sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.51 2008/08/26 12:05:10 sephe Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42 #include "opt_mpls.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/proc.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/thread2.h>
  55 #include <sys/in_cksum.h>
  56
  57 #include <net/if.h>
  58 #include <net/netisr.h>
  59 #include <net/pfil.h>
  60 #include <net/route.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/in_systm.h>
  64 #include <netinet/ip.h>
  65 #include <netinet/in_pcb.h>
  66 #include <netinet/in_var.h>
  67 #include <netinet/ip_var.h>
  68
  69 #include <netproto/mpls/mpls_var.h>
  70
  71 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  72
  73 #ifdef IPSEC
  74 #include <netinet6/ipsec.h>
  75 #include <netproto/key/key.h>
  76 #ifdef IPSEC_DEBUG
  77 #include <netproto/key/key_debug.h>
  78 #else
  79 #define KEYDEBUG(lev,arg)
  80 #endif
  81 #endif /*IPSEC*/
  82
  83 #ifdef FAST_IPSEC
  84 #include <netproto/ipsec/ipsec.h>
  85 #include <netproto/ipsec/xform.h>
  86 #include <netproto/ipsec/key.h>
  87 #endif /*FAST_IPSEC*/
  88
  89 #include <net/ipfw/ip_fw.h>
  90 #include <net/dummynet/ip_dummynet.h>
  91
  92 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  93                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  94                                   (ntohl(a.s_addr)>>16)&0xFF,\
  95                                   (ntohl(a.s_addr)>>8)&0xFF,\
  96                                   (ntohl(a.s_addr))&0xFF, y);
  97
  98 u_short ip_id;
  99
 100 #ifdef MBUF_STRESS_TEST
 101 int mbuf_frag_size = 0;
 102 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 103         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 104 #endif
 105
 106 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 107 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 108 static void     ip_mloopback
 109         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 110 static int      ip_getmoptions
 111         (struct sockopt *, struct ip_moptions *);
 112 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 113 static int      ip_setmoptions
 114         (struct sockopt *, struct ip_moptions **);
 115
 116 int     ip_optcopy(struct ip *, struct ip *);
 117
 118
 119 extern  struct protosw inetsw[];
 120
 121 /*
 122  * IP output.  The packet in mbuf chain m contains a skeletal IP
 123  * header (with len, off, ttl, proto, tos, src, dst).
 124  * The mbuf chain containing the packet will be freed.
 125  * The mbuf opt, if present, will not be freed.
 126  */
 127 int
 128 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 129           int flags, struct ip_moptions *imo, struct inpcb *inp)
 130 {
 131         struct ip *ip;
 132         struct ifnet *ifp = NULL;       /* keep compiler happy */
 133         struct mbuf *m;
 134         int hlen = sizeof(struct ip);
 135         int len, off, error = 0;
 136         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 137         struct in_ifaddr *ia = NULL;
 138         int isbroadcast, sw_csum;
 139         struct in_addr pkt_dst;
 140         struct route iproute;
 141         struct m_tag *dn_mtag = NULL, *mtag;
 142 #ifdef IPSEC
 143         struct secpolicy *sp = NULL;
 144         struct socket *so = inp ? inp->inp_socket : NULL;
 145 #endif
 146 #ifdef FAST_IPSEC
 147         struct secpolicy *sp = NULL;
 148         struct tdb_ident *tdbi;
 149 #endif /* FAST_IPSEC */
 150         struct ip_fw_args args;
 151         struct sockaddr_in *next_hop = NULL;
 152         int src_was_INADDR_ANY = 0;     /* as the name says... */
 153
 154         args.eh = NULL;
 155         args.rule = NULL;
 156
 157         m = m0;
 158         M_ASSERTPKTHDR(m);
 159
 160         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 161                 /* Next hop */
 162                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 163                 KKASSERT(mtag != NULL);
 164                 next_hop = m_tag_data(mtag);
 165         }
 166
 167         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 168                 struct dn_pkt *dn_pkt;
 169
 170                 /* Extract info from dummynet tag */
 171                 dn_mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 172                 KKASSERT(dn_mtag != NULL);
 173                 dn_pkt = m_tag_data(dn_mtag);
 174
 175                 /*
 176                  * The packet was already tagged, so part of the
 177                  * processing was already done, and we need to go down.
 178                  * Get parameters from the tag.
 179                  */
 180                 args.rule = dn_pkt->dn_priv;
 181                 KKASSERT(args.rule != NULL);
 182                 opt = NULL;
 183                 ro = &dn_pkt->ro;
 184                 imo = NULL;
 185                 dst = dn_pkt->dn_dst;
 186                 ifp = dn_pkt->ifp;
 187                 flags = dn_pkt->flags;
 188
 189                 /*
 190                  * Don't delete the dummynet tag here, just unlink it,
 191                  * since some local variables (like 'ro' and 'dst') are
 192                  * still referencing certain parts of it.
 193                  * The dummynet tag will be freed at the end of the
 194                  * output process.
 195                  */
 196                 m_tag_unlink(m, dn_mtag);
 197                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 198         }
 199
 200         if (ro == NULL) {
 201                 ro = &iproute;
 202                 bzero(ro, sizeof *ro);
 203         }
 204
 205         if (args.rule != NULL) {        /* dummynet already saw us */
 206                 ip = mtod(m, struct ip *);
 207                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 208                 if (ro->ro_rt)
 209                         ia = ifatoia(ro->ro_rt->rt_ifa);
 210                 goto sendit;
 211         }
 212
 213         if (opt) {
 214                 len = 0;
 215                 m = ip_insertoptions(m, opt, &len);
 216                 if (len != 0)
 217                         hlen = len;
 218         }
 219         ip = mtod(m, struct ip *);
 220         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 221
 222         /*
 223          * Fill in IP header.
 224          */
 225         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 226                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 227                 ip->ip_off &= IP_DF;
 228                 ip->ip_id = ip_newid();
 229                 ipstat.ips_localout++;
 230         } else {
 231                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 232         }
 233
 234         dst = (struct sockaddr_in *)&ro->ro_dst;
 235         /*
 236          * If there is a cached route,
 237          * check that it is to the same destination
 238          * and is still up.  If not, free it and try again.
 239          * The address family should also be checked in case of sharing the
 240          * cache with IPv6.
 241          */
 242         if (ro->ro_rt &&
 243             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 244              dst->sin_family != AF_INET ||
 245              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 246                 rtfree(ro->ro_rt);
 247                 ro->ro_rt = (struct rtentry *)NULL;
 248         }
 249         if (ro->ro_rt == NULL) {
 250                 bzero(dst, sizeof *dst);
 251                 dst->sin_family = AF_INET;
 252                 dst->sin_len = sizeof *dst;
 253                 dst->sin_addr = pkt_dst;
 254         }
 255         /*
 256          * If routing to interface only,
 257          * short circuit routing lookup.
 258          */
 259         if (flags & IP_ROUTETOIF) {
 260                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 261                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 262                         ipstat.ips_noroute++;
 263                         error = ENETUNREACH;
 264                         goto bad;
 265                 }
 266                 ifp = ia->ia_ifp;
 267                 ip->ip_ttl = 1;
 268                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 269         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 270                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 271                 /*
 272                  * Bypass the normal routing lookup for multicast
 273                  * packets if the interface is specified.
 274                  */
 275                 ifp = imo->imo_multicast_ifp;
 276                 ia = IFP_TO_IA(ifp);
 277                 isbroadcast = 0;        /* fool gcc */
 278         } else {
 279                 /*
 280                  * If this is the case, we probably don't want to allocate
 281                  * a protocol-cloned route since we didn't get one from the
 282                  * ULP.  This lets TCP do its thing, while not burdening
 283                  * forwarding or ICMP with the overhead of cloning a route.
 284                  * Of course, we still want to do any cloning requested by
 285                  * the link layer, as this is probably required in all cases
 286                  * for correct operation (as it is for ARP).
 287                  */
 288                 if (ro->ro_rt == NULL)
 289                         rtalloc_ign(ro, RTF_PRCLONING);
 290                 if (ro->ro_rt == NULL) {
 291                         ipstat.ips_noroute++;
 292                         error = EHOSTUNREACH;
 293                         goto bad;
 294                 }
 295                 ia = ifatoia(ro->ro_rt->rt_ifa);
 296                 ifp = ro->ro_rt->rt_ifp;
 297                 ro->ro_rt->rt_use++;
 298                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 299                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 300                 if (ro->ro_rt->rt_flags & RTF_HOST)
 301                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 302                 else
 303                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 304         }
 305         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 306                 struct in_multi *inm;
 307
 308                 m->m_flags |= M_MCAST;
 309                 /*
 310                  * IP destination address is multicast.  Make sure "dst"
 311                  * still points to the address in "ro".  (It may have been
 312                  * changed to point to a gateway address, above.)
