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 1.10 2004/02/14 21:12:39 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_pfil_hooks.h"
  46 #include "opt_random_ip_id.h"
  47 #include "opt_mbuf_stress_test.h"
  48
  49 #include <sys/param.h>
  50 #include <sys/systm.h>
  51 #include <sys/kernel.h>
  52 #include <sys/malloc.h>
  53 #include <sys/mbuf.h>
  54 #include <sys/protosw.h>
  55 #include <sys/socket.h>
  56 #include <sys/socketvar.h>
  57 #include <sys/proc.h>
  58 #include <sys/sysctl.h>
  59 #include <sys/in_cksum.h>
  60
  61 #include <net/if.h>
  62 #ifdef PFIL_HOOKS
  63 #include <net/pfil.h>
  64 #endif
  65 #include <net/route.h>
  66
  67 #include <netinet/in.h>
  68 #include <netinet/in_systm.h>
  69 #include <netinet/ip.h>
  70 #include <netinet/in_pcb.h>
  71 #include <netinet/in_var.h>
  72 #include <netinet/ip_var.h>
  73
  74 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  75
  76 #ifdef IPSEC
  77 #include <netinet6/ipsec.h>
  78 #include <netproto/key/key.h>
  79 #ifdef IPSEC_DEBUG
  80 #include <netproto/key/key_debug.h>
  81 #else
  82 #define KEYDEBUG(lev,arg)
  83 #endif
  84 #endif /*IPSEC*/
  85
  86 #ifdef FAST_IPSEC
  87 #include <netipsec/ipsec.h>
  88 #include <netipsec/xform.h>
  89 #include <netipsec/key.h>
  90 #endif /*FAST_IPSEC*/
  91
  92 #include <net/ipfw/ip_fw.h>
  93 #include <net/dummynet/ip_dummynet.h>
  94
  95 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
  96                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  97                                   (ntohl(a.s_addr)>>16)&0xFF,\
  98                                   (ntohl(a.s_addr)>>8)&0xFF,\
  99                                   (ntohl(a.s_addr))&0xFF, y);
 100
 101 u_short ip_id;
 102
 103 #ifdef MBUF_STRESS_TEST
 104 int mbuf_frag_size = 0;
 105 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 106         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 107 #endif
 108
 109 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 110 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 111 static void     ip_mloopback
 112         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 113 static int      ip_getmoptions
 114         (struct sockopt *, struct ip_moptions *);
 115 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 116 static int      ip_setmoptions
 117         (struct sockopt *, struct ip_moptions **);
 118
 119 int     ip_optcopy(struct ip *, struct ip *);
 120 extern int (*fr_checkp) (struct ip *, int, struct ifnet *, int, struct mbuf **);
 121
 122
 123 extern  struct protosw inetsw[];
 124
 125 /*
 126  * IP output.  The packet in mbuf chain m contains a skeletal IP
 127  * header (with len, off, ttl, proto, tos, src, dst).
 128  * The mbuf chain containing the packet will be freed.
 129  * The mbuf opt, if present, will not be freed.
 130  */
 131 int
 132 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 133         int flags, struct ip_moptions *imo, struct inpcb *inp)
 134 {
 135         struct ip *ip;
 136         struct ifnet *ifp = NULL;       /* keep compiler happy */
 137         struct mbuf *m;
 138         int hlen = sizeof (struct ip);
 139         int len, off, error = 0;
 140         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 141         struct in_ifaddr *ia = NULL;
 142         int isbroadcast, sw_csum;
 143         struct in_addr pkt_dst;
 144 #ifdef IPSEC
 145         struct route iproute;
 146         struct secpolicy *sp = NULL;
 147         struct socket *so = inp ? inp->inp_socket : NULL;
 148 #endif
 149 #ifdef FAST_IPSEC
 150         struct route iproute;
 151         struct m_tag *mtag;
 152         struct secpolicy *sp = NULL;
 153         struct tdb_ident *tdbi;
 154         int s;
 155 #endif /* FAST_IPSEC */
 156         struct ip_fw_args args;
 157         int src_was_INADDR_ANY = 0;     /* as the name says... */
 158
 159         args.eh = NULL;
 160         args.rule = NULL;
 161         args.next_hop = NULL;
 162         args.divert_rule = 0;                   /* divert cookie */
 163
 164         /* Grab info from MT_TAG mbufs prepended to the chain. */
 165         for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
 166                 switch(m0->_m_tag_id) {
 167                 default:
 168                         printf("ip_output: unrecognised MT_TAG tag %d\n",
 169                             m0->_m_tag_id);
 170                         break;
 171
 172                 case PACKET_TAG_DUMMYNET:
 173                         /*
 174                          * the packet was already tagged, so part of the
 175                          * processing was already done, and we need to go down.
 176                          * Get parameters from the header.
 177                          */
 178                         args.rule = ((struct dn_pkt *)m0)->rule;
 179                         opt = NULL ;
 180                         ro = & ( ((struct dn_pkt *)m0)->ro ) ;
 181                         imo = NULL ;
 182                         dst = ((struct dn_pkt *)m0)->dn_dst ;
 183                         ifp = ((struct dn_pkt *)m0)->ifp ;
 184                         flags = ((struct dn_pkt *)m0)->flags ;
 185                         break;
 186
 187                 case PACKET_TAG_DIVERT:
 188                         args.divert_rule = (int)m0->m_data & 0xffff;
 189                         break;
 190
 191                 case PACKET_TAG_IPFORWARD:
 192                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 193                         break;
 194                 }
 195         }
 196         m = m0;
 197
 198         KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
 199 #ifndef FAST_IPSEC
 200         KASSERT(ro != NULL, ("ip_output: no route, proto %d",
 201             mtod(m, struct ip *)->ip_p));
 202 #endif
 203
 204         if (args.rule != NULL) {        /* dummynet already saw us */
 205                 ip = mtod(m, struct ip *);
 206                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 207                 if (ro->ro_rt)
 208                         ia = ifatoia(ro->ro_rt->rt_ifa);
 209                 goto sendit;
 210         }
 211
 212         if (opt) {
 213                 len = 0;
 214                 m = ip_insertoptions(m, opt, &len);
 215                 if (len != 0)
 216                         hlen = len;
 217         }
 218         ip = mtod(m, struct ip *);
 219         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 220
 221         /*
 222          * Fill in IP header.
 223          */
 224         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 225                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 226                 ip->ip_off &= IP_DF;
 227 #ifdef RANDOM_IP_ID
 228                 ip->ip_id = ip_randomid();
 229 #else
 230                 ip->ip_id = htons(ip_id++);
 231 #endif
 232                 ipstat.ips_localout++;
 233         } else {
 234                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 235         }
 236
 237 #ifdef FAST_IPSEC
 238         if (ro == NULL) {
 239                 ro = &iproute;
 240                 bzero(ro, sizeof (*ro));
 241         }
 242 #endif /* FAST_IPSEC */
 243         dst = (struct sockaddr_in *)&ro->ro_dst;
 244         /*
 245          * If there is a cached route,
 246          * check that it is to the same destination
 247          * and is still up.  If not, free it and try again.
 248          * The address family should also be checked in case of sharing the
 249          * cache with IPv6.
 250          */
 251         if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
 252                           dst->sin_family != AF_INET ||
 253                           dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 254                 RTFREE(ro->ro_rt);
 255                 ro->ro_rt = (struct rtentry *)0;
 256         }
 257         if (ro->ro_rt == 0) {
 258                 bzero(dst, sizeof(*dst));
 259                 dst->sin_family = AF_INET;
 260                 dst->sin_len = sizeof(*dst);
 261                 dst->sin_addr = pkt_dst;
 262         }
 263         /*
 264          * If routing to interface only,
 265          * short circuit routing lookup.
 266          */
 267         if (flags & IP_ROUTETOIF) {
 268                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
 269                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
 270                         ipstat.ips_noroute++;
 271                         error = ENETUNREACH;
 272                         goto bad;
 273                 }
 274                 ifp = ia->ia_ifp;
 275                 ip->ip_ttl = 1;
 276                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 277         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 278             imo != NULL && imo->imo_multicast_ifp != NULL) {
 279                 /*
 280                  * Bypass the normal routing lookup for multicast
 281                  * packets if the interface is specified.
 282                  */
 283                 ifp = imo->imo_multicast_ifp;
 284                 IFP_TO_IA(ifp, ia);
 285                 isbroadcast = 0;        /* fool gcc */
 286         } else {
 287                 /*
 288                  * If this is the case, we probably don't want to allocate
 289                  * a protocol-cloned route since we didn't get one from the
 290                  * ULP.  This lets TCP do its thing, while not burdening
 291                  * forwarding or ICMP with the overhead of cloning a route.
 292                  * Of course, we still want to do any cloning requested by
 293                  * the link layer, as this is probably required in all cases
 294                  * for correct operation (as it is for ARP).
