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 1.46 2008/07/07 22:02:10 nant 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 #include "opt_mpls.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/proc.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/thread2.h>
  55 #include <sys/in_cksum.h>
  56
  57 #include <net/if.h>
  58 #include <net/netisr.h>
  59 #include <net/pfil.h>
  60 #include <net/route.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/in_systm.h>
  64 #include <netinet/ip.h>
  65 #include <netinet/in_pcb.h>
  66 #include <netinet/in_var.h>
  67 #include <netinet/ip_var.h>
  68
  69 #include <netproto/mpls/mpls_var.h>
  70
  71 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  72
  73 #ifdef IPSEC
  74 #include <netinet6/ipsec.h>
  75 #include <netproto/key/key.h>
  76 #ifdef IPSEC_DEBUG
  77 #include <netproto/key/key_debug.h>
  78 #else
  79 #define KEYDEBUG(lev,arg)
  80 #endif
  81 #endif /*IPSEC*/
  82
  83 #ifdef FAST_IPSEC
  84 #include <netproto/ipsec/ipsec.h>
  85 #include <netproto/ipsec/xform.h>
  86 #include <netproto/ipsec/key.h>
  87 #endif /*FAST_IPSEC*/
  88
  89 #include <net/ipfw/ip_fw.h>
  90 #include <net/dummynet/ip_dummynet.h>
  91
  92 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  93                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  94                                   (ntohl(a.s_addr)>>16)&0xFF,\
  95                                   (ntohl(a.s_addr)>>8)&0xFF,\
  96                                   (ntohl(a.s_addr))&0xFF, y);
  97
  98 u_short ip_id;
  99
 100 #ifdef MBUF_STRESS_TEST
 101 int mbuf_frag_size = 0;
 102 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 103         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 104 #endif
 105
 106 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 107 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 108 static void     ip_mloopback
 109         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 110 static int      ip_getmoptions
 111         (struct sockopt *, struct ip_moptions *);
 112 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 113 static int      ip_setmoptions
 114         (struct sockopt *, struct ip_moptions **);
 115
 116 int     ip_optcopy(struct ip *, struct ip *);
 117
 118
 119 extern  struct protosw inetsw[];
 120
 121 /*
 122  * IP output.  The packet in mbuf chain m contains a skeletal IP
 123  * header (with len, off, ttl, proto, tos, src, dst).
 124  * The mbuf chain containing the packet will be freed.
 125  * The mbuf opt, if present, will not be freed.
 126  */
 127 int
 128 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 129           int flags, struct ip_moptions *imo, struct inpcb *inp)
 130 {
 131         struct ip *ip;
 132         struct ifnet *ifp = NULL;       /* keep compiler happy */
 133         struct mbuf *m;
 134         int hlen = sizeof(struct ip);
 135         int len, off, error = 0;
 136         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 137         struct in_ifaddr *ia = NULL;
 138         int isbroadcast, sw_csum;
 139         struct in_addr pkt_dst;
 140         struct route iproute;
 141         struct m_tag *dn_mtag = NULL;
 142 #ifdef IPSEC
 143         struct secpolicy *sp = NULL;
 144         struct socket *so = inp ? inp->inp_socket : NULL;
 145 #endif
 146 #ifdef FAST_IPSEC
 147         struct m_tag *mtag;
 148         struct secpolicy *sp = NULL;
 149         struct tdb_ident *tdbi;
 150 #endif /* FAST_IPSEC */
 151         struct ip_fw_args args;
 152         int src_was_INADDR_ANY = 0;     /* as the name says... */
 153
 154         args.eh = NULL;
 155         args.rule = NULL;
 156         args.next_hop = NULL;
 157
 158         /* Grab info from MT_TAG mbufs prepended to the chain. */
 159         while (m0 != NULL && m0->m_type == MT_TAG) {
 160                 switch(m0->_m_tag_id) {
 161                 case PACKET_TAG_IPFORWARD:
 162                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 163                         break;
 164                 default:
 165                         kprintf("ip_output: unrecognised MT_TAG tag %d\n",
 166                             m0->_m_tag_id);
 167                         break;
 168                 }
 169                 m0 = m0->m_next;
 170         }
 171         m = m0;
 172         M_ASSERTPKTHDR(m);
 173
 174         /* Extract info from dummynet tag */
 175         dn_mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 176         if (dn_mtag != NULL) {
 177                 struct dn_pkt *dn_pkt = m_tag_data(dn_mtag);
 178
 179                 /*
 180                  * The packet was already tagged, so part of the
 181                  * processing was already done, and we need to go down.
 182                  * Get parameters from the tag.
 183                  */
 184                 args.rule = dn_pkt->dn_priv;
 185                 opt = NULL;
 186                 ro = &dn_pkt->ro;
 187                 imo = NULL;
 188                 dst = dn_pkt->dn_dst;
 189                 ifp = dn_pkt->ifp;
 190                 flags = dn_pkt->flags;
 191
 192                 /*
 193                  * Don't delete the dummynet tag here, just unlink it,
 194                  * since some local variables (like 'ro' and 'dst') are
 195                  * still referencing certain parts of it.
 196                  * The dummynet tag will be freed at the end of the
 197                  * output process.
 198                  */
 199                 m_tag_unlink(m, dn_mtag);
 200         }
 201
 202         if (ro == NULL) {
 203                 ro = &iproute;
 204                 bzero(ro, sizeof *ro);
 205         }
 206
 207         if (args.rule != NULL) {        /* dummynet already saw us */
 208                 ip = mtod(m, struct ip *);
 209                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 210                 if (ro->ro_rt)
 211                         ia = ifatoia(ro->ro_rt->rt_ifa);
 212                 goto sendit;
 213         }
 214
 215         if (opt) {
 216                 len = 0;
 217                 m = ip_insertoptions(m, opt, &len);
 218                 if (len != 0)
 219                         hlen = len;
 220         }
 221         ip = mtod(m, struct ip *);
 222         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 223
 224         /*
 225          * Fill in IP header.
 226          */
 227         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 228                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 229                 ip->ip_off &= IP_DF;
 230                 ip->ip_id = ip_newid();
 231                 ipstat.ips_localout++;
 232         } else {
 233                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 234         }
 235
 236         dst = (struct sockaddr_in *)&ro->ro_dst;
 237         /*
 238          * If there is a cached route,
 239          * check that it is to the same destination
 240          * and is still up.  If not, free it and try again.
 241          * The address family should also be checked in case of sharing the
 242          * cache with IPv6.
 243          */
 244         if (ro->ro_rt &&
 245             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 246              dst->sin_family != AF_INET ||
 247              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 248                 rtfree(ro->ro_rt);
 249                 ro->ro_rt = (struct rtentry *)NULL;
 250         }
 251         if (ro->ro_rt == NULL) {
 252                 bzero(dst, sizeof *dst);
 253                 dst->sin_family = AF_INET;
 254                 dst->sin_len = sizeof *dst;
 255                 dst->sin_addr = pkt_dst;
 256         }
 257         /*
 258          * If routing to interface only,
 259          * short circuit routing lookup.
 260          */
 261         if (flags & IP_ROUTETOIF) {
 262                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 263                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 264                         ipstat.ips_noroute++;
 265                         error = ENETUNREACH;
 266                         goto bad;
 267                 }
 268                 ifp = ia->ia_ifp;
 269                 ip->ip_ttl = 1;
 270                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 271         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 272                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 273                 /*
 274                  * Bypass the normal routing lookup for multicast
 275                  * packets if the interface is specified.
 276                  */
 277                 ifp = imo->imo_multicast_ifp;
 278                 ia = IFP_TO_IA(ifp);
 279                 isbroadcast = 0;        /* fool gcc */
 280         } else {
 281                 /*
 282                  * If this is the case, we probably don't want to allocate
 283                  * a protocol-cloned route since we didn't get one from the
 284                  * ULP.  This lets TCP do its thing, while not burdening
 285                  * forwarding or ICMP with the overhead of cloning a route.
 286                  * Of course, we still want to do any cloning requested by
 287                  * the link layer, as this is probably required in all cases
 288                  * for correct operation (as it is for ARP).
 289                  */
 290                 if (ro->ro_rt == NULL)
 291                         rtalloc_ign(ro, RTF_PRCLONING);
 292                 if (ro->ro_rt == NULL) {
 293                         ipstat.ips_noroute++;
 294                         error = EHOSTUNREACH;
 295                         goto bad;
 296                 }
 297                 ia = ifatoia(ro->ro_rt->rt_ifa);
 298                 ifp = ro->ro_rt->rt_ifp;
 299                 ro->ro_rt->rt_use++;
 300                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 301                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 302                 if (ro->ro_rt->rt_flags & RTF_HOST)
 303                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 304                 else
 305                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 306         }
 307         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 308                 struct in_multi *inm;
 309
 310                 m->m_flags |= M_MCAST;
 311                 /*
 312                  * IP destination address is multicast.  Make sure "dst"
 313                  * still points to the address in "ro".  (It may have been
 314                  * changed to point to a gateway address, above.)