 313                  */
 314                 dst = (struct sockaddr_in *)&ro->ro_dst;
 315                 /*
 316                  * See if the caller provided any multicast options
 317                  */
 318                 if (imo != NULL) {
 319                         ip->ip_ttl = imo->imo_multicast_ttl;
 320                         if (imo->imo_multicast_vif != -1)
 321                                 ip->ip_src.s_addr =
 322                                     ip_mcast_src ?
 323                                     ip_mcast_src(imo->imo_multicast_vif) :
 324                                     INADDR_ANY;
 325                 } else
 326                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 327                 /*
 328                  * Confirm that the outgoing interface supports multicast.
 329                  */
 330                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 331                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 332                                 ipstat.ips_noroute++;
 333                                 error = ENETUNREACH;
 334                                 goto bad;
 335                         }
 336                 }
 337                 /*
 338                  * If source address not specified yet, use address
 339                  * of outgoing interface.
 340                  */
 341                 if (ip->ip_src.s_addr == INADDR_ANY) {
 342                         /* Interface may have no addresses. */
 343                         if (ia != NULL)
 344                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 345                 }
 346
 347                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 348                 if (inm != NULL &&
 349                    (imo == NULL || imo->imo_multicast_loop)) {
 350                         /*
 351                          * If we belong to the destination multicast group
 352                          * on the outgoing interface, and the caller did not
 353                          * forbid loopback, loop back a copy.
 354                          */
 355                         ip_mloopback(ifp, m, dst, hlen);
 356                 }
 357                 else {
 358                         /*
 359                          * If we are acting as a multicast router, perform
 360                          * multicast forwarding as if the packet had just
 361                          * arrived on the interface to which we are about
 362                          * to send.  The multicast forwarding function
 363                          * recursively calls this function, using the
 364                          * IP_FORWARDING flag to prevent infinite recursion.
 365                          *
 366                          * Multicasts that are looped back by ip_mloopback(),
 367                          * above, will be forwarded by the ip_input() routine,
 368                          * if necessary.
 369                          */
 370                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 371                                 /*
 372                                  * If rsvp daemon is not running, do not
 373                                  * set ip_moptions. This ensures that the packet
 374                                  * is multicast and not just sent down one link
 375                                  * as prescribed by rsvpd.
 376                                  */
 377                                 if (!rsvp_on)
 378                                         imo = NULL;
 379                                 if (ip_mforward &&
 380                                     ip_mforward(ip, ifp, m, imo) != 0) {
 381                                         m_freem(m);
 382                                         goto done;
 383                                 }
 384                         }
 385                 }
 386
 387                 /*
 388                  * Multicasts with a time-to-live of zero may be looped-
 389                  * back, above, but must not be transmitted on a network.
 390                  * Also, multicasts addressed to the loopback interface
 391                  * are not sent -- the above call to ip_mloopback() will
 392                  * loop back a copy if this host actually belongs to the
 393                  * destination group on the loopback interface.
 394                  */
 395                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 396                         m_freem(m);
 397                         goto done;
 398                 }
 399
 400                 goto sendit;
 401         } else {
 402                 m->m_flags &= ~M_MCAST;
 403         }
 404 #ifndef notdef
 405         /*
 406          * If the source address is not specified yet, use the address
 407          * of the outoing interface. In case, keep note we did that, so
 408          * if the the firewall changes the next-hop causing the output
 409          * interface to change, we can fix that.
 410          */
 411         if (ip->ip_src.s_addr == INADDR_ANY) {
 412                 /* Interface may have no addresses. */
 413                 if (ia != NULL) {
 414                         ip->ip_src = IA_SIN(ia)->sin_addr;
 415                         src_was_INADDR_ANY = 1;
 416                 }
 417         }
 418 #endif /* notdef */
 419 #ifdef ALTQ
 420         /*
 421          * Disable packet drop hack.
 422          * Packetdrop should be done by queueing.
 423          */
 424 #else /* !ALTQ */
 425         /*
 426          * Verify that we have any chance at all of being able to queue
 427          *      the packet or packet fragments
 428          */
 429         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 430                 ifp->if_snd.ifq_maxlen) {
 431                         error = ENOBUFS;
 432                         ipstat.ips_odropped++;
 433                         goto bad;
 434         }
 435 #endif /* !ALTQ */
 436
 437         /*
 438          * Look for broadcast address and
 439          * verify user is allowed to send
 440          * such a packet.
 441          */
 442         if (isbroadcast) {
 443                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 444                         error = EADDRNOTAVAIL;
 445                         goto bad;
 446                 }
 447                 if (!(flags & IP_ALLOWBROADCAST)) {
 448                         error = EACCES;
 449                         goto bad;
 450                 }
 451                 /* don't allow broadcast messages to be fragmented */
 452                 if (ip->ip_len > ifp->if_mtu) {
 453                         error = EMSGSIZE;
 454                         goto bad;
 455                 }
 456                 m->m_flags |= M_BCAST;
 457         } else {
 458                 m->m_flags &= ~M_BCAST;
 459         }
 460
 461 sendit:
 462 #ifdef IPSEC
 463         /* get SP for this packet */
 464         if (so == NULL)
 465                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 466         else
 467                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 468
 469         if (sp == NULL) {
 470                 ipsecstat.out_inval++;
 471                 goto bad;
 472         }
 473
 474         error = 0;
 475
 476         /* check policy */
 477         switch (sp->policy) {
 478         case IPSEC_POLICY_DISCARD:
 479                 /*
 480                  * This packet is just discarded.
 481                  */
 482                 ipsecstat.out_polvio++;
 483                 goto bad;
 484
 485         case IPSEC_POLICY_BYPASS:
 486         case IPSEC_POLICY_NONE:
 487                 /* no need to do IPsec. */
 488                 goto skip_ipsec;
 489
 490         case IPSEC_POLICY_IPSEC:
 491                 if (sp->req == NULL) {
 492                         /* acquire a policy */
 493                         error = key_spdacquire(sp);
 494                         goto bad;
 495                 }
 496                 break;
 497
 498         case IPSEC_POLICY_ENTRUST:
 499         default:
 500                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 501         }
 502     {
 503         struct ipsec_output_state state;
 504         bzero(&state, sizeof state);
 505         state.m = m;
 506         if (flags & IP_ROUTETOIF) {
 507                 state.ro = &iproute;
 508                 bzero(&iproute, sizeof iproute);
 509         } else
 510                 state.ro = ro;
 511         state.dst = (struct sockaddr *)dst;
 512
 513         ip->ip_sum = 0;
 514
 515         /*
 516          * XXX
 517          * delayed checksums are not currently compatible with IPsec
 518          */
 519         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 520                 in_delayed_cksum(m);
 521                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 522         }
 523
 524         ip->ip_len = htons(ip->ip_len);
 525         ip->ip_off = htons(ip->ip_off);
 526
 527         error = ipsec4_output(&state, sp, flags);
 528
 529         m = state.m;
 530         if (flags & IP_ROUTETOIF) {
 531                 /*
 532                  * if we have tunnel mode SA, we may need to ignore
 533                  * IP_ROUTETOIF.
 534                  */
 535                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 536                         flags &= ~IP_ROUTETOIF;
 537                         ro = state.ro;
 538                 }
 539         } else
 540                 ro = state.ro;
 541         dst = (struct sockaddr_in *)state.dst;
 542         if (error) {
 543                 /* mbuf is already reclaimed in ipsec4_output. */
 544                 m0 = NULL;
 545                 switch (error) {
 546                 case EHOSTUNREACH:
 547                 case ENETUNREACH:
 548                 case EMSGSIZE:
 549                 case ENOBUFS:
 550                 case ENOMEM:
 551                         break;
 552                 default:
 553                         kprintf("ip4_output (ipsec): error code %d\n", error);
 554                         /*fall through*/
 555                 case ENOENT:
 556                         /* don't show these error codes to the user */
 557                         error = 0;
 558                         break;
 559                 }
 560                 goto bad;
 561         }
 562     }
 563
 564         /* be sure to update variables that are affected by ipsec4_output() */
 565         ip = mtod(m, struct ip *);
 566 #ifdef _IP_VHL
 567         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 568 #else
 569         hlen = ip->ip_hl << 2;
 570 #endif
 571         if (ro->ro_rt == NULL) {
 572                 if (!(flags & IP_ROUTETOIF)) {
 573                         kprintf("ip_output: "
 574                                 "can't update route after IPsec processing\n");
 575                         error = EHOSTUNREACH;   /*XXX*/
 576                         goto bad;
 577                 }
 578         } else {
 579                 ia = ifatoia(ro->ro_rt->rt_ifa);
 580                 ifp = ro->ro_rt->rt_ifp;
 581         }
 582
 583         /* make it flipped, again. */
 584         ip->ip_len = ntohs(ip->ip_len);
 585         ip->ip_off = ntohs(ip->ip_off);
 586 skip_ipsec:
 587 #endif /*IPSEC*/
 588 #ifdef FAST_IPSEC
 589         /*
 590          * Check the security policy (SP) for the packet and, if
 591          * required, do IPsec-related processing.  There are two
 592          * cases here; the first time a packet is sent through
 593          * it will be untagged and handled by ipsec4_checkpolicy.