 295                  */
 296                 if (ro->ro_rt == 0)
 297                         rtalloc_ign(ro, RTF_PRCLONING);
 298                 if (ro->ro_rt == 0) {
 299                         ipstat.ips_noroute++;
 300                         error = EHOSTUNREACH;
 301                         goto bad;
 302                 }
 303                 ia = ifatoia(ro->ro_rt->rt_ifa);
 304                 ifp = ro->ro_rt->rt_ifp;
 305                 ro->ro_rt->rt_use++;
 306                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 307                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 308                 if (ro->ro_rt->rt_flags & RTF_HOST)
 309                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 310                 else
 311                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 312         }
 313         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 314                 struct in_multi *inm;
 315
 316                 m->m_flags |= M_MCAST;
 317                 /*
 318                  * IP destination address is multicast.  Make sure "dst"
 319                  * still points to the address in "ro".  (It may have been
 320                  * changed to point to a gateway address, above.)
 321                  */
 322                 dst = (struct sockaddr_in *)&ro->ro_dst;
 323                 /*
 324                  * See if the caller provided any multicast options
 325                  */
 326                 if (imo != NULL) {
 327                         ip->ip_ttl = imo->imo_multicast_ttl;
 328                         if (imo->imo_multicast_vif != -1)
 329                                 ip->ip_src.s_addr =
 330                                     ip_mcast_src ?
 331                                     ip_mcast_src(imo->imo_multicast_vif) :
 332                                     INADDR_ANY;
 333                 } else
 334                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 335                 /*
 336                  * Confirm that the outgoing interface supports multicast.
 337                  */
 338                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 339                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 340                                 ipstat.ips_noroute++;
 341                                 error = ENETUNREACH;
 342                                 goto bad;
 343                         }
 344                 }
 345                 /*
 346                  * If source address not specified yet, use address
 347                  * of outgoing interface.
 348                  */
 349                 if (ip->ip_src.s_addr == INADDR_ANY) {
 350                         /* Interface may have no addresses. */
 351                         if (ia != NULL)
 352                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 353                 }
 354
 355                 if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 356                         /*
 357                          * XXX
 358                          * delayed checksums are not currently
 359                          * compatible with IP multicast routing
 360                          */
 361                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 362                                 in_delayed_cksum(m);
 363                                 m->m_pkthdr.csum_flags &=
 364                                         ~CSUM_DELAY_DATA;
 365                         }
 366                 }
 367                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 368                 if (inm != NULL &&
 369                    (imo == NULL || imo->imo_multicast_loop)) {
 370                         /*
 371                          * If we belong to the destination multicast group
 372                          * on the outgoing interface, and the caller did not
 373                          * forbid loopback, loop back a copy.
 374                          */
 375                         ip_mloopback(ifp, m, dst, hlen);
 376                 }
 377                 else {
 378                         /*
 379                          * If we are acting as a multicast router, perform
 380                          * multicast forwarding as if the packet had just
 381                          * arrived on the interface to which we are about
 382                          * to send.  The multicast forwarding function
 383                          * recursively calls this function, using the
 384                          * IP_FORWARDING flag to prevent infinite recursion.
 385                          *
 386                          * Multicasts that are looped back by ip_mloopback(),
 387                          * above, will be forwarded by the ip_input() routine,
 388                          * if necessary.
 389                          */
 390                         if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 391                                 /*
 392                                  * If rsvp daemon is not running, do not
 393                                  * set ip_moptions. This ensures that the packet
 394                                  * is multicast and not just sent down one link
 395                                  * as prescribed by rsvpd.
 396                                  */
 397                                 if (!rsvp_on)
 398                                         imo = NULL;
 399                                 if (ip_mforward &&
 400                                     ip_mforward(ip, ifp, m, imo) != 0) {
 401                                         m_freem(m);
 402                                         goto done;
 403                                 }
 404                         }
 405                 }
 406
 407                 /*
 408                  * Multicasts with a time-to-live of zero may be looped-
 409                  * back, above, but must not be transmitted on a network.
 410                  * Also, multicasts addressed to the loopback interface
 411                  * are not sent -- the above call to ip_mloopback() will
 412                  * loop back a copy if this host actually belongs to the
 413                  * destination group on the loopback interface.
 414                  */
 415                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 416                         m_freem(m);
 417                         goto done;
 418                 }
 419
 420                 goto sendit;
 421         }
 422 #ifndef notdef
 423         /*
 424          * If the source address is not specified yet, use the address
 425          * of the outoing interface. In case, keep note we did that, so
 426          * if the the firewall changes the next-hop causing the output
 427          * interface to change, we can fix that.
 428          */
 429         if (ip->ip_src.s_addr == INADDR_ANY) {
 430                 /* Interface may have no addresses. */
 431                 if (ia != NULL) {
 432                         ip->ip_src = IA_SIN(ia)->sin_addr;
 433                         src_was_INADDR_ANY = 1;
 434                 }
 435         }
 436 #endif /* notdef */
 437         /*
 438          * Verify that we have any chance at all of being able to queue
 439          *      the packet or packet fragments
 440          */
 441         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 442                 ifp->if_snd.ifq_maxlen) {
 443                         error = ENOBUFS;
 444                         ipstat.ips_odropped++;
 445                         goto bad;
 446         }
 447
 448         /*
 449          * Look for broadcast address and
 450          * verify user is allowed to send
 451          * such a packet.
 452          */
 453         if (isbroadcast) {
 454                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 455                         error = EADDRNOTAVAIL;
 456                         goto bad;
 457                 }
 458                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 459                         error = EACCES;
 460                         goto bad;
 461                 }
 462                 /* don't allow broadcast messages to be fragmented */
 463                 if (ip->ip_len > ifp->if_mtu) {
 464                         error = EMSGSIZE;
 465                         goto bad;
 466                 }
 467                 m->m_flags |= M_BCAST;
 468         } else {
 469                 m->m_flags &= ~M_BCAST;
 470         }
 471
 472 sendit:
 473 #ifdef IPSEC
 474         /* get SP for this packet */
 475         if (so == NULL)
 476                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 477         else
 478                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 479
 480         if (sp == NULL) {
 481                 ipsecstat.out_inval++;
 482                 goto bad;
 483         }
 484
 485         error = 0;
 486
 487         /* check policy */
 488         switch (sp->policy) {
 489         case IPSEC_POLICY_DISCARD:
 490                 /*
 491                  * This packet is just discarded.
 492                  */
 493                 ipsecstat.out_polvio++;
 494                 goto bad;
 495
 496         case IPSEC_POLICY_BYPASS:
 497         case IPSEC_POLICY_NONE:
 498                 /* no need to do IPsec. */
 499                 goto skip_ipsec;
 500
 501         case IPSEC_POLICY_IPSEC:
 502                 if (sp->req == NULL) {
 503                         /* acquire a policy */
 504                         error = key_spdacquire(sp);
 505                         goto bad;
 506                 }
 507                 break;
 508
 509         case IPSEC_POLICY_ENTRUST:
 510         default:
 511                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
 512         }
 513     {
 514         struct ipsec_output_state state;
 515         bzero(&state, sizeof(state));
 516         state.m = m;
 517         if (flags & IP_ROUTETOIF) {
 518                 state.ro = &iproute;
 519                 bzero(&iproute, sizeof(iproute));
 520         } else
 521                 state.ro = ro;
 522         state.dst = (struct sockaddr *)dst;
 523
 524         ip->ip_sum = 0;
 525
 526         /*
 527          * XXX
 528          * delayed checksums are not currently compatible with IPsec
 529          */
 530         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 531                 in_delayed_cksum(m);
 532                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 533         }
 534
 535         ip->ip_len = htons(ip->ip_len);
 536         ip->ip_off = htons(ip->ip_off);
 537
 538         error = ipsec4_output(&state, sp, flags);
 539
 540         m = state.m;
 541         if (flags & IP_ROUTETOIF) {
 542                 /*
 543                  * if we have tunnel mode SA, we may need to ignore
 544                  * IP_ROUTETOIF.
 545                  */
 546                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 547                         flags &= ~IP_ROUTETOIF;
 548                         ro = state.ro;
 549                 }
 550         } else
 551                 ro = state.ro;
 552         dst = (struct sockaddr_in *)state.dst;
 553         if (error) {
 554                 /* mbuf is already reclaimed in ipsec4_output. */
 555                 m0 = NULL;
 556                 switch (error) {
 557                 case EHOSTUNREACH:
 558                 case ENETUNREACH:
 559                 case EMSGSIZE:
 560                 case ENOBUFS:
 561                 case ENOMEM:
 562                         break;
 563                 default:
 564                         printf("ip4_output (ipsec): error code %d\n", error);
 565                         /*fall through*/
 566                 case ENOENT:
 567                         /* don't show these error codes to the user */
 568                         error = 0;
 569                         break;
 570                 }
 571                 goto bad;
 572         }
 573     }
 574
 575         /* be sure to update variables that are affected by ipsec4_output() */
 576         ip = mtod(m, struct ip *);
 577 #ifdef _IP_VHL
 578         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 579 #else
 580         hlen = ip->ip_hl << 2;
 581 #endif
 582         if (ro->ro_rt == NULL) {
 583                 if ((flags & IP_ROUTETOIF) == 0) {
 584                         printf("ip_output: "
 585                                 "can't update route after IPsec processing\n");
 586                         error = EHOSTUNREACH;   /*XXX*/
 587                         goto bad;
 588                 }
 589         } else {
 590                 ia = ifatoia(ro->ro_rt->rt_ifa);
 591                 ifp = ro->ro_rt->rt_ifp;
 592         }
 593
 594         /* make it flipped, again. */
 595         ip->ip_len = ntohs(ip->ip_len);
 596         ip->ip_off = ntohs(ip->ip_off);
 597 skip_ipsec:
 598 #endif /*IPSEC*/
 599 #ifdef FAST_IPSEC
 600         /*
 601          * Check the security policy (SP) for the packet and, if
 602          * required, do IPsec-related processing.  There are two
 603          * cases here; the first time a packet is sent through
 604          * it will be untagged and handled by ipsec4_checkpolicy.
