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  * 4. 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  * $FreeBSD$
  31  */
  32
  33 #include "opt_ipfw.h"
  34 #include "opt_ipsec.h"
  35 #include "opt_mac.h"
  36 #include "opt_mbuf_stress_test.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/kernel.h>
  41 #include <sys/malloc.h>
  42 #include <sys/mbuf.h>
  43 #include <sys/priv.h>
  44 #include <sys/protosw.h>
  45 #include <sys/socket.h>
  46 #include <sys/socketvar.h>
  47 #include <sys/sysctl.h>
  48
  49 #include <net/if.h>
  50 #include <net/netisr.h>
  51 #include <net/pfil.h>
  52 #include <net/route.h>
  53
  54 #include <netinet/in.h>
  55 #include <netinet/in_systm.h>
  56 #include <netinet/ip.h>
  57 #include <netinet/in_pcb.h>
  58 #include <netinet/in_var.h>
  59 #include <netinet/ip_var.h>
  60 #include <netinet/ip_options.h>
  61
  62 #if defined(IPSEC) || defined(FAST_IPSEC)
  63 #include <netinet/ip_ipsec.h>
  64 #ifdef IPSEC
  65 #include <netinet6/ipsec.h>
  66 #endif
  67 #ifdef FAST_IPSEC
  68 #include <netipsec/ipsec.h>
  69 #endif
  70 #endif /*IPSEC*/
  71
  72 #include <machine/in_cksum.h>
  73
  74 #include <security/mac/mac_framework.h>
  75
  76 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  77
  78 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
  79                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  80                                   (ntohl(a.s_addr)>>16)&0xFF,\
  81                                   (ntohl(a.s_addr)>>8)&0xFF,\
  82                                   (ntohl(a.s_addr))&0xFF, y);
  83
  84 u_short ip_id;
  85
  86 #ifdef MBUF_STRESS_TEST
  87 int mbuf_frag_size = 0;
  88 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
  89         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
  90 #endif
  91
  92 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
  93 static void     ip_mloopback
  94         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
  95 static int      ip_getmoptions(struct inpcb *, struct sockopt *);
  96 static int      ip_setmoptions(struct inpcb *, struct sockopt *);
  97
  98
  99 extern  struct protosw inetsw[];
 100
 101 /*
 102  * IP output.  The packet in mbuf chain m contains a skeletal IP
 103  * header (with len, off, ttl, proto, tos, src, dst).
 104  * The mbuf chain containing the packet will be freed.
 105  * The mbuf opt, if present, will not be freed.
 106  * In the IP forwarding case, the packet will arrive with options already
 107  * inserted, so must have a NULL opt pointer.
 108  */
 109 int
 110 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 111     struct ip_moptions *imo, struct inpcb *inp)
 112 {
 113         struct ip *ip;
 114         struct ifnet *ifp = NULL;       /* keep compiler happy */
 115         struct mbuf *m0;
 116         int hlen = sizeof (struct ip);
 117         int mtu;
 118         int len, error = 0;
 119         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 120         struct in_ifaddr *ia = NULL;
 121         int isbroadcast, sw_csum;
 122         struct route iproute;
 123         struct in_addr odst;
 124 #ifdef IPFIREWALL_FORWARD
 125         struct m_tag *fwd_tag = NULL;
 126 #endif
 127         M_ASSERTPKTHDR(m);
 128
 129         if (ro == NULL) {
 130                 ro = &iproute;
 131                 bzero(ro, sizeof (*ro));
 132         }
 133
 134         if (inp != NULL)
 135                 INP_LOCK_ASSERT(inp);
 136
 137         if (opt) {
 138                 len = 0;
 139                 m = ip_insertoptions(m, opt, &len);
 140                 if (len != 0)
 141                         hlen = len;
 142         }
 143         ip = mtod(m, struct ip *);
 144
 145         /*
 146          * Fill in IP header.  If we are not allowing fragmentation,
 147          * then the ip_id field is meaningless, but we don't set it
 148          * to zero.  Doing so causes various problems when devices along
 149          * the path (routers, load balancers, firewalls, etc.) illegally
 150          * disable DF on our packet.  Note that a 16-bit counter
 151          * will wrap around in less than 10 seconds at 100 Mbit/s on a
 152          * medium with MTU 1500.  See Steven M. Bellovin, "A Technique
 153          * for Counting NATted Hosts", Proc. IMW'02, available at
 154          * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>.
 155          */
 156         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 157                 ip->ip_v = IPVERSION;
 158                 ip->ip_hl = hlen >> 2;
 159                 ip->ip_id = ip_newid();
 160                 ipstat.ips_localout++;
 161         } else {
 162                 hlen = ip->ip_hl << 2;
 163         }
 164
 165         dst = (struct sockaddr_in *)&ro->ro_dst;
 166 again:
 167         /*
 168          * If there is a cached route,
 169          * check that it is to the same destination
 170          * and is still up.  If not, free it and try again.
 171          * The address family should also be checked in case of sharing the
 172          * cache with IPv6.
 173          */
 174         if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
 175                           dst->sin_family != AF_INET ||
 176                           dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
 177                 RTFREE(ro->ro_rt);
 178                 ro->ro_rt = (struct rtentry *)NULL;
 179         }
 180 #ifdef IPFIREWALL_FORWARD
 181         if (ro->ro_rt == NULL && fwd_tag == NULL) {
 182 #else
 183         if (ro->ro_rt == NULL) {
 184 #endif
 185                 bzero(dst, sizeof(*dst));
 186                 dst->sin_family = AF_INET;
 187                 dst->sin_len = sizeof(*dst);
 188                 dst->sin_addr = ip->ip_dst;
 189         }
 190         /*
 191          * If routing to interface only, short circuit routing lookup.
 192          * The use of an all-ones broadcast address implies this; an
 193          * interface is specified by the broadcast address of an interface,
 194          * or the destination address of a ptp interface.
 195          */
 196         if (flags & IP_SENDONES) {
 197                 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL &&
 198                     (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) {
 199                         ipstat.ips_noroute++;
 200                         error = ENETUNREACH;
 201                         goto bad;
 202                 }
 203                 ip->ip_dst.s_addr = INADDR_BROADCAST;
 204                 dst->sin_addr = ip->ip_dst;
 205                 ifp = ia->ia_ifp;
 206                 ip->ip_ttl = 1;
 207                 isbroadcast = 1;
 208         } else if (flags & IP_ROUTETOIF) {
 209                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 210                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 211                         ipstat.ips_noroute++;
 212                         error = ENETUNREACH;
 213                         goto bad;
 214                 }
 215                 ifp = ia->ia_ifp;
 216                 ip->ip_ttl = 1;
 217                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 218         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 219             imo != NULL && imo->imo_multicast_ifp != NULL) {
 220                 /*
 221                  * Bypass the normal routing lookup for multicast
 222                  * packets if the interface is specified.
