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 1.11 2004/03/06 01:58:55 hsu 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 #include <net/netisr.h>
  63 #ifdef PFIL_HOOKS
  64 #include <net/pfil.h>
  65 #endif
  66 #include <net/route.h>
  67
  68 #include <netinet/in.h>
  69 #include <netinet/in_systm.h>
  70 #include <netinet/ip.h>
  71 #include <netinet/in_pcb.h>
  72 #include <netinet/in_var.h>
  73 #include <netinet/ip_var.h>
  74
  75 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  76
  77 #ifdef IPSEC
  78 #include <netinet6/ipsec.h>
  79 #include <netproto/key/key.h>
  80 #ifdef IPSEC_DEBUG
  81 #include <netproto/key/key_debug.h>
  82 #else
  83 #define KEYDEBUG(lev,arg)
  84 #endif
  85 #endif /*IPSEC*/
  86
  87 #ifdef FAST_IPSEC
  88 #include <netipsec/ipsec.h>
  89 #include <netipsec/xform.h>
  90 #include <netipsec/key.h>
  91 #endif /*FAST_IPSEC*/
  92
  93 #include <net/ipfw/ip_fw.h>
  94 #include <net/dummynet/ip_dummynet.h>
  95
  96 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
  97                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  98                                   (ntohl(a.s_addr)>>16)&0xFF,\
  99                                   (ntohl(a.s_addr)>>8)&0xFF,\
 100                                   (ntohl(a.s_addr))&0xFF, y);
 101
 102 u_short ip_id;
 103
 104 #ifdef MBUF_STRESS_TEST
 105 int mbuf_frag_size = 0;
 106 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 107         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 108 #endif
 109
 110 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 111 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 112 static void     ip_mloopback
 113         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 114 static int      ip_getmoptions
 115         (struct sockopt *, struct ip_moptions *);
 116 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 117 static int      ip_setmoptions
 118         (struct sockopt *, struct ip_moptions **);
 119
 120 int     ip_optcopy(struct ip *, struct ip *);
 121 extern int (*fr_checkp) (struct ip *, int, struct ifnet *, int, struct mbuf **);
 122
 123
 124 extern  struct protosw inetsw[];
 125
 126 /*
 127  * IP output.  The packet in mbuf chain m contains a skeletal IP
 128  * header (with len, off, ttl, proto, tos, src, dst).
 129  * The mbuf chain containing the packet will be freed.
 130  * The mbuf opt, if present, will not be freed.
 131  */
 132 int
 133 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 134         int flags, struct ip_moptions *imo, struct inpcb *inp)
 135 {
 136         struct ip *ip;
 137         struct ifnet *ifp = NULL;       /* keep compiler happy */
 138         struct mbuf *m;
 139         int hlen = sizeof (struct ip);
 140         int len, off, error = 0;
 141         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 142         struct in_ifaddr *ia = NULL;
 143         int isbroadcast, sw_csum;
 144         struct in_addr pkt_dst;
 145 #ifdef IPSEC
 146         struct route iproute;
 147         struct secpolicy *sp = NULL;
 148         struct socket *so = inp ? inp->inp_socket : NULL;
 149 #endif
 150 #ifdef FAST_IPSEC
 151         struct route iproute;
 152         struct m_tag *mtag;
 153         struct secpolicy *sp = NULL;
 154         struct tdb_ident *tdbi;
 155         int s;
 156 #endif /* FAST_IPSEC */
 157         struct ip_fw_args args;
 158         int src_was_INADDR_ANY = 0;     /* as the name says... */
 159
 160         args.eh = NULL;
 161         args.rule = NULL;
 162         args.next_hop = NULL;
 163         args.divert_rule = 0;                   /* divert cookie */
 164
 165         /* Grab info from MT_TAG mbufs prepended to the chain. */
 166         for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
 167                 switch(m0->_m_tag_id) {
 168                 default:
 169                         printf("ip_output: unrecognised MT_TAG tag %d\n",
 170                             m0->_m_tag_id);
 171                         break;
 172
 173                 case PACKET_TAG_DUMMYNET:
 174                         /*
 175                          * the packet was already tagged, so part of the
 176                          * processing was already done, and we need to go down.
 177                          * Get parameters from the header.
 178                          */
 179                         args.rule = ((struct dn_pkt *)m0)->rule;
 180                         opt = NULL ;
 181                         ro = & ( ((struct dn_pkt *)m0)->ro ) ;
 182                         imo = NULL ;
 183                         dst = ((struct dn_pkt *)m0)->dn_dst ;
 184                         ifp = ((struct dn_pkt *)m0)->ifp ;
 185                         flags = ((struct dn_pkt *)m0)->flags ;
 186                         break;
 187
 188                 case PACKET_TAG_DIVERT:
 189                         args.divert_rule = (int)m0->m_data & 0xffff;
 190                         break;
 191
 192                 case PACKET_TAG_IPFORWARD:
 193                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 194                         break;
 195                 }
 196         }
 197         m = m0;
 198
 199         KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
 200 #ifndef FAST_IPSEC
 201         KASSERT(ro != NULL, ("ip_output: no route, proto %d",
 202             mtod(m, struct ip *)->ip_p));
 203 #endif
 204
 205         if (args.rule != NULL) {        /* dummynet already saw us */
 206                 ip = mtod(m, struct ip *);
 207                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 208                 if (ro->ro_rt)
 209                         ia = ifatoia(ro->ro_rt->rt_ifa);
 210                 goto sendit;
 211         }
 212
 213         if (opt) {
 214                 len = 0;
 215                 m = ip_insertoptions(m, opt, &len);
 216                 if (len != 0)
 217                         hlen = len;
 218         }
 219         ip = mtod(m, struct ip *);
 220         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 221
 222         /*
 223          * Fill in IP header.
 224          */
 225         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 226                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 227                 ip->ip_off &= IP_DF;
 228 #ifdef RANDOM_IP_ID
 229                 ip->ip_id = ip_randomid();
 230 #else
 231                 ip->ip_id = htons(ip_id++);
 232 #endif
 233                 ipstat.ips_localout++;
 234         } else {
 235                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 236         }
 237
 238 #ifdef FAST_IPSEC
 239         if (ro == NULL) {
 240                 ro = &iproute;
 241                 bzero(ro, sizeof (*ro));
 242         }
 243 #endif /* FAST_IPSEC */
 244         dst = (struct sockaddr_in *)&ro->ro_dst;
 245         /*
 246          * If there is a cached route,
 247          * check that it is to the same destination
 248          * and is still up.  If not, free it and try again.
 249          * The address family should also be checked in case of sharing the
 250          * cache with IPv6.
 251          */
 252         if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
 253                           dst->sin_family != AF_INET ||
 254                           dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 255                 RTFREE(ro->ro_rt);
 256                 ro->ro_rt = (struct rtentry *)0;
 257         }
 258         if (ro->ro_rt == 0) {
 259                 bzero(dst, sizeof(*dst));
 260                 dst->sin_family = AF_INET;
 261                 dst->sin_len = sizeof(*dst);
 262                 dst->sin_addr = pkt_dst;
 263         }
 264         /*
 265          * If routing to interface only,
 266          * short circuit routing lookup.
 267          */
 268         if (flags & IP_ROUTETOIF) {
 269                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
 270                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
 271                         ipstat.ips_noroute++;
 272                         error = ENETUNREACH;
 273                         goto bad;
 274                 }
 275                 ifp = ia->ia_ifp;
 276                 ip->ip_ttl = 1;
 277                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 278         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 279             imo != NULL && imo->imo_multicast_ifp != NULL) {
 280                 /*
 281                  * Bypass the normal routing lookup for multicast
 282                  * packets if the interface is specified.
 283                  */
 284                 ifp = imo->imo_multicast_ifp;
 285                 IFP_TO_IA(ifp, ia);
 286                 isbroadcast = 0;        /* fool gcc */
 287         } else {
 288                 /*
 289                  * If this is the case, we probably don't want to allocate
 290                  * a protocol-cloned route since we didn't get one from the
 291                  * ULP.  This lets TCP do its thing, while not burdening
 292                  * forwarding or ICMP with the overhead of cloning a route.
 293                  * Of course, we still want to do any cloning requested by
 294                  * the link layer, as this is probably required in all cases
 295                  * for correct operation (as it is for ARP).
 296                  */
 297                 if (ro->ro_rt == 0)
 298                         rtalloc_ign(ro, RTF_PRCLONING);
 299                 if (ro->ro_rt == 0) {
 300                         ipstat.ips_noroute++;
 301                         error = EHOSTUNREACH;
 302                         goto bad;
 303                 }
 304                 ia = ifatoia(ro->ro_rt->rt_ifa);
 305                 ifp = ro->ro_rt->rt_ifp;
 306                 ro->ro_rt->rt_use++;
 307                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 308                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 309                 if (ro->ro_rt->rt_flags & RTF_HOST)
 310                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 311                 else
 312                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 313         }
 314         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 315                 struct in_multi *inm;
 316
 317                 m->m_flags |= M_MCAST;
 318                 /*
 319                  * IP destination address is multicast.  Make sure "dst"
 320                  * still points to the address in "ro".  (It may have been
 321                  * changed to point to a gateway address, above.)
