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 1.50 2008/08/23 04:12:23 sephe Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42 #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, *mtag;
 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 secpolicy *sp = NULL;
 148         struct tdb_ident *tdbi;
 149 #endif /* FAST_IPSEC */
 150         struct ip_fw_args args;
 151         struct sockaddr_in *next_hop = NULL;
 152         int src_was_INADDR_ANY = 0;     /* as the name says... */
 153
 154         args.eh = NULL;
 155         args.rule = NULL;
 156
 157         m = m0;
 158         M_ASSERTPKTHDR(m);
 159
 160         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 161                 /* Next hop */
 162                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 163                 KKASSERT(mtag != NULL);
 164                 next_hop = m_tag_data(mtag);
 165         }
 166
 167         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 168                 struct dn_pkt *dn_pkt;
 169
 170                 /* Extract info from dummynet tag */
 171                 dn_mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 172                 KKASSERT(dn_mtag != NULL);
 173                 dn_pkt = m_tag_data(dn_mtag);
 174
 175                 /*
 176                  * The packet was already tagged, so part of the
 177                  * processing was already done, and we need to go down.
 178                  * Get parameters from the tag.
 179                  */
 180                 args.rule = dn_pkt->dn_priv;
 181                 KKASSERT(args.rule != NULL);
 182                 opt = NULL;
 183                 ro = &dn_pkt->ro;
 184                 imo = NULL;
 185                 dst = dn_pkt->dn_dst;
 186                 ifp = dn_pkt->ifp;
 187                 flags = dn_pkt->flags;
 188
 189                 /*
 190                  * Don't delete the dummynet tag here, just unlink it,
 191                  * since some local variables (like 'ro' and 'dst') are
 192                  * still referencing certain parts of it.
 193                  * The dummynet tag will be freed at the end of the
 194                  * output process.
 195                  */
 196                 m_tag_unlink(m, dn_mtag);
 197                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 198         }
 199
 200         if (ro == NULL) {
 201                 ro = &iproute;
 202                 bzero(ro, sizeof *ro);
 203         }
 204
 205         if (args.rule != NULL) {        /* dummynet already saw us */
 206                 ip = mtod(m, struct ip *);
 207                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 208                 if (ro->ro_rt)
 209                         ia = ifatoia(ro->ro_rt->rt_ifa);
 210                 goto sendit;
 211         }
 212
 213         if (opt) {
 214                 len = 0;
 215                 m = ip_insertoptions(m, opt, &len);
 216                 if (len != 0)
 217                         hlen = len;
 218         }
 219         ip = mtod(m, struct ip *);
 220         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 221
 222         /*
 223          * Fill in IP header.
 224          */
 225         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 226                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 227                 ip->ip_off &= IP_DF;
 228                 ip->ip_id = ip_newid();
 229                 ipstat.ips_localout++;
 230         } else {
 231                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 232         }
 233
 234         dst = (struct sockaddr_in *)&ro->ro_dst;
 235         /*
 236          * If there is a cached route,
 237          * check that it is to the same destination
 238          * and is still up.  If not, free it and try again.
 239          * The address family should also be checked in case of sharing the
 240          * cache with IPv6.
 241          */
 242         if (ro->ro_rt &&
 243             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 244              dst->sin_family != AF_INET ||
 245              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 246                 rtfree(ro->ro_rt);
 247                 ro->ro_rt = (struct rtentry *)NULL;
 248         }
 249         if (ro->ro_rt == NULL) {
 250                 bzero(dst, sizeof *dst);
 251                 dst->sin_family = AF_INET;
 252                 dst->sin_len = sizeof *dst;
 253                 dst->sin_addr = pkt_dst;
 254         }
 255         /*
 256          * If routing to interface only,
 257          * short circuit routing lookup.
 258          */
 259         if (flags & IP_ROUTETOIF) {
 260                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 261                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 262                         ipstat.ips_noroute++;
 263                         error = ENETUNREACH;
 264                         goto bad;
 265                 }
 266                 ifp = ia->ia_ifp;
 267                 ip->ip_ttl = 1;
 268                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 269         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 270                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 271                 /*
 272                  * Bypass the normal routing lookup for multicast
 273                  * packets if the interface is specified.
 274                  */
 275                 ifp = imo->imo_multicast_ifp;
 276                 ia = IFP_TO_IA(ifp);
 277                 isbroadcast = 0;        /* fool gcc */
 278         } else {
 279                 /*
 280                  * If this is the case, we probably don't want to allocate
 281                  * a protocol-cloned route since we didn't get one from the
 282                  * ULP.  This lets TCP do its thing, while not burdening
 283                  * forwarding or ICMP with the overhead of cloning a route.
 284                  * Of course, we still want to do any cloning requested by
 285                  * the link layer, as this is probably required in all cases
 286                  * for correct operation (as it is for ARP).
 287                  */
 288                 if (ro->ro_rt == NULL)
 289                         rtalloc_ign(ro, RTF_PRCLONING);
 290                 if (ro->ro_rt == NULL) {
 291                         ipstat.ips_noroute++;
 292                         error = EHOSTUNREACH;
 293                         goto bad;
 294                 }
 295                 ia = ifatoia(ro->ro_rt->rt_ifa);
 296                 ifp = ro->ro_rt->rt_ifp;
 297                 ro->ro_rt->rt_use++;
 298                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 299                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 300                 if (ro->ro_rt->rt_flags & RTF_HOST)
 301                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 302                 else
 303                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 304         }
 305         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 306                 struct in_multi *inm;
 307
 308                 m->m_flags |= M_MCAST;
 309                 /*
 310                  * IP destination address is multicast.  Make sure "dst"
 311                  * still points to the address in "ro".  (It may have been
 312                  * changed to point to a gateway address, above.)
 313                  */
 314                 dst = (struct sockaddr_in *)&ro->ro_dst;
 315                 /*
 316                  * See if the caller provided any multicast options
 317                  */
 318                 if (imo != NULL) {
 319                         ip->ip_ttl = imo->imo_multicast_ttl;
 320                         if (imo->imo_multicast_vif != -1)
 321                                 ip->ip_src.s_addr =
 322                                     ip_mcast_src ?
 323                                     ip_mcast_src(imo->imo_multicast_vif) :
 324                                     INADDR_ANY;
 325                 } else
 326                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 327                 /*
 328                  * Confirm that the outgoing interface supports multicast.
 329                  */
 330                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 331                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 332                                 ipstat.ips_noroute++;
 333                                 error = ENETUNREACH;
 334                                 goto bad;
 335                         }
 336                 }
 337                 /*
 338                  * If source address not specified yet, use address
 339                  * of outgoing interface.
 340                  */
 341                 if (ip->ip_src.s_addr == INADDR_ANY) {
 342                         /* Interface may have no addresses. */
 343                         if (ia != NULL)
 344                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 345                 }
 346
 347                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 348                 if (inm != NULL &&
 349                    (imo == NULL || imo->imo_multicast_loop)) {
 350                         /*
 351                          * If we belong to the destination multicast group
 352                          * on the outgoing interface, and the caller did not
 353                          * forbid loopback, loop back a copy.
 354                          */
 355                         ip_mloopback(ifp, m, dst, hlen);
 356                 }
 357                 else {
 358                         /*
 359                          * If we are acting as a multicast router, perform
 360                          * multicast forwarding as if the packet had just
 361                          * arrived on the interface to which we are about
 362                          * to send.  The multicast forwarding function
 363                          * recursively calls this function, using the
 364                          * IP_FORWARDING flag to prevent infinite recursion.
 365                          *
 366                          * Multicasts that are looped back by ip_mloopback(),
 367                          * above, will be forwarded by the ip_input() routine,
 368                          * if necessary.
 369                          */
 370                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 371                                 /*
 372                                  * If rsvp daemon is not running, do not
 373                                  * set ip_moptions. This ensures that the packet
 374                                  * is multicast and not just sent down one link
 375                                  * as prescribed by rsvpd.
 376                                  */
 377                                 if (!rsvp_on)
 378                                         imo = NULL;
 379                                 if (ip_mforward &&
 380                                     ip_mforward(ip, ifp, m, imo) != 0) {
 381                                         m_freem(m);
 382                                         goto done;
 383                                 }
 384                         }
 385                 }
 386
 387                 /*
 388                  * Multicasts with a time-to-live of zero may be looped-
 389                  * back, above, but must not be transmitted on a network.
