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  */
  32
  33 #define _IP_VHL
  34
  35 #include "opt_ipdn.h"
  36 #include "opt_ipdivert.h"
  37 #include "opt_ipsec.h"
  38 #include "opt_mbuf_stress_test.h"
  39 #include "opt_mpls.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/kernel.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mbuf.h>
  46 #include <sys/protosw.h>
  47 #include <sys/socket.h>
  48 #include <sys/socketvar.h>
  49 #include <sys/proc.h>
  50 #include <sys/priv.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/in_cksum.h>
  53 #include <sys/lock.h>
  54
  55 #include <sys/thread2.h>
  56 #include <sys/mplock2.h>
  57 #include <sys/msgport2.h>
  58
  59 #include <net/if.h>
  60 #include <net/netisr.h>
  61 #include <net/pfil.h>
  62 #include <net/route.h>
  63
  64 #include <netinet/in.h>
  65 #include <netinet/in_systm.h>
  66 #include <netinet/ip.h>
  67 #include <netinet/in_pcb.h>
  68 #include <netinet/in_var.h>
  69 #include <netinet/ip_var.h>
  70
  71 #include <netproto/mpls/mpls_var.h>
  72
  73 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  74
  75 #ifdef IPSEC
  76 #include <netinet6/ipsec.h>
  77 #include <netproto/key/key.h>
  78 #ifdef IPSEC_DEBUG
  79 #include <netproto/key/key_debug.h>
  80 #else
  81 #define KEYDEBUG(lev,arg)
  82 #endif
  83 #endif /*IPSEC*/
  84
  85 #ifdef FAST_IPSEC
  86 #include <netproto/ipsec/ipsec.h>
  87 #include <netproto/ipsec/xform.h>
  88 #include <netproto/ipsec/key.h>
  89 #endif /*FAST_IPSEC*/
  90
  91 #include <net/ipfw/ip_fw.h>
  92 #include <net/dummynet/ip_dummynet.h>
  93
  94 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  95                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  96                                   (ntohl(a.s_addr)>>16)&0xFF,\
  97                                   (ntohl(a.s_addr)>>8)&0xFF,\
  98                                   (ntohl(a.s_addr))&0xFF, y);
  99
 100 u_short ip_id;
 101
 102 #ifdef MBUF_STRESS_TEST
 103 int mbuf_frag_size = 0;
 104 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 105         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 106 #endif
 107
 108 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 109 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 110 static void     ip_mloopback
 111         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 112 static int      ip_getmoptions
 113         (struct sockopt *, struct ip_moptions *);
 114 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 115 static int      ip_setmoptions
 116         (struct sockopt *, struct ip_moptions **);
 117
 118 int     ip_optcopy(struct ip *, struct ip *);
 119
 120 extern  int route_assert_owner_access;
 121
 122 extern  struct protosw inetsw[];
 123
 124 static int
 125 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
 126 {
 127         struct in_ifaddr_container *iac;
 128
 129         /*
 130          * We need to figure out if we have been forwarded to a local
 131          * socket.  If so, then we should somehow "loop back" to
 132          * ip_input(), and get directed to the PCB as if we had received
 133          * this packet.  This is because it may be difficult to identify
 134          * the packets you want to forward until they are being output
 135          * and have selected an interface (e.g. locally initiated
 136          * packets).  If we used the loopback inteface, we would not be
 137          * able to control what happens as the packet runs through
 138          * ip_input() as it is done through a ISR.
 139          */
 140         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 141                 /*
 142                  * If the addr to forward to is one of ours, we pretend
 143                  * to be the destination for this packet.
 144                  */
 145                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 146                         break;
 147         }
 148         if (iac != NULL) {
 149                 struct ip *ip;
 150
 151                 if (m->m_pkthdr.rcvif == NULL)
 152                         m->m_pkthdr.rcvif = ifunit("lo0");
 153                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 154                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 155                                                   CSUM_PSEUDO_HDR;
 156                         m->m_pkthdr.csum_data = 0xffff;
 157                 }
 158                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 159
 160                 /*
 161                  * Make sure that the IP header is in one mbuf,
 162                  * required by ip_input
 163                  */
 164                 if (m->m_len < hlen) {
 165                         m = m_pullup(m, hlen);
 166                         if (m == NULL) {
 167                                 /* The packet was freed; we are done */
 168                                 return 1;
 169                         }
 170                 }
 171                 ip = mtod(m, struct ip *);
 172
 173                 ip->ip_len = htons(ip->ip_len);
 174                 ip->ip_off = htons(ip->ip_off);
 175                 ip_input(m);
 176
 177                 return 1; /* The packet gets forwarded locally */
 178         }
 179         return 0;
 180 }
 181
 182 /*
 183  * IP output.  The packet in mbuf chain m contains a skeletal IP
 184  * header (with len, off, ttl, proto, tos, src, dst).
 185  * The mbuf chain containing the packet will be freed.
 186  * The mbuf opt, if present, will not be freed.
 187  */
 188 int
 189 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 190           int flags, struct ip_moptions *imo, struct inpcb *inp)
 191 {
 192         struct ip *ip;
 193         struct ifnet *ifp = NULL;       /* keep compiler happy */
 194         struct mbuf *m;
 195         int hlen = sizeof(struct ip);
 196         int len, error = 0;
 197         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 198         struct in_ifaddr *ia = NULL;
 199         int isbroadcast, sw_csum;
 200         struct in_addr pkt_dst;
 201         struct route iproute;
 202         struct m_tag *mtag;
 203 #ifdef IPSEC
 204         struct secpolicy *sp = NULL;
 205         struct socket *so = inp ? inp->inp_socket : NULL;
 206 #endif
 207 #ifdef FAST_IPSEC
 208         struct secpolicy *sp = NULL;
 209         struct tdb_ident *tdbi;
 210 #endif /* FAST_IPSEC */
 211         struct sockaddr_in *next_hop = NULL;
 212         int src_was_INADDR_ANY = 0;     /* as the name says... */
 213
 214         m = m0;
 215         M_ASSERTPKTHDR(m);
 216
 217         if (ro == NULL) {
 218                 ro = &iproute;
 219                 bzero(ro, sizeof *ro);
 220         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 221                 if (flags & IP_DEBUGROUTE) {
 222                         if (route_assert_owner_access) {
 223                                 panic("ip_output: "
 224                                       "rt rt_cpuid %d accessed on cpu %d\n",
 225                                       ro->ro_rt->rt_cpuid, mycpuid);
 226                         } else {
 227                                 kprintf("ip_output: "
 228                                         "rt rt_cpuid %d accessed on cpu %d\n",
 229                                         ro->ro_rt->rt_cpuid, mycpuid);
 230                                 print_backtrace(-1);
 231                         }
 232                 }
 233
 234                 /*
 235                  * XXX
 236                  * If the cached rtentry's owner CPU is not the current CPU,
 237                  * then don't touch the cached rtentry (remote free is too
 238                  * expensive in this context); just relocate the route.
 239                  */
 240                 ro = &iproute;
 241                 bzero(ro, sizeof *ro);
 242         }
 243
 244         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 245                 /* Next hop */
 246                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 247                 KKASSERT(mtag != NULL);
 248                 next_hop = m_tag_data(mtag);
 249         }
 250
 251         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 252                 struct dn_pkt *dn_pkt;
 253
 254                 /* Extract info from dummynet tag */
 255                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 256                 KKASSERT(mtag != NULL);
 257                 dn_pkt = m_tag_data(mtag);
 258
 259                 /*
 260                  * The packet was already tagged, so part of the
 261                  * processing was already done, and we need to go down.
 262                  * Get the calculated parameters from the tag.