 315                  */
 316                 dst = (struct sockaddr_in *)&ro->ro_dst;
 317                 /*
 318                  * See if the caller provided any multicast options
 319                  */
 320                 if (imo != NULL) {
 321                         ip->ip_ttl = imo->imo_multicast_ttl;
 322                         if (imo->imo_multicast_vif != -1)
 323                                 ip->ip_src.s_addr =
 324                                     ip_mcast_src ?
 325                                     ip_mcast_src(imo->imo_multicast_vif) :
 326                                     INADDR_ANY;
 327                 } else
 328                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 329                 /*
 330                  * Confirm that the outgoing interface supports multicast.
 331                  */
 332                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 333                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 334                                 ipstat.ips_noroute++;
 335                                 error = ENETUNREACH;
 336                                 goto bad;
 337                         }
 338                 }
 339                 /*
 340                  * If source address not specified yet, use address
 341                  * of outgoing interface.
 342                  */
 343                 if (ip->ip_src.s_addr == INADDR_ANY) {
 344                         /* Interface may have no addresses. */
 345                         if (ia != NULL)
 346                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 347                 }
 348
 349                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 350                 if (inm != NULL &&
 351                    (imo == NULL || imo->imo_multicast_loop)) {
 352                         /*
 353                          * If we belong to the destination multicast group
 354                          * on the outgoing interface, and the caller did not
 355                          * forbid loopback, loop back a copy.
 356                          */
 357                         ip_mloopback(ifp, m, dst, hlen);
 358                 }
 359                 else {
 360                         /*
 361                          * If we are acting as a multicast router, perform
 362                          * multicast forwarding as if the packet had just
 363                          * arrived on the interface to which we are about
 364                          * to send.  The multicast forwarding function
 365                          * recursively calls this function, using the
 366                          * IP_FORWARDING flag to prevent infinite recursion.
 367                          *
 368                          * Multicasts that are looped back by ip_mloopback(),
 369                          * above, will be forwarded by the ip_input() routine,
 370                          * if necessary.
 371                          */
 372                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 373                                 /*
 374                                  * If rsvp daemon is not running, do not
 375                                  * set ip_moptions. This ensures that the packet
 376                                  * is multicast and not just sent down one link
 377                                  * as prescribed by rsvpd.
 378                                  */
 379                                 if (!rsvp_on)
 380                                         imo = NULL;
 381                                 if (ip_mforward &&
 382                                     ip_mforward(ip, ifp, m, imo) != 0) {
 383                                         m_freem(m);
 384                                         goto done;
 385                                 }
 386                         }
 387                 }
 388
 389                 /*
 390                  * Multicasts with a time-to-live of zero may be looped-
 391                  * back, above, but must not be transmitted on a network.
 392                  * Also, multicasts addressed to the loopback interface
 393                  * are not sent -- the above call to ip_mloopback() will
 394                  * loop back a copy if this host actually belongs to the
 395                  * destination group on the loopback interface.
 396                  */
 397                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 398                         m_freem(m);
 399                         goto done;
 400                 }
 401
 402                 goto sendit;
 403         }
 404 #ifndef notdef
 405         /*
 406          * If the source address is not specified yet, use the address
 407          * of the outoing interface. In case, keep note we did that, so
 408          * if the the firewall changes the next-hop causing the output
 409          * interface to change, we can fix that.
 410          */
 411         if (ip->ip_src.s_addr == INADDR_ANY) {
 412                 /* Interface may have no addresses. */
 413                 if (ia != NULL) {
 414                         ip->ip_src = IA_SIN(ia)->sin_addr;
 415                         src_was_INADDR_ANY = 1;
 416                 }
 417         }
 418 #endif /* notdef */
 419 #ifdef ALTQ
 420         /*
 421          * Disable packet drop hack.
 422          * Packetdrop should be done by queueing.
 423          */
 424 #else /* !ALTQ */
 425         /*
 426          * Verify that we have any chance at all of being able to queue
 427          *      the packet or packet fragments
 428          */
 429         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 430                 ifp->if_snd.ifq_maxlen) {
 431                         error = ENOBUFS;
 432                         ipstat.ips_odropped++;
 433                         goto bad;
 434         }
 435 #endif /* !ALTQ */
 436
 437         /*
 438          * Look for broadcast address and
 439          * verify user is allowed to send
 440          * such a packet.
 441          */
 442         if (isbroadcast) {
 443                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 444                         error = EADDRNOTAVAIL;
 445                         goto bad;
 446                 }
 447                 if (!(flags & IP_ALLOWBROADCAST)) {
 448                         error = EACCES;
 449                         goto bad;
 450                 }
 451                 /* don't allow broadcast messages to be fragmented */
 452                 if (ip->ip_len > ifp->if_mtu) {
 453                         error = EMSGSIZE;
 454                         goto bad;
 455                 }
 456                 m->m_flags |= M_BCAST;
 457         } else {
 458                 m->m_flags &= ~M_BCAST;
 459         }
 460
 461 sendit:
 462 #ifdef IPSEC
 463         /* get SP for this packet */
 464         if (so == NULL)
 465                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 466         else
 467                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 468
 469         if (sp == NULL) {
 470                 ipsecstat.out_inval++;
 471                 goto bad;
 472         }
 473
 474         error = 0;
 475
 476         /* check policy */
 477         switch (sp->policy) {
 478         case IPSEC_POLICY_DISCARD:
 479                 /*
 480                  * This packet is just discarded.
 481                  */
 482                 ipsecstat.out_polvio++;
 483                 goto bad;
 484
 485         case IPSEC_POLICY_BYPASS:
 486         case IPSEC_POLICY_NONE:
 487                 /* no need to do IPsec. */
 488                 goto skip_ipsec;
 489
 490         case IPSEC_POLICY_IPSEC:
 491                 if (sp->req == NULL) {
 492                         /* acquire a policy */
 493                         error = key_spdacquire(sp);
 494                         goto bad;
 495                 }
 496                 break;
 497
 498         case IPSEC_POLICY_ENTRUST:
 499         default:
 500                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 501         }
 502     {
 503         struct ipsec_output_state state;
 504         bzero(&state, sizeof state);
 505         state.m = m;
 506         if (flags & IP_ROUTETOIF) {
 507                 state.ro = &iproute;
 508                 bzero(&iproute, sizeof iproute);
 509         } else
 510                 state.ro = ro;
 511         state.dst = (struct sockaddr *)dst;
 512
 513         ip->ip_sum = 0;
 514
 515         /*
 516          * XXX
 517          * delayed checksums are not currently compatible with IPsec
 518          */
 519         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 520                 in_delayed_cksum(m);
 521                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 522         }
 523
 524         ip->ip_len = htons(ip->ip_len);
 525         ip->ip_off = htons(ip->ip_off);
 526
 527         error = ipsec4_output(&state, sp, flags);
 528
 529         m = state.m;
 530         if (flags & IP_ROUTETOIF) {
 531                 /*
 532                  * if we have tunnel mode SA, we may need to ignore
 533                  * IP_ROUTETOIF.
 534                  */
 535                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 536                         flags &= ~IP_ROUTETOIF;
 537                         ro = state.ro;
 538                 }
 539         } else
 540                 ro = state.ro;
 541         dst = (struct sockaddr_in *)state.dst;
 542         if (error) {
 543                 /* mbuf is already reclaimed in ipsec4_output. */
 544                 m0 = NULL;
 545                 switch (error) {
 546                 case EHOSTUNREACH:
 547                 case ENETUNREACH:
 548                 case EMSGSIZE:
 549                 case ENOBUFS:
 550                 case ENOMEM:
 551                         break;
 552                 default:
 553                         kprintf("ip4_output (ipsec): error code %d\n", error);
 554                         /*fall through*/
 555                 case ENOENT:
 556                         /* don't show these error codes to the user */
 557                         error = 0;
 558                         break;
 559                 }
 560                 goto bad;
 561         }
 562     }
 563
 564         /* be sure to update variables that are affected by ipsec4_output() */
 565         ip = mtod(m, struct ip *);
 566 #ifdef _IP_VHL
 567         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 568 #else
 569         hlen = ip->ip_hl << 2;
 570 #endif
 571         if (ro->ro_rt == NULL) {
 572                 if (!(flags & IP_ROUTETOIF)) {
 573                         kprintf("ip_output: "
 574                                 "can't update route after IPsec processing\n");
 575                         error = EHOSTUNREACH;   /*XXX*/
 576                         goto bad;
 577                 }
 578         } else {
 579                 ia = ifatoia(ro->ro_rt->rt_ifa);
 580                 ifp = ro->ro_rt->rt_ifp;
 581         }
 582
 583         /* make it flipped, again. */
 584         ip->ip_len = ntohs(ip->ip_len);
 585         ip->ip_off = ntohs(ip->ip_off);
 586 skip_ipsec:
 587 #endif /*IPSEC*/
 588 #ifdef FAST_IPSEC
 589         /*
 590          * Check the security policy (SP) for the packet and, if
 591          * required, do IPsec-related processing.  There are two
 592          * cases here; the first time a packet is sent through
 593          * it will be untagged and handled by ipsec4_checkpolicy.
