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 1.63 2008/09/13 12:57:07 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 extern  int route_assert_owner_access;
 119 extern  void db_print_backtrace(void);
 120
 121 extern  struct protosw inetsw[];
 122
 123 static int
 124 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst)
 125 {
 126         struct in_ifaddr_container *iac;
 127
 128         /*
 129          * We need to figure out if we have been forwarded to a local
 130          * socket.  If so, then we should somehow "loop back" to
 131          * ip_input(), and get directed to the PCB as if we had received
 132          * this packet.  This is because it may be difficult to identify
 133          * the packets you want to forward until they are being output
 134          * and have selected an interface (e.g. locally initiated
 135          * packets).  If we used the loopback inteface, we would not be
 136          * able to control what happens as the packet runs through
 137          * ip_input() as it is done through a ISR.
 138          */
 139         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 140                 /*
 141                  * If the addr to forward to is one of ours, we pretend
 142                  * to be the destination for this packet.
 143                  */
 144                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 145                         break;
 146         }
 147         if (iac != NULL) {
 148                 struct ip *ip = mtod(m, struct ip *);
 149
 150                 if (m->m_pkthdr.rcvif == NULL)
 151                         m->m_pkthdr.rcvif = ifunit("lo0");
 152                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 153                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 154                                                   CSUM_PSEUDO_HDR;
 155                         m->m_pkthdr.csum_data = 0xffff;
 156                 }
 157                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 158
 159                 ip->ip_len = htons(ip->ip_len);
 160                 ip->ip_off = htons(ip->ip_off);
 161                 ip_input(m);
 162
 163                 return 1; /* Packet gets forwarded locally */
 164         }
 165         return 0;
 166 }
 167
 168 /*
 169  * IP output.  The packet in mbuf chain m contains a skeletal IP
 170  * header (with len, off, ttl, proto, tos, src, dst).
 171  * The mbuf chain containing the packet will be freed.
 172  * The mbuf opt, if present, will not be freed.
 173  */
 174 int
 175 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 176           int flags, struct ip_moptions *imo, struct inpcb *inp)
 177 {
 178         struct ip *ip;
 179         struct ifnet *ifp = NULL;       /* keep compiler happy */
 180         struct mbuf *m;
 181         int hlen = sizeof(struct ip);
 182         int len, error = 0;
 183         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 184         struct in_ifaddr *ia = NULL;
 185         int isbroadcast, sw_csum;
 186         struct in_addr pkt_dst;
 187         struct route iproute;
 188         struct m_tag *mtag;
 189 #ifdef IPSEC
 190         struct secpolicy *sp = NULL;
 191         struct socket *so = inp ? inp->inp_socket : NULL;
 192 #endif
 193 #ifdef FAST_IPSEC
 194         struct secpolicy *sp = NULL;
 195         struct tdb_ident *tdbi;
 196 #endif /* FAST_IPSEC */
 197         struct sockaddr_in *next_hop = NULL;
 198         int src_was_INADDR_ANY = 0;     /* as the name says... */
 199
 200         m = m0;
 201         M_ASSERTPKTHDR(m);
 202
 203         if (ro == NULL) {
 204                 ro = &iproute;
 205                 bzero(ro, sizeof *ro);
 206         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 207                 if (flags & IP_DEBUGROUTE) {
 208                         if (route_assert_owner_access) {
 209                                 panic("ip_output: "
 210                                       "rt rt_cpuid %d accessed on cpu %d\n",
 211                                       ro->ro_rt->rt_cpuid, mycpuid);
 212                         } else {
 213                                 kprintf("ip_output: "
 214                                         "rt rt_cpuid %d accessed on cpu %d\n",
 215                                         ro->ro_rt->rt_cpuid, mycpuid);
 216                                 db_print_backtrace();
 217                         }
 218                 }
 219
 220                 /*
 221                  * XXX
 222                  * If the cached rtentry's owner CPU is not the current CPU,
 223                  * then don't touch the cached rtentry (remote free is too
 224                  * expensive in this context); just relocate the route.
 225                  */
 226                 ro = &iproute;
 227                 bzero(ro, sizeof *ro);
 228         }
 229
 230         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 231                 /* Next hop */
 232                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 233                 KKASSERT(mtag != NULL);
 234                 next_hop = m_tag_data(mtag);
 235         }
 236
 237         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 238                 struct dn_pkt *dn_pkt;
 239
 240                 /* Extract info from dummynet tag */
 241                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 242                 KKASSERT(mtag != NULL);
 243                 dn_pkt = m_tag_data(mtag);
 244
 245                 /*
 246                  * The packet was already tagged, so part of the
 247                  * processing was already done, and we need to go down.
 248                  * Get the calculated parameters from the tag.
 249                  */
 250                 ifp = dn_pkt->ifp;
 251
 252                 KKASSERT(ro == &iproute);
 253                 *ro = dn_pkt->ro; /* structure copy */
 254                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 255
 256                 dst = dn_pkt->dn_dst;
 257                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 258                         /* If 'dst' points into dummynet tag, adjust it */
 259                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 260                 }
 261
 262                 ip = mtod(m, struct ip *);
 263                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 264                 if (ro->ro_rt)
 265                         ia = ifatoia(ro->ro_rt->rt_ifa);
 266                 goto sendit;
 267         }
 268
 269         if (opt) {
 270                 len = 0;
 271                 m = ip_insertoptions(m, opt, &len);
 272                 if (len != 0)
 273                         hlen = len;
 274         }
 275         ip = mtod(m, struct ip *);
 276
 277         /*
 278          * Fill in IP header.
 279          */
 280         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 281                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 282                 ip->ip_off &= IP_DF;
 283                 ip->ip_id = ip_newid();
 284                 ipstat.ips_localout++;
 285         } else {
 286                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 287         }
 288
 289 reroute:
 290         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 291
 292         dst = (struct sockaddr_in *)&ro->ro_dst;
 293         /*
 294          * If there is a cached route,
 295          * check that it is to the same destination
 296          * and is still up.  If not, free it and try again.
 297          * The address family should also be checked in case of sharing the
 298          * cache with IPv6.
 299          */
 300         if (ro->ro_rt &&
 301             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 302              dst->sin_family != AF_INET ||
 303              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 304                 rtfree(ro->ro_rt);
 305                 ro->ro_rt = (struct rtentry *)NULL;
 306         }
 307         if (ro->ro_rt == NULL) {
 308                 bzero(dst, sizeof *dst);
 309                 dst->sin_family = AF_INET;
 310                 dst->sin_len = sizeof *dst;
 311                 dst->sin_addr = pkt_dst;
 312         }
 313         /*
 314          * If routing to interface only,
 315          * short circuit routing lookup.
 316          */
 317         if (flags & IP_ROUTETOIF) {
 318                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 319                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 320                         ipstat.ips_noroute++;
 321                         error = ENETUNREACH;
 322                         goto bad;
 323                 }
 324                 ifp = ia->ia_ifp;
 325                 ip->ip_ttl = 1;
 326                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 327         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 328                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 329                 /*
 330                  * Bypass the normal routing lookup for multicast
 331                  * packets if the interface is specified.
 332                  */
 333                 ifp = imo->imo_multicast_ifp;
 334                 ia = IFP_TO_IA(ifp);
 335                 isbroadcast = 0;        /* fool gcc */
 336         } else {
 337                 /*
 338                  * If this is the case, we probably don't want to allocate
 339                  * a protocol-cloned route since we didn't get one from the
 340                  * ULP.  This lets TCP do its thing, while not burdening
 341                  * forwarding or ICMP with the overhead of cloning a route.
 342                  * Of course, we still want to do any cloning requested by
 343                  * the link layer, as this is probably required in all cases
 344                  * for correct operation (as it is for ARP).