 223                  */
 224                 ifp = imo->imo_multicast_ifp;
 225                 IFP_TO_IA(ifp, ia);
 226                 isbroadcast = 0;        /* fool gcc */
 227         } else {
 228                 /*
 229                  * We want to do any cloning requested by the link layer,
 230                  * as this is probably required in all cases for correct
 231                  * operation (as it is for ARP).
 232                  */
 233                 if (ro->ro_rt == NULL)
 234                         rtalloc_ign(ro, 0);
 235                 if (ro->ro_rt == NULL) {
 236                         ipstat.ips_noroute++;
 237                         error = EHOSTUNREACH;
 238                         goto bad;
 239                 }
 240                 ia = ifatoia(ro->ro_rt->rt_ifa);
 241                 ifp = ro->ro_rt->rt_ifp;
 242                 ro->ro_rt->rt_rmx.rmx_pksent++;
 243                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 244                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 245                 if (ro->ro_rt->rt_flags & RTF_HOST)
 246                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 247                 else
 248                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 249         }
 250         /*
 251          * Calculate MTU.  If we have a route that is up, use that,
 252          * otherwise use the interface's MTU.
 253          */
 254         if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) {
 255                 /*
 256                  * This case can happen if the user changed the MTU
 257                  * of an interface after enabling IP on it.  Because
 258                  * most netifs don't keep track of routes pointing to
 259                  * them, there is no way for one to update all its
 260                  * routes when the MTU is changed.
 261                  */
 262                 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)
 263                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 264                 mtu = ro->ro_rt->rt_rmx.rmx_mtu;
 265         } else {
 266                 mtu = ifp->if_mtu;
 267         }
 268         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 269                 struct in_multi *inm;
 270
 271                 m->m_flags |= M_MCAST;
 272                 /*
 273                  * IP destination address is multicast.  Make sure "dst"
 274                  * still points to the address in "ro".  (It may have been
 275                  * changed to point to a gateway address, above.)
 276                  */
 277                 dst = (struct sockaddr_in *)&ro->ro_dst;
 278                 /*
 279                  * See if the caller provided any multicast options
 280                  */
 281                 if (imo != NULL) {
 282                         ip->ip_ttl = imo->imo_multicast_ttl;
 283                         if (imo->imo_multicast_vif != -1)
 284                                 ip->ip_src.s_addr =
 285                                     ip_mcast_src ?
 286                                     ip_mcast_src(imo->imo_multicast_vif) :
 287                                     INADDR_ANY;
 288                 } else
 289                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 290                 /*
 291                  * Confirm that the outgoing interface supports multicast.
 292                  */
 293                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 294                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 295                                 ipstat.ips_noroute++;
 296                                 error = ENETUNREACH;
 297                                 goto bad;
 298                         }
 299                 }
 300                 /*
 301                  * If source address not specified yet, use address
 302                  * of outgoing interface.
 303                  */
 304                 if (ip->ip_src.s_addr == INADDR_ANY) {
 305                         /* Interface may have no addresses. */
 306                         if (ia != NULL)
 307                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 308                 }
 309
 310                 IN_MULTI_LOCK();
 311                 IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm);
 312                 if (inm != NULL &&
 313                    (imo == NULL || imo->imo_multicast_loop)) {
 314                         IN_MULTI_UNLOCK();
 315                         /*
 316                          * If we belong to the destination multicast group
 317                          * on the outgoing interface, and the caller did not
 318                          * forbid loopback, loop back a copy.
 319                          */
 320                         ip_mloopback(ifp, m, dst, hlen);
 321                 }
 322                 else {
 323                         IN_MULTI_UNLOCK();
 324                         /*
 325                          * If we are acting as a multicast router, perform
 326                          * multicast forwarding as if the packet had just
 327                          * arrived on the interface to which we are about
 328                          * to send.  The multicast forwarding function
 329                          * recursively calls this function, using the
 330                          * IP_FORWARDING flag to prevent infinite recursion.
 331                          *
 332                          * Multicasts that are looped back by ip_mloopback(),
 333                          * above, will be forwarded by the ip_input() routine,
 334                          * if necessary.
 335                          */
 336                         if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
 337                                 /*
 338                                  * If rsvp daemon is not running, do not
 339                                  * set ip_moptions. This ensures that the packet
 340                                  * is multicast and not just sent down one link
 341                                  * as prescribed by rsvpd.
 342                                  */
 343                                 if (!rsvp_on)
 344                                         imo = NULL;
 345                                 if (ip_mforward &&
 346                                     ip_mforward(ip, ifp, m, imo) != 0) {
 347                                         m_freem(m);
 348                                         goto done;
 349                                 }
 350                         }
 351                 }
 352
 353                 /*
 354                  * Multicasts with a time-to-live of zero may be looped-
 355                  * back, above, but must not be transmitted on a network.
 356                  * Also, multicasts addressed to the loopback interface
 357                  * are not sent -- the above call to ip_mloopback() will
 358                  * loop back a copy if this host actually belongs to the
 359                  * destination group on the loopback interface.
 360                  */
 361                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 362                         m_freem(m);
 363                         goto done;
 364                 }
 365
 366                 goto sendit;
 367         }
 368
 369         /*
 370          * If the source address is not specified yet, use the address
 371          * of the outoing interface.
 372          */
 373         if (ip->ip_src.s_addr == INADDR_ANY) {
 374                 /* Interface may have no addresses. */
 375                 if (ia != NULL) {
 376                         ip->ip_src = IA_SIN(ia)->sin_addr;
 377                 }
 378         }
 379
 380         /*
 381          * Verify that we have any chance at all of being able to queue the
 382          * packet or packet fragments, unless ALTQ is enabled on the given
 383          * interface in which case packetdrop should be done by queueing.
 384          */
 385 #ifdef ALTQ
 386         if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) &&
 387             ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >=
 388             ifp->if_snd.ifq_maxlen))
 389 #else
 390         if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >=
 391             ifp->if_snd.ifq_maxlen)
 392 #endif /* ALTQ */
 393         {
 394                 error = ENOBUFS;
 395                 ipstat.ips_odropped++;
 396                 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1);
 397                 goto bad;
 398         }
 399
 400         /*
 401          * Look for broadcast address and
 402          * verify user is allowed to send
 403          * such a packet.
