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 1.43 2008/06/08 08:38:05 sephe 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         struct m_tag *dn_mtag = NULL;
 139 #ifdef IPSEC
 140         struct secpolicy *sp = NULL;
 141         struct socket *so = inp ? inp->inp_socket : NULL;
 142 #endif
 143 #ifdef FAST_IPSEC
 144         struct m_tag *mtag;
 145         struct secpolicy *sp = NULL;
 146         struct tdb_ident *tdbi;
 147 #endif /* FAST_IPSEC */
 148         struct ip_fw_args args;
 149         int src_was_INADDR_ANY = 0;     /* as the name says... */
 150
 151         args.eh = NULL;
 152         args.rule = NULL;
 153         args.next_hop = NULL;
 154
 155         /* Grab info from MT_TAG mbufs prepended to the chain. */
 156         while (m0 != NULL && m0->m_type == MT_TAG) {
 157                 switch(m0->_m_tag_id) {
 158                 case PACKET_TAG_IPFORWARD:
 159                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 160                         break;
 161                 default:
 162                         kprintf("ip_output: unrecognised MT_TAG tag %d\n",
 163                             m0->_m_tag_id);
 164                         break;
 165                 }
 166                 m0 = m0->m_next;
 167         }
 168         m = m0;
 169         M_ASSERTPKTHDR(m);
 170
 171         /* Extract info from dummynet tag */
 172         dn_mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 173         if (dn_mtag != NULL) {
 174                 struct dn_pkt *dn_pkt = m_tag_data(dn_mtag);
 175
 176                 /*
 177                  * The packet was already tagged, so part of the
 178                  * processing was already done, and we need to go down.
 179                  * Get parameters from the tag.
 180                  */
 181                 args.rule = dn_pkt->dn_priv;
 182                 opt = NULL;
 183                 ro = &dn_pkt->ro;
 184                 imo = NULL;
 185                 dst = dn_pkt->dn_dst;
 186                 ifp = dn_pkt->ifp;
 187                 flags = dn_pkt->flags;
 188
 189                 /*
 190                  * Don't delete the dummynet tag here, just unlink it,
 191                  * since some local variables (like 'ro' and 'dst') are
 192                  * still referencing certain parts of it.
 193                  * The dummynet tag will be freed at the end of the
 194                  * output process.
 195                  */
 196                 m_tag_unlink(m, dn_mtag);
 197         }
 198
 199         if (ro == NULL) {
 200                 ro = &iproute;
 201                 bzero(ro, sizeof *ro);
 202         }
 203
 204         if (args.rule != NULL) {        /* dummynet already saw us */
 205                 ip = mtod(m, struct ip *);
 206                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 207                 if (ro->ro_rt)
 208                         ia = ifatoia(ro->ro_rt->rt_ifa);
 209                 goto sendit;
 210         }
 211
 212         if (opt) {
 213                 len = 0;
 214                 m = ip_insertoptions(m, opt, &len);
 215                 if (len != 0)
 216                         hlen = len;
 217         }
 218         ip = mtod(m, struct ip *);
 219         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 220
 221         /*
 222          * Fill in IP header.
 223          */
 224         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 225                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 226                 ip->ip_off &= IP_DF;
 227                 ip->ip_id = ip_newid();
 228                 ipstat.ips_localout++;
 229         } else {
 230                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 231         }
 232
 233         dst = (struct sockaddr_in *)&ro->ro_dst;
 234         /*
 235          * If there is a cached route,
 236          * check that it is to the same destination
 237          * and is still up.  If not, free it and try again.
 238          * The address family should also be checked in case of sharing the
 239          * cache with IPv6.
 240          */
 241         if (ro->ro_rt &&
 242             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 243              dst->sin_family != AF_INET ||
 244              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 245                 rtfree(ro->ro_rt);
 246                 ro->ro_rt = (struct rtentry *)NULL;
 247         }
 248         if (ro->ro_rt == NULL) {
 249                 bzero(dst, sizeof *dst);
 250                 dst->sin_family = AF_INET;
 251                 dst->sin_len = sizeof *dst;
 252                 dst->sin_addr = pkt_dst;
 253         }
 254         /*
 255          * If routing to interface only,
 256          * short circuit routing lookup.
 257          */
 258         if (flags & IP_ROUTETOIF) {
 259                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 260                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 261                         ipstat.ips_noroute++;
 262                         error = ENETUNREACH;
 263                         goto bad;
 264                 }
 265                 ifp = ia->ia_ifp;
 266                 ip->ip_ttl = 1;
 267                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 268         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 269                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 270                 /*
 271                  * Bypass the normal routing lookup for multicast
 272                  * packets if the interface is specified.
 273                  */
 274                 ifp = imo->imo_multicast_ifp;
 275                 ia = IFP_TO_IA(ifp);
 276                 isbroadcast = 0;        /* fool gcc */
 277         } else {
 278                 /*
 279                  * If this is the case, we probably don't want to allocate
 280                  * a protocol-cloned route since we didn't get one from the
 281                  * ULP.  This lets TCP do its thing, while not burdening
 282                  * forwarding or ICMP with the overhead of cloning a route.
 283                  * Of course, we still want to do any cloning requested by
 284                  * the link layer, as this is probably required in all cases
 285                  * for correct operation (as it is for ARP).
 286                  */
 287                 if (ro->ro_rt == NULL)
 288                         rtalloc_ign(ro, RTF_PRCLONING);
 289                 if (ro->ro_rt == NULL) {
 290                         ipstat.ips_noroute++;
 291                         error = EHOSTUNREACH;
 292                         goto bad;
 293                 }
 294                 ia = ifatoia(ro->ro_rt->rt_ifa);
 295                 ifp = ro->ro_rt->rt_ifp;
 296                 ro->ro_rt->rt_use++;
 297                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 298                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 299                 if (ro->ro_rt->rt_flags & RTF_HOST)
 300                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 301                 else
 302                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 303         }
 304         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 305                 struct in_multi *inm;
 306
 307                 m->m_flags |= M_MCAST;
 308                 /*
 309                  * IP destination address is multicast.  Make sure "dst"
 310                  * still points to the address in "ro".  (It may have been
 311                  * changed to point to a gateway address, above.)
 312                  */
 313                 dst = (struct sockaddr_in *)&ro->ro_dst;
 314                 /*
 315                  * See if the caller provided any multicast options
 316                  */
 317                 if (imo != NULL) {
 318                         ip->ip_ttl = imo->imo_multicast_ttl;
 319                         if (imo->imo_multicast_vif != -1)
 320                                 ip->ip_src.s_addr =
 321                                     ip_mcast_src ?
 322                                     ip_mcast_src(imo->imo_multicast_vif) :
 323                                     INADDR_ANY;
 324                 } else
 325                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 326                 /*
 327                  * Confirm that the outgoing interface supports multicast.
 328                  */
 329                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 330                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 331                                 ipstat.ips_noroute++;
 332                                 error = ENETUNREACH;
 333                                 goto bad;
 334                         }
 335                 }
 336                 /*
 337                  * If source address not specified yet, use address
 338                  * of outgoing interface.
 339                  */
 340                 if (ip->ip_src.s_addr == INADDR_ANY) {
 341                         /* Interface may have no addresses. */
 342                         if (ia != NULL)
 343                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 344                 }
 345
 346                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 347                 if (inm != NULL &&
 348                    (imo == NULL || imo->imo_multicast_loop)) {
 349                         /*
 350                          * If we belong to the destination multicast group
 351                          * on the outgoing interface, and the caller did not
 352                          * forbid loopback, loop back a copy.
 353                          */
 354                         ip_mloopback(ifp, m, dst, hlen);
 355                 }
 356                 else {
 357                         /*
 358                          * If we are acting as a multicast router, perform
 359                          * multicast forwarding as if the packet had just
 360                          * arrived on the interface to which we are about
 361                          * to send.  The multicast forwarding function
 362                          * recursively calls this function, using the
 363                          * IP_FORWARDING flag to prevent infinite recursion.
 364                          *
 365                          * Multicasts that are looped back by ip_mloopback(),
 366                          * above, will be forwarded by the ip_input() routine,
 367                          * if necessary.
 368                          */
 369                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 370                                 /*
 371                                  * If rsvp daemon is not running, do not
 372                                  * set ip_moptions. This ensures that the packet
 373                                  * is multicast and not just sent down one link
 374                                  * as prescribed by rsvpd.