 263                  */
 264                 ifp = dn_pkt->ifp;
 265
 266                 KKASSERT(ro == &iproute);
 267                 *ro = dn_pkt->ro; /* structure copy */
 268                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 269
 270                 dst = dn_pkt->dn_dst;
 271                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 272                         /* If 'dst' points into dummynet tag, adjust it */
 273                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 274                 }
 275
 276                 ip = mtod(m, struct ip *);
 277                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 278                 if (ro->ro_rt)
 279                         ia = ifatoia(ro->ro_rt->rt_ifa);
 280                 goto sendit;
 281         }
 282
 283         if (opt) {
 284                 len = 0;
 285                 m = ip_insertoptions(m, opt, &len);
 286                 if (len != 0)
 287                         hlen = len;
 288         }
 289         ip = mtod(m, struct ip *);
 290
 291         /*
 292          * Fill in IP header.
 293          */
 294         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 295                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 296                 ip->ip_off &= IP_DF;
 297                 ip->ip_id = ip_newid();
 298                 ipstat.ips_localout++;
 299         } else {
 300                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 301         }
 302
 303 reroute:
 304         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 305
 306         dst = (struct sockaddr_in *)&ro->ro_dst;
 307         /*
 308          * If there is a cached route,
 309          * check that it is to the same destination
 310          * and is still up.  If not, free it and try again.
 311          * The address family should also be checked in case of sharing the
 312          * cache with IPv6.
 313          */
 314         if (ro->ro_rt &&
 315             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 316              dst->sin_family != AF_INET ||
 317              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 318                 rtfree(ro->ro_rt);
 319                 ro->ro_rt = NULL;
 320         }
 321         if (ro->ro_rt == NULL) {
 322                 bzero(dst, sizeof *dst);
 323                 dst->sin_family = AF_INET;
 324                 dst->sin_len = sizeof *dst;
 325                 dst->sin_addr = pkt_dst;
 326         }
 327         /*
 328          * If routing to interface only,
 329          * short circuit routing lookup.
 330          */
 331         if (flags & IP_ROUTETOIF) {
 332                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 333                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 334                         ipstat.ips_noroute++;
 335                         error = ENETUNREACH;
 336                         goto bad;
 337                 }
 338                 ifp = ia->ia_ifp;
 339                 ip->ip_ttl = 1;
 340                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 341         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 342                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 343                 /*
 344                  * Bypass the normal routing lookup for multicast
 345                  * packets if the interface is specified.
 346                  */
 347                 ifp = imo->imo_multicast_ifp;
 348                 ia = IFP_TO_IA(ifp);
 349                 isbroadcast = 0;        /* fool gcc */
 350         } else {
 351                 /*
 352                  * If this is the case, we probably don't want to allocate
 353                  * a protocol-cloned route since we didn't get one from the
 354                  * ULP.  This lets TCP do its thing, while not burdening
 355                  * forwarding or ICMP with the overhead of cloning a route.
 356                  * Of course, we still want to do any cloning requested by
 357                  * the link layer, as this is probably required in all cases
 358                  * for correct operation (as it is for ARP).
 359                  */
 360                 if (ro->ro_rt == NULL)
 361                         rtalloc_ign(ro, RTF_PRCLONING);
 362                 if (ro->ro_rt == NULL) {
 363                         ipstat.ips_noroute++;
 364                         error = EHOSTUNREACH;
 365                         goto bad;
 366                 }
 367                 ia = ifatoia(ro->ro_rt->rt_ifa);
 368                 ifp = ro->ro_rt->rt_ifp;
 369                 ro->ro_rt->rt_use++;
 370                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 371                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 372                 if (ro->ro_rt->rt_flags & RTF_HOST)
 373                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 374                 else
 375                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 376         }
 377         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 378                 struct in_multi *inm;
 379
 380                 m->m_flags |= M_MCAST;
 381                 /*
 382                  * IP destination address is multicast.  Make sure "dst"
 383                  * still points to the address in "ro".  (It may have been
 384                  * changed to point to a gateway address, above.)
 385                  */
 386                 dst = (struct sockaddr_in *)&ro->ro_dst;
 387                 /*
 388                  * See if the caller provided any multicast options
 389                  */
 390                 if (imo != NULL) {
 391                         ip->ip_ttl = imo->imo_multicast_ttl;
 392                         if (imo->imo_multicast_vif != -1) {
 393                                 ip->ip_src.s_addr =
 394                                     ip_mcast_src ?
 395                                     ip_mcast_src(imo->imo_multicast_vif) :
 396                                     INADDR_ANY;
 397                         }
 398                 } else {
 399                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 400                 }
 401                 /*
 402                  * Confirm that the outgoing interface supports multicast.
 403                  */
 404                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 405                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 406                                 ipstat.ips_noroute++;
 407                                 error = ENETUNREACH;
 408                                 goto bad;
 409                         }
 410                 }
 411                 /*
 412                  * If source address not specified yet, use address
 413                  * of outgoing interface.
 414                  */
 415                 if (ip->ip_src.s_addr == INADDR_ANY) {
 416                         /* Interface may have no addresses. */
 417                         if (ia != NULL)
 418                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 419                 }
 420
 421                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 422                 if (inm != NULL &&
 423                     (imo == NULL || imo->imo_multicast_loop)) {
 424                         /*
 425                          * If we belong to the destination multicast group
 426                          * on the outgoing interface, and the caller did not
 427                          * forbid loopback, loop back a copy.
 428                          */
 429                         ip_mloopback(ifp, m, dst, hlen);
 430                 } else {
 431                         /*
 432                          * If we are acting as a multicast router, perform
 433                          * multicast forwarding as if the packet had just
 434                          * arrived on the interface to which we are about
 435                          * to send.  The multicast forwarding function
 436                          * recursively calls this function, using the
 437                          * IP_FORWARDING flag to prevent infinite recursion.
 438                          *
 439                          * Multicasts that are looped back by ip_mloopback(),
 440                          * above, will be forwarded by the ip_input() routine,
 441                          * if necessary.
 442                          */
 443                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 444                                 /*
 445                                  * If rsvp daemon is not running, do not
 446                                  * set ip_moptions. This ensures that the packet
 447                                  * is multicast and not just sent down one link
 448                                  * as prescribed by rsvpd.
 449                                  */
 450                                 if (!rsvp_on)
 451                                         imo = NULL;
 452                                 if (ip_mforward) {
 453                                         get_mplock();
 454                                         if (ip_mforward(ip, ifp, m, imo) != 0) {
 455                                                 m_freem(m);
 456                                                 rel_mplock();
 457                                                 goto done;
 458                                         }
 459                                         rel_mplock();
 460                                 }
 461                         }
 462                 }
 463
 464                 /*
 465                  * Multicasts with a time-to-live of zero may be looped-
 466                  * back, above, but must not be transmitted on a network.
 467                  * Also, multicasts addressed to the loopback interface
 468                  * are not sent -- the above call to ip_mloopback() will
 469                  * loop back a copy if this host actually belongs to the
 470                  * destination group on the loopback interface.
 471                  */
 472                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 473                         m_freem(m);
 474                         goto done;
 475                 }
 476
 477                 goto sendit;
 478         } else {
 479                 m->m_flags &= ~M_MCAST;
 480         }
 481
 482         /*
 483          * If the source address is not specified yet, use the address
 484          * of the outoing interface. In case, keep note we did that, so
 485          * if the the firewall changes the next-hop causing the output
 486          * interface to change, we can fix that.
 487          */
 488         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 489                 /* Interface may have no addresses. */
 490                 if (ia != NULL) {
 491                         ip->ip_src = IA_SIN(ia)->sin_addr;
 492                         src_was_INADDR_ANY = 1;
 493                 }
 494         }
 495
 496 #ifdef ALTQ
 497         /*
 498          * Disable packet drop hack.
 499          * Packetdrop should be done by queueing.
