a valid IP multicast address.
1899                  */
1900                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1901                 if (error)
1902                         break;
1903
1904                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1905                         error = EINVAL;
1906                         break;
1907                 }
1908                 crit_enter();
1909                 /*
1910                  * If no interface address was provided, use the interface of
1911                  * the route to the given multicast address.
1912                  */
1913                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1914                         struct sockaddr_in dst;
1915                         struct rtentry *rt;
1916
1917                         bzero(&dst, sizeof(struct sockaddr_in));
1918                         dst.sin_len = sizeof(struct sockaddr_in);
1919                         dst.sin_family = AF_INET;
1920                         dst.sin_addr = mreq.imr_multiaddr;
1921                         rt = rtlookup((struct sockaddr *)&dst);
1922                         if (rt == NULL) {
1923                                 error = EADDRNOTAVAIL;
1924                                 crit_exit();
1925                                 break;
1926                         }
1927                         --rt->rt_refcnt;
1928                         ifp = rt->rt_ifp;
1929                 } else {
1930                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1931                 }
1932
1933                 /*
1934                  * See if we found an interface, and confirm that it
1935                  * supports multicast.
1936                  */
1937                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1938                         error = EADDRNOTAVAIL;
1939                         crit_exit();
1940                         break;
1941                 }
1942                 /*
1943                  * See if the membership already exists or if all the
1944                  * membership slots are full.
1945                  */
1946                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1947                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1948                             imo->imo_membership[i]->inm_addr.s_addr
1949                                                 == mreq.imr_multiaddr.s_addr)
1950                                 break;
1951                 }
1952                 if (i < imo->imo_num_memberships) {
1953                         error = EADDRINUSE;
1954                         crit_exit();
1955                         break;
1956                 }
1957                 if (i == IP_MAX_MEMBERSHIPS) {
1958                         error = ETOOMANYREFS;
1959                         crit_exit();
1960                         break;
1961                 }
1962                 /*
1963                  * Everything looks good; add a new record to the multicast
1964                  * address list for the given interface.
1965                  */
1966                 if ((imo->imo_membership[i] =
1967                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1968                         error = ENOBUFS;
1969                         crit_exit();
1970                         break;
1971                 }
1972                 ++imo->imo_num_memberships;
1973                 crit_exit();
1974                 break;
1975
1976         case IP_DROP_MEMBERSHIP:
1977                 /*
1978                  * Drop a multicast group membership.
1979                  * Group must be a valid IP multicast address.
1980                  */
1981                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1982                 if (error)
1983                         break;
1984
1985                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1986                         error = EINVAL;
1987                         break;
1988                 }
1989
1990                 crit_enter();
1991                 /*
1992                  * If an interface address was specified, get a pointer
1993                  * to its ifnet structure.
1994                  */
1995                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1996                         ifp = NULL;
1997                 else {
1998                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1999                         if (ifp == NULL) {
2000                                 error = EADDRNOTAVAIL;
2001                                 crit_exit();
2002                                 break;
2003                         }
2004                 }
2005                 /*
2006                  * Find the membership in the membership array.
2007                  */
2008                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2009                         if ((ifp == NULL ||
2010                              imo->imo_membership[i]->inm_ifp == ifp) &&
2011                             imo->imo_membership[i]->inm_addr.s_addr ==
2012                             mreq.imr_multiaddr.s_addr)
2013                                 break;
2014                 }
2015                 if (i == imo->imo_num_memberships) {
2016                         error = EADDRNOTAVAIL;
2017                         crit_exit();
2018                         break;
2019                 }
2020                 /*
2021                  * Give up the multicast address record to which the
2022                  * membership points.
2023                  */
2024                 in_delmulti(imo->imo_membership[i]);
2025                 /*
2026                  * Remove the gap in the membership array.
2027                  */
2028                 for (++i; i < imo->imo_num_memberships; ++i)
2029                         imo->imo_membership[i-1] = imo->imo_membership[i];
2030                 --imo->imo_num_memberships;
2031                 crit_exit();
2032                 break;
2033
2034         default:
2035                 error = EOPNOTSUPP;
2036                 break;
2037         }
2038
2039         /*
2040          * If all options have default values, no need to keep the mbuf.
2041          */
2042         if (imo->imo_multicast_ifp == NULL &&
2043             imo->imo_multicast_vif == -1 &&
2044             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2045             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2046             imo->imo_num_memberships == 0) {
2047                 free(*imop, M_IPMOPTS);
2048                 *imop = NULL;
2049         }
2050
2051         return (error);
2052 }
2053
2054 /*
2055  * Return the IP multicast options in response to user getsockopt().
