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 1.67 2008/10/28 03:07:28 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/priv.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/in_cksum.h>
  56 #include <sys/lock.h>
  57
  58 #include <sys/thread2.h>
  59 #include <sys/mplock2.h>
  60
  61 #include <net/if.h>
  62 #include <net/netisr.h>
  63 #include <net/pfil.h>
  64 #include <net/route.h>
  65
  66 #include <netinet/in.h>
  67 #include <netinet/in_systm.h>
  68 #include <netinet/ip.h>
  69 #include <netinet/in_pcb.h>
  70 #include <netinet/in_var.h>
  71 #include <netinet/ip_var.h>
  72
  73 #include <netproto/mpls/mpls_var.h>
  74
  75 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  76
  77 #ifdef IPSEC
  78 #include <netinet6/ipsec.h>
  79 #include <netproto/key/key.h>
  80 #ifdef IPSEC_DEBUG
  81 #include <netproto/key/key_debug.h>
  82 #else
  83 #define KEYDEBUG(lev,arg)
  84 #endif
  85 #endif /*IPSEC*/
  86
  87 #ifdef FAST_IPSEC
  88 #include <netproto/ipsec/ipsec.h>
  89 #include <netproto/ipsec/xform.h>
  90 #include <netproto/ipsec/key.h>
  91 #endif /*FAST_IPSEC*/
  92
  93 #include <net/ipfw/ip_fw.h>
  94 #include <net/dummynet/ip_dummynet.h>
  95
  96 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  97                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  98                                   (ntohl(a.s_addr)>>16)&0xFF,\
  99                                   (ntohl(a.s_addr)>>8)&0xFF,\
 100                                   (ntohl(a.s_addr))&0xFF, y);
 101
 102 u_short ip_id;
 103
 104 #ifdef MBUF_STRESS_TEST
 105 int mbuf_frag_size = 0;
 106 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 107         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 108 #endif
 109
 110 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 111 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 112 static void     ip_mloopback
 113         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 114 static int      ip_getmoptions
 115         (struct sockopt *, struct ip_moptions *);
 116 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 117 static int      ip_setmoptions
 118         (struct sockopt *, struct ip_moptions **);
 119
 120 int     ip_optcopy(struct ip *, struct ip *);
 121
 122 extern  int route_assert_owner_access;
 123
 124 extern  struct protosw inetsw[];
 125
 126 static int
 127 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
 128 {
 129         struct in_ifaddr_container *iac;
 130
 131         /*
 132          * We need to figure out if we have been forwarded to a local
 133          * socket.  If so, then we should somehow "loop back" to
 134          * ip_input(), and get directed to the PCB as if we had received
 135          * this packet.  This is because it may be difficult to identify
 136          * the packets you want to forward until they are being output
 137          * and have selected an interface (e.g. locally initiated
 138          * packets).  If we used the loopback inteface, we would not be
 139          * able to control what happens as the packet runs through
 140          * ip_input() as it is done through a ISR.
 141          */
 142         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 143                 /*
 144                  * If the addr to forward to is one of ours, we pretend
 145                  * to be the destination for this packet.
 146                  */
 147                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 148                         break;
 149         }
 150         if (iac != NULL) {
 151                 struct ip *ip;
 152
 153                 if (m->m_pkthdr.rcvif == NULL)
 154                         m->m_pkthdr.rcvif = ifunit("lo0");
 155                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 156                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 157                                                   CSUM_PSEUDO_HDR;
 158                         m->m_pkthdr.csum_data = 0xffff;
 159                 }
 160                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 161
 162                 /*
 163                  * Make sure that the IP header is in one mbuf,
 164                  * required by ip_input
 165                  */
 166                 if (m->m_len < hlen) {
 167                         m = m_pullup(m, hlen);
 168                         if (m == NULL) {
 169                                 /* The packet was freed; we are done */
 170                                 return 1;
 171                         }
 172                 }
 173                 ip = mtod(m, struct ip *);
 174
 175                 ip->ip_len = htons(ip->ip_len);
 176                 ip->ip_off = htons(ip->ip_off);
 177                 ip_input(m);
 178
 179                 return 1; /* The packet gets forwarded locally */
 180         }
 181         return 0;
 182 }
 183
 184 /*
 185  * IP output.  The packet in mbuf chain m contains a skeletal IP
 186  * header (with len, off, ttl, proto, tos, src, dst).
 187  * The mbuf chain containing the packet will be freed.
 188  * The mbuf opt, if present, will not be freed.
 189  */
 190 int
 191 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 192           int flags, struct ip_moptions *imo, struct inpcb *inp)
 193 {
 194         struct ip *ip;
 195         struct ifnet *ifp = NULL;       /* keep compiler happy */
 196         struct mbuf *m;
 197         int hlen = sizeof(struct ip);
 198         int len, error = 0;
 199         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 200         struct in_ifaddr *ia = NULL;
 201         int isbroadcast, sw_csum;
 202         struct in_addr pkt_dst;
 203         struct route iproute;
 204         struct m_tag *mtag;
 205 #ifdef IPSEC
 206         struct secpolicy *sp = NULL;
 207         struct socket *so = inp ? inp->inp_socket : NULL;
 208 #endif
 209 #ifdef FAST_IPSEC
 210         struct secpolicy *sp = NULL;
 211         struct tdb_ident *tdbi;
 212 #endif /* FAST_IPSEC */
 213         struct sockaddr_in *next_hop = NULL;
 214         int src_was_INADDR_ANY = 0;     /* as the name says... */
 215
 216         m = m0;
 217         M_ASSERTPKTHDR(m);
 218
 219         if (ro == NULL) {
 220                 ro = &iproute;
 221                 bzero(ro, sizeof *ro);
 222         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 223                 if (flags & IP_DEBUGROUTE) {
 224                         if (route_assert_owner_access) {
 225                                 panic("ip_output: "
 226                                       "rt rt_cpuid %d accessed on cpu %d\n",
 227                                       ro->ro_rt->rt_cpuid, mycpuid);
 228                         } else {
 229                                 kprintf("ip_output: "
 230                                         "rt rt_cpuid %d accessed on cpu %d\n",
 231                                         ro->ro_rt->rt_cpuid, mycpuid);
 232                                 print_backtrace(-1);
 233                         }
 234                 }
 235
 236                 /*
 237                  * XXX
 238                  * If the cached rtentry's owner CPU is not the current CPU,
 239                  * then don't touch the cached rtentry (remote free is too
 240                  * expensive in this context); just relocate the route.
 241                  */
 242                 ro = &iproute;
 243                 bzero(ro, sizeof *ro);
 244         }
 245
 246         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 247                 /* Next hop */
 248                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 249                 KKASSERT(mtag != NULL);
 250                 next_hop = m_tag_data(mtag);
 251         }
 252
 253         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 254                 struct dn_pkt *dn_pkt;
 255
 256                 /* Extract info from dummynet tag */
 257                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 258                 KKASSERT(mtag != NULL);
 259                 dn_pkt = m_tag_data(mtag);
 260
 261                 /*
 262                  * The packet was already tagged, so part of the
 263                  * processing was already done, and we need to go down.
 264                  * Get the calculated parameters from the tag.
 265                  */
 266                 ifp = dn_pkt->ifp;
 267
 268                 KKASSERT(ro == &iproute);
 269                 *ro = dn_pkt->ro; /* structure copy */
 270                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 271
 272                 dst = dn_pkt->dn_dst;
 273                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 274                         /* If 'dst' points into dummynet tag, adjust it */
 275                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 276                 }
 277
 278                 ip = mtod(m, struct ip *);
 279                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 280                 if (ro->ro_rt)
 281                         ia = ifatoia(ro->ro_rt->rt_ifa);
 282                 goto sendit;
 283         }
 284
 285         if (opt) {
 286                 len = 0;
 287                 m = ip_insertoptions(m, opt, &len);
 288                 if (len != 0)
 289                         hlen = len;
 290         }
 291         ip = mtod(m, struct ip *);
 292
 293         /*
 294          * Fill in IP header.
 295          */
 296         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 297                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 298                 ip->ip_off &= IP_DF;
 299                 ip->ip_id = ip_newid();
 300                 ipstat.ips_localout++;
 301         } else {
 302                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 303         }
 304
 305 reroute:
 306         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 307
 308 #ifdef INVARIANTS
 309         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 310                 /*
 311                  * XXX
 312                  * Multicast is not MPSAFE yet.  Caller must hold
 313                  * BGL when output a multicast IP packet.
