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 1.37 2007/04/04 06:13:26 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)        kprintf("%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                         kprintf("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                 kprintf("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                         kprintf("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                         kprintf("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                                         m->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                 lwkt_serialize_enter(ifp->if_serializer);
1006                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1007                                           ro->ro_rt);
1008                 lwkt_serialize_exit(ifp->if_serializer);
1009                 goto done;
1010         }
1011
1012         if (ip->ip_off & IP_DF) {
1013                 error = EMSGSIZE;
1014                 /*
1015                  * This case can happen if the user changed the MTU
1016                  * of an interface after enabling IP on it.  Because
1017                  * most netifs don't keep track of routes pointing to
1018                  * them, there is no way for one to update all its
1019                  * routes when the MTU is changed.
1020                  */
1021                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1022                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1023                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1024                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1025                 }
1026                 ipstat.ips_cantfrag++;
1027                 goto bad;
1028         }
1029
1030         /*
1031          * Too large for interface; fragment if possible. If successful,
1032          * on return, m will point to a list of packets to be sent.
1033          */
1034         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1035         if (error)
1036                 goto bad;
1037         for (; m; m = m0) {
1038                 m0 = m->m_nextpkt;
1039                 m->m_nextpkt = NULL;
1040 #ifdef IPSEC
1041                 /* clean ipsec history once it goes out of the node */
1042                 ipsec_delaux(m);
1043 #endif
1044                 if (error == 0) {
1045                         /* Record statistics for this interface address. */
1046                         if (ia != NULL) {
1047                                 ia->ia_ifa.if_opackets++;
1048                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1049                         }
1050                         lwkt_serialize_enter(ifp->if_serializer);
1051                         error = (*ifp->if_output)(ifp, m,
1052                                                   (struct sockaddr *)dst,
1053                                                   ro->ro_rt);
1054                         lwkt_serialize_exit(ifp->if_serializer);
1055                 } else {
1056                         m_freem(m);
1057                 }
1058         }
1059
1060         if (error == 0)
1061                 ipstat.ips_fragmented++;
1062
1063 done:
1064         if (ro == &iproute && ro->ro_rt != NULL) {
1065                 RTFREE(ro->ro_rt);
1066                 ro->ro_rt = NULL;
1067         }
1068 #ifdef IPSEC
1069         if (sp != NULL) {
1070                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1071                         kprintf("DP ip_output call free SP:%p\n", sp));
1072                 key_freesp(sp);
1073         }
1074 #endif
1075 #ifdef FAST_IPSEC
1076         if (sp != NULL)
1077                 KEY_FREESP(&sp);
1078 #endif
1079         return (error);
1080 bad:
1081         m_freem(m);
1082         goto done;
1083 }
1084
1085 /*
1086  * Create a chain of fragments which fit the given mtu. m_frag points to the
1087  * mbuf to be fragmented; on return it points to the chain with the fragments.
1088  * Return 0 if no error. If error, m_frag may contain a partially built
1089  * chain of fragments that should be freed by the caller.
1090  *
1091  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1092  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1093  */
1094 int
1095 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1096             u_long if_hwassist_flags, int sw_csum)
1097 {
1098         int error = 0;
1099         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1100         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1101         int off;
1102         struct mbuf *m0 = *m_frag;      /* the original packet          */
1103         int firstlen;
1104         struct mbuf **mnext;
1105         int nfrags;
1106
1107         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1108                 ipstat.ips_cantfrag++;
1109                 return EMSGSIZE;
1110         }
1111
1112         /*
1113          * Must be able to put at least 8 bytes per fragment.
1114          */
1115         if (len < 8)
1116                 return EMSGSIZE;
1117
1118         /*
1119          * If the interface will not calculate checksums on
1120          * fragmented packets, then do it here.
1121          */
1122         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1123             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1124                 in_delayed_cksum(m0);
1125                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1126         }
1127
1128         if (len > PAGE_SIZE) {
1129                 /*
1130                  * Fragment large datagrams such that each segment
1131                  * contains a multiple of PAGE_SIZE amount of data,
1132                  * plus headers. This enables a receiver to perform
1133                  * page-flipping zero-copy optimizations.
1134                  *
1135                  * XXX When does this help given that sender and receiver
1136                  * could have different page sizes, and also mtu could
1137                  * be less than the receiver's page size ?
1138                  */
1139                 int newlen;
1140                 struct mbuf *m;
1141
1142                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1143                         off += m->m_len;
1144
1145                 /*
1146                  * firstlen (off - hlen) must be aligned on an
1147                  * 8-byte boundary
1148                  */
1149                 if (off < hlen)
1150                         goto smart_frag_failure;
1151                 off = ((off - hlen) & ~7) + hlen;
1152                 newlen = (~PAGE_MASK) & mtu;
1153                 if ((newlen + sizeof(struct ip)) > mtu) {
1154                         /* we failed, go back the default */
1155 smart_frag_failure:
1156                         newlen = len;
1157                         off = hlen + len;
1158                 }
1159                 len = newlen;
1160
1161         } else {
1162                 off = hlen + len;
1163         }
1164
1165         firstlen = off - hlen;
1166         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1167
1168         /*
1169          * Loop through length of segment after first fragment,
1170          * make new header and copy data of each part and link onto chain.