 375                                  */
 376                                 if (!rsvp_on)
 377                                         imo = NULL;
 378                                 if (ip_mforward &&
 379                                     ip_mforward(ip, ifp, m, imo) != 0) {
 380                                         m_freem(m);
 381                                         goto done;
 382                                 }
 383                         }
 384                 }
 385
 386                 /*
 387                  * Multicasts with a time-to-live of zero may be looped-
 388                  * back, above, but must not be transmitted on a network.
 389                  * Also, multicasts addressed to the loopback interface
 390                  * are not sent -- the above call to ip_mloopback() will
 391                  * loop back a copy if this host actually belongs to the
 392                  * destination group on the loopback interface.
 393                  */
 394                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 395                         m_freem(m);
 396                         goto done;
 397                 }
 398
 399                 goto sendit;
 400         }
 401 #ifndef notdef
 402         /*
 403          * If the source address is not specified yet, use the address
 404          * of the outoing interface. In case, keep note we did that, so
 405          * if the the firewall changes the next-hop causing the output
 406          * interface to change, we can fix that.
 407          */
 408         if (ip->ip_src.s_addr == INADDR_ANY) {
 409                 /* Interface may have no addresses. */
 410                 if (ia != NULL) {
 411                         ip->ip_src = IA_SIN(ia)->sin_addr;
 412                         src_was_INADDR_ANY = 1;
 413                 }
 414         }
 415 #endif /* notdef */
 416 #ifdef ALTQ
 417         /*
 418          * Disable packet drop hack.
 419          * Packetdrop should be done by queueing.
 420          */
 421 #else /* !ALTQ */
 422         /*
 423          * Verify that we have any chance at all of being able to queue
 424          *      the packet or packet fragments
 425          */
 426         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 427                 ifp->if_snd.ifq_maxlen) {
 428                         error = ENOBUFS;
 429                         ipstat.ips_odropped++;
 430                         goto bad;
 431         }
 432 #endif /* !ALTQ */
 433
 434         /*
 435          * Look for broadcast address and
 436          * verify user is allowed to send
 437          * such a packet.
 438          */
 439         if (isbroadcast) {
 440                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 441                         error = EADDRNOTAVAIL;
 442                         goto bad;
 443                 }
 444                 if (!(flags & IP_ALLOWBROADCAST)) {
 445                         error = EACCES;
 446                         goto bad;
 447                 }
 448                 /* don't allow broadcast messages to be fragmented */
 449                 if (ip->ip_len > ifp->if_mtu) {
 450                         error = EMSGSIZE;
 451                         goto bad;
 452                 }
 453                 m->m_flags |= M_BCAST;
 454         } else {
 455                 m->m_flags &= ~M_BCAST;
 456         }
 457
 458 sendit:
 459 #ifdef IPSEC
 460         /* get SP for this packet */
 461         if (so == NULL)
 462                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 463         else
 464                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 465
 466         if (sp == NULL) {
 467                 ipsecstat.out_inval++;
 468                 goto bad;
 469         }
 470
 471         error = 0;
 472
 473         /* check policy */
 474         switch (sp->policy) {
 475         case IPSEC_POLICY_DISCARD:
 476                 /*
 477                  * This packet is just discarded.
 478                  */
 479                 ipsecstat.out_polvio++;
 480                 goto bad;
 481
 482         case IPSEC_POLICY_BYPASS:
 483         case IPSEC_POLICY_NONE:
 484                 /* no need to do IPsec. */
 485                 goto skip_ipsec;
 486
 487         case IPSEC_POLICY_IPSEC:
 488                 if (sp->req == NULL) {
 489                         /* acquire a policy */
 490                         error = key_spdacquire(sp);
 491                         goto bad;
 492                 }
 493                 break;
 494
 495         case IPSEC_POLICY_ENTRUST:
 496         default:
 497                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 498         }
 499     {
 500         struct ipsec_output_state state;
 501         bzero(&state, sizeof state);
 502         state.m = m;
 503         if (flags & IP_ROUTETOIF) {
 504                 state.ro = &iproute;
 505                 bzero(&iproute, sizeof iproute);
 506         } else
 507                 state.ro = ro;
 508         state.dst = (struct sockaddr *)dst;
 509
 510         ip->ip_sum = 0;
 511
 512         /*
 513          * XXX
 514          * delayed checksums are not currently compatible with IPsec
 515          */
 516         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 517                 in_delayed_cksum(m);
 518                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 519         }
 520
 521         ip->ip_len = htons(ip->ip_len);
 522         ip->ip_off = htons(ip->ip_off);
 523
 524         error = ipsec4_output(&state, sp, flags);
 525
 526         m = state.m;
 527         if (flags & IP_ROUTETOIF) {
 528                 /*
 529                  * if we have tunnel mode SA, we may need to ignore
 530                  * IP_ROUTETOIF.
 531                  */
 532                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 533                         flags &= ~IP_ROUTETOIF;
 534                         ro = state.ro;
 535                 }
 536         } else
 537                 ro = state.ro;
 538         dst = (struct sockaddr_in *)state.dst;
 539         if (error) {
 540                 /* mbuf is already reclaimed in ipsec4_output. */
 541                 m0 = NULL;
 542                 switch (error) {
 543                 case EHOSTUNREACH:
 544                 case ENETUNREACH:
 545                 case EMSGSIZE:
 546                 case ENOBUFS:
 547                 case ENOMEM:
 548                         break;
 549                 default:
 550                         kprintf("ip4_output (ipsec): error code %d\n", error);
 551                         /*fall through*/
 552                 case ENOENT:
 553                         /* don't show these error codes to the user */
 554                         error = 0;
 555                         break;
 556                 }
 557                 goto bad;
 558         }
 559     }
 560
 561         /* be sure to update variables that are affected by ipsec4_output() */
 562         ip = mtod(m, struct ip *);
 563 #ifdef _IP_VHL
 564         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 565 #else
 566         hlen = ip->ip_hl << 2;
 567 #endif
 568         if (ro->ro_rt == NULL) {
 569                 if (!(flags & IP_ROUTETOIF)) {
 570                         kprintf("ip_output: "
 571                                 "can't update route after IPsec processing\n");
 572                         error = EHOSTUNREACH;   /*XXX*/
 573                         goto bad;
 574                 }
 575         } else {
 576                 ia = ifatoia(ro->ro_rt->rt_ifa);
 577                 ifp = ro->ro_rt->rt_ifp;
 578         }
 579
 580         /* make it flipped, again. */
 581         ip->ip_len = ntohs(ip->ip_len);
 582         ip->ip_off = ntohs(ip->ip_off);
 583 skip_ipsec:
 584 #endif /*IPSEC*/
 585 #ifdef FAST_IPSEC
 586         /*
 587          * Check the security policy (SP) for the packet and, if
 588          * required, do IPsec-related processing.  There are two
 589          * cases here; the first time a packet is sent through
 590          * it will be untagged and handled by ipsec4_checkpolicy.
 591          * If the packet is resubmitted to ip_output (e.g. after
 592          * AH, ESP, etc. processing), there will be a tag to bypass
 593          * the lookup and related policy checking.
 594          */
 595         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 596         crit_enter();
 597         if (mtag != NULL) {
 598                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 599                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 600                 if (sp == NULL)
 601                         error = -EINVAL;        /* force silent drop */
 602                 m_tag_delete(m, mtag);
 603         } else {
 604                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 605                                         &error, inp);
 606         }
 607         /*
 608          * There are four return cases:
 609          *    sp != NULL                    apply IPsec policy
 610          *    sp == NULL, error == 0        no IPsec handling needed
 611          *    sp == NULL, error == -EINVAL  discard packet w/o error
 612          *    sp == NULL, error != 0        discard packet, report error
 613          */
 614         if (sp != NULL) {
 615                 /* Loop detection, check if ipsec processing already done */
 616                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 617                 for (mtag = m_tag_first(m); mtag != NULL;
 618                      mtag = m_tag_next(m, mtag)) {
 619                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 620                                 continue;
 621                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 622                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 623                                 continue;
 624                         /*
 625                          * Check if policy has an SA associated with it.
 626                          * This can happen when an SP has yet to acquire
 627                          * an SA; e.g. on first reference.  If it occurs,
 628                          * then we let ipsec4_process_packet do its thing.
 629                          */
 630                         if (sp->req->sav == NULL)
 631                                 break;
 632                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 633                         if (tdbi->spi == sp->req->sav->spi &&
 634                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 635                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 636                                  sizeof(union sockaddr_union)) == 0) {
 637                                 /*
 638                                  * No IPsec processing is needed, free
 639                                  * reference to SP.