 314                  */
 315                 ASSERT_MP_LOCK_HELD(curthread);
 316         }
 317 #endif
 318
 319         dst = (struct sockaddr_in *)&ro->ro_dst;
 320         /*
 321          * If there is a cached route,
 322          * check that it is to the same destination
 323          * and is still up.  If not, free it and try again.
 324          * The address family should also be checked in case of sharing the
 325          * cache with IPv6.
 326          */
 327         if (ro->ro_rt &&
 328             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 329              dst->sin_family != AF_INET ||
 330              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 331                 rtfree(ro->ro_rt);
 332                 ro->ro_rt = NULL;
 333         }
 334         if (ro->ro_rt == NULL) {
 335                 bzero(dst, sizeof *dst);
 336                 dst->sin_family = AF_INET;
 337                 dst->sin_len = sizeof *dst;
 338                 dst->sin_addr = pkt_dst;
 339         }
 340         /*
 341          * If routing to interface only,
 342          * short circuit routing lookup.
 343          */
 344         if (flags & IP_ROUTETOIF) {
 345                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 346                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 347                         ipstat.ips_noroute++;
 348                         error = ENETUNREACH;
 349                         goto bad;
 350                 }
 351                 ifp = ia->ia_ifp;
 352                 ip->ip_ttl = 1;
 353                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 354         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 355                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 356                 /*
 357                  * Bypass the normal routing lookup for multicast
 358                  * packets if the interface is specified.
 359                  */
 360                 ifp = imo->imo_multicast_ifp;
 361                 ia = IFP_TO_IA(ifp);
 362                 isbroadcast = 0;        /* fool gcc */
 363         } else {
 364                 /*
 365                  * If this is the case, we probably don't want to allocate
 366                  * a protocol-cloned route since we didn't get one from the
 367                  * ULP.  This lets TCP do its thing, while not burdening
 368                  * forwarding or ICMP with the overhead of cloning a route.
 369                  * Of course, we still want to do any cloning requested by
 370                  * the link layer, as this is probably required in all cases
 371                  * for correct operation (as it is for ARP).
 372                  */
 373                 if (ro->ro_rt == NULL)
 374                         rtalloc_ign(ro, RTF_PRCLONING);
 375                 if (ro->ro_rt == NULL) {
 376                         ipstat.ips_noroute++;
 377                         error = EHOSTUNREACH;
 378                         goto bad;
 379                 }
 380                 ia = ifatoia(ro->ro_rt->rt_ifa);
 381                 ifp = ro->ro_rt->rt_ifp;
 382                 ro->ro_rt->rt_use++;
 383                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 384                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 385                 if (ro->ro_rt->rt_flags & RTF_HOST)
 386                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 387                 else
 388                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 389         }
 390         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 391                 struct in_multi *inm;
 392
 393                 m->m_flags |= M_MCAST;
 394                 /*
 395                  * IP destination address is multicast.  Make sure "dst"
 396                  * still points to the address in "ro".  (It may have been
 397                  * changed to point to a gateway address, above.)
 398                  */
 399                 dst = (struct sockaddr_in *)&ro->ro_dst;
 400                 /*
 401                  * See if the caller provided any multicast options
 402                  */
 403                 if (imo != NULL) {
 404                         ip->ip_ttl = imo->imo_multicast_ttl;
 405                         if (imo->imo_multicast_vif != -1) {
 406                                 ip->ip_src.s_addr =
 407                                     ip_mcast_src ?
 408                                     ip_mcast_src(imo->imo_multicast_vif) :
 409                                     INADDR_ANY;
 410                         }
 411                 } else {
 412                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 413                 }
 414                 /*
 415                  * Confirm that the outgoing interface supports multicast.
 416                  */
 417                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 418                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 419                                 ipstat.ips_noroute++;
 420                                 error = ENETUNREACH;
 421                                 goto bad;
 422                         }
 423                 }
 424                 /*
 425                  * If source address not specified yet, use address
 426                  * of outgoing interface.
 427                  */
 428                 if (ip->ip_src.s_addr == INADDR_ANY) {
 429                         /* Interface may have no addresses. */
 430                         if (ia != NULL)
 431                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 432                 }
 433
 434                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 435                 if (inm != NULL &&
 436                     (imo == NULL || imo->imo_multicast_loop)) {
 437                         /*
 438                          * If we belong to the destination multicast group
 439                          * on the outgoing interface, and the caller did not
 440                          * forbid loopback, loop back a copy.
 441                          */
 442                         ip_mloopback(ifp, m, dst, hlen);
 443                 } else {
 444                         /*
 445                          * If we are acting as a multicast router, perform
 446                          * multicast forwarding as if the packet had just
 447                          * arrived on the interface to which we are about
 448                          * to send.  The multicast forwarding function
 449                          * recursively calls this function, using the
 450                          * IP_FORWARDING flag to prevent infinite recursion.
 451                          *
 452                          * Multicasts that are looped back by ip_mloopback(),
 453                          * above, will be forwarded by the ip_input() routine,
 454                          * if necessary.
 455                          */
 456                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 457                                 /*
 458                                  * If rsvp daemon is not running, do not
 459                                  * set ip_moptions. This ensures that the packet
 460                                  * is multicast and not just sent down one link
 461                                  * as prescribed by rsvpd.
 462                                  */
 463                                 if (!rsvp_on)
 464                                         imo = NULL;
 465                                 if (ip_mforward &&
 466                                     ip_mforward(ip, ifp, m, imo) != 0) {
 467                                         m_freem(m);
 468                                         goto done;
 469                                 }
 470                         }
 471                 }
 472
 473                 /*
 474                  * Multicasts with a time-to-live of zero may be looped-
 475                  * back, above, but must not be transmitted on a network.
 476                  * Also, multicasts addressed to the loopback interface
 477                  * are not sent -- the above call to ip_mloopback() will
 478                  * loop back a copy if this host actually belongs to the
 479                  * destination group on the loopback interface.
 480                  */
 481                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 482                         m_freem(m);
 483                         goto done;
 484                 }
 485
 486                 goto sendit;
 487         } else {
 488                 m->m_flags &= ~M_MCAST;
 489         }
 490
 491         /*
 492          * If the source address is not specified yet, use the address
 493          * of the outoing interface. In case, keep note we did that, so
 494          * if the the firewall changes the next-hop causing the output
 495          * interface to change, we can fix that.
 496          */
 497         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 498                 /* Interface may have no addresses. */
 499                 if (ia != NULL) {
 500                         ip->ip_src = IA_SIN(ia)->sin_addr;
 501                         src_was_INADDR_ANY = 1;
 502                 }
 503         }
 504
 505 #ifdef ALTQ
 506         /*
 507          * Disable packet drop hack.
 508          * Packetdrop should be done by queueing.
 509          */
 510 #else /* !ALTQ */
 511         /*
 512          * Verify that we have any chance at all of being able to queue
 513          *      the packet or packet fragments
 514          */
 515         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 516             ifp->if_snd.ifq_maxlen) {
 517                 error = ENOBUFS;
 518                 ipstat.ips_odropped++;
 519                 goto bad;
 520         }
 521 #endif /* !ALTQ */
 522
 523         /*
 524          * Look for broadcast address and
 525          * verify user is allowed to send
 526          * such a packet.
 527          */
 528         if (isbroadcast) {
 529                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 530                         error = EADDRNOTAVAIL;
 531                         goto bad;
 532                 }
 533                 if (!(flags & IP_ALLOWBROADCAST)) {
 534                         error = EACCES;
 535                         goto bad;
 536                 }
 537                 /* don't allow broadcast messages to be fragmented */
 538                 if (ip->ip_len > ifp->if_mtu) {
 539                         error = EMSGSIZE;
 540                         goto bad;
 541                 }
 542                 m->m_flags |= M_BCAST;
 543         } else {
 544                 m->m_flags &= ~M_BCAST;
 545         }
 546
 547 sendit:
 548 #ifdef IPSEC
 549         /* get SP for this packet */
 550         if (so == NULL)
 551                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 552         else
 553                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 554
 555         if (sp == NULL) {
 556                 ipsecstat.out_inval++;
 557                 goto bad;
 558         }
 559
 560         error = 0;
 561
 562         /* check policy */
 563         switch (sp->policy) {
 564         case IPSEC_POLICY_DISCARD:
 565                 /*
 566                  * This packet is just discarded.