 594          * If the packet is resubmitted to ip_output (e.g. after
 595          * AH, ESP, etc. processing), there will be a tag to bypass
 596          * the lookup and related policy checking.
 597          */
 598         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 599         crit_enter();
 600         if (mtag != NULL) {
 601                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 602                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 603                 if (sp == NULL)
 604                         error = -EINVAL;        /* force silent drop */
 605                 m_tag_delete(m, mtag);
 606         } else {
 607                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 608                                         &error, inp);
 609         }
 610         /*
 611          * There are four return cases:
 612          *    sp != NULL                    apply IPsec policy
 613          *    sp == NULL, error == 0        no IPsec handling needed
 614          *    sp == NULL, error == -EINVAL  discard packet w/o error
 615          *    sp == NULL, error != 0        discard packet, report error
 616          */
 617         if (sp != NULL) {
 618                 /* Loop detection, check if ipsec processing already done */
 619                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 620                 for (mtag = m_tag_first(m); mtag != NULL;
 621                      mtag = m_tag_next(m, mtag)) {
 622                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 623                                 continue;
 624                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 625                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 626                                 continue;
 627                         /*
 628                          * Check if policy has an SA associated with it.
 629                          * This can happen when an SP has yet to acquire
 630                          * an SA; e.g. on first reference.  If it occurs,
 631                          * then we let ipsec4_process_packet do its thing.
 632                          */
 633                         if (sp->req->sav == NULL)
 634                                 break;
 635                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 636                         if (tdbi->spi == sp->req->sav->spi &&
 637                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 638                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 639                                  sizeof(union sockaddr_union)) == 0) {
 640                                 /*
 641                                  * No IPsec processing is needed, free
 642                                  * reference to SP.
 643                                  *
 644                                  * NB: null pointer to avoid free at
 645                                  *     done: below.
 646                                  */
 647                                 KEY_FREESP(&sp), sp = NULL;
 648                                 crit_exit();
 649                                 goto spd_done;
 650                         }
 651                 }
 652
 653                 /*
 654                  * Do delayed checksums now because we send before
 655                  * this is done in the normal processing path.
 656                  */
 657                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 658                         in_delayed_cksum(m);
 659                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 660                 }
 661
 662                 ip->ip_len = htons(ip->ip_len);
 663                 ip->ip_off = htons(ip->ip_off);
 664
 665                 /* NB: callee frees mbuf */
 666                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 667                 /*
 668                  * Preserve KAME behaviour: ENOENT can be returned
 669                  * when an SA acquire is in progress.  Don't propagate
 670                  * this to user-level; it confuses applications.
 671                  *
 672                  * XXX this will go away when the SADB is redone.
 673                  */
 674                 if (error == ENOENT)
 675                         error = 0;
 676                 crit_exit();
 677                 goto done;
 678         } else {
 679                 crit_exit();
 680
 681                 if (error != 0) {
 682                         /*
 683                          * Hack: -EINVAL is used to signal that a packet
 684                          * should be silently discarded.  This is typically
 685                          * because we asked key management for an SA and
 686                          * it was delayed (e.g. kicked up to IKE).
 687                          */
 688                         if (error == -EINVAL)
 689                                 error = 0;
 690                         goto bad;
 691                 } else {
 692                         /* No IPsec processing for this packet. */
 693                 }
 694 #ifdef notyet
 695                 /*
 696                  * If deferred crypto processing is needed, check that
 697                  * the interface supports it.
 698                  */
 699                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 700                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 701                         /* notify IPsec to do its own crypto */
 702                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 703                         error = EHOSTUNREACH;
 704                         goto bad;
 705                 }
 706 #endif
 707         }
 708 spd_done:
 709 #endif /* FAST_IPSEC */
 710         /*
 711          * IpHack's section.
 712          * - Xlate: translate packet's addr/port (NAT).
 713          * - Firewall: deny/allow/etc.
 714          * - Wrap: fake packet's addr/port <unimpl.>
 715          * - Encapsulate: put it in another IP and send out. <unimp.>
 716          */
 717
 718         /*
 719          * Run through list of hooks for output packets.
 720          */
 721         if (pfil_has_hooks(&inet_pfil_hook)) {
 722                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 723                 if (error != 0 || m == NULL)
 724                         goto done;
 725                 ip = mtod(m, struct ip *);
 726         }
 727
 728         /*
 729          * Check with the firewall...
 730          * but not if we are already being fwd'd from a firewall.
 731          */
 732         if (fw_enable && IPFW_LOADED && !next_hop) {
 733                 struct sockaddr_in *old = dst;
 734
 735                 args.m = m;
 736                 args.oif = ifp;
 737                 off = ip_fw_chk_ptr(&args);
 738                 m = args.m;
 739
 740                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 741                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 742                         KKASSERT(mtag != NULL);
 743                         next_hop = m_tag_data(mtag);
 744                         dst = next_hop;
 745                 }
 746
 747                 /*
 748                  * On return we must do the following:
 749                  * m == NULL    -> drop the pkt (old interface, deprecated)
 750                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 751                  * 1<=off<= 0xffff              -> DIVERT
 752                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 753                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 754                  * dst != old                   -> IPFIREWALL_FORWARD
 755                  * off==0, dst==old             -> accept
 756                  * If some of the above modules are not compiled in, then
 757                  * we should't have to check the corresponding condition
 758                  * (because the ipfw control socket should not accept
 759                  * unsupported rules), but better play safe and drop
 760                  * packets in case of doubt.
 761                  */
 762                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 763                         if (m)
 764                                 m_freem(m);
 765                         error = EACCES;
 766                         goto done;
 767                 }
 768                 ip = mtod(m, struct ip *);
 769                 if (off == 0 && dst == old)             /* common case */
 770                         goto pass;
 771                 if (off & IP_FW_PORT_DYNT_FLAG) {
 772                         /*
 773                          * pass the pkt to dummynet. Need to include
 774                          * pipe number, m, ifp, ro, dst because these are
 775                          * not recomputed in the next pass.
 776                          * All other parameters have been already used and
 777                          * so they are not needed anymore.
 778                          * XXX note: if the ifp or ro entry are deleted
 779                          * while a pkt is in dummynet, we are in trouble!
 780                          */
 781                         args.ro = ro;
 782                         args.dst = dst;
 783                         args.flags = flags;
 784
 785                         error = 0;
 786                         ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
 787                         goto done;
 788                 }
 789 #ifdef IPDIVERT
 790                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
 791                         struct mbuf *clone = NULL;
 792
 793                         /* Clone packet if we're doing a 'tee' */
 794                         if ((off & IP_FW_PORT_TEE_FLAG))
 795                                 clone = m_dup(m, MB_DONTWAIT);
 796
 797                         /*
 798                          * XXX
 799                          * delayed checksums are not currently compatible
 800                          * with divert sockets.
 801                          */
 802                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 803                                 in_delayed_cksum(m);
 804                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 805                         }
 806
 807                         /* Restore packet header fields to original values */
 808                         ip->ip_len = htons(ip->ip_len);
 809                         ip->ip_off = htons(ip->ip_off);
 810
 811                         /* Deliver packet to divert input routine */
 812                         divert_packet(m, 0, off & 0xffff);
 813
 814                         /* If 'tee', continue with original packet */
 815                         if (clone != NULL) {
 816                                 m = clone;
 817                                 ip = mtod(m, struct ip *);
 818                                 goto pass;
 819                         }
 820                         goto done;
 821                 }
 822 #endif
 823
 824                 /* IPFIREWALL_FORWARD */
 825                 /*
 826                  * Check dst to make sure it is directly reachable on the
 827                  * interface we previously thought it was.
 828                  * If it isn't (which may be likely in some situations) we have
 829                  * to re-route it (ie, find a route for the next-hop and the
 830                  * associated interface) and set them here. This is nested
 831                  * forwarding which in most cases is undesirable, except where
 832                  * such control is nigh impossible. So we do it here.
 833                  * And I'm babbling.