 605          * If the packet is resubmitted to ip_output (e.g. after
 606          * AH, ESP, etc. processing), there will be a tag to bypass
 607          * the lookup and related policy checking.
 608          */
 609         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 610         s = splnet();
 611         if (mtag != NULL) {
 612                 tdbi = (struct tdb_ident *)(mtag + 1);
 613                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 614                 if (sp == NULL)
 615                         error = -EINVAL;        /* force silent drop */
 616                 m_tag_delete(m, mtag);
 617         } else {
 618                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 619                                         &error, inp);
 620         }
 621         /*
 622          * There are four return cases:
 623          *    sp != NULL                    apply IPsec policy
 624          *    sp == NULL, error == 0        no IPsec handling needed
 625          *    sp == NULL, error == -EINVAL  discard packet w/o error
 626          *    sp == NULL, error != 0        discard packet, report error
 627          */
 628         if (sp != NULL) {
 629                 /* Loop detection, check if ipsec processing already done */
 630                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 631                 for (mtag = m_tag_first(m); mtag != NULL;
 632                      mtag = m_tag_next(m, mtag)) {
 633                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 634                                 continue;
 635                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 636                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 637                                 continue;
 638                         /*
 639                          * Check if policy has an SA associated with it.
 640                          * This can happen when an SP has yet to acquire
 641                          * an SA; e.g. on first reference.  If it occurs,
 642                          * then we let ipsec4_process_packet do its thing.
 643                          */
 644                         if (sp->req->sav == NULL)
 645                                 break;
 646                         tdbi = (struct tdb_ident *)(mtag + 1);
 647                         if (tdbi->spi == sp->req->sav->spi &&
 648                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 649                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 650                                  sizeof (union sockaddr_union)) == 0) {
 651                                 /*
 652                                  * No IPsec processing is needed, free
 653                                  * reference to SP.
 654                                  *
 655                                  * NB: null pointer to avoid free at
 656                                  *     done: below.
 657                                  */
 658                                 KEY_FREESP(&sp), sp = NULL;
 659                                 splx(s);
 660                                 goto spd_done;
 661                         }
 662                 }
 663
 664                 /*
 665                  * Do delayed checksums now because we send before
 666                  * this is done in the normal processing path.
 667                  */
 668                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 669                         in_delayed_cksum(m);
 670                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 671                 }
 672
 673                 ip->ip_len = htons(ip->ip_len);
 674                 ip->ip_off = htons(ip->ip_off);
 675
 676                 /* NB: callee frees mbuf */
 677                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 678                 /*
 679                  * Preserve KAME behaviour: ENOENT can be returned
 680                  * when an SA acquire is in progress.  Don't propagate
 681                  * this to user-level; it confuses applications.
 682                  *
 683                  * XXX this will go away when the SADB is redone.
 684                  */
 685                 if (error == ENOENT)
 686                         error = 0;
 687                 splx(s);
 688                 goto done;
 689         } else {
 690                 splx(s);
 691
 692                 if (error != 0) {
 693                         /*
 694                          * Hack: -EINVAL is used to signal that a packet
 695                          * should be silently discarded.  This is typically
 696                          * because we asked key management for an SA and
 697                          * it was delayed (e.g. kicked up to IKE).
 698                          */
 699                         if (error == -EINVAL)
 700                                 error = 0;
 701                         goto bad;
 702                 } else {
 703                         /* No IPsec processing for this packet. */
 704                 }
 705 #ifdef notyet
 706                 /*
 707                  * If deferred crypto processing is needed, check that
 708                  * the interface supports it.
 709                  */
 710                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 711                 if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
 712                         /* notify IPsec to do its own crypto */
 713                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
 714                         error = EHOSTUNREACH;
 715                         goto bad;
 716                 }
 717 #endif
 718         }
 719 spd_done:
 720 #endif /* FAST_IPSEC */
 721         /*
 722          * IpHack's section.
 723          * - Xlate: translate packet's addr/port (NAT).
 724          * - Firewall: deny/allow/etc.
 725          * - Wrap: fake packet's addr/port <unimpl.>
 726          * - Encapsulate: put it in another IP and send out. <unimp.>
 727          */
 728 #ifdef PFIL_HOOKS
 729         /*
 730          * Run through list of hooks for output packets.
 731          */
 732         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 733         if (error != 0 || m == NULL)
 734                 goto done;
 735         ip = mtod(m, struct ip *);
 736 #endif /* PFIL_HOOKS */
 737
 738         /*
 739          * Check with the firewall...
 740          * but not if we are already being fwd'd from a firewall.
 741          */
 742         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 743                 struct sockaddr_in *old = dst;
 744
 745                 args.m = m;
 746                 args.next_hop = dst;
 747                 args.oif = ifp;
 748                 off = ip_fw_chk_ptr(&args);
 749                 m = args.m;
 750                 dst = args.next_hop;
 751
 752                 /*
 753                  * On return we must do the following:
 754                  * m == NULL    -> drop the pkt (old interface, deprecated)
 755                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 756                  * 1<=off<= 0xffff              -> DIVERT
 757                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 758                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 759                  * dst != old                   -> IPFIREWALL_FORWARD
 760                  * off==0, dst==old             -> accept
 761                  * If some of the above modules are not compiled in, then
 762                  * we should't have to check the corresponding condition
 763                  * (because the ipfw control socket should not accept
 764                  * unsupported rules), but better play safe and drop
 765                  * packets in case of doubt.
 766                  */
 767                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 768                         if (m)
 769                                 m_freem(m);
 770                         error = EACCES;
 771                         goto done;
 772                 }
 773                 ip = mtod(m, struct ip *);
 774                 if (off == 0 && dst == old)             /* common case */
 775                         goto pass;
 776                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
 777                         /*
 778                          * pass the pkt to dummynet. Need to include
 779                          * pipe number, m, ifp, ro, dst because these are
 780                          * not recomputed in the next pass.
 781                          * All other parameters have been already used and
 782                          * so they are not needed anymore.
 783                          * XXX note: if the ifp or ro entry are deleted
 784                          * while a pkt is in dummynet, we are in trouble!
 785                          */
 786                         args.ro = ro;
 787                         args.dst = dst;
 788                         args.flags = flags;
 789
 790                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
 791                                 &args);
 792                         goto done;
 793                 }
 794 #ifdef IPDIVERT
 795                 if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
 796                         struct mbuf *clone = NULL;
 797
 798                         /* Clone packet if we're doing a 'tee' */
 799                         if ((off & IP_FW_PORT_TEE_FLAG) != 0)
 800                                 clone = m_dup(m, M_DONTWAIT);
 801
 802                         /*
 803                          * XXX
 804                          * delayed checksums are not currently compatible
 805                          * with divert sockets.
 806                          */
 807                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 808                                 in_delayed_cksum(m);
 809                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 810                         }
 811
 812                         /* Restore packet header fields to original values */
 813                         ip->ip_len = htons(ip->ip_len);
 814                         ip->ip_off = htons(ip->ip_off);
 815
 816                         /* Deliver packet to divert input routine */
 817                         divert_packet(m, 0, off & 0xffff, args.divert_rule);
 818
 819                         /* If 'tee', continue with original packet */
 820                         if (clone != NULL) {
 821                                 m = clone;
 822                                 ip = mtod(m, struct ip *);
 823                                 goto pass;
 824                         }
 825                         goto done;
 826                 }
 827 #endif
 828
 829                 /* IPFIREWALL_FORWARD */
 830                 /*
 831                  * Check dst to make sure it is directly reachable on the
 832                  * interface we previously thought it was.
 833                  * If it isn't (which may be likely in some situations) we have
 834                  * to re-route it (ie, find a route for the next-hop and the
 835                  * associated interface) and set them here. This is nested
 836                  * forwarding which in most cases is undesirable, except where
 837                  * such control is nigh impossible. So we do it here.
 838                  * And I'm babbling.
