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. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  34  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  35  * $DragonFly: src/sys/netinet/ip_output.c,v 1.17 2004/07/26 06:32:58 asmodai Exp $
  36  */
  37
  38 #define _IP_VHL
  39
  40 #include "opt_ipfw.h"
  41 #include "opt_ipdn.h"
  42 #include "opt_ipdivert.h"
  43 #include "opt_ipfilter.h"
  44 #include "opt_ipsec.h"
  45 #include "opt_random_ip_id.h"
  46 #include "opt_mbuf_stress_test.h"
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/kernel.h>
  51 #include <sys/malloc.h>
  52 #include <sys/mbuf.h>
  53 #include <sys/protosw.h>
  54 #include <sys/socket.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/proc.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/in_cksum.h>
  59
  60 #include <net/if.h>
  61 #include <net/netisr.h>
  62 #include <net/pfil.h>
  63 #include <net/route.h>
  64
  65 #include <netinet/in.h>
  66 #include <netinet/in_systm.h>
  67 #include <netinet/ip.h>
  68 #include <netinet/in_pcb.h>
  69 #include <netinet/in_var.h>
  70 #include <netinet/ip_var.h>
  71
  72 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  73
  74 #ifdef IPSEC
  75 #include <netinet6/ipsec.h>
  76 #include <netproto/key/key.h>
  77 #ifdef IPSEC_DEBUG
  78 #include <netproto/key/key_debug.h>
  79 #else
  80 #define KEYDEBUG(lev,arg)
  81 #endif
  82 #endif /*IPSEC*/
  83
  84 #ifdef FAST_IPSEC
  85 #include <netipsec/ipsec.h>
  86 #include <netipsec/xform.h>
  87 #include <netipsec/key.h>
  88 #endif /*FAST_IPSEC*/
  89
  90 #include <net/ipfw/ip_fw.h>
  91 #include <net/dummynet/ip_dummynet.h>
  92
  93 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
  94                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  95                                   (ntohl(a.s_addr)>>16)&0xFF,\
  96                                   (ntohl(a.s_addr)>>8)&0xFF,\
  97                                   (ntohl(a.s_addr))&0xFF, y);
  98
  99 u_short ip_id;
 100
 101 #ifdef MBUF_STRESS_TEST
 102 int mbuf_frag_size = 0;
 103 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 104         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 105 #endif
 106
 107 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 108 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 109 static void     ip_mloopback
 110         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 111 static int      ip_getmoptions
 112         (struct sockopt *, struct ip_moptions *);
 113 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 114 static int      ip_setmoptions
 115         (struct sockopt *, struct ip_moptions **);
 116
 117 int     ip_optcopy(struct ip *, struct ip *);
 118
 119
 120 extern  struct protosw inetsw[];
 121
 122 /*
 123  * IP output.  The packet in mbuf chain m contains a skeletal IP
 124  * header (with len, off, ttl, proto, tos, src, dst).
 125  * The mbuf chain containing the packet will be freed.
 126  * The mbuf opt, if present, will not be freed.
 127  */
 128 int
 129 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 130         int flags, struct ip_moptions *imo, struct inpcb *inp)
 131 {
 132         struct ip *ip;
 133         struct ifnet *ifp = NULL;       /* keep compiler happy */
 134         struct mbuf *m;
 135         int hlen = sizeof (struct ip);
 136         int len, off, error = 0;
 137         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 138         struct in_ifaddr *ia = NULL;
 139         int isbroadcast, sw_csum;
 140         struct in_addr pkt_dst;
 141 #ifdef IPSEC
 142         struct route iproute;
 143         struct secpolicy *sp = NULL;
 144         struct socket *so = inp ? inp->inp_socket : NULL;
 145 #endif
 146 #ifdef FAST_IPSEC
 147         struct route iproute;
 148         struct m_tag *mtag;
 149         struct secpolicy *sp = NULL;
 150         struct tdb_ident *tdbi;
 151         int s;
 152 #endif /* FAST_IPSEC */
 153         struct ip_fw_args args;
 154         int src_was_INADDR_ANY = 0;     /* as the name says... */
 155
 156         args.eh = NULL;
 157         args.rule = NULL;
 158         args.next_hop = NULL;
 159         args.divert_rule = 0;                   /* divert cookie */
 160
 161         /* Grab info from MT_TAG mbufs prepended to the chain. */
 162         for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
 163                 switch(m0->_m_tag_id) {
 164                 default:
 165                         printf("ip_output: unrecognised MT_TAG tag %d\n",
 166                             m0->_m_tag_id);
 167                         break;
 168
 169                 case PACKET_TAG_DUMMYNET:
 170                         /*
 171                          * the packet was already tagged, so part of the
 172                          * processing was already done, and we need to go down.
 173                          * Get parameters from the header.
 174                          */
 175                         args.rule = ((struct dn_pkt *)m0)->rule;
 176                         opt = NULL ;
 177                         ro = & ( ((struct dn_pkt *)m0)->ro ) ;
 178                         imo = NULL ;
 179                         dst = ((struct dn_pkt *)m0)->dn_dst ;
 180                         ifp = ((struct dn_pkt *)m0)->ifp ;
 181                         flags = ((struct dn_pkt *)m0)->flags ;
 182                         break;
 183
 184                 case PACKET_TAG_DIVERT:
 185                         args.divert_rule = (int)m0->m_data & 0xffff;
 186                         break;
 187
 188                 case PACKET_TAG_IPFORWARD:
 189                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 190                         break;
 191                 }
 192         }
 193         m = m0;
 194
 195         KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
 196 #ifndef FAST_IPSEC
 197         KASSERT(ro != NULL, ("ip_output: no route, proto %d",
 198             mtod(m, struct ip *)->ip_p));
 199 #endif
 200
 201         if (args.rule != NULL) {        /* dummynet already saw us */
 202                 ip = mtod(m, struct ip *);
 203                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 204                 if (ro->ro_rt)
 205                         ia = ifatoia(ro->ro_rt->rt_ifa);
 206                 goto sendit;
 207         }
 208
 209         if (opt) {
 210                 len = 0;
 211                 m = ip_insertoptions(m, opt, &len);
 212                 if (len != 0)
 213                         hlen = len;
 214         }
 215         ip = mtod(m, struct ip *);
 216         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 217
 218         /*
 219          * Fill in IP header.
 220          */
 221         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 222                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 223                 ip->ip_off &= IP_DF;
 224 #ifdef RANDOM_IP_ID
 225                 ip->ip_id = ip_randomid();
 226 #else
 227                 ip->ip_id = htons(ip_id++);
 228 #endif
 229                 ipstat.ips_localout++;
 230         } else {
 231                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 232         }
 233
 234 #ifdef FAST_IPSEC
 235         if (ro == NULL) {
 236                 ro = &iproute;
 237                 bzero(ro, sizeof (*ro));
 238         }
 239 #endif /* FAST_IPSEC */
 240         dst = (struct sockaddr_in *)&ro->ro_dst;
 241         /*
 242          * If there is a cached route,
 243          * check that it is to the same destination
 244          * and is still up.  If not, free it and try again.
 245          * The address family should also be checked in case of sharing the
 246          * cache with IPv6.
 247          */
 248         if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
 249                           dst->sin_family != AF_INET ||
 250                           dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 251                 RTFREE(ro->ro_rt);
 252                 ro->ro_rt = (struct rtentry *)0;
 253         }
 254         if (ro->ro_rt == 0) {
 255                 bzero(dst, sizeof(*dst));
 256                 dst->sin_family = AF_INET;
 257                 dst->sin_len = sizeof(*dst);
 258                 dst->sin_addr = pkt_dst;
 259         }
 260         /*
 261          * If routing to interface only,
 262          * short circuit routing lookup.
 263          */
 264         if (flags & IP_ROUTETOIF) {
 265                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
 266                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
 267                         ipstat.ips_noroute++;
 268                         error = ENETUNREACH;
 269                         goto bad;
 270                 }
 271                 ifp = ia->ia_ifp;
 272                 ip->ip_ttl = 1;
 273                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 274         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 275             imo != NULL && imo->imo_multicast_ifp != NULL) {
 276                 /*
 277                  * Bypass the normal routing lookup for multicast
 278                  * packets if the interface is specified.
 279                  */
 280                 ifp = imo->imo_multicast_ifp;
 281                 IFP_TO_IA(ifp, ia);
 282                 isbroadcast = 0;        /* fool gcc */
 283         } else {
 284                 /*
 285                  * If this is the case, we probably don't want to allocate
 286                  * a protocol-cloned route since we didn't get one from the
 287                  * ULP.  This lets TCP do its thing, while not burdening
 288                  * forwarding or ICMP with the overhead of cloning a route.
 289                  * Of course, we still want to do any cloning requested by
 290                  * the link layer, as this is probably required in all cases
 291                  * for correct operation (as it is for ARP).