 345                  */
 346                 if (ro->ro_rt == NULL)
 347                         rtalloc_ign(ro, RTF_PRCLONING);
 348                 if (ro->ro_rt == NULL) {
 349                         ipstat.ips_noroute++;
 350                         error = EHOSTUNREACH;
 351                         goto bad;
 352                 }
 353                 ia = ifatoia(ro->ro_rt->rt_ifa);
 354                 ifp = ro->ro_rt->rt_ifp;
 355                 ro->ro_rt->rt_use++;
 356                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 357                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 358                 if (ro->ro_rt->rt_flags & RTF_HOST)
 359                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 360                 else
 361                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 362         }
 363         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 364                 struct in_multi *inm;
 365
 366                 m->m_flags |= M_MCAST;
 367                 /*
 368                  * IP destination address is multicast.  Make sure "dst"
 369                  * still points to the address in "ro".  (It may have been
 370                  * changed to point to a gateway address, above.)
 371                  */
 372                 dst = (struct sockaddr_in *)&ro->ro_dst;
 373                 /*
 374                  * See if the caller provided any multicast options
 375                  */
 376                 if (imo != NULL) {
 377                         ip->ip_ttl = imo->imo_multicast_ttl;
 378                         if (imo->imo_multicast_vif != -1)
 379                                 ip->ip_src.s_addr =
 380                                     ip_mcast_src ?
 381                                     ip_mcast_src(imo->imo_multicast_vif) :
 382                                     INADDR_ANY;
 383                 } else
 384                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 385                 /*
 386                  * Confirm that the outgoing interface supports multicast.
 387                  */
 388                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 389                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 390                                 ipstat.ips_noroute++;
 391                                 error = ENETUNREACH;
 392                                 goto bad;
 393                         }
 394                 }
 395                 /*
 396                  * If source address not specified yet, use address
 397                  * of outgoing interface.
 398                  */
 399                 if (ip->ip_src.s_addr == INADDR_ANY) {
 400                         /* Interface may have no addresses. */
 401                         if (ia != NULL)
 402                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 403                 }
 404
 405                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 406                 if (inm != NULL &&
 407                    (imo == NULL || imo->imo_multicast_loop)) {
 408                         /*
 409                          * If we belong to the destination multicast group
 410                          * on the outgoing interface, and the caller did not
 411                          * forbid loopback, loop back a copy.
 412                          */
 413                         ip_mloopback(ifp, m, dst, hlen);
 414                 }
 415                 else {
 416                         /*
 417                          * If we are acting as a multicast router, perform
 418                          * multicast forwarding as if the packet had just
 419                          * arrived on the interface to which we are about
 420                          * to send.  The multicast forwarding function
 421                          * recursively calls this function, using the
 422                          * IP_FORWARDING flag to prevent infinite recursion.
 423                          *
 424                          * Multicasts that are looped back by ip_mloopback(),
 425                          * above, will be forwarded by the ip_input() routine,
 426                          * if necessary.
 427                          */
 428                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 429                                 /*
 430                                  * If rsvp daemon is not running, do not
 431                                  * set ip_moptions. This ensures that the packet
 432                                  * is multicast and not just sent down one link
 433                                  * as prescribed by rsvpd.
 434                                  */
 435                                 if (!rsvp_on)
 436                                         imo = NULL;
 437                                 if (ip_mforward &&
 438                                     ip_mforward(ip, ifp, m, imo) != 0) {
 439                                         m_freem(m);
 440                                         goto done;
 441                                 }
 442                         }
 443                 }
 444
 445                 /*
 446                  * Multicasts with a time-to-live of zero may be looped-
 447                  * back, above, but must not be transmitted on a network.
 448                  * Also, multicasts addressed to the loopback interface
 449                  * are not sent -- the above call to ip_mloopback() will
 450                  * loop back a copy if this host actually belongs to the
 451                  * destination group on the loopback interface.
 452                  */
 453                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 454                         m_freem(m);
 455                         goto done;
 456                 }
 457
 458                 goto sendit;
 459         } else {
 460                 m->m_flags &= ~M_MCAST;
 461         }
 462
 463         /*
 464          * If the source address is not specified yet, use the address
 465          * of the outoing interface. In case, keep note we did that, so
 466          * if the the firewall changes the next-hop causing the output
 467          * interface to change, we can fix that.
 468          */
 469         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 470                 /* Interface may have no addresses. */
 471                 if (ia != NULL) {
 472                         ip->ip_src = IA_SIN(ia)->sin_addr;
 473                         src_was_INADDR_ANY = 1;
 474                 }
 475         }
 476
 477 #ifdef ALTQ
 478         /*
 479          * Disable packet drop hack.
 480          * Packetdrop should be done by queueing.
 481          */
 482 #else /* !ALTQ */
 483         /*
 484          * Verify that we have any chance at all of being able to queue
 485          *      the packet or packet fragments
 486          */
 487         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 488                 ifp->if_snd.ifq_maxlen) {
 489                         error = ENOBUFS;
 490                         ipstat.ips_odropped++;
 491                         goto bad;
 492         }
 493 #endif /* !ALTQ */
 494
 495         /*
 496          * Look for broadcast address and
 497          * verify user is allowed to send
 498          * such a packet.
 499          */
 500         if (isbroadcast) {
 501                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 502                         error = EADDRNOTAVAIL;
 503                         goto bad;
 504                 }
 505                 if (!(flags & IP_ALLOWBROADCAST)) {
 506                         error = EACCES;
 507                         goto bad;
 508                 }
 509                 /* don't allow broadcast messages to be fragmented */
 510                 if (ip->ip_len > ifp->if_mtu) {
 511                         error = EMSGSIZE;
 512                         goto bad;
 513                 }
 514                 m->m_flags |= M_BCAST;
 515         } else {
 516                 m->m_flags &= ~M_BCAST;
 517         }
 518
 519 sendit:
 520 #ifdef IPSEC
 521         /* get SP for this packet */
 522         if (so == NULL)
 523                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 524         else
 525                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 526
 527         if (sp == NULL) {
 528                 ipsecstat.out_inval++;
 529                 goto bad;
 530         }
 531
 532         error = 0;
 533
 534         /* check policy */
 535         switch (sp->policy) {
 536         case IPSEC_POLICY_DISCARD:
 537                 /*
 538                  * This packet is just discarded.
 539                  */
 540                 ipsecstat.out_polvio++;
 541                 goto bad;
 542
 543         case IPSEC_POLICY_BYPASS:
 544         case IPSEC_POLICY_NONE:
 545                 /* no need to do IPsec. */
 546                 goto skip_ipsec;
 547
 548         case IPSEC_POLICY_IPSEC:
 549                 if (sp->req == NULL) {
 550                         /* acquire a policy */
 551                         error = key_spdacquire(sp);
 552                         goto bad;
 553                 }
 554                 break;
 555
 556         case IPSEC_POLICY_ENTRUST:
 557         default:
 558                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 559         }
 560     {
 561         struct ipsec_output_state state;
 562         bzero(&state, sizeof state);
 563         state.m = m;
 564         if (flags & IP_ROUTETOIF) {
 565                 state.ro = &iproute;
 566                 bzero(&iproute, sizeof iproute);
 567         } else
 568                 state.ro = ro;
 569         state.dst = (struct sockaddr *)dst;
 570
 571         ip->ip_sum = 0;
 572
 573         /*
 574          * XXX
 575          * delayed checksums are not currently compatible with IPsec
 576          */
 577         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 578                 in_delayed_cksum(m);
 579                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 580         }
 581
 582         ip->ip_len = htons(ip->ip_len);
 583         ip->ip_off = htons(ip->ip_off);
 584
 585         error = ipsec4_output(&state, sp, flags);
 586
 587         m = state.m;
 588         if (flags & IP_ROUTETOIF) {
 589                 /*
 590                  * if we have tunnel mode SA, we may need to ignore
 591                  * IP_ROUTETOIF.