 390                  * Also, multicasts addressed to the loopback interface
 391                  * are not sent -- the above call to ip_mloopback() will
 392                  * loop back a copy if this host actually belongs to the
 393                  * destination group on the loopback interface.
 394                  */
 395                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 396                         m_freem(m);
 397                         goto done;
 398                 }
 399
 400                 goto sendit;
 401         }
 402 #ifndef notdef
 403         /*
 404          * If the source address is not specified yet, use the address
 405          * of the outoing interface. In case, keep note we did that, so
 406          * if the the firewall changes the next-hop causing the output
 407          * interface to change, we can fix that.
 408          */
 409         if (ip->ip_src.s_addr == INADDR_ANY) {
 410                 /* Interface may have no addresses. */
 411                 if (ia != NULL) {
 412                         ip->ip_src = IA_SIN(ia)->sin_addr;
 413                         src_was_INADDR_ANY = 1;
 414                 }
 415         }
 416 #endif /* notdef */
 417 #ifdef ALTQ
 418         /*
 419          * Disable packet drop hack.
 420          * Packetdrop should be done by queueing.
 421          */
 422 #else /* !ALTQ */
 423         /*
 424          * Verify that we have any chance at all of being able to queue
 425          *      the packet or packet fragments
 426          */
 427         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 428                 ifp->if_snd.ifq_maxlen) {
 429                         error = ENOBUFS;
 430                         ipstat.ips_odropped++;
 431                         goto bad;
 432         }
 433 #endif /* !ALTQ */
 434
 435         /*
 436          * Look for broadcast address and
 437          * verify user is allowed to send
 438          * such a packet.
 439          */
 440         if (isbroadcast) {
 441                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 442                         error = EADDRNOTAVAIL;
 443                         goto bad;
 444                 }
 445                 if (!(flags & IP_ALLOWBROADCAST)) {
 446                         error = EACCES;
 447                         goto bad;
 448                 }
 449                 /* don't allow broadcast messages to be fragmented */
 450                 if (ip->ip_len > ifp->if_mtu) {
 451                         error = EMSGSIZE;
 452                         goto bad;
 453                 }
 454                 m->m_flags |= M_BCAST;
 455         } else {
 456                 m->m_flags &= ~M_BCAST;
 457         }
 458
 459 sendit:
 460 #ifdef IPSEC
 461         /* get SP for this packet */
 462         if (so == NULL)
 463                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 464         else
 465                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 466
 467         if (sp == NULL) {
 468                 ipsecstat.out_inval++;
 469                 goto bad;
 470         }
 471
 472         error = 0;
 473
 474         /* check policy */
 475         switch (sp->policy) {
 476         case IPSEC_POLICY_DISCARD:
 477                 /*
 478                  * This packet is just discarded.
 479                  */
 480                 ipsecstat.out_polvio++;
 481                 goto bad;
 482
 483         case IPSEC_POLICY_BYPASS:
 484         case IPSEC_POLICY_NONE:
 485                 /* no need to do IPsec. */
 486                 goto skip_ipsec;
 487
 488         case IPSEC_POLICY_IPSEC:
 489                 if (sp->req == NULL) {
 490                         /* acquire a policy */
 491                         error = key_spdacquire(sp);
 492                         goto bad;
 493                 }
 494                 break;
 495
 496         case IPSEC_POLICY_ENTRUST:
 497         default:
 498                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 499         }
 500     {
 501         struct ipsec_output_state state;
 502         bzero(&state, sizeof state);
 503         state.m = m;
 504         if (flags & IP_ROUTETOIF) {
 505                 state.ro = &iproute;
 506                 bzero(&iproute, sizeof iproute);
 507         } else
 508                 state.ro = ro;
 509         state.dst = (struct sockaddr *)dst;
 510
 511         ip->ip_sum = 0;
 512
 513         /*
 514          * XXX
 515          * delayed checksums are not currently compatible with IPsec
 516          */
 517         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 518                 in_delayed_cksum(m);
 519                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 520         }
 521
 522         ip->ip_len = htons(ip->ip_len);
 523         ip->ip_off = htons(ip->ip_off);
 524
 525         error = ipsec4_output(&state, sp, flags);
 526
 527         m = state.m;
 528         if (flags & IP_ROUTETOIF) {
 529                 /*
 530                  * if we have tunnel mode SA, we may need to ignore
 531                  * IP_ROUTETOIF.
 532                  */
 533                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 534                         flags &= ~IP_ROUTETOIF;
 535                         ro = state.ro;
 536                 }
 537         } else
 538                 ro = state.ro;
 539         dst = (struct sockaddr_in *)state.dst;
 540         if (error) {
 541                 /* mbuf is already reclaimed in ipsec4_output. */
 542                 m0 = NULL;
 543                 switch (error) {
 544                 case EHOSTUNREACH:
 545                 case ENETUNREACH:
 546                 case EMSGSIZE:
 547                 case ENOBUFS:
 548                 case ENOMEM:
 549                         break;
 550                 default:
 551                         kprintf("ip4_output (ipsec): error code %d\n", error);
 552                         /*fall through*/
 553                 case ENOENT:
 554                         /* don't show these error codes to the user */
 555                         error = 0;
 556                         break;
 557                 }
 558                 goto bad;
 559         }
 560     }
 561
 562         /* be sure to update variables that are affected by ipsec4_output() */
 563         ip = mtod(m, struct ip *);
 564 #ifdef _IP_VHL
 565         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 566 #else
 567         hlen = ip->ip_hl << 2;
 568 #endif
 569         if (ro->ro_rt == NULL) {
 570                 if (!(flags & IP_ROUTETOIF)) {
 571                         kprintf("ip_output: "
 572                                 "can't update route after IPsec processing\n");
 573                         error = EHOSTUNREACH;   /*XXX*/
 574                         goto bad;
 575                 }
 576         } else {
 577                 ia = ifatoia(ro->ro_rt->rt_ifa);
 578                 ifp = ro->ro_rt->rt_ifp;
 579         }
 580
 581         /* make it flipped, again. */
 582         ip->ip_len = ntohs(ip->ip_len);
 583         ip->ip_off = ntohs(ip->ip_off);
 584 skip_ipsec:
 585 #endif /*IPSEC*/
 586 #ifdef FAST_IPSEC
 587         /*
 588          * Check the security policy (SP) for the packet and, if
 589          * required, do IPsec-related processing.  There are two
 590          * cases here; the first time a packet is sent through
 591          * it will be untagged and handled by ipsec4_checkpolicy.
 592          * If the packet is resubmitted to ip_output (e.g. after
 593          * AH, ESP, etc. processing), there will be a tag to bypass
 594          * the lookup and related policy checking.
 595          */
 596         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 597         crit_enter();
 598         if (mtag != NULL) {
 599                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 600                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 601                 if (sp == NULL)
 602                         error = -EINVAL;        /* force silent drop */
 603                 m_tag_delete(m, mtag);
 604         } else {
 605                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 606                                         &error, inp);
 607         }
 608         /*
 609          * There are four return cases:
 610          *    sp != NULL                    apply IPsec policy
 611          *    sp == NULL, error == 0        no IPsec handling needed
 612          *    sp == NULL, error == -EINVAL  discard packet w/o error
 613          *    sp == NULL, error != 0        discard packet, report error
 614          */
 615         if (sp != NULL) {
 616                 /* Loop detection, check if ipsec processing already done */
 617                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 618                 for (mtag = m_tag_first(m); mtag != NULL;
 619                      mtag = m_tag_next(m, mtag)) {
 620                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 621                                 continue;
 622                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 623                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 624                                 continue;
 625                         /*
 626                          * Check if policy has an SA associated with it.
 627                          * This can happen when an SP has yet to acquire
 628                          * an SA; e.g. on first reference.  If it occurs,
 629                          * then we let ipsec4_process_packet do its thing.
 630                          */
 631                         if (sp->req->sav == NULL)
 632                                 break;
 633                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 634                         if (tdbi->spi == sp->req->sav->spi &&
 635                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 636                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 637                                  sizeof(union sockaddr_union)) == 0) {
 638                                 /*
 639                                  * No IPsec processing is needed, free
 640                                  * reference to SP.
 641                                  *
 642                                  * NB: null pointer to avoid free at
 643                                  *     done: below.