 292                  */
 293                 if (ro->ro_rt == 0)
 294                         rtalloc_ign(ro, RTF_PRCLONING);
 295                 if (ro->ro_rt == 0) {
 296                         ipstat.ips_noroute++;
 297                         error = EHOSTUNREACH;
 298                         goto bad;
 299                 }
 300                 ia = ifatoia(ro->ro_rt->rt_ifa);
 301                 ifp = ro->ro_rt->rt_ifp;
 302                 ro->ro_rt->rt_use++;
 303                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 304                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 305                 if (ro->ro_rt->rt_flags & RTF_HOST)
 306                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 307                 else
 308                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 309         }
 310         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 311                 struct in_multi *inm;
 312
 313                 m->m_flags |= M_MCAST;
 314                 /*
 315                  * IP destination address is multicast.  Make sure "dst"
 316                  * still points to the address in "ro".  (It may have been
 317                  * changed to point to a gateway address, above.)
 318                  */
 319                 dst = (struct sockaddr_in *)&ro->ro_dst;
 320                 /*
 321                  * See if the caller provided any multicast options
 322                  */
 323                 if (imo != NULL) {
 324                         ip->ip_ttl = imo->imo_multicast_ttl;
 325                         if (imo->imo_multicast_vif != -1)
 326                                 ip->ip_src.s_addr =
 327                                     ip_mcast_src ?
 328                                     ip_mcast_src(imo->imo_multicast_vif) :
 329                                     INADDR_ANY;
 330                 } else
 331                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 332                 /*
 333                  * Confirm that the outgoing interface supports multicast.
 334                  */
 335                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 336                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 337                                 ipstat.ips_noroute++;
 338                                 error = ENETUNREACH;
 339                                 goto bad;
 340                         }
 341                 }
 342                 /*
 343                  * If source address not specified yet, use address
 344                  * of outgoing interface.
 345                  */
 346                 if (ip->ip_src.s_addr == INADDR_ANY) {
 347                         /* Interface may have no addresses. */
 348                         if (ia != NULL)
 349                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 350                 }
 351
 352                 if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 353                         /*
 354                          * XXX
 355                          * delayed checksums are not currently
 356                          * compatible with IP multicast routing
 357                          */
 358                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 359                                 in_delayed_cksum(m);
 360                                 m->m_pkthdr.csum_flags &=
 361                                         ~CSUM_DELAY_DATA;
 362                         }
 363                 }
 364                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 365                 if (inm != NULL &&
 366                    (imo == NULL || imo->imo_multicast_loop)) {
 367                         /*
 368                          * If we belong to the destination multicast group
 369                          * on the outgoing interface, and the caller did not
 370                          * forbid loopback, loop back a copy.
 371                          */
 372                         ip_mloopback(ifp, m, dst, hlen);
 373                 }
 374                 else {
 375                         /*
 376                          * If we are acting as a multicast router, perform
 377                          * multicast forwarding as if the packet had just
 378                          * arrived on the interface to which we are about
 379                          * to send.  The multicast forwarding function
 380                          * recursively calls this function, using the
 381                          * IP_FORWARDING flag to prevent infinite recursion.
 382                          *
 383                          * Multicasts that are looped back by ip_mloopback(),
 384                          * above, will be forwarded by the ip_input() routine,
 385                          * if necessary.
 386                          */
 387                         if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 388                                 /*
 389                                  * If rsvp daemon is not running, do not
 390                                  * set ip_moptions. This ensures that the packet
 391                                  * is multicast and not just sent down one link
 392                                  * as prescribed by rsvpd.
 393                                  */
 394                                 if (!rsvp_on)
 395                                         imo = NULL;
 396                                 if (ip_mforward &&
 397                                     ip_mforward(ip, ifp, m, imo) != 0) {
 398                                         m_freem(m);
 399                                         goto done;
 400                                 }
 401                         }
 402                 }
 403
 404                 /*
 405                  * Multicasts with a time-to-live of zero may be looped-
 406                  * back, above, but must not be transmitted on a network.
 407                  * Also, multicasts addressed to the loopback interface
 408                  * are not sent -- the above call to ip_mloopback() will
 409                  * loop back a copy if this host actually belongs to the
 410                  * destination group on the loopback interface.
 411                  */
 412                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 413                         m_freem(m);
 414                         goto done;
 415                 }
 416
 417                 goto sendit;
 418         }
 419 #ifndef notdef
 420         /*
 421          * If the source address is not specified yet, use the address
 422          * of the outoing interface. In case, keep note we did that, so
 423          * if the the firewall changes the next-hop causing the output
 424          * interface to change, we can fix that.
 425          */
 426         if (ip->ip_src.s_addr == INADDR_ANY) {
 427                 /* Interface may have no addresses. */
 428                 if (ia != NULL) {
 429                         ip->ip_src = IA_SIN(ia)->sin_addr;
 430                         src_was_INADDR_ANY = 1;
 431                 }
 432         }
 433 #endif /* notdef */
 434         /*
 435          * Verify that we have any chance at all of being able to queue
 436          *      the packet or packet fragments
 437          */
 438         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 439                 ifp->if_snd.ifq_maxlen) {
 440                         error = ENOBUFS;
 441                         ipstat.ips_odropped++;
 442                         goto bad;
 443         }
 444
 445         /*
 446          * Look for broadcast address and
 447          * verify user is allowed to send
 448          * such a packet.
 449          */
 450         if (isbroadcast) {
 451                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 452                         error = EADDRNOTAVAIL;
 453                         goto bad;
 454                 }
 455                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 456                         error = EACCES;
 457                         goto bad;
 458                 }
 459                 /* don't allow broadcast messages to be fragmented */
 460                 if (ip->ip_len > ifp->if_mtu) {
 461                         error = EMSGSIZE;
 462                         goto bad;
 463                 }
 464                 m->m_flags |= M_BCAST;
 465         } else {
 466                 m->m_flags &= ~M_BCAST;
 467         }
 468
 469 sendit:
 470 #ifdef IPSEC
 471         /* get SP for this packet */
 472         if (so == NULL)
 473                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 474         else
 475                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 476
 477         if (sp == NULL) {
 478                 ipsecstat.out_inval++;
 479                 goto bad;
 480         }
 481
 482         error = 0;
 483
 484         /* check policy */
 485         switch (sp->policy) {
 486         case IPSEC_POLICY_DISCARD:
 487                 /*
 488                  * This packet is just discarded.
 489                  */
 490                 ipsecstat.out_polvio++;
 491                 goto bad;
 492
 493         case IPSEC_POLICY_BYPASS:
 494         case IPSEC_POLICY_NONE:
 495                 /* no need to do IPsec. */
 496                 goto skip_ipsec;
 497
 498         case IPSEC_POLICY_IPSEC:
 499                 if (sp->req == NULL) {
 500                         /* acquire a policy */
 501                         error = key_spdacquire(sp);
 502                         goto bad;
 503                 }
 504                 break;
 505
 506         case IPSEC_POLICY_ENTRUST:
 507         default:
 508                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
 509         }
 510     {
 511         struct ipsec_output_state state;
 512         bzero(&state, sizeof(state));
 513         state.m = m;
 514         if (flags & IP_ROUTETOIF) {
 515                 state.ro = &iproute;
 516                 bzero(&iproute, sizeof(iproute));
 517         } else
 518                 state.ro = ro;
 519         state.dst = (struct sockaddr *)dst;
 520
 521         ip->ip_sum = 0;
 522
 523         /*
 524          * XXX
 525          * delayed checksums are not currently compatible with IPsec
 526          */
 527         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 528                 in_delayed_cksum(m);
 529                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 530         }
 531
 532         ip->ip_len = htons(ip->ip_len);
 533         ip->ip_off = htons(ip->ip_off);
 534
 535         error = ipsec4_output(&state, sp, flags);
 536
 537         m = state.m;
 538         if (flags & IP_ROUTETOIF) {
 539                 /*
 540                  * if we have tunnel mode SA, we may need to ignore
 541                  * IP_ROUTETOIF.
 542                  */
 543                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 544                         flags &= ~IP_ROUTETOIF;
 545                         ro = state.ro;
 546                 }
 547         } else
 548                 ro = state.ro;
 549         dst = (struct sockaddr_in *)state.dst;
 550         if (error) {
 551                 /* mbuf is already reclaimed in ipsec4_output. */
 552                 m0 = NULL;
 553                 switch (error) {
 554                 case EHOSTUNREACH:
 555                 case ENETUNREACH:
 556                 case EMSGSIZE:
 557                 case ENOBUFS:
 558                 case ENOMEM:
 559                         break;
 560                 default:
 561                         printf("ip4_output (ipsec): error code %d\n", error);
 562                         /*fall through*/
 563                 case ENOENT:
 564                         /* don't show these error codes to the user */
 565                         error = 0;
 566                         break;
 567                 }
 568                 goto bad;
 569         }
 570     }
 571
 572         /* be sure to update variables that are affected by ipsec4_output() */
 573         ip = mtod(m, struct ip *);
 574 #ifdef _IP_VHL
 575         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 576 #else
 577         hlen = ip->ip_hl << 2;
 578 #endif
 579         if (ro->ro_rt == NULL) {
 580                 if ((flags & IP_ROUTETOIF) == 0) {
 581                         printf("ip_output: "
 582                                 "can't update route after IPsec processing\n");
 583                         error = EHOSTUNREACH;   /*XXX*/
 584                         goto bad;
 585                 }
 586         } else {
 587                 ia = ifatoia(ro->ro_rt->rt_ifa);
 588                 ifp = ro->ro_rt->rt_ifp;
 589         }
 590
 591         /* make it flipped, again. */
 592         ip->ip_len = ntohs(ip->ip_len);
 593         ip->ip_off = ntohs(ip->ip_off);
 594 skip_ipsec:
 595 #endif /*IPSEC*/
 596 #ifdef FAST_IPSEC
 597         /*
 598          * Check the security policy (SP) for the packet and, if
 599          * required, do IPsec-related processing.  There are two
 600          * cases here; the first time a packet is sent through
 601          * it will be untagged and handled by ipsec4_checkpolicy.
