sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.32 2005/08/15 16:46:21 dillon Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/kernel.h>
  46 #include <sys/malloc.h>
  47 #include <sys/mbuf.h>
  48 #include <sys/protosw.h>
  49 #include <sys/socket.h>
  50 #include <sys/socketvar.h>
  51 #include <sys/proc.h>
  52 #include <sys/sysctl.h>
  53 #include <sys/thread2.h>
  54 #include <sys/in_cksum.h>
  55
  56 #include <net/if.h>
  57 #include <net/netisr.h>
  58 #include <net/pfil.h>
  59 #include <net/route.h>
  60
  61 #include <netinet/in.h>
  62 #include <netinet/in_systm.h>
  63 #include <netinet/ip.h>
  64 #include <netinet/in_pcb.h>
  65 #include <netinet/in_var.h>
  66 #include <netinet/ip_var.h>
  67
  68 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  69
  70 #ifdef IPSEC
  71 #include <netinet6/ipsec.h>
  72 #include <netproto/key/key.h>
  73 #ifdef IPSEC_DEBUG
  74 #include <netproto/key/key_debug.h>
  75 #else
  76 #define KEYDEBUG(lev,arg)
  77 #endif
  78 #endif /*IPSEC*/
  79
  80 #ifdef FAST_IPSEC
  81 #include <netproto/ipsec/ipsec.h>
  82 #include <netproto/ipsec/xform.h>
  83 #include <netproto/ipsec/key.h>
  84 #endif /*FAST_IPSEC*/
  85
  86 #include <net/ipfw/ip_fw.h>
  87 #include <net/dummynet/ip_dummynet.h>
  88
  89 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
  90                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  91                                   (ntohl(a.s_addr)>>16)&0xFF,\
  92                                   (ntohl(a.s_addr)>>8)&0xFF,\
  93                                   (ntohl(a.s_addr))&0xFF, y);
  94
  95 u_short ip_id;
  96
  97 #ifdef MBUF_STRESS_TEST
  98 int mbuf_frag_size = 0;
  99 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 100         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 101 #endif
 102
 103 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 104 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 105 static void     ip_mloopback
 106         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 107 static int      ip_getmoptions
 108         (struct sockopt *, struct ip_moptions *);
 109 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 110 static int      ip_setmoptions
 111         (struct sockopt *, struct ip_moptions **);
 112
 113 int     ip_optcopy(struct ip *, struct ip *);
 114
 115
 116 extern  struct protosw inetsw[];
 117
 118 /*
 119  * IP output.  The packet in mbuf chain m contains a skeletal IP
 120  * header (with len, off, ttl, proto, tos, src, dst).
 121  * The mbuf chain containing the packet will be freed.
 122  * The mbuf opt, if present, will not be freed.
 123  */
 124 int
 125 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 126           int flags, struct ip_moptions *imo, struct inpcb *inp)
 127 {
 128         struct ip *ip;
 129         struct ifnet *ifp = NULL;       /* keep compiler happy */
 130         struct mbuf *m;
 131         int hlen = sizeof(struct ip);
 132         int len, off, error = 0;
 133         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 134         struct in_ifaddr *ia = NULL;
 135         int isbroadcast, sw_csum;
 136         struct in_addr pkt_dst;
 137         struct route iproute;
 138 #ifdef IPSEC
 139         struct secpolicy *sp = NULL;
 140         struct socket *so = inp ? inp->inp_socket : NULL;
 141 #endif
 142 #ifdef FAST_IPSEC
 143         struct m_tag *mtag;
 144         struct secpolicy *sp = NULL;
 145         struct tdb_ident *tdbi;
 146 #endif /* FAST_IPSEC */
 147         struct ip_fw_args args;
 148         int src_was_INADDR_ANY = 0;     /* as the name says... */
 149
 150         args.eh = NULL;
 151         args.rule = NULL;
 152         args.next_hop = NULL;
 153
 154         /* Grab info from MT_TAG mbufs prepended to the chain. */
 155         while (m0 != NULL && m0->m_type == MT_TAG) {
 156                 switch(m0->_m_tag_id) {
 157                 case PACKET_TAG_DUMMYNET:
 158                         /*
 159                          * the packet was already tagged, so part of the
 160                          * processing was already done, and we need to go down.
 161                          * Get parameters from the header.
 162                          */
 163                         args.rule = ((struct dn_pkt *)m0)->rule;
 164                         opt = NULL ;
 165                         ro = &((struct dn_pkt *)m0)->ro;
 166                         imo = NULL ;
 167                         dst = ((struct dn_pkt *)m0)->dn_dst ;
 168                         ifp = ((struct dn_pkt *)m0)->ifp ;
 169                         flags = ((struct dn_pkt *)m0)->flags ;
 170                         break;
 171                 case PACKET_TAG_IPFORWARD:
 172                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 173                         break;
 174                 default:
 175                         printf("ip_output: unrecognised MT_TAG tag %d\n",
 176                             m0->_m_tag_id);
 177                         break;
 178                 }
 179                 m0 = m0->m_next;
 180         }
 181         m = m0;
 182         KASSERT(m != NULL && (m->m_flags & M_PKTHDR), ("ip_output: no HDR"));
 183
 184         if (ro == NULL) {
 185                 ro = &iproute;
 186                 bzero(ro, sizeof *ro);
 187         }
 188
 189         if (args.rule != NULL) {        /* dummynet already saw us */
 190                 ip = mtod(m, struct ip *);
 191                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 192                 if (ro->ro_rt)
 193                         ia = ifatoia(ro->ro_rt->rt_ifa);
 194                 goto sendit;
 195         }
 196
 197         if (opt) {
 198                 len = 0;
 199                 m = ip_insertoptions(m, opt, &len);
 200                 if (len != 0)
 201                         hlen = len;
 202         }
 203         ip = mtod(m, struct ip *);
 204         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 205
 206         /*
 207          * Fill in IP header.
 208          */
 209         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 210                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 211                 ip->ip_off &= IP_DF;
 212                 ip->ip_id = ip_newid();
 213                 ipstat.ips_localout++;
 214         } else {
 215                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 216         }
 217
 218         dst = (struct sockaddr_in *)&ro->ro_dst;
 219         /*
 220          * If there is a cached route,
 221          * check that it is to the same destination
 222          * and is still up.  If not, free it and try again.
 223          * The address family should also be checked in case of sharing the
 224          * cache with IPv6.
 225          */
 226         if (ro->ro_rt &&
 227             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 228              dst->sin_family != AF_INET ||
 229              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 230                 rtfree(ro->ro_rt);
 231                 ro->ro_rt = (struct rtentry *)NULL;
 232         }
 233         if (ro->ro_rt == NULL) {
 234                 bzero(dst, sizeof *dst);
 235                 dst->sin_family = AF_INET;
 236                 dst->sin_len = sizeof *dst;
 237                 dst->sin_addr = pkt_dst;
 238         }
 239         /*
 240          * If routing to interface only,
 241          * short circuit routing lookup.
 242          */
 243         if (flags & IP_ROUTETOIF) {
 244                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 245                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 246                         ipstat.ips_noroute++;
 247                         error = ENETUNREACH;
 248                         goto bad;
 249                 }
 250                 ifp = ia->ia_ifp;
 251                 ip->ip_ttl = 1;
 252                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 253         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 254                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 255                 /*
 256                  * Bypass the normal routing lookup for multicast
 257                  * packets if the interface is specified.
 258                  */
 259                 ifp = imo->imo_multicast_ifp;
 260                 IFP_TO_IA(ifp, ia);
 261                 isbroadcast = 0;        /* fool gcc */
 262         } else {
 263                 /*
 264                  * If this is the case, we probably don't want to allocate
 265                  * a protocol-cloned route since we didn't get one from the
 266                  * ULP.  This lets TCP do its thing, while not burdening
 267                  * forwarding or ICMP with the overhead of cloning a route.
 268                  * Of course, we still want to do any cloning requested by
 269                  * the link layer, as this is probably required in all cases
 270                  * for correct operation (as it is for ARP).
 271                  */
 272                 if (ro->ro_rt == NULL)
 273                         rtalloc_ign(ro, RTF_PRCLONING);
 274                 if (ro->ro_rt == NULL) {
 275                         ipstat.ips_noroute++;
 276                         error = EHOSTUNREACH;
 277                         goto bad;
 278                 }
 279                 ia = ifatoia(ro->ro_rt->rt_ifa);
 280                 ifp = ro->ro_rt->rt_ifp;
 281                 ro->ro_rt->rt_use++;
 282                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 283                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 284                 if (ro->ro_rt->rt_flags & RTF_HOST)
 285                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 286                 else
 287                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 288         }
 289         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 290                 struct in_multi *inm;
 291
 292                 m->m_flags |= M_MCAST;
 293                 /*
 294                  * IP destination address is multicast.  Make sure "dst"
 295                  * still points to the address in "ro".  (It may have been
 296                  * changed to point to a gateway address, above.)