 567                  */
 568                 ipsecstat.out_polvio++;
 569                 goto bad;
 570
 571         case IPSEC_POLICY_BYPASS:
 572         case IPSEC_POLICY_NONE:
 573                 /* no need to do IPsec. */
 574                 goto skip_ipsec;
 575
 576         case IPSEC_POLICY_IPSEC:
 577                 if (sp->req == NULL) {
 578                         /* acquire a policy */
 579                         error = key_spdacquire(sp);
 580                         goto bad;
 581                 }
 582                 break;
 583
 584         case IPSEC_POLICY_ENTRUST:
 585         default:
 586                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 587         }
 588     {
 589         struct ipsec_output_state state;
 590         bzero(&state, sizeof state);
 591         state.m = m;
 592         if (flags & IP_ROUTETOIF) {
 593                 state.ro = &iproute;
 594                 bzero(&iproute, sizeof iproute);
 595         } else
 596                 state.ro = ro;
 597         state.dst = (struct sockaddr *)dst;
 598
 599         ip->ip_sum = 0;
 600
 601         /*
 602          * XXX
 603          * delayed checksums are not currently compatible with IPsec
 604          */
 605         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 606                 in_delayed_cksum(m);
 607                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 608         }
 609
 610         ip->ip_len = htons(ip->ip_len);
 611         ip->ip_off = htons(ip->ip_off);
 612
 613         error = ipsec4_output(&state, sp, flags);
 614
 615         m = state.m;
 616         if (flags & IP_ROUTETOIF) {
 617                 /*
 618                  * if we have tunnel mode SA, we may need to ignore
 619                  * IP_ROUTETOIF.
 620                  */
 621                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 622                         flags &= ~IP_ROUTETOIF;
 623                         ro = state.ro;
 624                 }
 625         } else
 626                 ro = state.ro;
 627         dst = (struct sockaddr_in *)state.dst;
 628         if (error) {
 629                 /* mbuf is already reclaimed in ipsec4_output. */
 630                 m0 = NULL;
 631                 switch (error) {
 632                 case EHOSTUNREACH:
 633                 case ENETUNREACH:
 634                 case EMSGSIZE:
 635                 case ENOBUFS:
 636                 case ENOMEM:
 637                         break;
 638                 default:
 639                         kprintf("ip4_output (ipsec): error code %d\n", error);
 640                         /*fall through*/
 641                 case ENOENT:
 642                         /* don't show these error codes to the user */
 643                         error = 0;
 644                         break;
 645                 }
 646                 goto bad;
 647         }
 648     }
 649
 650         /* be sure to update variables that are affected by ipsec4_output() */
 651         ip = mtod(m, struct ip *);
 652 #ifdef _IP_VHL
 653         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 654 #else
 655         hlen = ip->ip_hl << 2;
 656 #endif
 657         if (ro->ro_rt == NULL) {
 658                 if (!(flags & IP_ROUTETOIF)) {
 659                         kprintf("ip_output: "
 660                                 "can't update route after IPsec processing\n");
 661                         error = EHOSTUNREACH;   /*XXX*/
 662                         goto bad;
 663                 }
 664         } else {
 665                 ia = ifatoia(ro->ro_rt->rt_ifa);
 666                 ifp = ro->ro_rt->rt_ifp;
 667         }
 668
 669         /* make it flipped, again. */
 670         ip->ip_len = ntohs(ip->ip_len);
 671         ip->ip_off = ntohs(ip->ip_off);
 672 skip_ipsec:
 673 #endif /*IPSEC*/
 674 #ifdef FAST_IPSEC
 675         /*
 676          * Check the security policy (SP) for the packet and, if
 677          * required, do IPsec-related processing.  There are two
 678          * cases here; the first time a packet is sent through
 679          * it will be untagged and handled by ipsec4_checkpolicy.
 680          * If the packet is resubmitted to ip_output (e.g. after
 681          * AH, ESP, etc. processing), there will be a tag to bypass
 682          * the lookup and related policy checking.
 683          */
 684         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 685         crit_enter();
 686         if (mtag != NULL) {
 687                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 688                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 689                 if (sp == NULL)
 690                         error = -EINVAL;        /* force silent drop */
 691                 m_tag_delete(m, mtag);
 692         } else {
 693                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 694                                         &error, inp);
 695         }
 696         /*
 697          * There are four return cases:
 698          *    sp != NULL                    apply IPsec policy
 699          *    sp == NULL, error == 0        no IPsec handling needed
 700          *    sp == NULL, error == -EINVAL  discard packet w/o error
 701          *    sp == NULL, error != 0        discard packet, report error
 702          */
 703         if (sp != NULL) {
 704                 /* Loop detection, check if ipsec processing already done */
 705                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 706                 for (mtag = m_tag_first(m); mtag != NULL;
 707                      mtag = m_tag_next(m, mtag)) {
 708                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 709                                 continue;
 710                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 711                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 712                                 continue;
 713                         /*
 714                          * Check if policy has an SA associated with it.
 715                          * This can happen when an SP has yet to acquire
 716                          * an SA; e.g. on first reference.  If it occurs,
 717                          * then we let ipsec4_process_packet do its thing.
 718                          */
 719                         if (sp->req->sav == NULL)
 720                                 break;
 721                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 722                         if (tdbi->spi == sp->req->sav->spi &&
 723                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 724                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 725                                  sizeof(union sockaddr_union)) == 0) {
 726                                 /*
 727                                  * No IPsec processing is needed, free
 728                                  * reference to SP.
 729                                  *
 730                                  * NB: null pointer to avoid free at
 731                                  *     done: below.
 732                                  */
 733                                 KEY_FREESP(&sp), sp = NULL;
 734                                 crit_exit();
 735                                 goto spd_done;
 736                         }
 737                 }
 738
 739                 /*
 740                  * Do delayed checksums now because we send before
 741                  * this is done in the normal processing path.
 742                  */
 743                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 744                         in_delayed_cksum(m);
 745                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 746                 }
 747
 748                 ip->ip_len = htons(ip->ip_len);
 749                 ip->ip_off = htons(ip->ip_off);
 750
 751                 /* NB: callee frees mbuf */
 752                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 753                 /*
 754                  * Preserve KAME behaviour: ENOENT can be returned
 755                  * when an SA acquire is in progress.  Don't propagate
 756                  * this to user-level; it confuses applications.
 757                  *
 758                  * XXX this will go away when the SADB is redone.
 759                  */
 760                 if (error == ENOENT)
 761                         error = 0;
 762                 crit_exit();
 763                 goto done;
 764         } else {
 765                 crit_exit();
 766
 767                 if (error != 0) {
 768                         /*
 769                          * Hack: -EINVAL is used to signal that a packet
 770                          * should be silently discarded.  This is typically
 771                          * because we asked key management for an SA and
 772                          * it was delayed (e.g. kicked up to IKE).
 773                          */
 774                         if (error == -EINVAL)
 775                                 error = 0;
 776                         goto bad;
 777                 } else {
 778                         /* No IPsec processing for this packet. */
 779                 }
 780 #ifdef notyet
 781                 /*
 782                  * If deferred crypto processing is needed, check that
 783                  * the interface supports it.
 784                  */
 785                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 786                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 787                         /* notify IPsec to do its own crypto */
 788                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 789                         error = EHOSTUNREACH;
 790                         goto bad;
 791                 }
 792 #endif
 793         }
 794 spd_done:
 795 #endif /* FAST_IPSEC */
 796
 797         /* We are already being fwd'd from a firewall. */
 798         if (next_hop != NULL)
 799                 goto pass;
 800
 801         /* No pfil hooks */
 802         if (!pfil_has_hooks(&inet_pfil_hook)) {
 803                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 804                         /*
 805                          * Strip dummynet tags from stranded packets
 806                          */
 807                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 808                         KKASSERT(mtag != NULL);
 809                         m_tag_delete(m, mtag);
 810                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 811                 }
 812                 goto pass;
 813         }
 814
 815         /*
 816          * IpHack's section.
 817          * - Xlate: translate packet's addr/port (NAT).
 818          * - Firewall: deny/allow/etc.
 819          * - Wrap: fake packet's addr/port <unimpl.>
 820          * - Encapsulate: put it in another IP and send out. <unimp.>
 821          */
 822
 823         /*
 824          * Run through list of hooks for output packets.
 825          */
 826         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 827         if (error != 0 || m == NULL)
 828                 goto done;
 829         ip = mtod(m, struct ip *);
 830
 831         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 832                 /*
 833                  * Check dst to make sure it is directly reachable on the
 834                  * interface we previously thought it was.