 404          */
 405         if (isbroadcast) {
 406                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 407                         error = EADDRNOTAVAIL;
 408                         goto bad;
 409                 }
 410                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 411                         error = EACCES;
 412                         goto bad;
 413                 }
 414                 /* don't allow broadcast messages to be fragmented */
 415                 if (ip->ip_len > mtu) {
 416                         error = EMSGSIZE;
 417                         goto bad;
 418                 }
 419                 m->m_flags |= M_BCAST;
 420         } else {
 421                 m->m_flags &= ~M_BCAST;
 422         }
 423
 424 sendit:
 425 #if defined(IPSEC) || defined(FAST_IPSEC)
 426         switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) {
 427         case 1:
 428                 goto bad;
 429         case -1:
 430                 goto done;
 431         case 0:
 432         default:
 433                 break;  /* Continue with packet processing. */
 434         }
 435         /* Update variables that are affected by ipsec4_output(). */
 436         ip = mtod(m, struct ip *);
 437         hlen = ip->ip_hl << 2;
 438 #endif /* IPSEC */
 439
 440         /* Jump over all PFIL processing if hooks are not active. */
 441         if (!PFIL_HOOKED(&inet_pfil_hook))
 442                 goto passout;
 443
 444         /* Run through list of hooks for output packets. */
 445         odst.s_addr = ip->ip_dst.s_addr;
 446         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp);
 447         if (error != 0 || m == NULL)
 448                 goto done;
 449
 450         ip = mtod(m, struct ip *);
 451
 452         /* See if destination IP address was changed by packet filter. */
 453         if (odst.s_addr != ip->ip_dst.s_addr) {
 454                 m->m_flags |= M_SKIP_FIREWALL;
 455                 /* If destination is now ourself drop to ip_input(). */
 456                 if (in_localip(ip->ip_dst)) {
 457                         m->m_flags |= M_FASTFWD_OURS;
 458                         if (m->m_pkthdr.rcvif == NULL)
 459                                 m->m_pkthdr.rcvif = loif;
 460                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 461                                 m->m_pkthdr.csum_flags |=
 462                                     CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 463                                 m->m_pkthdr.csum_data = 0xffff;
 464                         }
 465                         m->m_pkthdr.csum_flags |=
 466                             CSUM_IP_CHECKED | CSUM_IP_VALID;
 467
 468                         error = netisr_queue(NETISR_IP, m);
 469                         goto done;
 470                 } else
 471                         goto again;     /* Redo the routing table lookup. */
 472         }
 473
 474 #ifdef IPFIREWALL_FORWARD
 475         /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
 476         if (m->m_flags & M_FASTFWD_OURS) {
 477                 if (m->m_pkthdr.rcvif == NULL)
 478                         m->m_pkthdr.rcvif = loif;
 479                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 480                         m->m_pkthdr.csum_flags |=
 481                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 482                         m->m_pkthdr.csum_data = 0xffff;
 483                 }
 484                 m->m_pkthdr.csum_flags |=
 485                             CSUM_IP_CHECKED | CSUM_IP_VALID;
 486
 487                 error = netisr_queue(NETISR_IP, m);
 488                 goto done;
 489         }
 490         /* Or forward to some other address? */
 491         fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 492         if (fwd_tag) {
 493                 dst = (struct sockaddr_in *)&ro->ro_dst;
 494                 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
 495                 m->m_flags |= M_SKIP_FIREWALL;
 496                 m_tag_delete(m, fwd_tag);
 497                 goto again;
 498         }
 499 #endif /* IPFIREWALL_FORWARD */
 500
 501 passout:
 502         /* 127/8 must not appear on wire - RFC1122. */
 503         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 504             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 505                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 506                         ipstat.ips_badaddr++;
 507                         error = EADDRNOTAVAIL;
 508                         goto bad;
 509                 }
 510         }
 511
 512         m->m_pkthdr.csum_flags |= CSUM_IP;
 513         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 514         if (sw_csum & CSUM_DELAY_DATA) {
 515                 in_delayed_cksum(m);
 516                 sw_csum &= ~CSUM_DELAY_DATA;
 517         }
 518         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 519
 520         /*
 521          * If small enough for interface, or the interface will take
 522          * care of the fragmentation for us, we can just send directly.
 523          */
 524         if (ip->ip_len <= mtu ||
 525             (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
 526             ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
 527                 ip->ip_len = htons(ip->ip_len);
 528                 ip->ip_off = htons(ip->ip_off);
 529                 ip->ip_sum = 0;
 530                 if (sw_csum & CSUM_DELAY_IP)
 531                         ip->ip_sum = in_cksum(m, hlen);
 532
 533                 /*
 534                  * Record statistics for this interface address.
 535                  * With CSUM_TSO the byte/packet count will be slightly
 536                  * incorrect because we count the IP+TCP headers only
 537                  * once instead of for every generated packet.
 538                  */
 539                 if (!(flags & IP_FORWARDING) && ia) {
 540                         if (m->m_pkthdr.csum_flags & CSUM_TSO)
 541                                 ia->ia_ifa.if_opackets +=
 542                                     m->m_pkthdr.len / m->m_pkthdr.tso_segsz;
 543                         else
 544                                 ia->ia_ifa.if_opackets++;
 545                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 546                 }
 547 #ifdef IPSEC
 548                 /* clean ipsec history once it goes out of the node */
 549                 ipsec_delaux(m);
 550 #endif
 551 #ifdef MBUF_STRESS_TEST
 552                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
 553                         m = m_fragment(m, M_DONTWAIT, mbuf_frag_size);
 554 #endif
 555                 /*
 556                  * Reset layer specific mbuf flags
 557                  * to avoid confusing lower layers.
 558                  */
 559                 m->m_flags &= ~(M_PROTOFLAGS);
 560
 561                 error = (*ifp->if_output)(ifp, m,
 562                                 (struct sockaddr *)dst, ro->ro_rt);
 563                 goto done;
 564         }
 565
 566         /* Balk when DF bit is set or the interface didn't support TSO. */
 567         if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 568                 error = EMSGSIZE;
 569                 ipstat.ips_cantfrag++;
 570                 goto bad;
 571         }
 572
 573         /*
 574          * Too large for interface; fragment if possible. If successful,
 575          * on return, m will point to a list of packets to be sent.
 576          */
 577         error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum);
 578         if (error)
 579                 goto bad;
 580         for (; m; m = m0) {
 581                 m0 = m->m_nextpkt;
 582                 m->m_nextpkt = 0;
 583 #ifdef IPSEC
 584                 /* clean ipsec history once it goes out of the node */
 585                 ipsec_delaux(m);
 586 #endif
 587                 if (error == 0) {
 588                         /* Record statistics for this interface address. */
 589                         if (ia != NULL) {
 590                                 ia->ia_ifa.if_opackets++;
 591                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 592                         }
 593                         /*
 594                          * Reset layer specific mbuf flags
 595                          * to avoid confusing upper layers.
 596                          */
 597                         m->m_flags &= ~(M_PROTOFLAGS);
 598
 599                         error = (*ifp->if_output)(ifp, m,
 600                             (struct sockaddr *)dst, ro->ro_rt);
 601                 } else
 602                         m_freem(m);
 603         }
 604
 605         if (error == 0)
 606                 ipstat.ips_fragmented++;
 607
 608 done:
 609         if (ro == &iproute && ro->ro_rt) {
 610                 RTFREE(ro->ro_rt);
 611         }
 612         return (error);
 613 bad:
 614         m_freem(m);
 615         goto done;
 616 }
 617
 618 /*
 619  * Create a chain of fragments which fit the given mtu. m_frag points to the
 620  * mbuf to be fragmented; on return it points to the chain with the fragments.
 621  * Return 0 if no error. If error, m_frag may contain a partially built
 622  * chain of fragments that should be freed by the caller.