 297                  */
 298                 dst = (struct sockaddr_in *)&ro->ro_dst;
 299                 /*
 300                  * See if the caller provided any multicast options
 301                  */
 302                 if (imo != NULL) {
 303                         ip->ip_ttl = imo->imo_multicast_ttl;
 304                         if (imo->imo_multicast_vif != -1)
 305                                 ip->ip_src.s_addr =
 306                                     ip_mcast_src ?
 307                                     ip_mcast_src(imo->imo_multicast_vif) :
 308                                     INADDR_ANY;
 309                 } else
 310                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 311                 /*
 312                  * Confirm that the outgoing interface supports multicast.
 313                  */
 314                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 315                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 316                                 ipstat.ips_noroute++;
 317                                 error = ENETUNREACH;
 318                                 goto bad;
 319                         }
 320                 }
 321                 /*
 322                  * If source address not specified yet, use address
 323                  * of outgoing interface.
 324                  */
 325                 if (ip->ip_src.s_addr == INADDR_ANY) {
 326                         /* Interface may have no addresses. */
 327                         if (ia != NULL)
 328                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 329                 }
 330
 331                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 332                 if (inm != NULL &&
 333                    (imo == NULL || imo->imo_multicast_loop)) {
 334                         /*
 335                          * If we belong to the destination multicast group
 336                          * on the outgoing interface, and the caller did not
 337                          * forbid loopback, loop back a copy.
 338                          */
 339                         ip_mloopback(ifp, m, dst, hlen);
 340                 }
 341                 else {
 342                         /*
 343                          * If we are acting as a multicast router, perform
 344                          * multicast forwarding as if the packet had just
 345                          * arrived on the interface to which we are about
 346                          * to send.  The multicast forwarding function
 347                          * recursively calls this function, using the
 348                          * IP_FORWARDING flag to prevent infinite recursion.
 349                          *
 350                          * Multicasts that are looped back by ip_mloopback(),
 351                          * above, will be forwarded by the ip_input() routine,
 352                          * if necessary.
 353                          */
 354                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 355                                 /*
 356                                  * If rsvp daemon is not running, do not
 357                                  * set ip_moptions. This ensures that the packet
 358                                  * is multicast and not just sent down one link
 359                                  * as prescribed by rsvpd.
 360                                  */
 361                                 if (!rsvp_on)
 362                                         imo = NULL;
 363                                 if (ip_mforward &&
 364                                     ip_mforward(ip, ifp, m, imo) != 0) {
 365                                         m_freem(m);
 366                                         goto done;
 367                                 }
 368                         }
 369                 }
 370
 371                 /*
 372                  * Multicasts with a time-to-live of zero may be looped-
 373                  * back, above, but must not be transmitted on a network.
 374                  * Also, multicasts addressed to the loopback interface
 375                  * are not sent -- the above call to ip_mloopback() will
 376                  * loop back a copy if this host actually belongs to the
 377                  * destination group on the loopback interface.
 378                  */
 379                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 380                         m_freem(m);
 381                         goto done;
 382                 }
 383
 384                 goto sendit;
 385         }
 386 #ifndef notdef
 387         /*
 388          * If the source address is not specified yet, use the address
 389          * of the outoing interface. In case, keep note we did that, so
 390          * if the the firewall changes the next-hop causing the output
 391          * interface to change, we can fix that.
 392          */
 393         if (ip->ip_src.s_addr == INADDR_ANY) {
 394                 /* Interface may have no addresses. */
 395                 if (ia != NULL) {
 396                         ip->ip_src = IA_SIN(ia)->sin_addr;
 397                         src_was_INADDR_ANY = 1;
 398                 }
 399         }
 400 #endif /* notdef */
 401 #ifdef ALTQ
 402         /*
 403          * Disable packet drop hack.
 404          * Packetdrop should be done by queueing.
 405          */
 406 #else /* !ALTQ */
 407         /*
 408          * Verify that we have any chance at all of being able to queue
 409          *      the packet or packet fragments
 410          */
 411         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 412                 ifp->if_snd.ifq_maxlen) {
 413                         error = ENOBUFS;
 414                         ipstat.ips_odropped++;
 415                         goto bad;
 416         }
 417 #endif /* !ALTQ */
 418
 419         /*
 420          * Look for broadcast address and
 421          * verify user is allowed to send
 422          * such a packet.
 423          */
 424         if (isbroadcast) {
 425                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 426                         error = EADDRNOTAVAIL;
 427                         goto bad;
 428                 }
 429                 if (!(flags & IP_ALLOWBROADCAST)) {
 430                         error = EACCES;
 431                         goto bad;
 432                 }
 433                 /* don't allow broadcast messages to be fragmented */
 434                 if (ip->ip_len > ifp->if_mtu) {
 435                         error = EMSGSIZE;
 436                         goto bad;
 437                 }
 438                 m->m_flags |= M_BCAST;
 439         } else {
 440                 m->m_flags &= ~M_BCAST;
 441         }
 442
 443 sendit:
 444 #ifdef IPSEC
 445         /* get SP for this packet */
 446         if (so == NULL)
 447                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 448         else
 449                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 450
 451         if (sp == NULL) {
 452                 ipsecstat.out_inval++;
 453                 goto bad;
 454         }
 455
 456         error = 0;
 457
 458         /* check policy */
 459         switch (sp->policy) {
 460         case IPSEC_POLICY_DISCARD:
 461                 /*
 462                  * This packet is just discarded.
 463                  */
 464                 ipsecstat.out_polvio++;
 465                 goto bad;
 466
 467         case IPSEC_POLICY_BYPASS:
 468         case IPSEC_POLICY_NONE:
 469                 /* no need to do IPsec. */
 470                 goto skip_ipsec;
 471
 472         case IPSEC_POLICY_IPSEC:
 473                 if (sp->req == NULL) {
 474                         /* acquire a policy */
 475                         error = key_spdacquire(sp);
 476                         goto bad;
 477                 }
 478                 break;
 479
 480         case IPSEC_POLICY_ENTRUST:
 481         default:
 482                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
 483         }
 484     {
 485         struct ipsec_output_state state;
 486         bzero(&state, sizeof state);
 487         state.m = m;
 488         if (flags & IP_ROUTETOIF) {
 489                 state.ro = &iproute;
 490                 bzero(&iproute, sizeof iproute);
 491         } else
 492                 state.ro = ro;
 493         state.dst = (struct sockaddr *)dst;
 494
 495         ip->ip_sum = 0;
 496
 497         /*
 498          * XXX
 499          * delayed checksums are not currently compatible with IPsec
 500          */
 501         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 502                 in_delayed_cksum(m);
 503                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 504         }
 505
 506         ip->ip_len = htons(ip->ip_len);
 507         ip->ip_off = htons(ip->ip_off);
 508
 509         error = ipsec4_output(&state, sp, flags);
 510
 511         m = state.m;
 512         if (flags & IP_ROUTETOIF) {
 513                 /*
 514                  * if we have tunnel mode SA, we may need to ignore
 515                  * IP_ROUTETOIF.
 516                  */
 517                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 518                         flags &= ~IP_ROUTETOIF;
 519                         ro = state.ro;
 520                 }
 521         } else
 522                 ro = state.ro;
 523         dst = (struct sockaddr_in *)state.dst;
 524         if (error) {
 525                 /* mbuf is already reclaimed in ipsec4_output. */
 526                 m0 = NULL;
 527                 switch (error) {
 528                 case EHOSTUNREACH:
 529                 case ENETUNREACH:
 530                 case EMSGSIZE:
 531                 case ENOBUFS:
 532                 case ENOMEM:
 533                         break;
 534                 default:
 535                         printf("ip4_output (ipsec): error code %d\n", error);
 536                         /*fall through*/
 537                 case ENOENT:
 538                         /* don't show these error codes to the user */
 539                         error = 0;
 540                         break;
 541                 }
 542                 goto bad;
 543         }
 544     }
 545
 546         /* be sure to update variables that are affected by ipsec4_output() */
 547         ip = mtod(m, struct ip *);
 548 #ifdef _IP_VHL
 549         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 550 #else
 551         hlen = ip->ip_hl << 2;
 552 #endif
 553         if (ro->ro_rt == NULL) {
 554                 if (!(flags & IP_ROUTETOIF)) {
 555                         printf("ip_output: "
 556                                 "can't update route after IPsec processing\n");
 557                         error = EHOSTUNREACH;   /*XXX*/
 558                         goto bad;
 559                 }
 560         } else {
 561                 ia = ifatoia(ro->ro_rt->rt_ifa);
 562                 ifp = ro->ro_rt->rt_ifp;
 563         }
 564
 565         /* make it flipped, again. */
 566         ip->ip_len = ntohs(ip->ip_len);
 567         ip->ip_off = ntohs(ip->ip_off);
 568 skip_ipsec:
 569 #endif /*IPSEC*/
 570 #ifdef FAST_IPSEC
 571         /*
 572          * Check the security policy (SP) for the packet and, if
 573          * required, do IPsec-related processing.  There are two
 574          * cases here; the first time a packet is sent through
 575          * it will be untagged and handled by ipsec4_checkpolicy.