 602          * If the packet is resubmitted to ip_output (e.g. after
 603          * AH, ESP, etc. processing), there will be a tag to bypass
 604          * the lookup and related policy checking.
 605          */
 606         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 607         s = splnet();
 608         if (mtag != NULL) {
 609                 tdbi = (struct tdb_ident *)(mtag + 1);
 610                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 611                 if (sp == NULL)
 612                         error = -EINVAL;        /* force silent drop */
 613                 m_tag_delete(m, mtag);
 614         } else {
 615                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 616                                         &error, inp);
 617         }
 618         /*
 619          * There are four return cases:
 620          *    sp != NULL                    apply IPsec policy
 621          *    sp == NULL, error == 0        no IPsec handling needed
 622          *    sp == NULL, error == -EINVAL  discard packet w/o error
 623          *    sp == NULL, error != 0        discard packet, report error
 624          */
 625         if (sp != NULL) {
 626                 /* Loop detection, check if ipsec processing already done */
 627                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 628                 for (mtag = m_tag_first(m); mtag != NULL;
 629                      mtag = m_tag_next(m, mtag)) {
 630                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 631                                 continue;
 632                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 633                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 634                                 continue;
 635                         /*
 636                          * Check if policy has an SA associated with it.
 637                          * This can happen when an SP has yet to acquire
 638                          * an SA; e.g. on first reference.  If it occurs,
 639                          * then we let ipsec4_process_packet do its thing.
 640                          */
 641                         if (sp->req->sav == NULL)
 642                                 break;
 643                         tdbi = (struct tdb_ident *)(mtag + 1);
 644                         if (tdbi->spi == sp->req->sav->spi &&
 645                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 646                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 647                                  sizeof (union sockaddr_union)) == 0) {
 648                                 /*
 649                                  * No IPsec processing is needed, free
 650                                  * reference to SP.
 651                                  *
 652                                  * NB: null pointer to avoid free at
 653                                  *     done: below.
 654                                  */
 655                                 KEY_FREESP(&sp), sp = NULL;
 656                                 splx(s);
 657                                 goto spd_done;
 658                         }
 659                 }
 660
 661                 /*
 662                  * Do delayed checksums now because we send before
 663                  * this is done in the normal processing path.
 664                  */
 665                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 666                         in_delayed_cksum(m);
 667                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 668                 }
 669
 670                 ip->ip_len = htons(ip->ip_len);
 671                 ip->ip_off = htons(ip->ip_off);
 672
 673                 /* NB: callee frees mbuf */
 674                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 675                 /*
 676                  * Preserve KAME behaviour: ENOENT can be returned
 677                  * when an SA acquire is in progress.  Don't propagate
 678                  * this to user-level; it confuses applications.
 679                  *
 680                  * XXX this will go away when the SADB is redone.
 681                  */
 682                 if (error == ENOENT)
 683                         error = 0;
 684                 splx(s);
 685                 goto done;
 686         } else {
 687                 splx(s);
 688
 689                 if (error != 0) {
 690                         /*
 691                          * Hack: -EINVAL is used to signal that a packet
 692                          * should be silently discarded.  This is typically
 693                          * because we asked key management for an SA and
 694                          * it was delayed (e.g. kicked up to IKE).
 695                          */
 696                         if (error == -EINVAL)
 697                                 error = 0;
 698                         goto bad;
 699                 } else {
 700                         /* No IPsec processing for this packet. */
 701                 }
 702 #ifdef notyet
 703                 /*
 704                  * If deferred crypto processing is needed, check that
 705                  * the interface supports it.
 706                  */
 707                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 708                 if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
 709                         /* notify IPsec to do its own crypto */
 710                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
 711                         error = EHOSTUNREACH;
 712                         goto bad;
 713                 }
 714 #endif
 715         }
 716 spd_done:
 717 #endif /* FAST_IPSEC */
 718         /*
 719          * IpHack's section.
 720          * - Xlate: translate packet's addr/port (NAT).
 721          * - Firewall: deny/allow/etc.
 722          * - Wrap: fake packet's addr/port <unimpl.>
 723          * - Encapsulate: put it in another IP and send out. <unimp.>
 724          */
 725
 726         /*
 727          * Run through list of hooks for output packets.
 728          */
 729         if (pfil_has_hooks(&inet_pfil_hook)) {
 730                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 731                 if (error != 0 || m == NULL)
 732                         goto done;
 733                 ip = mtod(m, struct ip *);
 734         }
 735
 736         /*
 737          * Check with the firewall...
 738          * but not if we are already being fwd'd from a firewall.
 739          */
 740         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 741                 struct sockaddr_in *old = dst;
 742
 743                 args.m = m;
 744                 args.next_hop = dst;
 745                 args.oif = ifp;
 746                 off = ip_fw_chk_ptr(&args);
 747                 m = args.m;
 748                 dst = args.next_hop;
 749
 750                 /*
 751                  * On return we must do the following:
 752                  * m == NULL    -> drop the pkt (old interface, deprecated)
 753                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 754                  * 1<=off<= 0xffff              -> DIVERT
 755                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 756                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 757                  * dst != old                   -> IPFIREWALL_FORWARD
 758                  * off==0, dst==old             -> accept
 759                  * If some of the above modules are not compiled in, then
 760                  * we should't have to check the corresponding condition
 761                  * (because the ipfw control socket should not accept
 762                  * unsupported rules), but better play safe and drop
 763                  * packets in case of doubt.
 764                  */
 765                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 766                         if (m)
 767                                 m_freem(m);
 768                         error = EACCES;
 769                         goto done;
 770                 }
 771                 ip = mtod(m, struct ip *);
 772                 if (off == 0 && dst == old)             /* common case */
 773                         goto pass;
 774                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
 775                         /*
 776                          * pass the pkt to dummynet. Need to include
 777                          * pipe number, m, ifp, ro, dst because these are
 778                          * not recomputed in the next pass.
 779                          * All other parameters have been already used and
 780                          * so they are not needed anymore.
 781                          * XXX note: if the ifp or ro entry are deleted
 782                          * while a pkt is in dummynet, we are in trouble!
 783                          */
 784                         args.ro = ro;
 785                         args.dst = dst;
 786                         args.flags = flags;
 787
 788                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
 789                                 &args);
 790                         goto done;
 791                 }
 792 #ifdef IPDIVERT
 793                 if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
 794                         struct mbuf *clone = NULL;
 795
 796                         /* Clone packet if we're doing a 'tee' */
 797                         if ((off & IP_FW_PORT_TEE_FLAG) != 0)
 798                                 clone = m_dup(m, MB_DONTWAIT);
 799
 800                         /*
 801                          * XXX
 802                          * delayed checksums are not currently compatible
 803                          * with divert sockets.
 804                          */
 805                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 806                                 in_delayed_cksum(m);
 807                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 808                         }
 809
 810                         /* Restore packet header fields to original values */
 811                         ip->ip_len = htons(ip->ip_len);
 812                         ip->ip_off = htons(ip->ip_off);
 813
 814                         /* Deliver packet to divert input routine */
 815                         divert_packet(m, 0, off & 0xffff, args.divert_rule);
 816
 817                         /* If 'tee', continue with original packet */
 818                         if (clone != NULL) {
 819                                 m = clone;
 820                                 ip = mtod(m, struct ip *);
 821                                 goto pass;
 822                         }
 823                         goto done;
 824                 }
 825 #endif
 826
 827                 /* IPFIREWALL_FORWARD */
 828                 /*
 829                  * Check dst to make sure it is directly reachable on the
 830                  * interface we previously thought it was.
 831                  * If it isn't (which may be likely in some situations) we have
 832                  * to re-route it (ie, find a route for the next-hop and the
 833                  * associated interface) and set them here. This is nested
 834                  * forwarding which in most cases is undesirable, except where
 835                  * such control is nigh impossible. So we do it here.
 836                  * And I'm babbling.