 500          */
 501 #else /* !ALTQ */
 502         /*
 503          * Verify that we have any chance at all of being able to queue
 504          *      the packet or packet fragments
 505          */
 506         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 507             ifp->if_snd.ifq_maxlen) {
 508                 error = ENOBUFS;
 509                 ipstat.ips_odropped++;
 510                 goto bad;
 511         }
 512 #endif /* !ALTQ */
 513
 514         /*
 515          * Look for broadcast address and
 516          * verify user is allowed to send
 517          * such a packet.
 518          */
 519         if (isbroadcast) {
 520                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 521                         error = EADDRNOTAVAIL;
 522                         goto bad;
 523                 }
 524                 if (!(flags & IP_ALLOWBROADCAST)) {
 525                         error = EACCES;
 526                         goto bad;
 527                 }
 528                 /* don't allow broadcast messages to be fragmented */
 529                 if (ip->ip_len > ifp->if_mtu) {
 530                         error = EMSGSIZE;
 531                         goto bad;
 532                 }
 533                 m->m_flags |= M_BCAST;
 534         } else {
 535                 m->m_flags &= ~M_BCAST;
 536         }
 537
 538 sendit:
 539 #ifdef IPSEC
 540         /* get SP for this packet */
 541         if (so == NULL)
 542                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 543         else
 544                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 545
 546         if (sp == NULL) {
 547                 ipsecstat.out_inval++;
 548                 goto bad;
 549         }
 550
 551         error = 0;
 552
 553         /* check policy */
 554         switch (sp->policy) {
 555         case IPSEC_POLICY_DISCARD:
 556                 /*
 557                  * This packet is just discarded.
 558                  */
 559                 ipsecstat.out_polvio++;
 560                 goto bad;
 561
 562         case IPSEC_POLICY_BYPASS:
 563         case IPSEC_POLICY_NONE:
 564         case IPSEC_POLICY_TCP:
 565                 /* no need to do IPsec. */
 566                 goto skip_ipsec;
 567
 568         case IPSEC_POLICY_IPSEC:
 569                 if (sp->req == NULL) {
 570                         /* acquire a policy */
 571                         error = key_spdacquire(sp);
 572                         goto bad;
 573                 }
 574                 break;
 575
 576         case IPSEC_POLICY_ENTRUST:
 577         default:
 578                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 579         }
 580     {
 581         struct ipsec_output_state state;
 582         bzero(&state, sizeof state);
 583         state.m = m;
 584         if (flags & IP_ROUTETOIF) {
 585                 state.ro = &iproute;
 586                 bzero(&iproute, sizeof iproute);
 587         } else
 588                 state.ro = ro;
 589         state.dst = (struct sockaddr *)dst;
 590
 591         ip->ip_sum = 0;
 592
 593         /*
 594          * XXX
 595          * delayed checksums are not currently compatible with IPsec
 596          */
 597         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 598                 in_delayed_cksum(m);
 599                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 600         }
 601
 602         ip->ip_len = htons(ip->ip_len);
 603         ip->ip_off = htons(ip->ip_off);
 604
 605         error = ipsec4_output(&state, sp, flags);
 606
 607         m = state.m;
 608         if (flags & IP_ROUTETOIF) {
 609                 /*
 610                  * if we have tunnel mode SA, we may need to ignore
 611                  * IP_ROUTETOIF.
 612                  */
 613                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 614                         flags &= ~IP_ROUTETOIF;
 615                         ro = state.ro;
 616                 }
 617         } else
 618                 ro = state.ro;
 619         dst = (struct sockaddr_in *)state.dst;
 620         if (error) {
 621                 /* mbuf is already reclaimed in ipsec4_output. */
 622                 m0 = NULL;
 623                 switch (error) {
 624                 case EHOSTUNREACH:
 625                 case ENETUNREACH:
 626                 case EMSGSIZE:
 627                 case ENOBUFS:
 628                 case ENOMEM:
 629                         break;
 630                 default:
 631                         kprintf("ip4_output (ipsec): error code %d\n", error);
 632                         /*fall through*/
 633                 case ENOENT:
 634                         /* don't show these error codes to the user */
 635                         error = 0;
 636                         break;
 637                 }
 638                 goto bad;
 639         }
 640     }
 641
 642         /* be sure to update variables that are affected by ipsec4_output() */
 643         ip = mtod(m, struct ip *);
 644 #ifdef _IP_VHL
 645         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 646 #else
 647         hlen = ip->ip_hl << 2;
 648 #endif
 649         if (ro->ro_rt == NULL) {
 650                 if (!(flags & IP_ROUTETOIF)) {
 651                         kprintf("ip_output: "
 652                                 "can't update route after IPsec processing\n");
 653                         error = EHOSTUNREACH;   /*XXX*/
 654                         goto bad;
 655                 }
 656         } else {
 657                 ia = ifatoia(ro->ro_rt->rt_ifa);
 658                 ifp = ro->ro_rt->rt_ifp;
 659         }
 660
 661         /* make it flipped, again. */
 662         ip->ip_len = ntohs(ip->ip_len);
 663         ip->ip_off = ntohs(ip->ip_off);
 664 skip_ipsec:
 665 #endif /*IPSEC*/
 666 #ifdef FAST_IPSEC
 667         /*
 668          * Check the security policy (SP) for the packet and, if
 669          * required, do IPsec-related processing.  There are two
 670          * cases here; the first time a packet is sent through
 671          * it will be untagged and handled by ipsec4_checkpolicy.
 672          * If the packet is resubmitted to ip_output (e.g. after
 673          * AH, ESP, etc. processing), there will be a tag to bypass
 674          * the lookup and related policy checking.
 675          */
 676         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 677         crit_enter();
 678         if (mtag != NULL) {
 679                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 680                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 681                 if (sp == NULL)
 682                         error = -EINVAL;        /* force silent drop */
 683                 m_tag_delete(m, mtag);
 684         } else {
 685                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 686                                         &error, inp);
 687         }
 688         /*
 689          * There are four return cases:
 690          *    sp != NULL                    apply IPsec policy
 691          *    sp == NULL, error == 0        no IPsec handling needed
 692          *    sp == NULL, error == -EINVAL  discard packet w/o error
 693          *    sp == NULL, error != 0        discard packet, report error
 694          */
 695         if (sp != NULL) {
 696                 /* Loop detection, check if ipsec processing already done */
 697                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 698                 for (mtag = m_tag_first(m); mtag != NULL;
 699                      mtag = m_tag_next(m, mtag)) {
 700                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 701                                 continue;
 702                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 703                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 704                                 continue;
 705                         /*
 706                          * Check if policy has an SA associated with it.
 707                          * This can happen when an SP has yet to acquire
 708                          * an SA; e.g. on first reference.  If it occurs,
 709                          * then we let ipsec4_process_packet do its thing.
 710                          */
 711                         if (sp->req->sav == NULL)
 712                                 break;
 713                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 714                         if (tdbi->spi == sp->req->sav->spi &&
 715                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 716                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 717                                  sizeof(union sockaddr_union)) == 0) {
 718                                 /*
 719                                  * No IPsec processing is needed, free
 720                                  * reference to SP.
 721                                  *
 722                                  * NB: null pointer to avoid free at
 723                                  *     done: below.
 724                                  */
 725                                 KEY_FREESP(&sp), sp = NULL;
 726                                 crit_exit();
 727                                 goto spd_done;
 728                         }
 729                 }
 730
 731                 /*
 732                  * Do delayed checksums now because we send before
 733                  * this is done in the normal processing path.
 734                  */
 735                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 736                         in_delayed_cksum(m);
 737                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 738                 }
 739
 740                 ip->ip_len = htons(ip->ip_len);
 741                 ip->ip_off = htons(ip->ip_off);
 742
 743                 /* NB: callee frees mbuf */
 744                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 745                 /*
 746                  * Preserve KAME behaviour: ENOENT can be returned
 747                  * when an SA acquire is in progress.  Don't propagate
 748                  * this to user-level; it confuses applications.