 644                                  */
 645                                 KEY_FREESP(&sp), sp = NULL;
 646                                 crit_exit();
 647                                 goto spd_done;
 648                         }
 649                 }
 650
 651                 /*
 652                  * Do delayed checksums now because we send before
 653                  * this is done in the normal processing path.
 654                  */
 655                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 656                         in_delayed_cksum(m);
 657                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 658                 }
 659
 660                 ip->ip_len = htons(ip->ip_len);
 661                 ip->ip_off = htons(ip->ip_off);
 662
 663                 /* NB: callee frees mbuf */
 664                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 665                 /*
 666                  * Preserve KAME behaviour: ENOENT can be returned
 667                  * when an SA acquire is in progress.  Don't propagate
 668                  * this to user-level; it confuses applications.
 669                  *
 670                  * XXX this will go away when the SADB is redone.
 671                  */
 672                 if (error == ENOENT)
 673                         error = 0;
 674                 crit_exit();
 675                 goto done;
 676         } else {
 677                 crit_exit();
 678
 679                 if (error != 0) {
 680                         /*
 681                          * Hack: -EINVAL is used to signal that a packet
 682                          * should be silently discarded.  This is typically
 683                          * because we asked key management for an SA and
 684                          * it was delayed (e.g. kicked up to IKE).
 685                          */
 686                         if (error == -EINVAL)
 687                                 error = 0;
 688                         goto bad;
 689                 } else {
 690                         /* No IPsec processing for this packet. */
 691                 }
 692 #ifdef notyet
 693                 /*
 694                  * If deferred crypto processing is needed, check that
 695                  * the interface supports it.
 696                  */
 697                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 698                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 699                         /* notify IPsec to do its own crypto */
 700                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 701                         error = EHOSTUNREACH;
 702                         goto bad;
 703                 }
 704 #endif
 705         }
 706 spd_done:
 707 #endif /* FAST_IPSEC */
 708         /*
 709          * IpHack's section.
 710          * - Xlate: translate packet's addr/port (NAT).
 711          * - Firewall: deny/allow/etc.
 712          * - Wrap: fake packet's addr/port <unimpl.>
 713          * - Encapsulate: put it in another IP and send out. <unimp.>
 714          */
 715
 716         /*
 717          * Run through list of hooks for output packets.
 718          */
 719         if (pfil_has_hooks(&inet_pfil_hook)) {
 720                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 721                 if (error != 0 || m == NULL)
 722                         goto done;
 723                 ip = mtod(m, struct ip *);
 724         }
 725
 726         /*
 727          * Check with the firewall...
 728          * but not if we are already being fwd'd from a firewall.
 729          */
 730         if (fw_enable && IPFW_LOADED && !next_hop) {
 731                 struct sockaddr_in *old = dst;
 732
 733                 args.m = m;
 734                 args.oif = ifp;
 735                 off = ip_fw_chk_ptr(&args);
 736                 m = args.m;
 737
 738                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 739                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 740                         KKASSERT(mtag != NULL);
 741                         next_hop = m_tag_data(mtag);
 742                         dst = next_hop;
 743                 }
 744
 745                 /*
 746                  * On return we must do the following:
 747                  * m == NULL    -> drop the pkt (old interface, deprecated)
 748                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 749                  * 1<=off<= 0xffff              -> DIVERT
 750                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 751                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 752                  * dst != old                   -> IPFIREWALL_FORWARD
 753                  * off==0, dst==old             -> accept
 754                  * If some of the above modules are not compiled in, then
 755                  * we should't have to check the corresponding condition
 756                  * (because the ipfw control socket should not accept
 757                  * unsupported rules), but better play safe and drop
 758                  * packets in case of doubt.
 759                  */
 760                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 761                         if (m)
 762                                 m_freem(m);
 763                         error = EACCES;
 764                         goto done;
 765                 }
 766                 ip = mtod(m, struct ip *);
 767                 if (off == 0 && dst == old)             /* common case */
 768                         goto pass;
 769                 if (off & IP_FW_PORT_DYNT_FLAG) {
 770                         /*
 771                          * pass the pkt to dummynet. Need to include
 772                          * pipe number, m, ifp, ro, dst because these are
 773                          * not recomputed in the next pass.
 774                          * All other parameters have been already used and
 775                          * so they are not needed anymore.
 776                          * XXX note: if the ifp or ro entry are deleted
 777                          * while a pkt is in dummynet, we are in trouble!
 778                          */
 779                         args.ro = ro;
 780                         args.dst = dst;
 781                         args.flags = flags;
 782
 783                         error = 0;
 784                         ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
 785                         goto done;
 786                 }
 787 #ifdef IPDIVERT
 788                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
 789                         struct mbuf *clone = NULL;
 790
 791                         /* Clone packet if we're doing a 'tee' */
 792                         if ((off & IP_FW_PORT_TEE_FLAG))
 793                                 clone = m_dup(m, MB_DONTWAIT);
 794
 795                         /*
 796                          * XXX
 797                          * delayed checksums are not currently compatible
 798                          * with divert sockets.
 799                          */
 800                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 801                                 in_delayed_cksum(m);
 802                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 803                         }
 804
 805                         /* Restore packet header fields to original values */
 806                         ip->ip_len = htons(ip->ip_len);
 807                         ip->ip_off = htons(ip->ip_off);
 808
 809                         /* Deliver packet to divert input routine */
 810                         divert_packet(m, 0, off & 0xffff);
 811
 812                         /* If 'tee', continue with original packet */
 813                         if (clone != NULL) {
 814                                 m = clone;
 815                                 ip = mtod(m, struct ip *);
 816                                 goto pass;
 817                         }
 818                         goto done;
 819                 }
 820 #endif
 821
 822                 /* IPFIREWALL_FORWARD */
 823                 /*
 824                  * Check dst to make sure it is directly reachable on the
 825                  * interface we previously thought it was.
 826                  * If it isn't (which may be likely in some situations) we have
 827                  * to re-route it (ie, find a route for the next-hop and the
 828                  * associated interface) and set them here. This is nested
 829                  * forwarding which in most cases is undesirable, except where
 830                  * such control is nigh impossible. So we do it here.
 831                  * And I'm babbling.
 832                  */
 833                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 834 #if 0
 835                         /*
 836                          * XXX To improve readability, this block should be
 837                          * changed into a function call as below:
 838                          */
 839                         error = ip_ipforward(&m, &dst, &ifp);
 840                         if (error)
 841                                 goto bad;
 842                         if (m == NULL) /* ip_input consumed the mbuf */
 843                                 goto done;
 844 #else
 845                         struct in_ifaddr *ia;
 846                         struct in_ifaddr_container *iac;
 847
 848                         /*
 849                          * XXX sro_fwd below is static, and a pointer
 850                          * to it gets passed to routines downstream.
 851                          * This could have surprisingly bad results in
 852                          * practice, because its content is overwritten
 853                          * by subsequent packets.
 854                          */
 855                         /* There must be a better way to do this next line... */
 856                         static struct route sro_fwd;
 857                         struct route *ro_fwd = &sro_fwd;
 858
 859 #if 0
 860                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 861                             dst->sin_addr, "\n");
 862 #endif
 863
 864                         /*
 865                          * We need to figure out if we have been forwarded
 866                          * to a local socket. If so, then we should somehow
 867                          * "loop back" to ip_input, and get directed to the
 868                          * PCB as if we had received this packet. This is
 869                          * because it may be dificult to identify the packets
 870                          * you want to forward until they are being output
 871                          * and have selected an interface. (e.g. locally
 872                          * initiated packets) If we used the loopback inteface,
 873                          * we would not be able to control what happens
 874                          * as the packet runs through ip_input() as
 875                          * it is done through a ISR.
 876                          */
 877                         ia = NULL;
 878                         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr),
 879                                      ia_hash) {
 880                                 /*
 881                                  * If the addr to forward to is one
 882                                  * of ours, we pretend to
 883                                  * be the destination for this packet.
