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. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  34  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  35  * $DragonFly: src/sys/netinet/ip_output.c,v 1.18 2004/08/26 20:57:02 dillon Exp $
  36  */
  37
  38 #define _IP_VHL
  39
  40 #include "opt_ipfw.h"
  41 #include "opt_ipdn.h"
  42 #include "opt_ipdivert.h"
  43 #include "opt_ipfilter.h"
  44 #include "opt_ipsec.h"
  45 #include "opt_random_ip_id.h"
  46 #include "opt_mbuf_stress_test.h"
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/kernel.h>
  51 #include <sys/malloc.h>
  52 #include <sys/mbuf.h>
  53 #include <sys/protosw.h>
  54 #include <sys/socket.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/proc.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/in_cksum.h>
  59
  60 #include <net/if.h>
  61 #include <net/netisr.h>
  62 #include <net/pfil.h>
  63 #include <net/route.h>
  64
  65 #include <netinet/in.h>
  66 #include <netinet/in_systm.h>
  67 #include <netinet/ip.h>
  68 #include <netinet/in_pcb.h>
  69 #include <netinet/in_var.h>
  70 #include <netinet/ip_var.h>
  71
  72 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  73
  74 #ifdef IPSEC
  75 #include <netinet6/ipsec.h>
  76 #include <netproto/key/key.h>
  77 #ifdef IPSEC_DEBUG
  78 #include <netproto/key/key_debug.h>
  79 #else
  80 #define KEYDEBUG(lev,arg)
  81 #endif
  82 #endif /*IPSEC*/
  83
  84 #ifdef FAST_IPSEC
  85 #include <netipsec/ipsec.h>
  86 #include <netipsec/xform.h>
  87 #include <netipsec/key.h>
  88 #endif /*FAST_IPSEC*/
  89
  90 #include <net/ipfw/ip_fw.h>
  91 #include <net/dummynet/ip_dummynet.h>
  92
  93 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
  94                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  95                                   (ntohl(a.s_addr)>>16)&0xFF,\
  96                                   (ntohl(a.s_addr)>>8)&0xFF,\
  97                                   (ntohl(a.s_addr))&0xFF, y);
  98
  99 u_short ip_id;
 100
 101 #ifdef MBUF_STRESS_TEST
 102 int mbuf_frag_size = 0;
 103 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 104         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 105 #endif
 106
 107 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 108 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 109 static void     ip_mloopback
 110         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 111 static int      ip_getmoptions
 112         (struct sockopt *, struct ip_moptions *);
 113 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 114 static int      ip_setmoptions
 115         (struct sockopt *, struct ip_moptions **);
 116
 117 int     ip_optcopy(struct ip *, struct ip *);
 118
 119
 120 extern  struct protosw inetsw[];
 121
 122 /*
 123  * IP output.  The packet in mbuf chain m contains a skeletal IP
 124  * header (with len, off, ttl, proto, tos, src, dst).
 125  * The mbuf chain containing the packet will be freed.
 126  * The mbuf opt, if present, will not be freed.
 127  */
 128 int
 129 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 130         int flags, struct ip_moptions *imo, struct inpcb *inp)
 131 {
 132         struct ip *ip;
 133         struct ifnet *ifp = NULL;       /* keep compiler happy */
 134         struct mbuf *m;
 135         int hlen = sizeof (struct ip);
 136         int len, off, error = 0;
 137         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 138         struct in_ifaddr *ia = NULL;
 139         int isbroadcast, sw_csum;
 140         struct in_addr pkt_dst;
 141 #ifdef IPSEC
 142         struct route iproute;
 143         struct secpolicy *sp = NULL;
 144         struct socket *so = inp ? inp->inp_socket : NULL;
 145 #endif
 146 #ifdef FAST_IPSEC
 147         struct route iproute;
 148         struct m_tag *mtag;
 149         struct secpolicy *sp = NULL;
 150         struct tdb_ident *tdbi;
 151         int s;
 152 #endif /* FAST_IPSEC */
 153         struct ip_fw_args args;
 154         int src_was_INADDR_ANY = 0;     /* as the name says... */
 155
 156         args.eh = NULL;
 157         args.rule = NULL;
 158         args.next_hop = NULL;
 159         args.divert_rule = 0;                   /* divert cookie */
 160
 161         /* Grab info from MT_TAG mbufs prepended to the chain. */
 162         for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
 163                 switch(m0->_m_tag_id) {
 164                 default:
 165                         printf("ip_output: unrecognised MT_TAG tag %d\n",
 166                             m0->_m_tag_id);
 167                         break;
 168
 169                 case PACKET_TAG_DUMMYNET:
 170                         /*
 171                          * the packet was already tagged, so part of the
 172                          * processing was already done, and we need to go down.
 173                          * Get parameters from the header.
 174                          */
 175                         args.rule = ((struct dn_pkt *)m0)->rule;
 176                         opt = NULL ;
 177                         ro = & ( ((struct dn_pkt *)m0)->ro ) ;
 178                         imo = NULL ;
 179                         dst = ((struct dn_pkt *)m0)->dn_dst ;
 180                         ifp = ((struct dn_pkt *)m0)->ifp ;
 181                         flags = ((struct dn_pkt *)m0)->flags ;
 182                         break;
 183
 184                 case PACKET_TAG_DIVERT:
 185                         args.divert_rule = (int)m0->m_data & 0xffff;
 186                         break;
 187
 188                 case PACKET_TAG_IPFORWARD:
 189                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 190                         break;
 191                 }
 192         }
 193         m = m0;
 194
 195         KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
 196 #ifndef FAST_IPSEC
 197         KASSERT(ro != NULL, ("ip_output: no route, proto %d",
 198             mtod(m, struct ip *)->ip_p));
 199 #endif
 200
 201         if (args.rule != NULL) {        /* dummynet already saw us */
 202                 ip = mtod(m, struct ip *);
 203                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 204                 if (ro->ro_rt)
 205                         ia = ifatoia(ro->ro_rt->rt_ifa);
 206                 goto sendit;
 207         }
 208
 209         if (opt) {
 210                 len = 0;
 211                 m = ip_insertoptions(m, opt, &len);
 212                 if (len != 0)
 213                         hlen = len;
 214         }
 215         ip = mtod(m, struct ip *);
 216         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 217
 218         /*
 219          * Fill in IP header.
 220          */
 221         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 222                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 223                 ip->ip_off &= IP_DF;
 224 #ifdef RANDOM_IP_ID
 225                 ip->ip_id = ip_randomid();
 226 #else
 227                 ip->ip_id = htons(ip_id++);
 228 #endif
 229                 ipstat.ips_localout++;
 230         } else {
 231                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 232         }
 233
 234 #ifdef FAST_IPSEC
 235         if (ro == NULL) {
 236                 ro = &iproute;
 237                 bzero(ro, sizeof (*ro));
 238         }
 239 #endif /* FAST_IPSEC */
 240         dst = (struct sockaddr_in *)&ro->ro_dst;
 241         /*
 242          * If there is a cached route,
 243          * check that it is to the same destination
 244          * and is still up.  If not, free it and try again.
 245          * The address family should also be checked in case of sharing the
 246          * cache with IPv6.
 247          */
 248         if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
 249                           dst->sin_family != AF_INET ||
 250                           dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 251                 RTFREE(ro->ro_rt);
 252                 ro->ro_rt = (struct rtentry *)0;
 253         }
 254         if (ro->ro_rt == 0) {
 255                 bzero(dst, sizeof(*dst));
 256                 dst->sin_family = AF_INET;
 257                 dst->sin_len = sizeof(*dst);
 258                 dst->sin_addr = pkt_dst;
 259         }
 260         /*
 261          * If routing to interface only,
 262          * short circuit routing lookup.
 263          */
 264         if (flags & IP_ROUTETOIF) {
 265                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
 266                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
 267                         ipstat.ips_noroute++;
 268                         error = ENETUNREACH;
 269                         goto bad;
 270                 }
 271                 ifp = ia->ia_ifp;
 272                 ip->ip_ttl = 1;
 273                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 274         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 275             imo != NULL && imo->imo_multicast_ifp != NULL) {
 276                 /*
 277                  * Bypass the normal routing lookup for multicast
 278                  * packets if the interface is specified.
 279                  */
 280                 ifp = imo->imo_multicast_ifp;
 281                 IFP_TO_IA(ifp, ia);
 282                 isbroadcast = 0;        /* fool gcc */
 283         } else {
 284                 /*
 285                  * If this is the case, we probably don't want to allocate
 286                  * a protocol-cloned route since we didn't get one from the
 287                  * ULP.  This lets TCP do its thing, while not burdening
 288                  * forwarding or ICMP with the overhead of cloning a route.
 289                  * Of course, we still want to do any cloning requested by
 290                  * the link layer, as this is probably required in all cases
 291                  * for correct operation (as it is for ARP).