 592                  */
 593                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 594                         flags &= ~IP_ROUTETOIF;
 595                         ro = state.ro;
 596                 }
 597         } else
 598                 ro = state.ro;
 599         dst = (struct sockaddr_in *)state.dst;
 600         if (error) {
 601                 /* mbuf is already reclaimed in ipsec4_output. */
 602                 m0 = NULL;
 603                 switch (error) {
 604                 case EHOSTUNREACH:
 605                 case ENETUNREACH:
 606                 case EMSGSIZE:
 607                 case ENOBUFS:
 608                 case ENOMEM:
 609                         break;
 610                 default:
 611                         kprintf("ip4_output (ipsec): error code %d\n", error);
 612                         /*fall through*/
 613                 case ENOENT:
 614                         /* don't show these error codes to the user */
 615                         error = 0;
 616                         break;
 617                 }
 618                 goto bad;
 619         }
 620     }
 621
 622         /* be sure to update variables that are affected by ipsec4_output() */
 623         ip = mtod(m, struct ip *);
 624 #ifdef _IP_VHL
 625         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 626 #else
 627         hlen = ip->ip_hl << 2;
 628 #endif
 629         if (ro->ro_rt == NULL) {
 630                 if (!(flags & IP_ROUTETOIF)) {
 631                         kprintf("ip_output: "
 632                                 "can't update route after IPsec processing\n");
 633                         error = EHOSTUNREACH;   /*XXX*/
 634                         goto bad;
 635                 }
 636         } else {
 637                 ia = ifatoia(ro->ro_rt->rt_ifa);
 638                 ifp = ro->ro_rt->rt_ifp;
 639         }
 640
 641         /* make it flipped, again. */
 642         ip->ip_len = ntohs(ip->ip_len);
 643         ip->ip_off = ntohs(ip->ip_off);
 644 skip_ipsec:
 645 #endif /*IPSEC*/
 646 #ifdef FAST_IPSEC
 647         /*
 648          * Check the security policy (SP) for the packet and, if
 649          * required, do IPsec-related processing.  There are two
 650          * cases here; the first time a packet is sent through
 651          * it will be untagged and handled by ipsec4_checkpolicy.
 652          * If the packet is resubmitted to ip_output (e.g. after
 653          * AH, ESP, etc. processing), there will be a tag to bypass
 654          * the lookup and related policy checking.
 655          */
 656         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 657         crit_enter();
 658         if (mtag != NULL) {
 659                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 660                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 661                 if (sp == NULL)
 662                         error = -EINVAL;        /* force silent drop */
 663                 m_tag_delete(m, mtag);
 664         } else {
 665                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 666                                         &error, inp);
 667         }
 668         /*
 669          * There are four return cases:
 670          *    sp != NULL                    apply IPsec policy
 671          *    sp == NULL, error == 0        no IPsec handling needed
 672          *    sp == NULL, error == -EINVAL  discard packet w/o error
 673          *    sp == NULL, error != 0        discard packet, report error
 674          */
 675         if (sp != NULL) {
 676                 /* Loop detection, check if ipsec processing already done */
 677                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 678                 for (mtag = m_tag_first(m); mtag != NULL;
 679                      mtag = m_tag_next(m, mtag)) {
 680                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 681                                 continue;
 682                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 683                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 684                                 continue;
 685                         /*
 686                          * Check if policy has an SA associated with it.
 687                          * This can happen when an SP has yet to acquire
 688                          * an SA; e.g. on first reference.  If it occurs,
 689                          * then we let ipsec4_process_packet do its thing.
 690                          */
 691                         if (sp->req->sav == NULL)
 692                                 break;
 693                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 694                         if (tdbi->spi == sp->req->sav->spi &&
 695                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 696                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 697                                  sizeof(union sockaddr_union)) == 0) {
 698                                 /*
 699                                  * No IPsec processing is needed, free
 700                                  * reference to SP.
 701                                  *
 702                                  * NB: null pointer to avoid free at
 703                                  *     done: below.
 704                                  */
 705                                 KEY_FREESP(&sp), sp = NULL;
 706                                 crit_exit();
 707                                 goto spd_done;
 708                         }
 709                 }
 710
 711                 /*
 712                  * Do delayed checksums now because we send before
 713                  * this is done in the normal processing path.
 714                  */
 715                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 716                         in_delayed_cksum(m);
 717                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 718                 }
 719
 720                 ip->ip_len = htons(ip->ip_len);
 721                 ip->ip_off = htons(ip->ip_off);
 722
 723                 /* NB: callee frees mbuf */
 724                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 725                 /*
 726                  * Preserve KAME behaviour: ENOENT can be returned
 727                  * when an SA acquire is in progress.  Don't propagate
 728                  * this to user-level; it confuses applications.
 729                  *
 730                  * XXX this will go away when the SADB is redone.
 731                  */
 732                 if (error == ENOENT)
 733                         error = 0;
 734                 crit_exit();
 735                 goto done;
 736         } else {
 737                 crit_exit();
 738
 739                 if (error != 0) {
 740                         /*
 741                          * Hack: -EINVAL is used to signal that a packet
 742                          * should be silently discarded.  This is typically
 743                          * because we asked key management for an SA and
 744                          * it was delayed (e.g. kicked up to IKE).
 745                          */
 746                         if (error == -EINVAL)
 747                                 error = 0;
 748                         goto bad;
 749                 } else {
 750                         /* No IPsec processing for this packet. */
 751                 }
 752 #ifdef notyet
 753                 /*
 754                  * If deferred crypto processing is needed, check that
 755                  * the interface supports it.
 756                  */
 757                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 758                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 759                         /* notify IPsec to do its own crypto */
 760                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 761                         error = EHOSTUNREACH;
 762                         goto bad;
 763                 }
 764 #endif
 765         }
 766 spd_done:
 767 #endif /* FAST_IPSEC */
 768
 769         /* We are already being fwd'd from a firewall. */
 770         if (next_hop != NULL)
 771                 goto pass;
 772
 773         /*
 774          * IpHack's section.
 775          * - Xlate: translate packet's addr/port (NAT).
 776          * - Firewall: deny/allow/etc.
 777          * - Wrap: fake packet's addr/port <unimpl.>
 778          * - Encapsulate: put it in another IP and send out. <unimp.>
 779          */
 780
 781         /*
 782          * Run through list of hooks for output packets.
 783          */
 784         if (pfil_has_hooks(&inet_pfil_hook)) {
 785                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 786                 if (error != 0 || m == NULL)
 787                         goto done;
 788                 ip = mtod(m, struct ip *);
 789         }
 790
 791         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 792                 /*
 793                  * Check dst to make sure it is directly reachable on the
 794                  * interface we previously thought it was.
 795                  * If it isn't (which may be likely in some situations) we have
 796                  * to re-route it (ie, find a route for the next-hop and the
 797                  * associated interface) and set them here. This is nested
 798                  * forwarding which in most cases is undesirable, except where
 799                  * such control is nigh impossible. So we do it here.
 800                  * And I'm babbling.
 801                  */
 802                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 803                 KKASSERT(mtag != NULL);
 804                 next_hop = m_tag_data(mtag);
 805
 806                 /*
 807                  * Try local forwarding first
 808                  */
 809                 if (ip_localforward(m, next_hop))
 810                         goto done;
 811
 812                 /*
 813                  * Relocate the route based on next_hop.
 814                  * If the current route is inp's cache, keep it untouched.
 815                  */
 816                 if (ro == &iproute && ro->ro_rt != NULL) {
 817                         RTFREE(ro->ro_rt);
 818                         ro->ro_rt = NULL;
 819                 }
 820                 ro = &iproute;
 821                 bzero(ro, sizeof *ro);
 822
 823                 /*
 824                  * Forwarding to broadcast address is not allowed.
 825                  * XXX Should we follow IP_ROUTETOIF?
 826                  */
 827                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 828
 829                 /* We are doing forwarding now */
 830                 flags |= IP_FORWARDING;
 831
 832                 goto reroute;
 833         }
 834
 835         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 836                 struct dn_pkt *dn_pkt;
 837
 838                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 839                 KKASSERT(mtag != NULL);
 840                 dn_pkt = m_tag_data(mtag);
 841
 842                 /*
 843                  * Under certain cases it is not possible to recalculate
 844                  * 'ro' and 'dst', let alone 'flags', so just save them in
 845                  * dummynet tag and avoid the possible wrong reculcalation
 846                  * when we come back to ip_output() again.
 847                  *
 848                  * All other parameters have been already used and so they
 849                  * are not needed anymore.