 884                                  */
 885                                 if (IA_SIN(iac->ia)->sin_addr.s_addr ==
 886                                     dst->sin_addr.s_addr) {
 887                                         ia = iac->ia;
 888                                         break;
 889                                 }
 890                         }
 891                         if (ia != NULL) {    /* tell ip_input "dont filter" */
 892                                 if (m->m_pkthdr.rcvif == NULL)
 893                                         m->m_pkthdr.rcvif = ifunit("lo0");
 894                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 895                                         m->m_pkthdr.csum_flags |=
 896                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 897                                         m->m_pkthdr.csum_data = 0xffff;
 898                                 }
 899                                 m->m_pkthdr.csum_flags |=
 900                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 901                                 ip->ip_len = htons(ip->ip_len);
 902                                 ip->ip_off = htons(ip->ip_off);
 903                                 ip_input(m);
 904                                 goto done;
 905                         }
 906                         /* Some of the logic for this was nicked from above.
 907                          *
 908                          * This rewrites the cached route in a local PCB.
 909                          * Is this what we want to do?
 910                          */
 911                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
 912                         ro_fwd->ro_rt = NULL;
 913
 914                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 915                         if (ro_fwd->ro_rt == NULL) {
 916                                 ipstat.ips_noroute++;
 917                                 error = EHOSTUNREACH;
 918                                 goto bad;
 919                         }
 920
 921                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 922                         ifp = ro_fwd->ro_rt->rt_ifp;
 923                         ro_fwd->ro_rt->rt_use++;
 924                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 925                                 dst = (struct sockaddr_in *)
 926                                     ro_fwd->ro_rt->rt_gateway;
 927                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 928                                 isbroadcast =
 929                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 930                         else
 931                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 932                         if (ro->ro_rt != NULL)
 933                                 rtfree(ro->ro_rt);
 934                         ro->ro_rt = ro_fwd->ro_rt;
 935                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 936
 937 #endif  /* ... block to be put into a function */
 938                         /*
 939                          * If we added a default src ip earlier,
 940                          * which would have been gotten from the-then
 941                          * interface, do it again, from the new one.
 942                          */
 943                         if (src_was_INADDR_ANY)
 944                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 945                         goto pass ;
 946                 }
 947
 948                 /*
 949                  * if we get here, none of the above matches, and
 950                  * we have to drop the pkt
 951                  */
 952                 m_freem(m);
 953                 error = EACCES; /* not sure this is the right error msg */
 954                 goto done;
 955         }
 956
 957 pass:
 958         /* 127/8 must not appear on wire - RFC1122. */
 959         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 960             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 961                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 962                         ipstat.ips_badaddr++;
 963                         error = EADDRNOTAVAIL;
 964                         goto bad;
 965                 }
 966         }
 967
 968         m->m_pkthdr.csum_flags |= CSUM_IP;
 969         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 970         if (sw_csum & CSUM_DELAY_DATA) {
 971                 in_delayed_cksum(m);
 972                 sw_csum &= ~CSUM_DELAY_DATA;
 973         }
 974         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 975
 976         /*
 977          * If small enough for interface, or the interface will take
 978          * care of the fragmentation for us, can just send directly.
 979          */
 980         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 981             !(ip->ip_off & IP_DF))) {
 982                 ip->ip_len = htons(ip->ip_len);
 983                 ip->ip_off = htons(ip->ip_off);
 984                 ip->ip_sum = 0;
 985                 if (sw_csum & CSUM_DELAY_IP) {
 986                         if (ip->ip_vhl == IP_VHL_BORING) {
 987                                 ip->ip_sum = in_cksum_hdr(ip);
 988                         } else {
 989                                 ip->ip_sum = in_cksum(m, hlen);
 990                         }
 991                 }
 992
 993                 /* Record statistics for this interface address. */
 994                 if (!(flags & IP_FORWARDING) && ia) {
 995                         ia->ia_ifa.if_opackets++;
 996                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 997                 }
 998
 999 #ifdef IPSEC
1000                 /* clean ipsec history once it goes out of the node */
1001                 ipsec_delaux(m);
1002 #endif
1003
1004 #ifdef MBUF_STRESS_TEST
1005                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1006                         struct mbuf *m1, *m2;
1007                         int length, tmp;
1008
1009                         tmp = length = m->m_pkthdr.len;
1010
1011                         while ((length -= mbuf_frag_size) >= 1) {
1012                                 m1 = m_split(m, length, MB_DONTWAIT);
1013                                 if (m1 == NULL)
1014                                         break;
1015                                 m2 = m;
1016                                 while (m2->m_next != NULL)
1017                                         m2 = m2->m_next;
1018                                 m2->m_next = m1;
1019                         }
1020                         m->m_pkthdr.len = tmp;
1021                 }
1022 #endif
1023
1024 #ifdef MPLS
1025                 if (!mpls_output_process(m, ro->ro_rt))
1026                         goto done;
1027 #endif
1028                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1029                                        ro->ro_rt);
1030                 goto done;
1031         }
1032
1033         if (ip->ip_off & IP_DF) {
1034                 error = EMSGSIZE;
1035                 /*
1036                  * This case can happen if the user changed the MTU
1037                  * of an interface after enabling IP on it.  Because
1038                  * most netifs don't keep track of routes pointing to
1039                  * them, there is no way for one to update all its
1040                  * routes when the MTU is changed.
1041                  */
1042                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1043                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1044                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1045                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1046                 }
1047                 ipstat.ips_cantfrag++;
1048                 goto bad;
1049         }
1050
1051         /*
1052          * Too large for interface; fragment if possible. If successful,
1053          * on return, m will point to a list of packets to be sent.
1054          */
1055         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1056         if (error)
1057                 goto bad;
1058         for (; m; m = m0) {
1059                 m0 = m->m_nextpkt;
1060                 m->m_nextpkt = NULL;
1061 #ifdef IPSEC
1062                 /* clean ipsec history once it goes out of the node */
1063                 ipsec_delaux(m);
1064 #endif
1065                 if (error == 0) {
1066                         /* Record statistics for this interface address. */
1067                         if (ia != NULL) {
1068                                 ia->ia_ifa.if_opackets++;
1069                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1070                         }
1071 #ifdef MPLS
1072                         if (!mpls_output_process(m, ro->ro_rt))
1073                                 continue;
1074 #endif
1075                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1076                                                ro->ro_rt);
1077                 } else {
1078                         m_freem(m);
1079                 }
1080         }
1081
1082         if (error == 0)
1083                 ipstat.ips_fragmented++;
1084
1085 done:
1086         if (ro == &iproute && ro->ro_rt != NULL) {
1087                 RTFREE(ro->ro_rt);
1088                 ro->ro_rt = NULL;
1089         }
1090 #ifdef IPSEC
1091         if (sp != NULL) {
1092                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1093                         kprintf("DP ip_output call free SP:%p\n", sp));
1094                 key_freesp(sp);
1095         }
1096 #endif
1097 #ifdef FAST_IPSEC
1098         if (sp != NULL)
1099                 KEY_FREESP(&sp);
1100 #endif
1101         if (dn_mtag != NULL)
1102                 m_tag_free(dn_mtag);
1103
1104         return (error);
1105 bad:
1106         m_freem(m);
1107         goto done;
1108 }
1109
1110 /*
1111  * Create a chain of fragments which fit the given mtu. m_frag points to the
1112  * mbuf to be fragmented; on return it points to the chain with the fragments.
1113  * Return 0 if no error. If error, m_frag may contain a partially built
1114  * chain of fragments that should be freed by the caller.
1115  *
1116  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1117  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1118  */
1119 int
1120 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1121             u_long if_hwassist_flags, int sw_csum)
1122 {
1123         int error = 0;
1124         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1125         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1126         int off;
1127         struct mbuf *m0 = *m_frag;      /* the original packet          */
1128         int firstlen;
1129         struct mbuf **mnext;
1130         int nfrags;
1131
1132         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1133                 ipstat.ips_cantfrag++;
1134                 return EMSGSIZE;
1135         }
1136
1137         /*
1138          * Must be able to put at least 8 bytes per fragment.
1139          */
1140         if (len < 8)
1141                 return EMSGSIZE;
1142
1143         /*
1144          * If the interface will not calculate checksums on
1145          * fragmented packets, then do it here.
1146          */
1147         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1148             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1149                 in_delayed_cksum(m0);
1150                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1151         }
1152
1153         if (len > PAGE_SIZE) {
1154                 /*
1155                  * Fragment large datagrams such that each segment
1156                  * contains a multiple of PAGE_SIZE amount of data,
1157                  * plus headers. This enables a receiver to perform
1158                  * page-flipping zero-copy optimizations.
1159                  *
1160                  * XXX When does this help given that sender and receiver
1161                  * could have different page sizes, and also mtu could
1162                  * be less than the receiver's page size ?