 839                  */
 840                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 841 #if 0
 842                         /*
 843                          * XXX To improve readability, this block should be
 844                          * changed into a function call as below:
 845                          */
 846                         error = ip_ipforward(&m, &dst, &ifp);
 847                         if (error)
 848                                 goto bad;
 849                         if (m == NULL) /* ip_input consumed the mbuf */
 850                                 goto done;
 851 #else
 852                         struct in_ifaddr *ia;
 853
 854                         /*
 855                          * XXX sro_fwd below is static, and a pointer
 856                          * to it gets passed to routines downstream.
 857                          * This could have surprisingly bad results in
 858                          * practice, because its content is overwritten
 859                          * by subsequent packets.
 860                          */
 861                         /* There must be a better way to do this next line... */
 862                         static struct route sro_fwd;
 863                         struct route *ro_fwd = &sro_fwd;
 864
 865 #if 0
 866                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 867                             dst->sin_addr, "\n");
 868 #endif
 869
 870                         /*
 871                          * We need to figure out if we have been forwarded
 872                          * to a local socket. If so, then we should somehow
 873                          * "loop back" to ip_input, and get directed to the
 874                          * PCB as if we had received this packet. This is
 875                          * because it may be dificult to identify the packets
 876                          * you want to forward until they are being output
 877                          * and have selected an interface. (e.g. locally
 878                          * initiated packets) If we used the loopback inteface,
 879                          * we would not be able to control what happens
 880                          * as the packet runs through ip_input() as
 881                          * it is done through a ISR.
 882                          */
 883                         LIST_FOREACH(ia,
 884                             INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 885                                 /*
 886                                  * If the addr to forward to is one
 887                                  * of ours, we pretend to
 888                                  * be the destination for this packet.
 889                                  */
 890                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 891                                                  dst->sin_addr.s_addr)
 892                                         break;
 893                         }
 894                         if (ia) {       /* tell ip_input "dont filter" */
 895                                 struct m_hdr tag;
 896
 897                                 tag.mh_type = MT_TAG;
 898                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 899                                 tag.mh_data = (caddr_t)args.next_hop;
 900                                 tag.mh_next = m;
 901
 902                                 if (m->m_pkthdr.rcvif == NULL)
 903                                         m->m_pkthdr.rcvif = ifunit("lo0");
 904                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 905                                         m->m_pkthdr.csum_flags |=
 906                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 907                                         m0->m_pkthdr.csum_data = 0xffff;
 908                                 }
 909                                 m->m_pkthdr.csum_flags |=
 910                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 911                                 ip->ip_len = htons(ip->ip_len);
 912                                 ip->ip_off = htons(ip->ip_off);
 913                                 ip_input((struct mbuf *)&tag);
 914                                 goto done;
 915                         }
 916                         /* Some of the logic for this was
 917                          * nicked from above.
 918                          *
 919                          * This rewrites the cached route in a local PCB.
 920                          * Is this what we want to do?
 921                          */
 922                         bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
 923
 924                         ro_fwd->ro_rt = 0;
 925                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 926
 927                         if (ro_fwd->ro_rt == 0) {
 928                                 ipstat.ips_noroute++;
 929                                 error = EHOSTUNREACH;
 930                                 goto bad;
 931                         }
 932
 933                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 934                         ifp = ro_fwd->ro_rt->rt_ifp;
 935                         ro_fwd->ro_rt->rt_use++;
 936                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 937                                 dst = (struct sockaddr_in *)
 938                                         ro_fwd->ro_rt->rt_gateway;
 939                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 940                                 isbroadcast =
 941                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 942                         else
 943                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 944                         if (ro->ro_rt)
 945                                 RTFREE(ro->ro_rt);
 946                         ro->ro_rt = ro_fwd->ro_rt;
 947                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 948
 949 #endif  /* ... block to be put into a function */
 950                         /*
 951                          * If we added a default src ip earlier,
 952                          * which would have been gotten from the-then
 953                          * interface, do it again, from the new one.
 954                          */
 955                         if (src_was_INADDR_ANY)
 956                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 957                         goto pass ;
 958                 }
 959
 960                 /*
 961                  * if we get here, none of the above matches, and
 962                  * we have to drop the pkt
 963                  */
 964                 m_freem(m);
 965                 error = EACCES; /* not sure this is the right error msg */
 966                 goto done;
 967         }
 968
 969 pass:
 970         /* 127/8 must not appear on wire - RFC1122. */
 971         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 972             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 973                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 974                         ipstat.ips_badaddr++;
 975                         error = EADDRNOTAVAIL;
 976                         goto bad;
 977                 }
 978         }
 979
 980         m->m_pkthdr.csum_flags |= CSUM_IP;
 981         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 982         if (sw_csum & CSUM_DELAY_DATA) {
 983                 in_delayed_cksum(m);
 984                 sw_csum &= ~CSUM_DELAY_DATA;
 985         }
 986         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 987
 988         /*
 989          * If small enough for interface, or the interface will take
 990          * care of the fragmentation for us, can just send directly.
 991          */
 992         if (ip->ip_len <= ifp->if_mtu || ifp->if_hwassist & CSUM_FRAGMENT) {
 993                 ip->ip_len = htons(ip->ip_len);
 994                 ip->ip_off = htons(ip->ip_off);
 995                 ip->ip_sum = 0;
 996                 if (sw_csum & CSUM_DELAY_IP) {
 997                         if (ip->ip_vhl == IP_VHL_BORING) {
 998                                 ip->ip_sum = in_cksum_hdr(ip);
 999                         } else {
1000                                 ip->ip_sum = in_cksum(m, hlen);
1001                         }
1002                 }
1003
1004                 /* Record statistics for this interface address. */
1005                 if (!(flags & IP_FORWARDING) && ia) {
1006                         ia->ia_ifa.if_opackets++;
1007                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1008                 }
1009
1010 #ifdef IPSEC
1011                 /* clean ipsec history once it goes out of the node */
1012                 ipsec_delaux(m);
1013 #endif
1014
1015 #ifdef MBUF_STRESS_TEST
1016                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1017                         struct mbuf *m1, *m2;
1018                         int length, tmp;
1019
1020                         tmp = length = m->m_pkthdr.len;
1021
1022                         while ((length -= mbuf_frag_size) >= 1) {
1023                                 m1 = m_split(m, length, M_DONTWAIT);
1024                                 if (m1 == NULL)
1025                                         break;
1026                                 m1->m_flags &= ~M_PKTHDR;
1027                                 m2 = m;
1028                                 while (m2->m_next != NULL)
1029                                         m2 = m2->m_next;
1030                                 m2->m_next = m1;
1031                         }
1032                         m->m_pkthdr.len = tmp;
1033                 }
1034 #endif
1035                 error = (*ifp->if_output)(ifp, m,
1036                                 (struct sockaddr *)dst, ro->ro_rt);
1037                 goto done;
1038         }
1039
1040         if (ip->ip_off & IP_DF) {
1041                 error = EMSGSIZE;
1042                 /*
1043                  * This case can happen if the user changed the MTU
1044                  * of an interface after enabling IP on it.  Because
1045                  * most netifs don't keep track of routes pointing to
1046                  * them, there is no way for one to update all its
1047                  * routes when the MTU is changed.
1048                  */
1049                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1050                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1051                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1052                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1053                 }
1054                 ipstat.ips_cantfrag++;
1055                 goto bad;
1056         }
1057
1058         /*
1059          * Too large for interface; fragment if possible. If successful,
1060          * on return, m will point to a list of packets to be sent.
1061          */
1062         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1063         if (error)
1064                 goto bad;
1065         for (; m; m = m0) {
1066                 m0 = m->m_nextpkt;
1067                 m->m_nextpkt = 0;
1068 #ifdef IPSEC
1069                 /* clean ipsec history once it goes out of the node */
1070                 ipsec_delaux(m);
1071 #endif
1072                 if (error == 0) {
1073                         /* Record statistics for this interface address. */
1074                         if (ia != NULL) {
1075                                 ia->ia_ifa.if_opackets++;
1076                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1077                         }
1078
1079                         error = (*ifp->if_output)(ifp, m,
1080                             (struct sockaddr *)dst, ro->ro_rt);
1081                 } else
1082                         m_freem(m);
1083         }
1084
1085         if (error == 0)
1086                 ipstat.ips_fragmented++;
1087
1088 done:
1089 #ifdef IPSEC
1090         if (ro == &iproute && ro->ro_rt) {
1091                 RTFREE(ro->ro_rt);
1092                 ro->ro_rt = NULL;
1093         }
1094         if (sp != NULL) {
1095                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1096                         printf("DP ip_output call free SP:%p\n", sp));
1097                 key_freesp(sp);
1098         }
1099 #endif
1100 #ifdef FAST_IPSEC
1101         if (ro == &iproute && ro->ro_rt) {
1102                 RTFREE(ro->ro_rt);
1103                 ro->ro_rt = NULL;
1104         }
1105         if (sp != NULL)
1106                 KEY_FREESP(&sp);
1107 #endif
1108         return (error);
1109 bad:
1110         m_freem(m);
1111         goto done;
1112 }
1113
1114 /*
1115  * Create a chain of fragments which fit the given mtu. m_frag points to the
1116  * mbuf to be fragmented; on return it points to the chain with the fragments.