 850                  * XXX if the ifp is deleted while a pkt is in dummynet,
 851                  * we are in trouble! (TODO use ifnet_detach_event)
 852                  *
 853                  * We need to copy *ro because for ICMP pkts (and maybe
 854                  * others) the caller passed a pointer into the stack;
 855                  * dst might also be a pointer into *ro so it needs to
 856                  * be updated.
 857                  */
 858                 dn_pkt->ro = *ro;
 859                 if (ro->ro_rt)
 860                         ro->ro_rt->rt_refcnt++;
 861                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 862                         /* 'dst' points into 'ro' */
 863                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 864                 }
 865                 dn_pkt->dn_dst = dst;
 866                 dn_pkt->flags = flags;
 867
 868                 ip_dn_queue(m);
 869                 goto done;
 870         }
 871 pass:
 872         /* 127/8 must not appear on wire - RFC1122. */
 873         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 874             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 875                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 876                         ipstat.ips_badaddr++;
 877                         error = EADDRNOTAVAIL;
 878                         goto bad;
 879                 }
 880         }
 881
 882         m->m_pkthdr.csum_flags |= CSUM_IP;
 883         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 884         if (sw_csum & CSUM_DELAY_DATA) {
 885                 in_delayed_cksum(m);
 886                 sw_csum &= ~CSUM_DELAY_DATA;
 887         }
 888         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 889
 890         /*
 891          * If small enough for interface, or the interface will take
 892          * care of the fragmentation for us, can just send directly.
 893          */
 894         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 895             !(ip->ip_off & IP_DF))) {
 896                 ip->ip_len = htons(ip->ip_len);
 897                 ip->ip_off = htons(ip->ip_off);
 898                 ip->ip_sum = 0;
 899                 if (sw_csum & CSUM_DELAY_IP) {
 900                         if (ip->ip_vhl == IP_VHL_BORING) {
 901                                 ip->ip_sum = in_cksum_hdr(ip);
 902                         } else {
 903                                 ip->ip_sum = in_cksum(m, hlen);
 904                         }
 905                 }
 906
 907                 /* Record statistics for this interface address. */
 908                 if (!(flags & IP_FORWARDING) && ia) {
 909                         ia->ia_ifa.if_opackets++;
 910                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 911                 }
 912
 913 #ifdef IPSEC
 914                 /* clean ipsec history once it goes out of the node */
 915                 ipsec_delaux(m);
 916 #endif
 917
 918 #ifdef MBUF_STRESS_TEST
 919                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 920                         struct mbuf *m1, *m2;
 921                         int length, tmp;
 922
 923                         tmp = length = m->m_pkthdr.len;
 924
 925                         while ((length -= mbuf_frag_size) >= 1) {
 926                                 m1 = m_split(m, length, MB_DONTWAIT);
 927                                 if (m1 == NULL)
 928                                         break;
 929                                 m2 = m;
 930                                 while (m2->m_next != NULL)
 931                                         m2 = m2->m_next;
 932                                 m2->m_next = m1;
 933                         }
 934                         m->m_pkthdr.len = tmp;
 935                 }
 936 #endif
 937
 938 #ifdef MPLS
 939                 if (!mpls_output_process(m, ro->ro_rt))
 940                         goto done;
 941 #endif
 942                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 943                                        ro->ro_rt);
 944                 goto done;
 945         }
 946
 947         if (ip->ip_off & IP_DF) {
 948                 error = EMSGSIZE;
 949                 /*
 950                  * This case can happen if the user changed the MTU
 951                  * of an interface after enabling IP on it.  Because
 952                  * most netifs don't keep track of routes pointing to
 953                  * them, there is no way for one to update all its
 954                  * routes when the MTU is changed.
 955                  */
 956                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 957                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 958                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 959                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 960                 }
 961                 ipstat.ips_cantfrag++;
 962                 goto bad;
 963         }
 964
 965         /*
 966          * Too large for interface; fragment if possible. If successful,
 967          * on return, m will point to a list of packets to be sent.
 968          */
 969         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
 970         if (error)
 971                 goto bad;
 972         for (; m; m = m0) {
 973                 m0 = m->m_nextpkt;
 974                 m->m_nextpkt = NULL;
 975 #ifdef IPSEC
 976                 /* clean ipsec history once it goes out of the node */
 977                 ipsec_delaux(m);
 978 #endif
 979                 if (error == 0) {
 980                         /* Record statistics for this interface address. */
 981                         if (ia != NULL) {
 982                                 ia->ia_ifa.if_opackets++;
 983                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 984                         }
 985 #ifdef MPLS
 986                         if (!mpls_output_process(m, ro->ro_rt))
 987                                 continue;
 988 #endif
 989                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 990                                                ro->ro_rt);
 991                 } else {
 992                         m_freem(m);
 993                 }
 994         }
 995
 996         if (error == 0)
 997                 ipstat.ips_fragmented++;
 998
 999 done:
1000         if (ro == &iproute && ro->ro_rt != NULL) {
1001                 RTFREE(ro->ro_rt);
1002                 ro->ro_rt = NULL;
1003         }
1004 #ifdef IPSEC
1005         if (sp != NULL) {
1006                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1007                         kprintf("DP ip_output call free SP:%p\n", sp));
1008                 key_freesp(sp);
1009         }
1010 #endif
1011 #ifdef FAST_IPSEC
1012         if (sp != NULL)
1013                 KEY_FREESP(&sp);
1014 #endif
1015         return (error);
1016 bad:
1017         m_freem(m);
1018         goto done;
1019 }
1020
1021 /*
1022  * Create a chain of fragments which fit the given mtu. m_frag points to the
1023  * mbuf to be fragmented; on return it points to the chain with the fragments.
1024  * Return 0 if no error. If error, m_frag may contain a partially built
1025  * chain of fragments that should be freed by the caller.
1026  *
1027  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1028  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1029  */
1030 int
1031 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1032             u_long if_hwassist_flags, int sw_csum)
1033 {
1034         int error = 0;
1035         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1036         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1037         int off;
1038         struct mbuf *m0 = *m_frag;      /* the original packet          */
1039         int firstlen;
1040         struct mbuf **mnext;
1041         int nfrags;
1042
1043         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1044                 ipstat.ips_cantfrag++;
1045                 return EMSGSIZE;
1046         }
1047
1048         /*
1049          * Must be able to put at least 8 bytes per fragment.
1050          */
1051         if (len < 8)
1052                 return EMSGSIZE;
1053
1054         /*
1055          * If the interface will not calculate checksums on
1056          * fragmented packets, then do it here.
1057          */
1058         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1059             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1060                 in_delayed_cksum(m0);
1061                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1062         }
1063
1064         if (len > PAGE_SIZE) {
1065                 /*
1066                  * Fragment large datagrams such that each segment
1067                  * contains a multiple of PAGE_SIZE amount of data,
1068                  * plus headers. This enables a receiver to perform
1069                  * page-flipping zero-copy optimizations.
1070                  *
1071                  * XXX When does this help given that sender and receiver
1072                  * could have different page sizes, and also mtu could
1073                  * be less than the receiver's page size ?
1074                  */
1075                 int newlen;
1076                 struct mbuf *m;
1077
1078                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1079                         off += m->m_len;
1080
1081                 /*
1082                  * firstlen (off - hlen) must be aligned on an
1083                  * 8-byte boundary
1084                  */
1085                 if (off < hlen)
1086                         goto smart_frag_failure;
1087                 off = ((off - hlen) & ~7) + hlen;
1088                 newlen = (~PAGE_MASK) & mtu;
1089                 if ((newlen + sizeof(struct ip)) > mtu) {
1090                         /* we failed, go back the default */
1091 smart_frag_failure:
1092                         newlen = len;
1093                         off = hlen + len;
1094                 }
1095                 len = newlen;
1096
1097         } else {
1098                 off = hlen + len;
1099         }
1100
1101         firstlen = off - hlen;
1102         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1103
1104         /*
1105          * Loop through length of segment after first fragment,
1106          * make new header and copy data of each part and link onto chain.
1107          * Here, m0 is the original packet, m is the fragment being created.