 749                  *
 750                  * XXX this will go away when the SADB is redone.
 751                  */
 752                 if (error == ENOENT)
 753                         error = 0;
 754                 crit_exit();
 755                 goto done;
 756         } else {
 757                 crit_exit();
 758
 759                 if (error != 0) {
 760                         /*
 761                          * Hack: -EINVAL is used to signal that a packet
 762                          * should be silently discarded.  This is typically
 763                          * because we asked key management for an SA and
 764                          * it was delayed (e.g. kicked up to IKE).
 765                          */
 766                         if (error == -EINVAL)
 767                                 error = 0;
 768                         goto bad;
 769                 } else {
 770                         /* No IPsec processing for this packet. */
 771                 }
 772 #ifdef notyet
 773                 /*
 774                  * If deferred crypto processing is needed, check that
 775                  * the interface supports it.
 776                  */
 777                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 778                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 779                         /* notify IPsec to do its own crypto */
 780                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 781                         error = EHOSTUNREACH;
 782                         goto bad;
 783                 }
 784 #endif
 785         }
 786 spd_done:
 787 #endif /* FAST_IPSEC */
 788
 789         /* We are already being fwd'd from a firewall. */
 790         if (next_hop != NULL)
 791                 goto pass;
 792
 793         /* No pfil hooks */
 794         if (!pfil_has_hooks(&inet_pfil_hook)) {
 795                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 796                         /*
 797                          * Strip dummynet tags from stranded packets
 798                          */
 799                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 800                         KKASSERT(mtag != NULL);
 801                         m_tag_delete(m, mtag);
 802                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 803                 }
 804                 goto pass;
 805         }
 806
 807         /*
 808          * IpHack's section.
 809          * - Xlate: translate packet's addr/port (NAT).
 810          * - Firewall: deny/allow/etc.
 811          * - Wrap: fake packet's addr/port <unimpl.>
 812          * - Encapsulate: put it in another IP and send out. <unimp.>
 813          */
 814
 815         /*
 816          * Run through list of hooks for output packets.
 817          */
 818         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 819         if (error != 0 || m == NULL)
 820                 goto done;
 821         ip = mtod(m, struct ip *);
 822
 823         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 824                 /*
 825                  * Check dst to make sure it is directly reachable on the
 826                  * interface we previously thought it was.
 827                  * If it isn't (which may be likely in some situations) we have
 828                  * to re-route it (ie, find a route for the next-hop and the
 829                  * associated interface) and set them here. This is nested
 830                  * forwarding which in most cases is undesirable, except where
 831                  * such control is nigh impossible. So we do it here.
 832                  * And I'm babbling.
 833                  */
 834                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 835                 KKASSERT(mtag != NULL);
 836                 next_hop = m_tag_data(mtag);
 837
 838                 /*
 839                  * Try local forwarding first
 840                  */
 841                 if (ip_localforward(m, next_hop, hlen))
 842                         goto done;
 843
 844                 /*
 845                  * Relocate the route based on next_hop.
 846                  * If the current route is inp's cache, keep it untouched.
 847                  */
 848                 if (ro == &iproute && ro->ro_rt != NULL) {
 849                         RTFREE(ro->ro_rt);
 850                         ro->ro_rt = NULL;
 851                 }
 852                 ro = &iproute;
 853                 bzero(ro, sizeof *ro);
 854
 855                 /*
 856                  * Forwarding to broadcast address is not allowed.
 857                  * XXX Should we follow IP_ROUTETOIF?
 858                  */
 859                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 860
 861                 /* We are doing forwarding now */
 862                 flags |= IP_FORWARDING;
 863
 864                 goto reroute;
 865         }
 866
 867         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 868                 struct dn_pkt *dn_pkt;
 869
 870                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 871                 KKASSERT(mtag != NULL);
 872                 dn_pkt = m_tag_data(mtag);
 873
 874                 /*
 875                  * Under certain cases it is not possible to recalculate
 876                  * 'ro' and 'dst', let alone 'flags', so just save them in
 877                  * dummynet tag and avoid the possible wrong reculcalation
 878                  * when we come back to ip_output() again.
 879                  *
 880                  * All other parameters have been already used and so they
 881                  * are not needed anymore.
 882                  * XXX if the ifp is deleted while a pkt is in dummynet,
 883                  * we are in trouble! (TODO use ifnet_detach_event)
 884                  *
 885                  * We need to copy *ro because for ICMP pkts (and maybe
 886                  * others) the caller passed a pointer into the stack;
 887                  * dst might also be a pointer into *ro so it needs to
 888                  * be updated.
 889                  */
 890                 dn_pkt->ro = *ro;
 891                 if (ro->ro_rt)
 892                         ro->ro_rt->rt_refcnt++;
 893                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 894                         /* 'dst' points into 'ro' */
 895                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 896                 }
 897                 dn_pkt->dn_dst = dst;
 898                 dn_pkt->flags = flags;
 899
 900                 ip_dn_queue(m);
 901                 goto done;
 902         }
 903 pass:
 904         /* 127/8 must not appear on wire - RFC1122. */
 905         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 906             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 907                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 908                         ipstat.ips_badaddr++;
 909                         error = EADDRNOTAVAIL;
 910                         goto bad;
 911                 }
 912         }
 913
 914         if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
 915                 m->m_pkthdr.csum_flags |= CSUM_IP;
 916                 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 917                 if (sw_csum & CSUM_DELAY_DATA) {
 918                         in_delayed_cksum(m);
 919                         sw_csum &= ~CSUM_DELAY_DATA;
 920                 }
 921                 m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 922         } else {
 923                 sw_csum = 0;
 924         }
 925         m->m_pkthdr.csum_iphlen = hlen;
 926
 927         /*
 928          * If small enough for interface, or the interface will take
 929          * care of the fragmentation or segmentation for us, can just
 930          * send directly.
 931          */
 932         if (ip->ip_len <= ifp->if_mtu ||
 933             ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
 934             (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 935                 ip->ip_len = htons(ip->ip_len);
 936                 ip->ip_off = htons(ip->ip_off);
 937                 ip->ip_sum = 0;
 938                 if (sw_csum & CSUM_DELAY_IP) {
 939                         if (ip->ip_vhl == IP_VHL_BORING)
 940                                 ip->ip_sum = in_cksum_hdr(ip);
 941                         else
 942                                 ip->ip_sum = in_cksum(m, hlen);
 943                 }
 944
 945                 /* Record statistics for this interface address. */
 946                 if (!(flags & IP_FORWARDING) && ia) {
 947                         ia->ia_ifa.if_opackets++;
 948                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 949                 }
 950
 951 #ifdef IPSEC
 952                 /* clean ipsec history once it goes out of the node */
 953                 ipsec_delaux(m);
 954 #endif
 955
 956 #ifdef MBUF_STRESS_TEST
 957                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 958                         struct mbuf *m1, *m2;
 959                         int length, tmp;
 960
 961                         tmp = length = m->m_pkthdr.len;
 962
 963                         while ((length -= mbuf_frag_size) >= 1) {
 964                                 m1 = m_split(m, length, MB_DONTWAIT);
 965                                 if (m1 == NULL)
 966                                         break;
 967                                 m2 = m;
 968                                 while (m2->m_next != NULL)
 969                                         m2 = m2->m_next;
 970                                 m2->m_next = m1;
 971                         }
 972                         m->m_pkthdr.len = tmp;
 973                 }
 974 #endif
 975
 976 #ifdef MPLS
 977                 if (!mpls_output_process(m, ro->ro_rt))
 978                         goto done;
 979 #endif
 980                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 981                                        ro->ro_rt);
 982                 goto done;
 983         }
 984
 985         if (ip->ip_off & IP_DF) {
 986                 error = EMSGSIZE;
 987                 /*
 988                  * This case can happen if the user changed the MTU
 989                  * of an interface after enabling IP on it.  Because
 990                  * most netifs don't keep track of routes pointing to
 991                  * them, there is no way for one to update all its
 992                  * routes when the MTU is changed.