1117  * Return 0 if no error. If error, m_frag may contain a partially built
1118  * chain of fragments that should be freed by the caller.
1119  *
1120  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1121  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1122  */
1123 int
1124 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1125             u_long if_hwassist_flags, int sw_csum)
1126 {
1127         int error = 0;
1128         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1129         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1130         int off;
1131         struct mbuf *m0 = *m_frag;      /* the original packet          */
1132         int firstlen;
1133         struct mbuf **mnext;
1134         int nfrags;
1135
1136         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1137                 ipstat.ips_cantfrag++;
1138                 return EMSGSIZE;
1139         }
1140
1141         /*
1142          * Must be able to put at least 8 bytes per fragment.
1143          */
1144         if (len < 8)
1145                 return EMSGSIZE;
1146
1147         /*
1148          * If the interface will not calculate checksums on
1149          * fragmented packets, then do it here.
1150          */
1151         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1152             (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1153                 in_delayed_cksum(m0);
1154                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1155         }
1156
1157         if (len > PAGE_SIZE) {
1158                 /*
1159                  * Fragment large datagrams such that each segment
1160                  * contains a multiple of PAGE_SIZE amount of data,
1161                  * plus headers. This enables a receiver to perform
1162                  * page-flipping zero-copy optimizations.
1163                  *
1164                  * XXX When does this help given that sender and receiver
1165                  * could have different page sizes, and also mtu could
1166                  * be less than the receiver's page size ?
1167                  */
1168                 int newlen;
1169                 struct mbuf *m;
1170
1171                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1172                         off += m->m_len;
1173
1174                 /*
1175                  * firstlen (off - hlen) must be aligned on an
1176                  * 8-byte boundary
1177                  */
1178                 if (off < hlen)
1179                         goto smart_frag_failure;
1180                 off = ((off - hlen) & ~7) + hlen;
1181                 newlen = (~PAGE_MASK) & mtu;
1182                 if ((newlen + sizeof (struct ip)) > mtu) {
1183                         /* we failed, go back the default */
1184 smart_frag_failure:
1185                         newlen = len;
1186                         off = hlen + len;
1187                 }
1188                 len = newlen;
1189
1190         } else {
1191                 off = hlen + len;
1192         }
1193
1194         firstlen = off - hlen;
1195         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1196
1197         /*
1198          * Loop through length of segment after first fragment,
1199          * make new header and copy data of each part and link onto chain.
1200          * Here, m0 is the original packet, m is the fragment being created.
1201          * The fragments are linked off the m_nextpkt of the original
1202          * packet, which after processing serves as the first fragment.
1203          */
1204         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1205                 struct ip *mhip;        /* ip header on the fragment */
1206                 struct mbuf *m;
1207                 int mhlen = sizeof (struct ip);
1208
1209                 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1210                 if (m == 0) {
1211                         error = ENOBUFS;
1212                         ipstat.ips_odropped++;
1213                         goto done;
1214                 }
1215                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1216                 /*
1217                  * In the first mbuf, leave room for the link header, then
1218                  * copy the original IP header including options. The payload
1219                  * goes into an additional mbuf chain returned by m_copy().
1220                  */
1221                 m->m_data += max_linkhdr;
1222                 mhip = mtod(m, struct ip *);
1223                 *mhip = *ip;
1224                 if (hlen > sizeof (struct ip)) {
1225                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1226                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1227                 }
1228                 m->m_len = mhlen;
1229                 /* XXX do we need to add ip->ip_off below ? */
1230                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1231                 if (off + len >= ip->ip_len) {  /* last fragment */
1232                         len = ip->ip_len - off;
1233                         m->m_flags |= M_LASTFRAG;
1234                 } else
1235                         mhip->ip_off |= IP_MF;
1236                 mhip->ip_len = htons((u_short)(len + mhlen));
1237                 m->m_next = m_copy(m0, off, len);
1238                 if (m->m_next == 0) {           /* copy failed */
1239                         m_free(m);
1240                         error = ENOBUFS;        /* ??? */
1241                         ipstat.ips_odropped++;
1242                         goto done;
1243                 }
1244                 m->m_pkthdr.len = mhlen + len;
1245                 m->m_pkthdr.rcvif = (struct ifnet *)0;
1246                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1247                 mhip->ip_off = htons(mhip->ip_off);
1248                 mhip->ip_sum = 0;
1249                 if (sw_csum & CSUM_DELAY_IP)
1250                         mhip->ip_sum = in_cksum(m, mhlen);
1251                 *mnext = m;
1252                 mnext = &m->m_nextpkt;
1253         }
1254         ipstat.ips_ofragments += nfrags;
1255
1256         /* set first marker for fragment chain */
1257         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1258         m0->m_pkthdr.csum_data = nfrags;
1259
1260         /*
1261          * Update first fragment by trimming what's been copied out
1262          * and updating header.
1263          */
1264         m_adj(m0, hlen + firstlen - ip->ip_len);
1265         m0->m_pkthdr.len = hlen + firstlen;
1266         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1267         ip->ip_off |= IP_MF;
1268         ip->ip_off = htons(ip->ip_off);
1269         ip->ip_sum = 0;
1270         if (sw_csum & CSUM_DELAY_IP)
1271                 ip->ip_sum = in_cksum(m0, hlen);
1272
1273 done:
1274         *m_frag = m0;
1275         return error;
1276 }
1277
1278 void
1279 in_delayed_cksum(struct mbuf *m)
1280 {
1281         struct ip *ip;
1282         u_short csum, offset;
1283
1284         ip = mtod(m, struct ip *);
1285         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1286         csum = in_cksum_skip(m, ip->ip_len, offset);
1287         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1288                 csum = 0xffff;
1289         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1290
1291         if (offset + sizeof(u_short) > m->m_len) {
1292                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1293                     m->m_len, offset, ip->ip_p);
1294                 /*
1295                  * XXX
1296                  * this shouldn't happen, but if it does, the
1297                  * correct behavior may be to insert the checksum
1298                  * in the existing chain instead of rearranging it.
1299                  */
1300                 m = m_pullup(m, offset + sizeof(u_short));
1301         }
1302         *(u_short *)(m->m_data + offset) = csum;
1303 }
1304
1305 /*
1306  * Insert IP options into preformed packet.
1307  * Adjust IP destination as required for IP source routing,
1308  * as indicated by a non-zero in_addr at the start of the options.
1309  *
1310  * XXX This routine assumes that the packet has no options in place.
1311  */
1312 static struct mbuf *
1313 ip_insertoptions(m, opt, phlen)
1314         struct mbuf *m;
1315         struct mbuf *opt;
1316         int *phlen;
1317 {
1318         struct ipoption *p = mtod(opt, struct ipoption *);
1319         struct mbuf *n;
1320         struct ip *ip = mtod(m, struct ip *);
1321         unsigned optlen;
1322
1323         optlen = opt->m_len - sizeof(p->ipopt_dst);
1324         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1325                 *phlen = 0;
1326                 return (m);             /* XXX should fail */
1327         }
1328         if (p->ipopt_dst.s_addr)
1329                 ip->ip_dst = p->ipopt_dst;
1330         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1331                 MGETHDR(n, M_DONTWAIT, MT_HEADER);
1332                 if (n == 0) {
1333                         *phlen = 0;
1334                         return (m);
1335                 }
1336                 n->m_pkthdr.rcvif = (struct ifnet *)0;
1337                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1338                 m->m_len -= sizeof(struct ip);
1339                 m->m_data += sizeof(struct ip);
1340                 n->m_next = m;
1341                 m = n;
1342                 m->m_len = optlen + sizeof(struct ip);
1343                 m->m_data += max_linkhdr;
1344                 (void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1345         } else {
1346                 m->m_data -= optlen;
1347                 m->m_len += optlen;
1348                 m->m_pkthdr.len += optlen;
1349                 ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1350         }
1351         ip = mtod(m, struct ip *);
1352         bcopy(p->ipopt_list, ip + 1, optlen);
1353         *phlen = sizeof(struct ip) + optlen;
1354         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1355         ip->ip_len += optlen;
1356         return (m);
1357 }
1358
1359 /*
1360  * Copy options from ip to jp,
1361  * omitting those not copied during fragmentation.
1362  */
1363 int
1364 ip_optcopy(ip, jp)
1365         struct ip *ip, *jp;
1366 {
1367         u_char *cp, *dp;
1368         int opt, optlen, cnt;
1369
1370         cp = (u_char *)(ip + 1);
1371         dp = (u_char *)(jp + 1);
1372         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1373         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1374                 opt = cp[0];
1375                 if (opt == IPOPT_EOL)
1376                         break;
1377                 if (opt == IPOPT_NOP) {
1378                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1379                         *dp++ = IPOPT_NOP;
1380                         optlen = 1;
1381                         continue;
1382                 }
1383
1384                 KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1385                     ("ip_optcopy: malformed ipv4 option"));
1386                 optlen = cp[IPOPT_OLEN];
1387                 KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1388                     ("ip_optcopy: malformed ipv4 option"));
1389
1390                 /* bogus lengths should have been caught by ip_dooptions */
1391                 if (optlen > cnt)
1392                         optlen = cnt;
1393                 if (IPOPT_COPIED(opt)) {
1394                         bcopy(cp, dp, optlen);
1395                         dp += optlen;
1396                 }
1397         }
1398         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1399                 *dp++ = IPOPT_EOL;
1400         return (optlen);
1401 }
1402
1403 /*
1404  * IP socket option processing.