 292                  */
 293                 if (ro->ro_rt == 0)
 294                         rtalloc_ign(ro, RTF_PRCLONING);
 295                 if (ro->ro_rt == 0) {
 296                         ipstat.ips_noroute++;
 297                         error = EHOSTUNREACH;
 298                         goto bad;
 299                 }
 300                 ia = ifatoia(ro->ro_rt->rt_ifa);
 301                 ifp = ro->ro_rt->rt_ifp;
 302                 ro->ro_rt->rt_use++;
 303                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 304                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 305                 if (ro->ro_rt->rt_flags & RTF_HOST)
 306                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 307                 else
 308                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 309         }
 310         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 311                 struct in_multi *inm;
 312
 313                 m->m_flags |= M_MCAST;
 314                 /*
 315                  * IP destination address is multicast.  Make sure "dst"
 316                  * still points to the address in "ro".  (It may have been
 317                  * changed to point to a gateway address, above.)
 318                  */
 319                 dst = (struct sockaddr_in *)&ro->ro_dst;
 320                 /*
 321                  * See if the caller provided any multicast options
 322                  */
 323                 if (imo != NULL) {
 324                         ip->ip_ttl = imo->imo_multicast_ttl;
 325                         if (imo->imo_multicast_vif != -1)
 326                                 ip->ip_src.s_addr =
 327                                     ip_mcast_src ?
 328                                     ip_mcast_src(imo->imo_multicast_vif) :
 329                                     INADDR_ANY;
 330                 } else
 331                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 332                 /*
 333                  * Confirm that the outgoing interface supports multicast.
 334                  */
 335                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 336                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 337                                 ipstat.ips_noroute++;
 338                                 error = ENETUNREACH;
 339                                 goto bad;
 340                         }
 341                 }
 342                 /*
 343                  * If source address not specified yet, use address
 344                  * of outgoing interface.
 345                  */
 346                 if (ip->ip_src.s_addr == INADDR_ANY) {
 347                         /* Interface may have no addresses. */
 348                         if (ia != NULL)
 349                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 350                 }
 351
 352                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 353                 if (inm != NULL &&
 354                    (imo == NULL || imo->imo_multicast_loop)) {
 355                         /*
 356                          * If we belong to the destination multicast group
 357                          * on the outgoing interface, and the caller did not
 358                          * forbid loopback, loop back a copy.
 359                          */
 360                         ip_mloopback(ifp, m, dst, hlen);
 361                 }
 362                 else {
 363                         /*
 364                          * If we are acting as a multicast router, perform
 365                          * multicast forwarding as if the packet had just
 366                          * arrived on the interface to which we are about
 367                          * to send.  The multicast forwarding function
 368                          * recursively calls this function, using the
 369                          * IP_FORWARDING flag to prevent infinite recursion.
 370                          *
 371                          * Multicasts that are looped back by ip_mloopback(),
 372                          * above, will be forwarded by the ip_input() routine,
 373                          * if necessary.
 374                          */
 375                         if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 376                                 /*
 377                                  * If rsvp daemon is not running, do not
 378                                  * set ip_moptions. This ensures that the packet
 379                                  * is multicast and not just sent down one link
 380                                  * as prescribed by rsvpd.
 381                                  */
 382                                 if (!rsvp_on)
 383                                         imo = NULL;
 384                                 if (ip_mforward &&
 385                                     ip_mforward(ip, ifp, m, imo) != 0) {
 386                                         m_freem(m);
 387                                         goto done;
 388                                 }
 389                         }
 390                 }
 391
 392                 /*
 393                  * Multicasts with a time-to-live of zero may be looped-
 394                  * back, above, but must not be transmitted on a network.
 395                  * Also, multicasts addressed to the loopback interface
 396                  * are not sent -- the above call to ip_mloopback() will
 397                  * loop back a copy if this host actually belongs to the
 398                  * destination group on the loopback interface.
 399                  */
 400                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 401                         m_freem(m);
 402                         goto done;
 403                 }
 404
 405                 goto sendit;
 406         }
 407 #ifndef notdef
 408         /*
 409          * If the source address is not specified yet, use the address
 410          * of the outoing interface. In case, keep note we did that, so
 411          * if the the firewall changes the next-hop causing the output
 412          * interface to change, we can fix that.
 413          */
 414         if (ip->ip_src.s_addr == INADDR_ANY) {
 415                 /* Interface may have no addresses. */
 416                 if (ia != NULL) {
 417                         ip->ip_src = IA_SIN(ia)->sin_addr;
 418                         src_was_INADDR_ANY = 1;
 419                 }
 420         }
 421 #endif /* notdef */
 422         /*
 423          * Verify that we have any chance at all of being able to queue
 424          *      the packet or packet fragments
 425          */
 426         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 427                 ifp->if_snd.ifq_maxlen) {
 428                         error = ENOBUFS;
 429                         ipstat.ips_odropped++;
 430                         goto bad;
 431         }
 432
 433         /*
 434          * Look for broadcast address and
 435          * verify user is allowed to send
 436          * such a packet.
 437          */
 438         if (isbroadcast) {
 439                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 440                         error = EADDRNOTAVAIL;
 441                         goto bad;
 442                 }
 443                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 444                         error = EACCES;
 445                         goto bad;
 446                 }
 447                 /* don't allow broadcast messages to be fragmented */
 448                 if (ip->ip_len > ifp->if_mtu) {
 449                         error = EMSGSIZE;
 450                         goto bad;
 451                 }
 452                 m->m_flags |= M_BCAST;
 453         } else {
 454                 m->m_flags &= ~M_BCAST;
 455         }
 456
 457 sendit:
 458 #ifdef IPSEC
 459         /* get SP for this packet */
 460         if (so == NULL)
 461                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 462         else
 463                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 464
 465         if (sp == NULL) {
 466                 ipsecstat.out_inval++;
 467                 goto bad;
 468         }
 469
 470         error = 0;
 471
 472         /* check policy */
 473         switch (sp->policy) {
 474         case IPSEC_POLICY_DISCARD:
 475                 /*
 476                  * This packet is just discarded.
 477                  */
 478                 ipsecstat.out_polvio++;
 479                 goto bad;
 480
 481         case IPSEC_POLICY_BYPASS:
 482         case IPSEC_POLICY_NONE:
 483                 /* no need to do IPsec. */
 484                 goto skip_ipsec;
 485
 486         case IPSEC_POLICY_IPSEC:
 487                 if (sp->req == NULL) {
 488                         /* acquire a policy */
 489                         error = key_spdacquire(sp);
 490                         goto bad;
 491                 }
 492                 break;
 493
 494         case IPSEC_POLICY_ENTRUST:
 495         default:
 496                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
 497         }
 498     {
 499         struct ipsec_output_state state;
 500         bzero(&state, sizeof(state));
 501         state.m = m;
 502         if (flags & IP_ROUTETOIF) {
 503                 state.ro = &iproute;
 504                 bzero(&iproute, sizeof(iproute));
 505         } else
 506                 state.ro = ro;
 507         state.dst = (struct sockaddr *)dst;
 508
 509         ip->ip_sum = 0;
 510
 511         /*
 512          * XXX
 513          * delayed checksums are not currently compatible with IPsec
 514          */
 515         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 516                 in_delayed_cksum(m);
 517                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 518         }
 519
 520         ip->ip_len = htons(ip->ip_len);
 521         ip->ip_off = htons(ip->ip_off);
 522
 523         error = ipsec4_output(&state, sp, flags);
 524
 525         m = state.m;
 526         if (flags & IP_ROUTETOIF) {
 527                 /*
 528                  * if we have tunnel mode SA, we may need to ignore
 529                  * IP_ROUTETOIF.
 530                  */
 531                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 532                         flags &= ~IP_ROUTETOIF;
 533                         ro = state.ro;
 534                 }
 535         } else
 536                 ro = state.ro;
 537         dst = (struct sockaddr_in *)state.dst;
 538         if (error) {
 539                 /* mbuf is already reclaimed in ipsec4_output. */
 540                 m0 = NULL;
 541                 switch (error) {
 542                 case EHOSTUNREACH:
 543                 case ENETUNREACH:
 544                 case EMSGSIZE:
 545                 case ENOBUFS:
 546                 case ENOMEM:
 547                         break;
 548                 default:
 549                         printf("ip4_output (ipsec): error code %d\n", error);
 550                         /*fall through*/
 551                 case ENOENT:
 552                         /* don't show these error codes to the user */
 553                         error = 0;
 554                         break;
 555                 }
 556                 goto bad;
 557         }
 558     }
 559
 560         /* be sure to update variables that are affected by ipsec4_output() */
 561         ip = mtod(m, struct ip *);
 562 #ifdef _IP_VHL
 563         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 564 #else
 565         hlen = ip->ip_hl << 2;
 566 #endif
 567         if (ro->ro_rt == NULL) {
 568                 if ((flags & IP_ROUTETOIF) == 0) {
 569                         printf("ip_output: "
 570                                 "can't update route after IPsec processing\n");
 571                         error = EHOSTUNREACH;   /*XXX*/
 572                         goto bad;
 573                 }
 574         } else {
 575                 ia = ifatoia(ro->ro_rt->rt_ifa);
 576                 ifp = ro->ro_rt->rt_ifp;
 577         }
 578
 579         /* make it flipped, again. */
 580         ip->ip_len = ntohs(ip->ip_len);
 581         ip->ip_off = ntohs(ip->ip_off);
 582 skip_ipsec:
 583 #endif /*IPSEC*/
 584 #ifdef FAST_IPSEC
 585         /*
 586          * Check the security policy (SP) for the packet and, if
 587          * required, do IPsec-related processing.  There are two
 588          * cases here; the first time a packet is sent through
 589          * it will be untagged and handled by ipsec4_checkpolicy.