 623  *
 624  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
 625  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
 626  */
 627 int
 628 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
 629     u_long if_hwassist_flags, int sw_csum)
 630 {
 631         int error = 0;
 632         int hlen = ip->ip_hl << 2;
 633         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
 634         int off;
 635         struct mbuf *m0 = *m_frag;      /* the original packet          */
 636         int firstlen;
 637         struct mbuf **mnext;
 638         int nfrags;
 639
 640         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
 641                 ipstat.ips_cantfrag++;
 642                 return EMSGSIZE;
 643         }
 644
 645         /*
 646          * Must be able to put at least 8 bytes per fragment.
 647          */
 648         if (len < 8)
 649                 return EMSGSIZE;
 650
 651         /*
 652          * If the interface will not calculate checksums on
 653          * fragmented packets, then do it here.
 654          */
 655         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
 656             (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
 657                 in_delayed_cksum(m0);
 658                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 659         }
 660
 661         if (len > PAGE_SIZE) {
 662                 /*
 663                  * Fragment large datagrams such that each segment
 664                  * contains a multiple of PAGE_SIZE amount of data,
 665                  * plus headers. This enables a receiver to perform
 666                  * page-flipping zero-copy optimizations.
 667                  *
 668                  * XXX When does this help given that sender and receiver
 669                  * could have different page sizes, and also mtu could
 670                  * be less than the receiver's page size ?
 671                  */
 672                 int newlen;
 673                 struct mbuf *m;
 674
 675                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
 676                         off += m->m_len;
 677
 678                 /*
 679                  * firstlen (off - hlen) must be aligned on an
 680                  * 8-byte boundary
 681                  */
 682                 if (off < hlen)
 683                         goto smart_frag_failure;
 684                 off = ((off - hlen) & ~7) + hlen;
 685                 newlen = (~PAGE_MASK) & mtu;
 686                 if ((newlen + sizeof (struct ip)) > mtu) {
 687                         /* we failed, go back the default */
 688 smart_frag_failure:
 689                         newlen = len;
 690                         off = hlen + len;
 691                 }
 692                 len = newlen;
 693
 694         } else {
 695                 off = hlen + len;
 696         }
 697
 698         firstlen = off - hlen;
 699         mnext = &m0->m_nextpkt;         /* pointer to next packet */
 700
 701         /*
 702          * Loop through length of segment after first fragment,
 703          * make new header and copy data of each part and link onto chain.
 704          * Here, m0 is the original packet, m is the fragment being created.
 705          * The fragments are linked off the m_nextpkt of the original
 706          * packet, which after processing serves as the first fragment.
 707          */
 708         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
 709                 struct ip *mhip;        /* ip header on the fragment */
 710                 struct mbuf *m;
 711                 int mhlen = sizeof (struct ip);
 712
 713                 MGETHDR(m, M_DONTWAIT, MT_DATA);
 714                 if (m == NULL) {
 715                         error = ENOBUFS;
 716                         ipstat.ips_odropped++;
 717                         goto done;
 718                 }
 719                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
 720                 /*
 721                  * In the first mbuf, leave room for the link header, then
 722                  * copy the original IP header including options. The payload
 723                  * goes into an additional mbuf chain returned by m_copy().
 724                  */
 725                 m->m_data += max_linkhdr;
 726                 mhip = mtod(m, struct ip *);
 727                 *mhip = *ip;
 728                 if (hlen > sizeof (struct ip)) {
 729                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
 730                         mhip->ip_v = IPVERSION;
 731                         mhip->ip_hl = mhlen >> 2;
 732                 }
 733                 m->m_len = mhlen;
 734                 /* XXX do we need to add ip->ip_off below ? */
 735                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
 736                 if (off + len >= ip->ip_len) {  /* last fragment */
 737                         len = ip->ip_len - off;
 738                         m->m_flags |= M_LASTFRAG;
 739                 } else
 740                         mhip->ip_off |= IP_MF;
 741                 mhip->ip_len = htons((u_short)(len + mhlen));
 742                 m->m_next = m_copy(m0, off, len);
 743                 if (m->m_next == NULL) {        /* copy failed */
 744                         m_free(m);
 745                         error = ENOBUFS;        /* ??? */
 746                         ipstat.ips_odropped++;
 747                         goto done;
 748                 }
 749                 m->m_pkthdr.len = mhlen + len;
 750                 m->m_pkthdr.rcvif = NULL;
 751 #ifdef MAC
 752                 mac_create_fragment(m0, m);
 753 #endif
 754                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
 755                 mhip->ip_off = htons(mhip->ip_off);
 756                 mhip->ip_sum = 0;
 757                 if (sw_csum & CSUM_DELAY_IP)
 758                         mhip->ip_sum = in_cksum(m, mhlen);
 759                 *mnext = m;
 760                 mnext = &m->m_nextpkt;
 761         }
 762         ipstat.ips_ofragments += nfrags;
 763
 764         /* set first marker for fragment chain */
 765         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
 766         m0->m_pkthdr.csum_data = nfrags;
 767
 768         /*
 769          * Update first fragment by trimming what's been copied out
 770          * and updating header.
 771          */
 772         m_adj(m0, hlen + firstlen - ip->ip_len);
 773         m0->m_pkthdr.len = hlen + firstlen;
 774         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
 775         ip->ip_off |= IP_MF;
 776         ip->ip_off = htons(ip->ip_off);
 777         ip->ip_sum = 0;
 778         if (sw_csum & CSUM_DELAY_IP)
 779                 ip->ip_sum = in_cksum(m0, hlen);
 780
 781 done:
 782         *m_frag = m0;
 783         return error;
 784 }
 785
 786 void
 787 in_delayed_cksum(struct mbuf *m)
 788 {
 789         struct ip *ip;
 790         u_short csum, offset;
 791
 792         ip = mtod(m, struct ip *);
 793         offset = ip->ip_hl << 2 ;
 794         csum = in_cksum_skip(m, ip->ip_len, offset);
 795         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
 796                 csum = 0xffff;
 797         offset += m->m_pkthdr.csum_data;        /* checksum offset */
 798
 799         if (offset + sizeof(u_short) > m->m_len) {
 800                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
 801                     m->m_len, offset, ip->ip_p);
 802                 /*
 803                  * XXX
 804                  * this shouldn't happen, but if it does, the
 805                  * correct behavior may be to insert the checksum
 806                  * in the appropriate next mbuf in the chain.
 807                  */
 808                 return;
 809         }
 810         *(u_short *)(m->m_data + offset) = csum;
 811 }
 812
 813 /*
 814  * IP socket option processing.