1171          * Here, m0 is the original packet, m is the fragment being created.
1172          * The fragments are linked off the m_nextpkt of the original
1173          * packet, which after processing serves as the first fragment.
1174          */
1175         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1176                 struct ip *mhip;        /* ip header on the fragment */
1177                 struct mbuf *m;
1178                 int mhlen = sizeof(struct ip);
1179
1180                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1181                 if (m == NULL) {
1182                         error = ENOBUFS;
1183                         ipstat.ips_odropped++;
1184                         goto done;
1185                 }
1186                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1187                 /*
1188                  * In the first mbuf, leave room for the link header, then
1189                  * copy the original IP header including options. The payload
1190                  * goes into an additional mbuf chain returned by m_copy().
1191                  */
1192                 m->m_data += max_linkhdr;
1193                 mhip = mtod(m, struct ip *);
1194                 *mhip = *ip;
1195                 if (hlen > sizeof(struct ip)) {
1196                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1197                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1198                 }
1199                 m->m_len = mhlen;
1200                 /* XXX do we need to add ip->ip_off below ? */
1201                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1202                 if (off + len >= ip->ip_len) {  /* last fragment */
1203                         len = ip->ip_len - off;
1204                         m->m_flags |= M_LASTFRAG;
1205                 } else
1206                         mhip->ip_off |= IP_MF;
1207                 mhip->ip_len = htons((u_short)(len + mhlen));
1208                 m->m_next = m_copy(m0, off, len);
1209                 if (m->m_next == NULL) {                /* copy failed */
1210                         m_free(m);
1211                         error = ENOBUFS;        /* ??? */
1212                         ipstat.ips_odropped++;
1213                         goto done;
1214                 }
1215                 m->m_pkthdr.len = mhlen + len;
1216                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1217                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1218                 mhip->ip_off = htons(mhip->ip_off);
1219                 mhip->ip_sum = 0;
1220                 if (sw_csum & CSUM_DELAY_IP)
1221                         mhip->ip_sum = in_cksum(m, mhlen);
1222                 *mnext = m;
1223                 mnext = &m->m_nextpkt;
1224         }
1225         ipstat.ips_ofragments += nfrags;
1226
1227         /* set first marker for fragment chain */
1228         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1229         m0->m_pkthdr.csum_data = nfrags;
1230
1231         /*
1232          * Update first fragment by trimming what's been copied out
1233          * and updating header.
1234          */
1235         m_adj(m0, hlen + firstlen - ip->ip_len);
1236         m0->m_pkthdr.len = hlen + firstlen;
1237         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1238         ip->ip_off |= IP_MF;
1239         ip->ip_off = htons(ip->ip_off);
1240         ip->ip_sum = 0;
1241         if (sw_csum & CSUM_DELAY_IP)
1242                 ip->ip_sum = in_cksum(m0, hlen);
1243
1244 done:
1245         *m_frag = m0;
1246         return error;
1247 }
1248
1249 void
1250 in_delayed_cksum(struct mbuf *m)
1251 {
1252         struct ip *ip;
1253         u_short csum, offset;
1254
1255         ip = mtod(m, struct ip *);
1256         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1257         csum = in_cksum_skip(m, ip->ip_len, offset);
1258         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1259                 csum = 0xffff;
1260         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1261
1262         if (offset + sizeof(u_short) > m->m_len) {
1263                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1264                     m->m_len, offset, ip->ip_p);
1265                 /*
1266                  * XXX
1267                  * this shouldn't happen, but if it does, the
1268                  * correct behavior may be to insert the checksum
1269                  * in the existing chain instead of rearranging it.
1270                  */
1271                 m = m_pullup(m, offset + sizeof(u_short));
1272         }
1273         *(u_short *)(m->m_data + offset) = csum;
1274 }
1275
1276 /*
1277  * Insert IP options into preformed packet.
1278  * Adjust IP destination as required for IP source routing,
1279  * as indicated by a non-zero in_addr at the start of the options.
1280  *
1281  * XXX This routine assumes that the packet has no options in place.