 993                  */
 994                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 995                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 996                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 997                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 998                 }
 999                 ipstat.ips_cantfrag++;
1000                 goto bad;
1001         }
1002
1003         /*
1004          * Too large for interface; fragment if possible. If successful,
1005          * on return, m will point to a list of packets to be sent.
1006          */
1007         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1008         if (error)
1009                 goto bad;
1010         for (; m; m = m0) {
1011                 m0 = m->m_nextpkt;
1012                 m->m_nextpkt = NULL;
1013 #ifdef IPSEC
1014                 /* clean ipsec history once it goes out of the node */
1015                 ipsec_delaux(m);
1016 #endif
1017                 if (error == 0) {
1018                         /* Record statistics for this interface address. */
1019                         if (ia != NULL) {
1020                                 ia->ia_ifa.if_opackets++;
1021                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1022                         }
1023 #ifdef MPLS
1024                         if (!mpls_output_process(m, ro->ro_rt))
1025                                 continue;
1026 #endif
1027                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1028                                                ro->ro_rt);
1029                 } else {
1030                         m_freem(m);
1031                 }
1032         }
1033
1034         if (error == 0)
1035                 ipstat.ips_fragmented++;
1036
1037 done:
1038         if (ro == &iproute && ro->ro_rt != NULL) {
1039                 RTFREE(ro->ro_rt);
1040                 ro->ro_rt = NULL;
1041         }
1042 #ifdef IPSEC
1043         if (sp != NULL) {
1044                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1045                         kprintf("DP ip_output call free SP:%p\n", sp));
1046                 key_freesp(sp);
1047         }
1048 #endif
1049 #ifdef FAST_IPSEC
1050         if (sp != NULL)
1051                 KEY_FREESP(&sp);
1052 #endif
1053         return (error);
1054 bad:
1055         m_freem(m);
1056         goto done;
1057 }
1058
1059 /*
1060  * Create a chain of fragments which fit the given mtu. m_frag points to the
1061  * mbuf to be fragmented; on return it points to the chain with the fragments.
1062  * Return 0 if no error. If error, m_frag may contain a partially built
1063  * chain of fragments that should be freed by the caller.
1064  *
1065  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1066  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1067  */
1068 int
1069 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1070             u_long if_hwassist_flags, int sw_csum)
1071 {
1072         int error = 0;
1073         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1074         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1075         int off;
1076         struct mbuf *m0 = *m_frag;      /* the original packet          */
1077         int firstlen;
1078         struct mbuf **mnext;
1079         int nfrags;
1080
1081         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1082                 ipstat.ips_cantfrag++;
1083                 return EMSGSIZE;
1084         }
1085
1086         /*
1087          * Must be able to put at least 8 bytes per fragment.
1088          */
1089         if (len < 8)
1090                 return EMSGSIZE;
1091
1092         /*
1093          * If the interface will not calculate checksums on
1094          * fragmented packets, then do it here.
1095          */
1096         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1097             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1098                 in_delayed_cksum(m0);
1099                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1100         }
1101
1102         if (len > PAGE_SIZE) {
1103                 /*
1104                  * Fragment large datagrams such that each segment
1105                  * contains a multiple of PAGE_SIZE amount of data,
1106                  * plus headers. This enables a receiver to perform
1107                  * page-flipping zero-copy optimizations.
1108                  *
1109                  * XXX When does this help given that sender and receiver
1110                  * could have different page sizes, and also mtu could
1111                  * be less than the receiver's page size ?
1112                  */
1113                 int newlen;
1114                 struct mbuf *m;
1115
1116                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1117                         off += m->m_len;
1118
1119                 /*
1120                  * firstlen (off - hlen) must be aligned on an
1121                  * 8-byte boundary
1122                  */
1123                 if (off < hlen)
1124                         goto smart_frag_failure;
1125                 off = ((off - hlen) & ~7) + hlen;
1126                 newlen = (~PAGE_MASK) & mtu;
1127                 if ((newlen + sizeof(struct ip)) > mtu) {
1128                         /* we failed, go back the default */
1129 smart_frag_failure:
1130                         newlen = len;
1131                         off = hlen + len;
1132                 }
1133                 len = newlen;
1134
1135         } else {
1136                 off = hlen + len;
1137         }
1138
1139         firstlen = off - hlen;
1140         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1141
1142         /*
1143          * Loop through length of segment after first fragment,
1144          * make new header and copy data of each part and link onto chain.
1145          * Here, m0 is the original packet, m is the fragment being created.
1146          * The fragments are linked off the m_nextpkt of the original
1147          * packet, which after processing serves as the first fragment.
1148          */
1149         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1150                 struct ip *mhip;        /* ip header on the fragment */
1151                 struct mbuf *m;
1152                 int mhlen = sizeof(struct ip);
1153
1154                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1155                 if (m == NULL) {
1156                         error = ENOBUFS;
1157                         ipstat.ips_odropped++;
1158                         goto done;
1159                 }
1160                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1161                 /*
1162                  * In the first mbuf, leave room for the link header, then
1163                  * copy the original IP header including options. The payload
1164                  * goes into an additional mbuf chain returned by m_copy().
1165                  */
1166                 m->m_data += max_linkhdr;
1167                 mhip = mtod(m, struct ip *);
1168                 *mhip = *ip;
1169                 if (hlen > sizeof(struct ip)) {
1170                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1171                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1172                 }
1173                 m->m_len = mhlen;
1174                 /* XXX do we need to add ip->ip_off below ? */
1175                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1176                 if (off + len >= ip->ip_len) {  /* last fragment */
1177                         len = ip->ip_len - off;
1178                         m->m_flags |= M_LASTFRAG;
1179                 } else
1180                         mhip->ip_off |= IP_MF;
1181                 mhip->ip_len = htons((u_short)(len + mhlen));
1182                 m->m_next = m_copy(m0, off, len);
1183                 if (m->m_next == NULL) {                /* copy failed */
1184                         m_free(m);
1185                         error = ENOBUFS;        /* ??? */
1186                         ipstat.ips_odropped++;
1187                         goto done;
1188                 }
1189                 m->m_pkthdr.len = mhlen + len;
1190                 m->m_pkthdr.rcvif = NULL;
1191                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1192                 m->m_pkthdr.csum_iphlen = mhlen;
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 void
1349 ip_ctloutput(netmsg_t msg)
1350 {
1351         struct socket *so = msg->base.nm_so;
1352         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1353         struct  inpcb *inp = so->so_pcb;
1354         int     error, optval;
1355
1356         error = optval = 0;
1357         if (sopt->sopt_level != IPPROTO_IP) {
1358                 error = EINVAL;
1359                 goto done;
1360         }
1361
1362         switch (sopt->sopt_dir) {
1363         case SOPT_SET:
1364                 switch (sopt->sopt_name) {
1365                 case IP_OPTIONS:
1366 #ifdef notyet
1367                 case IP_RETOPTS:
1368 #endif
1369                 {
1370                         struct mbuf *m;
1371                         if (sopt->sopt_valsize > MLEN) {
1372                                 error = EMSGSIZE;
1373                                 break;
1374                         }
1375                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1376                         if (m == NULL) {
1377                                 error = ENOBUFS;
1378                                 break;
1379                         }
1380                         m->m_len = sopt->sopt_valsize;
1381                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1382                                               m->m_len);
1383                         error = ip_pcbopts(sopt->sopt_name,
1384                                            &inp->inp_options, m);
1385                         goto done;