1163                  */
1164                 int newlen;
1165                 struct mbuf *m;
1166
1167                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1168                         off += m->m_len;
1169
1170                 /*
1171                  * firstlen (off - hlen) must be aligned on an
1172                  * 8-byte boundary
1173                  */
1174                 if (off < hlen)
1175                         goto smart_frag_failure;
1176                 off = ((off - hlen) & ~7) + hlen;
1177                 newlen = (~PAGE_MASK) & mtu;
1178                 if ((newlen + sizeof(struct ip)) > mtu) {
1179                         /* we failed, go back the default */
1180 smart_frag_failure:
1181                         newlen = len;
1182                         off = hlen + len;
1183                 }
1184                 len = newlen;
1185
1186         } else {
1187                 off = hlen + len;
1188         }
1189
1190         firstlen = off - hlen;
1191         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1192
1193         /*
1194          * Loop through length of segment after first fragment,
1195          * make new header and copy data of each part and link onto chain.
1196          * Here, m0 is the original packet, m is the fragment being created.
1197          * The fragments are linked off the m_nextpkt of the original
1198          * packet, which after processing serves as the first fragment.
1199          */
1200         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1201                 struct ip *mhip;        /* ip header on the fragment */
1202                 struct mbuf *m;
1203                 int mhlen = sizeof(struct ip);
1204
1205                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1206                 if (m == NULL) {
1207                         error = ENOBUFS;
1208                         ipstat.ips_odropped++;
1209                         goto done;
1210                 }
1211                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1212                 /*
1213                  * In the first mbuf, leave room for the link header, then
1214                  * copy the original IP header including options. The payload
1215                  * goes into an additional mbuf chain returned by m_copy().
1216                  */
1217                 m->m_data += max_linkhdr;
1218                 mhip = mtod(m, struct ip *);
1219                 *mhip = *ip;
1220                 if (hlen > sizeof(struct ip)) {
1221                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1222                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1223                 }
1224                 m->m_len = mhlen;
1225                 /* XXX do we need to add ip->ip_off below ? */
1226                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1227                 if (off + len >= ip->ip_len) {  /* last fragment */
1228                         len = ip->ip_len - off;
1229                         m->m_flags |= M_LASTFRAG;
1230                 } else
1231                         mhip->ip_off |= IP_MF;
1232                 mhip->ip_len = htons((u_short)(len + mhlen));
1233                 m->m_next = m_copy(m0, off, len);
1234                 if (m->m_next == NULL) {                /* copy failed */
1235                         m_free(m);
1236                         error = ENOBUFS;        /* ??? */
1237                         ipstat.ips_odropped++;
1238                         goto done;
1239                 }
1240                 m->m_pkthdr.len = mhlen + len;
1241                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1242                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1243                 mhip->ip_off = htons(mhip->ip_off);
1244                 mhip->ip_sum = 0;
1245                 if (sw_csum & CSUM_DELAY_IP)
1246                         mhip->ip_sum = in_cksum(m, mhlen);
1247                 *mnext = m;
1248                 mnext = &m->m_nextpkt;
1249         }
1250         ipstat.ips_ofragments += nfrags;
1251
1252         /* set first marker for fragment chain */
1253         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1254         m0->m_pkthdr.csum_data = nfrags;
1255
1256         /*
1257          * Update first fragment by trimming what's been copied out
1258          * and updating header.
1259          */
1260         m_adj(m0, hlen + firstlen - ip->ip_len);
1261         m0->m_pkthdr.len = hlen + firstlen;
1262         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1263         ip->ip_off |= IP_MF;
1264         ip->ip_off = htons(ip->ip_off);
1265         ip->ip_sum = 0;
1266         if (sw_csum & CSUM_DELAY_IP)
1267                 ip->ip_sum = in_cksum(m0, hlen);
1268
1269 done:
1270         *m_frag = m0;
1271         return error;
1272 }
1273
1274 void
1275 in_delayed_cksum(struct mbuf *m)
1276 {
1277         struct ip *ip;
1278         u_short csum, offset;
1279
1280         ip = mtod(m, struct ip *);
1281         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1282         csum = in_cksum_skip(m, ip->ip_len, offset);
1283         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1284                 csum = 0xffff;
1285         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1286
1287         if (offset + sizeof(u_short) > m->m_len) {
1288                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1289                     m->m_len, offset, ip->ip_p);
1290                 /*
1291                  * XXX
1292                  * this shouldn't happen, but if it does, the
1293                  * correct behavior may be to insert the checksum
1294                  * in the existing chain instead of rearranging it.
1295                  */
1296                 m = m_pullup(m, offset + sizeof(u_short));
1297         }
1298         *(u_short *)(m->m_data + offset) = csum;
1299 }
1300
1301 /*
1302  * Insert IP options into preformed packet.
1303  * Adjust IP destination as required for IP source routing,
1304  * as indicated by a non-zero in_addr at the start of the options.
1305  *
1306  * XXX This routine assumes that the packet has no options in place.
1307  */
1308 static struct mbuf *
1309 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1310 {
1311         struct ipoption *p = mtod(opt, struct ipoption *);
1312         struct mbuf *n;
1313         struct ip *ip = mtod(m, struct ip *);
1314         unsigned optlen;
1315
1316         optlen = opt->m_len - sizeof p->ipopt_dst;
1317         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1318                 *phlen = 0;
1319                 return (m);             /* XXX should fail */
1320         }
1321         if (p->ipopt_dst.s_addr)
1322                 ip->ip_dst = p->ipopt_dst;
1323         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1324                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1325                 if (n == NULL) {
1326                         *phlen = 0;
1327                         return (m);
1328                 }
1329                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1330                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1331                 m->m_len -= sizeof(struct ip);
1332                 m->m_data += sizeof(struct ip);
1333                 n->m_next = m;
1334                 m = n;
1335                 m->m_len = optlen + sizeof(struct ip);
1336                 m->m_data += max_linkhdr;
1337                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1338         } else {
1339                 m->m_data -= optlen;
1340                 m->m_len += optlen;
1341                 m->m_pkthdr.len += optlen;
1342                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1343         }
1344         ip = mtod(m, struct ip *);
1345         bcopy(p->ipopt_list, ip + 1, optlen);
1346         *phlen = sizeof(struct ip) + optlen;
1347         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1348         ip->ip_len += optlen;
1349         return (m);
1350 }
1351
1352 /*
1353  * Copy options from ip to jp,
1354  * omitting those not copied during fragmentation.
1355  */
1356 int
1357 ip_optcopy(struct ip *ip, struct ip *jp)
1358 {
1359         u_char *cp, *dp;
1360         int opt, optlen, cnt;
1361
1362         cp = (u_char *)(ip + 1);
1363         dp = (u_char *)(jp + 1);
1364         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1365         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1366                 opt = cp[0];
1367                 if (opt == IPOPT_EOL)
1368                         break;
1369                 if (opt == IPOPT_NOP) {
1370                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1371                         *dp++ = IPOPT_NOP;
1372                         optlen = 1;
1373                         continue;
1374                 }
1375
1376                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1377                     ("ip_optcopy: malformed ipv4 option"));
1378                 optlen = cp[IPOPT_OLEN];
1379                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1380                     ("ip_optcopy: malformed ipv4 option"));
1381
1382                 /* bogus lengths should have been caught by ip_dooptions */
1383                 if (optlen > cnt)
1384                         optlen = cnt;
1385                 if (IPOPT_COPIED(opt)) {
1386                         bcopy(cp, dp, optlen);
1387                         dp += optlen;
1388                 }
1389         }
1390         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1391                 *dp++ = IPOPT_EOL;
1392         return (optlen);
1393 }
1394
1395 /*
1396  * IP socket option processing.