 815  */
 816 int
 817 ip_ctloutput(struct socket *so, struct sockopt *sopt)
 818 {
 819         struct  inpcb *inp = sotoinpcb(so);
 820         int     error, optval;
 821
 822         error = optval = 0;
 823         if (sopt->sopt_level != IPPROTO_IP) {
 824                 return (EINVAL);
 825         }
 826
 827         switch (sopt->sopt_dir) {
 828         case SOPT_SET:
 829                 switch (sopt->sopt_name) {
 830                 case IP_OPTIONS:
 831 #ifdef notyet
 832                 case IP_RETOPTS:
 833 #endif
 834                 {
 835                         struct mbuf *m;
 836                         if (sopt->sopt_valsize > MLEN) {
 837                                 error = EMSGSIZE;
 838                                 break;
 839                         }
 840                         MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
 841                         if (m == NULL) {
 842                                 error = ENOBUFS;
 843                                 break;
 844                         }
 845                         m->m_len = sopt->sopt_valsize;
 846                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
 847                                             m->m_len);
 848                         if (error) {
 849                                 m_free(m);
 850                                 break;
 851                         }
 852                         INP_LOCK(inp);
 853                         error = ip_pcbopts(inp, sopt->sopt_name, m);
 854                         INP_UNLOCK(inp);
 855                         return (error);
 856                 }
 857
 858                 case IP_TOS:
 859                 case IP_TTL:
 860                 case IP_MINTTL:
 861                 case IP_RECVOPTS:
 862                 case IP_RECVRETOPTS:
 863                 case IP_RECVDSTADDR:
 864                 case IP_RECVTTL:
 865                 case IP_RECVIF:
 866                 case IP_FAITH:
 867                 case IP_ONESBCAST:
 868                 case IP_DONTFRAG:
 869                         error = sooptcopyin(sopt, &optval, sizeof optval,
 870                                             sizeof optval);
 871                         if (error)
 872                                 break;
 873
 874                         switch (sopt->sopt_name) {
 875                         case IP_TOS:
 876                                 inp->inp_ip_tos = optval;
 877                                 break;
 878
 879                         case IP_TTL:
 880                                 inp->inp_ip_ttl = optval;
 881                                 break;
 882
 883                         case IP_MINTTL:
 884                                 if (optval > 0 && optval <= MAXTTL)
 885                                         inp->inp_ip_minttl = optval;
 886                                 else
 887                                         error = EINVAL;
 888                                 break;
 889
 890 #define OPTSET(bit) do {                                                \
 891         INP_LOCK(inp);                                                  \
 892         if (optval)                                                     \
 893                 inp->inp_flags |= bit;                                  \
 894         else                                                            \
 895                 inp->inp_flags &= ~bit;                                 \
 896         INP_UNLOCK(inp);                                                \
 897 } while (0)
 898
 899                         case IP_RECVOPTS:
 900                                 OPTSET(INP_RECVOPTS);
 901                                 break;
 902
 903                         case IP_RECVRETOPTS:
 904                                 OPTSET(INP_RECVRETOPTS);
 905                                 break;
 906
 907                         case IP_RECVDSTADDR:
 908                                 OPTSET(INP_RECVDSTADDR);
 909                                 break;
 910
 911                         case IP_RECVTTL:
 912                                 OPTSET(INP_RECVTTL);
 913                                 break;
 914
 915                         case IP_RECVIF:
 916                                 OPTSET(INP_RECVIF);
 917                                 break;
 918
 919                         case IP_FAITH:
 920                                 OPTSET(INP_FAITH);
 921                                 break;
 922
 923                         case IP_ONESBCAST:
 924                                 OPTSET(INP_ONESBCAST);
 925                                 break;
 926                         case IP_DONTFRAG:
 927                                 OPTSET(INP_DONTFRAG);
 928                                 break;
 929                         }
 930                         break;
 931 #undef OPTSET
 932
 933                 case IP_MULTICAST_IF:
 934                 case IP_MULTICAST_VIF:
 935                 case IP_MULTICAST_TTL:
 936                 case IP_MULTICAST_LOOP:
 937                 case IP_ADD_MEMBERSHIP:
 938                 case IP_DROP_MEMBERSHIP:
 939                         error = ip_setmoptions(inp, sopt);
 940                         break;
 941
 942                 case IP_PORTRANGE:
 943                         error = sooptcopyin(sopt, &optval, sizeof optval,
 944                                             sizeof optval);
 945                         if (error)
 946                                 break;
 947
 948                         INP_LOCK(inp);
 949                         switch (optval) {
 950                         case IP_PORTRANGE_DEFAULT:
 951                                 inp->inp_flags &= ~(INP_LOWPORT);
 952                                 inp->inp_flags &= ~(INP_HIGHPORT);
 953                                 break;
 954
 955                         case IP_PORTRANGE_HIGH:
 956                                 inp->inp_flags &= ~(INP_LOWPORT);
 957                                 inp->inp_flags |= INP_HIGHPORT;
 958                                 break;
 959
 960                         case IP_PORTRANGE_LOW:
 961                                 inp->inp_flags &= ~(INP_HIGHPORT);
 962                                 inp->inp_flags |= INP_LOWPORT;
 963                                 break;
 964
 965                         default:
 966                                 error = EINVAL;
 967                                 break;
 968                         }
 969                         INP_UNLOCK(inp);
 970                         break;
 971
 972 #if defined(IPSEC) || defined(FAST_IPSEC)
 973                 case IP_IPSEC_POLICY:
 974                 {
 975                         caddr_t req;
 976                         size_t len = 0;
 977                         int priv;
 978                         struct mbuf *m;
 979                         int optname;
 980
 981                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
 982                                 break;
 983                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
 984                                 break;
 985                         if (sopt->sopt_td != NULL) {
 986                                 /*
 987                                  * XXXRW: Would be more desirable to do this
 988                                  * one layer down so that we only exercise
 989                                  * privilege if it is needed.