 640                                  *
 641                                  * NB: null pointer to avoid free at
 642                                  *     done: below.
 643                                  */
 644                                 KEY_FREESP(&sp), sp = NULL;
 645                                 crit_exit();
 646                                 goto spd_done;
 647                         }
 648                 }
 649
 650                 /*
 651                  * Do delayed checksums now because we send before
 652                  * this is done in the normal processing path.
 653                  */
 654                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 655                         in_delayed_cksum(m);
 656                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 657                 }
 658
 659                 ip->ip_len = htons(ip->ip_len);
 660                 ip->ip_off = htons(ip->ip_off);
 661
 662                 /* NB: callee frees mbuf */
 663                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 664                 /*
 665                  * Preserve KAME behaviour: ENOENT can be returned
 666                  * when an SA acquire is in progress.  Don't propagate
 667                  * this to user-level; it confuses applications.
 668                  *
 669                  * XXX this will go away when the SADB is redone.
 670                  */
 671                 if (error == ENOENT)
 672                         error = 0;
 673                 crit_exit();
 674                 goto done;
 675         } else {
 676                 crit_exit();
 677
 678                 if (error != 0) {
 679                         /*
 680                          * Hack: -EINVAL is used to signal that a packet
 681                          * should be silently discarded.  This is typically
 682                          * because we asked key management for an SA and
 683                          * it was delayed (e.g. kicked up to IKE).
 684                          */
 685                         if (error == -EINVAL)
 686                                 error = 0;
 687                         goto bad;
 688                 } else {
 689                         /* No IPsec processing for this packet. */
 690                 }
 691 #ifdef notyet
 692                 /*
 693                  * If deferred crypto processing is needed, check that
 694                  * the interface supports it.
 695                  */
 696                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 697                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 698                         /* notify IPsec to do its own crypto */
 699                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 700                         error = EHOSTUNREACH;
 701                         goto bad;
 702                 }
 703 #endif
 704         }
 705 spd_done:
 706 #endif /* FAST_IPSEC */
 707         /*
 708          * IpHack's section.
 709          * - Xlate: translate packet's addr/port (NAT).
 710          * - Firewall: deny/allow/etc.
 711          * - Wrap: fake packet's addr/port <unimpl.>
 712          * - Encapsulate: put it in another IP and send out. <unimp.>
 713          */
 714
 715         /*
 716          * Run through list of hooks for output packets.
 717          */
 718         if (pfil_has_hooks(&inet_pfil_hook)) {
 719                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 720                 if (error != 0 || m == NULL)
 721                         goto done;
 722                 ip = mtod(m, struct ip *);
 723         }
 724
 725         /*
 726          * Check with the firewall...
 727          * but not if we are already being fwd'd from a firewall.
 728          */
 729         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 730                 struct sockaddr_in *old = dst;
 731
 732                 args.m = m;
 733                 args.next_hop = dst;
 734                 args.oif = ifp;
 735                 off = ip_fw_chk_ptr(&args);
 736                 m = args.m;
 737                 dst = args.next_hop;
 738
 739                 /*
 740                  * On return we must do the following:
 741                  * m == NULL    -> drop the pkt (old interface, deprecated)
 742                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 743                  * 1<=off<= 0xffff              -> DIVERT
 744                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 745                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 746                  * dst != old                   -> IPFIREWALL_FORWARD
 747                  * off==0, dst==old             -> accept
 748                  * If some of the above modules are not compiled in, then
 749                  * we should't have to check the corresponding condition
 750                  * (because the ipfw control socket should not accept
 751                  * unsupported rules), but better play safe and drop
 752                  * packets in case of doubt.
 753                  */
 754                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 755                         if (m)
 756                                 m_freem(m);
 757                         error = EACCES;
 758                         goto done;
 759                 }
 760                 ip = mtod(m, struct ip *);
 761                 if (off == 0 && dst == old)             /* common case */
 762                         goto pass;
 763                 if (off & IP_FW_PORT_DYNT_FLAG) {
 764                         /*
 765                          * pass the pkt to dummynet. Need to include
 766                          * pipe number, m, ifp, ro, dst because these are
 767                          * not recomputed in the next pass.
 768                          * All other parameters have been already used and
 769                          * so they are not needed anymore.
 770                          * XXX note: if the ifp or ro entry are deleted
 771                          * while a pkt is in dummynet, we are in trouble!
 772                          */
 773                         args.ro = ro;
 774                         args.dst = dst;
 775                         args.flags = flags;
 776
 777                         error = 0;
 778                         ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
 779                         goto done;
 780                 }
 781 #ifdef IPDIVERT
 782                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
 783                         struct mbuf *clone = NULL;
 784
 785                         /* Clone packet if we're doing a 'tee' */
 786                         if ((off & IP_FW_PORT_TEE_FLAG))
 787                                 clone = m_dup(m, MB_DONTWAIT);
 788
 789                         /*
 790                          * XXX
 791                          * delayed checksums are not currently compatible
 792                          * with divert sockets.
 793                          */
 794                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 795                                 in_delayed_cksum(m);
 796                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 797                         }
 798
 799                         /* Restore packet header fields to original values */
 800                         ip->ip_len = htons(ip->ip_len);
 801                         ip->ip_off = htons(ip->ip_off);
 802
 803                         /* Deliver packet to divert input routine */
 804                         divert_packet(m, 0, off & 0xffff);
 805
 806                         /* If 'tee', continue with original packet */
 807                         if (clone != NULL) {
 808                                 m = clone;
 809                                 ip = mtod(m, struct ip *);
 810                                 goto pass;
 811                         }
 812                         goto done;
 813                 }
 814 #endif
 815
 816                 /* IPFIREWALL_FORWARD */
 817                 /*
 818                  * Check dst to make sure it is directly reachable on the
 819                  * interface we previously thought it was.
 820                  * If it isn't (which may be likely in some situations) we have
 821                  * to re-route it (ie, find a route for the next-hop and the
 822                  * associated interface) and set them here. This is nested
 823                  * forwarding which in most cases is undesirable, except where
 824                  * such control is nigh impossible. So we do it here.
 825                  * And I'm babbling.
 826                  */
 827                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 828 #if 0
 829                         /*
 830                          * XXX To improve readability, this block should be
 831                          * changed into a function call as below:
 832                          */
 833                         error = ip_ipforward(&m, &dst, &ifp);
 834                         if (error)
 835                                 goto bad;
 836                         if (m == NULL) /* ip_input consumed the mbuf */
 837                                 goto done;
 838 #else
 839                         struct in_ifaddr *ia;
 840
 841                         /*
 842                          * XXX sro_fwd below is static, and a pointer
 843                          * to it gets passed to routines downstream.
 844                          * This could have surprisingly bad results in
 845                          * practice, because its content is overwritten
 846                          * by subsequent packets.
 847                          */
 848                         /* There must be a better way to do this next line... */
 849                         static struct route sro_fwd;
 850                         struct route *ro_fwd = &sro_fwd;
 851
 852 #if 0
 853                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 854                             dst->sin_addr, "\n");
 855 #endif
 856
 857                         /*
 858                          * We need to figure out if we have been forwarded
 859                          * to a local socket. If so, then we should somehow
 860                          * "loop back" to ip_input, and get directed to the
 861                          * PCB as if we had received this packet. This is
 862                          * because it may be dificult to identify the packets
 863                          * you want to forward until they are being output
 864                          * and have selected an interface. (e.g. locally
 865                          * initiated packets) If we used the loopback inteface,
 866                          * we would not be able to control what happens
 867                          * as the packet runs through ip_input() as
 868                          * it is done through a ISR.
 869                          */
 870                         LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
 871                                      ia_hash) {
 872                                 /*
 873                                  * If the addr to forward to is one
 874                                  * of ours, we pretend to
 875                                  * be the destination for this packet.