1386                 }
1387
1388                 case IP_TOS:
1389                 case IP_TTL:
1390                 case IP_MINTTL:
1391                 case IP_RECVOPTS:
1392                 case IP_RECVRETOPTS:
1393                 case IP_RECVDSTADDR:
1394                 case IP_RECVIF:
1395                 case IP_RECVTTL:
1396                 case IP_FAITH:
1397                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1398                                              sizeof optval);
1399                         if (error)
1400                                 break;
1401                         switch (sopt->sopt_name) {
1402                         case IP_TOS:
1403                                 inp->inp_ip_tos = optval;
1404                                 break;
1405
1406                         case IP_TTL:
1407                                 inp->inp_ip_ttl = optval;
1408                                 break;
1409                         case IP_MINTTL:
1410                                 if (optval >= 0 && optval <= MAXTTL)
1411                                         inp->inp_ip_minttl = optval;
1412                                 else
1413                                         error = EINVAL;
1414                                 break;
1415 #define OPTSET(bit) \
1416         if (optval) \
1417                 inp->inp_flags |= bit; \
1418         else \
1419                 inp->inp_flags &= ~bit;
1420
1421                         case IP_RECVOPTS:
1422                                 OPTSET(INP_RECVOPTS);
1423                                 break;
1424
1425                         case IP_RECVRETOPTS:
1426                                 OPTSET(INP_RECVRETOPTS);
1427                                 break;
1428
1429                         case IP_RECVDSTADDR:
1430                                 OPTSET(INP_RECVDSTADDR);
1431                                 break;
1432
1433                         case IP_RECVIF:
1434                                 OPTSET(INP_RECVIF);
1435                                 break;
1436
1437                         case IP_RECVTTL:
1438                                 OPTSET(INP_RECVTTL);
1439                                 break;
1440
1441                         case IP_FAITH:
1442                                 OPTSET(INP_FAITH);
1443                                 break;
1444                         }
1445                         break;
1446 #undef OPTSET
1447
1448                 case IP_MULTICAST_IF:
1449                 case IP_MULTICAST_VIF:
1450                 case IP_MULTICAST_TTL:
1451                 case IP_MULTICAST_LOOP:
1452                 case IP_ADD_MEMBERSHIP:
1453                 case IP_DROP_MEMBERSHIP:
1454                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1455                         break;
1456
1457                 case IP_PORTRANGE:
1458                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1459                                             sizeof optval);
1460                         if (error)
1461                                 break;
1462
1463                         switch (optval) {
1464                         case IP_PORTRANGE_DEFAULT:
1465                                 inp->inp_flags &= ~(INP_LOWPORT);
1466                                 inp->inp_flags &= ~(INP_HIGHPORT);
1467                                 break;
1468
1469                         case IP_PORTRANGE_HIGH:
1470                                 inp->inp_flags &= ~(INP_LOWPORT);
1471                                 inp->inp_flags |= INP_HIGHPORT;
1472                                 break;
1473
1474                         case IP_PORTRANGE_LOW:
1475                                 inp->inp_flags &= ~(INP_HIGHPORT);
1476                                 inp->inp_flags |= INP_LOWPORT;
1477                                 break;
1478
1479                         default:
1480                                 error = EINVAL;
1481                                 break;
1482                         }
1483                         break;
1484
1485 #if defined(IPSEC) || defined(FAST_IPSEC)
1486                 case IP_IPSEC_POLICY:
1487                 {
1488                         caddr_t req;
1489                         size_t len = 0;
1490                         int priv;
1491                         struct mbuf *m;
1492                         int optname;
1493
1494                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1495                                 break;
1496                         soopt_to_mbuf(sopt, m);
1497                         priv = (sopt->sopt_td != NULL &&
1498                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1499                         req = mtod(m, caddr_t);
1500                         len = m->m_len;
1501                         optname = sopt->sopt_name;
1502                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1503                         m_freem(m);
1504                         break;
1505                 }
1506 #endif /*IPSEC*/
1507
1508                 default:
1509                         error = ENOPROTOOPT;
1510                         break;
1511                 }
1512                 break;
1513
1514         case SOPT_GET:
1515                 switch (sopt->sopt_name) {
1516                 case IP_OPTIONS:
1517                 case IP_RETOPTS:
1518                         if (inp->inp_options)
1519                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1520                                                            char *),
1521                                                 inp->inp_options->m_len);
1522                         else
1523                                 sopt->sopt_valsize = 0;
1524                         break;
1525
1526                 case IP_TOS:
1527                 case IP_TTL:
1528                 case IP_MINTTL:
1529                 case IP_RECVOPTS:
1530                 case IP_RECVRETOPTS:
1531                 case IP_RECVDSTADDR:
1532                 case IP_RECVTTL:
1533                 case IP_RECVIF:
1534                 case IP_PORTRANGE:
1535                 case IP_FAITH:
1536                         switch (sopt->sopt_name) {
1537
1538                         case IP_TOS:
1539                                 optval = inp->inp_ip_tos;
1540                                 break;
1541
1542                         case IP_TTL:
1543                                 optval = inp->inp_ip_ttl;
1544                                 break;
1545                         case IP_MINTTL:
1546                                 optval = inp->inp_ip_minttl;
1547                                 break;
1548
1549 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1550
1551                         case IP_RECVOPTS:
1552                                 optval = OPTBIT(INP_RECVOPTS);
1553                                 break;
1554
1555                         case IP_RECVRETOPTS:
1556                                 optval = OPTBIT(INP_RECVRETOPTS);
1557                                 break;
1558
1559                         case IP_RECVDSTADDR:
1560                                 optval = OPTBIT(INP_RECVDSTADDR);
1561                                 break;
1562
1563                         case IP_RECVTTL:
1564                                 optval = OPTBIT(INP_RECVTTL);
1565                                 break;
1566
1567                         case IP_RECVIF:
1568                                 optval = OPTBIT(INP_RECVIF);
1569                                 break;
1570
1571                         case IP_PORTRANGE:
1572                                 if (inp->inp_flags & INP_HIGHPORT)
1573                                         optval = IP_PORTRANGE_HIGH;
1574                                 else if (inp->inp_flags & INP_LOWPORT)
1575                                         optval = IP_PORTRANGE_LOW;
1576                                 else
1577                                         optval = 0;
1578                                 break;
1579
1580                         case IP_FAITH:
1581                                 optval = OPTBIT(INP_FAITH);
1582                                 break;
1583                         }
1584                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1585                         break;
1586
1587                 case IP_MULTICAST_IF:
1588                 case IP_MULTICAST_VIF:
1589                 case IP_MULTICAST_TTL:
1590                 case IP_MULTICAST_LOOP:
1591                 case IP_ADD_MEMBERSHIP:
1592                 case IP_DROP_MEMBERSHIP:
1593                         error = ip_getmoptions(sopt, inp->inp_moptions);
1594                         break;
1595
1596 #if defined(IPSEC) || defined(FAST_IPSEC)
1597                 case IP_IPSEC_POLICY:
1598                 {
1599                         struct mbuf *m = NULL;
1600                         caddr_t req = NULL;
1601                         size_t len = 0;
1602
1603                         if (m != NULL) {
1604                                 req = mtod(m, caddr_t);
1605                                 len = m->m_len;
1606                         }
1607                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1608                         if (error == 0)
1609                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1610                         if (error == 0)
1611                                 m_freem(m);
1612                         break;
1613                 }
1614 #endif /*IPSEC*/
1615
1616                 default:
1617                         error = ENOPROTOOPT;
1618                         break;
1619                 }
1620                 break;
1621         }
1622 done:
1623         lwkt_replymsg(&msg->lmsg, error);
1624 }
1625
1626 /*
1627  * Set up IP options in pcb for insertion in output packets.
1628  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1629  * with destination address if source routed.