 990                                  */
 991                                 error = priv_check(sopt->sopt_td,
 992                                     PRIV_NETINET_IPSEC);
 993                                 if (error)
 994                                         priv = 0;
 995                                 else
 996                                         priv = 1;
 997                         } else
 998                                 priv = 1;
 999                         req = mtod(m, caddr_t);
1000                         len = m->m_len;
1001                         optname = sopt->sopt_name;
1002                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1003                         m_freem(m);
1004                         break;
1005                 }
1006 #endif /*IPSEC*/
1007
1008                 default:
1009                         error = ENOPROTOOPT;
1010                         break;
1011                 }
1012                 break;
1013
1014         case SOPT_GET:
1015                 switch (sopt->sopt_name) {
1016                 case IP_OPTIONS:
1017                 case IP_RETOPTS:
1018                         if (inp->inp_options)
1019                                 error = sooptcopyout(sopt,
1020                                                      mtod(inp->inp_options,
1021                                                           char *),
1022                                                      inp->inp_options->m_len);
1023                         else
1024                                 sopt->sopt_valsize = 0;
1025                         break;
1026
1027                 case IP_TOS:
1028                 case IP_TTL:
1029                 case IP_MINTTL:
1030                 case IP_RECVOPTS:
1031                 case IP_RECVRETOPTS:
1032                 case IP_RECVDSTADDR:
1033                 case IP_RECVTTL:
1034                 case IP_RECVIF:
1035                 case IP_PORTRANGE:
1036                 case IP_FAITH:
1037                 case IP_ONESBCAST:
1038                 case IP_DONTFRAG:
1039                         switch (sopt->sopt_name) {
1040
1041                         case IP_TOS:
1042                                 optval = inp->inp_ip_tos;
1043                                 break;
1044
1045                         case IP_TTL:
1046                                 optval = inp->inp_ip_ttl;
1047                                 break;
1048
1049                         case IP_MINTTL:
1050                                 optval = inp->inp_ip_minttl;
1051                                 break;
1052
1053 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1054
1055                         case IP_RECVOPTS:
1056                                 optval = OPTBIT(INP_RECVOPTS);
1057                                 break;
1058
1059                         case IP_RECVRETOPTS:
1060                                 optval = OPTBIT(INP_RECVRETOPTS);
1061                                 break;
1062
1063                         case IP_RECVDSTADDR:
1064                                 optval = OPTBIT(INP_RECVDSTADDR);
1065                                 break;
1066
1067                         case IP_RECVTTL:
1068                                 optval = OPTBIT(INP_RECVTTL);
1069                                 break;
1070
1071                         case IP_RECVIF:
1072                                 optval = OPTBIT(INP_RECVIF);
1073                                 break;
1074
1075                         case IP_PORTRANGE:
1076                                 if (inp->inp_flags & INP_HIGHPORT)
1077                                         optval = IP_PORTRANGE_HIGH;
1078                                 else if (inp->inp_flags & INP_LOWPORT)
1079                                         optval = IP_PORTRANGE_LOW;
1080                                 else
1081                                         optval = 0;
1082                                 break;
1083
1084                         case IP_FAITH:
1085                                 optval = OPTBIT(INP_FAITH);
1086                                 break;
1087
1088                         case IP_ONESBCAST:
1089                                 optval = OPTBIT(INP_ONESBCAST);
1090                                 break;
1091                         case IP_DONTFRAG:
1092                                 optval = OPTBIT(INP_DONTFRAG);
1093                                 break;
1094                         }
1095                         error = sooptcopyout(sopt, &optval, sizeof optval);
1096                         break;
1097
1098                 case IP_MULTICAST_IF:
1099                 case IP_MULTICAST_VIF:
1100                 case IP_MULTICAST_TTL:
1101                 case IP_MULTICAST_LOOP:
1102                         error = ip_getmoptions(inp, sopt);
1103                         break;
1104
1105 #if defined(IPSEC) || defined(FAST_IPSEC)
1106                 case IP_IPSEC_POLICY:
1107                 {
1108                         struct mbuf *m = NULL;
1109                         caddr_t req = NULL;
1110                         size_t len = 0;
1111
1112                         if (m != 0) {
1113                                 req = mtod(m, caddr_t);
1114                                 len = m->m_len;
1115                         }
1116                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1117                         if (error == 0)
1118                                 error = soopt_mcopyout(sopt, m); /* XXX */
1119                         if (error == 0)
1120                                 m_freem(m);
1121                         break;
1122                 }
1123 #endif /*IPSEC*/
1124
1125                 default:
1126                         error = ENOPROTOOPT;
1127                         break;
1128                 }
1129                 break;
1130         }
1131         return (error);
1132 }
1133
1134 /*
1135  * XXX
1136  * The whole multicast option thing needs to be re-thought.
1137  * Several of these options are equally applicable to non-multicast
1138  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1139  * standard option (IP_TTL).
1140  */
1141
1142 /*
1143  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1144  */
1145 static struct ifnet *
1146 ip_multicast_if(struct in_addr *a, int *ifindexp)
1147 {
1148         int ifindex;
1149         struct ifnet *ifp;
1150
1151         if (ifindexp)
1152                 *ifindexp = 0;
1153         if (ntohl(a->s_addr) >> 24 == 0) {
1154                 ifindex = ntohl(a->s_addr) & 0xffffff;
1155                 if (ifindex < 0 || if_index < ifindex)
1156                         return NULL;
1157                 ifp = ifnet_byindex(ifindex);
1158                 if (ifindexp)
1159                         *ifindexp = ifindex;
1160         } else {
1161                 INADDR_TO_IFP(*a, ifp);
1162         }
1163         return ifp;
1164 }
1165
1166 /*
1167  * Given an inpcb, return its multicast options structure pointer.  Accepts
1168  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1169  */
1170 static struct ip_moptions *
1171 ip_findmoptions(struct inpcb *inp)
1172 {
1173         struct ip_moptions *imo;
1174         struct in_multi **immp;
1175
1176         INP_LOCK(inp);
1177         if (inp->inp_moptions != NULL)
1178                 return (inp->inp_moptions);
1179
1180         INP_UNLOCK(inp);
1181
1182         imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1183         immp = (struct in_multi **)malloc((sizeof(*immp) * IP_MIN_MEMBERSHIPS),
1184                                           M_IPMOPTS, M_WAITOK);
1185
1186         imo->imo_multicast_ifp = NULL;
1187         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1188         imo->imo_multicast_vif = -1;
1189         imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1190         imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1191         imo->imo_num_memberships = 0;
1192         imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1193         imo->imo_membership = immp;
1194
1195         INP_LOCK(inp);
1196         if (inp->inp_moptions != NULL) {
1197                 free(immp, M_IPMOPTS);
1198                 free(imo, M_IPMOPTS);
1199                 return (inp->inp_moptions);
1200         }
1201         inp->inp_moptions = imo;
1202         return (imo);
1203 }
1204
1205 /*
1206  * Set the IP multicast options in response to user setsockopt().
1207  */
1208 static int
1209 ip_setmoptions(struct inpcb *inp, struct sockopt *sopt)
1210 {
1211         int error = 0;
1212         int i;
1213         struct in_addr addr;
1214         struct ip_mreq mreq;
1215         struct ifnet *ifp;
1216         struct ip_moptions *imo;
1217         struct route ro;
1218         struct sockaddr_in *dst;
1219         int ifindex;
1220         int s;
1221
1222         switch (sopt->sopt_name) {
1223         /* store an index number for the vif you wanna use in the send */
1224         case IP_MULTICAST_VIF:
1225                 if (legal_vif_num == 0) {
1226                         error = EOPNOTSUPP;
1227                         break;
1228                 }
1229                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1230                 if (error)
1231                         break;
1232                 if (!legal_vif_num(i) && (i != -1)) {
1233                         error = EINVAL;
1234                         break;
1235                 }
1236                 imo = ip_findmoptions(inp);
1237                 imo->imo_multicast_vif = i;
1238                 INP_UNLOCK(inp);
1239                 break;
1240
1241         case IP_MULTICAST_IF:
1242                 /*
1243                  * Select the interface for outgoing multicast packets.
1244                  */
1245                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1246                 if (error)
1247                         break;
1248                 /*
1249                  * INADDR_ANY is used to remove a previous selection.
1250                  * When no interface is selected, a default one is
1251                  * chosen every time a multicast packet is sent.