 834                  */
 835                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 836 #if 0
 837                         /*
 838                          * XXX To improve readability, this block should be
 839                          * changed into a function call as below:
 840                          */
 841                         error = ip_ipforward(&m, &dst, &ifp);
 842                         if (error)
 843                                 goto bad;
 844                         if (m == NULL) /* ip_input consumed the mbuf */
 845                                 goto done;
 846 #else
 847                         struct in_ifaddr *ia;
 848                         struct in_ifaddr_container *iac;
 849
 850                         /*
 851                          * XXX sro_fwd below is static, and a pointer
 852                          * to it gets passed to routines downstream.
 853                          * This could have surprisingly bad results in
 854                          * practice, because its content is overwritten
 855                          * by subsequent packets.
 856                          */
 857                         /* There must be a better way to do this next line... */
 858                         static struct route sro_fwd;
 859                         struct route *ro_fwd = &sro_fwd;
 860
 861 #if 0
 862                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 863                             dst->sin_addr, "\n");
 864 #endif
 865
 866                         /*
 867                          * We need to figure out if we have been forwarded
 868                          * to a local socket. If so, then we should somehow
 869                          * "loop back" to ip_input, and get directed to the
 870                          * PCB as if we had received this packet. This is
 871                          * because it may be dificult to identify the packets
 872                          * you want to forward until they are being output
 873                          * and have selected an interface. (e.g. locally
 874                          * initiated packets) If we used the loopback inteface,
 875                          * we would not be able to control what happens
 876                          * as the packet runs through ip_input() as
 877                          * it is done through a ISR.
 878                          */
 879                         ia = NULL;
 880                         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr),
 881                                      ia_hash) {
 882                                 /*
 883                                  * If the addr to forward to is one
 884                                  * of ours, we pretend to
 885                                  * be the destination for this packet.
 886                                  */
 887                                 if (IA_SIN(iac->ia)->sin_addr.s_addr ==
 888                                     dst->sin_addr.s_addr) {
 889                                         ia = iac->ia;
 890                                         break;
 891                                 }
 892                         }
 893                         if (ia != NULL) {    /* tell ip_input "dont filter" */
 894                                 if (m->m_pkthdr.rcvif == NULL)
 895                                         m->m_pkthdr.rcvif = ifunit("lo0");
 896                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 897                                         m->m_pkthdr.csum_flags |=
 898                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 899                                         m->m_pkthdr.csum_data = 0xffff;
 900                                 }
 901                                 m->m_pkthdr.csum_flags |=
 902                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 903                                 ip->ip_len = htons(ip->ip_len);
 904                                 ip->ip_off = htons(ip->ip_off);
 905                                 ip_input(m);
 906                                 goto done;
 907                         }
 908                         /* Some of the logic for this was nicked from above.
 909                          *
 910                          * This rewrites the cached route in a local PCB.
 911                          * Is this what we want to do?
 912                          */
 913                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
 914                         ro_fwd->ro_rt = NULL;
 915
 916                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 917                         if (ro_fwd->ro_rt == NULL) {
 918                                 ipstat.ips_noroute++;
 919                                 error = EHOSTUNREACH;
 920                                 goto bad;
 921                         }
 922
 923                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 924                         ifp = ro_fwd->ro_rt->rt_ifp;
 925                         ro_fwd->ro_rt->rt_use++;
 926                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 927                                 dst = (struct sockaddr_in *)
 928                                     ro_fwd->ro_rt->rt_gateway;
 929                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 930                                 isbroadcast =
 931                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 932                         else
 933                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 934                         if (ro->ro_rt != NULL)
 935                                 rtfree(ro->ro_rt);
 936                         ro->ro_rt = ro_fwd->ro_rt;
 937                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 938
 939 #endif  /* ... block to be put into a function */
 940                         /*
 941                          * If we added a default src ip earlier,
 942                          * which would have been gotten from the-then
 943                          * interface, do it again, from the new one.
 944                          */
 945                         if (src_was_INADDR_ANY)
 946                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 947                         goto pass ;
 948                 }
 949
 950                 /*
 951                  * if we get here, none of the above matches, and
 952                  * we have to drop the pkt
 953                  */
 954                 m_freem(m);
 955                 error = EACCES; /* not sure this is the right error msg */
 956                 goto done;
 957         }
 958
 959 pass:
 960         /* 127/8 must not appear on wire - RFC1122. */
 961         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 962             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 963                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 964                         ipstat.ips_badaddr++;
 965                         error = EADDRNOTAVAIL;
 966                         goto bad;
 967                 }
 968         }
 969
 970         m->m_pkthdr.csum_flags |= CSUM_IP;
 971         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 972         if (sw_csum & CSUM_DELAY_DATA) {
 973                 in_delayed_cksum(m);
 974                 sw_csum &= ~CSUM_DELAY_DATA;
 975         }
 976         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 977
 978         /*
 979          * If small enough for interface, or the interface will take
 980          * care of the fragmentation for us, can just send directly.
 981          */
 982         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 983             !(ip->ip_off & IP_DF))) {
 984                 ip->ip_len = htons(ip->ip_len);
 985                 ip->ip_off = htons(ip->ip_off);
 986                 ip->ip_sum = 0;
 987                 if (sw_csum & CSUM_DELAY_IP) {
 988                         if (ip->ip_vhl == IP_VHL_BORING) {
 989                                 ip->ip_sum = in_cksum_hdr(ip);
 990                         } else {
 991                                 ip->ip_sum = in_cksum(m, hlen);
 992                         }
 993                 }
 994
 995                 /* Record statistics for this interface address. */
 996                 if (!(flags & IP_FORWARDING) && ia) {
 997                         ia->ia_ifa.if_opackets++;
 998                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 999                 }
1000
1001 #ifdef IPSEC
1002                 /* clean ipsec history once it goes out of the node */
1003                 ipsec_delaux(m);
1004 #endif
1005
1006 #ifdef MBUF_STRESS_TEST
1007                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1008                         struct mbuf *m1, *m2;
1009                         int length, tmp;
1010
1011                         tmp = length = m->m_pkthdr.len;
1012
1013                         while ((length -= mbuf_frag_size) >= 1) {
1014                                 m1 = m_split(m, length, MB_DONTWAIT);
1015                                 if (m1 == NULL)
1016                                         break;
1017                                 m2 = m;
1018                                 while (m2->m_next != NULL)
1019                                         m2 = m2->m_next;
1020                                 m2->m_next = m1;
1021                         }
1022                         m->m_pkthdr.len = tmp;
1023                 }
1024 #endif
1025
1026 #ifdef MPLS
1027                 if (!mpls_output_process(m, ro->ro_rt))
1028                         goto done;
1029 #endif
1030                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1031                                        ro->ro_rt);
1032                 goto done;
1033         }
1034
1035         if (ip->ip_off & IP_DF) {
1036                 error = EMSGSIZE;
1037                 /*
1038                  * This case can happen if the user changed the MTU
1039                  * of an interface after enabling IP on it.  Because
1040                  * most netifs don't keep track of routes pointing to
1041                  * them, there is no way for one to update all its
1042                  * routes when the MTU is changed.
1043                  */
1044                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1045                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1046                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1047                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1048                 }
1049                 ipstat.ips_cantfrag++;
1050                 goto bad;
1051         }
1052
1053         /*
1054          * Too large for interface; fragment if possible. If successful,
1055          * on return, m will point to a list of packets to be sent.
1056          */
1057         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1058         if (error)
1059                 goto bad;
1060         for (; m; m = m0) {
1061                 m0 = m->m_nextpkt;
1062                 m->m_nextpkt = NULL;
1063 #ifdef IPSEC
1064                 /* clean ipsec history once it goes out of the node */
1065                 ipsec_delaux(m);
1066 #endif
1067                 if (error == 0) {
1068                         /* Record statistics for this interface address. */
1069                         if (ia != NULL) {
1070                                 ia->ia_ifa.if_opackets++;
1071                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1072                         }
1073 #ifdef MPLS
1074                         if (!mpls_output_process(m, ro->ro_rt))
1075                                 continue;
1076 #endif
1077                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1078                                                ro->ro_rt);
1079                 } else {
1080                         m_freem(m);
1081                 }
1082         }
1083
1084         if (error == 0)
1085                 ipstat.ips_fragmented++;
1086
1087 done:
1088         if (ro == &iproute && ro->ro_rt != NULL) {
1089                 RTFREE(ro->ro_rt);
1090                 ro->ro_rt = NULL;
1091         }
1092 #ifdef IPSEC
1093         if (sp != NULL) {
1094                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1095                         kprintf("DP ip_output call free SP:%p\n", sp));
1096                 key_freesp(sp);
1097         }
1098 #endif
1099 #ifdef FAST_IPSEC
1100         if (sp != NULL)
1101                 KEY_FREESP(&sp);
1102 #endif
1103         if (dn_mtag != NULL)
1104                 m_tag_free(dn_mtag);
1105
1106         return (error);
1107 bad:
1108         m_freem(m);
1109         goto done;
1110 }
1111
1112 /*
1113  * Create a chain of fragments which fit the given mtu. m_frag points to the
1114  * mbuf to be fragmented; on return it points to the chain with the fragments.