 594          * If the packet is resubmitted to ip_output (e.g. after
 595          * AH, ESP, etc. processing), there will be a tag to bypass
 596          * the lookup and related policy checking.
 597          */
 598         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 599         crit_enter();
 600         if (mtag != NULL) {
 601                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 602                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 603                 if (sp == NULL)
 604                         error = -EINVAL;        /* force silent drop */
 605                 m_tag_delete(m, mtag);
 606         } else {
 607                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 608                                         &error, inp);
 609         }
 610         /*
 611          * There are four return cases:
 612          *    sp != NULL                    apply IPsec policy
 613          *    sp == NULL, error == 0        no IPsec handling needed
 614          *    sp == NULL, error == -EINVAL  discard packet w/o error
 615          *    sp == NULL, error != 0        discard packet, report error
 616          */
 617         if (sp != NULL) {
 618                 /* Loop detection, check if ipsec processing already done */
 619                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 620                 for (mtag = m_tag_first(m); mtag != NULL;
 621                      mtag = m_tag_next(m, mtag)) {
 622                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 623                                 continue;
 624                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 625                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 626                                 continue;
 627                         /*
 628                          * Check if policy has an SA associated with it.
 629                          * This can happen when an SP has yet to acquire
 630                          * an SA; e.g. on first reference.  If it occurs,
 631                          * then we let ipsec4_process_packet do its thing.
 632                          */
 633                         if (sp->req->sav == NULL)
 634                                 break;
 635                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 636                         if (tdbi->spi == sp->req->sav->spi &&
 637                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 638                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 639                                  sizeof(union sockaddr_union)) == 0) {
 640                                 /*
 641                                  * No IPsec processing is needed, free
 642                                  * reference to SP.
 643                                  *
 644                                  * NB: null pointer to avoid free at
 645                                  *     done: below.
 646                                  */
 647                                 KEY_FREESP(&sp), sp = NULL;
 648                                 crit_exit();
 649                                 goto spd_done;
 650                         }
 651                 }
 652
 653                 /*
 654                  * Do delayed checksums now because we send before
 655                  * this is done in the normal processing path.
 656                  */
 657                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 658                         in_delayed_cksum(m);
 659                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 660                 }
 661
 662                 ip->ip_len = htons(ip->ip_len);
 663                 ip->ip_off = htons(ip->ip_off);
 664
 665                 /* NB: callee frees mbuf */
 666                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 667                 /*
 668                  * Preserve KAME behaviour: ENOENT can be returned
 669                  * when an SA acquire is in progress.  Don't propagate
 670                  * this to user-level; it confuses applications.
 671                  *
 672                  * XXX this will go away when the SADB is redone.
 673                  */
 674                 if (error == ENOENT)
 675                         error = 0;
 676                 crit_exit();
 677                 goto done;
 678         } else {
 679                 crit_exit();
 680
 681                 if (error != 0) {
 682                         /*
 683                          * Hack: -EINVAL is used to signal that a packet
 684                          * should be silently discarded.  This is typically
 685                          * because we asked key management for an SA and
 686                          * it was delayed (e.g. kicked up to IKE).
 687                          */
 688                         if (error == -EINVAL)
 689                                 error = 0;
 690                         goto bad;
 691                 } else {
 692                         /* No IPsec processing for this packet. */
 693                 }
 694 #ifdef notyet
 695                 /*
 696                  * If deferred crypto processing is needed, check that
 697                  * the interface supports it.
 698                  */
 699                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 700                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 701                         /* notify IPsec to do its own crypto */
 702                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 703                         error = EHOSTUNREACH;
 704                         goto bad;
 705                 }
 706 #endif
 707         }
 708 spd_done:
 709 #endif /* FAST_IPSEC */
 710         /*
 711          * IpHack's section.
 712          * - Xlate: translate packet's addr/port (NAT).
 713          * - Firewall: deny/allow/etc.
 714          * - Wrap: fake packet's addr/port <unimpl.>
 715          * - Encapsulate: put it in another IP and send out. <unimp.>
 716          */
 717
 718         /*
 719          * Run through list of hooks for output packets.
 720          */
 721         if (pfil_has_hooks(&inet_pfil_hook)) {
 722                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 723                 if (error != 0 || m == NULL)
 724                         goto done;
 725                 ip = mtod(m, struct ip *);
 726         }
 727
 728         /*
 729          * Check with the firewall...
 730          * but not if we are already being fwd'd from a firewall.
 731          */
 732         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 733                 struct sockaddr_in *old = dst;
 734
 735                 args.m = m;
 736                 args.next_hop = dst;
 737                 args.oif = ifp;
 738                 off = ip_fw_chk_ptr(&args);
 739                 m = args.m;
 740                 dst = args.next_hop;
 741
 742                 /*
 743                  * On return we must do the following:
 744                  * m == NULL    -> drop the pkt (old interface, deprecated)
 745                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 746                  * 1<=off<= 0xffff              -> DIVERT
 747                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 748                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 749                  * dst != old                   -> IPFIREWALL_FORWARD
 750                  * off==0, dst==old             -> accept
 751                  * If some of the above modules are not compiled in, then
 752                  * we should't have to check the corresponding condition
 753                  * (because the ipfw control socket should not accept
 754                  * unsupported rules), but better play safe and drop
 755                  * packets in case of doubt.
 756                  */
 757                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 758                         if (m)
 759                                 m_freem(m);
 760                         error = EACCES;
 761                         goto done;
 762                 }
 763                 ip = mtod(m, struct ip *);
 764                 if (off == 0 && dst == old)             /* common case */
 765                         goto pass;
 766                 if (off & IP_FW_PORT_DYNT_FLAG) {
 767                         /*
 768                          * pass the pkt to dummynet. Need to include
 769                          * pipe number, m, ifp, ro, dst because these are
 770                          * not recomputed in the next pass.
 771                          * All other parameters have been already used and
 772                          * so they are not needed anymore.
 773                          * XXX note: if the ifp or ro entry are deleted
 774                          * while a pkt is in dummynet, we are in trouble!
 775                          */
 776                         args.ro = ro;
 777                         args.dst = dst;
 778                         args.flags = flags;
 779
 780                         error = 0;
 781                         ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
 782                         goto done;
 783                 }
 784 #ifdef IPDIVERT
 785                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
 786                         struct mbuf *clone = NULL;
 787
 788                         /* Clone packet if we're doing a 'tee' */
 789                         if ((off & IP_FW_PORT_TEE_FLAG))
 790                                 clone = m_dup(m, MB_DONTWAIT);
 791
 792                         /*
 793                          * XXX
 794                          * delayed checksums are not currently compatible
 795                          * with divert sockets.
 796                          */
 797                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 798                                 in_delayed_cksum(m);
 799                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 800                         }
 801
 802                         /* Restore packet header fields to original values */
 803                         ip->ip_len = htons(ip->ip_len);
 804                         ip->ip_off = htons(ip->ip_off);
 805
 806                         /* Deliver packet to divert input routine */
 807                         divert_packet(m, 0, off & 0xffff);
 808
 809                         /* If 'tee', continue with original packet */
 810                         if (clone != NULL) {
 811                                 m = clone;
 812                                 ip = mtod(m, struct ip *);
 813                                 goto pass;
 814                         }
 815                         goto done;
 816                 }
 817 #endif
 818
 819                 /* IPFIREWALL_FORWARD */
 820                 /*
 821                  * Check dst to make sure it is directly reachable on the
 822                  * interface we previously thought it was.
 823                  * If it isn't (which may be likely in some situations) we have
 824                  * to re-route it (ie, find a route for the next-hop and the
 825                  * associated interface) and set them here. This is nested
 826                  * forwarding which in most cases is undesirable, except where
 827                  * such control is nigh impossible. So we do it here.
 828                  * And I'm babbling.
 829                  */
 830                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 831 #if 0
 832                         /*
 833                          * XXX To improve readability, this block should be
 834                          * changed into a function call as below:
 835                          */
 836                         error = ip_ipforward(&m, &dst, &ifp);
 837                         if (error)
 838                                 goto bad;
 839                         if (m == NULL) /* ip_input consumed the mbuf */
 840                                 goto done;
 841 #else
 842                         struct in_ifaddr *ia;
 843                         struct in_ifaddr_container *iac;
 844
 845                         /*
 846                          * XXX sro_fwd below is static, and a pointer
 847                          * to it gets passed to routines downstream.