1108          * The fragments are linked off the m_nextpkt of the original
1109          * packet, which after processing serves as the first fragment.
1110          */
1111         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1112                 struct ip *mhip;        /* ip header on the fragment */
1113                 struct mbuf *m;
1114                 int mhlen = sizeof(struct ip);
1115
1116                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1117                 if (m == NULL) {
1118                         error = ENOBUFS;
1119                         ipstat.ips_odropped++;
1120                         goto done;
1121                 }
1122                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1123                 /*
1124                  * In the first mbuf, leave room for the link header, then
1125                  * copy the original IP header including options. The payload
1126                  * goes into an additional mbuf chain returned by m_copy().
1127                  */
1128                 m->m_data += max_linkhdr;
1129                 mhip = mtod(m, struct ip *);
1130                 *mhip = *ip;
1131                 if (hlen > sizeof(struct ip)) {
1132                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1133                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1134                 }
1135                 m->m_len = mhlen;
1136                 /* XXX do we need to add ip->ip_off below ? */
1137                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1138                 if (off + len >= ip->ip_len) {  /* last fragment */
1139                         len = ip->ip_len - off;
1140                         m->m_flags |= M_LASTFRAG;
1141                 } else
1142                         mhip->ip_off |= IP_MF;
1143                 mhip->ip_len = htons((u_short)(len + mhlen));
1144                 m->m_next = m_copy(m0, off, len);
1145                 if (m->m_next == NULL) {                /* copy failed */
1146                         m_free(m);
1147                         error = ENOBUFS;        /* ??? */
1148                         ipstat.ips_odropped++;
1149                         goto done;
1150                 }
1151                 m->m_pkthdr.len = mhlen + len;
1152                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1153                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1154                 mhip->ip_off = htons(mhip->ip_off);
1155                 mhip->ip_sum = 0;
1156                 if (sw_csum & CSUM_DELAY_IP)
1157                         mhip->ip_sum = in_cksum(m, mhlen);
1158                 *mnext = m;
1159                 mnext = &m->m_nextpkt;
1160         }
1161         ipstat.ips_ofragments += nfrags;
1162
1163         /* set first marker for fragment chain */
1164         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1165         m0->m_pkthdr.csum_data = nfrags;
1166
1167         /*
1168          * Update first fragment by trimming what's been copied out
1169          * and updating header.
1170          */
1171         m_adj(m0, hlen + firstlen - ip->ip_len);
1172         m0->m_pkthdr.len = hlen + firstlen;
1173         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1174         ip->ip_off |= IP_MF;
1175         ip->ip_off = htons(ip->ip_off);
1176         ip->ip_sum = 0;
1177         if (sw_csum & CSUM_DELAY_IP)
1178                 ip->ip_sum = in_cksum(m0, hlen);
1179
1180 done:
1181         *m_frag = m0;
1182         return error;
1183 }
1184
1185 void
1186 in_delayed_cksum(struct mbuf *m)
1187 {
1188         struct ip *ip;
1189         u_short csum, offset;
1190
1191         ip = mtod(m, struct ip *);
1192         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1193         csum = in_cksum_skip(m, ip->ip_len, offset);
1194         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1195                 csum = 0xffff;
1196         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1197
1198         if (offset + sizeof(u_short) > m->m_len) {
1199                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1200                     m->m_len, offset, ip->ip_p);
1201                 /*
1202                  * XXX
1203                  * this shouldn't happen, but if it does, the
1204                  * correct behavior may be to insert the checksum
1205                  * in the existing chain instead of rearranging it.
1206                  */
1207                 m = m_pullup(m, offset + sizeof(u_short));
1208         }
1209         *(u_short *)(m->m_data + offset) = csum;
1210 }
1211
1212 /*
1213  * Insert IP options into preformed packet.
1214  * Adjust IP destination as required for IP source routing,
1215  * as indicated by a non-zero in_addr at the start of the options.
1216  *
1217  * XXX This routine assumes that the packet has no options in place.
1218  */
1219 static struct mbuf *
1220 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1221 {
1222         struct ipoption *p = mtod(opt, struct ipoption *);
1223         struct mbuf *n;
1224         struct ip *ip = mtod(m, struct ip *);
1225         unsigned optlen;
1226
1227         optlen = opt->m_len - sizeof p->ipopt_dst;
1228         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1229                 *phlen = 0;
1230                 return (m);             /* XXX should fail */
1231         }
1232         if (p->ipopt_dst.s_addr)
1233                 ip->ip_dst = p->ipopt_dst;
1234         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1235                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1236                 if (n == NULL) {
1237                         *phlen = 0;
1238                         return (m);
1239                 }
1240                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1241                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1242                 m->m_len -= sizeof(struct ip);
1243                 m->m_data += sizeof(struct ip);
1244                 n->m_next = m;
1245                 m = n;
1246                 m->m_len = optlen + sizeof(struct ip);
1247                 m->m_data += max_linkhdr;
1248                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1249         } else {
1250                 m->m_data -= optlen;
1251                 m->m_len += optlen;
1252                 m->m_pkthdr.len += optlen;
1253                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1254         }
1255         ip = mtod(m, struct ip *);
1256         bcopy(p->ipopt_list, ip + 1, optlen);
1257         *phlen = sizeof(struct ip) + optlen;
1258         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1259         ip->ip_len += optlen;
1260         return (m);
1261 }
1262
1263 /*
1264  * Copy options from ip to jp,
1265  * omitting those not copied during fragmentation.
1266  */
1267 int
1268 ip_optcopy(struct ip *ip, struct ip *jp)
1269 {
1270         u_char *cp, *dp;
1271         int opt, optlen, cnt;
1272
1273         cp = (u_char *)(ip + 1);
1274         dp = (u_char *)(jp + 1);
1275         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1276         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1277                 opt = cp[0];
1278                 if (opt == IPOPT_EOL)
1279                         break;
1280                 if (opt == IPOPT_NOP) {
1281                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1282                         *dp++ = IPOPT_NOP;
1283                         optlen = 1;
1284                         continue;
1285                 }
1286
1287                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1288                     ("ip_optcopy: malformed ipv4 option"));
1289                 optlen = cp[IPOPT_OLEN];
1290                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1291                     ("ip_optcopy: malformed ipv4 option"));
1292
1293                 /* bogus lengths should have been caught by ip_dooptions */
1294                 if (optlen > cnt)
1295                         optlen = cnt;
1296                 if (IPOPT_COPIED(opt)) {
1297                         bcopy(cp, dp, optlen);
1298                         dp += optlen;
1299                 }
1300         }
1301         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1302                 *dp++ = IPOPT_EOL;
1303         return (optlen);
1304 }
1305
1306 /*
1307  * IP socket option processing.