 322                  */
 323                 dst = (struct sockaddr_in *)&ro->ro_dst;
 324                 /*
 325                  * See if the caller provided any multicast options
 326                  */
 327                 if (imo != NULL) {
 328                         ip->ip_ttl = imo->imo_multicast_ttl;
 329                         if (imo->imo_multicast_vif != -1)
 330                                 ip->ip_src.s_addr =
 331                                     ip_mcast_src ?
 332                                     ip_mcast_src(imo->imo_multicast_vif) :
 333                                     INADDR_ANY;
 334                 } else
 335                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 336                 /*
 337                  * Confirm that the outgoing interface supports multicast.
 338                  */
 339                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 340                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 341                                 ipstat.ips_noroute++;
 342                                 error = ENETUNREACH;
 343                                 goto bad;
 344                         }
 345                 }
 346                 /*
 347                  * If source address not specified yet, use address
 348                  * of outgoing interface.
 349                  */
 350                 if (ip->ip_src.s_addr == INADDR_ANY) {
 351                         /* Interface may have no addresses. */
 352                         if (ia != NULL)
 353                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 354                 }
 355
 356                 if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 357                         /*
 358                          * XXX
 359                          * delayed checksums are not currently
 360                          * compatible with IP multicast routing
 361                          */
 362                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 363                                 in_delayed_cksum(m);
 364                                 m->m_pkthdr.csum_flags &=
 365                                         ~CSUM_DELAY_DATA;
 366                         }
 367                 }
 368                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 369                 if (inm != NULL &&
 370                    (imo == NULL || imo->imo_multicast_loop)) {
 371                         /*
 372                          * If we belong to the destination multicast group
 373                          * on the outgoing interface, and the caller did not
 374                          * forbid loopback, loop back a copy.
 375                          */
 376                         ip_mloopback(ifp, m, dst, hlen);
 377                 }
 378                 else {
 379                         /*
 380                          * If we are acting as a multicast router, perform
 381                          * multicast forwarding as if the packet had just
 382                          * arrived on the interface to which we are about
 383                          * to send.  The multicast forwarding function
 384                          * recursively calls this function, using the
 385                          * IP_FORWARDING flag to prevent infinite recursion.
 386                          *
 387                          * Multicasts that are looped back by ip_mloopback(),
 388                          * above, will be forwarded by the ip_input() routine,
 389                          * if necessary.
 390                          */
 391                         if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 392                                 /*
 393                                  * If rsvp daemon is not running, do not
 394                                  * set ip_moptions. This ensures that the packet
 395                                  * is multicast and not just sent down one link
 396                                  * as prescribed by rsvpd.
 397                                  */
 398                                 if (!rsvp_on)
 399                                         imo = NULL;
 400                                 if (ip_mforward &&
 401                                     ip_mforward(ip, ifp, m, imo) != 0) {
 402                                         m_freem(m);
 403                                         goto done;
 404                                 }
 405                         }
 406                 }
 407
 408                 /*
 409                  * Multicasts with a time-to-live of zero may be looped-
 410                  * back, above, but must not be transmitted on a network.
 411                  * Also, multicasts addressed to the loopback interface
 412                  * are not sent -- the above call to ip_mloopback() will
 413                  * loop back a copy if this host actually belongs to the
 414                  * destination group on the loopback interface.
 415                  */
 416                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 417                         m_freem(m);
 418                         goto done;
 419                 }
 420
 421                 goto sendit;
 422         }
 423 #ifndef notdef
 424         /*
 425          * If the source address is not specified yet, use the address
 426          * of the outoing interface. In case, keep note we did that, so
 427          * if the the firewall changes the next-hop causing the output
 428          * interface to change, we can fix that.
 429          */
 430         if (ip->ip_src.s_addr == INADDR_ANY) {
 431                 /* Interface may have no addresses. */
 432                 if (ia != NULL) {
 433                         ip->ip_src = IA_SIN(ia)->sin_addr;
 434                         src_was_INADDR_ANY = 1;
 435                 }
 436         }
 437 #endif /* notdef */
 438         /*
 439          * Verify that we have any chance at all of being able to queue
 440          *      the packet or packet fragments
 441          */
 442         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 443                 ifp->if_snd.ifq_maxlen) {
 444                         error = ENOBUFS;
 445                         ipstat.ips_odropped++;
 446                         goto bad;
 447         }
 448
 449         /*
 450          * Look for broadcast address and
 451          * verify user is allowed to send
 452          * such a packet.
 453          */
 454         if (isbroadcast) {
 455                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 456                         error = EADDRNOTAVAIL;
 457                         goto bad;
 458                 }
 459                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 460                         error = EACCES;
 461                         goto bad;
 462                 }
 463                 /* don't allow broadcast messages to be fragmented */
 464                 if (ip->ip_len > ifp->if_mtu) {
 465                         error = EMSGSIZE;
 466                         goto bad;
 467                 }
 468                 m->m_flags |= M_BCAST;
 469         } else {
 470                 m->m_flags &= ~M_BCAST;
 471         }
 472
 473 sendit:
 474 #ifdef IPSEC
 475         /* get SP for this packet */
 476         if (so == NULL)
 477                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 478         else
 479                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 480
 481         if (sp == NULL) {
 482                 ipsecstat.out_inval++;
 483                 goto bad;
 484         }
 485
 486         error = 0;
 487
 488         /* check policy */
 489         switch (sp->policy) {
 490         case IPSEC_POLICY_DISCARD:
 491                 /*
 492                  * This packet is just discarded.
 493                  */
 494                 ipsecstat.out_polvio++;
 495                 goto bad;
 496
 497         case IPSEC_POLICY_BYPASS:
 498         case IPSEC_POLICY_NONE:
 499                 /* no need to do IPsec. */
 500                 goto skip_ipsec;
 501
 502         case IPSEC_POLICY_IPSEC:
 503                 if (sp->req == NULL) {
 504                         /* acquire a policy */
 505                         error = key_spdacquire(sp);
 506                         goto bad;
 507                 }
 508                 break;
 509
 510         case IPSEC_POLICY_ENTRUST:
 511         default:
 512                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
 513         }
 514     {
 515         struct ipsec_output_state state;
 516         bzero(&state, sizeof(state));
 517         state.m = m;
 518         if (flags & IP_ROUTETOIF) {
 519                 state.ro = &iproute;
 520                 bzero(&iproute, sizeof(iproute));
 521         } else
 522                 state.ro = ro;
 523         state.dst = (struct sockaddr *)dst;
 524
 525         ip->ip_sum = 0;
 526
 527         /*
 528          * XXX
 529          * delayed checksums are not currently compatible with IPsec
 530          */
 531         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 532                 in_delayed_cksum(m);
 533                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 534         }
 535
 536         ip->ip_len = htons(ip->ip_len);
 537         ip->ip_off = htons(ip->ip_off);
 538
 539         error = ipsec4_output(&state, sp, flags);
 540
 541         m = state.m;
 542         if (flags & IP_ROUTETOIF) {
 543                 /*
 544                  * if we have tunnel mode SA, we may need to ignore
 545                  * IP_ROUTETOIF.
 546                  */
 547                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 548                         flags &= ~IP_ROUTETOIF;
 549                         ro = state.ro;
 550                 }
 551         } else
 552                 ro = state.ro;
 553         dst = (struct sockaddr_in *)state.dst;
 554         if (error) {
 555                 /* mbuf is already reclaimed in ipsec4_output. */
 556                 m0 = NULL;
 557                 switch (error) {
 558                 case EHOSTUNREACH:
 559                 case ENETUNREACH:
 560                 case EMSGSIZE:
 561                 case ENOBUFS:
 562                 case ENOMEM:
 563                         break;
 564                 default:
 565                         printf("ip4_output (ipsec): error code %d\n", error);
 566                         /*fall through*/
 567                 case ENOENT:
 568                         /* don't show these error codes to the user */
 569                         error = 0;
 570                         break;
 571                 }
 572                 goto bad;
 573         }
 574     }
 575
 576         /* be sure to update variables that are affected by ipsec4_output() */
 577         ip = mtod(m, struct ip *);
 578 #ifdef _IP_VHL
 579         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 580 #else
 581         hlen = ip->ip_hl << 2;
 582 #endif
 583         if (ro->ro_rt == NULL) {
 584                 if ((flags & IP_ROUTETOIF) == 0) {
 585                         printf("ip_output: "
 586                                 "can't update route after IPsec processing\n");
 587                         error = EHOSTUNREACH;   /*XXX*/
 588                         goto bad;
 589                 }
 590         } else {
 591                 ia = ifatoia(ro->ro_rt->rt_ifa);
 592                 ifp = ro->ro_rt->rt_ifp;
 593         }
 594
 595         /* make it flipped, again. */
 596         ip->ip_len = ntohs(ip->ip_len);
 597         ip->ip_off = ntohs(ip->ip_off);
 598 skip_ipsec:
 599 #endif /*IPSEC*/
 600 #ifdef FAST_IPSEC
 601         /*
 602          * Check the security policy (SP) for the packet and, if
 603          * required, do IPsec-related processing.  There are two
 604          * cases here; the first time a packet is sent through
 605          * it will be untagged and handled by ipsec4_checkpolicy.