 576          * If the packet is resubmitted to ip_output (e.g. after
 577          * AH, ESP, etc. processing), there will be a tag to bypass
 578          * the lookup and related policy checking.
 579          */
 580         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 581         crit_enter();
 582         if (mtag != NULL) {
 583                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 584                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 585                 if (sp == NULL)
 586                         error = -EINVAL;        /* force silent drop */
 587                 m_tag_delete(m, mtag);
 588         } else {
 589                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 590                                         &error, inp);
 591         }
 592         /*
 593          * There are four return cases:
 594          *    sp != NULL                    apply IPsec policy
 595          *    sp == NULL, error == 0        no IPsec handling needed
 596          *    sp == NULL, error == -EINVAL  discard packet w/o error
 597          *    sp == NULL, error != 0        discard packet, report error
 598          */
 599         if (sp != NULL) {
 600                 /* Loop detection, check if ipsec processing already done */
 601                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 602                 for (mtag = m_tag_first(m); mtag != NULL;
 603                      mtag = m_tag_next(m, mtag)) {
 604                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 605                                 continue;
 606                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 607                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 608                                 continue;
 609                         /*
 610                          * Check if policy has an SA associated with it.
 611                          * This can happen when an SP has yet to acquire
 612                          * an SA; e.g. on first reference.  If it occurs,
 613                          * then we let ipsec4_process_packet do its thing.
 614                          */
 615                         if (sp->req->sav == NULL)
 616                                 break;
 617                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 618                         if (tdbi->spi == sp->req->sav->spi &&
 619                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 620                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 621                                  sizeof(union sockaddr_union)) == 0) {
 622                                 /*
 623                                  * No IPsec processing is needed, free
 624                                  * reference to SP.
 625                                  *
 626                                  * NB: null pointer to avoid free at
 627                                  *     done: below.
 628                                  */
 629                                 KEY_FREESP(&sp), sp = NULL;
 630                                 crit_exit();
 631                                 goto spd_done;
 632                         }
 633                 }
 634
 635                 /*
 636                  * Do delayed checksums now because we send before
 637                  * this is done in the normal processing path.
 638                  */
 639                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 640                         in_delayed_cksum(m);
 641                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 642                 }
 643
 644                 ip->ip_len = htons(ip->ip_len);
 645                 ip->ip_off = htons(ip->ip_off);
 646
 647                 /* NB: callee frees mbuf */
 648                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 649                 /*
 650                  * Preserve KAME behaviour: ENOENT can be returned
 651                  * when an SA acquire is in progress.  Don't propagate
 652                  * this to user-level; it confuses applications.
 653                  *
 654                  * XXX this will go away when the SADB is redone.
 655                  */
 656                 if (error == ENOENT)
 657                         error = 0;
 658                 crit_exit();
 659                 goto done;
 660         } else {
 661                 crit_exit();
 662
 663                 if (error != 0) {
 664                         /*
 665                          * Hack: -EINVAL is used to signal that a packet
 666                          * should be silently discarded.  This is typically
 667                          * because we asked key management for an SA and
 668                          * it was delayed (e.g. kicked up to IKE).
 669                          */
 670                         if (error == -EINVAL)
 671                                 error = 0;
 672                         goto bad;
 673                 } else {
 674                         /* No IPsec processing for this packet. */
 675                 }
 676 #ifdef notyet
 677                 /*
 678                  * If deferred crypto processing is needed, check that
 679                  * the interface supports it.
 680                  */
 681                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 682                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 683                         /* notify IPsec to do its own crypto */
 684                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 685                         error = EHOSTUNREACH;
 686                         goto bad;
 687                 }
 688 #endif
 689         }
 690 spd_done:
 691 #endif /* FAST_IPSEC */
 692         /*
 693          * IpHack's section.
 694          * - Xlate: translate packet's addr/port (NAT).
 695          * - Firewall: deny/allow/etc.
 696          * - Wrap: fake packet's addr/port <unimpl.>
 697          * - Encapsulate: put it in another IP and send out. <unimp.>
 698          */
 699
 700         /*
 701          * Run through list of hooks for output packets.
 702          */
 703         if (pfil_has_hooks(&inet_pfil_hook)) {
 704                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 705                 if (error != 0 || m == NULL)
 706                         goto done;
 707                 ip = mtod(m, struct ip *);
 708         }
 709
 710         /*
 711          * Check with the firewall...
 712          * but not if we are already being fwd'd from a firewall.
 713          */
 714         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 715                 struct sockaddr_in *old = dst;
 716
 717                 args.m = m;
 718                 args.next_hop = dst;
 719                 args.oif = ifp;
 720                 off = ip_fw_chk_ptr(&args);
 721                 m = args.m;
 722                 dst = args.next_hop;
 723
 724                 /*
 725                  * On return we must do the following:
 726                  * m == NULL    -> drop the pkt (old interface, deprecated)
 727                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 728                  * 1<=off<= 0xffff              -> DIVERT
 729                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 730                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 731                  * dst != old                   -> IPFIREWALL_FORWARD
 732                  * off==0, dst==old             -> accept
 733                  * If some of the above modules are not compiled in, then
 734                  * we should't have to check the corresponding condition
 735                  * (because the ipfw control socket should not accept
 736                  * unsupported rules), but better play safe and drop
 737                  * packets in case of doubt.
 738                  */
 739                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 740                         if (m)
 741                                 m_freem(m);
 742                         error = EACCES;
 743                         goto done;
 744                 }
 745                 ip = mtod(m, struct ip *);
 746                 if (off == 0 && dst == old)             /* common case */
 747                         goto pass;
 748                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG)) {
 749                         /*
 750                          * pass the pkt to dummynet. Need to include
 751                          * pipe number, m, ifp, ro, dst because these are
 752                          * not recomputed in the next pass.
 753                          * All other parameters have been already used and
 754                          * so they are not needed anymore.
 755                          * XXX note: if the ifp or ro entry are deleted
 756                          * while a pkt is in dummynet, we are in trouble!
 757                          */
 758                         args.ro = ro;
 759                         args.dst = dst;
 760                         args.flags = flags;
 761
 762                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
 763                                 &args);
 764                         goto done;
 765                 }
 766 #ifdef IPDIVERT
 767                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
 768                         struct mbuf *clone = NULL;
 769
 770                         /* Clone packet if we're doing a 'tee' */
 771                         if ((off & IP_FW_PORT_TEE_FLAG))
 772                                 clone = m_dup(m, MB_DONTWAIT);
 773
 774                         /*
 775                          * XXX
 776                          * delayed checksums are not currently compatible
 777                          * with divert sockets.
 778                          */
 779                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 780                                 in_delayed_cksum(m);
 781                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 782                         }
 783
 784                         /* Restore packet header fields to original values */
 785                         ip->ip_len = htons(ip->ip_len);
 786                         ip->ip_off = htons(ip->ip_off);
 787
 788                         /* Deliver packet to divert input routine */
 789                         divert_packet(m, 0, off & 0xffff);
 790
 791                         /* If 'tee', continue with original packet */
 792                         if (clone != NULL) {
 793                                 m = clone;
 794                                 ip = mtod(m, struct ip *);
 795                                 goto pass;
 796                         }
 797                         goto done;
 798                 }
 799 #endif
 800
 801                 /* IPFIREWALL_FORWARD */
 802                 /*
 803                  * Check dst to make sure it is directly reachable on the
 804                  * interface we previously thought it was.
 805                  * If it isn't (which may be likely in some situations) we have
 806                  * to re-route it (ie, find a route for the next-hop and the
 807                  * associated interface) and set them here. This is nested
 808                  * forwarding which in most cases is undesirable, except where
 809                  * such control is nigh impossible. So we do it here.
 810                  * And I'm babbling.
 811                  */
 812                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 813 #if 0
 814                         /*
 815                          * XXX To improve readability, this block should be
 816                          * changed into a function call as below:
 817                          */
 818                         error = ip_ipforward(&m, &dst, &ifp);
 819                         if (error)
 820                                 goto bad;
 821                         if (m == NULL) /* ip_input consumed the mbuf */
 822                                 goto done;
 823 #else
 824                         struct in_ifaddr *ia;
 825
 826                         /*
 827                          * XXX sro_fwd below is static, and a pointer
 828                          * to it gets passed to routines downstream.