1308  */
1309 int
1310 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1311 {
1312         struct  inpcb *inp = so->so_pcb;
1313         int     error, optval;
1314
1315         error = optval = 0;
1316         if (sopt->sopt_level != IPPROTO_IP) {
1317                 return (EINVAL);
1318         }
1319
1320         switch (sopt->sopt_dir) {
1321         case SOPT_SET:
1322                 switch (sopt->sopt_name) {
1323                 case IP_OPTIONS:
1324 #ifdef notyet
1325                 case IP_RETOPTS:
1326 #endif
1327                 {
1328                         struct mbuf *m;
1329                         if (sopt->sopt_valsize > MLEN) {
1330                                 error = EMSGSIZE;
1331                                 break;
1332                         }
1333                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1334                         if (m == NULL) {
1335                                 error = ENOBUFS;
1336                                 break;
1337                         }
1338                         m->m_len = sopt->sopt_valsize;
1339                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1340                                               m->m_len);
1341                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1342                                            m));
1343                 }
1344
1345                 case IP_TOS:
1346                 case IP_TTL:
1347                 case IP_MINTTL:
1348                 case IP_RECVOPTS:
1349                 case IP_RECVRETOPTS:
1350                 case IP_RECVDSTADDR:
1351                 case IP_RECVIF:
1352                 case IP_RECVTTL:
1353                 case IP_FAITH:
1354                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1355                                              sizeof optval);
1356                         if (error)
1357                                 break;
1358                         switch (sopt->sopt_name) {
1359                         case IP_TOS:
1360                                 inp->inp_ip_tos = optval;
1361                                 break;
1362
1363                         case IP_TTL:
1364                                 inp->inp_ip_ttl = optval;
1365                                 break;
1366                         case IP_MINTTL:
1367                                 if (optval > 0 && optval <= MAXTTL)
1368                                         inp->inp_ip_minttl = optval;
1369                                 else
1370                                         error = EINVAL;
1371                                 break;
1372 #define OPTSET(bit) \
1373         if (optval) \
1374                 inp->inp_flags |= bit; \
1375         else \
1376                 inp->inp_flags &= ~bit;
1377
1378                         case IP_RECVOPTS:
1379                                 OPTSET(INP_RECVOPTS);
1380                                 break;
1381
1382                         case IP_RECVRETOPTS:
1383                                 OPTSET(INP_RECVRETOPTS);
1384                                 break;
1385
1386                         case IP_RECVDSTADDR:
1387                                 OPTSET(INP_RECVDSTADDR);
1388                                 break;
1389
1390                         case IP_RECVIF:
1391                                 OPTSET(INP_RECVIF);
1392                                 break;
1393
1394                         case IP_RECVTTL:
1395                                 OPTSET(INP_RECVTTL);
1396                                 break;
1397
1398                         case IP_FAITH:
1399                                 OPTSET(INP_FAITH);
1400                                 break;
1401                         }
1402                         break;
1403 #undef OPTSET
1404
1405                 case IP_MULTICAST_IF:
1406                 case IP_MULTICAST_VIF:
1407                 case IP_MULTICAST_TTL:
1408                 case IP_MULTICAST_LOOP:
1409                 case IP_ADD_MEMBERSHIP:
1410                 case IP_DROP_MEMBERSHIP:
1411                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1412                         break;
1413
1414                 case IP_PORTRANGE:
1415                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1416                                             sizeof optval);
1417                         if (error)
1418                                 break;
1419
1420                         switch (optval) {
1421                         case IP_PORTRANGE_DEFAULT:
1422                                 inp->inp_flags &= ~(INP_LOWPORT);
1423                                 inp->inp_flags &= ~(INP_HIGHPORT);
1424                                 break;
1425
1426                         case IP_PORTRANGE_HIGH:
1427                                 inp->inp_flags &= ~(INP_LOWPORT);
1428                                 inp->inp_flags |= INP_HIGHPORT;
1429                                 break;
1430
1431                         case IP_PORTRANGE_LOW:
1432                                 inp->inp_flags &= ~(INP_HIGHPORT);
1433                                 inp->inp_flags |= INP_LOWPORT;
1434                                 break;
1435
1436                         default:
1437                                 error = EINVAL;
1438                                 break;
1439                         }
1440                         break;
1441
1442 #if defined(IPSEC) || defined(FAST_IPSEC)
1443                 case IP_IPSEC_POLICY:
1444                 {
1445                         caddr_t req;
1446                         size_t len = 0;
1447                         int priv;
1448                         struct mbuf *m;
1449                         int optname;
1450
1451                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1452                                 break;
1453                         soopt_to_mbuf(sopt, m);
1454                         priv = (sopt->sopt_td != NULL &&
1455                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1456                         req = mtod(m, caddr_t);
1457                         len = m->m_len;
1458                         optname = sopt->sopt_name;
1459                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1460                         m_freem(m);
1461                         break;
1462                 }
1463 #endif /*IPSEC*/
1464
1465                 default:
1466                         error = ENOPROTOOPT;
1467                         break;
1468                 }
1469                 break;
1470
1471         case SOPT_GET:
1472                 switch (sopt->sopt_name) {
1473                 case IP_OPTIONS:
1474                 case IP_RETOPTS:
1475                         if (inp->inp_options)
1476                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1477                                                            char *),
1478                                                 inp->inp_options->m_len);
1479                         else
1480                                 sopt->sopt_valsize = 0;
1481                         break;
1482
1483                 case IP_TOS:
1484                 case IP_TTL:
1485                 case IP_MINTTL:
1486                 case IP_RECVOPTS:
1487                 case IP_RECVRETOPTS:
1488                 case IP_RECVDSTADDR:
1489                 case IP_RECVTTL:
1490                 case IP_RECVIF:
1491                 case IP_PORTRANGE:
1492                 case IP_FAITH:
1493                         switch (sopt->sopt_name) {
1494
1495                         case IP_TOS:
1496                                 optval = inp->inp_ip_tos;
1497                                 break;
1498
1499                         case IP_TTL:
1500                                 optval = inp->inp_ip_ttl;
1501                                 break;
1502                         case IP_MINTTL:
1503                                 optval = inp->inp_ip_minttl;
1504                                 break;
1505
1506 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1507
1508                         case IP_RECVOPTS:
1509                                 optval = OPTBIT(INP_RECVOPTS);
1510                                 break;
1511
1512                         case IP_RECVRETOPTS:
1513                                 optval = OPTBIT(INP_RECVRETOPTS);
1514                                 break;
1515
1516                         case IP_RECVDSTADDR:
1517                                 optval = OPTBIT(INP_RECVDSTADDR);
1518                                 break;
1519
1520                         case IP_RECVTTL:
1521                                 optval = OPTBIT(INP_RECVTTL);
1522                                 break;
1523
1524                         case IP_RECVIF:
1525                                 optval = OPTBIT(INP_RECVIF);
1526                                 break;
1527
1528                         case IP_PORTRANGE:
1529                                 if (inp->inp_flags & INP_HIGHPORT)
1530                                         optval = IP_PORTRANGE_HIGH;
1531                                 else if (inp->inp_flags & INP_LOWPORT)
1532                                         optval = IP_PORTRANGE_LOW;
1533                                 else
1534                                         optval = 0;
1535                                 break;
1536
1537                         case IP_FAITH:
1538                                 optval = OPTBIT(INP_FAITH);
1539                                 break;
1540                         }
1541                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1542                         break;
1543
1544                 case IP_MULTICAST_IF:
1545                 case IP_MULTICAST_VIF:
1546                 case IP_MULTICAST_TTL:
1547                 case IP_MULTICAST_LOOP:
1548                 case IP_ADD_MEMBERSHIP:
1549                 case IP_DROP_MEMBERSHIP:
1550                         error = ip_getmoptions(sopt, inp->inp_moptions);
1551                         break;
1552
1553 #if defined(IPSEC) || defined(FAST_IPSEC)
1554                 case IP_IPSEC_POLICY:
1555                 {
1556                         struct mbuf *m = NULL;
1557                         caddr_t req = NULL;
1558                         size_t len = 0;
1559
1560                         if (m != NULL) {
1561                                 req = mtod(m, caddr_t);
1562                                 len = m->m_len;
1563                         }
1564                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1565                         if (error == 0)
1566                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1567                         if (error == 0)
1568                                 m_freem(m);
1569                         break;
1570                 }
1571 #endif /*IPSEC*/
1572
1573                 default:
1574                         error = ENOPROTOOPT;
1575                         break;
1576                 }
1577                 break;
1578         }
1579         return (error);
1580 }
1581
1582 /*
1583  * Set up IP options in pcb for insertion in output packets.
1584  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1585  * with destination address if source routed.
1586  */
1587 static int
1588 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1589 {
1590         int cnt, optlen;
1591         u_char *cp;
1592         u_char opt;
1593
1594         /* turn off any old options */
1595         if (*pcbopt)
1596                 m_free(*pcbopt);
1597         *pcbopt = 0;
1598         if (m == NULL || m->m_len == 0) {
1599                 /*
1600                  * Only turning off any previous options.
1601                  */
1602                 if (m != NULL)
1603                         m_free(m);
1604                 return (0);
1605         }
1606
1607         if (m->m_len % sizeof(int32_t))
1608                 goto bad;
1609         /*
1610          * IP first-hop destination address will be stored before
1611          * actual options; move other options back
1612          * and clear it when none present.