 606          * If the packet is resubmitted to ip_output (e.g. after
 607          * AH, ESP, etc. processing), there will be a tag to bypass
 608          * the lookup and related policy checking.
 609          */
 610         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 611         s = splnet();
 612         if (mtag != NULL) {
 613                 tdbi = (struct tdb_ident *)(mtag + 1);
 614                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 615                 if (sp == NULL)
 616                         error = -EINVAL;        /* force silent drop */
 617                 m_tag_delete(m, mtag);
 618         } else {
 619                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 620                                         &error, inp);
 621         }
 622         /*
 623          * There are four return cases:
 624          *    sp != NULL                    apply IPsec policy
 625          *    sp == NULL, error == 0        no IPsec handling needed
 626          *    sp == NULL, error == -EINVAL  discard packet w/o error
 627          *    sp == NULL, error != 0        discard packet, report error
 628          */
 629         if (sp != NULL) {
 630                 /* Loop detection, check if ipsec processing already done */
 631                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 632                 for (mtag = m_tag_first(m); mtag != NULL;
 633                      mtag = m_tag_next(m, mtag)) {
 634                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 635                                 continue;
 636                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 637                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 638                                 continue;
 639                         /*
 640                          * Check if policy has an SA associated with it.
 641                          * This can happen when an SP has yet to acquire
 642                          * an SA; e.g. on first reference.  If it occurs,
 643                          * then we let ipsec4_process_packet do its thing.
 644                          */
 645                         if (sp->req->sav == NULL)
 646                                 break;
 647                         tdbi = (struct tdb_ident *)(mtag + 1);
 648                         if (tdbi->spi == sp->req->sav->spi &&
 649                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 650                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 651                                  sizeof (union sockaddr_union)) == 0) {
 652                                 /*
 653                                  * No IPsec processing is needed, free
 654                                  * reference to SP.
 655                                  *
 656                                  * NB: null pointer to avoid free at
 657                                  *     done: below.
 658                                  */
 659                                 KEY_FREESP(&sp), sp = NULL;
 660                                 splx(s);
 661                                 goto spd_done;
 662                         }
 663                 }
 664
 665                 /*
 666                  * Do delayed checksums now because we send before
 667                  * this is done in the normal processing path.
 668                  */
 669                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 670                         in_delayed_cksum(m);
 671                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 672                 }
 673
 674                 ip->ip_len = htons(ip->ip_len);
 675                 ip->ip_off = htons(ip->ip_off);
 676
 677                 /* NB: callee frees mbuf */
 678                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 679                 /*
 680                  * Preserve KAME behaviour: ENOENT can be returned
 681                  * when an SA acquire is in progress.  Don't propagate
 682                  * this to user-level; it confuses applications.
 683                  *
 684                  * XXX this will go away when the SADB is redone.
 685                  */
 686                 if (error == ENOENT)
 687                         error = 0;
 688                 splx(s);
 689                 goto done;
 690         } else {
 691                 splx(s);
 692
 693                 if (error != 0) {
 694                         /*
 695                          * Hack: -EINVAL is used to signal that a packet
 696                          * should be silently discarded.  This is typically
 697                          * because we asked key management for an SA and
 698                          * it was delayed (e.g. kicked up to IKE).
 699                          */
 700                         if (error == -EINVAL)
 701                                 error = 0;
 702                         goto bad;
 703                 } else {
 704                         /* No IPsec processing for this packet. */
 705                 }
 706 #ifdef notyet
 707                 /*
 708                  * If deferred crypto processing is needed, check that
 709                  * the interface supports it.
 710                  */
 711                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 712                 if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
 713                         /* notify IPsec to do its own crypto */
 714                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
 715                         error = EHOSTUNREACH;
 716                         goto bad;
 717                 }
 718 #endif
 719         }
 720 spd_done:
 721 #endif /* FAST_IPSEC */
 722         /*
 723          * IpHack's section.
 724          * - Xlate: translate packet's addr/port (NAT).
 725          * - Firewall: deny/allow/etc.
 726          * - Wrap: fake packet's addr/port <unimpl.>
 727          * - Encapsulate: put it in another IP and send out. <unimp.>
 728          */
 729 #ifdef PFIL_HOOKS
 730         /*
 731          * Run through list of hooks for output packets.
 732          */
 733         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 734         if (error != 0 || m == NULL)
 735                 goto done;
 736         ip = mtod(m, struct ip *);
 737 #endif /* PFIL_HOOKS */
 738
 739         /*
 740          * Check with the firewall...
 741          * but not if we are already being fwd'd from a firewall.
 742          */
 743         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 744                 struct sockaddr_in *old = dst;
 745
 746                 args.m = m;
 747                 args.next_hop = dst;
 748                 args.oif = ifp;
 749                 off = ip_fw_chk_ptr(&args);
 750                 m = args.m;
 751                 dst = args.next_hop;
 752
 753                 /*
 754                  * On return we must do the following:
 755                  * m == NULL    -> drop the pkt (old interface, deprecated)
 756                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 757                  * 1<=off<= 0xffff              -> DIVERT
 758                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 759                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 760                  * dst != old                   -> IPFIREWALL_FORWARD
 761                  * off==0, dst==old             -> accept
 762                  * If some of the above modules are not compiled in, then
 763                  * we should't have to check the corresponding condition
 764                  * (because the ipfw control socket should not accept
 765                  * unsupported rules), but better play safe and drop
 766                  * packets in case of doubt.
 767                  */
 768                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 769                         if (m)
 770                                 m_freem(m);
 771                         error = EACCES;
 772                         goto done;
 773                 }
 774                 ip = mtod(m, struct ip *);
 775                 if (off == 0 && dst == old)             /* common case */
 776                         goto pass;
 777                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
 778                         /*
 779                          * pass the pkt to dummynet. Need to include
 780                          * pipe number, m, ifp, ro, dst because these are
 781                          * not recomputed in the next pass.
 782                          * All other parameters have been already used and
 783                          * so they are not needed anymore.
 784                          * XXX note: if the ifp or ro entry are deleted
 785                          * while a pkt is in dummynet, we are in trouble!
 786                          */
 787                         args.ro = ro;
 788                         args.dst = dst;
 789                         args.flags = flags;
 790
 791                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
 792                                 &args);
 793                         goto done;
 794                 }
 795 #ifdef IPDIVERT
 796                 if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
 797                         struct mbuf *clone = NULL;
 798
 799                         /* Clone packet if we're doing a 'tee' */
 800                         if ((off & IP_FW_PORT_TEE_FLAG) != 0)
 801                                 clone = m_dup(m, M_DONTWAIT);
 802
 803                         /*
 804                          * XXX
 805                          * delayed checksums are not currently compatible
 806                          * with divert sockets.
 807                          */
 808                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 809                                 in_delayed_cksum(m);
 810                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 811                         }
 812
 813                         /* Restore packet header fields to original values */
 814                         ip->ip_len = htons(ip->ip_len);
 815                         ip->ip_off = htons(ip->ip_off);
 816
 817                         /* Deliver packet to divert input routine */
 818                         divert_packet(m, 0, off & 0xffff, args.divert_rule);
 819
 820                         /* If 'tee', continue with original packet */
 821                         if (clone != NULL) {
 822                                 m = clone;
 823                                 ip = mtod(m, struct ip *);
 824                                 goto pass;
 825                         }
 826                         goto done;
 827                 }
 828 #endif
 829
 830                 /* IPFIREWALL_FORWARD */
 831                 /*
 832                  * Check dst to make sure it is directly reachable on the
 833                  * interface we previously thought it was.
 834                  * If it isn't (which may be likely in some situations) we have
 835                  * to re-route it (ie, find a route for the next-hop and the
 836                  * associated interface) and set them here. This is nested
 837                  * forwarding which in most cases is undesirable, except where
 838                  * such control is nigh impossible. So we do it here.
 839                  * And I'm babbling.
 840                  */
 841                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 842 #if 0
 843                         /*
 844                          * XXX To improve readability, this block should be
 845                          * changed into a function call as below:
 846                          */
 847                         error = ip_ipforward(&m, &dst, &ifp);
 848                         if (error)
 849                                 goto bad;
 850                         if (m == NULL) /* ip_input consumed the mbuf */
 851                                 goto done;
 852 #else
 853                         struct in_ifaddr *ia;
 854
 855                         /*
 856                          * XXX sro_fwd below is static, and a pointer
 857                          * to it gets passed to routines downstream.
 858                          * This could have surprisingly bad results in
 859                          * practice, because its content is overwritten
 860                          * by subsequent packets.