 590          * If the packet is resubmitted to ip_output (e.g. after
 591          * AH, ESP, etc. processing), there will be a tag to bypass
 592          * the lookup and related policy checking.
 593          */
 594         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 595         s = splnet();
 596         if (mtag != NULL) {
 597                 tdbi = (struct tdb_ident *)(mtag + 1);
 598                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 599                 if (sp == NULL)
 600                         error = -EINVAL;        /* force silent drop */
 601                 m_tag_delete(m, mtag);
 602         } else {
 603                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 604                                         &error, inp);
 605         }
 606         /*
 607          * There are four return cases:
 608          *    sp != NULL                    apply IPsec policy
 609          *    sp == NULL, error == 0        no IPsec handling needed
 610          *    sp == NULL, error == -EINVAL  discard packet w/o error
 611          *    sp == NULL, error != 0        discard packet, report error
 612          */
 613         if (sp != NULL) {
 614                 /* Loop detection, check if ipsec processing already done */
 615                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 616                 for (mtag = m_tag_first(m); mtag != NULL;
 617                      mtag = m_tag_next(m, mtag)) {
 618                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 619                                 continue;
 620                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 621                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 622                                 continue;
 623                         /*
 624                          * Check if policy has an SA associated with it.
 625                          * This can happen when an SP has yet to acquire
 626                          * an SA; e.g. on first reference.  If it occurs,
 627                          * then we let ipsec4_process_packet do its thing.
 628                          */
 629                         if (sp->req->sav == NULL)
 630                                 break;
 631                         tdbi = (struct tdb_ident *)(mtag + 1);
 632                         if (tdbi->spi == sp->req->sav->spi &&
 633                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 634                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 635                                  sizeof (union sockaddr_union)) == 0) {
 636                                 /*
 637                                  * No IPsec processing is needed, free
 638                                  * reference to SP.
 639                                  *
 640                                  * NB: null pointer to avoid free at
 641                                  *     done: below.
 642                                  */
 643                                 KEY_FREESP(&sp), sp = NULL;
 644                                 splx(s);
 645                                 goto spd_done;
 646                         }
 647                 }
 648
 649                 /*
 650                  * Do delayed checksums now because we send before
 651                  * this is done in the normal processing path.
 652                  */
 653                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 654                         in_delayed_cksum(m);
 655                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 656                 }
 657
 658                 ip->ip_len = htons(ip->ip_len);
 659                 ip->ip_off = htons(ip->ip_off);
 660
 661                 /* NB: callee frees mbuf */
 662                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 663                 /*
 664                  * Preserve KAME behaviour: ENOENT can be returned
 665                  * when an SA acquire is in progress.  Don't propagate
 666                  * this to user-level; it confuses applications.
 667                  *
 668                  * XXX this will go away when the SADB is redone.
 669                  */
 670                 if (error == ENOENT)
 671                         error = 0;
 672                 splx(s);
 673                 goto done;
 674         } else {
 675                 splx(s);
 676
 677                 if (error != 0) {
 678                         /*
 679                          * Hack: -EINVAL is used to signal that a packet
 680                          * should be silently discarded.  This is typically
 681                          * because we asked key management for an SA and
 682                          * it was delayed (e.g. kicked up to IKE).
 683                          */
 684                         if (error == -EINVAL)
 685                                 error = 0;
 686                         goto bad;
 687                 } else {
 688                         /* No IPsec processing for this packet. */
 689                 }
 690 #ifdef notyet
 691                 /*
 692                  * If deferred crypto processing is needed, check that
 693                  * the interface supports it.
 694                  */
 695                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 696                 if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
 697                         /* notify IPsec to do its own crypto */
 698                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
 699                         error = EHOSTUNREACH;
 700                         goto bad;
 701                 }
 702 #endif
 703         }
 704 spd_done:
 705 #endif /* FAST_IPSEC */
 706         /*
 707          * IpHack's section.
 708          * - Xlate: translate packet's addr/port (NAT).
 709          * - Firewall: deny/allow/etc.
 710          * - Wrap: fake packet's addr/port <unimpl.>
 711          * - Encapsulate: put it in another IP and send out. <unimp.>
 712          */
 713
 714         /*
 715          * Run through list of hooks for output packets.
 716          */
 717         if (pfil_has_hooks(&inet_pfil_hook)) {
 718                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 719                 if (error != 0 || m == NULL)
 720                         goto done;
 721                 ip = mtod(m, struct ip *);
 722         }
 723
 724         /*
 725          * Check with the firewall...
 726          * but not if we are already being fwd'd from a firewall.
 727          */
 728         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 729                 struct sockaddr_in *old = dst;
 730
 731                 args.m = m;
 732                 args.next_hop = dst;
 733                 args.oif = ifp;
 734                 off = ip_fw_chk_ptr(&args);
 735                 m = args.m;
 736                 dst = args.next_hop;
 737
 738                 /*
 739                  * On return we must do the following:
 740                  * m == NULL    -> drop the pkt (old interface, deprecated)
 741                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 742                  * 1<=off<= 0xffff              -> DIVERT
 743                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 744                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 745                  * dst != old                   -> IPFIREWALL_FORWARD
 746                  * off==0, dst==old             -> accept
 747                  * If some of the above modules are not compiled in, then
 748                  * we should't have to check the corresponding condition
 749                  * (because the ipfw control socket should not accept
 750                  * unsupported rules), but better play safe and drop
 751                  * packets in case of doubt.
 752                  */
 753                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 754                         if (m)
 755                                 m_freem(m);
 756                         error = EACCES;
 757                         goto done;
 758                 }
 759                 ip = mtod(m, struct ip *);
 760                 if (off == 0 && dst == old)             /* common case */
 761                         goto pass;
 762                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
 763                         /*
 764                          * pass the pkt to dummynet. Need to include
 765                          * pipe number, m, ifp, ro, dst because these are
 766                          * not recomputed in the next pass.
 767                          * All other parameters have been already used and
 768                          * so they are not needed anymore.
 769                          * XXX note: if the ifp or ro entry are deleted
 770                          * while a pkt is in dummynet, we are in trouble!
 771                          */
 772                         args.ro = ro;
 773                         args.dst = dst;
 774                         args.flags = flags;
 775
 776                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
 777                                 &args);
 778                         goto done;
 779                 }
 780 #ifdef IPDIVERT
 781                 if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
 782                         struct mbuf *clone = NULL;
 783
 784                         /* Clone packet if we're doing a 'tee' */
 785                         if ((off & IP_FW_PORT_TEE_FLAG) != 0)
 786                                 clone = m_dup(m, MB_DONTWAIT);
 787
 788                         /*
 789                          * XXX
 790                          * delayed checksums are not currently compatible
 791                          * with divert sockets.
 792                          */
 793                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 794                                 in_delayed_cksum(m);
 795                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 796                         }
 797
 798                         /* Restore packet header fields to original values */
 799                         ip->ip_len = htons(ip->ip_len);
 800                         ip->ip_off = htons(ip->ip_off);
 801
 802                         /* Deliver packet to divert input routine */
 803                         divert_packet(m, 0, off & 0xffff, args.divert_rule);
 804
 805                         /* If 'tee', continue with original packet */
 806                         if (clone != NULL) {
 807                                 m = clone;
 808                                 ip = mtod(m, struct ip *);
 809                                 goto pass;
 810                         }
 811                         goto done;
 812                 }
 813 #endif
 814
 815                 /* IPFIREWALL_FORWARD */
 816                 /*
 817                  * Check dst to make sure it is directly reachable on the
 818                  * interface we previously thought it was.
 819                  * If it isn't (which may be likely in some situations) we have
 820                  * to re-route it (ie, find a route for the next-hop and the
 821                  * associated interface) and set them here. This is nested
 822                  * forwarding which in most cases is undesirable, except where
 823                  * such control is nigh impossible. So we do it here.
 824                  * And I'm babbling.
 825                  */
 826                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 827 #if 0
 828                         /*
 829                          * XXX To improve readability, this block should be
 830                          * changed into a function call as below:
 831                          */
 832                         error = ip_ipforward(&m, &dst, &ifp);
 833                         if (error)
 834                                 goto bad;
 835                         if (m == NULL) /* ip_input consumed the mbuf */
 836                                 goto done;
 837 #else
 838                         struct in_ifaddr *ia;
 839
 840                         /*
 841                          * XXX sro_fwd below is static, and a pointer
 842                          * to it gets passed to routines downstream.