 837                  */
 838                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 839 #if 0
 840                         /*
 841                          * XXX To improve readability, this block should be
 842                          * changed into a function call as below:
 843                          */
 844                         error = ip_ipforward(&m, &dst, &ifp);
 845                         if (error)
 846                                 goto bad;
 847                         if (m == NULL) /* ip_input consumed the mbuf */
 848                                 goto done;
 849 #else
 850                         struct in_ifaddr *ia;
 851
 852                         /*
 853                          * XXX sro_fwd below is static, and a pointer
 854                          * to it gets passed to routines downstream.
 855                          * This could have surprisingly bad results in
 856                          * practice, because its content is overwritten
 857                          * by subsequent packets.
 858                          */
 859                         /* There must be a better way to do this next line... */
 860                         static struct route sro_fwd;
 861                         struct route *ro_fwd = &sro_fwd;
 862
 863 #if 0
 864                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 865                             dst->sin_addr, "\n");
 866 #endif
 867
 868                         /*
 869                          * We need to figure out if we have been forwarded
 870                          * to a local socket. If so, then we should somehow
 871                          * "loop back" to ip_input, and get directed to the
 872                          * PCB as if we had received this packet. This is
 873                          * because it may be dificult to identify the packets
 874                          * you want to forward until they are being output
 875                          * and have selected an interface. (e.g. locally
 876                          * initiated packets) If we used the loopback inteface,
 877                          * we would not be able to control what happens
 878                          * as the packet runs through ip_input() as
 879                          * it is done through a ISR.
 880                          */
 881                         LIST_FOREACH(ia,
 882                             INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 883                                 /*
 884                                  * If the addr to forward to is one
 885                                  * of ours, we pretend to
 886                                  * be the destination for this packet.
 887                                  */
 888                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 889                                                  dst->sin_addr.s_addr)
 890                                         break;
 891                         }
 892                         if (ia) {       /* tell ip_input "dont filter" */
 893                                 struct m_hdr tag;
 894
 895                                 tag.mh_type = MT_TAG;
 896                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 897                                 tag.mh_data = (caddr_t)args.next_hop;
 898                                 tag.mh_next = m;
 899
 900                                 if (m->m_pkthdr.rcvif == NULL)
 901                                         m->m_pkthdr.rcvif = ifunit("lo0");
 902                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 903                                         m->m_pkthdr.csum_flags |=
 904                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 905                                         m0->m_pkthdr.csum_data = 0xffff;
 906                                 }
 907                                 m->m_pkthdr.csum_flags |=
 908                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 909                                 ip->ip_len = htons(ip->ip_len);
 910                                 ip->ip_off = htons(ip->ip_off);
 911                                 ip_input((struct mbuf *)&tag);
 912                                 goto done;
 913                         }
 914                         /* Some of the logic for this was
 915                          * nicked from above.
 916                          *
 917                          * This rewrites the cached route in a local PCB.
 918                          * Is this what we want to do?
 919                          */
 920                         bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
 921
 922                         ro_fwd->ro_rt = 0;
 923                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 924
 925                         if (ro_fwd->ro_rt == 0) {
 926                                 ipstat.ips_noroute++;
 927                                 error = EHOSTUNREACH;
 928                                 goto bad;
 929                         }
 930
 931                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 932                         ifp = ro_fwd->ro_rt->rt_ifp;
 933                         ro_fwd->ro_rt->rt_use++;
 934                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 935                                 dst = (struct sockaddr_in *)
 936                                         ro_fwd->ro_rt->rt_gateway;
 937                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 938                                 isbroadcast =
 939                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 940                         else
 941                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 942                         if (ro->ro_rt)
 943                                 RTFREE(ro->ro_rt);
 944                         ro->ro_rt = ro_fwd->ro_rt;
 945                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 946
 947 #endif  /* ... block to be put into a function */
 948                         /*
 949                          * If we added a default src ip earlier,
 950                          * which would have been gotten from the-then
 951                          * interface, do it again, from the new one.
 952                          */
 953                         if (src_was_INADDR_ANY)
 954                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 955                         goto pass ;
 956                 }
 957
 958                 /*
 959                  * if we get here, none of the above matches, and
 960                  * we have to drop the pkt
 961                  */
 962                 m_freem(m);
 963                 error = EACCES; /* not sure this is the right error msg */
 964                 goto done;
 965         }
 966
 967 pass:
 968         /* 127/8 must not appear on wire - RFC1122. */
 969         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 970             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 971                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 972                         ipstat.ips_badaddr++;
 973                         error = EADDRNOTAVAIL;
 974                         goto bad;
 975                 }
 976         }
 977
 978         m->m_pkthdr.csum_flags |= CSUM_IP;
 979         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 980         if (sw_csum & CSUM_DELAY_DATA) {
 981                 in_delayed_cksum(m);
 982                 sw_csum &= ~CSUM_DELAY_DATA;
 983         }
 984         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 985
 986         /*
 987          * If small enough for interface, or the interface will take
 988          * care of the fragmentation for us, can just send directly.
 989          */
 990         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 991             (ip->ip_off & IP_DF) == 0)) {
 992                 ip->ip_len = htons(ip->ip_len);
 993                 ip->ip_off = htons(ip->ip_off);
 994                 ip->ip_sum = 0;
 995                 if (sw_csum & CSUM_DELAY_IP) {
 996                         if (ip->ip_vhl == IP_VHL_BORING) {
 997                                 ip->ip_sum = in_cksum_hdr(ip);
 998                         } else {
 999                                 ip->ip_sum = in_cksum(m, hlen);
1000                         }
1001                 }
1002
1003                 /* Record statistics for this interface address. */
1004                 if (!(flags & IP_FORWARDING) && ia) {
1005                         ia->ia_ifa.if_opackets++;
1006                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1007                 }
1008
1009 #ifdef IPSEC
1010                 /* clean ipsec history once it goes out of the node */
1011                 ipsec_delaux(m);
1012 #endif
1013
1014 #ifdef MBUF_STRESS_TEST
1015                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1016                         struct mbuf *m1, *m2;
1017                         int length, tmp;
1018
1019                         tmp = length = m->m_pkthdr.len;
1020
1021                         while ((length -= mbuf_frag_size) >= 1) {
1022                                 m1 = m_split(m, length, MB_DONTWAIT);
1023                                 if (m1 == NULL)
1024                                         break;
1025                                 m1->m_flags &= ~M_PKTHDR;
1026                                 m2 = m;
1027                                 while (m2->m_next != NULL)
1028                                         m2 = m2->m_next;
1029                                 m2->m_next = m1;
1030                         }
1031                         m->m_pkthdr.len = tmp;
1032                 }
1033 #endif
1034                 error = (*ifp->if_output)(ifp, m,
1035                                 (struct sockaddr *)dst, ro->ro_rt);
1036                 goto done;
1037         }
1038
1039         if (ip->ip_off & IP_DF) {
1040                 error = EMSGSIZE;
1041                 /*
1042                  * This case can happen if the user changed the MTU
1043                  * of an interface after enabling IP on it.  Because
1044                  * most netifs don't keep track of routes pointing to
1045                  * them, there is no way for one to update all its
1046                  * routes when the MTU is changed.
1047                  */
1048                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1049                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1050                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1051                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1052                 }
1053                 ipstat.ips_cantfrag++;
1054                 goto bad;
1055         }
1056
1057         /*
1058          * Too large for interface; fragment if possible. If successful,
1059          * on return, m will point to a list of packets to be sent.
1060          */
1061         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1062         if (error)
1063                 goto bad;
1064         for (; m; m = m0) {
1065                 m0 = m->m_nextpkt;
1066                 m->m_nextpkt = 0;
1067 #ifdef IPSEC
1068                 /* clean ipsec history once it goes out of the node */
1069                 ipsec_delaux(m);
1070 #endif
1071                 if (error == 0) {
1072                         /* Record statistics for this interface address. */
1073                         if (ia != NULL) {
1074                                 ia->ia_ifa.if_opackets++;
1075                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1076                         }
1077
1078                         error = (*ifp->if_output)(ifp, m,
1079                             (struct sockaddr *)dst, ro->ro_rt);
1080                 } else
1081                         m_freem(m);
1082         }
1083
1084         if (error == 0)
1085                 ipstat.ips_fragmented++;
1086
1087 done:
1088 #ifdef IPSEC
1089         if (ro == &iproute && ro->ro_rt) {
1090                 RTFREE(ro->ro_rt);
1091                 ro->ro_rt = NULL;
1092         }
1093         if (sp != NULL) {
1094                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1095                         printf("DP ip_output call free SP:%p\n", sp));
1096                 key_freesp(sp);
1097         }
1098 #endif
1099 #ifdef FAST_IPSEC
1100         if (ro == &iproute && ro->ro_rt) {
1101                 RTFREE(ro->ro_rt);
1102                 ro->ro_rt = NULL;
1103         }
1104         if (sp != NULL)
1105                 KEY_FREESP(&sp);
1106 #endif
1107         return (error);
1108 bad:
1109         m_freem(m);
1110         goto done;
1111 }
1112
1113 /*
1114  * Create a chain of fragments which fit the given mtu. m_frag points to the
1115  * mbuf to be fragmented; on return it points to the chain with the fragments.