 861                          */
 862                         /* There must be a better way to do this next line... */
 863                         static struct route sro_fwd;
 864                         struct route *ro_fwd = &sro_fwd;
 865
 866 #if 0
 867                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 868                             dst->sin_addr, "\n");
 869 #endif
 870
 871                         /*
 872                          * We need to figure out if we have been forwarded
 873                          * to a local socket. If so, then we should somehow
 874                          * "loop back" to ip_input, and get directed to the
 875                          * PCB as if we had received this packet. This is
 876                          * because it may be dificult to identify the packets
 877                          * you want to forward until they are being output
 878                          * and have selected an interface. (e.g. locally
 879                          * initiated packets) If we used the loopback inteface,
 880                          * we would not be able to control what happens
 881                          * as the packet runs through ip_input() as
 882                          * it is done through a ISR.
 883                          */
 884                         LIST_FOREACH(ia,
 885                             INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 886                                 /*
 887                                  * If the addr to forward to is one
 888                                  * of ours, we pretend to
 889                                  * be the destination for this packet.
 890                                  */
 891                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 892                                                  dst->sin_addr.s_addr)
 893                                         break;
 894                         }
 895                         if (ia) {       /* tell ip_input "dont filter" */
 896                                 struct m_hdr tag;
 897                                 struct netmsg_packet msg;
 898
 899                                 tag.mh_type = MT_TAG;
 900                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 901                                 tag.mh_data = (caddr_t)args.next_hop;
 902                                 tag.mh_next = m;
 903
 904                                 if (m->m_pkthdr.rcvif == NULL)
 905                                         m->m_pkthdr.rcvif = ifunit("lo0");
 906                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 907                                         m->m_pkthdr.csum_flags |=
 908                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 909                                         m0->m_pkthdr.csum_data = 0xffff;
 910                                 }
 911                                 m->m_pkthdr.csum_flags |=
 912                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 913                                 ip->ip_len = htons(ip->ip_len);
 914                                 ip->ip_off = htons(ip->ip_off);
 915                                 msg.nm_packet = (struct mbuf *)&tag;
 916                                 ip_input((struct netmsg *)&msg);
 917                                 goto done;
 918                         }
 919                         /* Some of the logic for this was
 920                          * nicked from above.
 921                          *
 922                          * This rewrites the cached route in a local PCB.
 923                          * Is this what we want to do?
 924                          */
 925                         bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
 926
 927                         ro_fwd->ro_rt = 0;
 928                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 929
 930                         if (ro_fwd->ro_rt == 0) {
 931                                 ipstat.ips_noroute++;
 932                                 error = EHOSTUNREACH;
 933                                 goto bad;
 934                         }
 935
 936                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 937                         ifp = ro_fwd->ro_rt->rt_ifp;
 938                         ro_fwd->ro_rt->rt_use++;
 939                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 940                                 dst = (struct sockaddr_in *)
 941                                         ro_fwd->ro_rt->rt_gateway;
 942                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 943                                 isbroadcast =
 944                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 945                         else
 946                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 947                         if (ro->ro_rt)
 948                                 RTFREE(ro->ro_rt);
 949                         ro->ro_rt = ro_fwd->ro_rt;
 950                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 951
 952 #endif  /* ... block to be put into a function */
 953                         /*
 954                          * If we added a default src ip earlier,
 955                          * which would have been gotten from the-then
 956                          * interface, do it again, from the new one.
 957                          */
 958                         if (src_was_INADDR_ANY)
 959                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 960                         goto pass ;
 961                 }
 962
 963                 /*
 964                  * if we get here, none of the above matches, and
 965                  * we have to drop the pkt
 966                  */
 967                 m_freem(m);
 968                 error = EACCES; /* not sure this is the right error msg */
 969                 goto done;
 970         }
 971
 972 pass:
 973         /* 127/8 must not appear on wire - RFC1122. */
 974         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 975             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 976                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 977                         ipstat.ips_badaddr++;
 978                         error = EADDRNOTAVAIL;
 979                         goto bad;
 980                 }
 981         }
 982
 983         m->m_pkthdr.csum_flags |= CSUM_IP;
 984         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 985         if (sw_csum & CSUM_DELAY_DATA) {
 986                 in_delayed_cksum(m);
 987                 sw_csum &= ~CSUM_DELAY_DATA;
 988         }
 989         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 990
 991         /*
 992          * If small enough for interface, or the interface will take
 993          * care of the fragmentation for us, can just send directly.
 994          */
 995         if (ip->ip_len <= ifp->if_mtu || ifp->if_hwassist & CSUM_FRAGMENT) {
 996                 ip->ip_len = htons(ip->ip_len);
 997                 ip->ip_off = htons(ip->ip_off);
 998                 ip->ip_sum = 0;
 999                 if (sw_csum & CSUM_DELAY_IP) {
1000                         if (ip->ip_vhl == IP_VHL_BORING) {
1001                                 ip->ip_sum = in_cksum_hdr(ip);
1002                         } else {
1003                                 ip->ip_sum = in_cksum(m, hlen);
1004                         }
1005                 }
1006
1007                 /* Record statistics for this interface address. */
1008                 if (!(flags & IP_FORWARDING) && ia) {
1009                         ia->ia_ifa.if_opackets++;
1010                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1011                 }
1012
1013 #ifdef IPSEC
1014                 /* clean ipsec history once it goes out of the node */
1015                 ipsec_delaux(m);
1016 #endif
1017
1018 #ifdef MBUF_STRESS_TEST
1019                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1020                         struct mbuf *m1, *m2;
1021                         int length, tmp;
1022
1023                         tmp = length = m->m_pkthdr.len;
1024
1025                         while ((length -= mbuf_frag_size) >= 1) {
1026                                 m1 = m_split(m, length, M_DONTWAIT);
1027                                 if (m1 == NULL)
1028                                         break;
1029                                 m1->m_flags &= ~M_PKTHDR;
1030                                 m2 = m;
1031                                 while (m2->m_next != NULL)
1032                                         m2 = m2->m_next;
1033                                 m2->m_next = m1;
1034                         }
1035                         m->m_pkthdr.len = tmp;
1036                 }
1037 #endif
1038                 error = (*ifp->if_output)(ifp, m,
1039                                 (struct sockaddr *)dst, ro->ro_rt);
1040                 goto done;
1041         }
1042
1043         if (ip->ip_off & IP_DF) {
1044                 error = EMSGSIZE;
1045                 /*
1046                  * This case can happen if the user changed the MTU
1047                  * of an interface after enabling IP on it.  Because
1048                  * most netifs don't keep track of routes pointing to
1049                  * them, there is no way for one to update all its
1050                  * routes when the MTU is changed.
1051                  */
1052                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1053                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1054                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1055                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1056                 }
1057                 ipstat.ips_cantfrag++;
1058                 goto bad;
1059         }
1060
1061         /*
1062          * Too large for interface; fragment if possible. If successful,
1063          * on return, m will point to a list of packets to be sent.
1064          */
1065         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1066         if (error)
1067                 goto bad;
1068         for (; m; m = m0) {
1069                 m0 = m->m_nextpkt;
1070                 m->m_nextpkt = 0;
1071 #ifdef IPSEC
1072                 /* clean ipsec history once it goes out of the node */
1073                 ipsec_delaux(m);
1074 #endif
1075                 if (error == 0) {
1076                         /* Record statistics for this interface address. */
1077                         if (ia != NULL) {
1078                                 ia->ia_ifa.if_opackets++;
1079                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1080                         }
1081
1082                         error = (*ifp->if_output)(ifp, m,
1083                             (struct sockaddr *)dst, ro->ro_rt);
1084                 } else
1085                         m_freem(m);
1086         }
1087
1088         if (error == 0)
1089                 ipstat.ips_fragmented++;
1090
1091 done:
1092 #ifdef IPSEC
1093         if (ro == &iproute && ro->ro_rt) {
1094                 RTFREE(ro->ro_rt);
1095                 ro->ro_rt = NULL;
1096         }
1097         if (sp != NULL) {
1098                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1099                         printf("DP ip_output call free SP:%p\n", sp));
1100                 key_freesp(sp);
1101         }
1102 #endif
1103 #ifdef FAST_IPSEC
1104         if (ro == &iproute && ro->ro_rt) {
1105                 RTFREE(ro->ro_rt);
1106                 ro->ro_rt = NULL;
1107         }
1108         if (sp != NULL)
1109                 KEY_FREESP(&sp);
1110 #endif
1111         return (error);
1112 bad:
1113         m_freem(m);
1114         goto done;
1115 }
1116
1117 /*
1118  * Create a chain of fragments which fit the given mtu. m_frag points to the
1119  * mbuf to be fragmented; on return it points to the chain with the fragments.
1120  * Return 0 if no error. If error, m_frag may contain a partially built
1121  * chain of fragments that should be freed by the caller.