1630  */
1631 static int
1632 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1633 {
1634         int cnt, optlen;
1635         u_char *cp;
1636         u_char opt;
1637
1638         /* turn off any old options */
1639         if (*pcbopt)
1640                 m_free(*pcbopt);
1641         *pcbopt = NULL;
1642         if (m == NULL || m->m_len == 0) {
1643                 /*
1644                  * Only turning off any previous options.
1645                  */
1646                 if (m != NULL)
1647                         m_free(m);
1648                 return (0);
1649         }
1650
1651         if (m->m_len % sizeof(int32_t))
1652                 goto bad;
1653         /*
1654          * IP first-hop destination address will be stored before
1655          * actual options; move other options back
1656          * and clear it when none present.
1657          */
1658         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1659                 goto bad;
1660         cnt = m->m_len;
1661         m->m_len += sizeof(struct in_addr);
1662         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1663         ovbcopy(mtod(m, caddr_t), cp, cnt);
1664         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1665
1666         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1667                 opt = cp[IPOPT_OPTVAL];
1668                 if (opt == IPOPT_EOL)
1669                         break;
1670                 if (opt == IPOPT_NOP)
1671                         optlen = 1;
1672                 else {
1673                         if (cnt < IPOPT_OLEN + sizeof *cp)
1674                                 goto bad;
1675                         optlen = cp[IPOPT_OLEN];
1676                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1677                                 goto bad;
1678                 }
1679                 switch (opt) {
1680
1681                 default:
1682                         break;
1683
1684                 case IPOPT_LSRR:
1685                 case IPOPT_SSRR:
1686                         /*
1687                          * user process specifies route as:
1688                          *      ->A->B->C->D
1689                          * D must be our final destination (but we can't
1690                          * check that since we may not have connected yet).
1691                          * A is first hop destination, which doesn't appear in
1692                          * actual IP option, but is stored before the options.
1693                          */
1694                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1695                                 goto bad;
1696                         m->m_len -= sizeof(struct in_addr);
1697                         cnt -= sizeof(struct in_addr);
1698                         optlen -= sizeof(struct in_addr);
1699                         cp[IPOPT_OLEN] = optlen;
1700                         /*
1701                          * Move first hop before start of options.
1702                          */
1703                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1704                               sizeof(struct in_addr));
1705                         /*
1706                          * Then copy rest of options back
1707                          * to close up the deleted entry.
1708                          */
1709                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1710                                 &cp[IPOPT_OFFSET+1],
1711                                 cnt - (IPOPT_MINOFF - 1));
1712                         break;
1713                 }
1714         }
1715         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1716                 goto bad;
1717         *pcbopt = m;
1718         return (0);
1719
1720 bad:
1721         m_free(m);
1722         return (EINVAL);
1723 }
1724
1725 /*
1726  * XXX
1727  * The whole multicast option thing needs to be re-thought.
1728  * Several of these options are equally applicable to non-multicast
1729  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1730  * standard option (IP_TTL).
1731  */
1732
1733 /*
1734  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1735  */
1736 static struct ifnet *
1737 ip_multicast_if(struct in_addr *a, int *ifindexp)
1738 {
1739         int ifindex;
1740         struct ifnet *ifp;
1741
1742         if (ifindexp)
1743                 *ifindexp = 0;
1744         if (ntohl(a->s_addr) >> 24 == 0) {
1745                 ifindex = ntohl(a->s_addr) & 0xffffff;
1746                 if (ifindex < 0 || if_index < ifindex)
1747                         return NULL;
1748                 ifp = ifindex2ifnet[ifindex];
1749                 if (ifindexp)
1750                         *ifindexp = ifindex;
1751         } else {
1752                 ifp = INADDR_TO_IFP(a);
1753         }
1754         return ifp;
1755 }
1756
1757 /*
1758  * Set the IP multicast options in response to user setsockopt().
1759  */
1760 static int
1761 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1762 {
1763         int error = 0;
1764         int i;
1765         struct in_addr addr;
1766         struct ip_mreq mreq;
1767         struct ifnet *ifp;
1768         struct ip_moptions *imo = *imop;
1769         int ifindex;
1770
1771         if (imo == NULL) {
1772                 /*
1773                  * No multicast option buffer attached to the pcb;
1774                  * allocate one and initialize to default values.
1775                  */
1776                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1777
1778                 *imop = imo;
1779                 imo->imo_multicast_ifp = NULL;
1780                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1781                 imo->imo_multicast_vif = -1;
1782                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1783                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1784                 imo->imo_num_memberships = 0;
1785         }
1786         switch (sopt->sopt_name) {
1787         /* store an index number for the vif you wanna use in the send */
1788         case IP_MULTICAST_VIF:
1789                 if (legal_vif_num == 0) {
1790                         error = EOPNOTSUPP;
1791                         break;
1792                 }
1793                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1794                 if (error)
1795                         break;
1796                 if (!legal_vif_num(i) && (i != -1)) {
1797                         error = EINVAL;
1798                         break;
1799                 }
1800                 imo->imo_multicast_vif = i;
1801                 break;
1802
1803         case IP_MULTICAST_IF:
1804                 /*
1805                  * Select the interface for outgoing multicast packets.
1806                  */
1807                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1808                 if (error)
1809                         break;
1810
1811                 /*
1812                  * INADDR_ANY is used to remove a previous selection.
1813                  * When no interface is selected, a default one is
1814                  * chosen every time a multicast packet is sent.
1815                  */
1816                 if (addr.s_addr == INADDR_ANY) {
1817                         imo->imo_multicast_ifp = NULL;
1818                         break;
1819                 }
1820                 /*
1821                  * The selected interface is identified by its local
1822                  * IP address.  Find the interface and confirm that
1823                  * it supports multicasting.
1824                  */
1825                 crit_enter();
1826                 ifp = ip_multicast_if(&addr, &ifindex);
1827                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1828                         crit_exit();
1829                         error = EADDRNOTAVAIL;
1830                         break;
1831                 }
1832                 imo->imo_multicast_ifp = ifp;
1833                 if (ifindex)
1834                         imo->imo_multicast_addr = addr;
1835                 else
1836                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1837                 crit_exit();
1838                 break;
1839
1840         case IP_MULTICAST_TTL:
1841                 /*
1842                  * Set the IP time-to-live for outgoing multicast packets.
1843                  * The original multicast API required a char argument,
1844                  * which is inconsistent with the rest of the socket API.
1845                  * We allow either a char or an int.
1846                  */
1847                 if (sopt->sopt_valsize == 1) {
1848                         u_char ttl;
1849                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1850                         if (error)
1851                                 break;
1852                         imo->imo_multicast_ttl = ttl;
1853                 } else {
1854                         u_int ttl;
1855                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1856                         if (error)
1857                                 break;
1858                         if (ttl > 255)
1859                                 error = EINVAL;
1860                         else
1861                                 imo->imo_multicast_ttl = ttl;
1862                 }
1863                 break;
1864
1865         case IP_MULTICAST_LOOP:
1866                 /*
1867                  * Set the loopback flag for outgoing multicast packets.
1868                  * Must be zero or one.  The original multicast API required a
1869                  * char argument, which is inconsistent with the rest
1870                  * of the socket API.  We allow either a char or an int.
1871                  */
1872                 if (sopt->sopt_valsize == 1) {
1873                         u_char loop;
1874
1875                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1876                         if (error)
1877                                 break;
1878                         imo->imo_multicast_loop = !!loop;
1879                 } else {
1880                         u_int loop;
1881
1882                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1883                                             sizeof loop);
1884                         if (error)
1885                                 break;
1886                         imo->imo_multicast_loop = !!loop;
1887                 }
1888                 break;
1889
1890         case IP_ADD_MEMBERSHIP:
1891                 /*
1892                  * Add a multicast group membership.
1893                  * Group must be a valid IP multicast address.