 835                  * If it isn't (which may be likely in some situations) we have
 836                  * to re-route it (ie, find a route for the next-hop and the
 837                  * associated interface) and set them here. This is nested
 838                  * forwarding which in most cases is undesirable, except where
 839                  * such control is nigh impossible. So we do it here.
 840                  * And I'm babbling.
 841                  */
 842                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 843                 KKASSERT(mtag != NULL);
 844                 next_hop = m_tag_data(mtag);
 845
 846                 /*
 847                  * Try local forwarding first
 848                  */
 849                 if (ip_localforward(m, next_hop, hlen))
 850                         goto done;
 851
 852                 /*
 853                  * Relocate the route based on next_hop.
 854                  * If the current route is inp's cache, keep it untouched.
 855                  */
 856                 if (ro == &iproute && ro->ro_rt != NULL) {
 857                         RTFREE(ro->ro_rt);
 858                         ro->ro_rt = NULL;
 859                 }
 860                 ro = &iproute;
 861                 bzero(ro, sizeof *ro);
 862
 863                 /*
 864                  * Forwarding to broadcast address is not allowed.
 865                  * XXX Should we follow IP_ROUTETOIF?
 866                  */
 867                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 868
 869                 /* We are doing forwarding now */
 870                 flags |= IP_FORWARDING;
 871
 872                 goto reroute;
 873         }
 874
 875         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 876                 struct dn_pkt *dn_pkt;
 877
 878                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 879                 KKASSERT(mtag != NULL);
 880                 dn_pkt = m_tag_data(mtag);
 881
 882                 /*
 883                  * Under certain cases it is not possible to recalculate
 884                  * 'ro' and 'dst', let alone 'flags', so just save them in
 885                  * dummynet tag and avoid the possible wrong reculcalation
 886                  * when we come back to ip_output() again.
 887                  *
 888                  * All other parameters have been already used and so they
 889                  * are not needed anymore.
 890                  * XXX if the ifp is deleted while a pkt is in dummynet,
 891                  * we are in trouble! (TODO use ifnet_detach_event)
 892                  *
 893                  * We need to copy *ro because for ICMP pkts (and maybe
 894                  * others) the caller passed a pointer into the stack;
 895                  * dst might also be a pointer into *ro so it needs to
 896                  * be updated.
 897                  */
 898                 dn_pkt->ro = *ro;
 899                 if (ro->ro_rt)
 900                         ro->ro_rt->rt_refcnt++;
 901                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 902                         /* 'dst' points into 'ro' */
 903                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 904                 }
 905                 dn_pkt->dn_dst = dst;
 906                 dn_pkt->flags = flags;
 907
 908                 ip_dn_queue(m);
 909                 goto done;
 910         }
 911 pass:
 912         /* 127/8 must not appear on wire - RFC1122. */
 913         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 914             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 915                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 916                         ipstat.ips_badaddr++;
 917                         error = EADDRNOTAVAIL;
 918                         goto bad;
 919                 }
 920         }
 921
 922         m->m_pkthdr.csum_flags |= CSUM_IP;
 923         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 924         if (sw_csum & CSUM_DELAY_DATA) {
 925                 in_delayed_cksum(m);
 926                 sw_csum &= ~CSUM_DELAY_DATA;
 927         }
 928         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 929
 930         /*
 931          * If small enough for interface, or the interface will take
 932          * care of the fragmentation for us, can just send directly.
 933          */
 934         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 935             !(ip->ip_off & IP_DF))) {
 936                 ip->ip_len = htons(ip->ip_len);
 937                 ip->ip_off = htons(ip->ip_off);
 938                 ip->ip_sum = 0;
 939                 if (sw_csum & CSUM_DELAY_IP) {
 940                         if (ip->ip_vhl == IP_VHL_BORING)
 941                                 ip->ip_sum = in_cksum_hdr(ip);
 942                         else
 943                                 ip->ip_sum = in_cksum(m, hlen);
 944                 }
 945
 946                 /* Record statistics for this interface address. */
 947                 if (!(flags & IP_FORWARDING) && ia) {
 948                         ia->ia_ifa.if_opackets++;
 949                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 950                 }
 951
 952 #ifdef IPSEC
 953                 /* clean ipsec history once it goes out of the node */
 954                 ipsec_delaux(m);
 955 #endif
 956
 957 #ifdef MBUF_STRESS_TEST
 958                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 959                         struct mbuf *m1, *m2;
 960                         int length, tmp;
 961
 962                         tmp = length = m->m_pkthdr.len;
 963
 964                         while ((length -= mbuf_frag_size) >= 1) {
 965                                 m1 = m_split(m, length, MB_DONTWAIT);
 966                                 if (m1 == NULL)
 967                                         break;
 968                                 m2 = m;
 969                                 while (m2->m_next != NULL)
 970                                         m2 = m2->m_next;
 971                                 m2->m_next = m1;
 972                         }
 973                         m->m_pkthdr.len = tmp;
 974                 }
 975 #endif
 976
 977 #ifdef MPLS
 978                 if (!mpls_output_process(m, ro->ro_rt))
 979                         goto done;
 980 #endif
 981                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 982                                        ro->ro_rt);
 983                 goto done;
 984         }
 985
 986         if (ip->ip_off & IP_DF) {
 987                 error = EMSGSIZE;
 988                 /*
 989                  * This case can happen if the user changed the MTU
 990                  * of an interface after enabling IP on it.  Because
 991                  * most netifs don't keep track of routes pointing to
 992                  * them, there is no way for one to update all its
 993                  * routes when the MTU is changed.
 994                  */
 995                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 996                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 997                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 998                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 999                 }
1000                 ipstat.ips_cantfrag++;
1001                 goto bad;
1002         }
1003
1004         /*
1005          * Too large for interface; fragment if possible. If successful,
1006          * on return, m will point to a list of packets to be sent.
1007          */
1008         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1009         if (error)
1010                 goto bad;
1011         for (; m; m = m0) {
1012                 m0 = m->m_nextpkt;
1013                 m->m_nextpkt = NULL;
1014 #ifdef IPSEC
1015                 /* clean ipsec history once it goes out of the node */
1016                 ipsec_delaux(m);
1017 #endif
1018                 if (error == 0) {
1019                         /* Record statistics for this interface address. */
1020                         if (ia != NULL) {
1021                                 ia->ia_ifa.if_opackets++;
1022                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1023                         }
1024 #ifdef MPLS
1025                         if (!mpls_output_process(m, ro->ro_rt))
1026                                 continue;
1027 #endif
1028                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1029                                                ro->ro_rt);
1030                 } else {
1031                         m_freem(m);
1032                 }
1033         }
1034
1035         if (error == 0)
1036                 ipstat.ips_fragmented++;
1037
1038 done:
1039         if (ro == &iproute && ro->ro_rt != NULL) {
1040                 RTFREE(ro->ro_rt);
1041                 ro->ro_rt = NULL;
1042         }
1043 #ifdef IPSEC
1044         if (sp != NULL) {
1045                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1046                         kprintf("DP ip_output call free SP:%p\n", sp));
1047                 key_freesp(sp);
1048         }
1049 #endif
1050 #ifdef FAST_IPSEC
1051         if (sp != NULL)
1052                 KEY_FREESP(&sp);
1053 #endif
1054         return (error);
1055 bad:
1056         m_freem(m);
1057         goto done;
1058 }
1059
1060 /*
1061  * Create a chain of fragments which fit the given mtu. m_frag points to the
1062  * mbuf to be fragmented; on return it points to the chain with the fragments.
1063  * Return 0 if no error. If error, m_frag may contain a partially built
1064  * chain of fragments that should be freed by the caller.
1065  *
1066  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1067  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1068  */
1069 int
1070 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1071             u_long if_hwassist_flags, int sw_csum)
1072 {
1073         int error = 0;
1074         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1075         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1076         int off;
1077         struct mbuf *m0 = *m_frag;      /* the original packet          */
1078         int firstlen;
1079         struct mbuf **mnext;
1080         int nfrags;
1081
1082         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1083                 ipstat.ips_cantfrag++;
1084                 return EMSGSIZE;
1085         }
1086
1087         /*
1088          * Must be able to put at least 8 bytes per fragment.
1089          */
1090         if (len < 8)
1091                 return EMSGSIZE;
1092
1093         /*
1094          * If the interface will not calculate checksums on
1095          * fragmented packets, then do it here.