1122  *
1123  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1124  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1125  */
1126 int
1127 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1128             u_long if_hwassist_flags, int sw_csum)
1129 {
1130         int error = 0;
1131         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1132         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1133         int off;
1134         struct mbuf *m0 = *m_frag;      /* the original packet          */
1135         int firstlen;
1136         struct mbuf **mnext;
1137         int nfrags;
1138
1139         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1140                 ipstat.ips_cantfrag++;
1141                 return EMSGSIZE;
1142         }
1143
1144         /*
1145          * Must be able to put at least 8 bytes per fragment.
1146          */
1147         if (len < 8)
1148                 return EMSGSIZE;
1149
1150         /*
1151          * If the interface will not calculate checksums on
1152          * fragmented packets, then do it here.
1153          */
1154         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1155             (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1156                 in_delayed_cksum(m0);
1157                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1158         }
1159
1160         if (len > PAGE_SIZE) {
1161                 /*
1162                  * Fragment large datagrams such that each segment
1163                  * contains a multiple of PAGE_SIZE amount of data,
1164                  * plus headers. This enables a receiver to perform
1165                  * page-flipping zero-copy optimizations.
1166                  *
1167                  * XXX When does this help given that sender and receiver
1168                  * could have different page sizes, and also mtu could
1169                  * be less than the receiver's page size ?
1170                  */
1171                 int newlen;
1172                 struct mbuf *m;
1173
1174                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1175                         off += m->m_len;
1176
1177                 /*
1178                  * firstlen (off - hlen) must be aligned on an
1179                  * 8-byte boundary
1180                  */
1181                 if (off < hlen)
1182                         goto smart_frag_failure;
1183                 off = ((off - hlen) & ~7) + hlen;
1184                 newlen = (~PAGE_MASK) & mtu;
1185                 if ((newlen + sizeof (struct ip)) > mtu) {
1186                         /* we failed, go back the default */
1187 smart_frag_failure:
1188                         newlen = len;
1189                         off = hlen + len;
1190                 }
1191                 len = newlen;
1192
1193         } else {
1194                 off = hlen + len;
1195         }
1196
1197         firstlen = off - hlen;
1198         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1199
1200         /*
1201          * Loop through length of segment after first fragment,
1202          * make new header and copy data of each part and link onto chain.
1203          * Here, m0 is the original packet, m is the fragment being created.
1204          * The fragments are linked off the m_nextpkt of the original
1205          * packet, which after processing serves as the first fragment.
1206          */
1207         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1208                 struct ip *mhip;        /* ip header on the fragment */
1209                 struct mbuf *m;
1210                 int mhlen = sizeof (struct ip);
1211
1212                 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1213                 if (m == 0) {
1214                         error = ENOBUFS;
1215                         ipstat.ips_odropped++;
1216                         goto done;
1217                 }
1218                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1219                 /*
1220                  * In the first mbuf, leave room for the link header, then
1221                  * copy the original IP header including options. The payload
1222                  * goes into an additional mbuf chain returned by m_copy().
1223                  */
1224                 m->m_data += max_linkhdr;
1225                 mhip = mtod(m, struct ip *);
1226                 *mhip = *ip;
1227                 if (hlen > sizeof (struct ip)) {
1228                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1229                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1230                 }
1231                 m->m_len = mhlen;
1232                 /* XXX do we need to add ip->ip_off below ? */
1233                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1234                 if (off + len >= ip->ip_len) {  /* last fragment */
1235                         len = ip->ip_len - off;
1236                         m->m_flags |= M_LASTFRAG;
1237                 } else
1238                         mhip->ip_off |= IP_MF;
1239                 mhip->ip_len = htons((u_short)(len + mhlen));
1240                 m->m_next = m_copy(m0, off, len);
1241                 if (m->m_next == 0) {           /* copy failed */
1242                         m_free(m);
1243                         error = ENOBUFS;        /* ??? */
1244                         ipstat.ips_odropped++;
1245                         goto done;
1246                 }
1247                 m->m_pkthdr.len = mhlen + len;
1248                 m->m_pkthdr.rcvif = (struct ifnet *)0;
1249                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1250                 mhip->ip_off = htons(mhip->ip_off);
1251                 mhip->ip_sum = 0;
1252                 if (sw_csum & CSUM_DELAY_IP)
1253                         mhip->ip_sum = in_cksum(m, mhlen);
1254                 *mnext = m;
1255                 mnext = &m->m_nextpkt;
1256         }
1257         ipstat.ips_ofragments += nfrags;
1258
1259         /* set first marker for fragment chain */
1260         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1261         m0->m_pkthdr.csum_data = nfrags;
1262
1263         /*
1264          * Update first fragment by trimming what's been copied out
1265          * and updating header.
1266          */
1267         m_adj(m0, hlen + firstlen - ip->ip_len);
1268         m0->m_pkthdr.len = hlen + firstlen;
1269         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1270         ip->ip_off |= IP_MF;
1271         ip->ip_off = htons(ip->ip_off);
1272         ip->ip_sum = 0;
1273         if (sw_csum & CSUM_DELAY_IP)
1274                 ip->ip_sum = in_cksum(m0, hlen);
1275
1276 done:
1277         *m_frag = m0;
1278         return error;
1279 }
1280
1281 void
1282 in_delayed_cksum(struct mbuf *m)
1283 {
1284         struct ip *ip;
1285         u_short csum, offset;
1286
1287         ip = mtod(m, struct ip *);
1288         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1289         csum = in_cksum_skip(m, ip->ip_len, offset);
1290         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1291                 csum = 0xffff;
1292         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1293
1294         if (offset + sizeof(u_short) > m->m_len) {
1295                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1296                     m->m_len, offset, ip->ip_p);
1297                 /*
1298                  * XXX
1299                  * this shouldn't happen, but if it does, the
1300                  * correct behavior may be to insert the checksum
1301                  * in the existing chain instead of rearranging it.
1302                  */
1303                 m = m_pullup(m, offset + sizeof(u_short));
1304         }
1305         *(u_short *)(m->m_data + offset) = csum;
1306 }
1307
1308 /*
1309  * Insert IP options into preformed packet.
1310  * Adjust IP destination as required for IP source routing,
1311  * as indicated by a non-zero in_addr at the start of the options.
1312  *
1313  * XXX This routine assumes that the packet has no options in place.
1314  */
1315 static struct mbuf *
1316 ip_insertoptions(m, opt, phlen)
1317         struct mbuf *m;
1318         struct mbuf *opt;
1319         int *phlen;
1320 {
1321         struct ipoption *p = mtod(opt, struct ipoption *);
1322         struct mbuf *n;
1323         struct ip *ip = mtod(m, struct ip *);
1324         unsigned optlen;
1325
1326         optlen = opt->m_len - sizeof(p->ipopt_dst);
1327         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1328                 *phlen = 0;
1329                 return (m);             /* XXX should fail */
1330         }
1331         if (p->ipopt_dst.s_addr)
1332                 ip->ip_dst = p->ipopt_dst;
1333         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1334                 MGETHDR(n, M_DONTWAIT, MT_HEADER);
1335                 if (n == 0) {
1336                         *phlen = 0;
1337                         return (m);
1338                 }
1339                 n->m_pkthdr.rcvif = (struct ifnet *)0;
1340                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1341                 m->m_len -= sizeof(struct ip);
1342                 m->m_data += sizeof(struct ip);
1343                 n->m_next = m;
1344                 m = n;
1345                 m->m_len = optlen + sizeof(struct ip);
1346                 m->m_data += max_linkhdr;
1347                 (void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1348         } else {
1349                 m->m_data -= optlen;
1350                 m->m_len += optlen;
1351                 m->m_pkthdr.len += optlen;
1352                 ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1353         }
1354         ip = mtod(m, struct ip *);
1355         bcopy(p->ipopt_list, ip + 1, optlen);
1356         *phlen = sizeof(struct ip) + optlen;
1357         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1358         ip->ip_len += optlen;
1359         return (m);
1360 }
1361
1362 /*
1363  * Copy options from ip to jp,
1364  * omitting those not copied during fragmentation.
1365  */
1366 int
1367 ip_optcopy(ip, jp)
1368         struct ip *ip, *jp;
1369 {
1370         u_char *cp, *dp;
1371         int opt, optlen, cnt;
1372
1373         cp = (u_char *)(ip + 1);
1374         dp = (u_char *)(jp + 1);
1375         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1376         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1377                 opt = cp[0];
1378                 if (opt == IPOPT_EOL)
1379                         break;
1380                 if (opt == IPOPT_NOP) {
1381                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1382                         *dp++ = IPOPT_NOP;
1383                         optlen = 1;
1384                         continue;
1385                 }
1386
1387                 KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1388                     ("ip_optcopy: malformed ipv4 option"));
1389                 optlen = cp[IPOPT_OLEN];
1390                 KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1391                     ("ip_optcopy: malformed ipv4 option"));
1392
1393                 /* bogus lengths should have been caught by ip_dooptions */
1394                 if (optlen > cnt)
1395                         optlen = cnt;
1396                 if (IPOPT_COPIED(opt)) {
1397                         bcopy(cp, dp, optlen);
1398                         dp += optlen;
1399                 }
1400         }
1401         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1402                 *dp++ = IPOPT_EOL;
1403         return (optlen);
1404 }
1405
1406 /*
1407  * IP socket option processing.