1282  */
1283 static struct mbuf *
1284 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1285 {
1286         struct ipoption *p = mtod(opt, struct ipoption *);
1287         struct mbuf *n;
1288         struct ip *ip = mtod(m, struct ip *);
1289         unsigned optlen;
1290
1291         optlen = opt->m_len - sizeof p->ipopt_dst;
1292         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1293                 *phlen = 0;
1294                 return (m);             /* XXX should fail */
1295         }
1296         if (p->ipopt_dst.s_addr)
1297                 ip->ip_dst = p->ipopt_dst;
1298         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1299                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1300                 if (n == NULL) {
1301                         *phlen = 0;
1302                         return (m);
1303                 }
1304                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1305                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1306                 m->m_len -= sizeof(struct ip);
1307                 m->m_data += sizeof(struct ip);
1308                 n->m_next = m;
1309                 m = n;
1310                 m->m_len = optlen + sizeof(struct ip);
1311                 m->m_data += max_linkhdr;
1312                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1313         } else {
1314                 m->m_data -= optlen;
1315                 m->m_len += optlen;
1316                 m->m_pkthdr.len += optlen;
1317                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1318         }
1319         ip = mtod(m, struct ip *);
1320         bcopy(p->ipopt_list, ip + 1, optlen);
1321         *phlen = sizeof(struct ip) + optlen;
1322         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1323         ip->ip_len += optlen;
1324         return (m);
1325 }
1326
1327 /*
1328  * Copy options from ip to jp,
1329  * omitting those not copied during fragmentation.
1330  */
1331 int
1332 ip_optcopy(struct ip *ip, struct ip *jp)
1333 {
1334         u_char *cp, *dp;
1335         int opt, optlen, cnt;
1336
1337         cp = (u_char *)(ip + 1);
1338         dp = (u_char *)(jp + 1);
1339         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1340         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1341                 opt = cp[0];
1342                 if (opt == IPOPT_EOL)
1343                         break;
1344                 if (opt == IPOPT_NOP) {
1345                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1346                         *dp++ = IPOPT_NOP;
1347                         optlen = 1;
1348                         continue;
1349                 }
1350
1351                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1352                     ("ip_optcopy: malformed ipv4 option"));
1353                 optlen = cp[IPOPT_OLEN];
1354                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1355                     ("ip_optcopy: malformed ipv4 option"));
1356
1357                 /* bogus lengths should have been caught by ip_dooptions */
1358                 if (optlen > cnt)
1359                         optlen = cnt;
1360                 if (IPOPT_COPIED(opt)) {
1361                         bcopy(cp, dp, optlen);
1362                         dp += optlen;
1363                 }
1364         }
1365         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1366                 *dp++ = IPOPT_EOL;
1367         return (optlen);
1368 }
1369
1370 /*
1371  * IP socket option processing.
1372  */
1373 int
1374 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1375 {
1376         struct  inpcb *inp = so->so_pcb;
1377         int     error, optval;
1378
1379         error = optval = 0;
1380         if (sopt->sopt_level != IPPROTO_IP) {
1381                 return (EINVAL);
1382         }
1383
1384         switch (sopt->sopt_dir) {
1385         case SOPT_SET:
1386                 switch (sopt->sopt_name) {
1387                 case IP_OPTIONS:
1388 #ifdef notyet
1389                 case IP_RETOPTS:
1390 #endif
1391                 {
1392                         struct mbuf *m;
1393                         if (sopt->sopt_valsize > MLEN) {
1394                                 error = EMSGSIZE;
1395                                 break;
1396                         }
1397                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1398                         if (m == NULL) {
1399                                 error = ENOBUFS;
1400                                 break;
1401                         }
1402                         m->m_len = sopt->sopt_valsize;
1403                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1404                                             m->m_len);
1405
1406                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1407                                            m));
1408                 }
1409
1410                 case IP_TOS:
1411                 case IP_TTL:
1412                 case IP_MINTTL:
1413                 case IP_RECVOPTS:
1414                 case IP_RECVRETOPTS:
1415                 case IP_RECVDSTADDR:
1416                 case IP_RECVIF:
1417                 case IP_RECVTTL:
1418                 case IP_FAITH:
1419                         error = sooptcopyin(sopt, &optval, sizeof optval,
1420                                             sizeof optval);
1421                         if (error)
1422                                 break;
1423
1424                         switch (sopt->sopt_name) {
1425                         case IP_TOS:
1426                                 inp->inp_ip_tos = optval;
1427                                 break;
1428
1429                         case IP_TTL:
1430                                 inp->inp_ip_ttl = optval;
1431                                 break;
1432                         case IP_MINTTL:
1433                                 if (optval > 0 && optval <= MAXTTL)
1434                                         inp->inp_ip_minttl = optval;
1435                                 else
1436                                         error = EINVAL;
1437                                 break;
1438 #define OPTSET(bit) \
1439         if (optval) \
1440                 inp->inp_flags |= bit; \
1441         else \
1442                 inp->inp_flags &= ~bit;