1096          */
1097         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1098             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1099                 in_delayed_cksum(m0);
1100                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1101         }
1102
1103         if (len > PAGE_SIZE) {
1104                 /*
1105                  * Fragment large datagrams such that each segment
1106                  * contains a multiple of PAGE_SIZE amount of data,
1107                  * plus headers. This enables a receiver to perform
1108                  * page-flipping zero-copy optimizations.
1109                  *
1110                  * XXX When does this help given that sender and receiver
1111                  * could have different page sizes, and also mtu could
1112                  * be less than the receiver's page size ?
1113                  */
1114                 int newlen;
1115                 struct mbuf *m;
1116
1117                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1118                         off += m->m_len;
1119
1120                 /*
1121                  * firstlen (off - hlen) must be aligned on an
1122                  * 8-byte boundary
1123                  */
1124                 if (off < hlen)
1125                         goto smart_frag_failure;
1126                 off = ((off - hlen) & ~7) + hlen;
1127                 newlen = (~PAGE_MASK) & mtu;
1128                 if ((newlen + sizeof(struct ip)) > mtu) {
1129                         /* we failed, go back the default */
1130 smart_frag_failure:
1131                         newlen = len;
1132                         off = hlen + len;
1133                 }
1134                 len = newlen;
1135
1136         } else {
1137                 off = hlen + len;
1138         }
1139
1140         firstlen = off - hlen;
1141         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1142
1143         /*
1144          * Loop through length of segment after first fragment,
1145          * make new header and copy data of each part and link onto chain.
1146          * Here, m0 is the original packet, m is the fragment being created.
1147          * The fragments are linked off the m_nextpkt of the original
1148          * packet, which after processing serves as the first fragment.
1149          */
1150         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1151                 struct ip *mhip;        /* ip header on the fragment */
1152                 struct mbuf *m;
1153                 int mhlen = sizeof(struct ip);
1154
1155                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1156                 if (m == NULL) {
1157                         error = ENOBUFS;
1158                         ipstat.ips_odropped++;
1159                         goto done;
1160                 }
1161                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1162                 /*
1163                  * In the first mbuf, leave room for the link header, then
1164                  * copy the original IP header including options. The payload
1165                  * goes into an additional mbuf chain returned by m_copy().
1166                  */
1167                 m->m_data += max_linkhdr;
1168                 mhip = mtod(m, struct ip *);
1169                 *mhip = *ip;
1170                 if (hlen > sizeof(struct ip)) {
1171                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1172                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1173                 }
1174                 m->m_len = mhlen;
1175                 /* XXX do we need to add ip->ip_off below ? */
1176                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1177                 if (off + len >= ip->ip_len) {  /* last fragment */
1178                         len = ip->ip_len - off;
1179                         m->m_flags |= M_LASTFRAG;
1180                 } else
1181                         mhip->ip_off |= IP_MF;
1182                 mhip->ip_len = htons((u_short)(len + mhlen));
1183                 m->m_next = m_copy(m0, off, len);
1184                 if (m->m_next == NULL) {                /* copy failed */
1185                         m_free(m);
1186                         error = ENOBUFS;        /* ??? */
1187                         ipstat.ips_odropped++;
1188                         goto done;
1189                 }
1190                 m->m_pkthdr.len = mhlen + len;
1191                 m->m_pkthdr.rcvif = NULL;
1192                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1193                 mhip->ip_off = htons(mhip->ip_off);
1194                 mhip->ip_sum = 0;
1195                 if (sw_csum & CSUM_DELAY_IP)
1196                         mhip->ip_sum = in_cksum(m, mhlen);
1197                 *mnext = m;
1198                 mnext = &m->m_nextpkt;
1199         }
1200         ipstat.ips_ofragments += nfrags;
1201
1202         /* set first marker for fragment chain */
1203         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1204         m0->m_pkthdr.csum_data = nfrags;
1205
1206         /*
1207          * Update first fragment by trimming what's been copied out
1208          * and updating header.
1209          */
1210         m_adj(m0, hlen + firstlen - ip->ip_len);
1211         m0->m_pkthdr.len = hlen + firstlen;
1212         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1213         ip->ip_off |= IP_MF;
1214         ip->ip_off = htons(ip->ip_off);
1215         ip->ip_sum = 0;
1216         if (sw_csum & CSUM_DELAY_IP)
1217                 ip->ip_sum = in_cksum(m0, hlen);
1218
1219 done:
1220         *m_frag = m0;
1221         return error;
1222 }
1223
1224 void
1225 in_delayed_cksum(struct mbuf *m)
1226 {
1227         struct ip *ip;
1228         u_short csum, offset;
1229
1230         ip = mtod(m, struct ip *);
1231         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1232         csum = in_cksum_skip(m, ip->ip_len, offset);
1233         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1234                 csum = 0xffff;
1235         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1236
1237         if (offset + sizeof(u_short) > m->m_len) {
1238                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1239                     m->m_len, offset, ip->ip_p);
1240                 /*
1241                  * XXX
1242                  * this shouldn't happen, but if it does, the
1243                  * correct behavior may be to insert the checksum
1244                  * in the existing chain instead of rearranging it.
1245                  */
1246                 m = m_pullup(m, offset + sizeof(u_short));
1247         }
1248         *(u_short *)(m->m_data + offset) = csum;
1249 }
1250
1251 /*
1252  * Insert IP options into preformed packet.
1253  * Adjust IP destination as required for IP source routing,
1254  * as indicated by a non-zero in_addr at the start of the options.
1255  *
1256  * XXX This routine assumes that the packet has no options in place.
1257  */
1258 static struct mbuf *
1259 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1260 {
1261         struct ipoption *p = mtod(opt, struct ipoption *);
1262         struct mbuf *n;
1263         struct ip *ip = mtod(m, struct ip *);
1264         unsigned optlen;
1265
1266         optlen = opt->m_len - sizeof p->ipopt_dst;
1267         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1268                 *phlen = 0;
1269                 return (m);             /* XXX should fail */
1270         }
1271         if (p->ipopt_dst.s_addr)
1272                 ip->ip_dst = p->ipopt_dst;
1273         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1274                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1275                 if (n == NULL) {
1276                         *phlen = 0;
1277                         return (m);
1278                 }
1279                 n->m_pkthdr.rcvif = NULL;
1280                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1281                 m->m_len -= sizeof(struct ip);
1282                 m->m_data += sizeof(struct ip);
1283                 n->m_next = m;
1284                 m = n;
1285                 m->m_len = optlen + sizeof(struct ip);
1286                 m->m_data += max_linkhdr;
1287                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1288         } else {
1289                 m->m_data -= optlen;
1290                 m->m_len += optlen;
1291                 m->m_pkthdr.len += optlen;
1292                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1293         }
1294         ip = mtod(m, struct ip *);
1295         bcopy(p->ipopt_list, ip + 1, optlen);
1296         *phlen = sizeof(struct ip) + optlen;
1297         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1298         ip->ip_len += optlen;
1299         return (m);
1300 }
1301
1302 /*
1303  * Copy options from ip to jp,
1304  * omitting those not copied during fragmentation.
1305  */
1306 int
1307 ip_optcopy(struct ip *ip, struct ip *jp)
1308 {
1309         u_char *cp, *dp;
1310         int opt, optlen, cnt;
1311
1312         cp = (u_char *)(ip + 1);
1313         dp = (u_char *)(jp + 1);
1314         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1315         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1316                 opt = cp[0];
1317                 if (opt == IPOPT_EOL)
1318                         break;
1319                 if (opt == IPOPT_NOP) {
1320                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1321                         *dp++ = IPOPT_NOP;
1322                         optlen = 1;
1323                         continue;
1324                 }
1325
1326                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1327                     ("ip_optcopy: malformed ipv4 option"));
1328                 optlen = cp[IPOPT_OLEN];
1329                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1330                     ("ip_optcopy: malformed ipv4 option"));
1331
1332                 /* bogus lengths should have been caught by ip_dooptions */
1333                 if (optlen > cnt)
1334                         optlen = cnt;
1335                 if (IPOPT_COPIED(opt)) {
1336                         bcopy(cp, dp, optlen);
1337                         dp += optlen;
1338                 }
1339         }
1340         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1341                 *dp++ = IPOPT_EOL;
1342         return (optlen);
1343 }
1344
1345 /*
1346  * IP socket option processing.