 843                          * This could have surprisingly bad results in
 844                          * practice, because its content is overwritten
 845                          * by subsequent packets.
 846                          */
 847                         /* There must be a better way to do this next line... */
 848                         static struct route sro_fwd;
 849                         struct route *ro_fwd = &sro_fwd;
 850
 851 #if 0
 852                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 853                             dst->sin_addr, "\n");
 854 #endif
 855
 856                         /*
 857                          * We need to figure out if we have been forwarded
 858                          * to a local socket. If so, then we should somehow
 859                          * "loop back" to ip_input, and get directed to the
 860                          * PCB as if we had received this packet. This is
 861                          * because it may be dificult to identify the packets
 862                          * you want to forward until they are being output
 863                          * and have selected an interface. (e.g. locally
 864                          * initiated packets) If we used the loopback inteface,
 865                          * we would not be able to control what happens
 866                          * as the packet runs through ip_input() as
 867                          * it is done through a ISR.
 868                          */
 869                         LIST_FOREACH(ia,
 870                             INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 871                                 /*
 872                                  * If the addr to forward to is one
 873                                  * of ours, we pretend to
 874                                  * be the destination for this packet.
 875                                  */
 876                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 877                                                  dst->sin_addr.s_addr)
 878                                         break;
 879                         }
 880                         if (ia) {       /* tell ip_input "dont filter" */
 881                                 struct m_hdr tag;
 882
 883                                 tag.mh_type = MT_TAG;
 884                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 885                                 tag.mh_data = (caddr_t)args.next_hop;
 886                                 tag.mh_next = m;
 887
 888                                 if (m->m_pkthdr.rcvif == NULL)
 889                                         m->m_pkthdr.rcvif = ifunit("lo0");
 890                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 891                                         m->m_pkthdr.csum_flags |=
 892                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 893                                         m0->m_pkthdr.csum_data = 0xffff;
 894                                 }
 895                                 m->m_pkthdr.csum_flags |=
 896                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 897                                 ip->ip_len = htons(ip->ip_len);
 898                                 ip->ip_off = htons(ip->ip_off);
 899                                 ip_input((struct mbuf *)&tag);
 900                                 goto done;
 901                         }
 902                         /* Some of the logic for this was
 903                          * nicked from above.
 904                          *
 905                          * This rewrites the cached route in a local PCB.
 906                          * Is this what we want to do?
 907                          */
 908                         bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
 909
 910                         ro_fwd->ro_rt = 0;
 911                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 912
 913                         if (ro_fwd->ro_rt == 0) {
 914                                 ipstat.ips_noroute++;
 915                                 error = EHOSTUNREACH;
 916                                 goto bad;
 917                         }
 918
 919                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 920                         ifp = ro_fwd->ro_rt->rt_ifp;
 921                         ro_fwd->ro_rt->rt_use++;
 922                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 923                                 dst = (struct sockaddr_in *)
 924                                         ro_fwd->ro_rt->rt_gateway;
 925                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 926                                 isbroadcast =
 927                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 928                         else
 929                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 930                         if (ro->ro_rt)
 931                                 RTFREE(ro->ro_rt);
 932                         ro->ro_rt = ro_fwd->ro_rt;
 933                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 934
 935 #endif  /* ... block to be put into a function */
 936                         /*
 937                          * If we added a default src ip earlier,
 938                          * which would have been gotten from the-then
 939                          * interface, do it again, from the new one.
 940                          */
 941                         if (src_was_INADDR_ANY)
 942                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 943                         goto pass ;
 944                 }
 945
 946                 /*
 947                  * if we get here, none of the above matches, and
 948                  * we have to drop the pkt
 949                  */
 950                 m_freem(m);
 951                 error = EACCES; /* not sure this is the right error msg */
 952                 goto done;
 953         }
 954
 955 pass:
 956         /* 127/8 must not appear on wire - RFC1122. */
 957         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 958             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 959                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 960                         ipstat.ips_badaddr++;
 961                         error = EADDRNOTAVAIL;
 962                         goto bad;
 963                 }
 964         }
 965
 966         m->m_pkthdr.csum_flags |= CSUM_IP;
 967         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 968         if (sw_csum & CSUM_DELAY_DATA) {
 969                 in_delayed_cksum(m);
 970                 sw_csum &= ~CSUM_DELAY_DATA;
 971         }
 972         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 973
 974         /*
 975          * If small enough for interface, or the interface will take
 976          * care of the fragmentation for us, can just send directly.
 977          */
 978         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 979             (ip->ip_off & IP_DF) == 0)) {
 980                 ip->ip_len = htons(ip->ip_len);
 981                 ip->ip_off = htons(ip->ip_off);
 982                 ip->ip_sum = 0;
 983                 if (sw_csum & CSUM_DELAY_IP) {
 984                         if (ip->ip_vhl == IP_VHL_BORING) {
 985                                 ip->ip_sum = in_cksum_hdr(ip);
 986                         } else {
 987                                 ip->ip_sum = in_cksum(m, hlen);
 988                         }
 989                 }
 990
 991                 /* Record statistics for this interface address. */
 992                 if (!(flags & IP_FORWARDING) && ia) {
 993                         ia->ia_ifa.if_opackets++;
 994                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 995                 }
 996
 997 #ifdef IPSEC
 998                 /* clean ipsec history once it goes out of the node */
 999                 ipsec_delaux(m);
1000 #endif
1001
1002 #ifdef MBUF_STRESS_TEST
1003                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1004                         struct mbuf *m1, *m2;
1005                         int length, tmp;
1006
1007                         tmp = length = m->m_pkthdr.len;
1008
1009                         while ((length -= mbuf_frag_size) >= 1) {
1010                                 m1 = m_split(m, length, MB_DONTWAIT);
1011                                 if (m1 == NULL)
1012                                         break;
1013                                 m1->m_flags &= ~M_PKTHDR;
1014                                 m2 = m;
1015                                 while (m2->m_next != NULL)
1016                                         m2 = m2->m_next;
1017                                 m2->m_next = m1;
1018                         }
1019                         m->m_pkthdr.len = tmp;
1020                 }
1021 #endif
1022                 error = (*ifp->if_output)(ifp, m,
1023                                 (struct sockaddr *)dst, ro->ro_rt);
1024                 goto done;
1025         }
1026
1027         if (ip->ip_off & IP_DF) {
1028                 error = EMSGSIZE;
1029                 /*
1030                  * This case can happen if the user changed the MTU
1031                  * of an interface after enabling IP on it.  Because
1032                  * most netifs don't keep track of routes pointing to
1033                  * them, there is no way for one to update all its
1034                  * routes when the MTU is changed.
1035                  */
1036                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1037                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1038                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1039                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1040                 }
1041                 ipstat.ips_cantfrag++;
1042                 goto bad;
1043         }
1044
1045         /*
1046          * Too large for interface; fragment if possible. If successful,
1047          * on return, m will point to a list of packets to be sent.
1048          */
1049         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1050         if (error)
1051                 goto bad;
1052         for (; m; m = m0) {
1053                 m0 = m->m_nextpkt;
1054                 m->m_nextpkt = 0;
1055 #ifdef IPSEC
1056                 /* clean ipsec history once it goes out of the node */
1057                 ipsec_delaux(m);
1058 #endif
1059                 if (error == 0) {
1060                         /* Record statistics for this interface address. */
1061                         if (ia != NULL) {
1062                                 ia->ia_ifa.if_opackets++;
1063                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1064                         }
1065
1066                         error = (*ifp->if_output)(ifp, m,
1067                             (struct sockaddr *)dst, ro->ro_rt);
1068                 } else
1069                         m_freem(m);
1070         }
1071
1072         if (error == 0)
1073                 ipstat.ips_fragmented++;
1074
1075 done:
1076 #ifdef IPSEC
1077         if (ro == &iproute && ro->ro_rt) {
1078                 RTFREE(ro->ro_rt);
1079                 ro->ro_rt = NULL;
1080         }
1081         if (sp != NULL) {
1082                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1083                         printf("DP ip_output call free SP:%p\n", sp));
1084                 key_freesp(sp);
1085         }
1086 #endif
1087 #ifdef FAST_IPSEC
1088         if (ro == &iproute && ro->ro_rt) {
1089                 RTFREE(ro->ro_rt);
1090                 ro->ro_rt = NULL;
1091         }
1092         if (sp != NULL)
1093                 KEY_FREESP(&sp);
1094 #endif
1095         return (error);
1096 bad:
1097         m_freem(m);
1098         goto done;
1099 }
1100
1101 /*
1102  * Create a chain of fragments which fit the given mtu. m_frag points to the
1103  * mbuf to be fragmented; on return it points to the chain with the fragments.
1104  * Return 0 if no error. If error, m_frag may contain a partially built
1105  * chain of fragments that should be freed by the caller.