 829                          * This could have surprisingly bad results in
 830                          * practice, because its content is overwritten
 831                          * by subsequent packets.
 832                          */
 833                         /* There must be a better way to do this next line... */
 834                         static struct route sro_fwd;
 835                         struct route *ro_fwd = &sro_fwd;
 836
 837 #if 0
 838                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 839                             dst->sin_addr, "\n");
 840 #endif
 841
 842                         /*
 843                          * We need to figure out if we have been forwarded
 844                          * to a local socket. If so, then we should somehow
 845                          * "loop back" to ip_input, and get directed to the
 846                          * PCB as if we had received this packet. This is
 847                          * because it may be dificult to identify the packets
 848                          * you want to forward until they are being output
 849                          * and have selected an interface. (e.g. locally
 850                          * initiated packets) If we used the loopback inteface,
 851                          * we would not be able to control what happens
 852                          * as the packet runs through ip_input() as
 853                          * it is done through a ISR.
 854                          */
 855                         LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
 856                                      ia_hash) {
 857                                 /*
 858                                  * If the addr to forward to is one
 859                                  * of ours, we pretend to
 860                                  * be the destination for this packet.
 861                                  */
 862                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 863                                                  dst->sin_addr.s_addr)
 864                                         break;
 865                         }
 866                         if (ia != NULL) {    /* tell ip_input "dont filter" */
 867                                 struct m_hdr tag;
 868
 869                                 tag.mh_type = MT_TAG;
 870                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 871                                 tag.mh_data = (caddr_t)args.next_hop;
 872                                 tag.mh_next = m;
 873
 874                                 if (m->m_pkthdr.rcvif == NULL)
 875                                         m->m_pkthdr.rcvif = ifunit("lo0");
 876                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 877                                         m->m_pkthdr.csum_flags |=
 878                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 879                                         m0->m_pkthdr.csum_data = 0xffff;
 880                                 }
 881                                 m->m_pkthdr.csum_flags |=
 882                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 883                                 ip->ip_len = htons(ip->ip_len);
 884                                 ip->ip_off = htons(ip->ip_off);
 885                                 ip_input((struct mbuf *)&tag);
 886                                 goto done;
 887                         }
 888                         /* Some of the logic for this was nicked from above.
 889                          *
 890                          * This rewrites the cached route in a local PCB.
 891                          * Is this what we want to do?
 892                          */
 893                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
 894                         ro_fwd->ro_rt = NULL;
 895
 896                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 897                         if (ro_fwd->ro_rt == NULL) {
 898                                 ipstat.ips_noroute++;
 899                                 error = EHOSTUNREACH;
 900                                 goto bad;
 901                         }
 902
 903                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 904                         ifp = ro_fwd->ro_rt->rt_ifp;
 905                         ro_fwd->ro_rt->rt_use++;
 906                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 907                                 dst = (struct sockaddr_in *)
 908                                     ro_fwd->ro_rt->rt_gateway;
 909                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 910                                 isbroadcast =
 911                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 912                         else
 913                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 914                         if (ro->ro_rt != NULL)
 915                                 rtfree(ro->ro_rt);
 916                         ro->ro_rt = ro_fwd->ro_rt;
 917                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 918
 919 #endif  /* ... block to be put into a function */
 920                         /*
 921                          * If we added a default src ip earlier,
 922                          * which would have been gotten from the-then
 923                          * interface, do it again, from the new one.
 924                          */
 925                         if (src_was_INADDR_ANY)
 926                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 927                         goto pass ;
 928                 }
 929
 930                 /*
 931                  * if we get here, none of the above matches, and
 932                  * we have to drop the pkt
 933                  */
 934                 m_freem(m);
 935                 error = EACCES; /* not sure this is the right error msg */
 936                 goto done;
 937         }
 938
 939 pass:
 940         /* 127/8 must not appear on wire - RFC1122. */
 941         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 942             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 943                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 944                         ipstat.ips_badaddr++;
 945                         error = EADDRNOTAVAIL;
 946                         goto bad;
 947                 }
 948         }
 949
 950         m->m_pkthdr.csum_flags |= CSUM_IP;
 951         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 952         if (sw_csum & CSUM_DELAY_DATA) {
 953                 in_delayed_cksum(m);
 954                 sw_csum &= ~CSUM_DELAY_DATA;
 955         }
 956         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 957
 958         /*
 959          * If small enough for interface, or the interface will take
 960          * care of the fragmentation for us, can just send directly.
 961          */
 962         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 963             !(ip->ip_off & IP_DF))) {
 964                 ip->ip_len = htons(ip->ip_len);
 965                 ip->ip_off = htons(ip->ip_off);
 966                 ip->ip_sum = 0;
 967                 if (sw_csum & CSUM_DELAY_IP) {
 968                         if (ip->ip_vhl == IP_VHL_BORING) {
 969                                 ip->ip_sum = in_cksum_hdr(ip);
 970                         } else {
 971                                 ip->ip_sum = in_cksum(m, hlen);
 972                         }
 973                 }
 974
 975                 /* Record statistics for this interface address. */
 976                 if (!(flags & IP_FORWARDING) && ia) {
 977                         ia->ia_ifa.if_opackets++;
 978                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 979                 }
 980
 981 #ifdef IPSEC
 982                 /* clean ipsec history once it goes out of the node */
 983                 ipsec_delaux(m);
 984 #endif
 985
 986 #ifdef MBUF_STRESS_TEST
 987                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 988                         struct mbuf *m1, *m2;
 989                         int length, tmp;
 990
 991                         tmp = length = m->m_pkthdr.len;
 992
 993                         while ((length -= mbuf_frag_size) >= 1) {
 994                                 m1 = m_split(m, length, MB_DONTWAIT);
 995                                 if (m1 == NULL)
 996                                         break;
 997                                 m2 = m;
 998                                 while (m2->m_next != NULL)
 999                                         m2 = m2->m_next;
1000                                 m2->m_next = m1;
1001                         }
1002                         m->m_pkthdr.len = tmp;
1003                 }
1004 #endif
1005                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1006                                           ro->ro_rt);
1007                 goto done;
1008         }
1009
1010         if (ip->ip_off & IP_DF) {
1011                 error = EMSGSIZE;
1012                 /*
1013                  * This case can happen if the user changed the MTU
1014                  * of an interface after enabling IP on it.  Because
1015                  * most netifs don't keep track of routes pointing to
1016                  * them, there is no way for one to update all its
1017                  * routes when the MTU is changed.
1018                  */
1019                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1020                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1021                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1022                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1023                 }
1024                 ipstat.ips_cantfrag++;
1025                 goto bad;
1026         }
1027
1028         /*
1029          * Too large for interface; fragment if possible. If successful,
1030          * on return, m will point to a list of packets to be sent.
1031          */
1032         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1033         if (error)
1034                 goto bad;
1035         for (; m; m = m0) {
1036                 m0 = m->m_nextpkt;
1037                 m->m_nextpkt = NULL;
1038 #ifdef IPSEC
1039                 /* clean ipsec history once it goes out of the node */
1040                 ipsec_delaux(m);
1041 #endif
1042                 if (error == 0) {
1043                         /* Record statistics for this interface address. */
1044                         if (ia != NULL) {
1045                                 ia->ia_ifa.if_opackets++;
1046                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1047                         }
1048                         error = (*ifp->if_output)(ifp, m,
1049                                                   (struct sockaddr *)dst,
1050                                                   ro->ro_rt);
1051                 } else
1052                         m_freem(m);
1053         }
1054
1055         if (error == 0)
1056                 ipstat.ips_fragmented++;
1057
1058 done:
1059         if (ro == &iproute && ro->ro_rt != NULL) {
1060                 RTFREE(ro->ro_rt);
1061                 ro->ro_rt = NULL;
1062         }
1063 #ifdef IPSEC
1064         if (sp != NULL) {
1065                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1066                         printf("DP ip_output call free SP:%p\n", sp));
1067                 key_freesp(sp);
1068         }
1069 #endif
1070 #ifdef FAST_IPSEC
1071         if (sp != NULL)
1072                 KEY_FREESP(&sp);
1073 #endif
1074         return (error);
1075 bad:
1076         m_freem(m);
1077         goto done;
1078 }
1079
1080 /*
1081  * Create a chain of fragments which fit the given mtu. m_frag points to the
1082  * mbuf to be fragmented; on return it points to the chain with the fragments.
1083  * Return 0 if no error. If error, m_frag may contain a partially built
1084  * chain of fragments that should be freed by the caller.