1252                  */
1253                 imo = ip_findmoptions(inp);
1254                 if (addr.s_addr == INADDR_ANY) {
1255                         imo->imo_multicast_ifp = NULL;
1256                         INP_UNLOCK(inp);
1257                         break;
1258                 }
1259                 /*
1260                  * The selected interface is identified by its local
1261                  * IP address.  Find the interface and confirm that
1262                  * it supports multicasting.
1263                  */
1264                 s = splimp();
1265                 ifp = ip_multicast_if(&addr, &ifindex);
1266                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1267                         INP_UNLOCK(inp);
1268                         splx(s);
1269                         error = EADDRNOTAVAIL;
1270                         break;
1271                 }
1272                 imo->imo_multicast_ifp = ifp;
1273                 if (ifindex)
1274                         imo->imo_multicast_addr = addr;
1275                 else
1276                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1277                 INP_UNLOCK(inp);
1278                 splx(s);
1279                 break;
1280
1281         case IP_MULTICAST_TTL:
1282                 /*
1283                  * Set the IP time-to-live for outgoing multicast packets.
1284                  * The original multicast API required a char argument,
1285                  * which is inconsistent with the rest of the socket API.
1286                  * We allow either a char or an int.
1287                  */
1288                 if (sopt->sopt_valsize == 1) {
1289                         u_char ttl;
1290                         error = sooptcopyin(sopt, &ttl, 1, 1);
1291                         if (error)
1292                                 break;
1293                         imo = ip_findmoptions(inp);
1294                         imo->imo_multicast_ttl = ttl;
1295                         INP_UNLOCK(inp);
1296                 } else {
1297                         u_int ttl;
1298                         error = sooptcopyin(sopt, &ttl, sizeof ttl,
1299                                             sizeof ttl);
1300                         if (error)
1301                                 break;
1302                         if (ttl > 255)
1303                                 error = EINVAL;
1304                         else {
1305                                 imo = ip_findmoptions(inp);
1306                                 imo->imo_multicast_ttl = ttl;
1307                                 INP_UNLOCK(inp);
1308                         }
1309                 }
1310                 break;
1311
1312         case IP_MULTICAST_LOOP:
1313                 /*
1314                  * Set the loopback flag for outgoing multicast packets.
1315                  * Must be zero or one.  The original multicast API required a
1316                  * char argument, which is inconsistent with the rest
1317                  * of the socket API.  We allow either a char or an int.
1318                  */
1319                 if (sopt->sopt_valsize == 1) {
1320                         u_char loop;
1321                         error = sooptcopyin(sopt, &loop, 1, 1);
1322                         if (error)
1323                                 break;
1324                         imo = ip_findmoptions(inp);
1325                         imo->imo_multicast_loop = !!loop;
1326                         INP_UNLOCK(inp);
1327                 } else {
1328                         u_int loop;
1329                         error = sooptcopyin(sopt, &loop, sizeof loop,
1330                                             sizeof loop);
1331                         if (error)
1332                                 break;
1333                         imo = ip_findmoptions(inp);
1334                         imo->imo_multicast_loop = !!loop;
1335                         INP_UNLOCK(inp);
1336                 }
1337                 break;
1338
1339         case IP_ADD_MEMBERSHIP:
1340                 /*
1341                  * Add a multicast group membership.
1342                  * Group must be a valid IP multicast address.
1343                  */
1344                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1345                 if (error)
1346                         break;
1347
1348                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1349                         error = EINVAL;
1350                         break;
1351                 }
1352                 s = splimp();
1353                 /*
1354                  * If no interface address was provided, use the interface of
1355                  * the route to the given multicast address.
1356                  */
1357                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1358                         bzero((caddr_t)&ro, sizeof(ro));
1359                         dst = (struct sockaddr_in *)&ro.ro_dst;
1360                         dst->sin_len = sizeof(*dst);
1361                         dst->sin_family = AF_INET;
1362                         dst->sin_addr = mreq.imr_multiaddr;
1363                         rtalloc_ign(&ro, RTF_CLONING);
1364                         if (ro.ro_rt == NULL) {
1365                                 error = EADDRNOTAVAIL;
1366                                 splx(s);
1367                                 break;
1368                         }
1369                         ifp = ro.ro_rt->rt_ifp;
1370                         RTFREE(ro.ro_rt);
1371                 }
1372                 else {
1373                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1374                 }
1375
1376                 /*
1377                  * See if we found an interface, and confirm that it
1378                  * supports multicast.
1379                  */
1380                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1381                         error = EADDRNOTAVAIL;
1382                         splx(s);
1383                         break;
1384                 }
1385                 /*
1386                  * See if the membership already exists or if all the
1387                  * membership slots are full.
1388                  */
1389                 imo = ip_findmoptions(inp);
1390                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1391                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1392                             imo->imo_membership[i]->inm_addr.s_addr
1393                                                 == mreq.imr_multiaddr.s_addr)
1394                                 break;
1395                 }
1396                 if (i < imo->imo_num_memberships) {
1397                         INP_UNLOCK(inp);
1398                         error = EADDRINUSE;
1399                         splx(s);
1400                         break;
1401                 }
1402                 if (imo->imo_num_memberships == imo->imo_max_memberships) {
1403                     struct in_multi **nmships, **omships;
1404                     size_t newmax;
1405                     /*
1406                      * Resize the vector to next power-of-two minus 1. If the
1407                      * size would exceed the maximum then we know we've really
1408                      * run out of entries. Otherwise, we realloc() the vector
1409                      * with the INP lock held to avoid introducing a race.
1410                      */
1411                     nmships = NULL;
1412                     omships = imo->imo_membership;
1413                     newmax = ((imo->imo_max_memberships + 1) * 2) - 1;
1414                     if (newmax <= IP_MAX_MEMBERSHIPS) {
1415                         nmships = (struct in_multi **)realloc(omships,
1416 sizeof(*nmships) * newmax, M_IPMOPTS, M_NOWAIT);
1417                         if (nmships != NULL) {
1418                             imo->imo_membership = nmships;
1419                             imo->imo_max_memberships = newmax;
1420                         }
1421                     }
1422                     if (nmships == NULL) {
1423                         INP_UNLOCK(inp);
1424                         error = ETOOMANYREFS;
1425                         splx(s);
1426                         break;
1427                     }
1428                 }
1429                 /*
1430                  * Everything looks good; add a new record to the multicast
1431                  * address list for the given interface.
1432                  */
1433                 if ((imo->imo_membership[i] =
1434                     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1435                         INP_UNLOCK(inp);
1436                         error = ENOBUFS;
1437                         splx(s);
1438                         break;
1439                 }
1440                 ++imo->imo_num_memberships;
1441                 INP_UNLOCK(inp);
1442                 splx(s);
1443                 break;
1444
1445         case IP_DROP_MEMBERSHIP:
1446                 /*
1447                  * Drop a multicast group membership.
1448                  * Group must be a valid IP multicast address.
1449                  */
1450                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1451                 if (error)
1452                         break;
1453
1454                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1455                         error = EINVAL;
1456                         break;
1457                 }
1458
1459                 s = splimp();
1460                 /*
1461                  * If an interface address was specified, get a pointer
1462                  * to its ifnet structure.