1106  *
1107  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1108  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1109  */
1110 int
1111 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1112             u_long if_hwassist_flags, int sw_csum)
1113 {
1114         int error = 0;
1115         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1116         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1117         int off;
1118         struct mbuf *m0 = *m_frag;      /* the original packet          */
1119         int firstlen;
1120         struct mbuf **mnext;
1121         int nfrags;
1122
1123         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1124                 ipstat.ips_cantfrag++;
1125                 return EMSGSIZE;
1126         }
1127
1128         /*
1129          * Must be able to put at least 8 bytes per fragment.
1130          */
1131         if (len < 8)
1132                 return EMSGSIZE;
1133
1134         /*
1135          * If the interface will not calculate checksums on
1136          * fragmented packets, then do it here.
1137          */
1138         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1139             (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1140                 in_delayed_cksum(m0);
1141                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1142         }
1143
1144         if (len > PAGE_SIZE) {
1145                 /*
1146                  * Fragment large datagrams such that each segment
1147                  * contains a multiple of PAGE_SIZE amount of data,
1148                  * plus headers. This enables a receiver to perform
1149                  * page-flipping zero-copy optimizations.
1150                  *
1151                  * XXX When does this help given that sender and receiver
1152                  * could have different page sizes, and also mtu could
1153                  * be less than the receiver's page size ?
1154                  */
1155                 int newlen;
1156                 struct mbuf *m;
1157
1158                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1159                         off += m->m_len;
1160
1161                 /*
1162                  * firstlen (off - hlen) must be aligned on an
1163                  * 8-byte boundary
1164                  */
1165                 if (off < hlen)
1166                         goto smart_frag_failure;
1167                 off = ((off - hlen) & ~7) + hlen;
1168                 newlen = (~PAGE_MASK) & mtu;
1169                 if ((newlen + sizeof (struct ip)) > mtu) {
1170                         /* we failed, go back the default */
1171 smart_frag_failure:
1172                         newlen = len;
1173                         off = hlen + len;
1174                 }
1175                 len = newlen;
1176
1177         } else {
1178                 off = hlen + len;
1179         }
1180
1181         firstlen = off - hlen;
1182         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1183
1184         /*
1185          * Loop through length of segment after first fragment,
1186          * make new header and copy data of each part and link onto chain.
1187          * Here, m0 is the original packet, m is the fragment being created.
1188          * The fragments are linked off the m_nextpkt of the original
1189          * packet, which after processing serves as the first fragment.
1190          */
1191         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1192                 struct ip *mhip;        /* ip header on the fragment */
1193                 struct mbuf *m;
1194                 int mhlen = sizeof (struct ip);
1195
1196                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1197                 if (m == 0) {
1198                         error = ENOBUFS;
1199                         ipstat.ips_odropped++;
1200                         goto done;
1201                 }
1202                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1203                 /*
1204                  * In the first mbuf, leave room for the link header, then
1205                  * copy the original IP header including options. The payload
1206                  * goes into an additional mbuf chain returned by m_copy().
1207                  */
1208                 m->m_data += max_linkhdr;
1209                 mhip = mtod(m, struct ip *);
1210                 *mhip = *ip;
1211                 if (hlen > sizeof (struct ip)) {
1212                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1213                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1214                 }
1215                 m->m_len = mhlen;
1216                 /* XXX do we need to add ip->ip_off below ? */
1217                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1218                 if (off + len >= ip->ip_len) {  /* last fragment */
1219                         len = ip->ip_len - off;
1220                         m->m_flags |= M_LASTFRAG;
1221                 } else
1222                         mhip->ip_off |= IP_MF;
1223                 mhip->ip_len = htons((u_short)(len + mhlen));
1224                 m->m_next = m_copy(m0, off, len);
1225                 if (m->m_next == 0) {           /* copy failed */
1226                         m_free(m);
1227                         error = ENOBUFS;        /* ??? */
1228                         ipstat.ips_odropped++;
1229                         goto done;
1230                 }
1231                 m->m_pkthdr.len = mhlen + len;
1232                 m->m_pkthdr.rcvif = (struct ifnet *)0;
1233                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1234                 mhip->ip_off = htons(mhip->ip_off);
1235                 mhip->ip_sum = 0;
1236                 if (sw_csum & CSUM_DELAY_IP)
1237                         mhip->ip_sum = in_cksum(m, mhlen);
1238                 *mnext = m;
1239                 mnext = &m->m_nextpkt;
1240         }
1241         ipstat.ips_ofragments += nfrags;
1242
1243         /* set first marker for fragment chain */
1244         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1245         m0->m_pkthdr.csum_data = nfrags;
1246
1247         /*
1248          * Update first fragment by trimming what's been copied out
1249          * and updating header.
1250          */
1251         m_adj(m0, hlen + firstlen - ip->ip_len);
1252         m0->m_pkthdr.len = hlen + firstlen;
1253         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1254         ip->ip_off |= IP_MF;
1255         ip->ip_off = htons(ip->ip_off);
1256         ip->ip_sum = 0;
1257         if (sw_csum & CSUM_DELAY_IP)
1258                 ip->ip_sum = in_cksum(m0, hlen);
1259
1260 done:
1261         *m_frag = m0;
1262         return error;
1263 }
1264
1265 void
1266 in_delayed_cksum(struct mbuf *m)
1267 {
1268         struct ip *ip;
1269         u_short csum, offset;
1270
1271         ip = mtod(m, struct ip *);
1272         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1273         csum = in_cksum_skip(m, ip->ip_len, offset);
1274         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1275                 csum = 0xffff;
1276         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1277
1278         if (offset + sizeof(u_short) > m->m_len) {
1279                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1280                     m->m_len, offset, ip->ip_p);
1281                 /*
1282                  * XXX
1283                  * this shouldn't happen, but if it does, the
1284                  * correct behavior may be to insert the checksum
1285                  * in the existing chain instead of rearranging it.
1286                  */
1287                 m = m_pullup(m, offset + sizeof(u_short));
1288         }
1289         *(u_short *)(m->m_data + offset) = csum;
1290 }
1291
1292 /*
1293  * Insert IP options into preformed packet.
1294  * Adjust IP destination as required for IP source routing,
1295  * as indicated by a non-zero in_addr at the start of the options.
1296  *
1297  * XXX This routine assumes that the packet has no options in place.
1298  */
1299 static struct mbuf *
1300 ip_insertoptions(m, opt, phlen)
1301         struct mbuf *m;
1302         struct mbuf *opt;
1303         int *phlen;
1304 {
1305         struct ipoption *p = mtod(opt, struct ipoption *);
1306         struct mbuf *n;
1307         struct ip *ip = mtod(m, struct ip *);
1308         unsigned optlen;
1309
1310         optlen = opt->m_len - sizeof(p->ipopt_dst);
1311         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1312                 *phlen = 0;
1313                 return (m);             /* XXX should fail */
1314         }
1315         if (p->ipopt_dst.s_addr)
1316                 ip->ip_dst = p->ipopt_dst;
1317         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1318                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1319                 if (n == 0) {
1320                         *phlen = 0;
1321                         return (m);
1322                 }
1323                 n->m_pkthdr.rcvif = (struct ifnet *)0;
1324                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1325                 m->m_len -= sizeof(struct ip);
1326                 m->m_data += sizeof(struct ip);
1327                 n->m_next = m;
1328                 m = n;
1329                 m->m_len = optlen + sizeof(struct ip);
1330                 m->m_data += max_linkhdr;
1331                 (void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1332         } else {
1333                 m->m_data -= optlen;
1334                 m->m_len += optlen;
1335                 m->m_pkthdr.len += optlen;
1336                 ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1337         }
1338         ip = mtod(m, struct ip *);
1339         bcopy(p->ipopt_list, ip + 1, optlen);
1340         *phlen = sizeof(struct ip) + optlen;
1341         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1342         ip->ip_len += optlen;
1343         return (m);
1344 }
1345
1346 /*
1347  * Copy options from ip to jp,
1348  * omitting those not copied during fragmentation.
1349  */
1350 int
1351 ip_optcopy(ip, jp)
1352         struct ip *ip, *jp;
1353 {
1354         u_char *cp, *dp;
1355         int opt, optlen, cnt;
1356
1357         cp = (u_char *)(ip + 1);
1358         dp = (u_char *)(jp + 1);
1359         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1360         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1361                 opt = cp[0];
1362                 if (opt == IPOPT_EOL)
1363                         break;
1364                 if (opt == IPOPT_NOP) {
1365                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1366                         *dp++ = IPOPT_NOP;
1367                         optlen = 1;
1368                         continue;
1369                 }
1370
1371                 KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1372                     ("ip_optcopy: malformed ipv4 option"));
1373                 optlen = cp[IPOPT_OLEN];
1374                 KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1375                     ("ip_optcopy: malformed ipv4 option"));
1376
1377                 /* bogus lengths should have been caught by ip_dooptions */
1378                 if (optlen > cnt)
1379                         optlen = cnt;
1380                 if (IPOPT_COPIED(opt)) {
1381                         bcopy(cp, dp, optlen);
1382                         dp += optlen;
1383                 }
1384         }
1385         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1386                 *dp++ = IPOPT_EOL;
1387         return (optlen);
1388 }
1389
1390 /*
1391  * IP socket option processing.