1116  * Return 0 if no error. If error, m_frag may contain a partially built
1117  * chain of fragments that should be freed by the caller.
1118  *
1119  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1120  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1121  */
1122 int
1123 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1124             u_long if_hwassist_flags, int sw_csum)
1125 {
1126         int error = 0;
1127         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1128         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1129         int off;
1130         struct mbuf *m0 = *m_frag;      /* the original packet          */
1131         int firstlen;
1132         struct mbuf **mnext;
1133         int nfrags;
1134
1135         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1136                 ipstat.ips_cantfrag++;
1137                 return EMSGSIZE;
1138         }
1139
1140         /*
1141          * Must be able to put at least 8 bytes per fragment.
1142          */
1143         if (len < 8)
1144                 return EMSGSIZE;
1145
1146         /*
1147          * If the interface will not calculate checksums on
1148          * fragmented packets, then do it here.
1149          */
1150         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1151             (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1152                 in_delayed_cksum(m0);
1153                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1154         }
1155
1156         if (len > PAGE_SIZE) {
1157                 /*
1158                  * Fragment large datagrams such that each segment
1159                  * contains a multiple of PAGE_SIZE amount of data,
1160                  * plus headers. This enables a receiver to perform
1161                  * page-flipping zero-copy optimizations.
1162                  *
1163                  * XXX When does this help given that sender and receiver
1164                  * could have different page sizes, and also mtu could
1165                  * be less than the receiver's page size ?
1166                  */
1167                 int newlen;
1168                 struct mbuf *m;
1169
1170                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1171                         off += m->m_len;
1172
1173                 /*
1174                  * firstlen (off - hlen) must be aligned on an
1175                  * 8-byte boundary
1176                  */
1177                 if (off < hlen)
1178                         goto smart_frag_failure;
1179                 off = ((off - hlen) & ~7) + hlen;
1180                 newlen = (~PAGE_MASK) & mtu;
1181                 if ((newlen + sizeof (struct ip)) > mtu) {
1182                         /* we failed, go back the default */
1183 smart_frag_failure:
1184                         newlen = len;
1185                         off = hlen + len;
1186                 }
1187                 len = newlen;
1188
1189         } else {
1190                 off = hlen + len;
1191         }
1192
1193         firstlen = off - hlen;
1194         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1195
1196         /*
1197          * Loop through length of segment after first fragment,
1198          * make new header and copy data of each part and link onto chain.
1199          * Here, m0 is the original packet, m is the fragment being created.
1200          * The fragments are linked off the m_nextpkt of the original
1201          * packet, which after processing serves as the first fragment.
1202          */
1203         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1204                 struct ip *mhip;        /* ip header on the fragment */
1205                 struct mbuf *m;
1206                 int mhlen = sizeof (struct ip);
1207
1208                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1209                 if (m == 0) {
1210                         error = ENOBUFS;
1211                         ipstat.ips_odropped++;
1212                         goto done;
1213                 }
1214                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1215                 /*
1216                  * In the first mbuf, leave room for the link header, then
1217                  * copy the original IP header including options. The payload
1218                  * goes into an additional mbuf chain returned by m_copy().
1219                  */
1220                 m->m_data += max_linkhdr;
1221                 mhip = mtod(m, struct ip *);
1222                 *mhip = *ip;
1223                 if (hlen > sizeof (struct ip)) {
1224                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1225                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1226                 }
1227                 m->m_len = mhlen;
1228                 /* XXX do we need to add ip->ip_off below ? */
1229                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1230                 if (off + len >= ip->ip_len) {  /* last fragment */
1231                         len = ip->ip_len - off;
1232                         m->m_flags |= M_LASTFRAG;
1233                 } else
1234                         mhip->ip_off |= IP_MF;
1235                 mhip->ip_len = htons((u_short)(len + mhlen));
1236                 m->m_next = m_copy(m0, off, len);
1237                 if (m->m_next == 0) {           /* copy failed */
1238                         m_free(m);
1239                         error = ENOBUFS;        /* ??? */
1240                         ipstat.ips_odropped++;
1241                         goto done;
1242                 }
1243                 m->m_pkthdr.len = mhlen + len;
1244                 m->m_pkthdr.rcvif = (struct ifnet *)0;
1245                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1246                 mhip->ip_off = htons(mhip->ip_off);
1247                 mhip->ip_sum = 0;
1248                 if (sw_csum & CSUM_DELAY_IP)
1249                         mhip->ip_sum = in_cksum(m, mhlen);
1250                 *mnext = m;
1251                 mnext = &m->m_nextpkt;
1252         }
1253         ipstat.ips_ofragments += nfrags;
1254
1255         /* set first marker for fragment chain */
1256         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1257         m0->m_pkthdr.csum_data = nfrags;
1258
1259         /*
1260          * Update first fragment by trimming what's been copied out
1261          * and updating header.
1262          */
1263         m_adj(m0, hlen + firstlen - ip->ip_len);
1264         m0->m_pkthdr.len = hlen + firstlen;
1265         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1266         ip->ip_off |= IP_MF;
1267         ip->ip_off = htons(ip->ip_off);
1268         ip->ip_sum = 0;
1269         if (sw_csum & CSUM_DELAY_IP)
1270                 ip->ip_sum = in_cksum(m0, hlen);
1271
1272 done:
1273         *m_frag = m0;
1274         return error;
1275 }
1276
1277 void
1278 in_delayed_cksum(struct mbuf *m)
1279 {
1280         struct ip *ip;
1281         u_short csum, offset;
1282
1283         ip = mtod(m, struct ip *);
1284         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1285         csum = in_cksum_skip(m, ip->ip_len, offset);
1286         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1287                 csum = 0xffff;
1288         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1289
1290         if (offset + sizeof(u_short) > m->m_len) {
1291                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1292                     m->m_len, offset, ip->ip_p);
1293                 /*
1294                  * XXX
1295                  * this shouldn't happen, but if it does, the
1296                  * correct behavior may be to insert the checksum
1297                  * in the existing chain instead of rearranging it.
1298                  */
1299                 m = m_pullup(m, offset + sizeof(u_short));
1300         }
1301         *(u_short *)(m->m_data + offset) = csum;
1302 }
1303
1304 /*
1305  * Insert IP options into preformed packet.
1306  * Adjust IP destination as required for IP source routing,
1307  * as indicated by a non-zero in_addr at the start of the options.
1308  *
1309  * XXX This routine assumes that the packet has no options in place.
1310  */
1311 static struct mbuf *
1312 ip_insertoptions(m, opt, phlen)
1313         struct mbuf *m;
1314         struct mbuf *opt;
1315         int *phlen;
1316 {
1317         struct ipoption *p = mtod(opt, struct ipoption *);
1318         struct mbuf *n;
1319         struct ip *ip = mtod(m, struct ip *);
1320         unsigned optlen;
1321
1322         optlen = opt->m_len - sizeof(p->ipopt_dst);
1323         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1324                 *phlen = 0;
1325                 return (m);             /* XXX should fail */
1326         }
1327         if (p->ipopt_dst.s_addr)
1328                 ip->ip_dst = p->ipopt_dst;
1329         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1330                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1331                 if (n == 0) {
1332                         *phlen = 0;
1333                         return (m);
1334                 }
1335                 n->m_pkthdr.rcvif = (struct ifnet *)0;
1336                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1337                 m->m_len -= sizeof(struct ip);
1338                 m->m_data += sizeof(struct ip);
1339                 n->m_next = m;
1340                 m = n;
1341                 m->m_len = optlen + sizeof(struct ip);
1342                 m->m_data += max_linkhdr;
1343                 (void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1344         } else {
1345                 m->m_data -= optlen;
1346                 m->m_len += optlen;
1347                 m->m_pkthdr.len += optlen;
1348                 ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1349         }
1350         ip = mtod(m, struct ip *);
1351         bcopy(p->ipopt_list, ip + 1, optlen);
1352         *phlen = sizeof(struct ip) + optlen;
1353         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1354         ip->ip_len += optlen;
1355         return (m);
1356 }
1357
1358 /*
1359  * Copy options from ip to jp,
1360  * omitting those not copied during fragmentation.
1361  */
1362 int
1363 ip_optcopy(ip, jp)
1364         struct ip *ip, *jp;
1365 {
1366         u_char *cp, *dp;
1367         int opt, optlen, cnt;
1368
1369         cp = (u_char *)(ip + 1);
1370         dp = (u_char *)(jp + 1);
1371         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1372         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1373                 opt = cp[0];
1374                 if (opt == IPOPT_EOL)
1375                         break;
1376                 if (opt == IPOPT_NOP) {
1377                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1378                         *dp++ = IPOPT_NOP;
1379                         optlen = 1;
1380                         continue;
1381                 }
1382
1383                 KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1384                     ("ip_optcopy: malformed ipv4 option"));
1385                 optlen = cp[IPOPT_OLEN];
1386                 KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1387                     ("ip_optcopy: malformed ipv4 option"));
1388
1389                 /* bogus lengths should have been caught by ip_dooptions */
1390                 if (optlen > cnt)
1391                         optlen = cnt;
1392                 if (IPOPT_COPIED(opt)) {
1393                         bcopy(cp, dp, optlen);
1394                         dp += optlen;
1395                 }
1396         }
1397         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1398                 *dp++ = IPOPT_EOL;
1399         return (optlen);
1400 }
1401
1402 /*
1403  * IP socket option processing.