1115  * Return 0 if no error. If error, m_frag may contain a partially built
1116  * chain of fragments that should be freed by the caller.
1117  *
1118  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1119  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1120  */
1121 int
1122 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1123             u_long if_hwassist_flags, int sw_csum)
1124 {
1125         int error = 0;
1126         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1127         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1128         int off;
1129         struct mbuf *m0 = *m_frag;      /* the original packet          */
1130         int firstlen;
1131         struct mbuf **mnext;
1132         int nfrags;
1133
1134         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1135                 ipstat.ips_cantfrag++;
1136                 return EMSGSIZE;
1137         }
1138
1139         /*
1140          * Must be able to put at least 8 bytes per fragment.
1141          */
1142         if (len < 8)
1143                 return EMSGSIZE;
1144
1145         /*
1146          * If the interface will not calculate checksums on
1147          * fragmented packets, then do it here.
1148          */
1149         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1150             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1151                 in_delayed_cksum(m0);
1152                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1153         }
1154
1155         if (len > PAGE_SIZE) {
1156                 /*
1157                  * Fragment large datagrams such that each segment
1158                  * contains a multiple of PAGE_SIZE amount of data,
1159                  * plus headers. This enables a receiver to perform
1160                  * page-flipping zero-copy optimizations.
1161                  *
1162                  * XXX When does this help given that sender and receiver
1163                  * could have different page sizes, and also mtu could
1164                  * be less than the receiver's page size ?
1165                  */
1166                 int newlen;
1167                 struct mbuf *m;
1168
1169                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1170                         off += m->m_len;
1171
1172                 /*
1173                  * firstlen (off - hlen) must be aligned on an
1174                  * 8-byte boundary
1175                  */
1176                 if (off < hlen)
1177                         goto smart_frag_failure;
1178                 off = ((off - hlen) & ~7) + hlen;
1179                 newlen = (~PAGE_MASK) & mtu;
1180                 if ((newlen + sizeof(struct ip)) > mtu) {
1181                         /* we failed, go back the default */
1182 smart_frag_failure:
1183                         newlen = len;
1184                         off = hlen + len;
1185                 }
1186                 len = newlen;
1187
1188         } else {
1189                 off = hlen + len;
1190         }
1191
1192         firstlen = off - hlen;
1193         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1194
1195         /*
1196          * Loop through length of segment after first fragment,
1197          * make new header and copy data of each part and link onto chain.
1198          * Here, m0 is the original packet, m is the fragment being created.
1199          * The fragments are linked off the m_nextpkt of the original
1200          * packet, which after processing serves as the first fragment.
1201          */
1202         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1203                 struct ip *mhip;        /* ip header on the fragment */
1204                 struct mbuf *m;
1205                 int mhlen = sizeof(struct ip);
1206
1207                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1208                 if (m == NULL) {
1209                         error = ENOBUFS;
1210                         ipstat.ips_odropped++;
1211                         goto done;
1212                 }
1213                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1214                 /*
1215                  * In the first mbuf, leave room for the link header, then
1216                  * copy the original IP header including options. The payload
1217                  * goes into an additional mbuf chain returned by m_copy().
1218                  */
1219                 m->m_data += max_linkhdr;
1220                 mhip = mtod(m, struct ip *);
1221                 *mhip = *ip;
1222                 if (hlen > sizeof(struct ip)) {
1223                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1224                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1225                 }
1226                 m->m_len = mhlen;
1227                 /* XXX do we need to add ip->ip_off below ? */
1228                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1229                 if (off + len >= ip->ip_len) {  /* last fragment */
1230                         len = ip->ip_len - off;
1231                         m->m_flags |= M_LASTFRAG;
1232                 } else
1233                         mhip->ip_off |= IP_MF;
1234                 mhip->ip_len = htons((u_short)(len + mhlen));
1235                 m->m_next = m_copy(m0, off, len);
1236                 if (m->m_next == NULL) {                /* copy failed */
1237                         m_free(m);
1238                         error = ENOBUFS;        /* ??? */
1239                         ipstat.ips_odropped++;
1240                         goto done;
1241                 }
1242                 m->m_pkthdr.len = mhlen + len;
1243                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1244                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1245                 mhip->ip_off = htons(mhip->ip_off);
1246                 mhip->ip_sum = 0;
1247                 if (sw_csum & CSUM_DELAY_IP)
1248                         mhip->ip_sum = in_cksum(m, mhlen);
1249                 *mnext = m;
1250                 mnext = &m->m_nextpkt;
1251         }
1252         ipstat.ips_ofragments += nfrags;
1253
1254         /* set first marker for fragment chain */
1255         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1256         m0->m_pkthdr.csum_data = nfrags;
1257
1258         /*
1259          * Update first fragment by trimming what's been copied out
1260          * and updating header.
1261          */
1262         m_adj(m0, hlen + firstlen - ip->ip_len);
1263         m0->m_pkthdr.len = hlen + firstlen;
1264         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1265         ip->ip_off |= IP_MF;
1266         ip->ip_off = htons(ip->ip_off);
1267         ip->ip_sum = 0;
1268         if (sw_csum & CSUM_DELAY_IP)
1269                 ip->ip_sum = in_cksum(m0, hlen);
1270
1271 done:
1272         *m_frag = m0;
1273         return error;
1274 }
1275
1276 void
1277 in_delayed_cksum(struct mbuf *m)
1278 {
1279         struct ip *ip;
1280         u_short csum, offset;
1281
1282         ip = mtod(m, struct ip *);
1283         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1284         csum = in_cksum_skip(m, ip->ip_len, offset);
1285         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1286                 csum = 0xffff;
1287         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1288
1289         if (offset + sizeof(u_short) > m->m_len) {
1290                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1291                     m->m_len, offset, ip->ip_p);
1292                 /*
1293                  * XXX
1294                  * this shouldn't happen, but if it does, the
1295                  * correct behavior may be to insert the checksum
1296                  * in the existing chain instead of rearranging it.
1297                  */
1298                 m = m_pullup(m, offset + sizeof(u_short));
1299         }
1300         *(u_short *)(m->m_data + offset) = csum;
1301 }
1302
1303 /*
1304  * Insert IP options into preformed packet.
1305  * Adjust IP destination as required for IP source routing,
1306  * as indicated by a non-zero in_addr at the start of the options.
1307  *
1308  * XXX This routine assumes that the packet has no options in place.
1309  */
1310 static struct mbuf *
1311 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1312 {
1313         struct ipoption *p = mtod(opt, struct ipoption *);
1314         struct mbuf *n;
1315         struct ip *ip = mtod(m, struct ip *);
1316         unsigned optlen;
1317
1318         optlen = opt->m_len - sizeof p->ipopt_dst;
1319         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1320                 *phlen = 0;
1321                 return (m);             /* XXX should fail */
1322         }
1323         if (p->ipopt_dst.s_addr)
1324                 ip->ip_dst = p->ipopt_dst;
1325         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1326                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1327                 if (n == NULL) {
1328                         *phlen = 0;
1329                         return (m);
1330                 }
1331                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1332                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1333                 m->m_len -= sizeof(struct ip);
1334                 m->m_data += sizeof(struct ip);
1335                 n->m_next = m;
1336                 m = n;
1337                 m->m_len = optlen + sizeof(struct ip);
1338                 m->m_data += max_linkhdr;
1339                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1340         } else {
1341                 m->m_data -= optlen;
1342                 m->m_len += optlen;
1343                 m->m_pkthdr.len += optlen;
1344                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1345         }
1346         ip = mtod(m, struct ip *);
1347         bcopy(p->ipopt_list, ip + 1, optlen);
1348         *phlen = sizeof(struct ip) + optlen;
1349         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1350         ip->ip_len += optlen;
1351         return (m);
1352 }
1353
1354 /*
1355  * Copy options from ip to jp,
1356  * omitting those not copied during fragmentation.