 848                          * This could have surprisingly bad results in
 849                          * practice, because its content is overwritten
 850                          * by subsequent packets.
 851                          */
 852                         /* There must be a better way to do this next line... */
 853                         static struct route sro_fwd;
 854                         struct route *ro_fwd = &sro_fwd;
 855
 856 #if 0
 857                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 858                             dst->sin_addr, "\n");
 859 #endif
 860
 861                         /*
 862                          * We need to figure out if we have been forwarded
 863                          * to a local socket. If so, then we should somehow
 864                          * "loop back" to ip_input, and get directed to the
 865                          * PCB as if we had received this packet. This is
 866                          * because it may be dificult to identify the packets
 867                          * you want to forward until they are being output
 868                          * and have selected an interface. (e.g. locally
 869                          * initiated packets) If we used the loopback inteface,
 870                          * we would not be able to control what happens
 871                          * as the packet runs through ip_input() as
 872                          * it is done through a ISR.
 873                          */
 874                         ia = NULL;
 875                         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr),
 876                                      ia_hash) {
 877                                 /*
 878                                  * If the addr to forward to is one
 879                                  * of ours, we pretend to
 880                                  * be the destination for this packet.
 881                                  */
 882                                 if (IA_SIN(iac->ia)->sin_addr.s_addr ==
 883                                     dst->sin_addr.s_addr) {
 884                                         ia = iac->ia;
 885                                         break;
 886                                 }
 887                         }
 888                         if (ia != NULL) {    /* tell ip_input "dont filter" */
 889                                 struct m_hdr tag;
 890
 891                                 tag.mh_type = MT_TAG;
 892                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 893                                 tag.mh_data = (caddr_t)args.next_hop;
 894                                 tag.mh_next = m;
 895
 896                                 if (m->m_pkthdr.rcvif == NULL)
 897                                         m->m_pkthdr.rcvif = ifunit("lo0");
 898                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 899                                         m->m_pkthdr.csum_flags |=
 900                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 901                                         m->m_pkthdr.csum_data = 0xffff;
 902                                 }
 903                                 m->m_pkthdr.csum_flags |=
 904                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 905                                 ip->ip_len = htons(ip->ip_len);
 906                                 ip->ip_off = htons(ip->ip_off);
 907                                 ip_input((struct mbuf *)&tag);
 908                                 goto done;
 909                         }
 910                         /* Some of the logic for this was nicked from above.
 911                          *
 912                          * This rewrites the cached route in a local PCB.
 913                          * Is this what we want to do?
 914                          */
 915                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
 916                         ro_fwd->ro_rt = NULL;
 917
 918                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 919                         if (ro_fwd->ro_rt == NULL) {
 920                                 ipstat.ips_noroute++;
 921                                 error = EHOSTUNREACH;
 922                                 goto bad;
 923                         }
 924
 925                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 926                         ifp = ro_fwd->ro_rt->rt_ifp;
 927                         ro_fwd->ro_rt->rt_use++;
 928                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 929                                 dst = (struct sockaddr_in *)
 930                                     ro_fwd->ro_rt->rt_gateway;
 931                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 932                                 isbroadcast =
 933                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 934                         else
 935                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 936                         if (ro->ro_rt != NULL)
 937                                 rtfree(ro->ro_rt);
 938                         ro->ro_rt = ro_fwd->ro_rt;
 939                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 940
 941 #endif  /* ... block to be put into a function */
 942                         /*
 943                          * If we added a default src ip earlier,
 944                          * which would have been gotten from the-then
 945                          * interface, do it again, from the new one.
 946                          */
 947                         if (src_was_INADDR_ANY)
 948                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 949                         goto pass ;
 950                 }
 951
 952                 /*
 953                  * if we get here, none of the above matches, and
 954                  * we have to drop the pkt
 955                  */
 956                 m_freem(m);
 957                 error = EACCES; /* not sure this is the right error msg */
 958                 goto done;
 959         }
 960
 961 pass:
 962         /* 127/8 must not appear on wire - RFC1122. */
 963         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 964             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 965                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 966                         ipstat.ips_badaddr++;
 967                         error = EADDRNOTAVAIL;
 968                         goto bad;
 969                 }
 970         }
 971
 972         m->m_pkthdr.csum_flags |= CSUM_IP;
 973         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 974         if (sw_csum & CSUM_DELAY_DATA) {
 975                 in_delayed_cksum(m);
 976                 sw_csum &= ~CSUM_DELAY_DATA;
 977         }
 978         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 979
 980         /*
 981          * If small enough for interface, or the interface will take
 982          * care of the fragmentation for us, can just send directly.
 983          */
 984         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 985             !(ip->ip_off & IP_DF))) {
 986                 ip->ip_len = htons(ip->ip_len);
 987                 ip->ip_off = htons(ip->ip_off);
 988                 ip->ip_sum = 0;
 989                 if (sw_csum & CSUM_DELAY_IP) {
 990                         if (ip->ip_vhl == IP_VHL_BORING) {
 991                                 ip->ip_sum = in_cksum_hdr(ip);
 992                         } else {
 993                                 ip->ip_sum = in_cksum(m, hlen);
 994                         }
 995                 }
 996
 997                 /* Record statistics for this interface address. */
 998                 if (!(flags & IP_FORWARDING) && ia) {
 999                         ia->ia_ifa.if_opackets++;
1000                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1001                 }
1002
1003 #ifdef IPSEC
1004                 /* clean ipsec history once it goes out of the node */
1005                 ipsec_delaux(m);
1006 #endif
1007
1008 #ifdef MBUF_STRESS_TEST
1009                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1010                         struct mbuf *m1, *m2;
1011                         int length, tmp;
1012
1013                         tmp = length = m->m_pkthdr.len;
1014
1015                         while ((length -= mbuf_frag_size) >= 1) {
1016                                 m1 = m_split(m, length, MB_DONTWAIT);
1017                                 if (m1 == NULL)
1018                                         break;
1019                                 m2 = m;
1020                                 while (m2->m_next != NULL)
1021                                         m2 = m2->m_next;
1022                                 m2->m_next = m1;
1023                         }
1024                         m->m_pkthdr.len = tmp;
1025                 }
1026 #endif
1027
1028 #ifdef MPLS
1029                 struct rtentry *send_route = ro->ro_rt; /* copy-in/copy-out parameter */
1030
1031                 if (!mpls_output_process(ifp, m, &dst, send_route))
1032                         goto done;
1033 #endif
1034                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1035                                        ro->ro_rt);
1036                 goto done;
1037         }
1038
1039         if (ip->ip_off & IP_DF) {
1040                 error = EMSGSIZE;
1041                 /*
1042                  * This case can happen if the user changed the MTU
1043                  * of an interface after enabling IP on it.  Because
1044                  * most netifs don't keep track of routes pointing to
1045                  * them, there is no way for one to update all its
1046                  * routes when the MTU is changed.
1047                  */
1048                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1049                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1050                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1051                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1052                 }
1053                 ipstat.ips_cantfrag++;
1054                 goto bad;
1055         }
1056
1057         /*
1058          * Too large for interface; fragment if possible. If successful,
1059          * on return, m will point to a list of packets to be sent.
1060          */
1061         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1062         if (error)
1063                 goto bad;
1064         for (; m; m = m0) {
1065                 m0 = m->m_nextpkt;
1066                 m->m_nextpkt = NULL;
1067 #ifdef IPSEC
1068                 /* clean ipsec history once it goes out of the node */
1069                 ipsec_delaux(m);
1070 #endif
1071                 if (error == 0) {
1072                         /* Record statistics for this interface address. */
1073                         if (ia != NULL) {
1074                                 ia->ia_ifa.if_opackets++;
1075                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1076                         }
1077 #ifdef MPLS
1078                 struct rtentry *send_route = ro->ro_rt; /* copy-in/copy-out parameter */
1079
1080                 if (!mpls_output_process(ifp, m, &dst, send_route))
1081                         goto done;
1082 #endif
1083                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1084                                                ro->ro_rt);
1085                 } else {
1086                         m_freem(m);
1087                 }
1088         }
1089
1090         if (error == 0)
1091                 ipstat.ips_fragmented++;
1092
1093 done:
1094         if (ro == &iproute && ro->ro_rt != NULL) {
1095                 RTFREE(ro->ro_rt);
1096                 ro->ro_rt = NULL;
1097         }
1098 #ifdef IPSEC
1099         if (sp != NULL) {
1100                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1101                         kprintf("DP ip_output call free SP:%p\n", sp));
1102                 key_freesp(sp);
1103         }
1104 #endif
1105 #ifdef FAST_IPSEC
1106         if (sp != NULL)
1107                 KEY_FREESP(&sp);
1108 #endif
1109         if (dn_mtag != NULL)
1110                 m_tag_free(dn_mtag);
1111
1112         return (error);
1113 bad:
1114         m_freem(m);
1115         goto done;
1116 }
1117
1118 /*
1119  * Create a chain of fragments which fit the given mtu. m_frag points to the
1120  * mbuf to be fragmented; on return it points to the chain with the fragments.