1404  */
1405 int
1406 ip_ctloutput(so, sopt)
1407         struct socket *so;
1408         struct sockopt *sopt;
1409 {
1410         struct  inpcb *inp = sotoinpcb(so);
1411         int     error, optval;
1412
1413         error = optval = 0;
1414         if (sopt->sopt_level != IPPROTO_IP) {
1415                 return (EINVAL);
1416         }
1417
1418         switch (sopt->sopt_dir) {
1419         case SOPT_SET:
1420                 switch (sopt->sopt_name) {
1421                 case IP_OPTIONS:
1422 #ifdef notyet
1423                 case IP_RETOPTS:
1424 #endif
1425                 {
1426                         struct mbuf *m;
1427                         if (sopt->sopt_valsize > MLEN) {
1428                                 error = EMSGSIZE;
1429                                 break;
1430                         }
1431                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1432                         if (m == 0) {
1433                                 error = ENOBUFS;
1434                                 break;
1435                         }
1436                         m->m_len = sopt->sopt_valsize;
1437                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1438                                             m->m_len);
1439
1440                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1441                                            m));
1442                 }
1443
1444                 case IP_TOS:
1445                 case IP_TTL:
1446                 case IP_RECVOPTS:
1447                 case IP_RECVRETOPTS:
1448                 case IP_RECVDSTADDR:
1449                 case IP_RECVIF:
1450                 case IP_FAITH:
1451                         error = sooptcopyin(sopt, &optval, sizeof optval,
1452                                             sizeof optval);
1453                         if (error)
1454                                 break;
1455
1456                         switch (sopt->sopt_name) {
1457                         case IP_TOS:
1458                                 inp->inp_ip_tos = optval;
1459                                 break;
1460
1461                         case IP_TTL:
1462                                 inp->inp_ip_ttl = optval;
1463                                 break;
1464 #define OPTSET(bit) \
1465         if (optval) \
1466                 inp->inp_flags |= bit; \
1467         else \
1468                 inp->inp_flags &= ~bit;
1469
1470                         case IP_RECVOPTS:
1471                                 OPTSET(INP_RECVOPTS);
1472                                 break;
1473
1474                         case IP_RECVRETOPTS:
1475                                 OPTSET(INP_RECVRETOPTS);
1476                                 break;
1477
1478                         case IP_RECVDSTADDR:
1479                                 OPTSET(INP_RECVDSTADDR);
1480                                 break;
1481
1482                         case IP_RECVIF:
1483                                 OPTSET(INP_RECVIF);
1484                                 break;
1485
1486                         case IP_FAITH:
1487                                 OPTSET(INP_FAITH);
1488                                 break;
1489                         }
1490                         break;
1491 #undef OPTSET
1492
1493                 case IP_MULTICAST_IF:
1494                 case IP_MULTICAST_VIF:
1495                 case IP_MULTICAST_TTL:
1496                 case IP_MULTICAST_LOOP:
1497                 case IP_ADD_MEMBERSHIP:
1498                 case IP_DROP_MEMBERSHIP:
1499                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1500                         break;
1501
1502                 case IP_PORTRANGE:
1503                         error = sooptcopyin(sopt, &optval, sizeof optval,
1504                                             sizeof optval);
1505                         if (error)
1506                                 break;
1507
1508                         switch (optval) {
1509                         case IP_PORTRANGE_DEFAULT:
1510                                 inp->inp_flags &= ~(INP_LOWPORT);
1511                                 inp->inp_flags &= ~(INP_HIGHPORT);
1512                                 break;
1513
1514                         case IP_PORTRANGE_HIGH:
1515                                 inp->inp_flags &= ~(INP_LOWPORT);
1516                                 inp->inp_flags |= INP_HIGHPORT;
1517                                 break;
1518
1519                         case IP_PORTRANGE_LOW:
1520                                 inp->inp_flags &= ~(INP_HIGHPORT);
1521                                 inp->inp_flags |= INP_LOWPORT;
1522                                 break;
1523
1524                         default:
1525                                 error = EINVAL;
1526                                 break;
1527                         }
1528                         break;
1529
1530 #if defined(IPSEC) || defined(FAST_IPSEC)
1531                 case IP_IPSEC_POLICY:
1532                 {
1533                         caddr_t req;
1534                         size_t len = 0;
1535                         int priv;
1536                         struct mbuf *m;
1537                         int optname;
1538
1539                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1540                                 break;
1541                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1542                                 break;
1543                         priv = (sopt->sopt_td != NULL &&
1544                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1545                         req = mtod(m, caddr_t);
1546                         len = m->m_len;
1547                         optname = sopt->sopt_name;
1548                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1549                         m_freem(m);
1550                         break;
1551                 }
1552 #endif /*IPSEC*/
1553
1554                 default:
1555                         error = ENOPROTOOPT;
1556                         break;
1557                 }
1558                 break;
1559
1560         case SOPT_GET:
1561                 switch (sopt->sopt_name) {
1562                 case IP_OPTIONS:
1563                 case IP_RETOPTS:
1564                         if (inp->inp_options)
1565                                 error = sooptcopyout(sopt,
1566                                                      mtod(inp->inp_options,
1567                                                           char *),
1568                                                      inp->inp_options->m_len);
1569                         else
1570                                 sopt->sopt_valsize = 0;
1571                         break;
1572
1573                 case IP_TOS:
1574                 case IP_TTL:
1575                 case IP_RECVOPTS:
1576                 case IP_RECVRETOPTS:
1577                 case IP_RECVDSTADDR:
1578                 case IP_RECVIF:
1579                 case IP_PORTRANGE:
1580                 case IP_FAITH:
1581                         switch (sopt->sopt_name) {
1582
1583                         case IP_TOS:
1584                                 optval = inp->inp_ip_tos;
1585                                 break;
1586
1587                         case IP_TTL:
1588                                 optval = inp->inp_ip_ttl;
1589                                 break;
1590
1591 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1592
1593                         case IP_RECVOPTS:
1594                                 optval = OPTBIT(INP_RECVOPTS);
1595                                 break;
1596
1597                         case IP_RECVRETOPTS:
1598                                 optval = OPTBIT(INP_RECVRETOPTS);
1599                                 break;
1600
1601                         case IP_RECVDSTADDR:
1602                                 optval = OPTBIT(INP_RECVDSTADDR);
1603                                 break;
1604
1605                         case IP_RECVIF:
1606                                 optval = OPTBIT(INP_RECVIF);
1607                                 break;
1608
1609                         case IP_PORTRANGE:
1610                                 if (inp->inp_flags & INP_HIGHPORT)
1611                                         optval = IP_PORTRANGE_HIGH;
1612                                 else if (inp->inp_flags & INP_LOWPORT)
1613                                         optval = IP_PORTRANGE_LOW;
1614                                 else
1615                                         optval = 0;
1616                                 break;
1617
1618                         case IP_FAITH:
1619                                 optval = OPTBIT(INP_FAITH);
1620                                 break;
1621                         }
1622                         error = sooptcopyout(sopt, &optval, sizeof optval);
1623                         break;
1624
1625                 case IP_MULTICAST_IF:
1626                 case IP_MULTICAST_VIF:
1627                 case IP_MULTICAST_TTL:
1628                 case IP_MULTICAST_LOOP:
1629                 case IP_ADD_MEMBERSHIP:
1630                 case IP_DROP_MEMBERSHIP:
1631                         error = ip_getmoptions(sopt, inp->inp_moptions);
1632                         break;
1633
1634 #if defined(IPSEC) || defined(FAST_IPSEC)
1635                 case IP_IPSEC_POLICY:
1636                 {
1637                         struct mbuf *m = NULL;
1638                         caddr_t req = NULL;
1639                         size_t len = 0;
1640
1641                         if (m != 0) {
1642                                 req = mtod(m, caddr_t);
1643                                 len = m->m_len;
1644                         }
1645                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1646                         if (error == 0)
1647                                 error = soopt_mcopyout(sopt, m); /* XXX */
1648                         if (error == 0)
1649                                 m_freem(m);
1650                         break;
1651                 }
1652 #endif /*IPSEC*/
1653
1654                 default:
1655                         error = ENOPROTOOPT;
1656                         break;
1657                 }
1658                 break;
1659         }
1660         return (error);
1661 }
1662
1663 /*
1664  * Set up IP options in pcb for insertion in output packets.
1665  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1666  * with destination address if source routed.