 876                                  */
 877                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 878                                                  dst->sin_addr.s_addr)
 879                                         break;
 880                         }
 881                         if (ia != NULL) {    /* tell ip_input "dont filter" */
 882                                 struct m_hdr tag;
 883
 884                                 tag.mh_type = MT_TAG;
 885                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 886                                 tag.mh_data = (caddr_t)args.next_hop;
 887                                 tag.mh_next = m;
 888
 889                                 if (m->m_pkthdr.rcvif == NULL)
 890                                         m->m_pkthdr.rcvif = ifunit("lo0");
 891                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 892                                         m->m_pkthdr.csum_flags |=
 893                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 894                                         m->m_pkthdr.csum_data = 0xffff;
 895                                 }
 896                                 m->m_pkthdr.csum_flags |=
 897                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 898                                 ip->ip_len = htons(ip->ip_len);
 899                                 ip->ip_off = htons(ip->ip_off);
 900                                 ip_input((struct mbuf *)&tag);
 901                                 goto done;
 902                         }
 903                         /* Some of the logic for this was nicked from above.
 904                          *
 905                          * This rewrites the cached route in a local PCB.
 906                          * Is this what we want to do?
 907                          */
 908                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
 909                         ro_fwd->ro_rt = NULL;
 910
 911                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 912                         if (ro_fwd->ro_rt == NULL) {
 913                                 ipstat.ips_noroute++;
 914                                 error = EHOSTUNREACH;
 915                                 goto bad;
 916                         }
 917
 918                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 919                         ifp = ro_fwd->ro_rt->rt_ifp;
 920                         ro_fwd->ro_rt->rt_use++;
 921                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 922                                 dst = (struct sockaddr_in *)
 923                                     ro_fwd->ro_rt->rt_gateway;
 924                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 925                                 isbroadcast =
 926                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 927                         else
 928                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 929                         if (ro->ro_rt != NULL)
 930                                 rtfree(ro->ro_rt);
 931                         ro->ro_rt = ro_fwd->ro_rt;
 932                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 933
 934 #endif  /* ... block to be put into a function */
 935                         /*
 936                          * If we added a default src ip earlier,
 937                          * which would have been gotten from the-then
 938                          * interface, do it again, from the new one.
 939                          */
 940                         if (src_was_INADDR_ANY)
 941                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 942                         goto pass ;
 943                 }
 944
 945                 /*
 946                  * if we get here, none of the above matches, and
 947                  * we have to drop the pkt
 948                  */
 949                 m_freem(m);
 950                 error = EACCES; /* not sure this is the right error msg */
 951                 goto done;
 952         }
 953
 954 pass:
 955         /* 127/8 must not appear on wire - RFC1122. */
 956         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 957             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 958                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 959                         ipstat.ips_badaddr++;
 960                         error = EADDRNOTAVAIL;
 961                         goto bad;
 962                 }
 963         }
 964
 965         m->m_pkthdr.csum_flags |= CSUM_IP;
 966         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 967         if (sw_csum & CSUM_DELAY_DATA) {
 968                 in_delayed_cksum(m);
 969                 sw_csum &= ~CSUM_DELAY_DATA;
 970         }
 971         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 972
 973         /*
 974          * If small enough for interface, or the interface will take
 975          * care of the fragmentation for us, can just send directly.
 976          */
 977         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 978             !(ip->ip_off & IP_DF))) {
 979                 ip->ip_len = htons(ip->ip_len);
 980                 ip->ip_off = htons(ip->ip_off);
 981                 ip->ip_sum = 0;
 982                 if (sw_csum & CSUM_DELAY_IP) {
 983                         if (ip->ip_vhl == IP_VHL_BORING) {
 984                                 ip->ip_sum = in_cksum_hdr(ip);
 985                         } else {
 986                                 ip->ip_sum = in_cksum(m, hlen);
 987                         }
 988                 }
 989
 990                 /* Record statistics for this interface address. */
 991                 if (!(flags & IP_FORWARDING) && ia) {
 992                         ia->ia_ifa.if_opackets++;
 993                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 994                 }
 995
 996 #ifdef IPSEC
 997                 /* clean ipsec history once it goes out of the node */
 998                 ipsec_delaux(m);
 999 #endif
1000
1001 #ifdef MBUF_STRESS_TEST
1002                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1003                         struct mbuf *m1, *m2;
1004                         int length, tmp;
1005
1006                         tmp = length = m->m_pkthdr.len;
1007
1008                         while ((length -= mbuf_frag_size) >= 1) {
1009                                 m1 = m_split(m, length, MB_DONTWAIT);
1010                                 if (m1 == NULL)
1011                                         break;
1012                                 m2 = m;
1013                                 while (m2->m_next != NULL)
1014                                         m2 = m2->m_next;
1015                                 m2->m_next = m1;
1016                         }
1017                         m->m_pkthdr.len = tmp;
1018                 }
1019 #endif
1020                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1021                                        ro->ro_rt);
1022                 goto done;
1023         }
1024
1025         if (ip->ip_off & IP_DF) {
1026                 error = EMSGSIZE;
1027                 /*
1028                  * This case can happen if the user changed the MTU
1029                  * of an interface after enabling IP on it.  Because
1030                  * most netifs don't keep track of routes pointing to
1031                  * them, there is no way for one to update all its
1032                  * routes when the MTU is changed.
1033                  */
1034                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1035                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1036                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1037                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1038                 }
1039                 ipstat.ips_cantfrag++;
1040                 goto bad;
1041         }
1042
1043         /*
1044          * Too large for interface; fragment if possible. If successful,
1045          * on return, m will point to a list of packets to be sent.
1046          */
1047         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1048         if (error)
1049                 goto bad;
1050         for (; m; m = m0) {
1051                 m0 = m->m_nextpkt;
1052                 m->m_nextpkt = NULL;
1053 #ifdef IPSEC
1054                 /* clean ipsec history once it goes out of the node */
1055                 ipsec_delaux(m);
1056 #endif
1057                 if (error == 0) {
1058                         /* Record statistics for this interface address. */
1059                         if (ia != NULL) {
1060                                 ia->ia_ifa.if_opackets++;
1061                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1062                         }
1063                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1064                                                ro->ro_rt);
1065                 } else {
1066                         m_freem(m);
1067                 }
1068         }
1069
1070         if (error == 0)
1071                 ipstat.ips_fragmented++;
1072
1073 done:
1074         if (ro == &iproute && ro->ro_rt != NULL) {
1075                 RTFREE(ro->ro_rt);
1076                 ro->ro_rt = NULL;
1077         }
1078 #ifdef IPSEC
1079         if (sp != NULL) {
1080                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1081                         kprintf("DP ip_output call free SP:%p\n", sp));
1082                 key_freesp(sp);
1083         }
1084 #endif
1085 #ifdef FAST_IPSEC
1086         if (sp != NULL)
1087                 KEY_FREESP(&sp);
1088 #endif
1089         if (dn_mtag != NULL)
1090                 m_tag_free(dn_mtag);
1091
1092         return (error);
1093 bad:
1094         m_freem(m);
1095         goto done;
1096 }
1097
1098 /*
1099  * Create a chain of fragments which fit the given mtu. m_frag points to the
1100  * mbuf to be fragmented; on return it points to the chain with the fragments.
1101  * Return 0 if no error. If error, m_frag may contain a partially built
1102  * chain of fragments that should be freed by the caller.
1103  *
1104  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1105  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1106  */
1107 int
1108 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1109             u_long if_hwassist_flags, int sw_csum)
1110 {
1111         int error = 0;
1112         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1113         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1114         int off;
1115         struct mbuf *m0 = *m_frag;      /* the original packet          */
1116         int firstlen;
1117         struct mbuf **mnext;
1118         int nfrags;
1119
1120         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1121                 ipstat.ips_cantfrag++;
1122                 return EMSGSIZE;
1123         }
1124
1125         /*
1126          * Must be able to put at least 8 bytes per fragment.
1127          */
1128         if (len < 8)
1129                 return EMSGSIZE;
1130
1131         /*
1132          * If the interface will not calculate checksums on
1133          * fragmented packets, then do it here.
1134          */
1135         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1136             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1137                 in_delayed_cksum(m0);
1138                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1139         }
1140
1141         if (len > PAGE_SIZE) {
1142                 /*
1143                  * Fragment large datagrams such that each segment
1144                  * contains a multiple of PAGE_SIZE amount of data,
1145                  * plus headers. This enables a receiver to perform
1146                  * page-flipping zero-copy optimizations.
1147                  *
1148                  * XXX When does this help given that sender and receiver
1149                  * could have different page sizes, and also mtu could
1150                  * be less than the receiver's page size ?