1121  * Return 0 if no error. If error, m_frag may contain a partially built
1122  * chain of fragments that should be freed by the caller.
1123  *
1124  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1125  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1126  */
1127 int
1128 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1129             u_long if_hwassist_flags, int sw_csum)
1130 {
1131         int error = 0;
1132         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1133         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1134         int off;
1135         struct mbuf *m0 = *m_frag;      /* the original packet          */
1136         int firstlen;
1137         struct mbuf **mnext;
1138         int nfrags;
1139
1140         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1141                 ipstat.ips_cantfrag++;
1142                 return EMSGSIZE;
1143         }
1144
1145         /*
1146          * Must be able to put at least 8 bytes per fragment.
1147          */
1148         if (len < 8)
1149                 return EMSGSIZE;
1150
1151         /*
1152          * If the interface will not calculate checksums on
1153          * fragmented packets, then do it here.
1154          */
1155         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1156             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1157                 in_delayed_cksum(m0);
1158                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1159         }
1160
1161         if (len > PAGE_SIZE) {
1162                 /*
1163                  * Fragment large datagrams such that each segment
1164                  * contains a multiple of PAGE_SIZE amount of data,
1165                  * plus headers. This enables a receiver to perform
1166                  * page-flipping zero-copy optimizations.
1167                  *
1168                  * XXX When does this help given that sender and receiver
1169                  * could have different page sizes, and also mtu could
1170                  * be less than the receiver's page size ?
1171                  */
1172                 int newlen;
1173                 struct mbuf *m;
1174
1175                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1176                         off += m->m_len;
1177
1178                 /*
1179                  * firstlen (off - hlen) must be aligned on an
1180                  * 8-byte boundary
1181                  */
1182                 if (off < hlen)
1183                         goto smart_frag_failure;
1184                 off = ((off - hlen) & ~7) + hlen;
1185                 newlen = (~PAGE_MASK) & mtu;
1186                 if ((newlen + sizeof(struct ip)) > mtu) {
1187                         /* we failed, go back the default */
1188 smart_frag_failure:
1189                         newlen = len;
1190                         off = hlen + len;
1191                 }
1192                 len = newlen;
1193
1194         } else {
1195                 off = hlen + len;
1196         }
1197
1198         firstlen = off - hlen;
1199         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1200
1201         /*
1202          * Loop through length of segment after first fragment,
1203          * make new header and copy data of each part and link onto chain.
1204          * Here, m0 is the original packet, m is the fragment being created.
1205          * The fragments are linked off the m_nextpkt of the original
1206          * packet, which after processing serves as the first fragment.
1207          */
1208         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1209                 struct ip *mhip;        /* ip header on the fragment */
1210                 struct mbuf *m;
1211                 int mhlen = sizeof(struct ip);
1212
1213                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1214                 if (m == NULL) {
1215                         error = ENOBUFS;
1216                         ipstat.ips_odropped++;
1217                         goto done;
1218                 }
1219                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1220                 /*
1221                  * In the first mbuf, leave room for the link header, then
1222                  * copy the original IP header including options. The payload
1223                  * goes into an additional mbuf chain returned by m_copy().
1224                  */
1225                 m->m_data += max_linkhdr;
1226                 mhip = mtod(m, struct ip *);
1227                 *mhip = *ip;
1228                 if (hlen > sizeof(struct ip)) {
1229                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1230                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1231                 }
1232                 m->m_len = mhlen;
1233                 /* XXX do we need to add ip->ip_off below ? */
1234                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1235                 if (off + len >= ip->ip_len) {  /* last fragment */
1236                         len = ip->ip_len - off;
1237                         m->m_flags |= M_LASTFRAG;
1238                 } else
1239                         mhip->ip_off |= IP_MF;
1240                 mhip->ip_len = htons((u_short)(len + mhlen));
1241                 m->m_next = m_copy(m0, off, len);
1242                 if (m->m_next == NULL) {                /* copy failed */
1243                         m_free(m);
1244                         error = ENOBUFS;        /* ??? */
1245                         ipstat.ips_odropped++;
1246                         goto done;
1247                 }
1248                 m->m_pkthdr.len = mhlen + len;
1249                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1250                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1251                 mhip->ip_off = htons(mhip->ip_off);
1252                 mhip->ip_sum = 0;
1253                 if (sw_csum & CSUM_DELAY_IP)
1254                         mhip->ip_sum = in_cksum(m, mhlen);
1255                 *mnext = m;
1256                 mnext = &m->m_nextpkt;
1257         }
1258         ipstat.ips_ofragments += nfrags;
1259
1260         /* set first marker for fragment chain */
1261         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1262         m0->m_pkthdr.csum_data = nfrags;
1263
1264         /*
1265          * Update first fragment by trimming what's been copied out
1266          * and updating header.
1267          */
1268         m_adj(m0, hlen + firstlen - ip->ip_len);
1269         m0->m_pkthdr.len = hlen + firstlen;
1270         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1271         ip->ip_off |= IP_MF;
1272         ip->ip_off = htons(ip->ip_off);
1273         ip->ip_sum = 0;
1274         if (sw_csum & CSUM_DELAY_IP)
1275                 ip->ip_sum = in_cksum(m0, hlen);
1276
1277 done:
1278         *m_frag = m0;
1279         return error;
1280 }
1281
1282 void
1283 in_delayed_cksum(struct mbuf *m)
1284 {
1285         struct ip *ip;
1286         u_short csum, offset;
1287
1288         ip = mtod(m, struct ip *);
1289         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1290         csum = in_cksum_skip(m, ip->ip_len, offset);
1291         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1292                 csum = 0xffff;
1293         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1294
1295         if (offset + sizeof(u_short) > m->m_len) {
1296                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1297                     m->m_len, offset, ip->ip_p);
1298                 /*
1299                  * XXX
1300                  * this shouldn't happen, but if it does, the
1301                  * correct behavior may be to insert the checksum
1302                  * in the existing chain instead of rearranging it.
1303                  */
1304                 m = m_pullup(m, offset + sizeof(u_short));
1305         }
1306         *(u_short *)(m->m_data + offset) = csum;
1307 }
1308
1309 /*
1310  * Insert IP options into preformed packet.
1311  * Adjust IP destination as required for IP source routing,
1312  * as indicated by a non-zero in_addr at the start of the options.
1313  *
1314  * XXX This routine assumes that the packet has no options in place.
1315  */
1316 static struct mbuf *
1317 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1318 {
1319         struct ipoption *p = mtod(opt, struct ipoption *);
1320         struct mbuf *n;
1321         struct ip *ip = mtod(m, struct ip *);
1322         unsigned optlen;
1323
1324         optlen = opt->m_len - sizeof p->ipopt_dst;
1325         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1326                 *phlen = 0;
1327                 return (m);             /* XXX should fail */
1328         }
1329         if (p->ipopt_dst.s_addr)
1330                 ip->ip_dst = p->ipopt_dst;
1331         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1332                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1333                 if (n == NULL) {
1334                         *phlen = 0;
1335                         return (m);
1336                 }
1337                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1338                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1339                 m->m_len -= sizeof(struct ip);
1340                 m->m_data += sizeof(struct ip);
1341                 n->m_next = m;
1342                 m = n;
1343                 m->m_len = optlen + sizeof(struct ip);
1344                 m->m_data += max_linkhdr;
1345                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1346         } else {
1347                 m->m_data -= optlen;
1348                 m->m_len += optlen;
1349                 m->m_pkthdr.len += optlen;
1350                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1351         }
1352         ip = mtod(m, struct ip *);
1353         bcopy(p->ipopt_list, ip + 1, optlen);
1354         *phlen = sizeof(struct ip) + optlen;
1355         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1356         ip->ip_len += optlen;
1357         return (m);
1358 }
1359
1360 /*
1361  * Copy options from ip to jp,
1362  * omitting those not copied during fragmentation.