1443
1444                         case IP_RECVOPTS:
1445                                 OPTSET(INP_RECVOPTS);
1446                                 break;
1447
1448                         case IP_RECVRETOPTS:
1449                                 OPTSET(INP_RECVRETOPTS);
1450                                 break;
1451
1452                         case IP_RECVDSTADDR:
1453                                 OPTSET(INP_RECVDSTADDR);
1454                                 break;
1455
1456                         case IP_RECVIF:
1457                                 OPTSET(INP_RECVIF);
1458                                 break;
1459
1460                         case IP_RECVTTL:
1461                                 OPTSET(INP_RECVTTL);
1462                                 break;
1463
1464                         case IP_FAITH:
1465                                 OPTSET(INP_FAITH);
1466                                 break;
1467                         }
1468                         break;
1469 #undef OPTSET
1470
1471                 case IP_MULTICAST_IF:
1472                 case IP_MULTICAST_VIF:
1473                 case IP_MULTICAST_TTL:
1474                 case IP_MULTICAST_LOOP:
1475                 case IP_ADD_MEMBERSHIP:
1476                 case IP_DROP_MEMBERSHIP:
1477                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1478                         break;
1479
1480                 case IP_PORTRANGE:
1481                         error = sooptcopyin(sopt, &optval, sizeof optval,
1482                                             sizeof optval);
1483                         if (error)
1484                                 break;
1485
1486                         switch (optval) {
1487                         case IP_PORTRANGE_DEFAULT:
1488                                 inp->inp_flags &= ~(INP_LOWPORT);
1489                                 inp->inp_flags &= ~(INP_HIGHPORT);
1490                                 break;
1491
1492                         case IP_PORTRANGE_HIGH:
1493                                 inp->inp_flags &= ~(INP_LOWPORT);
1494                                 inp->inp_flags |= INP_HIGHPORT;
1495                                 break;
1496
1497                         case IP_PORTRANGE_LOW:
1498                                 inp->inp_flags &= ~(INP_HIGHPORT);
1499                                 inp->inp_flags |= INP_LOWPORT;
1500                                 break;
1501
1502                         default:
1503                                 error = EINVAL;
1504                                 break;
1505                         }
1506                         break;
1507
1508 #if defined(IPSEC) || defined(FAST_IPSEC)
1509                 case IP_IPSEC_POLICY:
1510                 {
1511                         caddr_t req;
1512                         size_t len = 0;
1513                         int priv;
1514                         struct mbuf *m;
1515                         int optname;
1516
1517                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1518                                 break;
1519                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1520                                 break;
1521                         priv = (sopt->sopt_td != NULL &&
1522                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1523                         req = mtod(m, caddr_t);
1524                         len = m->m_len;
1525                         optname = sopt->sopt_name;
1526                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1527                         m_freem(m);
1528                         break;
1529                 }
1530 #endif /*IPSEC*/
1531
1532                 default:
1533                         error = ENOPROTOOPT;
1534                         break;
1535                 }
1536                 break;
1537
1538         case SOPT_GET:
1539                 switch (sopt->sopt_name) {
1540                 case IP_OPTIONS:
1541                 case IP_RETOPTS:
1542                         if (inp->inp_options)
1543                                 error = sooptcopyout(sopt,
1544                                                      mtod(inp->inp_options,
1545                                                           char *),
1546                                                      inp->inp_options->m_len);
1547                         else
1548                                 sopt->sopt_valsize = 0;
1549                         break;
1550
1551                 case IP_TOS:
1552                 case IP_TTL:
1553                 case IP_MINTTL:
1554                 case IP_RECVOPTS:
1555                 case IP_RECVRETOPTS:
1556                 case IP_RECVDSTADDR:
1557                 case IP_RECVTTL:
1558                 case IP_RECVIF:
1559                 case IP_PORTRANGE:
1560                 case IP_FAITH:
1561                         switch (sopt->sopt_name) {
1562
1563                         case IP_TOS:
1564                                 optval = inp->inp_ip_tos;
1565                                 break;
1566
1567                         case IP_TTL:
1568                                 optval = inp->inp_ip_ttl;
1569                                 break;
1570                         case IP_MINTTL:
1571                                 optval = inp->inp_ip_minttl;
1572                                 break;
1573
1574 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1575
1576                         case IP_RECVOPTS:
1577                                 optval = OPTBIT(INP_RECVOPTS);
1578                                 break;
1579
1580                         case IP_RECVRETOPTS:
1581                                 optval = OPTBIT(INP_RECVRETOPTS);
1582                                 break;
1583
1584                         case IP_RECVDSTADDR:
1585                                 optval = OPTBIT(INP_RECVDSTADDR);
1586                                 break;
1587
1588                         case IP_RECVTTL:
1589                                 optval = OPTBIT(INP_RECVTTL);
1590                                 break;
1591
1592                         case IP_RECVIF:
1593                                 optval = OPTBIT(INP_RECVIF);
1594                                 break;
1595
1596                         case IP_PORTRANGE:
1597                                 if (inp->inp_flags & INP_HIGHPORT)