1085  *
1086  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1087  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1088  */
1089 int
1090 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1091             u_long if_hwassist_flags, int sw_csum)
1092 {
1093         int error = 0;
1094         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1095         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1096         int off;
1097         struct mbuf *m0 = *m_frag;      /* the original packet          */
1098         int firstlen;
1099         struct mbuf **mnext;
1100         int nfrags;
1101
1102         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1103                 ipstat.ips_cantfrag++;
1104                 return EMSGSIZE;
1105         }
1106
1107         /*
1108          * Must be able to put at least 8 bytes per fragment.
1109          */
1110         if (len < 8)
1111                 return EMSGSIZE;
1112
1113         /*
1114          * If the interface will not calculate checksums on
1115          * fragmented packets, then do it here.
1116          */
1117         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1118             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1119                 in_delayed_cksum(m0);
1120                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1121         }
1122
1123         if (len > PAGE_SIZE) {
1124                 /*
1125                  * Fragment large datagrams such that each segment
1126                  * contains a multiple of PAGE_SIZE amount of data,
1127                  * plus headers. This enables a receiver to perform
1128                  * page-flipping zero-copy optimizations.
1129                  *
1130                  * XXX When does this help given that sender and receiver
1131                  * could have different page sizes, and also mtu could
1132                  * be less than the receiver's page size ?
1133                  */
1134                 int newlen;
1135                 struct mbuf *m;
1136
1137                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1138                         off += m->m_len;
1139
1140                 /*
1141                  * firstlen (off - hlen) must be aligned on an
1142                  * 8-byte boundary
1143                  */
1144                 if (off < hlen)
1145                         goto smart_frag_failure;
1146                 off = ((off - hlen) & ~7) + hlen;
1147                 newlen = (~PAGE_MASK) & mtu;
1148                 if ((newlen + sizeof(struct ip)) > mtu) {
1149                         /* we failed, go back the default */
1150 smart_frag_failure:
1151                         newlen = len;
1152                         off = hlen + len;
1153                 }
1154                 len = newlen;
1155
1156         } else {
1157                 off = hlen + len;
1158         }
1159
1160         firstlen = off - hlen;
1161         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1162
1163         /*
1164          * Loop through length of segment after first fragment,
1165          * make new header and copy data of each part and link onto chain.
1166          * Here, m0 is the original packet, m is the fragment being created.
1167          * The fragments are linked off the m_nextpkt of the original
1168          * packet, which after processing serves as the first fragment.
1169          */
1170         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1171                 struct ip *mhip;        /* ip header on the fragment */
1172                 struct mbuf *m;
1173                 int mhlen = sizeof(struct ip);
1174
1175                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1176                 if (m == NULL) {
1177                         error = ENOBUFS;
1178                         ipstat.ips_odropped++;
1179                         goto done;
1180                 }
1181                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1182                 /*
1183                  * In the first mbuf, leave room for the link header, then
1184                  * copy the original IP header including options. The payload
1185                  * goes into an additional mbuf chain returned by m_copy().
1186                  */
1187                 m->m_data += max_linkhdr;
1188                 mhip = mtod(m, struct ip *);
1189                 *mhip = *ip;
1190                 if (hlen > sizeof(struct ip)) {
1191                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1192                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1193                 }
1194                 m->m_len = mhlen;
1195                 /* XXX do we need to add ip->ip_off below ? */
1196                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1197                 if (off + len >= ip->ip_len) {  /* last fragment */
1198                         len = ip->ip_len - off;
1199                         m->m_flags |= M_LASTFRAG;
1200                 } else
1201                         mhip->ip_off |= IP_MF;
1202                 mhip->ip_len = htons((u_short)(len + mhlen));
1203                 m->m_next = m_copy(m0, off, len);
1204                 if (m->m_next == NULL) {                /* copy failed */
1205                         m_free(m);
1206                         error = ENOBUFS;        /* ??? */
1207                         ipstat.ips_odropped++;
1208                         goto done;
1209                 }
1210                 m->m_pkthdr.len = mhlen + len;
1211                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1212                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1213                 mhip->ip_off = htons(mhip->ip_off);
1214                 mhip->ip_sum = 0;
1215                 if (sw_csum & CSUM_DELAY_IP)
1216                         mhip->ip_sum = in_cksum(m, mhlen);
1217                 *mnext = m;
1218                 mnext = &m->m_nextpkt;
1219         }
1220         ipstat.ips_ofragments += nfrags;
1221
1222         /* set first marker for fragment chain */
1223         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1224         m0->m_pkthdr.csum_data = nfrags;
1225
1226         /*
1227          * Update first fragment by trimming what's been copied out
1228          * and updating header.
1229          */
1230         m_adj(m0, hlen + firstlen - ip->ip_len);
1231         m0->m_pkthdr.len = hlen + firstlen;
1232         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1233         ip->ip_off |= IP_MF;
1234         ip->ip_off = htons(ip->ip_off);
1235         ip->ip_sum = 0;
1236         if (sw_csum & CSUM_DELAY_IP)
1237                 ip->ip_sum = in_cksum(m0, hlen);
1238
1239 done:
1240         *m_frag = m0;
1241         return error;
1242 }
1243
1244 void
1245 in_delayed_cksum(struct mbuf *m)
1246 {
1247         struct ip *ip;
1248         u_short csum, offset;
1249
1250         ip = mtod(m, struct ip *);
1251         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1252         csum = in_cksum_skip(m, ip->ip_len, offset);
1253         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1254                 csum = 0xffff;
1255         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1256
1257         if (offset + sizeof(u_short) > m->m_len) {
1258                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1259                     m->m_len, offset, ip->ip_p);
1260                 /*
1261                  * XXX
1262                  * this shouldn't happen, but if it does, the
1263                  * correct behavior may be to insert the checksum
1264                  * in the existing chain instead of rearranging it.
1265                  */
1266                 m = m_pullup(m, offset + sizeof(u_short));
1267         }
1268         *(u_short *)(m->m_data + offset) = csum;
1269 }
1270
1271 /*
1272  * Insert IP options into preformed packet.
1273  * Adjust IP destination as required for IP source routing,
1274  * as indicated by a non-zero in_addr at the start of the options.
1275  *
1276  * XXX This routine assumes that the packet has no options in place.
1277  */
1278 static struct mbuf *
1279 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1280 {
1281         struct ipoption *p = mtod(opt, struct ipoption *);
1282         struct mbuf *n;
1283         struct ip *ip = mtod(m, struct ip *);
1284         unsigned optlen;
1285
1286         optlen = opt->m_len - sizeof p->ipopt_dst;
1287         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1288                 *phlen = 0;
1289                 return (m);             /* XXX should fail */
1290         }
1291         if (p->ipopt_dst.s_addr)
1292                 ip->ip_dst = p->ipopt_dst;
1293         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1294                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1295                 if (n == NULL) {
1296                         *phlen = 0;
1297                         return (m);
1298                 }
1299                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1300                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1301                 m->m_len -= sizeof(struct ip);
1302                 m->m_data += sizeof(struct ip);
1303                 n->m_next = m;
1304                 m = n;
1305                 m->m_len = optlen + sizeof(struct ip);
1306                 m->m_data += max_linkhdr;
1307                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1308         } else {
1309                 m->m_data -= optlen;
1310                 m->m_len += optlen;
1311                 m->m_pkthdr.len += optlen;
1312                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1313         }
1314         ip = mtod(m, struct ip *);
1315         bcopy(p->ipopt_list, ip + 1, optlen);
1316         *phlen = sizeof(struct ip) + optlen;
1317         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1318         ip->ip_len += optlen;
1319         return (m);
1320 }
1321
1322 /*
1323  * Copy options from ip to jp,
1324  * omitting those not copied during fragmentation.
1325  */
1326 int
1327 ip_optcopy(struct ip *ip, struct ip *jp)
1328 {
1329         u_char *cp, *dp;
1330         int opt, optlen, cnt;
1331
1332         cp = (u_char *)(ip + 1);
1333         dp = (u_char *)(jp + 1);
1334         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1335         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1336                 opt = cp[0];
1337                 if (opt == IPOPT_EOL)
1338                         break;
1339                 if (opt == IPOPT_NOP) {
1340                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1341                         *dp++ = IPOPT_NOP;
1342                         optlen = 1;
1343                         continue;
1344                 }
1345
1346                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1347                     ("ip_optcopy: malformed ipv4 option"));
1348                 optlen = cp[IPOPT_OLEN];
1349                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1350                     ("ip_optcopy: malformed ipv4 option"));
1351
1352                 /* bogus lengths should have been caught by ip_dooptions */
1353                 if (optlen > cnt)
1354                         optlen = cnt;
1355                 if (IPOPT_COPIED(opt)) {
1356                         bcopy(cp, dp, optlen);
1357                         dp += optlen;
1358                 }
1359         }
1360         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1361                 *dp++ = IPOPT_EOL;
1362         return (optlen);
1363 }
1364
1365 /*
1366  * IP socket option processing.