1408  */
1409 int
1410 ip_ctloutput(so, sopt)
1411         struct socket *so;
1412         struct sockopt *sopt;
1413 {
1414         struct  inpcb *inp = sotoinpcb(so);
1415         int     error, optval;
1416
1417         error = optval = 0;
1418         if (sopt->sopt_level != IPPROTO_IP) {
1419                 return (EINVAL);
1420         }
1421
1422         switch (sopt->sopt_dir) {
1423         case SOPT_SET:
1424                 switch (sopt->sopt_name) {
1425                 case IP_OPTIONS:
1426 #ifdef notyet
1427                 case IP_RETOPTS:
1428 #endif
1429                 {
1430                         struct mbuf *m;
1431                         if (sopt->sopt_valsize > MLEN) {
1432                                 error = EMSGSIZE;
1433                                 break;
1434                         }
1435                         MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_HEADER);
1436                         if (m == 0) {
1437                                 error = ENOBUFS;
1438                                 break;
1439                         }
1440                         m->m_len = sopt->sopt_valsize;
1441                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1442                                             m->m_len);
1443
1444                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1445                                            m));
1446                 }
1447
1448                 case IP_TOS:
1449                 case IP_TTL:
1450                 case IP_RECVOPTS:
1451                 case IP_RECVRETOPTS:
1452                 case IP_RECVDSTADDR:
1453                 case IP_RECVIF:
1454                 case IP_FAITH:
1455                         error = sooptcopyin(sopt, &optval, sizeof optval,
1456                                             sizeof optval);
1457                         if (error)
1458                                 break;
1459
1460                         switch (sopt->sopt_name) {
1461                         case IP_TOS:
1462                                 inp->inp_ip_tos = optval;
1463                                 break;
1464
1465                         case IP_TTL:
1466                                 inp->inp_ip_ttl = optval;
1467                                 break;
1468 #define OPTSET(bit) \
1469         if (optval) \
1470                 inp->inp_flags |= bit; \
1471         else \
1472                 inp->inp_flags &= ~bit;
1473
1474                         case IP_RECVOPTS:
1475                                 OPTSET(INP_RECVOPTS);
1476                                 break;
1477
1478                         case IP_RECVRETOPTS:
1479                                 OPTSET(INP_RECVRETOPTS);
1480                                 break;
1481
1482                         case IP_RECVDSTADDR:
1483                                 OPTSET(INP_RECVDSTADDR);
1484                                 break;
1485
1486                         case IP_RECVIF:
1487                                 OPTSET(INP_RECVIF);
1488                                 break;
1489
1490                         case IP_FAITH:
1491                                 OPTSET(INP_FAITH);
1492                                 break;
1493                         }
1494                         break;
1495 #undef OPTSET
1496
1497                 case IP_MULTICAST_IF:
1498                 case IP_MULTICAST_VIF:
1499                 case IP_MULTICAST_TTL:
1500                 case IP_MULTICAST_LOOP:
1501                 case IP_ADD_MEMBERSHIP:
1502                 case IP_DROP_MEMBERSHIP:
1503                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1504                         break;
1505
1506                 case IP_PORTRANGE:
1507                         error = sooptcopyin(sopt, &optval, sizeof optval,
1508                                             sizeof optval);
1509                         if (error)
1510                                 break;
1511
1512                         switch (optval) {
1513                         case IP_PORTRANGE_DEFAULT:
1514                                 inp->inp_flags &= ~(INP_LOWPORT);
1515                                 inp->inp_flags &= ~(INP_HIGHPORT);
1516                                 break;
1517
1518                         case IP_PORTRANGE_HIGH:
1519                                 inp->inp_flags &= ~(INP_LOWPORT);
1520                                 inp->inp_flags |= INP_HIGHPORT;
1521                                 break;
1522
1523                         case IP_PORTRANGE_LOW:
1524                                 inp->inp_flags &= ~(INP_HIGHPORT);
1525                                 inp->inp_flags |= INP_LOWPORT;
1526                                 break;
1527
1528                         default:
1529                                 error = EINVAL;
1530                                 break;
1531                         }
1532                         break;
1533
1534 #if defined(IPSEC) || defined(FAST_IPSEC)
1535                 case IP_IPSEC_POLICY:
1536                 {
1537                         caddr_t req;
1538                         size_t len = 0;
1539                         int priv;
1540                         struct mbuf *m;
1541                         int optname;
1542
1543                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1544                                 break;
1545                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1546                                 break;
1547                         priv = (sopt->sopt_td != NULL &&
1548                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1549                         req = mtod(m, caddr_t);
1550                         len = m->m_len;
1551                         optname = sopt->sopt_name;
1552                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1553                         m_freem(m);
1554                         break;
1555                 }
1556 #endif /*IPSEC*/
1557
1558                 default:
1559                         error = ENOPROTOOPT;
1560                         break;
1561                 }
1562                 break;
1563
1564         case SOPT_GET:
1565                 switch (sopt->sopt_name) {
1566                 case IP_OPTIONS:
1567                 case IP_RETOPTS:
1568                         if (inp->inp_options)
1569                                 error = sooptcopyout(sopt,
1570                                                      mtod(inp->inp_options,
1571                                                           char *),
1572                                                      inp->inp_options->m_len);
1573                         else
1574                                 sopt->sopt_valsize = 0;
1575                         break;
1576
1577                 case IP_TOS:
1578                 case IP_TTL:
1579                 case IP_RECVOPTS:
1580                 case IP_RECVRETOPTS:
1581                 case IP_RECVDSTADDR:
1582                 case IP_RECVIF:
1583                 case IP_PORTRANGE:
1584                 case IP_FAITH:
1585                         switch (sopt->sopt_name) {
1586
1587                         case IP_TOS:
1588                                 optval = inp->inp_ip_tos;
1589                                 break;
1590
1591                         case IP_TTL:
1592                                 optval = inp->inp_ip_ttl;
1593                                 break;
1594
1595 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1596
1597                         case IP_RECVOPTS:
1598                                 optval = OPTBIT(INP_RECVOPTS);
1599                                 break;
1600
1601                         case IP_RECVRETOPTS:
1602                                 optval = OPTBIT(INP_RECVRETOPTS);
1603                                 break;
1604
1605                         case IP_RECVDSTADDR:
1606                                 optval = OPTBIT(INP_RECVDSTADDR);
1607                                 break;
1608
1609                         case IP_RECVIF:
1610                                 optval = OPTBIT(INP_RECVIF);
1611                                 break;
1612
1613                         case IP_PORTRANGE:
1614                                 if (inp->inp_flags & INP_HIGHPORT)
1615                                         optval = IP_PORTRANGE_HIGH;
1616                                 else if (inp->inp_flags & INP_LOWPORT)
1617                                         optval = IP_PORTRANGE_LOW;
1618                                 else
1619                                         optval = 0;
1620                                 break;
1621
1622                         case IP_FAITH:
1623                                 optval = OPTBIT(INP_FAITH);
1624                                 break;
1625                         }
1626                         error = sooptcopyout(sopt, &optval, sizeof optval);
1627                         break;
1628
1629                 case IP_MULTICAST_IF:
1630                 case IP_MULTICAST_VIF:
1631                 case IP_MULTICAST_TTL:
1632                 case IP_MULTICAST_LOOP:
1633                 case IP_ADD_MEMBERSHIP:
1634                 case IP_DROP_MEMBERSHIP:
1635                         error = ip_getmoptions(sopt, inp->inp_moptions);
1636                         break;
1637
1638 #if defined(IPSEC) || defined(FAST_IPSEC)
1639                 case IP_IPSEC_POLICY:
1640                 {
1641                         struct mbuf *m = NULL;
1642                         caddr_t req = NULL;
1643                         size_t len = 0;
1644
1645                         if (m != 0) {
1646                                 req = mtod(m, caddr_t);
1647                                 len = m->m_len;
1648                         }
1649                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1650                         if (error == 0)
1651                                 error = soopt_mcopyout(sopt, m); /* XXX */
1652                         if (error == 0)
1653                                 m_freem(m);
1654                         break;
1655                 }
1656 #endif /*IPSEC*/
1657
1658                 default:
1659                         error = ENOPROTOOPT;
1660                         break;
1661                 }
1662                 break;
1663         }
1664         return (error);
1665 }
1666
1667 /*
1668  * Set up IP options in pcb for insertion in output packets.
1669  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1670  * with destination address if source routed.