1397  */
1398 int
1399 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1400 {
1401         struct  inpcb *inp = so->so_pcb;
1402         int     error, optval;
1403
1404         error = optval = 0;
1405         if (sopt->sopt_level != IPPROTO_IP) {
1406                 return (EINVAL);
1407         }
1408
1409         switch (sopt->sopt_dir) {
1410         case SOPT_SET:
1411                 switch (sopt->sopt_name) {
1412                 case IP_OPTIONS:
1413 #ifdef notyet
1414                 case IP_RETOPTS:
1415 #endif
1416                 {
1417                         struct mbuf *m;
1418                         if (sopt->sopt_valsize > MLEN) {
1419                                 error = EMSGSIZE;
1420                                 break;
1421                         }
1422                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1423                         if (m == NULL) {
1424                                 error = ENOBUFS;
1425                                 break;
1426                         }
1427                         m->m_len = sopt->sopt_valsize;
1428                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1429                                               m->m_len);
1430                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1431                                            m));
1432                 }
1433
1434                 case IP_TOS:
1435                 case IP_TTL:
1436                 case IP_MINTTL:
1437                 case IP_RECVOPTS:
1438                 case IP_RECVRETOPTS:
1439                 case IP_RECVDSTADDR:
1440                 case IP_RECVIF:
1441                 case IP_RECVTTL:
1442                 case IP_FAITH:
1443                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1444                                              sizeof optval);
1445                         if (error)
1446                                 break;
1447                         switch (sopt->sopt_name) {
1448                         case IP_TOS:
1449                                 inp->inp_ip_tos = optval;
1450                                 break;
1451
1452                         case IP_TTL:
1453                                 inp->inp_ip_ttl = optval;
1454                                 break;
1455                         case IP_MINTTL:
1456                                 if (optval > 0 && optval <= MAXTTL)
1457                                         inp->inp_ip_minttl = optval;
1458                                 else
1459                                         error = EINVAL;
1460                                 break;
1461 #define OPTSET(bit) \
1462         if (optval) \
1463                 inp->inp_flags |= bit; \
1464         else \
1465                 inp->inp_flags &= ~bit;
1466
1467                         case IP_RECVOPTS:
1468                                 OPTSET(INP_RECVOPTS);
1469                                 break;
1470
1471                         case IP_RECVRETOPTS:
1472                                 OPTSET(INP_RECVRETOPTS);
1473                                 break;
1474
1475                         case IP_RECVDSTADDR:
1476                                 OPTSET(INP_RECVDSTADDR);
1477                                 break;
1478
1479                         case IP_RECVIF:
1480                                 OPTSET(INP_RECVIF);
1481                                 break;
1482
1483                         case IP_RECVTTL:
1484                                 OPTSET(INP_RECVTTL);
1485                                 break;
1486
1487                         case IP_FAITH:
1488                                 OPTSET(INP_FAITH);
1489                                 break;
1490                         }
1491                         break;
1492 #undef OPTSET
1493
1494                 case IP_MULTICAST_IF:
1495                 case IP_MULTICAST_VIF:
1496                 case IP_MULTICAST_TTL:
1497                 case IP_MULTICAST_LOOP:
1498                 case IP_ADD_MEMBERSHIP:
1499                 case IP_DROP_MEMBERSHIP:
1500                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1501                         break;
1502
1503                 case IP_PORTRANGE:
1504                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1505                                             sizeof optval);
1506                         if (error)
1507                                 break;
1508
1509                         switch (optval) {
1510                         case IP_PORTRANGE_DEFAULT:
1511                                 inp->inp_flags &= ~(INP_LOWPORT);
1512                                 inp->inp_flags &= ~(INP_HIGHPORT);
1513                                 break;
1514
1515                         case IP_PORTRANGE_HIGH:
1516                                 inp->inp_flags &= ~(INP_LOWPORT);
1517                                 inp->inp_flags |= INP_HIGHPORT;
1518                                 break;
1519
1520                         case IP_PORTRANGE_LOW:
1521                                 inp->inp_flags &= ~(INP_HIGHPORT);
1522                                 inp->inp_flags |= INP_LOWPORT;
1523                                 break;
1524
1525                         default:
1526                                 error = EINVAL;
1527                                 break;
1528                         }
1529                         break;
1530
1531 #if defined(IPSEC) || defined(FAST_IPSEC)
1532                 case IP_IPSEC_POLICY:
1533                 {
1534                         caddr_t req;
1535                         size_t len = 0;
1536                         int priv;
1537                         struct mbuf *m;
1538                         int optname;
1539
1540                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1541                                 break;
1542                         soopt_to_mbuf(sopt, m);
1543                         priv = (sopt->sopt_td != NULL &&
1544                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1545                         req = mtod(m, caddr_t);
1546                         len = m->m_len;
1547                         optname = sopt->sopt_name;
1548                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1549                         m_freem(m);
1550                         break;
1551                 }
1552 #endif /*IPSEC*/
1553
1554                 default:
1555                         error = ENOPROTOOPT;
1556                         break;
1557                 }
1558                 break;
1559
1560         case SOPT_GET:
1561                 switch (sopt->sopt_name) {
1562                 case IP_OPTIONS:
1563                 case IP_RETOPTS:
1564                         if (inp->inp_options)
1565                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1566                                                            char *),
1567                                                 inp->inp_options->m_len);
1568                         else
1569                                 sopt->sopt_valsize = 0;
1570                         break;
1571
1572                 case IP_TOS:
1573                 case IP_TTL:
1574                 case IP_MINTTL:
1575                 case IP_RECVOPTS:
1576                 case IP_RECVRETOPTS:
1577                 case IP_RECVDSTADDR:
1578                 case IP_RECVTTL:
1579                 case IP_RECVIF:
1580                 case IP_PORTRANGE:
1581                 case IP_FAITH:
1582                         switch (sopt->sopt_name) {
1583
1584                         case IP_TOS:
1585                                 optval = inp->inp_ip_tos;
1586                                 break;
1587
1588                         case IP_TTL:
1589                                 optval = inp->inp_ip_ttl;
1590                                 break;
1591                         case IP_MINTTL:
1592                                 optval = inp->inp_ip_minttl;
1593                                 break;
1594
1595 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1596
1597                         case IP_RECVOPTS:
1598                                 optval = OPTBIT(INP_RECVOPTS);
1599                                 break;
1600
1601                         case IP_RECVRETOPTS:
1602                                 optval = OPTBIT(INP_RECVRETOPTS);
1603                                 break;
1604
1605                         case IP_RECVDSTADDR:
1606                                 optval = OPTBIT(INP_RECVDSTADDR);
1607                                 break;
1608
1609                         case IP_RECVTTL:
1610                                 optval = OPTBIT(INP_RECVTTL);
1611                                 break;
1612
1613                         case IP_RECVIF:
1614                                 optval = OPTBIT(INP_RECVIF);
1615                                 break;
1616
1617                         case IP_PORTRANGE:
1618                                 if (inp->inp_flags & INP_HIGHPORT)
1619                                         optval = IP_PORTRANGE_HIGH;
1620                                 else if (inp->inp_flags & INP_LOWPORT)
1621                                         optval = IP_PORTRANGE_LOW;
1622                                 else
1623                                         optval = 0;
1624                                 break;
1625
1626                         case IP_FAITH:
1627                                 optval = OPTBIT(INP_FAITH);
1628                                 break;
1629                         }
1630                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1631                         break;
1632
1633                 case IP_MULTICAST_IF:
1634                 case IP_MULTICAST_VIF:
1635                 case IP_MULTICAST_TTL:
1636                 case IP_MULTICAST_LOOP:
1637                 case IP_ADD_MEMBERSHIP:
1638                 case IP_DROP_MEMBERSHIP:
1639                         error = ip_getmoptions(sopt, inp->inp_moptions);
1640                         break;
1641
1642 #if defined(IPSEC) || defined(FAST_IPSEC)
1643                 case IP_IPSEC_POLICY:
1644                 {
1645                         struct mbuf *m = NULL;
1646                         caddr_t req = NULL;
1647                         size_t len = 0;
1648
1649                         if (m != NULL) {
1650                                 req = mtod(m, caddr_t);
1651                                 len = m->m_len;
1652                         }
1653                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1654                         if (error == 0)
1655                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1656                         if (error == 0)
1657                                 m_freem(m);
1658                         break;
1659                 }
1660 #endif /*IPSEC*/
1661
1662                 default:
1663                         error = ENOPROTOOPT;
1664                         break;
1665                 }
1666                 break;
1667         }
1668         return (error);
1669 }
1670
1671 /*
1672  * Set up IP options in pcb for insertion in output packets.
1673  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1674  * with destination address if source routed.
1675  */
1676 static int
1677 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1678 {
1679         int cnt, optlen;
1680         u_char *cp;
1681         u_char opt;
1682
1683         /* turn off any old options */
1684         if (*pcbopt)
1685                 m_free(*pcbopt);
1686         *pcbopt = 0;
1687         if (m == NULL || m->m_len == 0) {
1688                 /*
1689                  * Only turning off any previous options.