1367  */
1368 int
1369 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1370 {
1371         struct  inpcb *inp = so->so_pcb;
1372         int     error, optval;
1373
1374         error = optval = 0;
1375         if (sopt->sopt_level != IPPROTO_IP) {
1376                 return (EINVAL);
1377         }
1378
1379         switch (sopt->sopt_dir) {
1380         case SOPT_SET:
1381                 switch (sopt->sopt_name) {
1382                 case IP_OPTIONS:
1383 #ifdef notyet
1384                 case IP_RETOPTS:
1385 #endif
1386                 {
1387                         struct mbuf *m;
1388                         if (sopt->sopt_valsize > MLEN) {
1389                                 error = EMSGSIZE;
1390                                 break;
1391                         }
1392                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1393                         if (m == NULL) {
1394                                 error = ENOBUFS;
1395                                 break;
1396                         }
1397                         m->m_len = sopt->sopt_valsize;
1398                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1399                                             m->m_len);
1400
1401                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1402                                            m));
1403                 }
1404
1405                 case IP_TOS:
1406                 case IP_TTL:
1407                 case IP_RECVOPTS:
1408                 case IP_RECVRETOPTS:
1409                 case IP_RECVDSTADDR:
1410                 case IP_RECVIF:
1411                 case IP_FAITH:
1412                         error = sooptcopyin(sopt, &optval, sizeof optval,
1413                                             sizeof optval);
1414                         if (error)
1415                                 break;
1416
1417                         switch (sopt->sopt_name) {
1418                         case IP_TOS:
1419                                 inp->inp_ip_tos = optval;
1420                                 break;
1421
1422                         case IP_TTL:
1423                                 inp->inp_ip_ttl = optval;
1424                                 break;
1425 #define OPTSET(bit) \
1426         if (optval) \
1427                 inp->inp_flags |= bit; \
1428         else \
1429                 inp->inp_flags &= ~bit;
1430
1431                         case IP_RECVOPTS:
1432                                 OPTSET(INP_RECVOPTS);
1433                                 break;
1434
1435                         case IP_RECVRETOPTS:
1436                                 OPTSET(INP_RECVRETOPTS);
1437                                 break;
1438
1439                         case IP_RECVDSTADDR:
1440                                 OPTSET(INP_RECVDSTADDR);
1441                                 break;
1442
1443                         case IP_RECVIF:
1444                                 OPTSET(INP_RECVIF);
1445                                 break;
1446
1447                         case IP_FAITH:
1448                                 OPTSET(INP_FAITH);
1449                                 break;
1450                         }
1451                         break;
1452 #undef OPTSET
1453
1454                 case IP_MULTICAST_IF:
1455                 case IP_MULTICAST_VIF:
1456                 case IP_MULTICAST_TTL:
1457                 case IP_MULTICAST_LOOP:
1458                 case IP_ADD_MEMBERSHIP:
1459                 case IP_DROP_MEMBERSHIP:
1460                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1461                         break;
1462
1463                 case IP_PORTRANGE:
1464                         error = sooptcopyin(sopt, &optval, sizeof optval,
1465                                             sizeof optval);
1466                         if (error)
1467                                 break;
1468
1469                         switch (optval) {
1470                         case IP_PORTRANGE_DEFAULT:
1471                                 inp->inp_flags &= ~(INP_LOWPORT);
1472                                 inp->inp_flags &= ~(INP_HIGHPORT);
1473                                 break;
1474
1475                         case IP_PORTRANGE_HIGH:
1476                                 inp->inp_flags &= ~(INP_LOWPORT);
1477                                 inp->inp_flags |= INP_HIGHPORT;
1478                                 break;
1479
1480                         case IP_PORTRANGE_LOW:
1481                                 inp->inp_flags &= ~(INP_HIGHPORT);
1482                                 inp->inp_flags |= INP_LOWPORT;
1483                                 break;
1484
1485                         default:
1486                                 error = EINVAL;
1487                                 break;
1488                         }
1489                         break;
1490
1491 #if defined(IPSEC) || defined(FAST_IPSEC)
1492                 case IP_IPSEC_POLICY:
1493                 {
1494                         caddr_t req;
1495                         size_t len = 0;
1496                         int priv;
1497                         struct mbuf *m;
1498                         int optname;
1499
1500                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1501                                 break;
1502                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1503                                 break;
1504                         priv = (sopt->sopt_td != NULL &&
1505                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1506                         req = mtod(m, caddr_t);
1507                         len = m->m_len;
1508                         optname = sopt->sopt_name;
1509                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1510                         m_freem(m);
1511                         break;
1512                 }
1513 #endif /*IPSEC*/
1514
1515                 default:
1516                         error = ENOPROTOOPT;
1517                         break;
1518                 }
1519                 break;
1520
1521         case SOPT_GET:
1522                 switch (sopt->sopt_name) {
1523                 case IP_OPTIONS:
1524                 case IP_RETOPTS:
1525                         if (inp->inp_options)
1526                                 error = sooptcopyout(sopt,
1527                                                      mtod(inp->inp_options,
1528                                                           char *),
1529                                                      inp->inp_options->m_len);
1530                         else
1531                                 sopt->sopt_valsize = 0;
1532                         break;
1533
1534                 case IP_TOS:
1535                 case IP_TTL:
1536                 case IP_RECVOPTS:
1537                 case IP_RECVRETOPTS:
1538                 case IP_RECVDSTADDR:
1539                 case IP_RECVIF:
1540                 case IP_PORTRANGE:
1541                 case IP_FAITH:
1542                         switch (sopt->sopt_name) {
1543
1544                         case IP_TOS:
1545                                 optval = inp->inp_ip_tos;
1546                                 break;
1547
1548                         case IP_TTL:
1549                                 optval = inp->inp_ip_ttl;
1550                                 break;
1551
1552 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1553
1554                         case IP_RECVOPTS:
1555                                 optval = OPTBIT(INP_RECVOPTS);
1556                                 break;
1557
1558                         case IP_RECVRETOPTS:
1559                                 optval = OPTBIT(INP_RECVRETOPTS);
1560                                 break;
1561
1562                         case IP_RECVDSTADDR:
1563                                 optval = OPTBIT(INP_RECVDSTADDR);
1564                                 break;
1565
1566                         case IP_RECVIF:
1567                                 optval = OPTBIT(INP_RECVIF);
1568                                 break;
1569
1570                         case IP_PORTRANGE:
1571                                 if (inp->inp_flags & INP_HIGHPORT)
1572                                         optval = IP_PORTRANGE_HIGH;
1573                                 else if (inp->inp_flags & INP_LOWPORT)
1574                                         optval = IP_PORTRANGE_LOW;
1575                                 else
1576                                         optval = 0;
1577                                 break;
1578
1579                         case IP_FAITH:
1580                                 optval = OPTBIT(INP_FAITH);
1581                                 break;
1582                         }
1583                         error = sooptcopyout(sopt, &optval, sizeof optval);
1584                         break;
1585
1586                 case IP_MULTICAST_IF:
1587                 case IP_MULTICAST_VIF:
1588                 case IP_MULTICAST_TTL:
1589                 case IP_MULTICAST_LOOP:
1590                 case IP_ADD_MEMBERSHIP:
1591                 case IP_DROP_MEMBERSHIP:
1592                         error = ip_getmoptions(sopt, inp->inp_moptions);
1593                         break;
1594
1595 #if defined(IPSEC) || defined(FAST_IPSEC)
1596                 case IP_IPSEC_POLICY:
1597                 {
1598                         struct mbuf *m = NULL;
1599                         caddr_t req = NULL;
1600                         size_t len = 0;
1601
1602                         if (m != NULL) {
1603                                 req = mtod(m, caddr_t);
1604                                 len = m->m_len;
1605                         }
1606                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1607                         if (error == 0)
1608                                 error = soopt_mcopyout(sopt, m); /* XXX */
1609                         if (error == 0)
1610                                 m_freem(m);
1611                         break;
1612                 }
1613 #endif /*IPSEC*/
1614
1615                 default:
1616                         error = ENOPROTOOPT;
1617                         break;
1618                 }
1619                 break;
1620         }
1621         return (error);
1622 }
1623
1624 /*
1625  * Set up IP options in pcb for insertion in output packets.
1626  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1627  * with destination address if source routed.