1463                  */
1464                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1465                         ifp = NULL;
1466                 else {
1467                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1468                         if (ifp == NULL) {
1469                                 error = EADDRNOTAVAIL;
1470                                 splx(s);
1471                                 break;
1472                         }
1473                 }
1474                 /*
1475                  * Find the membership in the membership array.
1476                  */
1477                 imo = ip_findmoptions(inp);
1478                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1479                         if ((ifp == NULL ||
1480                              imo->imo_membership[i]->inm_ifp == ifp) &&
1481                              imo->imo_membership[i]->inm_addr.s_addr ==
1482                              mreq.imr_multiaddr.s_addr)
1483                                 break;
1484                 }
1485                 if (i == imo->imo_num_memberships) {
1486                         INP_UNLOCK(inp);
1487                         error = EADDRNOTAVAIL;
1488                         splx(s);
1489                         break;
1490                 }
1491                 /*
1492                  * Give up the multicast address record to which the
1493                  * membership points.
1494                  */
1495                 in_delmulti(imo->imo_membership[i]);
1496                 /*
1497                  * Remove the gap in the membership array.
1498                  */
1499                 for (++i; i < imo->imo_num_memberships; ++i)
1500                         imo->imo_membership[i-1] = imo->imo_membership[i];
1501                 --imo->imo_num_memberships;
1502                 INP_UNLOCK(inp);
1503                 splx(s);
1504                 break;
1505
1506         default:
1507                 error = EOPNOTSUPP;
1508                 break;
1509         }
1510
1511         return (error);
1512 }
1513
1514 /*
1515  * Return the IP multicast options in response to user getsockopt().
1516  */
1517 static int
1518 ip_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1519 {
1520         struct ip_moptions *imo;
1521         struct in_addr addr;
1522         struct in_ifaddr *ia;
1523         int error, optval;
1524         u_char coptval;
1525
1526         INP_LOCK(inp);
1527         imo = inp->inp_moptions;
1528
1529         error = 0;
1530         switch (sopt->sopt_name) {
1531         case IP_MULTICAST_VIF:
1532                 if (imo != NULL)
1533                         optval = imo->imo_multicast_vif;
1534                 else
1535                         optval = -1;
1536                 INP_UNLOCK(inp);
1537                 error = sooptcopyout(sopt, &optval, sizeof optval);
1538                 break;
1539
1540         case IP_MULTICAST_IF:
1541                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
1542                         addr.s_addr = INADDR_ANY;
1543                 else if (imo->imo_multicast_addr.s_addr) {
1544                         /* return the value user has set */
1545                         addr = imo->imo_multicast_addr;
1546                 } else {
1547                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
1548                         addr.s_addr = (ia == NULL) ? INADDR_ANY
1549                                 : IA_SIN(ia)->sin_addr.s_addr;
1550                 }
1551                 INP_UNLOCK(inp);
1552                 error = sooptcopyout(sopt, &addr, sizeof addr);
1553                 break;
1554
1555         case IP_MULTICAST_TTL:
1556                 if (imo == 0)
1557                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1558                 else
1559                         optval = coptval = imo->imo_multicast_ttl;
1560                 INP_UNLOCK(inp);
1561                 if (sopt->sopt_valsize == 1)
1562                         error = sooptcopyout(sopt, &coptval, 1);
1563                 else
1564                         error = sooptcopyout(sopt, &optval, sizeof optval);
1565                 break;
1566
1567         case IP_MULTICAST_LOOP:
1568                 if (imo == 0)
1569                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1570                 else
1571                         optval = coptval = imo->imo_multicast_loop;
1572                 INP_UNLOCK(inp);
1573                 if (sopt->sopt_valsize == 1)
1574                         error = sooptcopyout(sopt, &coptval, 1);
1575                 else
1576                         error = sooptcopyout(sopt, &optval, sizeof optval);
1577                 break;
1578
1579         default:
1580                 INP_UNLOCK(inp);
1581                 error = ENOPROTOOPT;
1582                 break;
1583         }
1584         INP_UNLOCK_ASSERT(inp);
1585
1586         return (error);
1587 }
1588
1589 /*
1590  * Discard the IP multicast options.
1591  */
1592 void
1593 ip_freemoptions(struct ip_moptions *imo)
1594 {
1595         register int i;
1596
1597         if (imo != NULL) {
1598                 for (i = 0; i < imo->imo_num_memberships; ++i)
1599                         in_delmulti(imo->imo_membership[i]);
1600                 free(imo->imo_membership, M_IPMOPTS);
1601                 free(imo, M_IPMOPTS);
1602         }
1603 }
1604
1605 /*
1606  * Routine called from ip_output() to loop back a copy of an IP multicast
1607  * packet to the input queue of a specified interface.  Note that this
1608  * calls the output routine of the loopback "driver", but with an interface
1609  * pointer that might NOT be a loopback interface -- evil, but easier than
1610  * replicating that code here.
1611  */
1612 static void
1613 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
1614     int hlen)
1615 {
1616         register struct ip *ip;
1617         struct mbuf *copym;
1618
1619         copym = m_copy(m, 0, M_COPYALL);
1620         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
1621                 copym = m_pullup(copym, hlen);
1622         if (copym != NULL) {
1623                 /* If needed, compute the checksum and mark it as valid. */
1624                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1625                         in_delayed_cksum(copym);
1626                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1627                         copym->m_pkthdr.csum_flags |=
1628                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1629                         copym->m_pkthdr.csum_data = 0xffff;
1630                 }
1631                 /*
1632                  * We don't bother to fragment if the IP length is greater
1633                  * than the interface's MTU.  Can this possibly matter?
1634                  */
1635                 ip = mtod(copym, struct ip *);
1636                 ip->ip_len = htons(ip->ip_len);
1637                 ip->ip_off = htons(ip->ip_off);
1638                 ip->ip_sum = 0;
1639                 ip->ip_sum = in_cksum(copym, hlen);
1640                 /*
1641                  * NB:
1642                  * It's not clear whether there are any lingering
1643                  * reentrancy problems in other areas which might
1644                  * be exposed by using ip_input directly (in
1645                  * particular, everything which modifies the packet
1646                  * in-place).  Yet another option is using the
1647                  * protosw directly to deliver the looped back
1648                  * packet.  For the moment, we'll err on the side
1649                  * of safety by using if_simloop().
1650                  */
1651 #if 1 /* XXX */
1652                 if (dst->sin_family != AF_INET) {
1653                         printf("ip_mloopback: bad address family %d\n",
1654                                                 dst->sin_family);
1655                         dst->sin_family = AF_INET;
1656                 }
1657 #endif
1658
1659 #ifdef notdef
1660                 copym->m_pkthdr.rcvif = ifp;
1661                 ip_input(copym);
1662 #else
1663                 if_simloop(ifp, copym, dst->sin_family, 0);
1664 #endif
1665         }
1666 }