1357  */
1358 int
1359 ip_optcopy(struct ip *ip, struct ip *jp)
1360 {
1361         u_char *cp, *dp;
1362         int opt, optlen, cnt;
1363
1364         cp = (u_char *)(ip + 1);
1365         dp = (u_char *)(jp + 1);
1366         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1367         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1368                 opt = cp[0];
1369                 if (opt == IPOPT_EOL)
1370                         break;
1371                 if (opt == IPOPT_NOP) {
1372                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1373                         *dp++ = IPOPT_NOP;
1374                         optlen = 1;
1375                         continue;
1376                 }
1377
1378                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1379                     ("ip_optcopy: malformed ipv4 option"));
1380                 optlen = cp[IPOPT_OLEN];
1381                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1382                     ("ip_optcopy: malformed ipv4 option"));
1383
1384                 /* bogus lengths should have been caught by ip_dooptions */
1385                 if (optlen > cnt)
1386                         optlen = cnt;
1387                 if (IPOPT_COPIED(opt)) {
1388                         bcopy(cp, dp, optlen);
1389                         dp += optlen;
1390                 }
1391         }
1392         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1393                 *dp++ = IPOPT_EOL;
1394         return (optlen);
1395 }
1396
1397 /*
1398  * IP socket option processing.
1399  */
1400 int
1401 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1402 {
1403         struct  inpcb *inp = so->so_pcb;
1404         int     error, optval;
1405
1406         error = optval = 0;
1407         if (sopt->sopt_level != IPPROTO_IP) {
1408                 return (EINVAL);
1409         }
1410
1411         switch (sopt->sopt_dir) {
1412         case SOPT_SET:
1413                 switch (sopt->sopt_name) {
1414                 case IP_OPTIONS:
1415 #ifdef notyet
1416                 case IP_RETOPTS:
1417 #endif
1418                 {
1419                         struct mbuf *m;
1420                         if (sopt->sopt_valsize > MLEN) {
1421                                 error = EMSGSIZE;
1422                                 break;
1423                         }
1424                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1425                         if (m == NULL) {
1426                                 error = ENOBUFS;
1427                                 break;
1428                         }
1429                         m->m_len = sopt->sopt_valsize;
1430                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1431                                               m->m_len);
1432                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1433                                            m));
1434                 }
1435
1436                 case IP_TOS:
1437                 case IP_TTL:
1438                 case IP_MINTTL:
1439                 case IP_RECVOPTS:
1440                 case IP_RECVRETOPTS:
1441                 case IP_RECVDSTADDR:
1442                 case IP_RECVIF:
1443                 case IP_RECVTTL:
1444                 case IP_FAITH:
1445                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1446                                              sizeof optval);
1447                         if (error)
1448                                 break;
1449                         switch (sopt->sopt_name) {
1450                         case IP_TOS:
1451                                 inp->inp_ip_tos = optval;
1452                                 break;
1453
1454                         case IP_TTL:
1455                                 inp->inp_ip_ttl = optval;
1456                                 break;
1457                         case IP_MINTTL:
1458                                 if (optval > 0 && optval <= MAXTTL)
1459                                         inp->inp_ip_minttl = optval;
1460                                 else
1461                                         error = EINVAL;
1462                                 break;
1463 #define OPTSET(bit) \
1464         if (optval) \
1465                 inp->inp_flags |= bit; \
1466         else \
1467                 inp->inp_flags &= ~bit;
1468
1469                         case IP_RECVOPTS:
1470                                 OPTSET(INP_RECVOPTS);
1471                                 break;
1472
1473                         case IP_RECVRETOPTS:
1474                                 OPTSET(INP_RECVRETOPTS);
1475                                 break;
1476
1477                         case IP_RECVDSTADDR:
1478                                 OPTSET(INP_RECVDSTADDR);
1479                                 break;
1480
1481                         case IP_RECVIF:
1482                                 OPTSET(INP_RECVIF);
1483                                 break;
1484
1485                         case IP_RECVTTL:
1486                                 OPTSET(INP_RECVTTL);
1487                                 break;
1488
1489                         case IP_FAITH:
1490                                 OPTSET(INP_FAITH);
1491                                 break;
1492                         }
1493                         break;
1494 #undef OPTSET
1495
1496                 case IP_MULTICAST_IF:
1497                 case IP_MULTICAST_VIF:
1498                 case IP_MULTICAST_TTL:
1499                 case IP_MULTICAST_LOOP:
1500                 case IP_ADD_MEMBERSHIP:
1501                 case IP_DROP_MEMBERSHIP:
1502                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1503                         break;
1504
1505                 case IP_PORTRANGE:
1506                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1507                                             sizeof optval);
1508                         if (error)
1509                                 break;
1510
1511                         switch (optval) {
1512                         case IP_PORTRANGE_DEFAULT:
1513                                 inp->inp_flags &= ~(INP_LOWPORT);
1514                                 inp->inp_flags &= ~(INP_HIGHPORT);
1515                                 break;
1516
1517                         case IP_PORTRANGE_HIGH:
1518                                 inp->inp_flags &= ~(INP_LOWPORT);
1519                                 inp->inp_flags |= INP_HIGHPORT;
1520                                 break;
1521
1522                         case IP_PORTRANGE_LOW:
1523                                 inp->inp_flags &= ~(INP_HIGHPORT);
1524                                 inp->inp_flags |= INP_LOWPORT;
1525                                 break;
1526
1527                         default:
1528                                 error = EINVAL;
1529                                 break;
1530                         }
1531                         break;
1532
1533 #if defined(IPSEC) || defined(FAST_IPSEC)
1534                 case IP_IPSEC_POLICY:
1535                 {
1536                         caddr_t req;
1537                         size_t len = 0;
1538                         int priv;
1539                         struct mbuf *m;
1540                         int optname;
1541
1542                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1543                                 break;
1544                         soopt_to_mbuf(sopt, m);
1545                         priv = (sopt->sopt_td != NULL &&
1546                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1547                         req = mtod(m, caddr_t);
1548                         len = m->m_len;
1549                         optname = sopt->sopt_name;
1550                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1551                         m_freem(m);
1552                         break;
1553                 }
1554 #endif /*IPSEC*/
1555
1556                 default:
1557                         error = ENOPROTOOPT;
1558                         break;
1559                 }
1560                 break;
1561
1562         case SOPT_GET:
1563                 switch (sopt->sopt_name) {
1564                 case IP_OPTIONS:
1565                 case IP_RETOPTS:
1566                         if (inp->inp_options)
1567                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1568                                                            char *),
1569                                                 inp->inp_options->m_len);
1570                         else
1571                                 sopt->sopt_valsize = 0;
1572                         break;
1573
1574                 case IP_TOS:
1575                 case IP_TTL:
1576                 case IP_MINTTL:
1577                 case IP_RECVOPTS:
1578                 case IP_RECVRETOPTS:
1579                 case IP_RECVDSTADDR:
1580                 case IP_RECVTTL:
1581                 case IP_RECVIF:
1582                 case IP_PORTRANGE:
1583                 case IP_FAITH:
1584                         switch (sopt->sopt_name) {
1585
1586                         case IP_TOS:
1587                                 optval = inp->inp_ip_tos;
1588                                 break;
1589
1590                         case IP_TTL:
1591                                 optval = inp->inp_ip_ttl;
1592                                 break;
1593                         case IP_MINTTL:
1594                                 optval = inp->inp_ip_minttl;
1595                                 break;
1596
1597 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1598
1599                         case IP_RECVOPTS:
1600                                 optval = OPTBIT(INP_RECVOPTS);
1601                                 break;
1602
1603                         case IP_RECVRETOPTS:
1604                                 optval = OPTBIT(INP_RECVRETOPTS);
1605                                 break;
1606
1607                         case IP_RECVDSTADDR:
1608                                 optval = OPTBIT(INP_RECVDSTADDR);
1609                                 break;
1610
1611                         case IP_RECVTTL:
1612                                 optval = OPTBIT(INP_RECVTTL);
1613                                 break;
1614
1615                         case IP_RECVIF:
1616                                 optval = OPTBIT(INP_RECVIF);
1617                                 break;
1618
1619                         case IP_PORTRANGE:
1620                                 if (inp->inp_flags & INP_HIGHPORT)
1621                                         optval = IP_PORTRANGE_HIGH;
1622                                 else if (inp->inp_flags & INP_LOWPORT)
1623                                         optval = IP_PORTRANGE_LOW;
1624                                 else
1625                                         optval = 0;
1626                                 break;
1627
1628                         case IP_FAITH:
1629                                 optval = OPTBIT(INP_FAITH);
1630                                 break;
1631                         }
1632                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1633                         break;
1634
1635                 case IP_MULTICAST_IF:
1636                 case IP_MULTICAST_VIF:
1637                 case IP_MULTICAST_TTL:
1638                 case IP_MULTICAST_LOOP:
1639                 case IP_ADD_MEMBERSHIP:
1640                 case IP_DROP_MEMBERSHIP:
1641                         error = ip_getmoptions(sopt, inp->inp_moptions);
1642                         break;
1643
1644 #if defined(IPSEC) || defined(FAST_IPSEC)
1645                 case IP_IPSEC_POLICY:
1646                 {
1647                         struct mbuf *m = NULL;
1648                         caddr_t req = NULL;
1649                         size_t len = 0;
1650
1651                         if (m != NULL) {
1652                                 req = mtod(m, caddr_t);
1653                                 len = m->m_len;
1654                         }
1655                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1656                         if (error == 0)
1657                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1658                         if (error == 0)
1659                                 m_freem(m);
1660                         break;
1661                 }
1662 #endif /*IPSEC*/
1663
1664                 default:
1665                         error = ENOPROTOOPT;
1666                         break;
1667                 }
1668                 break;
1669         }
1670         return (error);
1671 }
1672
1673 /*
1674  * Set up IP options in pcb for insertion in output packets.