1598                                         optval = IP_PORTRANGE_HIGH;
1599                                 else if (inp->inp_flags & INP_LOWPORT)
1600                                         optval = IP_PORTRANGE_LOW;
1601                                 else
1602                                         optval = 0;
1603                                 break;
1604
1605                         case IP_FAITH:
1606                                 optval = OPTBIT(INP_FAITH);
1607                                 break;
1608                         }
1609                         error = sooptcopyout(sopt, &optval, sizeof optval);
1610                         break;
1611
1612                 case IP_MULTICAST_IF:
1613                 case IP_MULTICAST_VIF:
1614                 case IP_MULTICAST_TTL:
1615                 case IP_MULTICAST_LOOP:
1616                 case IP_ADD_MEMBERSHIP:
1617                 case IP_DROP_MEMBERSHIP:
1618                         error = ip_getmoptions(sopt, inp->inp_moptions);
1619                         break;
1620
1621 #if defined(IPSEC) || defined(FAST_IPSEC)
1622                 case IP_IPSEC_POLICY:
1623                 {
1624                         struct mbuf *m = NULL;
1625                         caddr_t req = NULL;
1626                         size_t len = 0;
1627
1628                         if (m != NULL) {
1629                                 req = mtod(m, caddr_t);
1630                                 len = m->m_len;
1631                         }
1632                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1633                         if (error == 0)
1634                                 error = soopt_mcopyout(sopt, m); /* XXX */
1635                         if (error == 0)
1636                                 m_freem(m);
1637                         break;
1638                 }
1639 #endif /*IPSEC*/
1640
1641                 default:
1642                         error = ENOPROTOOPT;
1643                         break;
1644                 }
1645                 break;
1646         }
1647         return (error);
1648 }
1649
1650 /*
1651  * Set up IP options in pcb for insertion in output packets.
1652  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1653  * with destination address if source routed.
1654  */
1655 static int
1656 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1657 {
1658         int cnt, optlen;
1659         u_char *cp;
1660         u_char opt;
1661
1662         /* turn off any old options */
1663         if (*pcbopt)
1664                 m_free(*pcbopt);
1665         *pcbopt = 0;
1666         if (m == NULL || m->m_len == 0) {
1667                 /*
1668                  * Only turning off any previous options.
1669                  */
1670                 if (m != NULL)
1671                         m_free(m);
1672                 return (0);
1673         }
1674
1675         if (m->m_len % sizeof(int32_t))
1676                 goto bad;
1677         /*
1678          * IP first-hop destination address will be stored before
1679          * actual options; move other options back
1680          * and clear it when none present.
1681          */
1682         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1683                 goto bad;
1684         cnt = m->m_len;
1685         m->m_len += sizeof(struct in_addr);
1686         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1687         ovbcopy(mtod(m, caddr_t), cp, cnt);
1688         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1689
1690         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1691                 opt = cp[IPOPT_OPTVAL];
1692                 if (opt == IPOPT_EOL)
1693                         break;
1694                 if (opt == IPOPT_NOP)
1695                         optlen = 1;
1696                 else {
1697                         if (cnt < IPOPT_OLEN + sizeof *cp)
1698                                 goto bad;
1699                         optlen = cp[IPOPT_OLEN];
1700                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1701                                 goto bad;
1702                 }
1703                 switch (opt) {
1704
1705                 default:
1706                         break;
1707
1708                 case IPOPT_LSRR:
1709                 case IPOPT_SSRR:
1710                         /*
1711                          * user process specifies route as:
1712                          *      ->A->B->C->D
1713                          * D must be our final destination (but we can't
1714                          * check that since we may not have connected yet).
1715                          * A is first hop destination, which doesn't appear in
1716                          * actual IP option, but is stored before the options.
1717                          */
1718                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1719                                 goto bad;
1720                         m->m_len -= sizeof(struct in_addr);
1721                         cnt -= sizeof(struct in_addr);
1722                         optlen -= sizeof(struct in_addr);
1723                         cp[IPOPT_OLEN] = optlen;
1724                         /*
1725                          * Move first hop before start of options.
1726                          */
1727                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1728                               sizeof(struct in_addr));
1729                         /*
1730                          * Then copy rest of options back
1731                          * to close up the deleted entry.
1732                          */
1733                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1734                                 &cp[IPOPT_OFFSET+1],
1735                                 cnt - (IPOPT_MINOFF - 1));
1736                         break;
1737                 }
1738         }
1739         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1740                 goto bad;
1741         *pcbopt = m;
1742         return (0);
1743
1744 bad:
1745         m_free(m);
1746         return (EINVAL);
1747 }
1748
1749 /*
1750  * XXX
1751  * The whole multicast option thing needs to be re-thought.
1752  * Several of these options are equally applicable to non-multicast
1753  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1754  * standard option (IP_TTL).