1363  */
1364 int
1365 ip_optcopy(struct ip *ip, struct ip *jp)
1366 {
1367         u_char *cp, *dp;
1368         int opt, optlen, cnt;
1369
1370         cp = (u_char *)(ip + 1);
1371         dp = (u_char *)(jp + 1);
1372         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1373         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1374                 opt = cp[0];
1375                 if (opt == IPOPT_EOL)
1376                         break;
1377                 if (opt == IPOPT_NOP) {
1378                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1379                         *dp++ = IPOPT_NOP;
1380                         optlen = 1;
1381                         continue;
1382                 }
1383
1384                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1385                     ("ip_optcopy: malformed ipv4 option"));
1386                 optlen = cp[IPOPT_OLEN];
1387                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1388                     ("ip_optcopy: malformed ipv4 option"));
1389
1390                 /* bogus lengths should have been caught by ip_dooptions */
1391                 if (optlen > cnt)
1392                         optlen = cnt;
1393                 if (IPOPT_COPIED(opt)) {
1394                         bcopy(cp, dp, optlen);
1395                         dp += optlen;
1396                 }
1397         }
1398         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1399                 *dp++ = IPOPT_EOL;
1400         return (optlen);
1401 }
1402
1403 /*
1404  * IP socket option processing.
1405  */
1406 int
1407 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1408 {
1409         struct  inpcb *inp = so->so_pcb;
1410         int     error, optval;
1411
1412         error = optval = 0;
1413         if (sopt->sopt_level != IPPROTO_IP) {
1414                 return (EINVAL);
1415         }
1416
1417         switch (sopt->sopt_dir) {
1418         case SOPT_SET:
1419                 switch (sopt->sopt_name) {
1420                 case IP_OPTIONS:
1421 #ifdef notyet
1422                 case IP_RETOPTS:
1423 #endif
1424                 {
1425                         struct mbuf *m;
1426                         if (sopt->sopt_valsize > MLEN) {
1427                                 error = EMSGSIZE;
1428                                 break;
1429                         }
1430                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1431                         if (m == NULL) {
1432                                 error = ENOBUFS;
1433                                 break;
1434                         }
1435                         m->m_len = sopt->sopt_valsize;
1436                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1437                                               m->m_len);
1438                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1439                                            m));
1440                 }
1441
1442                 case IP_TOS:
1443                 case IP_TTL:
1444                 case IP_MINTTL:
1445                 case IP_RECVOPTS:
1446                 case IP_RECVRETOPTS:
1447                 case IP_RECVDSTADDR:
1448                 case IP_RECVIF:
1449                 case IP_RECVTTL:
1450                 case IP_FAITH:
1451                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1452                                              sizeof optval);
1453                         if (error)
1454                                 break;
1455                         switch (sopt->sopt_name) {
1456                         case IP_TOS:
1457                                 inp->inp_ip_tos = optval;
1458                                 break;
1459
1460                         case IP_TTL:
1461                                 inp->inp_ip_ttl = optval;
1462                                 break;
1463                         case IP_MINTTL:
1464                                 if (optval > 0 && optval <= MAXTTL)
1465                                         inp->inp_ip_minttl = optval;
1466                                 else
1467                                         error = EINVAL;
1468                                 break;
1469 #define OPTSET(bit) \
1470         if (optval) \
1471                 inp->inp_flags |= bit; \
1472         else \
1473                 inp->inp_flags &= ~bit;
1474
1475                         case IP_RECVOPTS:
1476                                 OPTSET(INP_RECVOPTS);
1477                                 break;
1478
1479                         case IP_RECVRETOPTS:
1480                                 OPTSET(INP_RECVRETOPTS);
1481                                 break;
1482
1483                         case IP_RECVDSTADDR:
1484                                 OPTSET(INP_RECVDSTADDR);
1485                                 break;
1486
1487                         case IP_RECVIF:
1488                                 OPTSET(INP_RECVIF);
1489                                 break;
1490
1491                         case IP_RECVTTL:
1492                                 OPTSET(INP_RECVTTL);
1493                                 break;
1494
1495                         case IP_FAITH:
1496                                 OPTSET(INP_FAITH);
1497                                 break;
1498                         }
1499                         break;
1500 #undef OPTSET
1501
1502                 case IP_MULTICAST_IF:
1503                 case IP_MULTICAST_VIF:
1504                 case IP_MULTICAST_TTL:
1505                 case IP_MULTICAST_LOOP:
1506                 case IP_ADD_MEMBERSHIP:
1507                 case IP_DROP_MEMBERSHIP:
1508                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1509                         break;
1510
1511                 case IP_PORTRANGE:
1512                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1513                                             sizeof optval);
1514                         if (error)
1515                                 break;
1516
1517                         switch (optval) {
1518                         case IP_PORTRANGE_DEFAULT:
1519                                 inp->inp_flags &= ~(INP_LOWPORT);
1520                                 inp->inp_flags &= ~(INP_HIGHPORT);
1521                                 break;
1522
1523                         case IP_PORTRANGE_HIGH:
1524                                 inp->inp_flags &= ~(INP_LOWPORT);
1525                                 inp->inp_flags |= INP_HIGHPORT;
1526                                 break;
1527
1528                         case IP_PORTRANGE_LOW:
1529                                 inp->inp_flags &= ~(INP_HIGHPORT);
1530                                 inp->inp_flags |= INP_LOWPORT;
1531                                 break;
1532
1533                         default:
1534                                 error = EINVAL;
1535                                 break;
1536                         }
1537                         break;
1538
1539 #if defined(IPSEC) || defined(FAST_IPSEC)
1540                 case IP_IPSEC_POLICY:
1541                 {
1542                         caddr_t req;
1543                         size_t len = 0;
1544                         int priv;
1545                         struct mbuf *m;
1546                         int optname;
1547
1548                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1549                                 break;
1550                         soopt_to_mbuf(sopt, m);
1551                         priv = (sopt->sopt_td != NULL &&
1552                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1553                         req = mtod(m, caddr_t);
1554                         len = m->m_len;
1555                         optname = sopt->sopt_name;
1556                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1557                         m_freem(m);
1558                         break;
1559                 }
1560 #endif /*IPSEC*/
1561
1562                 default:
1563                         error = ENOPROTOOPT;
1564                         break;
1565                 }
1566                 break;
1567
1568         case SOPT_GET:
1569                 switch (sopt->sopt_name) {
1570                 case IP_OPTIONS:
1571                 case IP_RETOPTS:
1572                         if (inp->inp_options)
1573                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1574                                                            char *),
1575                                                 inp->inp_options->m_len);
1576                         else
1577                                 sopt->sopt_valsize = 0;
1578                         break;
1579
1580                 case IP_TOS:
1581                 case IP_TTL:
1582                 case IP_MINTTL:
1583                 case IP_RECVOPTS:
1584                 case IP_RECVRETOPTS:
1585                 case IP_RECVDSTADDR:
1586                 case IP_RECVTTL:
1587                 case IP_RECVIF:
1588                 case IP_PORTRANGE:
1589                 case IP_FAITH:
1590                         switch (sopt->sopt_name) {
1591
1592                         case IP_TOS:
1593                                 optval = inp->inp_ip_tos;
1594                                 break;
1595
1596                         case IP_TTL:
1597                                 optval = inp->inp_ip_ttl;
1598                                 break;
1599                         case IP_MINTTL:
1600                                 optval = inp->inp_ip_minttl;
1601                                 break;
1602
1603 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1604
1605                         case IP_RECVOPTS:
1606                                 optval = OPTBIT(INP_RECVOPTS);
1607                                 break;
1608
1609                         case IP_RECVRETOPTS:
1610                                 optval = OPTBIT(INP_RECVRETOPTS);
1611                                 break;
1612
1613                         case IP_RECVDSTADDR:
1614                                 optval = OPTBIT(INP_RECVDSTADDR);
1615                                 break;
1616
1617                         case IP_RECVTTL:
1618                                 optval = OPTBIT(INP_RECVTTL);
1619                                 break;
1620
1621                         case IP_RECVIF:
1622                                 optval = OPTBIT(INP_RECVIF);
1623                                 break;
1624
1625                         case IP_PORTRANGE:
1626                                 if (inp->inp_flags & INP_HIGHPORT)
1627                                         optval = IP_PORTRANGE_HIGH;
1628                                 else if (inp->inp_flags & INP_LOWPORT)
1629                                         optval = IP_PORTRANGE_LOW;
1630                                 else
1631                                         optval = 0;
1632                                 break;
1633
1634                         case IP_FAITH:
1635                                 optval = OPTBIT(INP_FAITH);
1636                                 break;
1637                         }
1638                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1639                         break;
1640
1641                 case IP_MULTICAST_IF:
1642                 case IP_MULTICAST_VIF:
1643                 case IP_MULTICAST_TTL:
1644                 case IP_MULTICAST_LOOP:
1645                 case IP_ADD_MEMBERSHIP:
1646                 case IP_DROP_MEMBERSHIP:
1647                         error = ip_getmoptions(sopt, inp->inp_moptions);
1648                         break;
1649
1650 #if defined(IPSEC) || defined(FAST_IPSEC)
1651                 case IP_IPSEC_POLICY:
1652                 {
1653                         struct mbuf *m = NULL;
1654                         caddr_t req = NULL;
1655                         size_t len = 0;
1656
1657                         if (m != NULL) {
1658                                 req = mtod(m, caddr_t);
1659                                 len = m->m_len;
1660                         }
1661                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1662                         if (error == 0)
1663                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1664                         if (error == 0)
1665                                 m_freem(m);
1666                         break;
1667                 }
1668 #endif /*IPSEC*/
1669
1670                 default:
1671                         error = ENOPROTOOPT;
1672                         break;
1673                 }
1674                 break;
1675         }
1676         return (error);
1677 }
1678
1679 /*
1680  * Set up IP options in pcb for insertion in output packets.