1405  */
1406 int
1407 ip_ctloutput(so, sopt)
1408         struct socket *so;
1409         struct sockopt *sopt;
1410 {
1411         struct  inpcb *inp = sotoinpcb(so);
1412         int     error, optval;
1413
1414         error = optval = 0;
1415         if (sopt->sopt_level != IPPROTO_IP) {
1416                 return (EINVAL);
1417         }
1418
1419         switch (sopt->sopt_dir) {
1420         case SOPT_SET:
1421                 switch (sopt->sopt_name) {
1422                 case IP_OPTIONS:
1423 #ifdef notyet
1424                 case IP_RETOPTS:
1425 #endif
1426                 {
1427                         struct mbuf *m;
1428                         if (sopt->sopt_valsize > MLEN) {
1429                                 error = EMSGSIZE;
1430                                 break;
1431                         }
1432                         MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_HEADER);
1433                         if (m == 0) {
1434                                 error = ENOBUFS;
1435                                 break;
1436                         }
1437                         m->m_len = sopt->sopt_valsize;
1438                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1439                                             m->m_len);
1440
1441                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1442                                            m));
1443                 }
1444
1445                 case IP_TOS:
1446                 case IP_TTL:
1447                 case IP_RECVOPTS:
1448                 case IP_RECVRETOPTS:
1449                 case IP_RECVDSTADDR:
1450                 case IP_RECVIF:
1451                 case IP_FAITH:
1452                         error = sooptcopyin(sopt, &optval, sizeof optval,
1453                                             sizeof optval);
1454                         if (error)
1455                                 break;
1456
1457                         switch (sopt->sopt_name) {
1458                         case IP_TOS:
1459                                 inp->inp_ip_tos = optval;
1460                                 break;
1461
1462                         case IP_TTL:
1463                                 inp->inp_ip_ttl = optval;
1464                                 break;
1465 #define OPTSET(bit) \
1466         if (optval) \
1467                 inp->inp_flags |= bit; \
1468         else \
1469                 inp->inp_flags &= ~bit;
1470
1471                         case IP_RECVOPTS:
1472                                 OPTSET(INP_RECVOPTS);
1473                                 break;
1474
1475                         case IP_RECVRETOPTS:
1476                                 OPTSET(INP_RECVRETOPTS);
1477                                 break;
1478
1479                         case IP_RECVDSTADDR:
1480                                 OPTSET(INP_RECVDSTADDR);
1481                                 break;
1482
1483                         case IP_RECVIF:
1484                                 OPTSET(INP_RECVIF);
1485                                 break;
1486
1487                         case IP_FAITH:
1488                                 OPTSET(INP_FAITH);
1489                                 break;
1490                         }
1491                         break;
1492 #undef OPTSET
1493
1494                 case IP_MULTICAST_IF:
1495                 case IP_MULTICAST_VIF:
1496                 case IP_MULTICAST_TTL:
1497                 case IP_MULTICAST_LOOP:
1498                 case IP_ADD_MEMBERSHIP:
1499                 case IP_DROP_MEMBERSHIP:
1500                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1501                         break;
1502
1503                 case IP_PORTRANGE:
1504                         error = sooptcopyin(sopt, &optval, sizeof optval,
1505                                             sizeof optval);
1506                         if (error)
1507                                 break;
1508
1509                         switch (optval) {
1510                         case IP_PORTRANGE_DEFAULT:
1511                                 inp->inp_flags &= ~(INP_LOWPORT);
1512                                 inp->inp_flags &= ~(INP_HIGHPORT);
1513                                 break;
1514
1515                         case IP_PORTRANGE_HIGH:
1516                                 inp->inp_flags &= ~(INP_LOWPORT);
1517                                 inp->inp_flags |= INP_HIGHPORT;
1518                                 break;
1519
1520                         case IP_PORTRANGE_LOW:
1521                                 inp->inp_flags &= ~(INP_HIGHPORT);
1522                                 inp->inp_flags |= INP_LOWPORT;
1523                                 break;
1524
1525                         default:
1526                                 error = EINVAL;
1527                                 break;
1528                         }
1529                         break;
1530
1531 #if defined(IPSEC) || defined(FAST_IPSEC)
1532                 case IP_IPSEC_POLICY:
1533                 {
1534                         caddr_t req;
1535                         size_t len = 0;
1536                         int priv;
1537                         struct mbuf *m;
1538                         int optname;
1539
1540                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1541                                 break;
1542                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1543                                 break;
1544                         priv = (sopt->sopt_td != NULL &&
1545                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1546                         req = mtod(m, caddr_t);
1547                         len = m->m_len;
1548                         optname = sopt->sopt_name;
1549                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1550                         m_freem(m);
1551                         break;
1552                 }
1553 #endif /*IPSEC*/
1554
1555                 default:
1556                         error = ENOPROTOOPT;
1557                         break;
1558                 }
1559                 break;
1560
1561         case SOPT_GET:
1562                 switch (sopt->sopt_name) {
1563                 case IP_OPTIONS:
1564                 case IP_RETOPTS:
1565                         if (inp->inp_options)
1566                                 error = sooptcopyout(sopt,
1567                                                      mtod(inp->inp_options,
1568                                                           char *),
1569                                                      inp->inp_options->m_len);
1570                         else
1571                                 sopt->sopt_valsize = 0;
1572                         break;
1573
1574                 case IP_TOS:
1575                 case IP_TTL:
1576                 case IP_RECVOPTS:
1577                 case IP_RECVRETOPTS:
1578                 case IP_RECVDSTADDR:
1579                 case IP_RECVIF:
1580                 case IP_PORTRANGE:
1581                 case IP_FAITH:
1582                         switch (sopt->sopt_name) {
1583
1584                         case IP_TOS:
1585                                 optval = inp->inp_ip_tos;
1586                                 break;
1587
1588                         case IP_TTL:
1589                                 optval = inp->inp_ip_ttl;
1590                                 break;
1591
1592 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1593
1594                         case IP_RECVOPTS:
1595                                 optval = OPTBIT(INP_RECVOPTS);
1596                                 break;
1597
1598                         case IP_RECVRETOPTS:
1599                                 optval = OPTBIT(INP_RECVRETOPTS);
1600                                 break;
1601
1602                         case IP_RECVDSTADDR:
1603                                 optval = OPTBIT(INP_RECVDSTADDR);
1604                                 break;
1605
1606                         case IP_RECVIF:
1607                                 optval = OPTBIT(INP_RECVIF);
1608                                 break;
1609
1610                         case IP_PORTRANGE:
1611                                 if (inp->inp_flags & INP_HIGHPORT)
1612                                         optval = IP_PORTRANGE_HIGH;
1613                                 else if (inp->inp_flags & INP_LOWPORT)
1614                                         optval = IP_PORTRANGE_LOW;
1615                                 else
1616                                         optval = 0;
1617                                 break;
1618
1619                         case IP_FAITH:
1620                                 optval = OPTBIT(INP_FAITH);
1621                                 break;
1622                         }
1623                         error = sooptcopyout(sopt, &optval, sizeof optval);
1624                         break;
1625
1626                 case IP_MULTICAST_IF:
1627                 case IP_MULTICAST_VIF:
1628                 case IP_MULTICAST_TTL:
1629                 case IP_MULTICAST_LOOP:
1630                 case IP_ADD_MEMBERSHIP:
1631                 case IP_DROP_MEMBERSHIP:
1632                         error = ip_getmoptions(sopt, inp->inp_moptions);
1633                         break;
1634
1635 #if defined(IPSEC) || defined(FAST_IPSEC)
1636                 case IP_IPSEC_POLICY:
1637                 {
1638                         struct mbuf *m = NULL;
1639                         caddr_t req = NULL;
1640                         size_t len = 0;
1641
1642                         if (m != 0) {
1643                                 req = mtod(m, caddr_t);
1644                                 len = m->m_len;
1645                         }
1646                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1647                         if (error == 0)
1648                                 error = soopt_mcopyout(sopt, m); /* XXX */
1649                         if (error == 0)
1650                                 m_freem(m);
1651                         break;
1652                 }
1653 #endif /*IPSEC*/
1654
1655                 default:
1656                         error = ENOPROTOOPT;
1657                         break;
1658                 }
1659                 break;
1660         }
1661         return (error);
1662 }
1663
1664 /*
1665  * Set up IP options in pcb for insertion in output packets.
1666  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1667  * with destination address if source routed.