1755  */
1756
1757 /*
1758  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1759  */
1760 static struct ifnet *
1761 ip_multicast_if(struct in_addr *a, int *ifindexp)
1762 {
1763         int ifindex;
1764         struct ifnet *ifp;
1765
1766         if (ifindexp)
1767                 *ifindexp = 0;
1768         if (ntohl(a->s_addr) >> 24 == 0) {
1769                 ifindex = ntohl(a->s_addr) & 0xffffff;
1770                 if (ifindex < 0 || if_index < ifindex)
1771                         return NULL;
1772                 ifp = ifindex2ifnet[ifindex];
1773                 if (ifindexp)
1774                         *ifindexp = ifindex;
1775         } else {
1776                 INADDR_TO_IFP(*a, ifp);
1777         }
1778         return ifp;
1779 }
1780
1781 /*
1782  * Set the IP multicast options in response to user setsockopt().
1783  */
1784 static int
1785 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1786 {
1787         int error = 0;
1788         int i;
1789         struct in_addr addr;
1790         struct ip_mreq mreq;
1791         struct ifnet *ifp;
1792         struct ip_moptions *imo = *imop;
1793         int ifindex;
1794
1795         if (imo == NULL) {
1796                 /*
1797                  * No multicast option buffer attached to the pcb;
1798                  * allocate one and initialize to default values.
1799                  */
1800                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1801
1802                 if (imo == NULL)
1803                         return (ENOBUFS);
1804                 *imop = imo;
1805                 imo->imo_multicast_ifp = NULL;
1806                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1807                 imo->imo_multicast_vif = -1;
1808                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1809                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1810                 imo->imo_num_memberships = 0;
1811         }
1812
1813         switch (sopt->sopt_name) {
1814         /* store an index number for the vif you wanna use in the send */
1815         case IP_MULTICAST_VIF:
1816                 if (legal_vif_num == 0) {
1817                         error = EOPNOTSUPP;
1818                         break;
1819                 }
1820                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1821                 if (error)
1822                         break;
1823                 if (!legal_vif_num(i) && (i != -1)) {
1824                         error = EINVAL;
1825                         break;
1826                 }
1827                 imo->imo_multicast_vif = i;
1828                 break;
1829
1830         case IP_MULTICAST_IF:
1831                 /*
1832                  * Select the interface for outgoing multicast packets.
1833                  */
1834                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1835                 if (error)
1836                         break;
1837                 /*
1838                  * INADDR_ANY is used to remove a previous selection.
1839                  * When no interface is selected, a default one is
1840                  * chosen every time a multicast packet is sent.
1841                  */
1842                 if (addr.s_addr == INADDR_ANY) {
1843                         imo->imo_multicast_ifp = NULL;
1844                         break;
1845                 }
1846                 /*
1847                  * The selected interface is identified by its local
1848                  * IP address.  Find the interface and confirm that
1849                  * it supports multicasting.
1850                  */
1851                 crit_enter();
1852                 ifp = ip_multicast_if(&addr, &ifindex);
1853                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1854                         crit_exit();
1855                         error = EADDRNOTAVAIL;
1856                         break;
1857                 }
1858                 imo->imo_multicast_ifp = ifp;
1859                 if (ifindex)
1860                         imo->imo_multicast_addr = addr;
1861                 else
1862                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1863                 crit_exit();
1864                 break;
1865
1866         case IP_MULTICAST_TTL:
1867                 /*
1868                  * Set the IP time-to-live for outgoing multicast packets.
1869                  * The original multicast API required a char argument,
1870                  * which is inconsistent with the rest of the socket API.
1871                  * We allow either a char or an int.
1872                  */
1873                 if (sopt->sopt_valsize == 1) {
1874                         u_char ttl;
1875                         error = sooptcopyin(sopt, &ttl, 1, 1);
1876                         if (error)
1877                                 break;
1878                         imo->imo_multicast_ttl = ttl;
1879                 } else {
1880                         u_int ttl;
1881                         error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1882                         if (error)
1883                                 break;
1884                         if (ttl > 255)
1885                                 error = EINVAL;
1886                         else
1887                                 imo->imo_multicast_ttl = ttl;
1888                 }
1889                 break;
1890
1891         case IP_MULTICAST_LOOP:
1892                 /*
1893                  * Set the loopback flag for outgoing multicast packets.
1894                  * Must be zero or one.  The original multicast API required a
1895                  * char argument, which is inconsistent with the rest
1896                  * of the socket API.  We allow either a char or an int.
1897                  */
1898                 if (sopt->sopt_valsize == 1) {
1899                         u_char loop;
1900
1901                         error = sooptcopyin(sopt, &loop, 1, 1);
1902                         if (error)
1903                                 break;
1904                         imo->imo_multicast_loop = !!loop;
1905                 } else {
1906                         u_int loop;
1907
1908                         error = sooptcopyin(sopt, &loop, sizeof loop,
1909                                             sizeof loop);
1910                         if (error)
1911                                 break;
1912                         imo->imo_multicast_loop = !!loop;
1913                 }
1914                 break;
1915
1916         case IP_ADD_MEMBERSHIP:
1917                 /*
1918                  * Add a multicast group membership.