1671  */
1672 static int
1673 ip_pcbopts(optname, pcbopt, m)
1674         int optname;
1675         struct mbuf **pcbopt;
1676         struct mbuf *m;
1677 {
1678         int cnt, optlen;
1679         u_char *cp;
1680         u_char opt;
1681
1682         /* turn off any old options */
1683         if (*pcbopt)
1684                 (void)m_free(*pcbopt);
1685         *pcbopt = 0;
1686         if (m == (struct mbuf *)0 || m->m_len == 0) {
1687                 /*
1688                  * Only turning off any previous options.
1689                  */
1690                 if (m)
1691                         (void)m_free(m);
1692                 return (0);
1693         }
1694
1695         if (m->m_len % sizeof(int32_t))
1696                 goto bad;
1697         /*
1698          * IP first-hop destination address will be stored before
1699          * actual options; move other options back
1700          * and clear it when none present.
1701          */
1702         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1703                 goto bad;
1704         cnt = m->m_len;
1705         m->m_len += sizeof(struct in_addr);
1706         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1707         ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1708         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1709
1710         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1711                 opt = cp[IPOPT_OPTVAL];
1712                 if (opt == IPOPT_EOL)
1713                         break;
1714                 if (opt == IPOPT_NOP)
1715                         optlen = 1;
1716                 else {
1717                         if (cnt < IPOPT_OLEN + sizeof(*cp))
1718                                 goto bad;
1719                         optlen = cp[IPOPT_OLEN];
1720                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1721                                 goto bad;
1722                 }
1723                 switch (opt) {
1724
1725                 default:
1726                         break;
1727
1728                 case IPOPT_LSRR:
1729                 case IPOPT_SSRR:
1730                         /*
1731                          * user process specifies route as:
1732                          *      ->A->B->C->D
1733                          * D must be our final destination (but we can't
1734                          * check that since we may not have connected yet).
1735                          * A is first hop destination, which doesn't appear in
1736                          * actual IP option, but is stored before the options.
1737                          */
1738                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1739                                 goto bad;
1740                         m->m_len -= sizeof(struct in_addr);
1741                         cnt -= sizeof(struct in_addr);
1742                         optlen -= sizeof(struct in_addr);
1743                         cp[IPOPT_OLEN] = optlen;
1744                         /*
1745                          * Move first hop before start of options.
1746                          */
1747                         bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1748                             sizeof(struct in_addr));
1749                         /*
1750                          * Then copy rest of options back
1751                          * to close up the deleted entry.
1752                          */
1753                         ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1754                             sizeof(struct in_addr)),
1755                             (caddr_t)&cp[IPOPT_OFFSET+1],
1756                             (unsigned)cnt + sizeof(struct in_addr));
1757                         break;
1758                 }
1759         }
1760         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1761                 goto bad;
1762         *pcbopt = m;
1763         return (0);
1764
1765 bad:
1766         (void)m_free(m);
1767         return (EINVAL);
1768 }
1769
1770 /*
1771  * XXX
1772  * The whole multicast option thing needs to be re-thought.
1773  * Several of these options are equally applicable to non-multicast
1774  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1775  * standard option (IP_TTL).
1776  */
1777
1778 /*
1779  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1780  */
1781 static struct ifnet *
1782 ip_multicast_if(a, ifindexp)
1783         struct in_addr *a;
1784         int *ifindexp;
1785 {
1786         int ifindex;
1787         struct ifnet *ifp;
1788
1789         if (ifindexp)
1790                 *ifindexp = 0;
1791         if (ntohl(a->s_addr) >> 24 == 0) {
1792                 ifindex = ntohl(a->s_addr) & 0xffffff;
1793                 if (ifindex < 0 || if_index < ifindex)
1794                         return NULL;
1795                 ifp = ifindex2ifnet[ifindex];
1796                 if (ifindexp)
1797                         *ifindexp = ifindex;
1798         } else {
1799                 INADDR_TO_IFP(*a, ifp);
1800         }
1801         return ifp;
1802 }
1803
1804 /*
1805  * Set the IP multicast options in response to user setsockopt().
1806  */
1807 static int
1808 ip_setmoptions(sopt, imop)
1809         struct sockopt *sopt;
1810         struct ip_moptions **imop;
1811 {
1812         int error = 0;
1813         int i;
1814         struct in_addr addr;
1815         struct ip_mreq mreq;
1816         struct ifnet *ifp;
1817         struct ip_moptions *imo = *imop;
1818         struct route ro;
1819         struct sockaddr_in *dst;
1820         int ifindex;
1821         int s;
1822
1823         if (imo == NULL) {
1824                 /*
1825                  * No multicast option buffer attached to the pcb;
1826                  * allocate one and initialize to default values.
1827                  */
1828                 imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1829                     M_WAITOK);
1830
1831                 if (imo == NULL)
1832                         return (ENOBUFS);
1833                 *imop = imo;
1834                 imo->imo_multicast_ifp = NULL;
1835                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1836                 imo->imo_multicast_vif = -1;
1837                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1838                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1839                 imo->imo_num_memberships = 0;
1840         }
1841
1842         switch (sopt->sopt_name) {
1843         /* store an index number for the vif you wanna use in the send */
1844         case IP_MULTICAST_VIF:
1845                 if (legal_vif_num == 0) {
1846                         error = EOPNOTSUPP;
1847                         break;
1848                 }
1849                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1850                 if (error)
1851                         break;
1852                 if (!legal_vif_num(i) && (i != -1)) {
1853                         error = EINVAL;
1854                         break;
1855                 }
1856                 imo->imo_multicast_vif = i;
1857                 break;
1858
1859         case IP_MULTICAST_IF:
1860                 /*
1861                  * Select the interface for outgoing multicast packets.
1862                  */
1863                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1864                 if (error)
1865                         break;
1866                 /*
1867                  * INADDR_ANY is used to remove a previous selection.
1868                  * When no interface is selected, a default one is
1869                  * chosen every time a multicast packet is sent.
1870                  */
1871                 if (addr.s_addr == INADDR_ANY) {
1872                         imo->imo_multicast_ifp = NULL;
1873                         break;
1874                 }
1875                 /*
1876                  * The selected interface is identified by its local
1877                  * IP address.  Find the interface and confirm that
1878                  * it supports multicasting.
1879                  */
1880                 s = splimp();
1881                 ifp = ip_multicast_if(&addr, &ifindex);
1882                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1883                         splx(s);
1884                         error = EADDRNOTAVAIL;
1885                         break;
1886                 }
1887                 imo->imo_multicast_ifp = ifp;
1888                 if (ifindex)
1889                         imo->imo_multicast_addr = addr;
1890                 else
1891                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1892                 splx(s);
1893                 break;
1894
1895         case IP_MULTICAST_TTL:
1896                 /*
1897                  * Set the IP time-to-live for outgoing multicast packets.
1898                  * The original multicast API required a char argument,
1899                  * which is inconsistent with the rest of the socket API.
1900                  * We allow either a char or an int.
1901                  */
1902                 if (sopt->sopt_valsize == 1) {
1903                         u_char ttl;
1904                         error = sooptcopyin(sopt, &ttl, 1, 1);
1905                         if (error)
1906                                 break;
1907                         imo->imo_multicast_ttl = ttl;
1908                 } else {
1909                         u_int ttl;
1910                         error = sooptcopyin(sopt, &ttl, sizeof ttl,
1911                                             sizeof ttl);
1912                         if (error)
1913                                 break;
1914                         if (ttl > 255)
1915                                 error = EINVAL;
1916                         else
1917                                 imo->imo_multicast_ttl = ttl;
1918                 }
1919                 break;
1920
1921         case IP_MULTICAST_LOOP:
1922                 /*
1923                  * Set the loopback flag for outgoing multicast packets.
1924                  * Must be zero or one.  The original multicast API required a
1925                  * char argument, which is inconsistent with the rest
1926                  * of the socket API.  We allow either a char or an int.
1927                  */
1928                 if (sopt->sopt_valsize == 1) {
1929                         u_char loop;
1930                         error = sooptcopyin(sopt, &loop, 1, 1);
1931                         if (error)
1932                                 break;
1933                         imo->imo_multicast_loop = !!loop;
1934                 } else {
1935                         u_int loop;
1936                         error = sooptcopyin(sopt, &loop, sizeof loop,
1937                                             sizeof loop);
1938                         if (error)
1939                                 break;
1940                         imo->imo_multicast_loop = !!loop;
1941                 }
1942                 break;
1943
1944         case IP_ADD_MEMBERSHIP:
1945                 /*
1946                  * Add a multicast group membership.
1947                  * Group must be a valid IP multicast address.
1948                  */
1949                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1950                 if (error)
1951                         break;
1952
1953                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1954                         error = EINVAL;
1955                         break;
1956                 }
1957                 s = splimp();
1958                 /*
1959                  * If no interface address was provided, use the interface of
1960                  * the route to the given multicast address.