1628  */
1629 static int
1630 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1631 {
1632         int cnt, optlen;
1633         u_char *cp;
1634         u_char opt;
1635
1636         /* turn off any old options */
1637         if (*pcbopt)
1638                 m_free(*pcbopt);
1639         *pcbopt = 0;
1640         if (m == NULL || m->m_len == 0) {
1641                 /*
1642                  * Only turning off any previous options.
1643                  */
1644                 if (m != NULL)
1645                         m_free(m);
1646                 return (0);
1647         }
1648
1649         if (m->m_len % sizeof(int32_t))
1650                 goto bad;
1651         /*
1652          * IP first-hop destination address will be stored before
1653          * actual options; move other options back
1654          * and clear it when none present.
1655          */
1656         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1657                 goto bad;
1658         cnt = m->m_len;
1659         m->m_len += sizeof(struct in_addr);
1660         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1661         ovbcopy(mtod(m, caddr_t), cp, cnt);
1662         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1663
1664         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1665                 opt = cp[IPOPT_OPTVAL];
1666                 if (opt == IPOPT_EOL)
1667                         break;
1668                 if (opt == IPOPT_NOP)
1669                         optlen = 1;
1670                 else {
1671                         if (cnt < IPOPT_OLEN + sizeof *cp)
1672                                 goto bad;
1673                         optlen = cp[IPOPT_OLEN];
1674                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1675                                 goto bad;
1676                 }
1677                 switch (opt) {
1678
1679                 default:
1680                         break;
1681
1682                 case IPOPT_LSRR:
1683                 case IPOPT_SSRR:
1684                         /*
1685                          * user process specifies route as:
1686                          *      ->A->B->C->D
1687                          * D must be our final destination (but we can't
1688                          * check that since we may not have connected yet).
1689                          * A is first hop destination, which doesn't appear in
1690                          * actual IP option, but is stored before the options.
1691                          */
1692                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1693                                 goto bad;
1694                         m->m_len -= sizeof(struct in_addr);
1695                         cnt -= sizeof(struct in_addr);
1696                         optlen -= sizeof(struct in_addr);
1697                         cp[IPOPT_OLEN] = optlen;
1698                         /*
1699                          * Move first hop before start of options.
1700                          */
1701                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1702                               sizeof(struct in_addr));
1703                         /*
1704                          * Then copy rest of options back
1705                          * to close up the deleted entry.
1706                          */
1707                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1708                                 &cp[IPOPT_OFFSET+1],
1709                                 cnt - (IPOPT_MINOFF - 1));
1710                         break;
1711                 }
1712         }
1713         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1714                 goto bad;
1715         *pcbopt = m;
1716         return (0);
1717
1718 bad:
1719         m_free(m);
1720         return (EINVAL);
1721 }
1722
1723 /*
1724  * XXX
1725  * The whole multicast option thing needs to be re-thought.
1726  * Several of these options are equally applicable to non-multicast
1727  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1728  * standard option (IP_TTL).
1729  */
1730
1731 /*
1732  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1733  */
1734 static struct ifnet *
1735 ip_multicast_if(struct in_addr *a, int *ifindexp)
1736 {
1737         int ifindex;
1738         struct ifnet *ifp;
1739
1740         if (ifindexp)
1741                 *ifindexp = 0;
1742         if (ntohl(a->s_addr) >> 24 == 0) {
1743                 ifindex = ntohl(a->s_addr) & 0xffffff;
1744                 if (ifindex < 0 || if_index < ifindex)
1745                         return NULL;
1746                 ifp = ifindex2ifnet[ifindex];
1747                 if (ifindexp)
1748                         *ifindexp = ifindex;
1749         } else {
1750                 INADDR_TO_IFP(*a, ifp);
1751         }
1752         return ifp;
1753 }
1754
1755 /*
1756  * Set the IP multicast options in response to user setsockopt().
1757  */
1758 static int
1759 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1760 {
1761         int error = 0;
1762         int i;
1763         struct in_addr addr;
1764         struct ip_mreq mreq;
1765         struct ifnet *ifp;
1766         struct ip_moptions *imo = *imop;
1767         int ifindex;
1768
1769         if (imo == NULL) {
1770                 /*
1771                  * No multicast option buffer attached to the pcb;
1772                  * allocate one and initialize to default values.
1773                  */
1774                 imo = malloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1775
1776                 if (imo == NULL)
1777                         return (ENOBUFS);
1778                 *imop = imo;
1779                 imo->imo_multicast_ifp = NULL;
1780                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1781                 imo->imo_multicast_vif = -1;
1782                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1783                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1784                 imo->imo_num_memberships = 0;
1785         }
1786
1787         switch (sopt->sopt_name) {
1788         /* store an index number for the vif you wanna use in the send */
1789         case IP_MULTICAST_VIF:
1790                 if (legal_vif_num == 0) {
1791                         error = EOPNOTSUPP;
1792                         break;
1793                 }
1794                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1795                 if (error)
1796                         break;
1797                 if (!legal_vif_num(i) && (i != -1)) {
1798                         error = EINVAL;
1799                         break;
1800                 }
1801                 imo->imo_multicast_vif = i;
1802                 break;
1803
1804         case IP_MULTICAST_IF:
1805                 /*
1806                  * Select the interface for outgoing multicast packets.
1807                  */
1808                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1809                 if (error)
1810                         break;
1811                 /*
1812                  * INADDR_ANY is used to remove a previous selection.
1813                  * When no interface is selected, a default one is
1814                  * chosen every time a multicast packet is sent.
1815                  */
1816                 if (addr.s_addr == INADDR_ANY) {
1817                         imo->imo_multicast_ifp = NULL;
1818                         break;
1819                 }
1820                 /*
1821                  * The selected interface is identified by its local
1822                  * IP address.  Find the interface and confirm that
1823                  * it supports multicasting.
1824                  */
1825                 crit_enter();
1826                 ifp = ip_multicast_if(&addr, &ifindex);
1827                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1828                         crit_exit();
1829                         error = EADDRNOTAVAIL;
1830                         break;
1831                 }
1832                 imo->imo_multicast_ifp = ifp;
1833                 if (ifindex)
1834                         imo->imo_multicast_addr = addr;
1835                 else
1836                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1837                 crit_exit();
1838                 break;
1839
1840         case IP_MULTICAST_TTL:
1841                 /*
1842                  * Set the IP time-to-live for outgoing multicast packets.
1843                  * The original multicast API required a char argument,
1844                  * which is inconsistent with the rest of the socket API.
1845                  * We allow either a char or an int.
1846                  */
1847                 if (sopt->sopt_valsize == 1) {
1848                         u_char ttl;
1849                         error = sooptcopyin(sopt, &ttl, 1, 1);
1850                         if (error)
1851                                 break;
1852                         imo->imo_multicast_ttl = ttl;
1853                 } else {
1854                         u_int ttl;
1855                         error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1856                         if (error)
1857                                 break;
1858                         if (ttl > 255)
1859                                 error = EINVAL;
1860                         else
1861                                 imo->imo_multicast_ttl = ttl;
1862                 }
1863                 break;
1864
1865         case IP_MULTICAST_LOOP:
1866                 /*
1867                  * Set the loopback flag for outgoing multicast packets.
1868                  * Must be zero or one.  The original multicast API required a
1869                  * char argument, which is inconsistent with the rest
1870                  * of the socket API.  We allow either a char or an int.
1871                  */
1872                 if (sopt->sopt_valsize == 1) {
1873                         u_char loop;
1874
1875                         error = sooptcopyin(sopt, &loop, 1, 1);
1876                         if (error)
1877                                 break;
1878                         imo->imo_multicast_loop = !!loop;
1879                 } else {
1880                         u_int loop;
1881
1882                         error = sooptcopyin(sopt, &loop, sizeof loop,
1883                                             sizeof loop);
1884                         if (error)
1885                                 break;
1886                         imo->imo_multicast_loop = !!loop;
1887                 }
1888                 break;
1889
1890         case IP_ADD_MEMBERSHIP:
1891                 /*
1892                  * Add a multicast group membership.
1893                  * Group must be a valid IP multicast address.
1894                  */
1895                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1896                 if (error)
1897                         break;
1898
1899                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1900                         error = EINVAL;
1901                         break;
1902                 }
1903                 crit_enter();
1904                 /*
1905                  * If no interface address was provided, use the interface of
1906                  * the route to the given multicast address.
1907                  */
1908                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1909                         struct sockaddr_in dst;
1910                         struct rtentry *rt;
1911
1912                         bzero(&dst, sizeof(struct sockaddr_in));
1913                         dst.sin_len = sizeof(struct sockaddr_in);
1914                         dst.sin_family = AF_INET;
1915                         dst.sin_addr = mreq.imr_multiaddr;
1916                         rt = rtlookup((struct sockaddr *)&dst);
1917                         if (rt == NULL) {
1918                                 error = EADDRNOTAVAIL;
1919                                 crit_exit();
1920                                 break;
1921                         }
1922                         --rt->rt_refcnt;
1923                         ifp = rt->rt_ifp;
1924                 } else {
1925                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1926                 }
1927
1928                 /*
1929                  * See if we found an interface, and confirm that it
1930                  * supports multicast.