1613          */
1614         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1615                 goto bad;
1616         cnt = m->m_len;
1617         m->m_len += sizeof(struct in_addr);
1618         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1619         ovbcopy(mtod(m, caddr_t), cp, cnt);
1620         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1621
1622         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1623                 opt = cp[IPOPT_OPTVAL];
1624                 if (opt == IPOPT_EOL)
1625                         break;
1626                 if (opt == IPOPT_NOP)
1627                         optlen = 1;
1628                 else {
1629                         if (cnt < IPOPT_OLEN + sizeof *cp)
1630                                 goto bad;
1631                         optlen = cp[IPOPT_OLEN];
1632                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1633                                 goto bad;
1634                 }
1635                 switch (opt) {
1636
1637                 default:
1638                         break;
1639
1640                 case IPOPT_LSRR:
1641                 case IPOPT_SSRR:
1642                         /*
1643                          * user process specifies route as:
1644                          *      ->A->B->C->D
1645                          * D must be our final destination (but we can't
1646                          * check that since we may not have connected yet).
1647                          * A is first hop destination, which doesn't appear in
1648                          * actual IP option, but is stored before the options.
1649                          */
1650                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1651                                 goto bad;
1652                         m->m_len -= sizeof(struct in_addr);
1653                         cnt -= sizeof(struct in_addr);
1654                         optlen -= sizeof(struct in_addr);
1655                         cp[IPOPT_OLEN] = optlen;
1656                         /*
1657                          * Move first hop before start of options.
1658                          */
1659                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1660                               sizeof(struct in_addr));
1661                         /*
1662                          * Then copy rest of options back
1663                          * to close up the deleted entry.
1664                          */
1665                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1666                                 &cp[IPOPT_OFFSET+1],
1667                                 cnt - (IPOPT_MINOFF - 1));
1668                         break;
1669                 }
1670         }
1671         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1672                 goto bad;
1673         *pcbopt = m;
1674         return (0);
1675
1676 bad:
1677         m_free(m);
1678         return (EINVAL);
1679 }
1680
1681 /*
1682  * XXX
1683  * The whole multicast option thing needs to be re-thought.
1684  * Several of these options are equally applicable to non-multicast
1685  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1686  * standard option (IP_TTL).
1687  */
1688
1689 /*
1690  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1691  */
1692 static struct ifnet *
1693 ip_multicast_if(struct in_addr *a, int *ifindexp)
1694 {
1695         int ifindex;
1696         struct ifnet *ifp;
1697
1698         if (ifindexp)
1699                 *ifindexp = 0;
1700         if (ntohl(a->s_addr) >> 24 == 0) {
1701                 ifindex = ntohl(a->s_addr) & 0xffffff;
1702                 if (ifindex < 0 || if_index < ifindex)
1703                         return NULL;
1704                 ifp = ifindex2ifnet[ifindex];
1705                 if (ifindexp)
1706                         *ifindexp = ifindex;
1707         } else {
1708                 ifp = INADDR_TO_IFP(a);
1709         }
1710         return ifp;
1711 }
1712
1713 /*
1714  * Set the IP multicast options in response to user setsockopt().
1715  */
1716 static int
1717 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1718 {
1719         int error = 0;
1720         int i;
1721         struct in_addr addr;
1722         struct ip_mreq mreq;
1723         struct ifnet *ifp;
1724         struct ip_moptions *imo = *imop;
1725         int ifindex;
1726
1727         if (imo == NULL) {
1728                 /*
1729                  * No multicast option buffer attached to the pcb;
1730                  * allocate one and initialize to default values.
1731                  */
1732                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1733
1734                 *imop = imo;
1735                 imo->imo_multicast_ifp = NULL;
1736                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1737                 imo->imo_multicast_vif = -1;
1738                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1739                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1740                 imo->imo_num_memberships = 0;
1741         }
1742         switch (sopt->sopt_name) {
1743         /* store an index number for the vif you wanna use in the send */
1744         case IP_MULTICAST_VIF:
1745                 if (legal_vif_num == 0) {
1746                         error = EOPNOTSUPP;
1747                         break;
1748                 }
1749                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1750                 if (error)
1751                         break;
1752                 if (!legal_vif_num(i) && (i != -1)) {
1753                         error = EINVAL;
1754                         break;
1755                 }
1756                 imo->imo_multicast_vif = i;
1757                 break;
1758
1759         case IP_MULTICAST_IF:
1760                 /*
1761                  * Select the interface for outgoing multicast packets.
1762                  */
1763                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1764                 if (error)
1765                         break;
1766
1767                 /*
1768                  * INADDR_ANY is used to remove a previous selection.
1769                  * When no interface is selected, a default one is
1770                  * chosen every time a multicast packet is sent.
1771                  */
1772                 if (addr.s_addr == INADDR_ANY) {
1773                         imo->imo_multicast_ifp = NULL;
1774                         break;
1775                 }
1776                 /*
1777                  * The selected interface is identified by its local
1778                  * IP address.  Find the interface and confirm that
1779                  * it supports multicasting.
1780                  */
1781                 crit_enter();
1782                 ifp = ip_multicast_if(&addr, &ifindex);
1783                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1784                         crit_exit();
1785                         error = EADDRNOTAVAIL;
1786                         break;
1787                 }
1788                 imo->imo_multicast_ifp = ifp;
1789                 if (ifindex)
1790                         imo->imo_multicast_addr = addr;
1791                 else
1792                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1793                 crit_exit();
1794                 break;
1795
1796         case IP_MULTICAST_TTL:
1797                 /*
1798                  * Set the IP time-to-live for outgoing multicast packets.
1799                  * The original multicast API required a char argument,
1800                  * which is inconsistent with the rest of the socket API.
1801                  * We allow either a char or an int.
1802                  */
1803                 if (sopt->sopt_valsize == 1) {
1804                         u_char ttl;
1805                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1806                         if (error)
1807                                 break;
1808                         imo->imo_multicast_ttl = ttl;
1809                 } else {
1810                         u_int ttl;
1811                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1812                         if (error)
1813                                 break;
1814                         if (ttl > 255)
1815                                 error = EINVAL;
1816                         else
1817                                 imo->imo_multicast_ttl = ttl;
1818                 }
1819                 break;
1820
1821         case IP_MULTICAST_LOOP:
1822                 /*
1823                  * Set the loopback flag for outgoing multicast packets.
1824                  * Must be zero or one.  The original multicast API required a
1825                  * char argument, which is inconsistent with the rest
1826                  * of the socket API.  We allow either a char or an int.
1827                  */
1828                 if (sopt->sopt_valsize == 1) {
1829                         u_char loop;
1830
1831                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1832                         if (error)
1833                                 break;
1834                         imo->imo_multicast_loop = !!loop;
1835                 } else {
1836                         u_int loop;
1837
1838                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1839                                             sizeof loop);
1840                         if (error)
1841                                 break;
1842                         imo->imo_multicast_loop = !!loop;
1843                 }
1844                 break;
1845
1846         case IP_ADD_MEMBERSHIP:
1847                 /*
1848                  * Add a multicast group membership.
1849                  * Group must be a valid IP multicast address.
1850                  */
1851                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1852                 if (error)
1853                         break;
1854
1855                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1856                         error = EINVAL;
1857                         break;
1858                 }
1859                 crit_enter();
1860                 /*
1861                  * If no interface address was provided, use the interface of
1862                  * the route to the given multicast address.
1863                  */
1864                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1865                         struct sockaddr_in dst;
1866                         struct rtentry *rt;
1867
1868                         bzero(&dst, sizeof(struct sockaddr_in));
1869                         dst.sin_len = sizeof(struct sockaddr_in);
1870                         dst.sin_family = AF_INET;
1871                         dst.sin_addr = mreq.imr_multiaddr;
1872                         rt = rtlookup((struct sockaddr *)&dst);
1873                         if (rt == NULL) {
1874                                 error = EADDRNOTAVAIL;
1875                                 crit_exit();
1876                                 break;
1877                         }
1878                         --rt->rt_refcnt;
1879                         ifp = rt->rt_ifp;
1880                 } else {
1881                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1882                 }
1883
1884                 /*
1885                  * See if we found an interface, and confirm that it
1886                  * supports multicast.