1151                  */
1152                 int newlen;
1153                 struct mbuf *m;
1154
1155                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1156                         off += m->m_len;
1157
1158                 /*
1159                  * firstlen (off - hlen) must be aligned on an
1160                  * 8-byte boundary
1161                  */
1162                 if (off < hlen)
1163                         goto smart_frag_failure;
1164                 off = ((off - hlen) & ~7) + hlen;
1165                 newlen = (~PAGE_MASK) & mtu;
1166                 if ((newlen + sizeof(struct ip)) > mtu) {
1167                         /* we failed, go back the default */
1168 smart_frag_failure:
1169                         newlen = len;
1170                         off = hlen + len;
1171                 }
1172                 len = newlen;
1173
1174         } else {
1175                 off = hlen + len;
1176         }
1177
1178         firstlen = off - hlen;
1179         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1180
1181         /*
1182          * Loop through length of segment after first fragment,
1183          * make new header and copy data of each part and link onto chain.
1184          * Here, m0 is the original packet, m is the fragment being created.
1185          * The fragments are linked off the m_nextpkt of the original
1186          * packet, which after processing serves as the first fragment.
1187          */
1188         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1189                 struct ip *mhip;        /* ip header on the fragment */
1190                 struct mbuf *m;
1191                 int mhlen = sizeof(struct ip);
1192
1193                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1194                 if (m == NULL) {
1195                         error = ENOBUFS;
1196                         ipstat.ips_odropped++;
1197                         goto done;
1198                 }
1199                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1200                 /*
1201                  * In the first mbuf, leave room for the link header, then
1202                  * copy the original IP header including options. The payload
1203                  * goes into an additional mbuf chain returned by m_copy().
1204                  */
1205                 m->m_data += max_linkhdr;
1206                 mhip = mtod(m, struct ip *);
1207                 *mhip = *ip;
1208                 if (hlen > sizeof(struct ip)) {
1209                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1210                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1211                 }
1212                 m->m_len = mhlen;
1213                 /* XXX do we need to add ip->ip_off below ? */
1214                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1215                 if (off + len >= ip->ip_len) {  /* last fragment */
1216                         len = ip->ip_len - off;
1217                         m->m_flags |= M_LASTFRAG;
1218                 } else
1219                         mhip->ip_off |= IP_MF;
1220                 mhip->ip_len = htons((u_short)(len + mhlen));
1221                 m->m_next = m_copy(m0, off, len);
1222                 if (m->m_next == NULL) {                /* copy failed */
1223                         m_free(m);
1224                         error = ENOBUFS;        /* ??? */
1225                         ipstat.ips_odropped++;
1226                         goto done;
1227                 }
1228                 m->m_pkthdr.len = mhlen + len;
1229                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1230                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1231                 mhip->ip_off = htons(mhip->ip_off);
1232                 mhip->ip_sum = 0;
1233                 if (sw_csum & CSUM_DELAY_IP)
1234                         mhip->ip_sum = in_cksum(m, mhlen);
1235                 *mnext = m;
1236                 mnext = &m->m_nextpkt;
1237         }
1238         ipstat.ips_ofragments += nfrags;
1239
1240         /* set first marker for fragment chain */
1241         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1242         m0->m_pkthdr.csum_data = nfrags;
1243
1244         /*
1245          * Update first fragment by trimming what's been copied out
1246          * and updating header.
1247          */
1248         m_adj(m0, hlen + firstlen - ip->ip_len);
1249         m0->m_pkthdr.len = hlen + firstlen;
1250         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1251         ip->ip_off |= IP_MF;
1252         ip->ip_off = htons(ip->ip_off);
1253         ip->ip_sum = 0;
1254         if (sw_csum & CSUM_DELAY_IP)
1255                 ip->ip_sum = in_cksum(m0, hlen);
1256
1257 done:
1258         *m_frag = m0;
1259         return error;
1260 }
1261
1262 void
1263 in_delayed_cksum(struct mbuf *m)
1264 {
1265         struct ip *ip;
1266         u_short csum, offset;
1267
1268         ip = mtod(m, struct ip *);
1269         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1270         csum = in_cksum_skip(m, ip->ip_len, offset);
1271         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1272                 csum = 0xffff;
1273         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1274
1275         if (offset + sizeof(u_short) > m->m_len) {
1276                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1277                     m->m_len, offset, ip->ip_p);
1278                 /*
1279                  * XXX
1280                  * this shouldn't happen, but if it does, the
1281                  * correct behavior may be to insert the checksum
1282                  * in the existing chain instead of rearranging it.
1283                  */
1284                 m = m_pullup(m, offset + sizeof(u_short));
1285         }
1286         *(u_short *)(m->m_data + offset) = csum;
1287 }
1288
1289 /*
1290  * Insert IP options into preformed packet.
1291  * Adjust IP destination as required for IP source routing,
1292  * as indicated by a non-zero in_addr at the start of the options.
1293  *
1294  * XXX This routine assumes that the packet has no options in place.
1295  */
1296 static struct mbuf *
1297 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1298 {
1299         struct ipoption *p = mtod(opt, struct ipoption *);
1300         struct mbuf *n;
1301         struct ip *ip = mtod(m, struct ip *);
1302         unsigned optlen;
1303
1304         optlen = opt->m_len - sizeof p->ipopt_dst;
1305         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1306                 *phlen = 0;
1307                 return (m);             /* XXX should fail */
1308         }
1309         if (p->ipopt_dst.s_addr)
1310                 ip->ip_dst = p->ipopt_dst;
1311         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1312                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1313                 if (n == NULL) {
1314                         *phlen = 0;
1315                         return (m);
1316                 }
1317                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1318                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1319                 m->m_len -= sizeof(struct ip);
1320                 m->m_data += sizeof(struct ip);
1321                 n->m_next = m;
1322                 m = n;
1323                 m->m_len = optlen + sizeof(struct ip);
1324                 m->m_data += max_linkhdr;
1325                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1326         } else {
1327                 m->m_data -= optlen;
1328                 m->m_len += optlen;
1329                 m->m_pkthdr.len += optlen;
1330                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1331         }
1332         ip = mtod(m, struct ip *);
1333         bcopy(p->ipopt_list, ip + 1, optlen);
1334         *phlen = sizeof(struct ip) + optlen;
1335         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1336         ip->ip_len += optlen;
1337         return (m);
1338 }
1339
1340 /*
1341  * Copy options from ip to jp,
1342  * omitting those not copied during fragmentation.
1343  */
1344 int
1345 ip_optcopy(struct ip *ip, struct ip *jp)
1346 {
1347         u_char *cp, *dp;
1348         int opt, optlen, cnt;
1349
1350         cp = (u_char *)(ip + 1);
1351         dp = (u_char *)(jp + 1);
1352         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1353         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1354                 opt = cp[0];
1355                 if (opt == IPOPT_EOL)
1356                         break;
1357                 if (opt == IPOPT_NOP) {
1358                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1359                         *dp++ = IPOPT_NOP;
1360                         optlen = 1;
1361                         continue;
1362                 }
1363
1364                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1365                     ("ip_optcopy: malformed ipv4 option"));
1366                 optlen = cp[IPOPT_OLEN];
1367                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1368                     ("ip_optcopy: malformed ipv4 option"));
1369
1370                 /* bogus lengths should have been caught by ip_dooptions */
1371                 if (optlen > cnt)
1372                         optlen = cnt;
1373                 if (IPOPT_COPIED(opt)) {
1374                         bcopy(cp, dp, optlen);
1375                         dp += optlen;
1376                 }
1377         }
1378         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1379                 *dp++ = IPOPT_EOL;
1380         return (optlen);
1381 }
1382
1383 /*
1384  * IP socket option processing.