1931                  */
1932                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1933                         error = EADDRNOTAVAIL;
1934                         crit_exit();
1935                         break;
1936                 }
1937                 /*
1938                  * See if the membership already exists or if all the
1939                  * membership slots are full.
1940                  */
1941                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1942                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1943                             imo->imo_membership[i]->inm_addr.s_addr
1944                                                 == mreq.imr_multiaddr.s_addr)
1945                                 break;
1946                 }
1947                 if (i < imo->imo_num_memberships) {
1948                         error = EADDRINUSE;
1949                         crit_exit();
1950                         break;
1951                 }
1952                 if (i == IP_MAX_MEMBERSHIPS) {
1953                         error = ETOOMANYREFS;
1954                         crit_exit();
1955                         break;
1956                 }
1957                 /*
1958                  * Everything looks good; add a new record to the multicast
1959                  * address list for the given interface.
1960                  */
1961                 if ((imo->imo_membership[i] =
1962                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1963                         error = ENOBUFS;
1964                         crit_exit();
1965                         break;
1966                 }
1967                 ++imo->imo_num_memberships;
1968                 crit_exit();
1969                 break;
1970
1971         case IP_DROP_MEMBERSHIP:
1972                 /*
1973                  * Drop a multicast group membership.
1974                  * Group must be a valid IP multicast address.
1975                  */
1976                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1977                 if (error)
1978                         break;
1979
1980                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1981                         error = EINVAL;
1982                         break;
1983                 }
1984
1985                 crit_enter();
1986                 /*
1987                  * If an interface address was specified, get a pointer
1988                  * to its ifnet structure.
1989                  */
1990                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1991                         ifp = NULL;
1992                 else {
1993                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1994                         if (ifp == NULL) {
1995                                 error = EADDRNOTAVAIL;
1996                                 crit_exit();
1997                                 break;
1998                         }
1999                 }
2000                 /*
2001                  * Find the membership in the membership array.
2002                  */
2003                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2004                         if ((ifp == NULL ||
2005                              imo->imo_membership[i]->inm_ifp == ifp) &&
2006                             imo->imo_membership[i]->inm_addr.s_addr ==
2007                             mreq.imr_multiaddr.s_addr)
2008                                 break;
2009                 }
2010                 if (i == imo->imo_num_memberships) {
2011                         error = EADDRNOTAVAIL;
2012                         crit_exit();
2013                         break;
2014                 }
2015                 /*
2016                  * Give up the multicast address record to which the
2017                  * membership points.
2018                  */
2019                 in_delmulti(imo->imo_membership[i]);
2020                 /*
2021                  * Remove the gap in the membership array.
2022                  */
2023                 for (++i; i < imo->imo_num_memberships; ++i)
2024                         imo->imo_membership[i-1] = imo->imo_membership[i];
2025                 --imo->imo_num_memberships;
2026                 crit_exit();
2027                 break;
2028
2029         default:
2030                 error = EOPNOTSUPP;
2031                 break;
2032         }
2033
2034         /*
2035          * If all options have default values, no need to keep the mbuf.
2036          */
2037         if (imo->imo_multicast_ifp == NULL &&
2038             imo->imo_multicast_vif == -1 &&
2039             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2040             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2041             imo->imo_num_memberships == 0) {
2042                 free(*imop, M_IPMOPTS);
2043                 *imop = NULL;
2044         }
2045
2046         return (error);
2047 }
2048
2049 /*
2050  * Return the IP multicast options in response to user getsockopt().
2051  */
2052 static int
2053 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2054 {
2055         struct in_addr addr;
2056         struct in_ifaddr *ia;
2057         int error, optval;
2058         u_char coptval;
2059
2060         error = 0;
2061         switch (sopt->sopt_name) {
2062         case IP_MULTICAST_VIF:
2063                 if (imo != NULL)
2064                         optval = imo->imo_multicast_vif;
2065                 else
2066                         optval = -1;
2067                 error = sooptcopyout(sopt, &optval, sizeof optval);
2068                 break;
2069
2070         case IP_MULTICAST_IF:
2071                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2072                         addr.s_addr = INADDR_ANY;
2073                 else if (imo->imo_multicast_addr.s_addr) {
2074                         /* return the value user has set */
2075                         addr = imo->imo_multicast_addr;
2076                 } else {
2077                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2078                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2079                                 : IA_SIN(ia)->sin_addr.s_addr;
2080                 }
2081                 error = sooptcopyout(sopt, &addr, sizeof addr);
2082                 break;
2083
2084         case IP_MULTICAST_TTL:
2085                 if (imo == NULL)
2086                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2087                 else
2088                         optval = coptval = imo->imo_multicast_ttl;
2089                 if (sopt->sopt_valsize == 1)
2090                         error = sooptcopyout(sopt, &coptval, 1);
2091                 else
2092                         error = sooptcopyout(sopt, &optval, sizeof optval);
2093                 break;
2094
2095         case IP_MULTICAST_LOOP:
2096                 if (imo == NULL)
2097                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2098                 else
2099                         optval = coptval = imo->imo_multicast_loop;
2100                 if (sopt->sopt_valsize == 1)
2101                         error = sooptcopyout(sopt, &coptval, 1);
2102                 else
2103                         error = sooptcopyout(sopt, &optval, sizeof optval);
2104                 break;
2105
2106         default:
2107                 error = ENOPROTOOPT;
2108                 break;
2109         }
2110         return (error);
2111 }
2112
2113 /*
2114  * Discard the IP multicast options.
2115  */
2116 void
2117 ip_freemoptions(struct ip_moptions *imo)
2118 {
2119         int i;
2120
2121         if (imo != NULL) {
2122                 for (i = 0; i < imo->imo_num_memberships; ++i)
2123                         in_delmulti(imo->imo_membership[i]);
2124                 free(imo, M_IPMOPTS);
2125         }
2126 }
2127
2128 /*
2129  * Routine called from ip_output() to loop back a copy of an IP multicast
2130  * packet to the input queue of a specified interface.  Note that this
2131  * calls the output routine of the loopback "driver", but with an interface
2132  * pointer that might NOT be a loopback interface -- evil, but easier than
2133  * replicating that code here.
2134  */
2135 static void
2136 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2137              int hlen)
2138 {
2139         struct ip *ip;
2140         struct mbuf *copym;
2141
2142         copym = m_copypacket(m, MB_DONTWAIT);
2143         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2144                 copym = m_pullup(copym, hlen);
2145         if (copym != NULL) {
2146                 /*
2147                  * if the checksum hasn't been computed, mark it as valid
2148                  */
2149                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2150                         in_delayed_cksum(copym);
2151                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2152                         copym->m_pkthdr.csum_flags |=
2153                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2154                         copym->m_pkthdr.csum_data = 0xffff;
2155                 }
2156                 /*
2157                  * We don't bother to fragment if the IP length is greater
2158                  * than the interface's MTU.  Can this possibly matter?
2159                  */
2160                 ip = mtod(copym, struct ip *);
2161                 ip->ip_len = htons(ip->ip_len);
2162                 ip->ip_off = htons(ip->ip_off);
2163                 ip->ip_sum = 0;
2164                 if (ip->ip_vhl == IP_VHL_BORING) {
2165                         ip->ip_sum = in_cksum_hdr(ip);
2166                 } else {
2167                         ip->ip_sum = in_cksum(copym, hlen);
2168                 }
2169                 /*
2170                  * NB:
2171                  * It's not clear whether there are any lingering
2172                  * reentrancy problems in other areas which might
2173                  * be exposed by using ip_input directly (in
2174                  * particular, everything which modifies the packet
2175                  * in-place).  Yet another option is using the
2176                  * protosw directly to deliver the looped back
2177                  * packet.  For the moment, we'll err on the side
2178                  * of safety by using if_simloop().
2179                  */
2180 #if 1 /* XXX */
2181                 if (dst->sin_family != AF_INET) {
2182                         printf("ip_mloopback: bad address family %d\n",
2183                                                 dst->sin_family);
2184                         dst->sin_family = AF_INET;
2185                 }
2186 #endif
2187
2188 #ifdef notdef
2189                 copym->m_pkthdr.rcvif = ifp;
2190                 ip_input(copym);
2191 #else
2192                 if_simloop(ifp, copym, dst->sin_family, 0);
2193 #endif
2194         }
2195 }