1668  */
1669 static int
1670 ip_pcbopts(optname, pcbopt, m)
1671         int optname;
1672         struct mbuf **pcbopt;
1673         struct mbuf *m;
1674 {
1675         int cnt, optlen;
1676         u_char *cp;
1677         u_char opt;
1678
1679         /* turn off any old options */
1680         if (*pcbopt)
1681                 (void)m_free(*pcbopt);
1682         *pcbopt = 0;
1683         if (m == (struct mbuf *)0 || m->m_len == 0) {
1684                 /*
1685                  * Only turning off any previous options.
1686                  */
1687                 if (m)
1688                         (void)m_free(m);
1689                 return (0);
1690         }
1691
1692         if (m->m_len % sizeof(int32_t))
1693                 goto bad;
1694         /*
1695          * IP first-hop destination address will be stored before
1696          * actual options; move other options back
1697          * and clear it when none present.
1698          */
1699         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1700                 goto bad;
1701         cnt = m->m_len;
1702         m->m_len += sizeof(struct in_addr);
1703         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1704         ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1705         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1706
1707         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1708                 opt = cp[IPOPT_OPTVAL];
1709                 if (opt == IPOPT_EOL)
1710                         break;
1711                 if (opt == IPOPT_NOP)
1712                         optlen = 1;
1713                 else {
1714                         if (cnt < IPOPT_OLEN + sizeof(*cp))
1715                                 goto bad;
1716                         optlen = cp[IPOPT_OLEN];
1717                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1718                                 goto bad;
1719                 }
1720                 switch (opt) {
1721
1722                 default:
1723                         break;
1724
1725                 case IPOPT_LSRR:
1726                 case IPOPT_SSRR:
1727                         /*
1728                          * user process specifies route as:
1729                          *      ->A->B->C->D
1730                          * D must be our final destination (but we can't
1731                          * check that since we may not have connected yet).
1732                          * A is first hop destination, which doesn't appear in
1733                          * actual IP option, but is stored before the options.
1734                          */
1735                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1736                                 goto bad;
1737                         m->m_len -= sizeof(struct in_addr);
1738                         cnt -= sizeof(struct in_addr);
1739                         optlen -= sizeof(struct in_addr);
1740                         cp[IPOPT_OLEN] = optlen;
1741                         /*
1742                          * Move first hop before start of options.
1743                          */
1744                         bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1745                             sizeof(struct in_addr));
1746                         /*
1747                          * Then copy rest of options back
1748                          * to close up the deleted entry.
1749                          */
1750                         ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1751                             sizeof(struct in_addr)),
1752                             (caddr_t)&cp[IPOPT_OFFSET+1],
1753                             (unsigned)cnt + sizeof(struct in_addr));
1754                         break;
1755                 }
1756         }
1757         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1758                 goto bad;
1759         *pcbopt = m;
1760         return (0);
1761
1762 bad:
1763         (void)m_free(m);
1764         return (EINVAL);
1765 }
1766
1767 /*
1768  * XXX
1769  * The whole multicast option thing needs to be re-thought.
1770  * Several of these options are equally applicable to non-multicast
1771  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1772  * standard option (IP_TTL).
1773  */
1774
1775 /*
1776  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1777  */
1778 static struct ifnet *
1779 ip_multicast_if(a, ifindexp)
1780         struct in_addr *a;
1781         int *ifindexp;
1782 {
1783         int ifindex;
1784         struct ifnet *ifp;
1785
1786         if (ifindexp)
1787                 *ifindexp = 0;
1788         if (ntohl(a->s_addr) >> 24 == 0) {
1789                 ifindex = ntohl(a->s_addr) & 0xffffff;
1790                 if (ifindex < 0 || if_index < ifindex)
1791                         return NULL;
1792                 ifp = ifindex2ifnet[ifindex];
1793                 if (ifindexp)
1794                         *ifindexp = ifindex;
1795         } else {
1796                 INADDR_TO_IFP(*a, ifp);
1797         }
1798         return ifp;
1799 }
1800
1801 /*
1802  * Set the IP multicast options in response to user setsockopt().
1803  */
1804 static int
1805 ip_setmoptions(sopt, imop)
1806         struct sockopt *sopt;
1807         struct ip_moptions **imop;
1808 {
1809         int error = 0;
1810         int i;
1811         struct in_addr addr;
1812         struct ip_mreq mreq;
1813         struct ifnet *ifp;
1814         struct ip_moptions *imo = *imop;
1815         struct route ro;
1816         struct sockaddr_in *dst;
1817         int ifindex;
1818         int s;
1819
1820         if (imo == NULL) {
1821                 /*
1822                  * No multicast option buffer attached to the pcb;
1823                  * allocate one and initialize to default values.
1824                  */
1825                 imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1826                     M_WAITOK);
1827
1828                 if (imo == NULL)
1829                         return (ENOBUFS);
1830                 *imop = imo;
1831                 imo->imo_multicast_ifp = NULL;
1832                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1833                 imo->imo_multicast_vif = -1;
1834                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1835                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1836                 imo->imo_num_memberships = 0;
1837         }
1838
1839         switch (sopt->sopt_name) {
1840         /* store an index number for the vif you wanna use in the send */
1841         case IP_MULTICAST_VIF:
1842                 if (legal_vif_num == 0) {
1843                         error = EOPNOTSUPP;
1844                         break;
1845                 }
1846                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1847                 if (error)
1848                         break;
1849                 if (!legal_vif_num(i) && (i != -1)) {
1850                         error = EINVAL;
1851                         break;
1852                 }
1853                 imo->imo_multicast_vif = i;
1854                 break;
1855
1856         case IP_MULTICAST_IF:
1857                 /*
1858                  * Select the interface for outgoing multicast packets.
1859                  */
1860                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1861                 if (error)
1862                         break;
1863                 /*
1864                  * INADDR_ANY is used to remove a previous selection.
1865                  * When no interface is selected, a default one is
1866                  * chosen every time a multicast packet is sent.
1867                  */
1868                 if (addr.s_addr == INADDR_ANY) {
1869                         imo->imo_multicast_ifp = NULL;
1870                         break;
1871                 }
1872                 /*
1873                  * The selected interface is identified by its local
1874                  * IP address.  Find the interface and confirm that
1875                  * it supports multicasting.
1876                  */
1877                 s = splimp();
1878                 ifp = ip_multicast_if(&addr, &ifindex);
1879                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1880                         splx(s);
1881                         error = EADDRNOTAVAIL;
1882                         break;
1883                 }
1884                 imo->imo_multicast_ifp = ifp;
1885                 if (ifindex)
1886                         imo->imo_multicast_addr = addr;
1887                 else
1888                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1889                 splx(s);
1890                 break;
1891
1892         case IP_MULTICAST_TTL:
1893                 /*
1894                  * Set the IP time-to-live for outgoing multicast packets.
1895                  * The original multicast API required a char argument,
1896                  * which is inconsistent with the rest of the socket API.
1897                  * We allow either a char or an int.
1898                  */
1899                 if (sopt->sopt_valsize == 1) {
1900                         u_char ttl;
1901                         error = sooptcopyin(sopt, &ttl, 1, 1);
1902                         if (error)
1903                                 break;
1904                         imo->imo_multicast_ttl = ttl;
1905                 } else {
1906                         u_int ttl;
1907                         error = sooptcopyin(sopt, &ttl, sizeof ttl,
1908                                             sizeof ttl);
1909                         if (error)
1910                                 break;
1911                         if (ttl > 255)
1912                                 error = EINVAL;
1913                         else
1914                                 imo->imo_multicast_ttl = ttl;
1915                 }
1916                 break;
1917
1918         case IP_MULTICAST_LOOP:
1919                 /*
1920                  * Set the loopback flag for outgoing multicast packets.
1921                  * Must be zero or one.  The original multicast API required a
1922                  * char argument, which is inconsistent with the rest
1923                  * of the socket API.  We allow either a char or an int.
1924                  */
1925                 if (sopt->sopt_valsize == 1) {
1926                         u_char loop;
1927                         error = sooptcopyin(sopt, &loop, 1, 1);
1928                         if (error)
1929                                 break;
1930                         imo->imo_multicast_loop = !!loop;
1931                 } else {
1932                         u_int loop;
1933                         error = sooptcopyin(sopt, &loop, sizeof loop,
1934                                             sizeof loop);
1935                         if (error)
1936                                 break;
1937                         imo->imo_multicast_loop = !!loop;
1938                 }
1939                 break;
1940
1941         case IP_ADD_MEMBERSHIP:
1942                 /*
1943                  * Add a multicast group membership.
1944                  * Group must be a valid IP multicast address.
1945                  */
1946                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1947                 if (error)
1948                         break;
1949
1950                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1951                         error = EINVAL;
1952                         break;
1953                 }
1954                 s = splimp();
1955                 /*
1956                  * If no interface address was provided, use the interface of
1957                  * the route to the given multicast address.