1681  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1682  * with destination address if source routed.
1683  */
1684 static int
1685 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1686 {
1687         int cnt, optlen;
1688         u_char *cp;
1689         u_char opt;
1690
1691         /* turn off any old options */
1692         if (*pcbopt)
1693                 m_free(*pcbopt);
1694         *pcbopt = 0;
1695         if (m == NULL || m->m_len == 0) {
1696                 /*
1697                  * Only turning off any previous options.
1698                  */
1699                 if (m != NULL)
1700                         m_free(m);
1701                 return (0);
1702         }
1703
1704         if (m->m_len % sizeof(int32_t))
1705                 goto bad;
1706         /*
1707          * IP first-hop destination address will be stored before
1708          * actual options; move other options back
1709          * and clear it when none present.
1710          */
1711         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1712                 goto bad;
1713         cnt = m->m_len;
1714         m->m_len += sizeof(struct in_addr);
1715         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1716         ovbcopy(mtod(m, caddr_t), cp, cnt);
1717         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1718
1719         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1720                 opt = cp[IPOPT_OPTVAL];
1721                 if (opt == IPOPT_EOL)
1722                         break;
1723                 if (opt == IPOPT_NOP)
1724                         optlen = 1;
1725                 else {
1726                         if (cnt < IPOPT_OLEN + sizeof *cp)
1727                                 goto bad;
1728                         optlen = cp[IPOPT_OLEN];
1729                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1730                                 goto bad;
1731                 }
1732                 switch (opt) {
1733
1734                 default:
1735                         break;
1736
1737                 case IPOPT_LSRR:
1738                 case IPOPT_SSRR:
1739                         /*
1740                          * user process specifies route as:
1741                          *      ->A->B->C->D
1742                          * D must be our final destination (but we can't
1743                          * check that since we may not have connected yet).
1744                          * A is first hop destination, which doesn't appear in
1745                          * actual IP option, but is stored before the options.
1746                          */
1747                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1748                                 goto bad;
1749                         m->m_len -= sizeof(struct in_addr);
1750                         cnt -= sizeof(struct in_addr);
1751                         optlen -= sizeof(struct in_addr);
1752                         cp[IPOPT_OLEN] = optlen;
1753                         /*
1754                          * Move first hop before start of options.
1755                          */
1756                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1757                               sizeof(struct in_addr));
1758                         /*
1759                          * Then copy rest of options back
1760                          * to close up the deleted entry.
1761                          */
1762                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1763                                 &cp[IPOPT_OFFSET+1],
1764                                 cnt - (IPOPT_MINOFF - 1));
1765                         break;
1766                 }
1767         }
1768         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1769                 goto bad;
1770         *pcbopt = m;
1771         return (0);
1772
1773 bad:
1774         m_free(m);
1775         return (EINVAL);
1776 }
1777
1778 /*
1779  * XXX
1780  * The whole multicast option thing needs to be re-thought.
1781  * Several of these options are equally applicable to non-multicast
1782  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1783  * standard option (IP_TTL).
1784  */
1785
1786 /*
1787  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1788  */
1789 static struct ifnet *
1790 ip_multicast_if(struct in_addr *a, int *ifindexp)
1791 {
1792         int ifindex;
1793         struct ifnet *ifp;
1794
1795         if (ifindexp)
1796                 *ifindexp = 0;
1797         if (ntohl(a->s_addr) >> 24 == 0) {
1798                 ifindex = ntohl(a->s_addr) & 0xffffff;
1799                 if (ifindex < 0 || if_index < ifindex)
1800                         return NULL;
1801                 ifp = ifindex2ifnet[ifindex];
1802                 if (ifindexp)
1803                         *ifindexp = ifindex;
1804         } else {
1805                 ifp = INADDR_TO_IFP(a);
1806         }
1807         return ifp;
1808 }
1809
1810 /*
1811  * Set the IP multicast options in response to user setsockopt().
1812  */
1813 static int
1814 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1815 {
1816         int error = 0;
1817         int i;
1818         struct in_addr addr;
1819         struct ip_mreq mreq;
1820         struct ifnet *ifp;
1821         struct ip_moptions *imo = *imop;
1822         int ifindex;
1823
1824         if (imo == NULL) {
1825                 /*
1826                  * No multicast option buffer attached to the pcb;
1827                  * allocate one and initialize to default values.
1828                  */
1829                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1830
1831                 *imop = imo;
1832                 imo->imo_multicast_ifp = NULL;
1833                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1834                 imo->imo_multicast_vif = -1;
1835                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1836                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1837                 imo->imo_num_memberships = 0;
1838         }
1839         switch (sopt->sopt_name) {
1840         /* store an index number for the vif you wanna use in the send */
1841         case IP_MULTICAST_VIF:
1842                 if (legal_vif_num == 0) {
1843                         error = EOPNOTSUPP;
1844                         break;
1845                 }
1846                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1847                 if (error)
1848                         break;
1849                 if (!legal_vif_num(i) && (i != -1)) {
1850                         error = EINVAL;
1851                         break;
1852                 }
1853                 imo->imo_multicast_vif = i;
1854                 break;
1855
1856         case IP_MULTICAST_IF:
1857                 /*
1858                  * Select the interface for outgoing multicast packets.
1859                  */
1860                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1861                 if (error)
1862                         break;
1863
1864                 /*
1865                  * INADDR_ANY is used to remove a previous selection.
1866                  * When no interface is selected, a default one is
1867                  * chosen every time a multicast packet is sent.
1868                  */
1869                 if (addr.s_addr == INADDR_ANY) {
1870                         imo->imo_multicast_ifp = NULL;
1871                         break;
1872                 }
1873                 /*
1874                  * The selected interface is identified by its local
1875                  * IP address.  Find the interface and confirm that
1876                  * it supports multicasting.
1877                  */
1878                 crit_enter();
1879                 ifp = ip_multicast_if(&addr, &ifindex);
1880                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1881                         crit_exit();
1882                         error = EADDRNOTAVAIL;
1883                         break;
1884                 }
1885                 imo->imo_multicast_ifp = ifp;
1886                 if (ifindex)
1887                         imo->imo_multicast_addr = addr;
1888                 else
1889                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1890                 crit_exit();
1891                 break;
1892
1893         case IP_MULTICAST_TTL:
1894                 /*
1895                  * Set the IP time-to-live for outgoing multicast packets.
1896                  * The original multicast API required a char argument,
1897                  * which is inconsistent with the rest of the socket API.
1898                  * We allow either a char or an int.
1899                  */
1900                 if (sopt->sopt_valsize == 1) {
1901                         u_char ttl;
1902                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1903                         if (error)
1904                                 break;
1905                         imo->imo_multicast_ttl = ttl;
1906                 } else {
1907                         u_int ttl;
1908                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1909                         if (error)
1910                                 break;
1911                         if (ttl > 255)
1912                                 error = EINVAL;
1913                         else
1914                                 imo->imo_multicast_ttl = ttl;
1915                 }
1916                 break;
1917
1918         case IP_MULTICAST_LOOP:
1919                 /*
1920                  * Set the loopback flag for outgoing multicast packets.
1921                  * Must be zero or one.  The original multicast API required a
1922                  * char argument, which is inconsistent with the rest
1923                  * of the socket API.  We allow either a char or an int.
1924                  */
1925                 if (sopt->sopt_valsize == 1) {
1926                         u_char loop;
1927
1928                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1929                         if (error)
1930                                 break;
1931                         imo->imo_multicast_loop = !!loop;
1932                 } else {
1933                         u_int loop;
1934
1935                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1936                                             sizeof loop);
1937                         if (error)
1938                                 break;
1939                         imo->imo_multicast_loop = !!loop;
1940                 }
1941                 break;
1942
1943         case IP_ADD_MEMBERSHIP:
1944                 /*
1945                  * Add a multicast group membership.
1946                  * Group must be a valid IP multicast address.
1947                  */
1948                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1949                 if (error)
1950                         break;
1951
1952                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1953                         error = EINVAL;
1954                         break;
1955                 }
1956                 crit_enter();
1957                 /*
1958                  * If no interface address was provided, use the interface of
1959                  * the route to the given multicast address.