1675  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1676  * with destination address if source routed.
1677  */
1678 static int
1679 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1680 {
1681         int cnt, optlen;
1682         u_char *cp;
1683         u_char opt;
1684
1685         /* turn off any old options */
1686         if (*pcbopt)
1687                 m_free(*pcbopt);
1688         *pcbopt = 0;
1689         if (m == NULL || m->m_len == 0) {
1690                 /*
1691                  * Only turning off any previous options.
1692                  */
1693                 if (m != NULL)
1694                         m_free(m);
1695                 return (0);
1696         }
1697
1698         if (m->m_len % sizeof(int32_t))
1699                 goto bad;
1700         /*
1701          * IP first-hop destination address will be stored before
1702          * actual options; move other options back
1703          * and clear it when none present.
1704          */
1705         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1706                 goto bad;
1707         cnt = m->m_len;
1708         m->m_len += sizeof(struct in_addr);
1709         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1710         ovbcopy(mtod(m, caddr_t), cp, cnt);
1711         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1712
1713         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1714                 opt = cp[IPOPT_OPTVAL];
1715                 if (opt == IPOPT_EOL)
1716                         break;
1717                 if (opt == IPOPT_NOP)
1718                         optlen = 1;
1719                 else {
1720                         if (cnt < IPOPT_OLEN + sizeof *cp)
1721                                 goto bad;
1722                         optlen = cp[IPOPT_OLEN];
1723                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1724                                 goto bad;
1725                 }
1726                 switch (opt) {
1727
1728                 default:
1729                         break;
1730
1731                 case IPOPT_LSRR:
1732                 case IPOPT_SSRR:
1733                         /*
1734                          * user process specifies route as:
1735                          *      ->A->B->C->D
1736                          * D must be our final destination (but we can't
1737                          * check that since we may not have connected yet).
1738                          * A is first hop destination, which doesn't appear in
1739                          * actual IP option, but is stored before the options.
1740                          */
1741                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1742                                 goto bad;
1743                         m->m_len -= sizeof(struct in_addr);
1744                         cnt -= sizeof(struct in_addr);
1745                         optlen -= sizeof(struct in_addr);
1746                         cp[IPOPT_OLEN] = optlen;
1747                         /*
1748                          * Move first hop before start of options.
1749                          */
1750                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1751                               sizeof(struct in_addr));
1752                         /*
1753                          * Then copy rest of options back
1754                          * to close up the deleted entry.
1755                          */
1756                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1757                                 &cp[IPOPT_OFFSET+1],
1758                                 cnt - (IPOPT_MINOFF - 1));
1759                         break;
1760                 }
1761         }
1762         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1763                 goto bad;
1764         *pcbopt = m;
1765         return (0);
1766
1767 bad:
1768         m_free(m);
1769         return (EINVAL);
1770 }
1771
1772 /*
1773  * XXX
1774  * The whole multicast option thing needs to be re-thought.
1775  * Several of these options are equally applicable to non-multicast
1776  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1777  * standard option (IP_TTL).
1778  */
1779
1780 /*
1781  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1782  */
1783 static struct ifnet *
1784 ip_multicast_if(struct in_addr *a, 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                 ifp = INADDR_TO_IFP(a);
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(struct sockopt *sopt, struct ip_moptions **imop)
1809 {
1810         int error = 0;
1811         int i;
1812         struct in_addr addr;
1813         struct ip_mreq mreq;
1814         struct ifnet *ifp;
1815         struct ip_moptions *imo = *imop;
1816         int ifindex;
1817
1818         if (imo == NULL) {
1819                 /*
1820                  * No multicast option buffer attached to the pcb;
1821                  * allocate one and initialize to default values.
1822                  */
1823                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1824
1825                 *imop = imo;
1826                 imo->imo_multicast_ifp = NULL;
1827                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1828                 imo->imo_multicast_vif = -1;
1829                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1830                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1831                 imo->imo_num_memberships = 0;
1832         }
1833         switch (sopt->sopt_name) {
1834         /* store an index number for the vif you wanna use in the send */
1835         case IP_MULTICAST_VIF:
1836                 if (legal_vif_num == 0) {
1837                         error = EOPNOTSUPP;
1838                         break;
1839                 }
1840                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1841                 if (error)
1842                         break;
1843                 if (!legal_vif_num(i) && (i != -1)) {
1844                         error = EINVAL;
1845                         break;
1846                 }
1847                 imo->imo_multicast_vif = i;
1848                 break;
1849
1850         case IP_MULTICAST_IF:
1851                 /*
1852                  * Select the interface for outgoing multicast packets.
1853                  */
1854                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1855                 if (error)
1856                         break;
1857
1858                 /*
1859                  * INADDR_ANY is used to remove a previous selection.
1860                  * When no interface is selected, a default one is
1861                  * chosen every time a multicast packet is sent.
1862                  */
1863                 if (addr.s_addr == INADDR_ANY) {
1864                         imo->imo_multicast_ifp = NULL;
1865                         break;
1866                 }
1867                 /*
1868                  * The selected interface is identified by its local
1869                  * IP address.  Find the interface and confirm that
1870                  * it supports multicasting.
1871                  */
1872                 crit_enter();
1873                 ifp = ip_multicast_if(&addr, &ifindex);
1874                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1875                         crit_exit();
1876                         error = EADDRNOTAVAIL;
1877                         break;
1878                 }
1879                 imo->imo_multicast_ifp = ifp;
1880                 if (ifindex)
1881                         imo->imo_multicast_addr = addr;
1882                 else
1883                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1884                 crit_exit();
1885                 break;
1886
1887         case IP_MULTICAST_TTL:
1888                 /*
1889                  * Set the IP time-to-live for outgoing multicast packets.
1890                  * The original multicast API required a char argument,
1891                  * which is inconsistent with the rest of the socket API.
1892                  * We allow either a char or an int.
1893                  */
1894                 if (sopt->sopt_valsize == 1) {
1895                         u_char ttl;
1896                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1897                         if (error)
1898                                 break;
1899                         imo->imo_multicast_ttl = ttl;
1900                 } else {
1901                         u_int ttl;
1902                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1903                         if (error)
1904                                 break;
1905                         if (ttl > 255)
1906                                 error = EINVAL;
1907                         else
1908                                 imo->imo_multicast_ttl = ttl;
1909                 }
1910                 break;
1911
1912         case IP_MULTICAST_LOOP:
1913                 /*
1914                  * Set the loopback flag for outgoing multicast packets.
1915                  * Must be zero or one.  The original multicast API required a
1916                  * char argument, which is inconsistent with the rest
1917                  * of the socket API.  We allow either a char or an int.
1918                  */
1919                 if (sopt->sopt_valsize == 1) {
1920                         u_char loop;
1921
1922                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1923                         if (error)
1924                                 break;
1925                         imo->imo_multicast_loop = !!loop;
1926                 } else {
1927                         u_int loop;
1928
1929                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1930                                             sizeof loop);
1931                         if (error)
1932                                 break;
1933                         imo->imo_multicast_loop = !!loop;
1934                 }
1935                 break;
1936
1937         case IP_ADD_MEMBERSHIP:
1938                 /*
1939                  * Add a multicast group membership.
1940                  * Group must be a valid IP multicast address.
1941                  */
1942                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1943                 if (error)
1944                         break;
1945
1946                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1947                         error = EINVAL;
1948                         break;
1949                 }
1950                 crit_enter();
1951                 /*
1952                  * If no interface address was provided, use the interface of
1953                  * the route to the given multicast address.