1961                  */
1962                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1963                         bzero((caddr_t)&ro, sizeof(ro));
1964                         dst = (struct sockaddr_in *)&ro.ro_dst;
1965                         dst->sin_len = sizeof(*dst);
1966                         dst->sin_family = AF_INET;
1967                         dst->sin_addr = mreq.imr_multiaddr;
1968                         rtalloc(&ro);
1969                         if (ro.ro_rt == NULL) {
1970                                 error = EADDRNOTAVAIL;
1971                                 splx(s);
1972                                 break;
1973                         }
1974                         ifp = ro.ro_rt->rt_ifp;
1975                         rtfree(ro.ro_rt);
1976                 }
1977                 else {
1978                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1979                 }
1980
1981                 /*
1982                  * See if we found an interface, and confirm that it
1983                  * supports multicast.
1984                  */
1985                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1986                         error = EADDRNOTAVAIL;
1987                         splx(s);
1988                         break;
1989                 }
1990                 /*
1991                  * See if the membership already exists or if all the
1992                  * membership slots are full.
1993                  */
1994                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1995                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1996                             imo->imo_membership[i]->inm_addr.s_addr
1997                                                 == mreq.imr_multiaddr.s_addr)
1998                                 break;
1999                 }
2000                 if (i < imo->imo_num_memberships) {
2001                         error = EADDRINUSE;
2002                         splx(s);
2003                         break;
2004                 }
2005                 if (i == IP_MAX_MEMBERSHIPS) {
2006                         error = ETOOMANYREFS;
2007                         splx(s);
2008                         break;
2009                 }
2010                 /*
2011                  * Everything looks good; add a new record to the multicast
2012                  * address list for the given interface.
2013                  */
2014                 if ((imo->imo_membership[i] =
2015                     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2016                         error = ENOBUFS;
2017                         splx(s);
2018                         break;
2019                 }
2020                 ++imo->imo_num_memberships;
2021                 splx(s);
2022                 break;
2023
2024         case IP_DROP_MEMBERSHIP:
2025                 /*
2026                  * Drop a multicast group membership.
2027                  * Group must be a valid IP multicast address.
2028                  */
2029                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2030                 if (error)
2031                         break;
2032
2033                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2034                         error = EINVAL;
2035                         break;
2036                 }
2037
2038                 s = splimp();
2039                 /*
2040                  * If an interface address was specified, get a pointer
2041                  * to its ifnet structure.
2042                  */
2043                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2044                         ifp = NULL;
2045                 else {
2046                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2047                         if (ifp == NULL) {
2048                                 error = EADDRNOTAVAIL;
2049                                 splx(s);
2050                                 break;
2051                         }
2052                 }
2053                 /*
2054                  * Find the membership in the membership array.
2055                  */
2056                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2057                         if ((ifp == NULL ||
2058                              imo->imo_membership[i]->inm_ifp == ifp) &&
2059                              imo->imo_membership[i]->inm_addr.s_addr ==
2060                              mreq.imr_multiaddr.s_addr)
2061                                 break;
2062                 }
2063                 if (i == imo->imo_num_memberships) {
2064                         error = EADDRNOTAVAIL;
2065                         splx(s);
2066                         break;
2067                 }
2068                 /*
2069                  * Give up the multicast address record to which the
2070                  * membership points.
2071                  */
2072                 in_delmulti(imo->imo_membership[i]);
2073                 /*
2074                  * Remove the gap in the membership array.
2075                  */
2076                 for (++i; i < imo->imo_num_memberships; ++i)
2077                         imo->imo_membership[i-1] = imo->imo_membership[i];
2078                 --imo->imo_num_memberships;
2079                 splx(s);
2080                 break;
2081
2082         default:
2083                 error = EOPNOTSUPP;
2084                 break;
2085         }
2086
2087         /*
2088          * If all options have default values, no need to keep the mbuf.
2089          */
2090         if (imo->imo_multicast_ifp == NULL &&
2091             imo->imo_multicast_vif == -1 &&
2092             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2093             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2094             imo->imo_num_memberships == 0) {
2095                 free(*imop, M_IPMOPTS);
2096                 *imop = NULL;
2097         }
2098
2099         return (error);
2100 }
2101
2102 /*
2103  * Return the IP multicast options in response to user getsockopt().
2104  */
2105 static int
2106 ip_getmoptions(sopt, imo)
2107         struct sockopt *sopt;
2108         struct ip_moptions *imo;
2109 {
2110         struct in_addr addr;
2111         struct in_ifaddr *ia;
2112         int error, optval;
2113         u_char coptval;
2114
2115         error = 0;
2116         switch (sopt->sopt_name) {
2117         case IP_MULTICAST_VIF:
2118                 if (imo != NULL)
2119                         optval = imo->imo_multicast_vif;
2120                 else
2121                         optval = -1;
2122                 error = sooptcopyout(sopt, &optval, sizeof optval);
2123                 break;
2124
2125         case IP_MULTICAST_IF:
2126                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2127                         addr.s_addr = INADDR_ANY;
2128                 else if (imo->imo_multicast_addr.s_addr) {
2129                         /* return the value user has set */
2130                         addr = imo->imo_multicast_addr;
2131                 } else {
2132                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2133                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2134                                 : IA_SIN(ia)->sin_addr.s_addr;
2135                 }
2136                 error = sooptcopyout(sopt, &addr, sizeof addr);
2137                 break;
2138
2139         case IP_MULTICAST_TTL:
2140                 if (imo == 0)
2141                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2142                 else
2143                         optval = coptval = imo->imo_multicast_ttl;
2144                 if (sopt->sopt_valsize == 1)
2145                         error = sooptcopyout(sopt, &coptval, 1);
2146                 else
2147                         error = sooptcopyout(sopt, &optval, sizeof optval);
2148                 break;
2149
2150         case IP_MULTICAST_LOOP:
2151                 if (imo == 0)
2152                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2153                 else
2154                         optval = coptval = imo->imo_multicast_loop;
2155                 if (sopt->sopt_valsize == 1)
2156                         error = sooptcopyout(sopt, &coptval, 1);
2157                 else
2158                         error = sooptcopyout(sopt, &optval, sizeof optval);
2159                 break;
2160
2161         default:
2162                 error = ENOPROTOOPT;
2163                 break;
2164         }
2165         return (error);
2166 }
2167
2168 /*
2169  * Discard the IP multicast options.
2170  */
2171 void
2172 ip_freemoptions(imo)
2173         struct ip_moptions *imo;
2174 {
2175         int i;
2176
2177         if (imo != NULL) {
2178                 for (i = 0; i < imo->imo_num_memberships; ++i)
2179                         in_delmulti(imo->imo_membership[i]);
2180                 free(imo, M_IPMOPTS);
2181         }
2182 }
2183
2184 /*
2185  * Routine called from ip_output() to loop back a copy of an IP multicast
2186  * packet to the input queue of a specified interface.  Note that this
2187  * calls the output routine of the loopback "driver", but with an interface
2188  * pointer that might NOT be a loopback interface -- evil, but easier than
2189  * replicating that code here.
2190  */
2191 static void
2192 ip_mloopback(ifp, m, dst, hlen)
2193         struct ifnet *ifp;
2194         struct mbuf *m;
2195         struct sockaddr_in *dst;
2196         int hlen;
2197 {
2198         struct ip *ip;
2199         struct mbuf *copym;
2200
2201         copym = m_copy(m, 0, M_COPYALL);
2202         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2203                 copym = m_pullup(copym, hlen);
2204         if (copym != NULL) {
2205                 /*
2206                  * We don't bother to fragment if the IP length is greater
2207                  * than the interface's MTU.  Can this possibly matter?
2208                  */
2209                 ip = mtod(copym, struct ip *);
2210                 ip->ip_len = htons(ip->ip_len);
2211                 ip->ip_off = htons(ip->ip_off);
2212                 ip->ip_sum = 0;
2213                 if (ip->ip_vhl == IP_VHL_BORING) {
2214                         ip->ip_sum = in_cksum_hdr(ip);
2215                 } else {
2216                         ip->ip_sum = in_cksum(copym, hlen);
2217                 }
2218                 /*
2219                  * NB:
2220                  * It's not clear whether there are any lingering
2221                  * reentrancy problems in other areas which might
2222                  * be exposed by using ip_input directly (in
2223                  * particular, everything which modifies the packet
2224                  * in-place).  Yet another option is using the
2225                  * protosw directly to deliver the looped back
2226                  * packet.  For the moment, we'll err on the side
2227                  * of safety by using if_simloop().
2228                  */
2229 #if 1 /* XXX */
2230                 if (dst->sin_family != AF_INET) {
2231                         printf("ip_mloopback: bad address family %d\n",
2232                                                 dst->sin_family);
2233                         dst->sin_family = AF_INET;
2234                 }
2235 #endif
2236
2237 #ifdef notdef
2238                 copym->m_pkthdr.rcvif = ifp;
2239                 ip_input(copym);
2240 #else
2241                 /* if the checksum hasn't been computed, mark it as valid */
2242                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2243                         copym->m_pkthdr.csum_flags |=
2244                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2245                         copym->m_pkthdr.csum_data = 0xffff;
2246                 }
2247                 if_simloop(ifp, copym, dst->sin_family, 0);
2248 #endif
2249         }
2250 }