1347  */
1348 int
1349 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1350 {
1351         struct  inpcb *inp = so->so_pcb;
1352         int     error, optval;
1353
1354         error = optval = 0;
1355         if (sopt->sopt_level != IPPROTO_IP) {
1356                 return (EINVAL);
1357         }
1358
1359         switch (sopt->sopt_dir) {
1360         case SOPT_SET:
1361                 switch (sopt->sopt_name) {
1362                 case IP_OPTIONS:
1363 #ifdef notyet
1364                 case IP_RETOPTS:
1365 #endif
1366                 {
1367                         struct mbuf *m;
1368                         if (sopt->sopt_valsize > MLEN) {
1369                                 error = EMSGSIZE;
1370                                 break;
1371                         }
1372                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1373                         if (m == NULL) {
1374                                 error = ENOBUFS;
1375                                 break;
1376                         }
1377                         m->m_len = sopt->sopt_valsize;
1378                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1379                                               m->m_len);
1380                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1381                                            m));
1382                 }
1383
1384                 case IP_TOS:
1385                 case IP_TTL:
1386                 case IP_MINTTL:
1387                 case IP_RECVOPTS:
1388                 case IP_RECVRETOPTS:
1389                 case IP_RECVDSTADDR:
1390                 case IP_RECVIF:
1391                 case IP_RECVTTL:
1392                 case IP_FAITH:
1393                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1394                                              sizeof optval);
1395                         if (error)
1396                                 break;
1397                         switch (sopt->sopt_name) {
1398                         case IP_TOS:
1399                                 inp->inp_ip_tos = optval;
1400                                 break;
1401
1402                         case IP_TTL:
1403                                 inp->inp_ip_ttl = optval;
1404                                 break;
1405                         case IP_MINTTL:
1406                                 if (optval >= 0 && optval <= MAXTTL)
1407                                         inp->inp_ip_minttl = optval;
1408                                 else
1409                                         error = EINVAL;
1410                                 break;
1411 #define OPTSET(bit) \
1412         if (optval) \
1413                 inp->inp_flags |= bit; \
1414         else \
1415                 inp->inp_flags &= ~bit;
1416
1417                         case IP_RECVOPTS:
1418                                 OPTSET(INP_RECVOPTS);
1419                                 break;
1420
1421                         case IP_RECVRETOPTS:
1422                                 OPTSET(INP_RECVRETOPTS);
1423                                 break;
1424
1425                         case IP_RECVDSTADDR:
1426                                 OPTSET(INP_RECVDSTADDR);
1427                                 break;
1428
1429                         case IP_RECVIF:
1430                                 OPTSET(INP_RECVIF);
1431                                 break;
1432
1433                         case IP_RECVTTL:
1434                                 OPTSET(INP_RECVTTL);
1435                                 break;
1436
1437                         case IP_FAITH:
1438                                 OPTSET(INP_FAITH);
1439                                 break;
1440                         }
1441                         break;
1442 #undef OPTSET
1443
1444                 case IP_MULTICAST_IF:
1445                 case IP_MULTICAST_VIF:
1446                 case IP_MULTICAST_TTL:
1447                 case IP_MULTICAST_LOOP:
1448                 case IP_ADD_MEMBERSHIP:
1449                 case IP_DROP_MEMBERSHIP:
1450                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1451                         break;
1452
1453                 case IP_PORTRANGE:
1454                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1455                                             sizeof optval);
1456                         if (error)
1457                                 break;
1458
1459                         switch (optval) {
1460                         case IP_PORTRANGE_DEFAULT:
1461                                 inp->inp_flags &= ~(INP_LOWPORT);
1462                                 inp->inp_flags &= ~(INP_HIGHPORT);
1463                                 break;
1464
1465                         case IP_PORTRANGE_HIGH:
1466                                 inp->inp_flags &= ~(INP_LOWPORT);
1467                                 inp->inp_flags |= INP_HIGHPORT;
1468                                 break;
1469
1470                         case IP_PORTRANGE_LOW:
1471                                 inp->inp_flags &= ~(INP_HIGHPORT);
1472                                 inp->inp_flags |= INP_LOWPORT;
1473                                 break;
1474
1475                         default:
1476                                 error = EINVAL;
1477                                 break;
1478                         }
1479                         break;
1480
1481 #if defined(IPSEC) || defined(FAST_IPSEC)
1482                 case IP_IPSEC_POLICY:
1483                 {
1484                         caddr_t req;
1485                         size_t len = 0;
1486                         int priv;
1487                         struct mbuf *m;
1488                         int optname;
1489
1490                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1491                                 break;
1492                         soopt_to_mbuf(sopt, m);
1493                         priv = (sopt->sopt_td != NULL &&
1494                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1495                         req = mtod(m, caddr_t);
1496                         len = m->m_len;
1497                         optname = sopt->sopt_name;
1498                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1499                         m_freem(m);
1500                         break;
1501                 }
1502 #endif /*IPSEC*/
1503
1504                 default:
1505                         error = ENOPROTOOPT;
1506                         break;
1507                 }
1508                 break;
1509
1510         case SOPT_GET:
1511                 switch (sopt->sopt_name) {
1512                 case IP_OPTIONS:
1513                 case IP_RETOPTS:
1514                         if (inp->inp_options)
1515                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1516                                                            char *),
1517                                                 inp->inp_options->m_len);
1518                         else
1519                                 sopt->sopt_valsize = 0;
1520                         break;
1521
1522                 case IP_TOS:
1523                 case IP_TTL:
1524                 case IP_MINTTL:
1525                 case IP_RECVOPTS:
1526                 case IP_RECVRETOPTS:
1527                 case IP_RECVDSTADDR:
1528                 case IP_RECVTTL:
1529                 case IP_RECVIF:
1530                 case IP_PORTRANGE:
1531                 case IP_FAITH:
1532                         switch (sopt->sopt_name) {
1533
1534                         case IP_TOS:
1535                                 optval = inp->inp_ip_tos;
1536                                 break;
1537
1538                         case IP_TTL:
1539                                 optval = inp->inp_ip_ttl;
1540                                 break;
1541                         case IP_MINTTL:
1542                                 optval = inp->inp_ip_minttl;
1543                                 break;
1544
1545 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1546
1547                         case IP_RECVOPTS:
1548                                 optval = OPTBIT(INP_RECVOPTS);
1549                                 break;
1550
1551                         case IP_RECVRETOPTS:
1552                                 optval = OPTBIT(INP_RECVRETOPTS);
1553                                 break;
1554
1555                         case IP_RECVDSTADDR:
1556                                 optval = OPTBIT(INP_RECVDSTADDR);
1557                                 break;
1558
1559                         case IP_RECVTTL:
1560                                 optval = OPTBIT(INP_RECVTTL);
1561                                 break;
1562
1563                         case IP_RECVIF:
1564                                 optval = OPTBIT(INP_RECVIF);
1565                                 break;
1566
1567                         case IP_PORTRANGE:
1568                                 if (inp->inp_flags & INP_HIGHPORT)
1569                                         optval = IP_PORTRANGE_HIGH;
1570                                 else if (inp->inp_flags & INP_LOWPORT)
1571                                         optval = IP_PORTRANGE_LOW;
1572                                 else
1573                                         optval = 0;
1574                                 break;
1575
1576                         case IP_FAITH:
1577                                 optval = OPTBIT(INP_FAITH);
1578                                 break;
1579                         }
1580                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1581                         break;
1582
1583                 case IP_MULTICAST_IF:
1584                 case IP_MULTICAST_VIF:
1585                 case IP_MULTICAST_TTL:
1586                 case IP_MULTICAST_LOOP:
1587                 case IP_ADD_MEMBERSHIP:
1588                 case IP_DROP_MEMBERSHIP:
1589                         error = ip_getmoptions(sopt, inp->inp_moptions);
1590                         break;
1591
1592 #if defined(IPSEC) || defined(FAST_IPSEC)
1593                 case IP_IPSEC_POLICY:
1594                 {
1595                         struct mbuf *m = NULL;
1596                         caddr_t req = NULL;
1597                         size_t len = 0;
1598
1599                         if (m != NULL) {
1600                                 req = mtod(m, caddr_t);
1601                                 len = m->m_len;
1602                         }
1603                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1604                         if (error == 0)
1605                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1606                         if (error == 0)
1607                                 m_freem(m);
1608                         break;
1609                 }
1610 #endif /*IPSEC*/
1611
1612                 default:
1613                         error = ENOPROTOOPT;
1614                         break;
1615                 }
1616                 break;
1617         }
1618         return (error);
1619 }
1620
1621 /*
1622  * Set up IP options in pcb for insertion in output packets.