1958                  */
1959                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1960                         bzero((caddr_t)&ro, sizeof(ro));
1961                         dst = (struct sockaddr_in *)&ro.ro_dst;
1962                         dst->sin_len = sizeof(*dst);
1963                         dst->sin_family = AF_INET;
1964                         dst->sin_addr = mreq.imr_multiaddr;
1965                         rtalloc(&ro);
1966                         if (ro.ro_rt == NULL) {
1967                                 error = EADDRNOTAVAIL;
1968                                 splx(s);
1969                                 break;
1970                         }
1971                         ifp = ro.ro_rt->rt_ifp;
1972                         rtfree(ro.ro_rt);
1973                 }
1974                 else {
1975                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1976                 }
1977
1978                 /*
1979                  * See if we found an interface, and confirm that it
1980                  * supports multicast.
1981                  */
1982                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1983                         error = EADDRNOTAVAIL;
1984                         splx(s);
1985                         break;
1986                 }
1987                 /*
1988                  * See if the membership already exists or if all the
1989                  * membership slots are full.
1990                  */
1991                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1992                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1993                             imo->imo_membership[i]->inm_addr.s_addr
1994                                                 == mreq.imr_multiaddr.s_addr)
1995                                 break;
1996                 }
1997                 if (i < imo->imo_num_memberships) {
1998                         error = EADDRINUSE;
1999                         splx(s);
2000                         break;
2001                 }
2002                 if (i == IP_MAX_MEMBERSHIPS) {
2003                         error = ETOOMANYREFS;
2004                         splx(s);
2005                         break;
2006                 }
2007                 /*
2008                  * Everything looks good; add a new record to the multicast
2009                  * address list for the given interface.
2010                  */
2011                 if ((imo->imo_membership[i] =
2012                     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2013                         error = ENOBUFS;
2014                         splx(s);
2015                         break;
2016                 }
2017                 ++imo->imo_num_memberships;
2018                 splx(s);
2019                 break;
2020
2021         case IP_DROP_MEMBERSHIP:
2022                 /*
2023                  * Drop a multicast group membership.
2024                  * Group must be a valid IP multicast address.
2025                  */
2026                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2027                 if (error)
2028                         break;
2029
2030                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2031                         error = EINVAL;
2032                         break;
2033                 }
2034
2035                 s = splimp();
2036                 /*
2037                  * If an interface address was specified, get a pointer
2038                  * to its ifnet structure.
2039                  */
2040                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2041                         ifp = NULL;
2042                 else {
2043                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2044                         if (ifp == NULL) {
2045                                 error = EADDRNOTAVAIL;
2046                                 splx(s);
2047                                 break;
2048                         }
2049                 }
2050                 /*
2051                  * Find the membership in the membership array.
2052                  */
2053                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2054                         if ((ifp == NULL ||
2055                              imo->imo_membership[i]->inm_ifp == ifp) &&
2056                              imo->imo_membership[i]->inm_addr.s_addr ==
2057                              mreq.imr_multiaddr.s_addr)
2058                                 break;
2059                 }
2060                 if (i == imo->imo_num_memberships) {
2061                         error = EADDRNOTAVAIL;
2062                         splx(s);
2063                         break;
2064                 }
2065                 /*
2066                  * Give up the multicast address record to which the
2067                  * membership points.
2068                  */
2069                 in_delmulti(imo->imo_membership[i]);
2070                 /*
2071                  * Remove the gap in the membership array.
2072                  */
2073                 for (++i; i < imo->imo_num_memberships; ++i)
2074                         imo->imo_membership[i-1] = imo->imo_membership[i];
2075                 --imo->imo_num_memberships;
2076                 splx(s);
2077                 break;
2078
2079         default:
2080                 error = EOPNOTSUPP;
2081                 break;
2082         }
2083
2084         /*
2085          * If all options have default values, no need to keep the mbuf.
2086          */
2087         if (imo->imo_multicast_ifp == NULL &&
2088             imo->imo_multicast_vif == -1 &&
2089             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2090             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2091             imo->imo_num_memberships == 0) {
2092                 free(*imop, M_IPMOPTS);
2093                 *imop = NULL;
2094         }
2095
2096         return (error);
2097 }
2098
2099 /*
2100  * Return the IP multicast options in response to user getsockopt().
2101  */
2102 static int
2103 ip_getmoptions(sopt, imo)
2104         struct sockopt *sopt;
2105         struct ip_moptions *imo;
2106 {
2107         struct in_addr addr;
2108         struct in_ifaddr *ia;
2109         int error, optval;
2110         u_char coptval;
2111
2112         error = 0;
2113         switch (sopt->sopt_name) {
2114         case IP_MULTICAST_VIF:
2115                 if (imo != NULL)
2116                         optval = imo->imo_multicast_vif;
2117                 else
2118                         optval = -1;
2119                 error = sooptcopyout(sopt, &optval, sizeof optval);
2120                 break;
2121
2122         case IP_MULTICAST_IF:
2123                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2124                         addr.s_addr = INADDR_ANY;
2125                 else if (imo->imo_multicast_addr.s_addr) {
2126                         /* return the value user has set */
2127                         addr = imo->imo_multicast_addr;
2128                 } else {
2129                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2130                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2131                                 : IA_SIN(ia)->sin_addr.s_addr;
2132                 }
2133                 error = sooptcopyout(sopt, &addr, sizeof addr);
2134                 break;
2135
2136         case IP_MULTICAST_TTL:
2137                 if (imo == 0)
2138                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2139                 else
2140                         optval = coptval = imo->imo_multicast_ttl;
2141                 if (sopt->sopt_valsize == 1)
2142                         error = sooptcopyout(sopt, &coptval, 1);
2143                 else
2144                         error = sooptcopyout(sopt, &optval, sizeof optval);
2145                 break;
2146
2147         case IP_MULTICAST_LOOP:
2148                 if (imo == 0)
2149                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2150                 else
2151                         optval = coptval = imo->imo_multicast_loop;
2152                 if (sopt->sopt_valsize == 1)
2153                         error = sooptcopyout(sopt, &coptval, 1);
2154                 else
2155                         error = sooptcopyout(sopt, &optval, sizeof optval);
2156                 break;
2157
2158         default:
2159                 error = ENOPROTOOPT;
2160                 break;
2161         }
2162         return (error);
2163 }
2164
2165 /*
2166  * Discard the IP multicast options.
2167  */
2168 void
2169 ip_freemoptions(imo)
2170         struct ip_moptions *imo;
2171 {
2172         int i;
2173
2174         if (imo != NULL) {
2175                 for (i = 0; i < imo->imo_num_memberships; ++i)
2176                         in_delmulti(imo->imo_membership[i]);
2177                 free(imo, M_IPMOPTS);
2178         }
2179 }
2180
2181 /*
2182  * Routine called from ip_output() to loop back a copy of an IP multicast
2183  * packet to the input queue of a specified interface.  Note that this
2184  * calls the output routine of the loopback "driver", but with an interface
2185  * pointer that might NOT be a loopback interface -- evil, but easier than
2186  * replicating that code here.
2187  */
2188 static void
2189 ip_mloopback(ifp, m, dst, hlen)
2190         struct ifnet *ifp;
2191         struct mbuf *m;
2192         struct sockaddr_in *dst;
2193         int hlen;
2194 {
2195         struct ip *ip;
2196         struct mbuf *copym;
2197
2198         copym = m_copy(m, 0, M_COPYALL);
2199         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2200                 copym = m_pullup(copym, hlen);
2201         if (copym != NULL) {
2202                 /*
2203                  * We don't bother to fragment if the IP length is greater
2204                  * than the interface's MTU.  Can this possibly matter?
2205                  */
2206                 ip = mtod(copym, struct ip *);
2207                 ip->ip_len = htons(ip->ip_len);
2208                 ip->ip_off = htons(ip->ip_off);
2209                 ip->ip_sum = 0;
2210                 if (ip->ip_vhl == IP_VHL_BORING) {
2211                         ip->ip_sum = in_cksum_hdr(ip);
2212                 } else {
2213                         ip->ip_sum = in_cksum(copym, hlen);
2214                 }
2215                 /*
2216                  * NB:
2217                  * It's not clear whether there are any lingering
2218                  * reentrancy problems in other areas which might
2219                  * be exposed by using ip_input directly (in
2220                  * particular, everything which modifies the packet
2221                  * in-place).  Yet another option is using the
2222                  * protosw directly to deliver the looped back
2223                  * packet.  For the moment, we'll err on the side
2224                  * of safety by using if_simloop().
2225                  */
2226 #if 1 /* XXX */
2227                 if (dst->sin_family != AF_INET) {
2228                         printf("ip_mloopback: bad address family %d\n",
2229                                                 dst->sin_family);
2230                         dst->sin_family = AF_INET;
2231                 }
2232 #endif
2233
2234 #ifdef notdef
2235                 copym->m_pkthdr.rcvif = ifp;
2236                 ip_input(copym);
2237 #else
2238                 /* if the checksum hasn't been computed, mark it as valid */
2239                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2240                         copym->m_pkthdr.csum_flags |=
2241                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2242                         copym->m_pkthdr.csum_data = 0xffff;
2243                 }
2244                 if_simloop(ifp, copym, dst->sin_family, 0);
2245 #endif
2246         }
2247 }