1392  */
1393 int
1394 ip_ctloutput(so, sopt)
1395         struct socket *so;
1396         struct sockopt *sopt;
1397 {
1398         struct  inpcb *inp = sotoinpcb(so);
1399         int     error, optval;
1400
1401         error = optval = 0;
1402         if (sopt->sopt_level != IPPROTO_IP) {
1403                 return (EINVAL);
1404         }
1405
1406         switch (sopt->sopt_dir) {
1407         case SOPT_SET:
1408                 switch (sopt->sopt_name) {
1409                 case IP_OPTIONS:
1410 #ifdef notyet
1411                 case IP_RETOPTS:
1412 #endif
1413                 {
1414                         struct mbuf *m;
1415                         if (sopt->sopt_valsize > MLEN) {
1416                                 error = EMSGSIZE;
1417                                 break;
1418                         }
1419                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1420                         if (m == 0) {
1421                                 error = ENOBUFS;
1422                                 break;
1423                         }
1424                         m->m_len = sopt->sopt_valsize;
1425                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1426                                             m->m_len);
1427
1428                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1429                                            m));
1430                 }
1431
1432                 case IP_TOS:
1433                 case IP_TTL:
1434                 case IP_RECVOPTS:
1435                 case IP_RECVRETOPTS:
1436                 case IP_RECVDSTADDR:
1437                 case IP_RECVIF:
1438                 case IP_FAITH:
1439                         error = sooptcopyin(sopt, &optval, sizeof optval,
1440                                             sizeof optval);
1441                         if (error)
1442                                 break;
1443
1444                         switch (sopt->sopt_name) {
1445                         case IP_TOS:
1446                                 inp->inp_ip_tos = optval;
1447                                 break;
1448
1449                         case IP_TTL:
1450                                 inp->inp_ip_ttl = optval;
1451                                 break;
1452 #define OPTSET(bit) \
1453         if (optval) \
1454                 inp->inp_flags |= bit; \
1455         else \
1456                 inp->inp_flags &= ~bit;
1457
1458                         case IP_RECVOPTS:
1459                                 OPTSET(INP_RECVOPTS);
1460                                 break;
1461
1462                         case IP_RECVRETOPTS:
1463                                 OPTSET(INP_RECVRETOPTS);
1464                                 break;
1465
1466                         case IP_RECVDSTADDR:
1467                                 OPTSET(INP_RECVDSTADDR);
1468                                 break;
1469
1470                         case IP_RECVIF:
1471                                 OPTSET(INP_RECVIF);
1472                                 break;
1473
1474                         case IP_FAITH:
1475                                 OPTSET(INP_FAITH);
1476                                 break;
1477                         }
1478                         break;
1479 #undef OPTSET
1480
1481                 case IP_MULTICAST_IF:
1482                 case IP_MULTICAST_VIF:
1483                 case IP_MULTICAST_TTL:
1484                 case IP_MULTICAST_LOOP:
1485                 case IP_ADD_MEMBERSHIP:
1486                 case IP_DROP_MEMBERSHIP:
1487                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1488                         break;
1489
1490                 case IP_PORTRANGE:
1491                         error = sooptcopyin(sopt, &optval, sizeof optval,
1492                                             sizeof optval);
1493                         if (error)
1494                                 break;
1495
1496                         switch (optval) {
1497                         case IP_PORTRANGE_DEFAULT:
1498                                 inp->inp_flags &= ~(INP_LOWPORT);
1499                                 inp->inp_flags &= ~(INP_HIGHPORT);
1500                                 break;
1501
1502                         case IP_PORTRANGE_HIGH:
1503                                 inp->inp_flags &= ~(INP_LOWPORT);
1504                                 inp->inp_flags |= INP_HIGHPORT;
1505                                 break;
1506
1507                         case IP_PORTRANGE_LOW:
1508                                 inp->inp_flags &= ~(INP_HIGHPORT);
1509                                 inp->inp_flags |= INP_LOWPORT;
1510                                 break;
1511
1512                         default:
1513                                 error = EINVAL;
1514                                 break;
1515                         }
1516                         break;
1517
1518 #if defined(IPSEC) || defined(FAST_IPSEC)
1519                 case IP_IPSEC_POLICY:
1520                 {
1521                         caddr_t req;
1522                         size_t len = 0;
1523                         int priv;
1524                         struct mbuf *m;
1525                         int optname;
1526
1527                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1528                                 break;
1529                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1530                                 break;
1531                         priv = (sopt->sopt_td != NULL &&
1532                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1533                         req = mtod(m, caddr_t);
1534                         len = m->m_len;
1535                         optname = sopt->sopt_name;
1536                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1537                         m_freem(m);
1538                         break;
1539                 }
1540 #endif /*IPSEC*/
1541
1542                 default:
1543                         error = ENOPROTOOPT;
1544                         break;
1545                 }
1546                 break;
1547
1548         case SOPT_GET:
1549                 switch (sopt->sopt_name) {
1550                 case IP_OPTIONS:
1551                 case IP_RETOPTS:
1552                         if (inp->inp_options)
1553                                 error = sooptcopyout(sopt,
1554                                                      mtod(inp->inp_options,
1555                                                           char *),
1556                                                      inp->inp_options->m_len);
1557                         else
1558                                 sopt->sopt_valsize = 0;
1559                         break;
1560
1561                 case IP_TOS:
1562                 case IP_TTL:
1563                 case IP_RECVOPTS:
1564                 case IP_RECVRETOPTS:
1565                 case IP_RECVDSTADDR:
1566                 case IP_RECVIF:
1567                 case IP_PORTRANGE:
1568                 case IP_FAITH:
1569                         switch (sopt->sopt_name) {
1570
1571                         case IP_TOS:
1572                                 optval = inp->inp_ip_tos;
1573                                 break;
1574
1575                         case IP_TTL:
1576                                 optval = inp->inp_ip_ttl;
1577                                 break;
1578
1579 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1580
1581                         case IP_RECVOPTS:
1582                                 optval = OPTBIT(INP_RECVOPTS);
1583                                 break;
1584
1585                         case IP_RECVRETOPTS:
1586                                 optval = OPTBIT(INP_RECVRETOPTS);
1587                                 break;
1588
1589                         case IP_RECVDSTADDR:
1590                                 optval = OPTBIT(INP_RECVDSTADDR);
1591                                 break;
1592
1593                         case IP_RECVIF:
1594                                 optval = OPTBIT(INP_RECVIF);
1595                                 break;
1596
1597                         case IP_PORTRANGE:
1598                                 if (inp->inp_flags & INP_HIGHPORT)
1599                                         optval = IP_PORTRANGE_HIGH;
1600                                 else if (inp->inp_flags & INP_LOWPORT)
1601                                         optval = IP_PORTRANGE_LOW;
1602                                 else
1603                                         optval = 0;
1604                                 break;
1605
1606                         case IP_FAITH:
1607                                 optval = OPTBIT(INP_FAITH);
1608                                 break;
1609                         }
1610                         error = sooptcopyout(sopt, &optval, sizeof optval);
1611                         break;
1612
1613                 case IP_MULTICAST_IF:
1614                 case IP_MULTICAST_VIF:
1615                 case IP_MULTICAST_TTL:
1616                 case IP_MULTICAST_LOOP:
1617                 case IP_ADD_MEMBERSHIP:
1618                 case IP_DROP_MEMBERSHIP:
1619                         error = ip_getmoptions(sopt, inp->inp_moptions);
1620                         break;
1621
1622 #if defined(IPSEC) || defined(FAST_IPSEC)
1623                 case IP_IPSEC_POLICY:
1624                 {
1625                         struct mbuf *m = NULL;
1626                         caddr_t req = NULL;
1627                         size_t len = 0;
1628
1629                         if (m != 0) {
1630                                 req = mtod(m, caddr_t);
1631                                 len = m->m_len;
1632                         }
1633                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1634                         if (error == 0)
1635                                 error = soopt_mcopyout(sopt, m); /* XXX */
1636                         if (error == 0)
1637                                 m_freem(m);
1638                         break;
1639                 }
1640 #endif /*IPSEC*/
1641
1642                 default:
1643                         error = ENOPROTOOPT;
1644                         break;
1645                 }
1646                 break;
1647         }
1648         return (error);
1649 }
1650
1651 /*
1652  * Set up IP options in pcb for insertion in output packets.
1653  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1654  * with destination address if source routed.
1655  */
1656 static int
1657 ip_pcbopts(optname, pcbopt, m)
1658         int optname;
1659         struct mbuf **pcbopt;
1660         struct mbuf *m;
1661 {
1662         int cnt, optlen;
1663         u_char *cp;
1664         u_char opt;
1665
1666         /* turn off any old options */
1667         if (*pcbopt)
1668                 (void)m_free(*pcbopt);
1669         *pcbopt = 0;
1670         if (m == (struct mbuf *)0 || m->m_len == 0) {
1671                 /*
1672                  * Only turning off any previous options.