1960                  */
1961                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1962                         struct sockaddr_in dst;
1963                         struct rtentry *rt;
1964
1965                         bzero(&dst, sizeof(struct sockaddr_in));
1966                         dst.sin_len = sizeof(struct sockaddr_in);
1967                         dst.sin_family = AF_INET;
1968                         dst.sin_addr = mreq.imr_multiaddr;
1969                         rt = rtlookup((struct sockaddr *)&dst);
1970                         if (rt == NULL) {
1971                                 error = EADDRNOTAVAIL;
1972                                 crit_exit();
1973                                 break;
1974                         }
1975                         --rt->rt_refcnt;
1976                         ifp = rt->rt_ifp;
1977                 } else {
1978                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1979                 }
1980
1981                 /*
1982                  * See if we found an interface, and confirm that it
1983                  * supports multicast.
1984                  */
1985                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1986                         error = EADDRNOTAVAIL;
1987                         crit_exit();
1988                         break;
1989                 }
1990                 /*
1991                  * See if the membership already exists or if all the
1992                  * membership slots are full.
1993                  */
1994                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1995                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1996                             imo->imo_membership[i]->inm_addr.s_addr
1997                                                 == mreq.imr_multiaddr.s_addr)
1998                                 break;
1999                 }
2000                 if (i < imo->imo_num_memberships) {
2001                         error = EADDRINUSE;
2002                         crit_exit();
2003                         break;
2004                 }
2005                 if (i == IP_MAX_MEMBERSHIPS) {
2006                         error = ETOOMANYREFS;
2007                         crit_exit();
2008                         break;
2009                 }
2010                 /*
2011                  * Everything looks good; add a new record to the multicast
2012                  * address list for the given interface.
2013                  */
2014                 if ((imo->imo_membership[i] =
2015                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2016                         error = ENOBUFS;
2017                         crit_exit();
2018                         break;
2019                 }
2020                 ++imo->imo_num_memberships;
2021                 crit_exit();
2022                 break;
2023
2024         case IP_DROP_MEMBERSHIP:
2025                 /*
2026                  * Drop a multicast group membership.
2027                  * Group must be a valid IP multicast address.
2028                  */
2029                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2030                 if (error)
2031                         break;
2032
2033                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2034                         error = EINVAL;
2035                         break;
2036                 }
2037
2038                 crit_enter();
2039                 /*
2040                  * If an interface address was specified, get a pointer
2041                  * to its ifnet structure.
2042                  */
2043                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2044                         ifp = NULL;
2045                 else {
2046                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2047                         if (ifp == NULL) {
2048                                 error = EADDRNOTAVAIL;
2049                                 crit_exit();
2050                                 break;
2051                         }
2052                 }
2053                 /*
2054                  * Find the membership in the membership array.
2055                  */
2056                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2057                         if ((ifp == NULL ||
2058                              imo->imo_membership[i]->inm_ifp == ifp) &&
2059                             imo->imo_membership[i]->inm_addr.s_addr ==
2060                             mreq.imr_multiaddr.s_addr)
2061                                 break;
2062                 }
2063                 if (i == imo->imo_num_memberships) {
2064                         error = EADDRNOTAVAIL;
2065                         crit_exit();
2066                         break;
2067                 }
2068                 /*
2069                  * Give up the multicast address record to which the
2070                  * membership points.
2071                  */
2072                 in_delmulti(imo->imo_membership[i]);
2073                 /*
2074                  * Remove the gap in the membership array.
2075                  */
2076                 for (++i; i < imo->imo_num_memberships; ++i)
2077                         imo->imo_membership[i-1] = imo->imo_membership[i];
2078                 --imo->imo_num_memberships;
2079                 crit_exit();
2080                 break;
2081
2082         default:
2083                 error = EOPNOTSUPP;
2084                 break;
2085         }
2086
2087         /*
2088          * If all options have default values, no need to keep the mbuf.
2089          */
2090         if (imo->imo_multicast_ifp == NULL &&
2091             imo->imo_multicast_vif == -1 &&
2092             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2093             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2094             imo->imo_num_memberships == 0) {
2095                 kfree(*imop, M_IPMOPTS);
2096                 *imop = NULL;
2097         }
2098
2099         return (error);
2100 }
2101
2102 /*
2103  * Return the IP multicast options in response to user getsockopt().
2104  */
2105 static int
2106 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2107 {
2108         struct in_addr addr;
2109         struct in_ifaddr *ia;
2110         int error, optval;
2111         u_char coptval;
2112
2113         error = 0;
2114         switch (sopt->sopt_name) {
2115         case IP_MULTICAST_VIF:
2116                 if (imo != NULL)
2117                         optval = imo->imo_multicast_vif;
2118                 else
2119                         optval = -1;
2120                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2121                 break;
2122
2123         case IP_MULTICAST_IF:
2124                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2125                         addr.s_addr = INADDR_ANY;
2126                 else if (imo->imo_multicast_addr.s_addr) {
2127                         /* return the value user has set */
2128                         addr = imo->imo_multicast_addr;
2129                 } else {
2130                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2131                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2132                                 : IA_SIN(ia)->sin_addr.s_addr;
2133                 }
2134                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2135                 break;
2136
2137         case IP_MULTICAST_TTL:
2138                 if (imo == NULL)
2139                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2140                 else
2141                         optval = coptval = imo->imo_multicast_ttl;
2142                 if (sopt->sopt_valsize == 1)
2143                         soopt_from_kbuf(sopt, &coptval, 1);
2144                 else
2145                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2146                 break;
2147
2148         case IP_MULTICAST_LOOP:
2149                 if (imo == NULL)
2150                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2151                 else
2152                         optval = coptval = imo->imo_multicast_loop;
2153                 if (sopt->sopt_valsize == 1)
2154                         soopt_from_kbuf(sopt, &coptval, 1);
2155                 else
2156                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2157                 break;
2158
2159         default:
2160                 error = ENOPROTOOPT;
2161                 break;
2162         }
2163         return (error);
2164 }
2165
2166 /*
2167  * Discard the IP multicast options.
2168  */
2169 void
2170 ip_freemoptions(struct ip_moptions *imo)
2171 {
2172         int i;
2173
2174         if (imo != NULL) {
2175                 for (i = 0; i < imo->imo_num_memberships; ++i)
2176                         in_delmulti(imo->imo_membership[i]);
2177                 kfree(imo, M_IPMOPTS);
2178         }
2179 }
2180
2181 /*
2182  * Routine called from ip_output() to loop back a copy of an IP multicast
2183  * packet to the input queue of a specified interface.  Note that this
2184  * calls the output routine of the loopback "driver", but with an interface
2185  * pointer that might NOT be a loopback interface -- evil, but easier than
2186  * replicating that code here.
2187  */
2188 static void
2189 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2190              int hlen)
2191 {
2192         struct ip *ip;
2193         struct mbuf *copym;
2194
2195         copym = m_copypacket(m, MB_DONTWAIT);
2196         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2197                 copym = m_pullup(copym, hlen);
2198         if (copym != NULL) {
2199                 /*
2200                  * if the checksum hasn't been computed, mark it as valid
2201                  */
2202                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2203                         in_delayed_cksum(copym);
2204                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2205                         copym->m_pkthdr.csum_flags |=
2206                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2207                         copym->m_pkthdr.csum_data = 0xffff;
2208                 }
2209                 /*
2210                  * We don't bother to fragment if the IP length is greater
2211                  * than the interface's MTU.  Can this possibly matter?
2212                  */
2213                 ip = mtod(copym, struct ip *);
2214                 ip->ip_len = htons(ip->ip_len);
2215                 ip->ip_off = htons(ip->ip_off);
2216                 ip->ip_sum = 0;
2217                 if (ip->ip_vhl == IP_VHL_BORING) {
2218                         ip->ip_sum = in_cksum_hdr(ip);
2219                 } else {
2220                         ip->ip_sum = in_cksum(copym, hlen);
2221                 }
2222                 /*
2223                  * NB:
2224                  * It's not clear whether there are any lingering
2225                  * reentrancy problems in other areas which might
2226                  * be exposed by using ip_input directly (in
2227                  * particular, everything which modifies the packet
2228                  * in-place).  Yet another option is using the
2229                  * protosw directly to deliver the looped back
2230                  * packet.  For the moment, we'll err on the side
2231                  * of safety by using if_simloop().
2232                  */
2233 #if 1 /* XXX */
2234                 if (dst->sin_family != AF_INET) {
2235                         kprintf("ip_mloopback: bad address family %d\n",
2236                                                 dst->sin_family);
2237                         dst->sin_family = AF_INET;
2238                 }
2239 #endif
2240
2241 #ifdef notdef
2242                 copym->m_pkthdr.rcvif = ifp;
2243                 ip_input(copym);
2244 #else
2245                 if_simloop(ifp, copym, dst->sin_family, 0);
2246 #endif
2247         }
2248 }