1623  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1624  * with destination address if source routed.
1625  */
1626 static int
1627 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1628 {
1629         int cnt, optlen;
1630         u_char *cp;
1631         u_char opt;
1632
1633         /* turn off any old options */
1634         if (*pcbopt)
1635                 m_free(*pcbopt);
1636         *pcbopt = 0;
1637         if (m == NULL || m->m_len == 0) {
1638                 /*
1639                  * Only turning off any previous options.
1640                  */
1641                 if (m != NULL)
1642                         m_free(m);
1643                 return (0);
1644         }
1645
1646         if (m->m_len % sizeof(int32_t))
1647                 goto bad;
1648         /*
1649          * IP first-hop destination address will be stored before
1650          * actual options; move other options back
1651          * and clear it when none present.
1652          */
1653         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1654                 goto bad;
1655         cnt = m->m_len;
1656         m->m_len += sizeof(struct in_addr);
1657         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1658         ovbcopy(mtod(m, caddr_t), cp, cnt);
1659         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1660
1661         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1662                 opt = cp[IPOPT_OPTVAL];
1663                 if (opt == IPOPT_EOL)
1664                         break;
1665                 if (opt == IPOPT_NOP)
1666                         optlen = 1;
1667                 else {
1668                         if (cnt < IPOPT_OLEN + sizeof *cp)
1669                                 goto bad;
1670                         optlen = cp[IPOPT_OLEN];
1671                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1672                                 goto bad;
1673                 }
1674                 switch (opt) {
1675
1676                 default:
1677                         break;
1678
1679                 case IPOPT_LSRR:
1680                 case IPOPT_SSRR:
1681                         /*
1682                          * user process specifies route as:
1683                          *      ->A->B->C->D
1684                          * D must be our final destination (but we can't
1685                          * check that since we may not have connected yet).
1686                          * A is first hop destination, which doesn't appear in
1687                          * actual IP option, but is stored before the options.
1688                          */
1689                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1690                                 goto bad;
1691                         m->m_len -= sizeof(struct in_addr);
1692                         cnt -= sizeof(struct in_addr);
1693                         optlen -= sizeof(struct in_addr);
1694                         cp[IPOPT_OLEN] = optlen;
1695                         /*
1696                          * Move first hop before start of options.
1697                          */
1698                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1699                               sizeof(struct in_addr));
1700                         /*
1701                          * Then copy rest of options back
1702                          * to close up the deleted entry.
1703                          */
1704                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1705                                 &cp[IPOPT_OFFSET+1],
1706                                 cnt - (IPOPT_MINOFF - 1));
1707                         break;
1708                 }
1709         }
1710         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1711                 goto bad;
1712         *pcbopt = m;
1713         return (0);
1714
1715 bad:
1716         m_free(m);
1717         return (EINVAL);
1718 }
1719
1720 /*
1721  * XXX
1722  * The whole multicast option thing needs to be re-thought.
1723  * Several of these options are equally applicable to non-multicast
1724  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1725  * standard option (IP_TTL).
1726  */
1727
1728 /*
1729  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1730  */
1731 static struct ifnet *
1732 ip_multicast_if(struct in_addr *a, int *ifindexp)
1733 {
1734         int ifindex;
1735         struct ifnet *ifp;
1736
1737         if (ifindexp)
1738                 *ifindexp = 0;
1739         if (ntohl(a->s_addr) >> 24 == 0) {
1740                 ifindex = ntohl(a->s_addr) & 0xffffff;
1741                 if (ifindex < 0 || if_index < ifindex)
1742                         return NULL;
1743                 ifp = ifindex2ifnet[ifindex];
1744                 if (ifindexp)
1745                         *ifindexp = ifindex;
1746         } else {
1747                 ifp = INADDR_TO_IFP(a);
1748         }
1749         return ifp;
1750 }
1751
1752 /*
1753  * Set the IP multicast options in response to user setsockopt().
1754  */
1755 static int
1756 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1757 {
1758         int error = 0;
1759         int i;
1760         struct in_addr addr;
1761         struct ip_mreq mreq;
1762         struct ifnet *ifp;
1763         struct ip_moptions *imo = *imop;
1764         int ifindex;
1765
1766         if (imo == NULL) {
1767                 /*
1768                  * No multicast option buffer attached to the pcb;
1769                  * allocate one and initialize to default values.
1770                  */
1771                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1772
1773                 *imop = imo;
1774                 imo->imo_multicast_ifp = NULL;
1775                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1776                 imo->imo_multicast_vif = -1;
1777                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1778                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1779                 imo->imo_num_memberships = 0;
1780         }
1781         switch (sopt->sopt_name) {
1782         /* store an index number for the vif you wanna use in the send */
1783         case IP_MULTICAST_VIF:
1784                 if (legal_vif_num == 0) {
1785                         error = EOPNOTSUPP;
1786                         break;
1787                 }
1788                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1789                 if (error)
1790                         break;
1791                 if (!legal_vif_num(i) && (i != -1)) {
1792                         error = EINVAL;
1793                         break;
1794                 }
1795                 imo->imo_multicast_vif = i;
1796                 break;
1797
1798         case IP_MULTICAST_IF:
1799                 /*
1800                  * Select the interface for outgoing multicast packets.
1801                  */
1802                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1803                 if (error)
1804                         break;
1805
1806                 /*
1807                  * INADDR_ANY is used to remove a previous selection.
1808                  * When no interface is selected, a default one is
1809                  * chosen every time a multicast packet is sent.
1810                  */
1811                 if (addr.s_addr == INADDR_ANY) {
1812                         imo->imo_multicast_ifp = NULL;
1813                         break;
1814                 }
1815                 /*
1816                  * The selected interface is identified by its local
1817                  * IP address.  Find the interface and confirm that
1818                  * it supports multicasting.
1819                  */
1820                 crit_enter();
1821                 ifp = ip_multicast_if(&addr, &ifindex);
1822                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1823                         crit_exit();
1824                         error = EADDRNOTAVAIL;
1825                         break;
1826                 }
1827                 imo->imo_multicast_ifp = ifp;
1828                 if (ifindex)
1829                         imo->imo_multicast_addr = addr;
1830                 else
1831                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1832                 crit_exit();
1833                 break;
1834
1835         case IP_MULTICAST_TTL:
1836                 /*
1837                  * Set the IP time-to-live for outgoing multicast packets.
1838                  * The original multicast API required a char argument,
1839                  * which is inconsistent with the rest of the socket API.
1840                  * We allow either a char or an int.
1841                  */
1842                 if (sopt->sopt_valsize == 1) {
1843                         u_char ttl;
1844                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1845                         if (error)
1846                                 break;
1847                         imo->imo_multicast_ttl = ttl;
1848                 } else {
1849                         u_int ttl;
1850                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1851                         if (error)
1852                                 break;
1853                         if (ttl > 255)
1854                                 error = EINVAL;
1855                         else
1856                                 imo->imo_multicast_ttl = ttl;
1857                 }
1858                 break;
1859
1860         case IP_MULTICAST_LOOP:
1861                 /*
1862                  * Set the loopback flag for outgoing multicast packets.
1863                  * Must be zero or one.  The original multicast API required a
1864                  * char argument, which is inconsistent with the rest
1865                  * of the socket API.  We allow either a char or an int.
1866                  */
1867                 if (sopt->sopt_valsize == 1) {
1868                         u_char loop;
1869
1870                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1871                         if (error)
1872                                 break;
1873                         imo->imo_multicast_loop = !!loop;
1874                 } else {
1875                         u_int loop;
1876
1877                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1878                                             sizeof loop);
1879                         if (error)
1880                                 break;
1881                         imo->imo_multicast_loop = !!loop;
1882                 }
1883                 break;
1884
1885         case IP_ADD_MEMBERSHIP:
1886                 /*
1887                  * Add a multicast group membership.
1888                  * Group must be a valid IP multicast address.
1889                  */
1890                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1891                 if (error)
1892                         break;
1893
1894                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1895                         error = EINVAL;
1896                         break;
1897                 }
1898                 crit_enter();
1899                 /*
1900                  * If no interface address was provided, use the interface of
1901                  * the route to the given multicast address.