1954                  */
1955                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1956                         struct sockaddr_in dst;
1957                         struct rtentry *rt;
1958
1959                         bzero(&dst, sizeof(struct sockaddr_in));
1960                         dst.sin_len = sizeof(struct sockaddr_in);
1961                         dst.sin_family = AF_INET;
1962                         dst.sin_addr = mreq.imr_multiaddr;
1963                         rt = rtlookup((struct sockaddr *)&dst);
1964                         if (rt == NULL) {
1965                                 error = EADDRNOTAVAIL;
1966                                 crit_exit();
1967                                 break;
1968                         }
1969                         --rt->rt_refcnt;
1970                         ifp = rt->rt_ifp;
1971                 } else {
1972                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1973                 }
1974
1975                 /*
1976                  * See if we found an interface, and confirm that it
1977                  * supports multicast.
1978                  */
1979                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1980                         error = EADDRNOTAVAIL;
1981                         crit_exit();
1982                         break;
1983                 }
1984                 /*
1985                  * See if the membership already exists or if all the
1986                  * membership slots are full.
1987                  */
1988                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1989                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1990                             imo->imo_membership[i]->inm_addr.s_addr
1991                                                 == mreq.imr_multiaddr.s_addr)
1992                                 break;
1993                 }
1994                 if (i < imo->imo_num_memberships) {
1995                         error = EADDRINUSE;
1996                         crit_exit();
1997                         break;
1998                 }
1999                 if (i == IP_MAX_MEMBERSHIPS) {
2000                         error = ETOOMANYREFS;
2001                         crit_exit();
2002                         break;
2003                 }
2004                 /*
2005                  * Everything looks good; add a new record to the multicast
2006                  * address list for the given interface.
2007                  */
2008                 if ((imo->imo_membership[i] =
2009                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2010                         error = ENOBUFS;
2011                         crit_exit();
2012                         break;
2013                 }
2014                 ++imo->imo_num_memberships;
2015                 crit_exit();
2016                 break;
2017
2018         case IP_DROP_MEMBERSHIP:
2019                 /*
2020                  * Drop a multicast group membership.
2021                  * Group must be a valid IP multicast address.
2022                  */
2023                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2024                 if (error)
2025                         break;
2026
2027                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2028                         error = EINVAL;
2029                         break;
2030                 }
2031
2032                 crit_enter();
2033                 /*
2034                  * If an interface address was specified, get a pointer
2035                  * to its ifnet structure.
2036                  */
2037                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2038                         ifp = NULL;
2039                 else {
2040                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2041                         if (ifp == NULL) {
2042                                 error = EADDRNOTAVAIL;
2043                                 crit_exit();
2044                                 break;
2045                         }
2046                 }
2047                 /*
2048                  * Find the membership in the membership array.
2049                  */
2050                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2051                         if ((ifp == NULL ||
2052                              imo->imo_membership[i]->inm_ifp == ifp) &&
2053                             imo->imo_membership[i]->inm_addr.s_addr ==
2054                             mreq.imr_multiaddr.s_addr)
2055                                 break;
2056                 }
2057                 if (i == imo->imo_num_memberships) {
2058                         error = EADDRNOTAVAIL;
2059                         crit_exit();
2060                         break;
2061                 }
2062                 /*
2063                  * Give up the multicast address record to which the
2064                  * membership points.
2065                  */
2066                 in_delmulti(imo->imo_membership[i]);
2067                 /*
2068                  * Remove the gap in the membership array.
2069                  */
2070                 for (++i; i < imo->imo_num_memberships; ++i)
2071                         imo->imo_membership[i-1] = imo->imo_membership[i];
2072                 --imo->imo_num_memberships;
2073                 crit_exit();
2074                 break;
2075
2076         default:
2077                 error = EOPNOTSUPP;
2078                 break;
2079         }
2080
2081         /*
2082          * If all options have default values, no need to keep the mbuf.
2083          */
2084         if (imo->imo_multicast_ifp == NULL &&
2085             imo->imo_multicast_vif == -1 &&
2086             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2087             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2088             imo->imo_num_memberships == 0) {
2089                 kfree(*imop, M_IPMOPTS);
2090                 *imop = NULL;
2091         }
2092
2093         return (error);
2094 }
2095
2096 /*
2097  * Return the IP multicast options in response to user getsockopt().
2098  */
2099 static int
2100 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2101 {
2102         struct in_addr addr;
2103         struct in_ifaddr *ia;
2104         int error, optval;
2105         u_char coptval;
2106
2107         error = 0;
2108         switch (sopt->sopt_name) {
2109         case IP_MULTICAST_VIF:
2110                 if (imo != NULL)
2111                         optval = imo->imo_multicast_vif;
2112                 else
2113                         optval = -1;
2114                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2115                 break;
2116
2117         case IP_MULTICAST_IF:
2118                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2119                         addr.s_addr = INADDR_ANY;
2120                 else if (imo->imo_multicast_addr.s_addr) {
2121                         /* return the value user has set */
2122                         addr = imo->imo_multicast_addr;
2123                 } else {
2124                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2125                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2126                                 : IA_SIN(ia)->sin_addr.s_addr;
2127                 }
2128                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2129                 break;
2130
2131         case IP_MULTICAST_TTL:
2132                 if (imo == NULL)
2133                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2134                 else
2135                         optval = coptval = imo->imo_multicast_ttl;
2136                 if (sopt->sopt_valsize == 1)
2137                         soopt_from_kbuf(sopt, &coptval, 1);
2138                 else
2139                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2140                 break;
2141
2142         case IP_MULTICAST_LOOP:
2143                 if (imo == NULL)
2144                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2145                 else
2146                         optval = coptval = imo->imo_multicast_loop;
2147                 if (sopt->sopt_valsize == 1)
2148                         soopt_from_kbuf(sopt, &coptval, 1);
2149                 else
2150                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2151                 break;
2152
2153         default:
2154                 error = ENOPROTOOPT;
2155                 break;
2156         }
2157         return (error);
2158 }
2159
2160 /*
2161  * Discard the IP multicast options.
2162  */
2163 void
2164 ip_freemoptions(struct ip_moptions *imo)
2165 {
2166         int i;
2167
2168         if (imo != NULL) {
2169                 for (i = 0; i < imo->imo_num_memberships; ++i)
2170                         in_delmulti(imo->imo_membership[i]);
2171                 kfree(imo, M_IPMOPTS);
2172         }
2173 }
2174
2175 /*
2176  * Routine called from ip_output() to loop back a copy of an IP multicast
2177  * packet to the input queue of a specified interface.  Note that this
2178  * calls the output routine of the loopback "driver", but with an interface
2179  * pointer that might NOT be a loopback interface -- evil, but easier than
2180  * replicating that code here.
2181  */
2182 static void
2183 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2184              int hlen)
2185 {
2186         struct ip *ip;
2187         struct mbuf *copym;
2188
2189         copym = m_copypacket(m, MB_DONTWAIT);
2190         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2191                 copym = m_pullup(copym, hlen);
2192         if (copym != NULL) {
2193                 /*
2194                  * if the checksum hasn't been computed, mark it as valid
2195                  */
2196                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2197                         in_delayed_cksum(copym);
2198                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2199                         copym->m_pkthdr.csum_flags |=
2200                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2201                         copym->m_pkthdr.csum_data = 0xffff;
2202                 }
2203                 /*
2204                  * We don't bother to fragment if the IP length is greater
2205                  * than the interface's MTU.  Can this possibly matter?
2206                  */
2207                 ip = mtod(copym, struct ip *);
2208                 ip->ip_len = htons(ip->ip_len);
2209                 ip->ip_off = htons(ip->ip_off);
2210                 ip->ip_sum = 0;
2211                 if (ip->ip_vhl == IP_VHL_BORING) {
2212                         ip->ip_sum = in_cksum_hdr(ip);
2213                 } else {
2214                         ip->ip_sum = in_cksum(copym, hlen);
2215                 }
2216                 /*
2217                  * NB:
2218                  * It's not clear whether there are any lingering
2219                  * reentrancy problems in other areas which might
2220                  * be exposed by using ip_input directly (in
2221                  * particular, everything which modifies the packet
2222                  * in-place).  Yet another option is using the
2223                  * protosw directly to deliver the looped back
2224                  * packet.  For the moment, we'll err on the side
2225                  * of safety by using if_simloop().
2226                  */
2227 #if 1 /* XXX */
2228                 if (dst->sin_family != AF_INET) {
2229                         kprintf("ip_mloopback: bad address family %d\n",
2230                                                 dst->sin_family);
2231                         dst->sin_family = AF_INET;
2232                 }
2233 #endif
2234
2235 #ifdef notdef
2236                 copym->m_pkthdr.rcvif = ifp;
2237                 ip_input(copym);
2238 #else
2239                 if_simloop(ifp, copym, dst->sin_family, 0);
2240 #endif
2241         }
2242 }