1667  */
1668 static int
1669 ip_pcbopts(optname, pcbopt, m)
1670         int optname;
1671         struct mbuf **pcbopt;
1672         struct mbuf *m;
1673 {
1674         int cnt, optlen;
1675         u_char *cp;
1676         u_char opt;
1677
1678         /* turn off any old options */
1679         if (*pcbopt)
1680                 (void)m_free(*pcbopt);
1681         *pcbopt = 0;
1682         if (m == (struct mbuf *)0 || m->m_len == 0) {
1683                 /*
1684                  * Only turning off any previous options.
1685                  */
1686                 if (m)
1687                         (void)m_free(m);
1688                 return (0);
1689         }
1690
1691         if (m->m_len % sizeof(int32_t))
1692                 goto bad;
1693         /*
1694          * IP first-hop destination address will be stored before
1695          * actual options; move other options back
1696          * and clear it when none present.
1697          */
1698         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1699                 goto bad;
1700         cnt = m->m_len;
1701         m->m_len += sizeof(struct in_addr);
1702         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1703         ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1704         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1705
1706         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1707                 opt = cp[IPOPT_OPTVAL];
1708                 if (opt == IPOPT_EOL)
1709                         break;
1710                 if (opt == IPOPT_NOP)
1711                         optlen = 1;
1712                 else {
1713                         if (cnt < IPOPT_OLEN + sizeof(*cp))
1714                                 goto bad;
1715                         optlen = cp[IPOPT_OLEN];
1716                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1717                                 goto bad;
1718                 }
1719                 switch (opt) {
1720
1721                 default:
1722                         break;
1723
1724                 case IPOPT_LSRR:
1725                 case IPOPT_SSRR:
1726                         /*
1727                          * user process specifies route as:
1728                          *      ->A->B->C->D
1729                          * D must be our final destination (but we can't
1730                          * check that since we may not have connected yet).
1731                          * A is first hop destination, which doesn't appear in
1732                          * actual IP option, but is stored before the options.
1733                          */
1734                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1735                                 goto bad;
1736                         m->m_len -= sizeof(struct in_addr);
1737                         cnt -= sizeof(struct in_addr);
1738                         optlen -= sizeof(struct in_addr);
1739                         cp[IPOPT_OLEN] = optlen;
1740                         /*
1741                          * Move first hop before start of options.
1742                          */
1743                         bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1744                             sizeof(struct in_addr));
1745                         /*
1746                          * Then copy rest of options back
1747                          * to close up the deleted entry.
1748                          */
1749                         ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1750                             sizeof(struct in_addr)),
1751                             (caddr_t)&cp[IPOPT_OFFSET+1],
1752                             (unsigned)cnt + sizeof(struct in_addr));
1753                         break;
1754                 }
1755         }
1756         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1757                 goto bad;
1758         *pcbopt = m;
1759         return (0);
1760
1761 bad:
1762         (void)m_free(m);
1763         return (EINVAL);
1764 }
1765
1766 /*
1767  * XXX
1768  * The whole multicast option thing needs to be re-thought.
1769  * Several of these options are equally applicable to non-multicast
1770  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1771  * standard option (IP_TTL).
1772  */
1773
1774 /*
1775  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1776  */
1777 static struct ifnet *
1778 ip_multicast_if(a, ifindexp)
1779         struct in_addr *a;
1780         int *ifindexp;
1781 {
1782         int ifindex;
1783         struct ifnet *ifp;
1784
1785         if (ifindexp)
1786                 *ifindexp = 0;
1787         if (ntohl(a->s_addr) >> 24 == 0) {
1788                 ifindex = ntohl(a->s_addr) & 0xffffff;
1789                 if (ifindex < 0 || if_index < ifindex)
1790                         return NULL;
1791                 ifp = ifindex2ifnet[ifindex];
1792                 if (ifindexp)
1793                         *ifindexp = ifindex;
1794         } else {
1795                 INADDR_TO_IFP(*a, ifp);
1796         }
1797         return ifp;
1798 }
1799
1800 /*
1801  * Set the IP multicast options in response to user setsockopt().
1802  */
1803 static int
1804 ip_setmoptions(sopt, imop)
1805         struct sockopt *sopt;
1806         struct ip_moptions **imop;
1807 {
1808         int error = 0;
1809         int i;
1810         struct in_addr addr;
1811         struct ip_mreq mreq;
1812         struct ifnet *ifp;
1813         struct ip_moptions *imo = *imop;
1814         struct route ro;
1815         struct sockaddr_in *dst;
1816         int ifindex;
1817         int s;
1818
1819         if (imo == NULL) {
1820                 /*
1821                  * No multicast option buffer attached to the pcb;
1822                  * allocate one and initialize to default values.
1823                  */
1824                 imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1825                     M_WAITOK);
1826
1827                 if (imo == NULL)
1828                         return (ENOBUFS);
1829                 *imop = imo;
1830                 imo->imo_multicast_ifp = NULL;
1831                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1832                 imo->imo_multicast_vif = -1;
1833                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1834                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1835                 imo->imo_num_memberships = 0;
1836         }
1837
1838         switch (sopt->sopt_name) {
1839         /* store an index number for the vif you wanna use in the send */
1840         case IP_MULTICAST_VIF:
1841                 if (legal_vif_num == 0) {
1842                         error = EOPNOTSUPP;
1843                         break;
1844                 }
1845                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1846                 if (error)
1847                         break;
1848                 if (!legal_vif_num(i) && (i != -1)) {
1849                         error = EINVAL;
1850                         break;
1851                 }
1852                 imo->imo_multicast_vif = i;
1853                 break;
1854
1855         case IP_MULTICAST_IF:
1856                 /*
1857                  * Select the interface for outgoing multicast packets.
1858                  */
1859                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1860                 if (error)
1861                         break;
1862                 /*
1863                  * INADDR_ANY is used to remove a previous selection.
1864                  * When no interface is selected, a default one is
1865                  * chosen every time a multicast packet is sent.
1866                  */
1867                 if (addr.s_addr == INADDR_ANY) {
1868                         imo->imo_multicast_ifp = NULL;
1869                         break;
1870                 }
1871                 /*
1872                  * The selected interface is identified by its local
1873                  * IP address.  Find the interface and confirm that
1874                  * it supports multicasting.
1875                  */
1876                 s = splimp();
1877                 ifp = ip_multicast_if(&addr, &ifindex);
1878                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1879                         splx(s);
1880                         error = EADDRNOTAVAIL;
1881                         break;
1882                 }
1883                 imo->imo_multicast_ifp = ifp;
1884                 if (ifindex)
1885                         imo->imo_multicast_addr = addr;
1886                 else
1887                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1888                 splx(s);
1889                 break;
1890
1891         case IP_MULTICAST_TTL:
1892                 /*
1893                  * Set the IP time-to-live for outgoing multicast packets.
1894                  * The original multicast API required a char argument,
1895                  * which is inconsistent with the rest of the socket API.
1896                  * We allow either a char or an int.
1897                  */
1898                 if (sopt->sopt_valsize == 1) {
1899                         u_char ttl;
1900                         error = sooptcopyin(sopt, &ttl, 1, 1);
1901                         if (error)
1902                                 break;
1903                         imo->imo_multicast_ttl = ttl;
1904                 } else {
1905                         u_int ttl;
1906                         error = sooptcopyin(sopt, &ttl, sizeof ttl,
1907                                             sizeof ttl);
1908                         if (error)
1909                                 break;
1910                         if (ttl > 255)
1911                                 error = EINVAL;
1912                         else
1913                                 imo->imo_multicast_ttl = ttl;
1914                 }
1915                 break;
1916
1917         case IP_MULTICAST_LOOP:
1918                 /*
1919                  * Set the loopback flag for outgoing multicast packets.
1920                  * Must be zero or one.  The original multicast API required a
1921                  * char argument, which is inconsistent with the rest
1922                  * of the socket API.  We allow either a char or an int.
1923                  */
1924                 if (sopt->sopt_valsize == 1) {
1925                         u_char loop;
1926                         error = sooptcopyin(sopt, &loop, 1, 1);
1927                         if (error)
1928                                 break;
1929                         imo->imo_multicast_loop = !!loop;
1930                 } else {
1931                         u_int loop;
1932                         error = sooptcopyin(sopt, &loop, sizeof loop,
1933                                             sizeof loop);
1934                         if (error)
1935                                 break;
1936                         imo->imo_multicast_loop = !!loop;
1937                 }
1938                 break;
1939
1940         case IP_ADD_MEMBERSHIP:
1941                 /*
1942                  * Add a multicast group membership.
1943                  * Group must be a valid IP multicast address.
1944                  */
1945                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1946                 if (error)
1947                         break;
1948
1949                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1950                         error = EINVAL;
1951                         break;
1952                 }
1953                 s = splimp();
1954                 /*
1955                  * If no interface address was provided, use the interface of
1956                  * the route to the given multicast address.