1690                  */
1691                 if (m != NULL)
1692                         m_free(m);
1693                 return (0);
1694         }
1695
1696         if (m->m_len % sizeof(int32_t))
1697                 goto bad;
1698         /*
1699          * IP first-hop destination address will be stored before
1700          * actual options; move other options back
1701          * and clear it when none present.
1702          */
1703         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1704                 goto bad;
1705         cnt = m->m_len;
1706         m->m_len += sizeof(struct in_addr);
1707         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1708         ovbcopy(mtod(m, caddr_t), cp, cnt);
1709         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1710
1711         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1712                 opt = cp[IPOPT_OPTVAL];
1713                 if (opt == IPOPT_EOL)
1714                         break;
1715                 if (opt == IPOPT_NOP)
1716                         optlen = 1;
1717                 else {
1718                         if (cnt < IPOPT_OLEN + sizeof *cp)
1719                                 goto bad;
1720                         optlen = cp[IPOPT_OLEN];
1721                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1722                                 goto bad;
1723                 }
1724                 switch (opt) {
1725
1726                 default:
1727                         break;
1728
1729                 case IPOPT_LSRR:
1730                 case IPOPT_SSRR:
1731                         /*
1732                          * user process specifies route as:
1733                          *      ->A->B->C->D
1734                          * D must be our final destination (but we can't
1735                          * check that since we may not have connected yet).
1736                          * A is first hop destination, which doesn't appear in
1737                          * actual IP option, but is stored before the options.
1738                          */
1739                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1740                                 goto bad;
1741                         m->m_len -= sizeof(struct in_addr);
1742                         cnt -= sizeof(struct in_addr);
1743                         optlen -= sizeof(struct in_addr);
1744                         cp[IPOPT_OLEN] = optlen;
1745                         /*
1746                          * Move first hop before start of options.
1747                          */
1748                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1749                               sizeof(struct in_addr));
1750                         /*
1751                          * Then copy rest of options back
1752                          * to close up the deleted entry.
1753                          */
1754                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1755                                 &cp[IPOPT_OFFSET+1],
1756                                 cnt - (IPOPT_MINOFF - 1));
1757                         break;
1758                 }
1759         }
1760         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1761                 goto bad;
1762         *pcbopt = m;
1763         return (0);
1764
1765 bad:
1766         m_free(m);
1767         return (EINVAL);
1768 }
1769
1770 /*
1771  * XXX
1772  * The whole multicast option thing needs to be re-thought.
1773  * Several of these options are equally applicable to non-multicast
1774  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1775  * standard option (IP_TTL).
1776  */
1777
1778 /*
1779  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1780  */
1781 static struct ifnet *
1782 ip_multicast_if(struct in_addr *a, int *ifindexp)
1783 {
1784         int ifindex;
1785         struct ifnet *ifp;
1786
1787         if (ifindexp)
1788                 *ifindexp = 0;
1789         if (ntohl(a->s_addr) >> 24 == 0) {
1790                 ifindex = ntohl(a->s_addr) & 0xffffff;
1791                 if (ifindex < 0 || if_index < ifindex)
1792                         return NULL;
1793                 ifp = ifindex2ifnet[ifindex];
1794                 if (ifindexp)
1795                         *ifindexp = ifindex;
1796         } else {
1797                 ifp = INADDR_TO_IFP(a);
1798         }
1799         return ifp;
1800 }
1801
1802 /*
1803  * Set the IP multicast options in response to user setsockopt().
1804  */
1805 static int
1806 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1807 {
1808         int error = 0;
1809         int i;
1810         struct in_addr addr;
1811         struct ip_mreq mreq;
1812         struct ifnet *ifp;
1813         struct ip_moptions *imo = *imop;
1814         int ifindex;
1815
1816         if (imo == NULL) {
1817                 /*
1818                  * No multicast option buffer attached to the pcb;
1819                  * allocate one and initialize to default values.
1820                  */
1821                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1822
1823                 *imop = imo;
1824                 imo->imo_multicast_ifp = NULL;
1825                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1826                 imo->imo_multicast_vif = -1;
1827                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1828                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1829                 imo->imo_num_memberships = 0;
1830         }
1831         switch (sopt->sopt_name) {
1832         /* store an index number for the vif you wanna use in the send */
1833         case IP_MULTICAST_VIF:
1834                 if (legal_vif_num == 0) {
1835                         error = EOPNOTSUPP;
1836                         break;
1837                 }
1838                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1839                 if (error)
1840                         break;
1841                 if (!legal_vif_num(i) && (i != -1)) {
1842                         error = EINVAL;
1843                         break;
1844                 }
1845                 imo->imo_multicast_vif = i;
1846                 break;
1847
1848         case IP_MULTICAST_IF:
1849                 /*
1850                  * Select the interface for outgoing multicast packets.
1851                  */
1852                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1853                 if (error)
1854                         break;
1855
1856                 /*
1857                  * INADDR_ANY is used to remove a previous selection.
1858                  * When no interface is selected, a default one is
1859                  * chosen every time a multicast packet is sent.
1860                  */
1861                 if (addr.s_addr == INADDR_ANY) {
1862                         imo->imo_multicast_ifp = NULL;
1863                         break;
1864                 }
1865                 /*
1866                  * The selected interface is identified by its local
1867                  * IP address.  Find the interface and confirm that
1868                  * it supports multicasting.
1869                  */
1870                 crit_enter();
1871                 ifp = ip_multicast_if(&addr, &ifindex);
1872                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1873                         crit_exit();
1874                         error = EADDRNOTAVAIL;
1875                         break;
1876                 }
1877                 imo->imo_multicast_ifp = ifp;
1878                 if (ifindex)
1879                         imo->imo_multicast_addr = addr;
1880                 else
1881                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1882                 crit_exit();
1883                 break;
1884
1885         case IP_MULTICAST_TTL:
1886                 /*
1887                  * Set the IP time-to-live for outgoing multicast packets.
1888                  * The original multicast API required a char argument,
1889                  * which is inconsistent with the rest of the socket API.
1890                  * We allow either a char or an int.
1891                  */
1892                 if (sopt->sopt_valsize == 1) {
1893                         u_char ttl;
1894                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1895                         if (error)
1896                                 break;
1897                         imo->imo_multicast_ttl = ttl;
1898                 } else {
1899                         u_int ttl;
1900                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1901                         if (error)
1902                                 break;
1903                         if (ttl > 255)
1904                                 error = EINVAL;
1905                         else
1906                                 imo->imo_multicast_ttl = ttl;
1907                 }
1908                 break;
1909
1910         case IP_MULTICAST_LOOP:
1911                 /*
1912                  * Set the loopback flag for outgoing multicast packets.
1913                  * Must be zero or one.  The original multicast API required a
1914                  * char argument, which is inconsistent with the rest
1915                  * of the socket API.  We allow either a char or an int.
1916                  */
1917                 if (sopt->sopt_valsize == 1) {
1918                         u_char loop;
1919
1920                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1921                         if (error)
1922                                 break;
1923                         imo->imo_multicast_loop = !!loop;
1924                 } else {
1925                         u_int loop;
1926
1927                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1928                                             sizeof loop);
1929                         if (error)
1930                                 break;
1931                         imo->imo_multicast_loop = !!loop;
1932                 }
1933                 break;
1934
1935         case IP_ADD_MEMBERSHIP:
1936                 /*
1937                  * Add a multicast group membership.
1938                  * Group must be a valid IP multicast address.
1939                  */
1940                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1941                 if (error)
1942                         break;
1943
1944                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1945                         error = EINVAL;
1946                         break;
1947                 }
1948                 crit_enter();
1949                 /*
1950                  * If no interface address was provided, use the interface of
1951                  * the route to the given multicast address.
1952                  */
1953                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1954                         struct sockaddr_in dst;
1955                         struct rtentry *rt;
1956
1957                         bzero(&dst, sizeof(struct sockaddr_in));
1958                         dst.sin_len = sizeof(struct sockaddr_in);
1959                         dst.sin_family = AF_INET;
1960                         dst.sin_addr = mreq.imr_multiaddr;
1961                         rt = rtlookup((struct sockaddr *)&dst);
1962                         if (rt == NULL) {
1963                                 error = EADDRNOTAVAIL;
1964                                 crit_exit();
1965                                 break;
1966                         }
1967                         --rt->rt_refcnt;
1968                         ifp = rt->rt_ifp;
1969                 } else {
1970                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1971                 }
1972
1973                 /*
1974                  * See if we found an interface, and confirm that it
1975                  * supports multicast.