1673                  */
1674                 if (m)
1675                         (void)m_free(m);
1676                 return (0);
1677         }
1678
1679         if (m->m_len % sizeof(int32_t))
1680                 goto bad;
1681         /*
1682          * IP first-hop destination address will be stored before
1683          * actual options; move other options back
1684          * and clear it when none present.
1685          */
1686         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1687                 goto bad;
1688         cnt = m->m_len;
1689         m->m_len += sizeof(struct in_addr);
1690         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1691         ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1692         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1693
1694         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1695                 opt = cp[IPOPT_OPTVAL];
1696                 if (opt == IPOPT_EOL)
1697                         break;
1698                 if (opt == IPOPT_NOP)
1699                         optlen = 1;
1700                 else {
1701                         if (cnt < IPOPT_OLEN + sizeof(*cp))
1702                                 goto bad;
1703                         optlen = cp[IPOPT_OLEN];
1704                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1705                                 goto bad;
1706                 }
1707                 switch (opt) {
1708
1709                 default:
1710                         break;
1711
1712                 case IPOPT_LSRR:
1713                 case IPOPT_SSRR:
1714                         /*
1715                          * user process specifies route as:
1716                          *      ->A->B->C->D
1717                          * D must be our final destination (but we can't
1718                          * check that since we may not have connected yet).
1719                          * A is first hop destination, which doesn't appear in
1720                          * actual IP option, but is stored before the options.
1721                          */
1722                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1723                                 goto bad;
1724                         m->m_len -= sizeof(struct in_addr);
1725                         cnt -= sizeof(struct in_addr);
1726                         optlen -= sizeof(struct in_addr);
1727                         cp[IPOPT_OLEN] = optlen;
1728                         /*
1729                          * Move first hop before start of options.
1730                          */
1731                         bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1732                             sizeof(struct in_addr));
1733                         /*
1734                          * Then copy rest of options back
1735                          * to close up the deleted entry.
1736                          */
1737                         ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1738                             sizeof(struct in_addr)),
1739                             (caddr_t)&cp[IPOPT_OFFSET+1],
1740                             (unsigned)cnt + sizeof(struct in_addr));
1741                         break;
1742                 }
1743         }
1744         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1745                 goto bad;
1746         *pcbopt = m;
1747         return (0);
1748
1749 bad:
1750         (void)m_free(m);
1751         return (EINVAL);
1752 }
1753
1754 /*
1755  * XXX
1756  * The whole multicast option thing needs to be re-thought.
1757  * Several of these options are equally applicable to non-multicast
1758  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1759  * standard option (IP_TTL).
1760  */
1761
1762 /*
1763  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1764  */
1765 static struct ifnet *
1766 ip_multicast_if(a, ifindexp)
1767         struct in_addr *a;
1768         int *ifindexp;
1769 {
1770         int ifindex;
1771         struct ifnet *ifp;
1772
1773         if (ifindexp)
1774                 *ifindexp = 0;
1775         if (ntohl(a->s_addr) >> 24 == 0) {
1776                 ifindex = ntohl(a->s_addr) & 0xffffff;
1777                 if (ifindex < 0 || if_index < ifindex)
1778                         return NULL;
1779                 ifp = ifindex2ifnet[ifindex];
1780                 if (ifindexp)
1781                         *ifindexp = ifindex;
1782         } else {
1783                 INADDR_TO_IFP(*a, ifp);
1784         }
1785         return ifp;
1786 }
1787
1788 /*
1789  * Set the IP multicast options in response to user setsockopt().
1790  */
1791 static int
1792 ip_setmoptions(sopt, imop)
1793         struct sockopt *sopt;
1794         struct ip_moptions **imop;
1795 {
1796         int error = 0;
1797         int i;
1798         struct in_addr addr;
1799         struct ip_mreq mreq;
1800         struct ifnet *ifp;
1801         struct ip_moptions *imo = *imop;
1802         struct route ro;
1803         struct sockaddr_in *dst;
1804         int ifindex;
1805         int s;
1806
1807         if (imo == NULL) {
1808                 /*
1809                  * No multicast option buffer attached to the pcb;
1810                  * allocate one and initialize to default values.
1811                  */
1812                 imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1813                     M_WAITOK);
1814
1815                 if (imo == NULL)
1816                         return (ENOBUFS);
1817                 *imop = imo;
1818                 imo->imo_multicast_ifp = NULL;
1819                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1820                 imo->imo_multicast_vif = -1;
1821                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1822                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1823                 imo->imo_num_memberships = 0;
1824         }
1825
1826         switch (sopt->sopt_name) {
1827         /* store an index number for the vif you wanna use in the send */
1828         case IP_MULTICAST_VIF:
1829                 if (legal_vif_num == 0) {
1830                         error = EOPNOTSUPP;
1831                         break;
1832                 }
1833                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1834                 if (error)
1835                         break;
1836                 if (!legal_vif_num(i) && (i != -1)) {
1837                         error = EINVAL;
1838                         break;
1839                 }
1840                 imo->imo_multicast_vif = i;
1841                 break;
1842
1843         case IP_MULTICAST_IF:
1844                 /*
1845                  * Select the interface for outgoing multicast packets.
1846                  */
1847                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1848                 if (error)
1849                         break;
1850                 /*
1851                  * INADDR_ANY is used to remove a previous selection.
1852                  * When no interface is selected, a default one is
1853                  * chosen every time a multicast packet is sent.
1854                  */
1855                 if (addr.s_addr == INADDR_ANY) {
1856                         imo->imo_multicast_ifp = NULL;
1857                         break;
1858                 }
1859                 /*
1860                  * The selected interface is identified by its local
1861                  * IP address.  Find the interface and confirm that
1862                  * it supports multicasting.
1863                  */
1864                 s = splimp();
1865                 ifp = ip_multicast_if(&addr, &ifindex);
1866                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1867                         splx(s);
1868                         error = EADDRNOTAVAIL;
1869                         break;
1870                 }
1871                 imo->imo_multicast_ifp = ifp;
1872                 if (ifindex)
1873                         imo->imo_multicast_addr = addr;
1874                 else
1875                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1876                 splx(s);
1877                 break;
1878
1879         case IP_MULTICAST_TTL:
1880                 /*
1881                  * Set the IP time-to-live for outgoing multicast packets.
1882                  * The original multicast API required a char argument,
1883                  * which is inconsistent with the rest of the socket API.
1884                  * We allow either a char or an int.
1885                  */
1886                 if (sopt->sopt_valsize == 1) {
1887                         u_char ttl;
1888                         error = sooptcopyin(sopt, &ttl, 1, 1);
1889                         if (error)
1890                                 break;
1891                         imo->imo_multicast_ttl = ttl;
1892                 } else {
1893                         u_int ttl;
1894                         error = sooptcopyin(sopt, &ttl, sizeof ttl,
1895                                             sizeof ttl);
1896                         if (error)
1897                                 break;
1898                         if (ttl > 255)
1899                                 error = EINVAL;
1900                         else
1901                                 imo->imo_multicast_ttl = ttl;
1902                 }
1903                 break;
1904
1905         case IP_MULTICAST_LOOP:
1906                 /*
1907                  * Set the loopback flag for outgoing multicast packets.
1908                  * Must be zero or one.  The original multicast API required a
1909                  * char argument, which is inconsistent with the rest
1910                  * of the socket API.  We allow either a char or an int.
1911                  */
1912                 if (sopt->sopt_valsize == 1) {
1913                         u_char loop;
1914                         error = sooptcopyin(sopt, &loop, 1, 1);
1915                         if (error)
1916                                 break;
1917                         imo->imo_multicast_loop = !!loop;
1918                 } else {
1919                         u_int loop;
1920                         error = sooptcopyin(sopt, &loop, sizeof loop,
1921                                             sizeof loop);
1922                         if (error)
1923                                 break;
1924                         imo->imo_multicast_loop = !!loop;
1925                 }
1926                 break;
1927
1928         case IP_ADD_MEMBERSHIP:
1929                 /*
1930                  * Add a multicast group membership.
1931                  * Group must be a valid IP multicast address.
1932                  */
1933                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1934                 if (error)
1935                         break;
1936
1937                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1938                         error = EINVAL;
1939                         break;
1940                 }
1941                 s = splimp();
1942                 /*
1943                  * If no interface address was provided, use the interface of
1944                  * the route to the given multicast address.