2056  */
2057 static int
2058 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2059 {
2060         struct in_addr addr;
2061         struct in_ifaddr *ia;
2062         int error, optval;
2063         u_char coptval;
2064
2065         error = 0;
2066         switch (sopt->sopt_name) {
2067         case IP_MULTICAST_VIF:
2068                 if (imo != NULL)
2069                         optval = imo->imo_multicast_vif;
2070                 else
2071                         optval = -1;
2072                 error = sooptcopyout(sopt, &optval, sizeof optval);
2073                 break;
2074
2075         case IP_MULTICAST_IF:
2076                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2077                         addr.s_addr = INADDR_ANY;
2078                 else if (imo->imo_multicast_addr.s_addr) {
2079                         /* return the value user has set */
2080                         addr = imo->imo_multicast_addr;
2081                 } else {
2082                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2083                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2084                                 : IA_SIN(ia)->sin_addr.s_addr;
2085                 }
2086                 error = sooptcopyout(sopt, &addr, sizeof addr);
2087                 break;
2088
2089         case IP_MULTICAST_TTL:
2090                 if (imo == NULL)
2091                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2092                 else
2093                         optval = coptval = imo->imo_multicast_ttl;
2094                 if (sopt->sopt_valsize == 1)
2095                         error = sooptcopyout(sopt, &coptval, 1);
2096                 else
2097                         error = sooptcopyout(sopt, &optval, sizeof optval);
2098                 break;
2099
2100         case IP_MULTICAST_LOOP:
2101                 if (imo == NULL)
2102                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2103                 else
2104                         optval = coptval = imo->imo_multicast_loop;
2105                 if (sopt->sopt_valsize == 1)
2106                         error = sooptcopyout(sopt, &coptval, 1);
2107                 else
2108                         error = sooptcopyout(sopt, &optval, sizeof optval);
2109                 break;
2110
2111         default:
2112                 error = ENOPROTOOPT;
2113                 break;
2114         }
2115         return (error);
2116 }
2117
2118 /*
2119  * Discard the IP multicast options.
2120  */
2121 void
2122 ip_freemoptions(struct ip_moptions *imo)
2123 {
2124         int i;
2125
2126         if (imo != NULL) {
2127                 for (i = 0; i < imo->imo_num_memberships; ++i)
2128                         in_delmulti(imo->imo_membership[i]);
2129                 free(imo, M_IPMOPTS);
2130         }
2131 }
2132
2133 /*
2134  * Routine called from ip_output() to loop back a copy of an IP multicast
2135  * packet to the input queue of a specified interface.  Note that this
2136  * calls the output routine of the loopback "driver", but with an interface
2137  * pointer that might NOT be a loopback interface -- evil, but easier than
2138  * replicating that code here.
2139  */
2140 static void
2141 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2142              int hlen)
2143 {
2144         struct ip *ip;
2145         struct mbuf *copym;
2146
2147         copym = m_copypacket(m, MB_DONTWAIT);
2148         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2149                 copym = m_pullup(copym, hlen);
2150         if (copym != NULL) {
2151                 /*
2152                  * if the checksum hasn't been computed, mark it as valid
2153                  */
2154                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2155                         in_delayed_cksum(copym);
2156                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2157                         copym->m_pkthdr.csum_flags |=
2158                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2159                         copym->m_pkthdr.csum_data = 0xffff;
2160                 }
2161                 /*
2162                  * We don't bother to fragment if the IP length is greater
2163                  * than the interface's MTU.  Can this possibly matter?
2164                  */
2165                 ip = mtod(copym, struct ip *);
2166                 ip->ip_len = htons(ip->ip_len);
2167                 ip->ip_off = htons(ip->ip_off);
2168                 ip->ip_sum = 0;
2169                 if (ip->ip_vhl == IP_VHL_BORING) {
2170                         ip->ip_sum = in_cksum_hdr(ip);
2171                 } else {
2172                         ip->ip_sum = in_cksum(copym, hlen);
2173                 }
2174                 /*
2175                  * NB:
2176                  * It's not clear whether there are any lingering
2177                  * reentrancy problems in other areas which might
2178                  * be exposed by using ip_input directly (in
2179                  * particular, everything which modifies the packet
2180                  * in-place).  Yet another option is using the
2181                  * protosw directly to deliver the looped back
2182                  * packet.  For the moment, we'll err on the side
2183                  * of safety by using if_simloop().
2184                  */
2185 #if 1 /* XXX */
2186                 if (dst->sin_family != AF_INET) {
2187                         printf("ip_mloopback: bad address family %d\n",
2188                                                 dst->sin_family);
2189                         dst->sin_family = AF_INET;
2190                 }
2191 #endif
2192
2193 #ifdef notdef
2194                 copym->m_pkthdr.rcvif = ifp;
2195                 ip_input(copym);
2196 #else
2197                 if_simloop(ifp, copym, dst->sin_family, 0);
2198 #endif
2199         }
2200 }