1385  */
1386 int
1387 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1388 {
1389         struct  inpcb *inp = so->so_pcb;
1390         int     error, optval;
1391
1392         error = optval = 0;
1393         if (sopt->sopt_level != IPPROTO_IP) {
1394                 return (EINVAL);
1395         }
1396
1397         switch (sopt->sopt_dir) {
1398         case SOPT_SET:
1399                 switch (sopt->sopt_name) {
1400                 case IP_OPTIONS:
1401 #ifdef notyet
1402                 case IP_RETOPTS:
1403 #endif
1404                 {
1405                         struct mbuf *m;
1406                         if (sopt->sopt_valsize > MLEN) {
1407                                 error = EMSGSIZE;
1408                                 break;
1409                         }
1410                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1411                         if (m == NULL) {
1412                                 error = ENOBUFS;
1413                                 break;
1414                         }
1415                         m->m_len = sopt->sopt_valsize;
1416                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1417                                             m->m_len);
1418
1419                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1420                                            m));
1421                 }
1422
1423                 case IP_TOS:
1424                 case IP_TTL:
1425                 case IP_MINTTL:
1426                 case IP_RECVOPTS:
1427                 case IP_RECVRETOPTS:
1428                 case IP_RECVDSTADDR:
1429                 case IP_RECVIF:
1430                 case IP_RECVTTL:
1431                 case IP_FAITH:
1432                         error = sooptcopyin(sopt, &optval, sizeof optval,
1433                                             sizeof optval);
1434                         if (error)
1435                                 break;
1436
1437                         switch (sopt->sopt_name) {
1438                         case IP_TOS:
1439                                 inp->inp_ip_tos = optval;
1440                                 break;
1441
1442                         case IP_TTL:
1443                                 inp->inp_ip_ttl = optval;
1444                                 break;
1445                         case IP_MINTTL:
1446                                 if (optval > 0 && optval <= MAXTTL)
1447                                         inp->inp_ip_minttl = optval;
1448                                 else
1449                                         error = EINVAL;
1450                                 break;
1451 #define OPTSET(bit) \
1452         if (optval) \
1453                 inp->inp_flags |= bit; \
1454         else \
1455                 inp->inp_flags &= ~bit;
1456
1457                         case IP_RECVOPTS:
1458                                 OPTSET(INP_RECVOPTS);
1459                                 break;
1460
1461                         case IP_RECVRETOPTS:
1462                                 OPTSET(INP_RECVRETOPTS);
1463                                 break;
1464
1465                         case IP_RECVDSTADDR:
1466                                 OPTSET(INP_RECVDSTADDR);
1467                                 break;
1468
1469                         case IP_RECVIF:
1470                                 OPTSET(INP_RECVIF);
1471                                 break;
1472
1473                         case IP_RECVTTL:
1474                                 OPTSET(INP_RECVTTL);
1475                                 break;
1476
1477                         case IP_FAITH:
1478                                 OPTSET(INP_FAITH);
1479                                 break;
1480                         }
1481                         break;
1482 #undef OPTSET
1483
1484                 case IP_MULTICAST_IF:
1485                 case IP_MULTICAST_VIF:
1486                 case IP_MULTICAST_TTL:
1487                 case IP_MULTICAST_LOOP:
1488                 case IP_ADD_MEMBERSHIP:
1489                 case IP_DROP_MEMBERSHIP:
1490                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1491                         break;
1492
1493                 case IP_PORTRANGE:
1494                         error = sooptcopyin(sopt, &optval, sizeof optval,
1495                                             sizeof optval);
1496                         if (error)
1497                                 break;
1498
1499                         switch (optval) {
1500                         case IP_PORTRANGE_DEFAULT:
1501                                 inp->inp_flags &= ~(INP_LOWPORT);
1502                                 inp->inp_flags &= ~(INP_HIGHPORT);
1503                                 break;
1504
1505                         case IP_PORTRANGE_HIGH:
1506                                 inp->inp_flags &= ~(INP_LOWPORT);
1507                                 inp->inp_flags |= INP_HIGHPORT;
1508                                 break;
1509
1510                         case IP_PORTRANGE_LOW:
1511                                 inp->inp_flags &= ~(INP_HIGHPORT);
1512                                 inp->inp_flags |= INP_LOWPORT;
1513                                 break;
1514
1515                         default:
1516                                 error = EINVAL;
1517                                 break;
1518                         }
1519                         break;
1520
1521 #if defined(IPSEC) || defined(FAST_IPSEC)
1522                 case IP_IPSEC_POLICY:
1523                 {
1524                         caddr_t req;
1525                         size_t len = 0;
1526                         int priv;
1527                         struct mbuf *m;
1528                         int optname;
1529
1530                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1531                                 break;
1532                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1533                                 break;
1534                         priv = (sopt->sopt_td != NULL &&
1535                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1536                         req = mtod(m, caddr_t);
1537                         len = m->m_len;
1538                         optname = sopt->sopt_name;
1539                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1540                         m_freem(m);
1541                         break;
1542                 }
1543 #endif /*IPSEC*/
1544
1545                 default:
1546                         error = ENOPROTOOPT;
1547                         break;
1548                 }
1549                 break;
1550
1551         case SOPT_GET:
1552                 switch (sopt->sopt_name) {
1553                 case IP_OPTIONS:
1554                 case IP_RETOPTS:
1555                         if (inp->inp_options)
1556                                 error = sooptcopyout(sopt,
1557                                                      mtod(inp->inp_options,
1558                                                           char *),
1559                                                      inp->inp_options->m_len);
1560                         else
1561                                 sopt->sopt_valsize = 0;
1562                         break;
1563
1564                 case IP_TOS:
1565                 case IP_TTL:
1566                 case IP_MINTTL:
1567                 case IP_RECVOPTS:
1568                 case IP_RECVRETOPTS:
1569                 case IP_RECVDSTADDR:
1570                 case IP_RECVTTL:
1571                 case IP_RECVIF:
1572                 case IP_PORTRANGE:
1573                 case IP_FAITH:
1574                         switch (sopt->sopt_name) {
1575
1576                         case IP_TOS:
1577                                 optval = inp->inp_ip_tos;
1578                                 break;
1579
1580                         case IP_TTL:
1581                                 optval = inp->inp_ip_ttl;
1582                                 break;
1583                         case IP_MINTTL:
1584                                 optval = inp->inp_ip_minttl;
1585                                 break;
1586
1587 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1588
1589                         case IP_RECVOPTS:
1590                                 optval = OPTBIT(INP_RECVOPTS);
1591                                 break;
1592
1593                         case IP_RECVRETOPTS:
1594                                 optval = OPTBIT(INP_RECVRETOPTS);
1595                                 break;
1596
1597                         case IP_RECVDSTADDR:
1598                                 optval = OPTBIT(INP_RECVDSTADDR);
1599                                 break;
1600
1601                         case IP_RECVTTL:
1602                                 optval = OPTBIT(INP_RECVTTL);
1603                                 break;
1604
1605                         case IP_RECVIF:
1606                                 optval = OPTBIT(INP_RECVIF);
1607                                 break;
1608
1609                         case IP_PORTRANGE:
1610                                 if (inp->inp_flags & INP_HIGHPORT)
1611                                         optval = IP_PORTRANGE_HIGH;
1612                                 else if (inp->inp_flags & INP_LOWPORT)
1613                                         optval = IP_PORTRANGE_LOW;
1614                                 else
1615                                         optval = 0;
1616                                 break;
1617
1618                         case IP_FAITH:
1619                                 optval = OPTBIT(INP_FAITH);
1620                                 break;
1621                         }
1622                         error = sooptcopyout(sopt, &optval, sizeof optval);
1623                         break;
1624
1625                 case IP_MULTICAST_IF:
1626                 case IP_MULTICAST_VIF:
1627                 case IP_MULTICAST_TTL:
1628                 case IP_MULTICAST_LOOP:
1629                 case IP_ADD_MEMBERSHIP:
1630                 case IP_DROP_MEMBERSHIP:
1631                         error = ip_getmoptions(sopt, inp->inp_moptions);
1632                         break;
1633
1634 #if defined(IPSEC) || defined(FAST_IPSEC)
1635                 case IP_IPSEC_POLICY:
1636                 {
1637                         struct mbuf *m = NULL;
1638                         caddr_t req = NULL;
1639                         size_t len = 0;
1640
1641                         if (m != NULL) {
1642                                 req = mtod(m, caddr_t);
1643                                 len = m->m_len;
1644                         }
1645                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1646                         if (error == 0)
1647                                 error = soopt_mcopyout(sopt, m); /* XXX */
1648                         if (error == 0)
1649                                 m_freem(m);
1650                         break;
1651                 }
1652 #endif /*IPSEC*/
1653
1654                 default:
1655                         error = ENOPROTOOPT;
1656                         break;
1657                 }
1658                 break;
1659         }
1660         return (error);
1661 }
1662
1663 /*
1664  * Set up IP options in pcb for insertion in output packets.
1665  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1666  * with destination address if source routed.
1667  */
1668 static int
1669 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1670 {
1671         int cnt, optlen;
1672         u_char *cp;
1673         u_char opt;
1674
1675         /* turn off any old options */
1676         if (*pcbopt)
1677                 m_free(*pcbopt);
1678         *pcbopt = 0;
1679         if (m == NULL || m->m_len == 0) {
1680                 /*
1681                  * Only turning off any previous options.