1957                  */
1958                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1959                         bzero((caddr_t)&ro, sizeof(ro));
1960                         dst = (struct sockaddr_in *)&ro.ro_dst;
1961                         dst->sin_len = sizeof(*dst);
1962                         dst->sin_family = AF_INET;
1963                         dst->sin_addr = mreq.imr_multiaddr;
1964                         rtalloc(&ro);
1965                         if (ro.ro_rt == NULL) {
1966                                 error = EADDRNOTAVAIL;
1967                                 splx(s);
1968                                 break;
1969                         }
1970                         ifp = ro.ro_rt->rt_ifp;
1971                         rtfree(ro.ro_rt);
1972                 }
1973                 else {
1974                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1975                 }
1976
1977                 /*
1978                  * See if we found an interface, and confirm that it
1979                  * supports multicast.
1980                  */
1981                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1982                         error = EADDRNOTAVAIL;
1983                         splx(s);
1984                         break;
1985                 }
1986                 /*
1987                  * See if the membership already exists or if all the
1988                  * membership slots are full.
1989                  */
1990                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1991                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1992                             imo->imo_membership[i]->inm_addr.s_addr
1993                                                 == mreq.imr_multiaddr.s_addr)
1994                                 break;
1995                 }
1996                 if (i < imo->imo_num_memberships) {
1997                         error = EADDRINUSE;
1998                         splx(s);
1999                         break;
2000                 }
2001                 if (i == IP_MAX_MEMBERSHIPS) {
2002                         error = ETOOMANYREFS;
2003                         splx(s);
2004                         break;
2005                 }
2006                 /*
2007                  * Everything looks good; add a new record to the multicast
2008                  * address list for the given interface.
2009                  */
2010                 if ((imo->imo_membership[i] =
2011                     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2012                         error = ENOBUFS;
2013                         splx(s);
2014                         break;
2015                 }
2016                 ++imo->imo_num_memberships;
2017                 splx(s);
2018                 break;
2019
2020         case IP_DROP_MEMBERSHIP:
2021                 /*
2022                  * Drop a multicast group membership.
2023                  * Group must be a valid IP multicast address.
2024                  */
2025                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2026                 if (error)
2027                         break;
2028
2029                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2030                         error = EINVAL;
2031                         break;
2032                 }
2033
2034                 s = splimp();
2035                 /*
2036                  * If an interface address was specified, get a pointer
2037                  * to its ifnet structure.
2038                  */
2039                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2040                         ifp = NULL;
2041                 else {
2042                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2043                         if (ifp == NULL) {
2044                                 error = EADDRNOTAVAIL;
2045                                 splx(s);
2046                                 break;
2047                         }
2048                 }
2049                 /*
2050                  * Find the membership in the membership array.
2051                  */
2052                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2053                         if ((ifp == NULL ||
2054                              imo->imo_membership[i]->inm_ifp == ifp) &&
2055                              imo->imo_membership[i]->inm_addr.s_addr ==
2056                              mreq.imr_multiaddr.s_addr)
2057                                 break;
2058                 }
2059                 if (i == imo->imo_num_memberships) {
2060                         error = EADDRNOTAVAIL;
2061                         splx(s);
2062                         break;
2063                 }
2064                 /*
2065                  * Give up the multicast address record to which the
2066                  * membership points.
2067                  */
2068                 in_delmulti(imo->imo_membership[i]);
2069                 /*
2070                  * Remove the gap in the membership array.
2071                  */
2072                 for (++i; i < imo->imo_num_memberships; ++i)
2073                         imo->imo_membership[i-1] = imo->imo_membership[i];
2074                 --imo->imo_num_memberships;
2075                 splx(s);
2076                 break;
2077
2078         default:
2079                 error = EOPNOTSUPP;
2080                 break;
2081         }
2082
2083         /*
2084          * If all options have default values, no need to keep the mbuf.
2085          */
2086         if (imo->imo_multicast_ifp == NULL &&
2087             imo->imo_multicast_vif == -1 &&
2088             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2089             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2090             imo->imo_num_memberships == 0) {
2091                 free(*imop, M_IPMOPTS);
2092                 *imop = NULL;
2093         }
2094
2095         return (error);
2096 }
2097
2098 /*
2099  * Return the IP multicast options in response to user getsockopt().
2100  */
2101 static int
2102 ip_getmoptions(sopt, imo)
2103         struct sockopt *sopt;
2104         struct ip_moptions *imo;
2105 {
2106         struct in_addr addr;
2107         struct in_ifaddr *ia;
2108         int error, optval;
2109         u_char coptval;
2110
2111         error = 0;
2112         switch (sopt->sopt_name) {
2113         case IP_MULTICAST_VIF:
2114                 if (imo != NULL)
2115                         optval = imo->imo_multicast_vif;
2116                 else
2117                         optval = -1;
2118                 error = sooptcopyout(sopt, &optval, sizeof optval);
2119                 break;
2120
2121         case IP_MULTICAST_IF:
2122                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2123                         addr.s_addr = INADDR_ANY;
2124                 else if (imo->imo_multicast_addr.s_addr) {
2125                         /* return the value user has set */
2126                         addr = imo->imo_multicast_addr;
2127                 } else {
2128                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2129                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2130                                 : IA_SIN(ia)->sin_addr.s_addr;
2131                 }
2132                 error = sooptcopyout(sopt, &addr, sizeof addr);
2133                 break;
2134
2135         case IP_MULTICAST_TTL:
2136                 if (imo == 0)
2137                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2138                 else
2139                         optval = coptval = imo->imo_multicast_ttl;
2140                 if (sopt->sopt_valsize == 1)
2141                         error = sooptcopyout(sopt, &coptval, 1);
2142                 else
2143                         error = sooptcopyout(sopt, &optval, sizeof optval);
2144                 break;
2145
2146         case IP_MULTICAST_LOOP:
2147                 if (imo == 0)
2148                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2149                 else
2150                         optval = coptval = imo->imo_multicast_loop;
2151                 if (sopt->sopt_valsize == 1)
2152                         error = sooptcopyout(sopt, &coptval, 1);
2153                 else
2154                         error = sooptcopyout(sopt, &optval, sizeof optval);
2155                 break;
2156
2157         default:
2158                 error = ENOPROTOOPT;
2159                 break;
2160         }
2161         return (error);
2162 }
2163
2164 /*
2165  * Discard the IP multicast options.
2166  */
2167 void
2168 ip_freemoptions(imo)
2169         struct ip_moptions *imo;
2170 {
2171         int i;
2172
2173         if (imo != NULL) {
2174                 for (i = 0; i < imo->imo_num_memberships; ++i)
2175                         in_delmulti(imo->imo_membership[i]);
2176                 free(imo, M_IPMOPTS);
2177         }
2178 }
2179
2180 /*
2181  * Routine called from ip_output() to loop back a copy of an IP multicast
2182  * packet to the input queue of a specified interface.  Note that this
2183  * calls the output routine of the loopback "driver", but with an interface
2184  * pointer that might NOT be a loopback interface -- evil, but easier than
2185  * replicating that code here.
2186  */
2187 static void
2188 ip_mloopback(ifp, m, dst, hlen)
2189         struct ifnet *ifp;
2190         struct mbuf *m;
2191         struct sockaddr_in *dst;
2192         int hlen;
2193 {
2194         struct ip *ip;
2195         struct mbuf *copym;
2196
2197         copym = m_copy(m, 0, M_COPYALL);
2198         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2199                 copym = m_pullup(copym, hlen);
2200         if (copym != NULL) {
2201                 /*
2202                  * We don't bother to fragment if the IP length is greater
2203                  * than the interface's MTU.  Can this possibly matter?
2204                  */
2205                 ip = mtod(copym, struct ip *);
2206                 ip->ip_len = htons(ip->ip_len);
2207                 ip->ip_off = htons(ip->ip_off);
2208                 ip->ip_sum = 0;
2209                 if (ip->ip_vhl == IP_VHL_BORING) {
2210                         ip->ip_sum = in_cksum_hdr(ip);
2211                 } else {
2212                         ip->ip_sum = in_cksum(copym, hlen);
2213                 }
2214                 /*
2215                  * NB:
2216                  * It's not clear whether there are any lingering
2217                  * reentrancy problems in other areas which might
2218                  * be exposed by using ip_input directly (in
2219                  * particular, everything which modifies the packet
2220                  * in-place).  Yet another option is using the
2221                  * protosw directly to deliver the looped back
2222                  * packet.  For the moment, we'll err on the side
2223                  * of safety by using if_simloop().
2224                  */
2225 #if 1 /* XXX */
2226                 if (dst->sin_family != AF_INET) {
2227                         printf("ip_mloopback: bad address family %d\n",
2228                                                 dst->sin_family);
2229                         dst->sin_family = AF_INET;
2230                 }
2231 #endif
2232
2233 #ifdef notdef
2234                 copym->m_pkthdr.rcvif = ifp;
2235                 ip_input(copym);
2236 #else
2237                 /* if the checksum hasn't been computed, mark it as valid */
2238                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2239                         copym->m_pkthdr.csum_flags |=
2240                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2241                         copym->m_pkthdr.csum_data = 0xffff;
2242                 }
2243                 if_simloop(ifp, copym, dst->sin_family, 0);
2244 #endif
2245         }
2246 }