1919                  * Group must be a valid IP multicast address.
1920                  */
1921                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1922                 if (error)
1923                         break;
1924
1925                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1926                         error = EINVAL;
1927                         break;
1928                 }
1929                 crit_enter();
1930                 /*
1931                  * If no interface address was provided, use the interface of
1932                  * the route to the given multicast address.
1933                  */
1934                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1935                         struct sockaddr_in dst;
1936                         struct rtentry *rt;
1937
1938                         bzero(&dst, sizeof(struct sockaddr_in));
1939                         dst.sin_len = sizeof(struct sockaddr_in);
1940                         dst.sin_family = AF_INET;
1941                         dst.sin_addr = mreq.imr_multiaddr;
1942                         rt = rtlookup((struct sockaddr *)&dst);
1943                         if (rt == NULL) {
1944                                 error = EADDRNOTAVAIL;
1945                                 crit_exit();
1946                                 break;
1947                         }
1948                         --rt->rt_refcnt;
1949                         ifp = rt->rt_ifp;
1950                 } else {
1951                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1952                 }
1953
1954                 /*
1955                  * See if we found an interface, and confirm that it
1956                  * supports multicast.
1957                  */
1958                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1959                         error = EADDRNOTAVAIL;
1960                         crit_exit();
1961                         break;
1962                 }
1963                 /*
1964                  * See if the membership already exists or if all the
1965                  * membership slots are full.
1966                  */
1967                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1968                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1969                             imo->imo_membership[i]->inm_addr.s_addr
1970                                                 == mreq.imr_multiaddr.s_addr)
1971                                 break;
1972                 }
1973                 if (i < imo->imo_num_memberships) {
1974                         error = EADDRINUSE;
1975                         crit_exit();
1976                         break;
1977                 }
1978                 if (i == IP_MAX_MEMBERSHIPS) {
1979                         error = ETOOMANYREFS;
1980                         crit_exit();
1981                         break;
1982                 }
1983                 /*
1984                  * Everything looks good; add a new record to the multicast
1985                  * address list for the given interface.
1986                  */
1987                 if ((imo->imo_membership[i] =
1988                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1989                         error = ENOBUFS;
1990                         crit_exit();
1991                         break;
1992                 }
1993                 ++imo->imo_num_memberships;
1994                 crit_exit();
1995                 break;
1996
1997         case IP_DROP_MEMBERSHIP:
1998                 /*
1999                  * Drop a multicast group membership.
2000                  * Group must be a valid IP multicast address.
2001                  */
2002                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2003                 if (error)
2004                         break;
2005
2006                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2007                         error = EINVAL;
2008                         break;
2009                 }
2010
2011                 crit_enter();
2012                 /*
2013                  * If an interface address was specified, get a pointer
2014                  * to its ifnet structure.
2015                  */
2016                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2017                         ifp = NULL;
2018                 else {
2019                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2020                         if (ifp == NULL) {
2021                                 error = EADDRNOTAVAIL;
2022                                 crit_exit();
2023                                 break;
2024                         }
2025                 }
2026                 /*
2027                  * Find the membership in the membership array.
2028                  */
2029                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2030                         if ((ifp == NULL ||
2031                              imo->imo_membership[i]->inm_ifp == ifp) &&
2032                             imo->imo_membership[i]->inm_addr.s_addr ==
2033                             mreq.imr_multiaddr.s_addr)
2034                                 break;
2035                 }
2036                 if (i == imo->imo_num_memberships) {
2037                         error = EADDRNOTAVAIL;
2038                         crit_exit();
2039                         break;
2040                 }
2041                 /*
2042                  * Give up the multicast address record to which the
2043                  * membership points.
2044                  */
2045                 in_delmulti(imo->imo_membership[i]);
2046                 /*
2047                  * Remove the gap in the membership array.
2048                  */
2049                 for (++i; i < imo->imo_num_memberships; ++i)
2050                         imo->imo_membership[i-1] = imo->imo_membership[i];
2051                 --imo->imo_num_memberships;
2052                 crit_exit();
2053                 break;
2054
2055         default:
2056                 error = EOPNOTSUPP;
2057                 break;
2058         }
2059
2060         /*
2061          * If all options have default values, no need to keep the mbuf.
2062          */
2063         if (imo->imo_multicast_ifp == NULL &&
2064             imo->imo_multicast_vif == -1 &&
2065             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2066             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2067             imo->imo_num_memberships == 0) {
2068                 kfree(*imop, M_IPMOPTS);
2069                 *imop = NULL;
2070         }
2071
2072         return (error);
2073 }
2074
2075 /*
2076  * Return the IP multicast options in response to user getsockopt().