1682                  */
1683                 if (m != NULL)
1684                         m_free(m);
1685                 return (0);
1686         }
1687
1688         if (m->m_len % sizeof(int32_t))
1689                 goto bad;
1690         /*
1691          * IP first-hop destination address will be stored before
1692          * actual options; move other options back
1693          * and clear it when none present.
1694          */
1695         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1696                 goto bad;
1697         cnt = m->m_len;
1698         m->m_len += sizeof(struct in_addr);
1699         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1700         ovbcopy(mtod(m, caddr_t), cp, cnt);
1701         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1702
1703         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1704                 opt = cp[IPOPT_OPTVAL];
1705                 if (opt == IPOPT_EOL)
1706                         break;
1707                 if (opt == IPOPT_NOP)
1708                         optlen = 1;
1709                 else {
1710                         if (cnt < IPOPT_OLEN + sizeof *cp)
1711                                 goto bad;
1712                         optlen = cp[IPOPT_OLEN];
1713                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1714                                 goto bad;
1715                 }
1716                 switch (opt) {
1717
1718                 default:
1719                         break;
1720
1721                 case IPOPT_LSRR:
1722                 case IPOPT_SSRR:
1723                         /*
1724                          * user process specifies route as:
1725                          *      ->A->B->C->D
1726                          * D must be our final destination (but we can't
1727                          * check that since we may not have connected yet).
1728                          * A is first hop destination, which doesn't appear in
1729                          * actual IP option, but is stored before the options.
1730                          */
1731                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1732                                 goto bad;
1733                         m->m_len -= sizeof(struct in_addr);
1734                         cnt -= sizeof(struct in_addr);
1735                         optlen -= sizeof(struct in_addr);
1736                         cp[IPOPT_OLEN] = optlen;
1737                         /*
1738                          * Move first hop before start of options.
1739                          */
1740                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1741                               sizeof(struct in_addr));
1742                         /*
1743                          * Then copy rest of options back
1744                          * to close up the deleted entry.
1745                          */
1746                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1747                                 &cp[IPOPT_OFFSET+1],
1748                                 cnt - (IPOPT_MINOFF - 1));
1749                         break;
1750                 }
1751         }
1752         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1753                 goto bad;
1754         *pcbopt = m;
1755         return (0);
1756
1757 bad:
1758         m_free(m);
1759         return (EINVAL);
1760 }
1761
1762 /*
1763  * XXX
1764  * The whole multicast option thing needs to be re-thought.
1765  * Several of these options are equally applicable to non-multicast
1766  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1767  * standard option (IP_TTL).
1768  */
1769
1770 /*
1771  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1772  */
1773 static struct ifnet *
1774 ip_multicast_if(struct in_addr *a, int *ifindexp)
1775 {
1776         int ifindex;
1777         struct ifnet *ifp;
1778
1779         if (ifindexp)
1780                 *ifindexp = 0;
1781         if (ntohl(a->s_addr) >> 24 == 0) {
1782                 ifindex = ntohl(a->s_addr) & 0xffffff;
1783                 if (ifindex < 0 || if_index < ifindex)
1784                         return NULL;
1785                 ifp = ifindex2ifnet[ifindex];
1786                 if (ifindexp)
1787                         *ifindexp = ifindex;
1788         } else {
1789                 INADDR_TO_IFP(*a, ifp);
1790         }
1791         return ifp;
1792 }
1793
1794 /*
1795  * Set the IP multicast options in response to user setsockopt().
1796  */
1797 static int
1798 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1799 {
1800         int error = 0;
1801         int i;
1802         struct in_addr addr;
1803         struct ip_mreq mreq;
1804         struct ifnet *ifp;
1805         struct ip_moptions *imo = *imop;
1806         int ifindex;
1807
1808         if (imo == NULL) {
1809                 /*
1810                  * No multicast option buffer attached to the pcb;
1811                  * allocate one and initialize to default values.
1812                  */
1813                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1814
1815                 *imop = imo;
1816                 imo->imo_multicast_ifp = NULL;
1817                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1818                 imo->imo_multicast_vif = -1;
1819                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1820                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1821                 imo->imo_num_memberships = 0;
1822         }
1823
1824         switch (sopt->sopt_name) {
1825         /* store an index number for the vif you wanna use in the send */
1826         case IP_MULTICAST_VIF:
1827                 if (legal_vif_num == 0) {
1828                         error = EOPNOTSUPP;
1829                         break;
1830                 }
1831                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1832                 if (error)
1833                         break;
1834                 if (!legal_vif_num(i) && (i != -1)) {
1835                         error = EINVAL;
1836                         break;
1837                 }
1838                 imo->imo_multicast_vif = i;
1839                 break;
1840
1841         case IP_MULTICAST_IF:
1842                 /*
1843                  * Select the interface for outgoing multicast packets.
1844                  */
1845                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1846                 if (error)
1847                         break;
1848                 /*
1849                  * INADDR_ANY is used to remove a previous selection.
1850                  * When no interface is selected, a default one is
1851                  * chosen every time a multicast packet is sent.
1852                  */
1853                 if (addr.s_addr == INADDR_ANY) {
1854                         imo->imo_multicast_ifp = NULL;
1855                         break;
1856                 }
1857                 /*
1858                  * The selected interface is identified by its local
1859                  * IP address.  Find the interface and confirm that
1860                  * it supports multicasting.
1861                  */
1862                 crit_enter();
1863                 ifp = ip_multicast_if(&addr, &ifindex);
1864                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1865                         crit_exit();
1866                         error = EADDRNOTAVAIL;
1867                         break;
1868                 }
1869                 imo->imo_multicast_ifp = ifp;
1870                 if (ifindex)
1871                         imo->imo_multicast_addr = addr;
1872                 else
1873                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1874                 crit_exit();
1875                 break;
1876
1877         case IP_MULTICAST_TTL:
1878                 /*
1879                  * Set the IP time-to-live for outgoing multicast packets.
1880                  * The original multicast API required a char argument,
1881                  * which is inconsistent with the rest of the socket API.
1882                  * We allow either a char or an int.
1883                  */
1884                 if (sopt->sopt_valsize == 1) {
1885                         u_char ttl;
1886                         error = sooptcopyin(sopt, &ttl, 1, 1);
1887                         if (error)
1888                                 break;
1889                         imo->imo_multicast_ttl = ttl;
1890                 } else {
1891                         u_int ttl;
1892                         error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1893                         if (error)
1894                                 break;
1895                         if (ttl > 255)
1896                                 error = EINVAL;
1897                         else
1898                                 imo->imo_multicast_ttl = ttl;
1899                 }
1900                 break;
1901
1902         case IP_MULTICAST_LOOP:
1903                 /*
1904                  * Set the loopback flag for outgoing multicast packets.
1905                  * Must be zero or one.  The original multicast API required a
1906                  * char argument, which is inconsistent with the rest
1907                  * of the socket API.  We allow either a char or an int.
1908                  */
1909                 if (sopt->sopt_valsize == 1) {
1910                         u_char loop;
1911
1912                         error = sooptcopyin(sopt, &loop, 1, 1);
1913                         if (error)
1914                                 break;
1915                         imo->imo_multicast_loop = !!loop;
1916                 } else {
1917                         u_int loop;
1918
1919                         error = sooptcopyin(sopt, &loop, sizeof loop,
1920                                             sizeof loop);
1921                         if (error)
1922                                 break;
1923                         imo->imo_multicast_loop = !!loop;
1924                 }
1925                 break;
1926
1927         case IP_ADD_MEMBERSHIP:
1928                 /*
1929                  * Add a multicast group membership.
1930                  * Group must be a valid IP multicast address.
1931                  */
1932                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1933                 if (error)
1934                         break;
1935
1936                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1937                         error = EINVAL;
1938                         break;
1939                 }
1940                 crit_enter();
1941                 /*
1942                  * If no interface address was provided, use the interface of
1943                  * the route to the given multicast address.
1944                  */
1945                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1946                         struct sockaddr_in dst;
1947                         struct rtentry *rt;
1948
1949                         bzero(&dst, sizeof(struct sockaddr_in));
1950                         dst.sin_len = sizeof(struct sockaddr_in);
1951                         dst.sin_family = AF_INET;
1952                         dst.sin_addr = mreq.imr_multiaddr;
1953                         rt = rtlookup((struct sockaddr *)&dst);
1954                         if (rt == NULL) {
1955                                 error = EADDRNOTAVAIL;
1956                                 crit_exit();
1957                                 break;
1958                         }
1959                         --rt->rt_refcnt;
1960                         ifp = rt->rt_ifp;
1961                 } else {
1962                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1963                 }
1964
1965                 /*
1966                  * See if we found an interface, and confirm that it
1967                  * supports multicast.