1945                  */
1946                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1947                         bzero((caddr_t)&ro, sizeof(ro));
1948                         dst = (struct sockaddr_in *)&ro.ro_dst;
1949                         dst->sin_len = sizeof(*dst);
1950                         dst->sin_family = AF_INET;
1951                         dst->sin_addr = mreq.imr_multiaddr;
1952                         rtalloc(&ro);
1953                         if (ro.ro_rt == NULL) {
1954                                 error = EADDRNOTAVAIL;
1955                                 splx(s);
1956                                 break;
1957                         }
1958                         ifp = ro.ro_rt->rt_ifp;
1959                         rtfree(ro.ro_rt);
1960                 }
1961                 else {
1962                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1963                 }
1964
1965                 /*
1966                  * See if we found an interface, and confirm that it
1967                  * supports multicast.
1968                  */
1969                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1970                         error = EADDRNOTAVAIL;
1971                         splx(s);
1972                         break;
1973                 }
1974                 /*
1975                  * See if the membership already exists or if all the
1976                  * membership slots are full.
1977                  */
1978                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1979                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1980                             imo->imo_membership[i]->inm_addr.s_addr
1981                                                 == mreq.imr_multiaddr.s_addr)
1982                                 break;
1983                 }
1984                 if (i < imo->imo_num_memberships) {
1985                         error = EADDRINUSE;
1986                         splx(s);
1987                         break;
1988                 }
1989                 if (i == IP_MAX_MEMBERSHIPS) {
1990                         error = ETOOMANYREFS;
1991                         splx(s);
1992                         break;
1993                 }
1994                 /*
1995                  * Everything looks good; add a new record to the multicast
1996                  * address list for the given interface.
1997                  */
1998                 if ((imo->imo_membership[i] =
1999                     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2000                         error = ENOBUFS;
2001                         splx(s);
2002                         break;
2003                 }
2004                 ++imo->imo_num_memberships;
2005                 splx(s);
2006                 break;
2007
2008         case IP_DROP_MEMBERSHIP:
2009                 /*
2010                  * Drop a multicast group membership.
2011                  * Group must be a valid IP multicast address.
2012                  */
2013                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2014                 if (error)
2015                         break;
2016
2017                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2018                         error = EINVAL;
2019                         break;
2020                 }
2021
2022                 s = splimp();
2023                 /*
2024                  * If an interface address was specified, get a pointer
2025                  * to its ifnet structure.
2026                  */
2027                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2028                         ifp = NULL;
2029                 else {
2030                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2031                         if (ifp == NULL) {
2032                                 error = EADDRNOTAVAIL;
2033                                 splx(s);
2034                                 break;
2035                         }
2036                 }
2037                 /*
2038                  * Find the membership in the membership array.
2039                  */
2040                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2041                         if ((ifp == NULL ||
2042                              imo->imo_membership[i]->inm_ifp == ifp) &&
2043                              imo->imo_membership[i]->inm_addr.s_addr ==
2044                              mreq.imr_multiaddr.s_addr)
2045                                 break;
2046                 }
2047                 if (i == imo->imo_num_memberships) {
2048                         error = EADDRNOTAVAIL;
2049                         splx(s);
2050                         break;
2051                 }
2052                 /*
2053                  * Give up the multicast address record to which the
2054                  * membership points.
2055                  */
2056                 in_delmulti(imo->imo_membership[i]);
2057                 /*
2058                  * Remove the gap in the membership array.
2059                  */
2060                 for (++i; i < imo->imo_num_memberships; ++i)
2061                         imo->imo_membership[i-1] = imo->imo_membership[i];
2062                 --imo->imo_num_memberships;
2063                 splx(s);
2064                 break;
2065
2066         default:
2067                 error = EOPNOTSUPP;
2068                 break;
2069         }
2070
2071         /*
2072          * If all options have default values, no need to keep the mbuf.
2073          */
2074         if (imo->imo_multicast_ifp == NULL &&
2075             imo->imo_multicast_vif == -1 &&
2076             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2077             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2078             imo->imo_num_memberships == 0) {
2079                 free(*imop, M_IPMOPTS);
2080                 *imop = NULL;
2081         }
2082
2083         return (error);
2084 }
2085
2086 /*
2087  * Return the IP multicast options in response to user getsockopt().
2088  */
2089 static int
2090 ip_getmoptions(sopt, imo)
2091         struct sockopt *sopt;
2092         struct ip_moptions *imo;
2093 {
2094         struct in_addr addr;
2095         struct in_ifaddr *ia;
2096         int error, optval;
2097         u_char coptval;
2098
2099         error = 0;
2100         switch (sopt->sopt_name) {
2101         case IP_MULTICAST_VIF:
2102                 if (imo != NULL)
2103                         optval = imo->imo_multicast_vif;
2104                 else
2105                         optval = -1;
2106                 error = sooptcopyout(sopt, &optval, sizeof optval);
2107                 break;
2108
2109         case IP_MULTICAST_IF:
2110                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2111                         addr.s_addr = INADDR_ANY;
2112                 else if (imo->imo_multicast_addr.s_addr) {
2113                         /* return the value user has set */
2114                         addr = imo->imo_multicast_addr;
2115                 } else {
2116                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2117                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2118                                 : IA_SIN(ia)->sin_addr.s_addr;
2119                 }
2120                 error = sooptcopyout(sopt, &addr, sizeof addr);
2121                 break;
2122
2123         case IP_MULTICAST_TTL:
2124                 if (imo == 0)
2125                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2126                 else
2127                         optval = coptval = imo->imo_multicast_ttl;
2128                 if (sopt->sopt_valsize == 1)
2129                         error = sooptcopyout(sopt, &coptval, 1);
2130                 else
2131                         error = sooptcopyout(sopt, &optval, sizeof optval);
2132                 break;
2133
2134         case IP_MULTICAST_LOOP:
2135                 if (imo == 0)
2136                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2137                 else
2138                         optval = coptval = imo->imo_multicast_loop;
2139                 if (sopt->sopt_valsize == 1)
2140                         error = sooptcopyout(sopt, &coptval, 1);
2141                 else
2142                         error = sooptcopyout(sopt, &optval, sizeof optval);
2143                 break;
2144
2145         default:
2146                 error = ENOPROTOOPT;
2147                 break;
2148         }
2149         return (error);
2150 }
2151
2152 /*
2153  * Discard the IP multicast options.
2154  */
2155 void
2156 ip_freemoptions(imo)
2157         struct ip_moptions *imo;
2158 {
2159         int i;
2160
2161         if (imo != NULL) {
2162                 for (i = 0; i < imo->imo_num_memberships; ++i)
2163                         in_delmulti(imo->imo_membership[i]);
2164                 free(imo, M_IPMOPTS);
2165         }
2166 }
2167
2168 /*
2169  * Routine called from ip_output() to loop back a copy of an IP multicast
2170  * packet to the input queue of a specified interface.  Note that this
2171  * calls the output routine of the loopback "driver", but with an interface
2172  * pointer that might NOT be a loopback interface -- evil, but easier than
2173  * replicating that code here.
2174  */
2175 static void
2176 ip_mloopback(ifp, m, dst, hlen)
2177         struct ifnet *ifp;
2178         struct mbuf *m;
2179         struct sockaddr_in *dst;
2180         int hlen;
2181 {
2182         struct ip *ip;
2183         struct mbuf *copym;
2184
2185         copym = m_copy(m, 0, M_COPYALL);
2186         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2187                 copym = m_pullup(copym, hlen);
2188         if (copym != NULL) {
2189                 /*
2190                  * if the checksum hasn't been computed, mark it as valid
2191                  */
2192                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2193                         in_delayed_cksum(copym);
2194                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2195                         copym->m_pkthdr.csum_flags |=
2196                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2197                         copym->m_pkthdr.csum_data = 0xffff;
2198                 }
2199                 /*
2200                  * We don't bother to fragment if the IP length is greater
2201                  * than the interface's MTU.  Can this possibly matter?
2202                  */
2203                 ip = mtod(copym, struct ip *);
2204                 ip->ip_len = htons(ip->ip_len);
2205                 ip->ip_off = htons(ip->ip_off);
2206                 ip->ip_sum = 0;
2207                 if (ip->ip_vhl == IP_VHL_BORING) {
2208                         ip->ip_sum = in_cksum_hdr(ip);
2209                 } else {
2210                         ip->ip_sum = in_cksum(copym, hlen);
2211                 }
2212                 /*
2213                  * NB:
2214                  * It's not clear whether there are any lingering
2215                  * reentrancy problems in other areas which might
2216                  * be exposed by using ip_input directly (in
2217                  * particular, everything which modifies the packet
2218                  * in-place).  Yet another option is using the
2219                  * protosw directly to deliver the looped back
2220                  * packet.  For the moment, we'll err on the side
2221                  * of safety by using if_simloop().
2222                  */
2223 #if 1 /* XXX */
2224                 if (dst->sin_family != AF_INET) {
2225                         printf("ip_mloopback: bad address family %d\n",
2226                                                 dst->sin_family);
2227                         dst->sin_family = AF_INET;
2228                 }
2229 #endif
2230
2231 #ifdef notdef
2232                 copym->m_pkthdr.rcvif = ifp;
2233                 ip_input(copym);
2234 #else
2235                 if_simloop(ifp, copym, dst->sin_family, 0);
2236 #endif
2237         }
2238 }