2077  */
2078 static int
2079 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2080 {
2081         struct in_addr addr;
2082         struct in_ifaddr *ia;
2083         int error, optval;
2084         u_char coptval;
2085
2086         error = 0;
2087         switch (sopt->sopt_name) {
2088         case IP_MULTICAST_VIF:
2089                 if (imo != NULL)
2090                         optval = imo->imo_multicast_vif;
2091                 else
2092                         optval = -1;
2093                 error = sooptcopyout(sopt, &optval, sizeof optval);
2094                 break;
2095
2096         case IP_MULTICAST_IF:
2097                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2098                         addr.s_addr = INADDR_ANY;
2099                 else if (imo->imo_multicast_addr.s_addr) {
2100                         /* return the value user has set */
2101                         addr = imo->imo_multicast_addr;
2102                 } else {
2103                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2104                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2105                                 : IA_SIN(ia)->sin_addr.s_addr;
2106                 }
2107                 error = sooptcopyout(sopt, &addr, sizeof addr);
2108                 break;
2109
2110         case IP_MULTICAST_TTL:
2111                 if (imo == NULL)
2112                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2113                 else
2114                         optval = coptval = imo->imo_multicast_ttl;
2115                 if (sopt->sopt_valsize == 1)
2116                         error = sooptcopyout(sopt, &coptval, 1);
2117                 else
2118                         error = sooptcopyout(sopt, &optval, sizeof optval);
2119                 break;
2120
2121         case IP_MULTICAST_LOOP:
2122                 if (imo == NULL)
2123                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2124                 else
2125                         optval = coptval = imo->imo_multicast_loop;
2126                 if (sopt->sopt_valsize == 1)
2127                         error = sooptcopyout(sopt, &coptval, 1);
2128                 else
2129                         error = sooptcopyout(sopt, &optval, sizeof optval);
2130                 break;
2131
2132         default:
2133                 error = ENOPROTOOPT;
2134                 break;
2135         }
2136         return (error);
2137 }
2138
2139 /*
2140  * Discard the IP multicast options.
2141  */
2142 void
2143 ip_freemoptions(struct ip_moptions *imo)
2144 {
2145         int i;
2146
2147         if (imo != NULL) {
2148                 for (i = 0; i < imo->imo_num_memberships; ++i)
2149                         in_delmulti(imo->imo_membership[i]);
2150                 kfree(imo, M_IPMOPTS);
2151         }
2152 }
2153
2154 /*
2155  * Routine called from ip_output() to loop back a copy of an IP multicast
2156  * packet to the input queue of a specified interface.  Note that this
2157  * calls the output routine of the loopback "driver", but with an interface
2158  * pointer that might NOT be a loopback interface -- evil, but easier than
2159  * replicating that code here.
2160  */
2161 static void
2162 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2163              int hlen)
2164 {
2165         struct ip *ip;
2166         struct mbuf *copym;
2167
2168         copym = m_copypacket(m, MB_DONTWAIT);
2169         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2170                 copym = m_pullup(copym, hlen);
2171         if (copym != NULL) {
2172                 /*
2173                  * if the checksum hasn't been computed, mark it as valid
2174                  */
2175                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2176                         in_delayed_cksum(copym);
2177                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2178                         copym->m_pkthdr.csum_flags |=
2179                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2180                         copym->m_pkthdr.csum_data = 0xffff;
2181                 }
2182                 /*
2183                  * We don't bother to fragment if the IP length is greater
2184                  * than the interface's MTU.  Can this possibly matter?
2185                  */
2186                 ip = mtod(copym, struct ip *);
2187                 ip->ip_len = htons(ip->ip_len);
2188                 ip->ip_off = htons(ip->ip_off);
2189                 ip->ip_sum = 0;
2190                 if (ip->ip_vhl == IP_VHL_BORING) {
2191                         ip->ip_sum = in_cksum_hdr(ip);
2192                 } else {
2193                         ip->ip_sum = in_cksum(copym, hlen);
2194                 }
2195                 /*
2196                  * NB:
2197                  * It's not clear whether there are any lingering
2198                  * reentrancy problems in other areas which might
2199                  * be exposed by using ip_input directly (in
2200                  * particular, everything which modifies the packet
2201                  * in-place).  Yet another option is using the
2202                  * protosw directly to deliver the looped back
2203                  * packet.  For the moment, we'll err on the side
2204                  * of safety by using if_simloop().
2205                  */
2206 #if 1 /* XXX */
2207                 if (dst->sin_family != AF_INET) {
2208                         kprintf("ip_mloopback: bad address family %d\n",
2209                                                 dst->sin_family);
2210                         dst->sin_family = AF_INET;
2211                 }
2212 #endif
2213
2214 #ifdef notdef
2215                 copym->m_pkthdr.rcvif = ifp;
2216                 ip_input(copym);
2217 #else
2218                 if_simloop(ifp, copym, dst->sin_family, 0);
2219 #endif
2220         }
2221 }