1968                  */
1969                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1970                         error = EADDRNOTAVAIL;
1971                         crit_exit();
1972                         break;
1973                 }
1974                 /*
1975                  * See if the membership already exists or if all the
1976                  * membership slots are full.
1977                  */
1978                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1979                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1980                             imo->imo_membership[i]->inm_addr.s_addr
1981                                                 == mreq.imr_multiaddr.s_addr)
1982                                 break;
1983                 }
1984                 if (i < imo->imo_num_memberships) {
1985                         error = EADDRINUSE;
1986                         crit_exit();
1987                         break;
1988                 }
1989                 if (i == IP_MAX_MEMBERSHIPS) {
1990                         error = ETOOMANYREFS;
1991                         crit_exit();
1992                         break;
1993                 }
1994                 /*
1995                  * Everything looks good; add a new record to the multicast
1996                  * address list for the given interface.
1997                  */
1998                 if ((imo->imo_membership[i] =
1999                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2000                         error = ENOBUFS;
2001                         crit_exit();
2002                         break;
2003                 }
2004                 ++imo->imo_num_memberships;
2005                 crit_exit();
2006                 break;
2007
2008         case IP_DROP_MEMBERSHIP:
2009                 /*
2010                  * Drop a multicast group membership.
2011                  * Group must be a valid IP multicast address.
2012                  */
2013                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2014                 if (error)
2015                         break;
2016
2017                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2018                         error = EINVAL;
2019                         break;
2020                 }
2021
2022                 crit_enter();
2023                 /*
2024                  * If an interface address was specified, get a pointer
2025                  * to its ifnet structure.
2026                  */
2027                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2028                         ifp = NULL;
2029                 else {
2030                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2031                         if (ifp == NULL) {
2032                                 error = EADDRNOTAVAIL;
2033                                 crit_exit();
2034                                 break;
2035                         }
2036                 }
2037                 /*
2038                  * Find the membership in the membership array.
2039                  */
2040                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2041                         if ((ifp == NULL ||
2042                              imo->imo_membership[i]->inm_ifp == ifp) &&
2043                             imo->imo_membership[i]->inm_addr.s_addr ==
2044                             mreq.imr_multiaddr.s_addr)
2045                                 break;
2046                 }
2047                 if (i == imo->imo_num_memberships) {
2048                         error = EADDRNOTAVAIL;
2049                         crit_exit();
2050                         break;
2051                 }
2052                 /*
2053                  * Give up the multicast address record to which the
2054                  * membership points.
2055                  */
2056                 in_delmulti(imo->imo_membership[i]);
2057                 /*
2058                  * Remove the gap in the membership array.
2059                  */
2060                 for (++i; i < imo->imo_num_memberships; ++i)
2061                         imo->imo_membership[i-1] = imo->imo_membership[i];
2062                 --imo->imo_num_memberships;
2063                 crit_exit();
2064                 break;
2065
2066         default:
2067                 error = EOPNOTSUPP;
2068                 break;
2069         }
2070
2071         /*
2072          * If all options have default values, no need to keep the mbuf.
2073          */
2074         if (imo->imo_multicast_ifp == NULL &&
2075             imo->imo_multicast_vif == -1 &&
2076             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2077             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2078             imo->imo_num_memberships == 0) {
2079                 kfree(*imop, M_IPMOPTS);
2080                 *imop = NULL;
2081         }
2082
2083         return (error);
2084 }
2085
2086 /*
2087  * Return the IP multicast options in response to user getsockopt().
2088  */
2089 static int
2090 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2091 {
2092         struct in_addr addr;
2093         struct in_ifaddr *ia;
2094         int error, optval;
2095         u_char coptval;
2096
2097         error = 0;
2098         switch (sopt->sopt_name) {
2099         case IP_MULTICAST_VIF:
2100                 if (imo != NULL)
2101                         optval = imo->imo_multicast_vif;
2102                 else
2103                         optval = -1;
2104                 error = sooptcopyout(sopt, &optval, sizeof optval);
2105                 break;
2106
2107         case IP_MULTICAST_IF:
2108                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2109                         addr.s_addr = INADDR_ANY;
2110                 else if (imo->imo_multicast_addr.s_addr) {
2111                         /* return the value user has set */
2112                         addr = imo->imo_multicast_addr;
2113                 } else {
2114                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2115                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2116                                 : IA_SIN(ia)->sin_addr.s_addr;
2117                 }
2118                 error = sooptcopyout(sopt, &addr, sizeof addr);
2119                 break;
2120
2121         case IP_MULTICAST_TTL:
2122                 if (imo == NULL)
2123                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2124                 else
2125                         optval = coptval = imo->imo_multicast_ttl;
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         case IP_MULTICAST_LOOP:
2133                 if (imo == NULL)
2134                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2135                 else
2136                         optval = coptval = imo->imo_multicast_loop;
2137                 if (sopt->sopt_valsize == 1)
2138                         error = sooptcopyout(sopt, &coptval, 1);
2139                 else
2140                         error = sooptcopyout(sopt, &optval, sizeof optval);
2141                 break;
2142
2143         default:
2144                 error = ENOPROTOOPT;
2145                 break;
2146         }
2147         return (error);
2148 }
2149
2150 /*
2151  * Discard the IP multicast options.
2152  */
2153 void
2154 ip_freemoptions(struct ip_moptions *imo)
2155 {
2156         int i;
2157
2158         if (imo != NULL) {
2159                 for (i = 0; i < imo->imo_num_memberships; ++i)
2160                         in_delmulti(imo->imo_membership[i]);
2161                 kfree(imo, M_IPMOPTS);
2162         }
2163 }
2164
2165 /*
2166  * Routine called from ip_output() to loop back a copy of an IP multicast
2167  * packet to the input queue of a specified interface.  Note that this
2168  * calls the output routine of the loopback "driver", but with an interface
2169  * pointer that might NOT be a loopback interface -- evil, but easier than
2170  * replicating that code here.
2171  */
2172 static void
2173 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2174              int hlen)
2175 {
2176         struct ip *ip;
2177         struct mbuf *copym;
2178
2179         copym = m_copypacket(m, MB_DONTWAIT);
2180         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2181                 copym = m_pullup(copym, hlen);
2182         if (copym != NULL) {
2183                 /*
2184                  * if the checksum hasn't been computed, mark it as valid
2185                  */
2186                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2187                         in_delayed_cksum(copym);
2188                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2189                         copym->m_pkthdr.csum_flags |=
2190                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2191                         copym->m_pkthdr.csum_data = 0xffff;
2192                 }
2193                 /*
2194                  * We don't bother to fragment if the IP length is greater
2195                  * than the interface's MTU.  Can this possibly matter?
2196                  */
2197                 ip = mtod(copym, struct ip *);
2198                 ip->ip_len = htons(ip->ip_len);
2199                 ip->ip_off = htons(ip->ip_off);
2200                 ip->ip_sum = 0;
2201                 if (ip->ip_vhl == IP_VHL_BORING) {
2202                         ip->ip_sum = in_cksum_hdr(ip);
2203                 } else {
2204                         ip->ip_sum = in_cksum(copym, hlen);
2205                 }
2206                 /*
2207                  * NB:
2208                  * It's not clear whether there are any lingering
2209                  * reentrancy problems in other areas which might
2210                  * be exposed by using ip_input directly (in
2211                  * particular, everything which modifies the packet
2212                  * in-place).  Yet another option is using the
2213                  * protosw directly to deliver the looped back
2214                  * packet.  For the moment, we'll err on the side
2215                  * of safety by using if_simloop().
2216                  */
2217 #if 1 /* XXX */
2218                 if (dst->sin_family != AF_INET) {
2219                         kprintf("ip_mloopback: bad address family %d\n",
2220                                                 dst->sin_family);
2221                         dst->sin_family = AF_INET;
2222                 }
2223 #endif
2224
2225 #ifdef notdef
2226                 copym->m_pkthdr.rcvif = ifp;
2227                 ip_input(copym);
2228 #else
2229                 if_simloop(ifp, copym, dst->sin_family, 0);
2230 #endif
2231         }
2232 }