1902                  */
1903                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1904                         struct sockaddr_in dst;
1905                         struct rtentry *rt;
1906
1907                         bzero(&dst, sizeof(struct sockaddr_in));
1908                         dst.sin_len = sizeof(struct sockaddr_in);
1909                         dst.sin_family = AF_INET;
1910                         dst.sin_addr = mreq.imr_multiaddr;
1911                         rt = rtlookup((struct sockaddr *)&dst);
1912                         if (rt == NULL) {
1913                                 error = EADDRNOTAVAIL;
1914                                 crit_exit();
1915                                 break;
1916                         }
1917                         --rt->rt_refcnt;
1918                         ifp = rt->rt_ifp;
1919                 } else {
1920                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1921                 }
1922
1923                 /*
1924                  * See if we found an interface, and confirm that it
1925                  * supports multicast.
1926                  */
1927                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1928                         error = EADDRNOTAVAIL;
1929                         crit_exit();
1930                         break;
1931                 }
1932                 /*
1933                  * See if the membership already exists or if all the
1934                  * membership slots are full.
1935                  */
1936                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1937                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1938                             imo->imo_membership[i]->inm_addr.s_addr
1939                                                 == mreq.imr_multiaddr.s_addr)
1940                                 break;
1941                 }
1942                 if (i < imo->imo_num_memberships) {
1943                         error = EADDRINUSE;
1944                         crit_exit();
1945                         break;
1946                 }
1947                 if (i == IP_MAX_MEMBERSHIPS) {
1948                         error = ETOOMANYREFS;
1949                         crit_exit();
1950                         break;
1951                 }
1952                 /*
1953                  * Everything looks good; add a new record to the multicast
1954                  * address list for the given interface.
1955                  */
1956                 if ((imo->imo_membership[i] =
1957                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1958                         error = ENOBUFS;
1959                         crit_exit();
1960                         break;
1961                 }
1962                 ++imo->imo_num_memberships;
1963                 crit_exit();
1964                 break;
1965
1966         case IP_DROP_MEMBERSHIP:
1967                 /*
1968                  * Drop a multicast group membership.
1969                  * Group must be a valid IP multicast address.
1970                  */
1971                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1972                 if (error)
1973                         break;
1974
1975                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1976                         error = EINVAL;
1977                         break;
1978                 }
1979
1980                 crit_enter();
1981                 /*
1982                  * If an interface address was specified, get a pointer
1983                  * to its ifnet structure.
1984                  */
1985                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1986                         ifp = NULL;
1987                 else {
1988                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1989                         if (ifp == NULL) {
1990                                 error = EADDRNOTAVAIL;
1991                                 crit_exit();
1992                                 break;
1993                         }
1994                 }
1995                 /*
1996                  * Find the membership in the membership array.
1997                  */
1998                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1999                         if ((ifp == NULL ||
2000                              imo->imo_membership[i]->inm_ifp == ifp) &&
2001                             imo->imo_membership[i]->inm_addr.s_addr ==
2002                             mreq.imr_multiaddr.s_addr)
2003                                 break;
2004                 }
2005                 if (i == imo->imo_num_memberships) {
2006                         error = EADDRNOTAVAIL;
2007                         crit_exit();
2008                         break;
2009                 }
2010                 /*
2011                  * Give up the multicast address record to which the
2012                  * membership points.
2013                  */
2014                 in_delmulti(imo->imo_membership[i]);
2015                 /*
2016                  * Remove the gap in the membership array.
2017                  */
2018                 for (++i; i < imo->imo_num_memberships; ++i)
2019                         imo->imo_membership[i-1] = imo->imo_membership[i];
2020                 --imo->imo_num_memberships;
2021                 crit_exit();
2022                 break;
2023
2024         default:
2025                 error = EOPNOTSUPP;
2026                 break;
2027         }
2028
2029         /*
2030          * If all options have default values, no need to keep the mbuf.
2031          */
2032         if (imo->imo_multicast_ifp == NULL &&
2033             imo->imo_multicast_vif == -1 &&
2034             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2035             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2036             imo->imo_num_memberships == 0) {
2037                 kfree(*imop, M_IPMOPTS);
2038                 *imop = NULL;
2039         }
2040
2041         return (error);
2042 }
2043
2044 /*
2045  * Return the IP multicast options in response to user getsockopt().
2046  */
2047 static int
2048 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2049 {
2050         struct in_addr addr;
2051         struct in_ifaddr *ia;
2052         int error, optval;
2053         u_char coptval;
2054
2055         error = 0;
2056         switch (sopt->sopt_name) {
2057         case IP_MULTICAST_VIF:
2058                 if (imo != NULL)
2059                         optval = imo->imo_multicast_vif;
2060                 else
2061                         optval = -1;
2062                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2063                 break;
2064
2065         case IP_MULTICAST_IF:
2066                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2067                         addr.s_addr = INADDR_ANY;
2068                 else if (imo->imo_multicast_addr.s_addr) {
2069                         /* return the value user has set */
2070                         addr = imo->imo_multicast_addr;
2071                 } else {
2072                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2073                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2074                                 : IA_SIN(ia)->sin_addr.s_addr;
2075                 }
2076                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2077                 break;
2078
2079         case IP_MULTICAST_TTL:
2080                 if (imo == NULL)
2081                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2082                 else
2083                         optval = coptval = imo->imo_multicast_ttl;
2084                 if (sopt->sopt_valsize == 1)
2085                         soopt_from_kbuf(sopt, &coptval, 1);
2086                 else
2087                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2088                 break;
2089
2090         case IP_MULTICAST_LOOP:
2091                 if (imo == NULL)
2092                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2093                 else
2094                         optval = coptval = imo->imo_multicast_loop;
2095                 if (sopt->sopt_valsize == 1)
2096                         soopt_from_kbuf(sopt, &coptval, 1);
2097                 else
2098                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2099                 break;
2100
2101         default:
2102                 error = ENOPROTOOPT;
2103                 break;
2104         }
2105         return (error);
2106 }
2107
2108 /*
2109  * Discard the IP multicast options.
2110  */
2111 void
2112 ip_freemoptions(struct ip_moptions *imo)
2113 {
2114         int i;
2115
2116         if (imo != NULL) {
2117                 for (i = 0; i < imo->imo_num_memberships; ++i)
2118                         in_delmulti(imo->imo_membership[i]);
2119                 kfree(imo, M_IPMOPTS);
2120         }
2121 }
2122
2123 /*
2124  * Routine called from ip_output() to loop back a copy of an IP multicast
2125  * packet to the input queue of a specified interface.  Note that this
2126  * calls the output routine of the loopback "driver", but with an interface
2127  * pointer that might NOT be a loopback interface -- evil, but easier than
2128  * replicating that code here.
2129  */
2130 static void
2131 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2132              int hlen)
2133 {
2134         struct ip *ip;
2135         struct mbuf *copym;
2136
2137         copym = m_copypacket(m, MB_DONTWAIT);
2138         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2139                 copym = m_pullup(copym, hlen);
2140         if (copym != NULL) {
2141                 /*
2142                  * if the checksum hasn't been computed, mark it as valid
2143                  */
2144                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2145                         in_delayed_cksum(copym);
2146                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2147                         copym->m_pkthdr.csum_flags |=
2148                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2149                         copym->m_pkthdr.csum_data = 0xffff;
2150                 }
2151                 /*
2152                  * We don't bother to fragment if the IP length is greater
2153                  * than the interface's MTU.  Can this possibly matter?
2154                  */
2155                 ip = mtod(copym, struct ip *);
2156                 ip->ip_len = htons(ip->ip_len);
2157                 ip->ip_off = htons(ip->ip_off);
2158                 ip->ip_sum = 0;
2159                 if (ip->ip_vhl == IP_VHL_BORING) {
2160                         ip->ip_sum = in_cksum_hdr(ip);
2161                 } else {
2162                         ip->ip_sum = in_cksum(copym, hlen);
2163                 }
2164                 /*
2165                  * NB:
2166                  * It's not clear whether there are any lingering
2167                  * reentrancy problems in other areas which might
2168                  * be exposed by using ip_input directly (in
2169                  * particular, everything which modifies the packet
2170                  * in-place).  Yet another option is using the
2171                  * protosw directly to deliver the looped back
2172                  * packet.  For the moment, we'll err on the side
2173                  * of safety by using if_simloop().
2174                  */
2175 #if 1 /* XXX */
2176                 if (dst->sin_family != AF_INET) {
2177                         kprintf("ip_mloopback: bad address family %d\n",
2178                                                 dst->sin_family);
2179                         dst->sin_family = AF_INET;
2180                 }
2181 #endif
2182                 if_simloop(ifp, copym, dst->sin_family, 0);
2183         }
2184 }