1887                  */
1888                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1889                         error = EADDRNOTAVAIL;
1890                         crit_exit();
1891                         break;
1892                 }
1893                 /*
1894                  * See if the membership already exists or if all the
1895                  * membership slots are full.
1896                  */
1897                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1898                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1899                             imo->imo_membership[i]->inm_addr.s_addr
1900                                                 == mreq.imr_multiaddr.s_addr)
1901                                 break;
1902                 }
1903                 if (i < imo->imo_num_memberships) {
1904                         error = EADDRINUSE;
1905                         crit_exit();
1906                         break;
1907                 }
1908                 if (i == IP_MAX_MEMBERSHIPS) {
1909                         error = ETOOMANYREFS;
1910                         crit_exit();
1911                         break;
1912                 }
1913                 /*
1914                  * Everything looks good; add a new record to the multicast
1915                  * address list for the given interface.
1916                  */
1917                 if ((imo->imo_membership[i] =
1918                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1919                         error = ENOBUFS;
1920                         crit_exit();
1921                         break;
1922                 }
1923                 ++imo->imo_num_memberships;
1924                 crit_exit();
1925                 break;
1926
1927         case IP_DROP_MEMBERSHIP:
1928                 /*
1929                  * Drop a multicast group membership.
1930                  * Group must be a valid IP multicast address.
1931                  */
1932                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1933                 if (error)
1934                         break;
1935
1936                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1937                         error = EINVAL;
1938                         break;
1939                 }
1940
1941                 crit_enter();
1942                 /*
1943                  * If an interface address was specified, get a pointer
1944                  * to its ifnet structure.
1945                  */
1946                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1947                         ifp = NULL;
1948                 else {
1949                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1950                         if (ifp == NULL) {
1951                                 error = EADDRNOTAVAIL;
1952                                 crit_exit();
1953                                 break;
1954                         }
1955                 }
1956                 /*
1957                  * Find the membership in the membership array.
1958                  */
1959                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1960                         if ((ifp == NULL ||
1961                              imo->imo_membership[i]->inm_ifp == ifp) &&
1962                             imo->imo_membership[i]->inm_addr.s_addr ==
1963                             mreq.imr_multiaddr.s_addr)
1964                                 break;
1965                 }
1966                 if (i == imo->imo_num_memberships) {
1967                         error = EADDRNOTAVAIL;
1968                         crit_exit();
1969                         break;
1970                 }
1971                 /*
1972                  * Give up the multicast address record to which the
1973                  * membership points.
1974                  */
1975                 in_delmulti(imo->imo_membership[i]);
1976                 /*
1977                  * Remove the gap in the membership array.
1978                  */
1979                 for (++i; i < imo->imo_num_memberships; ++i)
1980                         imo->imo_membership[i-1] = imo->imo_membership[i];
1981                 --imo->imo_num_memberships;
1982                 crit_exit();
1983                 break;
1984
1985         default:
1986                 error = EOPNOTSUPP;
1987                 break;
1988         }
1989
1990         /*
1991          * If all options have default values, no need to keep the mbuf.
1992          */
1993         if (imo->imo_multicast_ifp == NULL &&
1994             imo->imo_multicast_vif == -1 &&
1995             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
1996             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
1997             imo->imo_num_memberships == 0) {
1998                 kfree(*imop, M_IPMOPTS);
1999                 *imop = NULL;
2000         }
2001
2002         return (error);
2003 }
2004
2005 /*
2006  * Return the IP multicast options in response to user getsockopt().
2007  */
2008 static int
2009 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2010 {
2011         struct in_addr addr;
2012         struct in_ifaddr *ia;
2013         int error, optval;
2014         u_char coptval;
2015
2016         error = 0;
2017         switch (sopt->sopt_name) {
2018         case IP_MULTICAST_VIF:
2019                 if (imo != NULL)
2020                         optval = imo->imo_multicast_vif;
2021                 else
2022                         optval = -1;
2023                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2024                 break;
2025
2026         case IP_MULTICAST_IF:
2027                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2028                         addr.s_addr = INADDR_ANY;
2029                 else if (imo->imo_multicast_addr.s_addr) {
2030                         /* return the value user has set */
2031                         addr = imo->imo_multicast_addr;
2032                 } else {
2033                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2034                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2035                                 : IA_SIN(ia)->sin_addr.s_addr;
2036                 }
2037                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2038                 break;
2039
2040         case IP_MULTICAST_TTL:
2041                 if (imo == NULL)
2042                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2043                 else
2044                         optval = coptval = imo->imo_multicast_ttl;
2045                 if (sopt->sopt_valsize == 1)
2046                         soopt_from_kbuf(sopt, &coptval, 1);
2047                 else
2048                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2049                 break;
2050
2051         case IP_MULTICAST_LOOP:
2052                 if (imo == NULL)
2053                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2054                 else
2055                         optval = coptval = imo->imo_multicast_loop;
2056                 if (sopt->sopt_valsize == 1)
2057                         soopt_from_kbuf(sopt, &coptval, 1);
2058                 else
2059                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2060                 break;
2061
2062         default:
2063                 error = ENOPROTOOPT;
2064                 break;
2065         }
2066         return (error);
2067 }
2068
2069 /*
2070  * Discard the IP multicast options.
2071  */
2072 void
2073 ip_freemoptions(struct ip_moptions *imo)
2074 {
2075         int i;
2076
2077         if (imo != NULL) {
2078                 for (i = 0; i < imo->imo_num_memberships; ++i)
2079                         in_delmulti(imo->imo_membership[i]);
2080                 kfree(imo, M_IPMOPTS);
2081         }
2082 }
2083
2084 /*
2085  * Routine called from ip_output() to loop back a copy of an IP multicast
2086  * packet to the input queue of a specified interface.  Note that this
2087  * calls the output routine of the loopback "driver", but with an interface
2088  * pointer that might NOT be a loopback interface -- evil, but easier than
2089  * replicating that code here.
2090  */
2091 static void
2092 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2093              int hlen)
2094 {
2095         struct ip *ip;
2096         struct mbuf *copym;
2097
2098         copym = m_copypacket(m, MB_DONTWAIT);
2099         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2100                 copym = m_pullup(copym, hlen);
2101         if (copym != NULL) {
2102                 /*
2103                  * if the checksum hasn't been computed, mark it as valid
2104                  */
2105                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2106                         in_delayed_cksum(copym);
2107                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2108                         copym->m_pkthdr.csum_flags |=
2109                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2110                         copym->m_pkthdr.csum_data = 0xffff;
2111                 }
2112                 /*
2113                  * We don't bother to fragment if the IP length is greater
2114                  * than the interface's MTU.  Can this possibly matter?
2115                  */
2116                 ip = mtod(copym, struct ip *);
2117                 ip->ip_len = htons(ip->ip_len);
2118                 ip->ip_off = htons(ip->ip_off);
2119                 ip->ip_sum = 0;
2120                 if (ip->ip_vhl == IP_VHL_BORING) {
2121                         ip->ip_sum = in_cksum_hdr(ip);
2122                 } else {
2123                         ip->ip_sum = in_cksum(copym, hlen);
2124                 }
2125                 /*
2126                  * NB:
2127                  * It's not clear whether there are any lingering
2128                  * reentrancy problems in other areas which might
2129                  * be exposed by using ip_input directly (in
2130                  * particular, everything which modifies the packet
2131                  * in-place).  Yet another option is using the
2132                  * protosw directly to deliver the looped back
2133                  * packet.  For the moment, we'll err on the side
2134                  * of safety by using if_simloop().
2135                  */
2136 #if 1 /* XXX */
2137                 if (dst->sin_family != AF_INET) {
2138                         kprintf("ip_mloopback: bad address family %d\n",
2139                                                 dst->sin_family);
2140                         dst->sin_family = AF_INET;
2141                 }
2142 #endif
2143
2144 #ifdef notdef
2145                 copym->m_pkthdr.rcvif = ifp;
2146                 ip_input(copym);
2147 #else
2148                 if_simloop(ifp, copym, dst->sin_family, 0);
2149 #endif
2150         }
2151 }