1976                  */
1977                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1978                         error = EADDRNOTAVAIL;
1979                         crit_exit();
1980                         break;
1981                 }
1982                 /*
1983                  * See if the membership already exists or if all the
1984                  * membership slots are full.
1985                  */
1986                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1987                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1988                             imo->imo_membership[i]->inm_addr.s_addr
1989                                                 == mreq.imr_multiaddr.s_addr)
1990                                 break;
1991                 }
1992                 if (i < imo->imo_num_memberships) {
1993                         error = EADDRINUSE;
1994                         crit_exit();
1995                         break;
1996                 }
1997                 if (i == IP_MAX_MEMBERSHIPS) {
1998                         error = ETOOMANYREFS;
1999                         crit_exit();
2000                         break;
2001                 }
2002                 /*
2003                  * Everything looks good; add a new record to the multicast
2004                  * address list for the given interface.
2005                  */
2006                 if ((imo->imo_membership[i] =
2007                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2008                         error = ENOBUFS;
2009                         crit_exit();
2010                         break;
2011                 }
2012                 ++imo->imo_num_memberships;
2013                 crit_exit();
2014                 break;
2015
2016         case IP_DROP_MEMBERSHIP:
2017                 /*
2018                  * Drop a multicast group membership.
2019                  * Group must be a valid IP multicast address.
2020                  */
2021                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2022                 if (error)
2023                         break;
2024
2025                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2026                         error = EINVAL;
2027                         break;
2028                 }
2029
2030                 crit_enter();
2031                 /*
2032                  * If an interface address was specified, get a pointer
2033                  * to its ifnet structure.
2034                  */
2035                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2036                         ifp = NULL;
2037                 else {
2038                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2039                         if (ifp == NULL) {
2040                                 error = EADDRNOTAVAIL;
2041                                 crit_exit();
2042                                 break;
2043                         }
2044                 }
2045                 /*
2046                  * Find the membership in the membership array.
2047                  */
2048                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2049                         if ((ifp == NULL ||
2050                              imo->imo_membership[i]->inm_ifp == ifp) &&
2051                             imo->imo_membership[i]->inm_addr.s_addr ==
2052                             mreq.imr_multiaddr.s_addr)
2053                                 break;
2054                 }
2055                 if (i == imo->imo_num_memberships) {
2056                         error = EADDRNOTAVAIL;
2057                         crit_exit();
2058                         break;
2059                 }
2060                 /*
2061                  * Give up the multicast address record to which the
2062                  * membership points.
2063                  */
2064                 in_delmulti(imo->imo_membership[i]);
2065                 /*
2066                  * Remove the gap in the membership array.
2067                  */
2068                 for (++i; i < imo->imo_num_memberships; ++i)
2069                         imo->imo_membership[i-1] = imo->imo_membership[i];
2070                 --imo->imo_num_memberships;
2071                 crit_exit();
2072                 break;
2073
2074         default:
2075                 error = EOPNOTSUPP;
2076                 break;
2077         }
2078
2079         /*
2080          * If all options have default values, no need to keep the mbuf.
2081          */
2082         if (imo->imo_multicast_ifp == NULL &&
2083             imo->imo_multicast_vif == -1 &&
2084             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2085             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2086             imo->imo_num_memberships == 0) {
2087                 kfree(*imop, M_IPMOPTS);
2088                 *imop = NULL;
2089         }
2090
2091         return (error);
2092 }
2093
2094 /*
2095  * Return the IP multicast options in response to user getsockopt().
2096  */
2097 static int
2098 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2099 {
2100         struct in_addr addr;
2101         struct in_ifaddr *ia;
2102         int error, optval;
2103         u_char coptval;
2104
2105         error = 0;
2106         switch (sopt->sopt_name) {
2107         case IP_MULTICAST_VIF:
2108                 if (imo != NULL)
2109                         optval = imo->imo_multicast_vif;
2110                 else
2111                         optval = -1;
2112                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2113                 break;
2114
2115         case IP_MULTICAST_IF:
2116                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2117                         addr.s_addr = INADDR_ANY;
2118                 else if (imo->imo_multicast_addr.s_addr) {
2119                         /* return the value user has set */
2120                         addr = imo->imo_multicast_addr;
2121                 } else {
2122                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2123                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2124                                 : IA_SIN(ia)->sin_addr.s_addr;
2125                 }
2126                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2127                 break;
2128
2129         case IP_MULTICAST_TTL:
2130                 if (imo == NULL)
2131                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2132                 else
2133                         optval = coptval = imo->imo_multicast_ttl;
2134                 if (sopt->sopt_valsize == 1)
2135                         soopt_from_kbuf(sopt, &coptval, 1);
2136                 else
2137                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2138                 break;
2139
2140         case IP_MULTICAST_LOOP:
2141                 if (imo == NULL)
2142                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2143                 else
2144                         optval = coptval = imo->imo_multicast_loop;
2145                 if (sopt->sopt_valsize == 1)
2146                         soopt_from_kbuf(sopt, &coptval, 1);
2147                 else
2148                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2149                 break;
2150
2151         default:
2152                 error = ENOPROTOOPT;
2153                 break;
2154         }
2155         return (error);
2156 }
2157
2158 /*
2159  * Discard the IP multicast options.
2160  */
2161 void
2162 ip_freemoptions(struct ip_moptions *imo)
2163 {
2164         int i;
2165
2166         if (imo != NULL) {
2167                 for (i = 0; i < imo->imo_num_memberships; ++i)
2168                         in_delmulti(imo->imo_membership[i]);
2169                 kfree(imo, M_IPMOPTS);
2170         }
2171 }
2172
2173 /*
2174  * Routine called from ip_output() to loop back a copy of an IP multicast
2175  * packet to the input queue of a specified interface.  Note that this
2176  * calls the output routine of the loopback "driver", but with an interface
2177  * pointer that might NOT be a loopback interface -- evil, but easier than
2178  * replicating that code here.
2179  */
2180 static void
2181 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2182              int hlen)
2183 {
2184         struct ip *ip;
2185         struct mbuf *copym;
2186
2187         copym = m_copypacket(m, MB_DONTWAIT);
2188         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2189                 copym = m_pullup(copym, hlen);
2190         if (copym != NULL) {
2191                 /*
2192                  * if the checksum hasn't been computed, mark it as valid
2193                  */
2194                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2195                         in_delayed_cksum(copym);
2196                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2197                         copym->m_pkthdr.csum_flags |=
2198                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2199                         copym->m_pkthdr.csum_data = 0xffff;
2200                 }
2201                 /*
2202                  * We don't bother to fragment if the IP length is greater
2203                  * than the interface's MTU.  Can this possibly matter?
2204                  */
2205                 ip = mtod(copym, struct ip *);
2206                 ip->ip_len = htons(ip->ip_len);
2207                 ip->ip_off = htons(ip->ip_off);
2208                 ip->ip_sum = 0;
2209                 if (ip->ip_vhl == IP_VHL_BORING) {
2210                         ip->ip_sum = in_cksum_hdr(ip);
2211                 } else {
2212                         ip->ip_sum = in_cksum(copym, hlen);
2213                 }
2214                 /*
2215                  * NB:
2216                  * It's not clear whether there are any lingering
2217                  * reentrancy problems in other areas which might
2218                  * be exposed by using ip_input directly (in
2219                  * particular, everything which modifies the packet
2220                  * in-place).  Yet another option is using the
2221                  * protosw directly to deliver the looped back
2222                  * packet.  For the moment, we'll err on the side
2223                  * of safety by using if_simloop().
2224                  */
2225 #if 1 /* XXX */
2226                 if (dst->sin_family != AF_INET) {
2227                         kprintf("ip_mloopback: bad address family %d\n",
2228                                                 dst->sin_family);
2229                         dst->sin_family = AF_INET;
2230                 }
2231 #endif
2232
2233 #ifdef notdef
2234                 copym->m_pkthdr.rcvif = ifp;
2235                 ip_input(copym);
2236 #else
2237                 if_simloop(ifp, copym, dst->sin_family, 0);
2238 #endif
2239         }
2240 }