1894                  */
1895                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1896                 if (error)
1897                         break;
1898
1899                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1900                         error = EINVAL;
1901                         break;
1902                 }
1903                 crit_enter();
1904                 /*
1905                  * If no interface address was provided, use the interface of
1906                  * the route to the given multicast address.
1907                  */
1908                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1909                         struct sockaddr_in dst;
1910                         struct rtentry *rt;
1911
1912                         bzero(&dst, sizeof(struct sockaddr_in));
1913                         dst.sin_len = sizeof(struct sockaddr_in);
1914                         dst.sin_family = AF_INET;
1915                         dst.sin_addr = mreq.imr_multiaddr;
1916                         rt = rtlookup((struct sockaddr *)&dst);
1917                         if (rt == NULL) {
1918                                 error = EADDRNOTAVAIL;
1919                                 crit_exit();
1920                                 break;
1921                         }
1922                         --rt->rt_refcnt;
1923                         ifp = rt->rt_ifp;
1924                 } else {
1925                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1926                 }
1927
1928                 /*
1929                  * See if we found an interface, and confirm that it
1930                  * supports multicast.
1931                  */
1932                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1933                         error = EADDRNOTAVAIL;
1934                         crit_exit();
1935                         break;
1936                 }
1937                 /*
1938                  * See if the membership already exists or if all the
1939                  * membership slots are full.
1940                  */
1941                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1942                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1943                             imo->imo_membership[i]->inm_addr.s_addr
1944                                                 == mreq.imr_multiaddr.s_addr)
1945                                 break;
1946                 }
1947                 if (i < imo->imo_num_memberships) {
1948                         error = EADDRINUSE;
1949                         crit_exit();
1950                         break;
1951                 }
1952                 if (i == IP_MAX_MEMBERSHIPS) {
1953                         error = ETOOMANYREFS;
1954                         crit_exit();
1955                         break;
1956                 }
1957                 /*
1958                  * Everything looks good; add a new record to the multicast
1959                  * address list for the given interface.
1960                  */
1961                 if ((imo->imo_membership[i] =
1962                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1963                         error = ENOBUFS;
1964                         crit_exit();
1965                         break;
1966                 }
1967                 ++imo->imo_num_memberships;
1968                 crit_exit();
1969                 break;
1970
1971         case IP_DROP_MEMBERSHIP:
1972                 /*
1973                  * Drop a multicast group membership.
1974                  * Group must be a valid IP multicast address.
1975                  */
1976                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1977                 if (error)
1978                         break;
1979
1980                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1981                         error = EINVAL;
1982                         break;
1983                 }
1984
1985                 crit_enter();
1986                 /*
1987                  * If an interface address was specified, get a pointer
1988                  * to its ifnet structure.
1989                  */
1990                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1991                         ifp = NULL;
1992                 else {
1993                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1994                         if (ifp == NULL) {
1995                                 error = EADDRNOTAVAIL;
1996                                 crit_exit();
1997                                 break;
1998                         }
1999                 }
2000                 /*
2001                  * Find the membership in the membership array.
2002                  */
2003                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2004                         if ((ifp == NULL ||
2005                              imo->imo_membership[i]->inm_ifp == ifp) &&
2006                             imo->imo_membership[i]->inm_addr.s_addr ==
2007                             mreq.imr_multiaddr.s_addr)
2008                                 break;
2009                 }
2010                 if (i == imo->imo_num_memberships) {
2011                         error = EADDRNOTAVAIL;
2012                         crit_exit();
2013                         break;
2014                 }
2015                 /*
2016                  * Give up the multicast address record to which the
2017                  * membership points.
2018                  */
2019                 in_delmulti(imo->imo_membership[i]);
2020                 /*
2021                  * Remove the gap in the membership array.
2022                  */
2023                 for (++i; i < imo->imo_num_memberships; ++i)
2024                         imo->imo_membership[i-1] = imo->imo_membership[i];
2025                 --imo->imo_num_memberships;
2026                 crit_exit();
2027                 break;
2028
2029         default:
2030                 error = EOPNOTSUPP;
2031                 break;
2032         }
2033
2034         /*
2035          * If all options have default values, no need to keep the mbuf.
2036          */
2037         if (imo->imo_multicast_ifp == NULL &&
2038             imo->imo_multicast_vif == -1 &&
2039             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2040             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2041             imo->imo_num_memberships == 0) {
2042                 kfree(*imop, M_IPMOPTS);
2043                 *imop = NULL;
2044         }
2045
2046         return (error);
2047 }
2048
2049 /*
2050  * Return the IP multicast options in response to user getsockopt().
2051  */
2052 static int
2053 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2054 {
2055         struct in_addr addr;
2056         struct in_ifaddr *ia;
2057         int error, optval;
2058         u_char coptval;
2059
2060         error = 0;
2061         switch (sopt->sopt_name) {
2062         case IP_MULTICAST_VIF:
2063                 if (imo != NULL)
2064                         optval = imo->imo_multicast_vif;
2065                 else
2066                         optval = -1;
2067                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2068                 break;
2069
2070         case IP_MULTICAST_IF:
2071                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2072                         addr.s_addr = INADDR_ANY;
2073                 else if (imo->imo_multicast_addr.s_addr) {
2074                         /* return the value user has set */
2075                         addr = imo->imo_multicast_addr;
2076                 } else {
2077                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2078                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2079                                 : IA_SIN(ia)->sin_addr.s_addr;
2080                 }
2081                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2082                 break;
2083
2084         case IP_MULTICAST_TTL:
2085                 if (imo == NULL)
2086                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2087                 else
2088                         optval = coptval = imo->imo_multicast_ttl;
2089                 if (sopt->sopt_valsize == 1)
2090                         soopt_from_kbuf(sopt, &coptval, 1);
2091                 else
2092                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2093                 break;
2094
2095         case IP_MULTICAST_LOOP:
2096                 if (imo == NULL)
2097                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2098                 else
2099                         optval = coptval = imo->imo_multicast_loop;
2100                 if (sopt->sopt_valsize == 1)
2101                         soopt_from_kbuf(sopt, &coptval, 1);
2102                 else
2103                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2104                 break;
2105
2106         default:
2107                 error = ENOPROTOOPT;
2108                 break;
2109         }
2110         return (error);
2111 }
2112
2113 /*
2114  * Discard the IP multicast options.
2115  */
2116 void
2117 ip_freemoptions(struct ip_moptions *imo)
2118 {
2119         int i;
2120
2121         if (imo != NULL) {
2122                 for (i = 0; i < imo->imo_num_memberships; ++i)
2123                         in_delmulti(imo->imo_membership[i]);
2124                 kfree(imo, M_IPMOPTS);
2125         }
2126 }
2127
2128 /*
2129  * Routine called from ip_output() to loop back a copy of an IP multicast
2130  * packet to the input queue of a specified interface.  Note that this
2131  * calls the output routine of the loopback "driver", but with an interface
2132  * pointer that might NOT be a loopback interface -- evil, but easier than
2133  * replicating that code here.
2134  */
2135 static void
2136 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2137              int hlen)
2138 {
2139         struct ip *ip;
2140         struct mbuf *copym;
2141
2142         copym = m_copypacket(m, MB_DONTWAIT);
2143         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2144                 copym = m_pullup(copym, hlen);
2145         if (copym != NULL) {
2146                 /*
2147                  * if the checksum hasn't been computed, mark it as valid
2148                  */
2149                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2150                         in_delayed_cksum(copym);
2151                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2152                         copym->m_pkthdr.csum_flags |=
2153                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2154                         copym->m_pkthdr.csum_data = 0xffff;
2155                 }
2156                 /*
2157                  * We don't bother to fragment if the IP length is greater
2158                  * than the interface's MTU.  Can this possibly matter?
2159                  */
2160                 ip = mtod(copym, struct ip *);
2161                 ip->ip_len = htons(ip->ip_len);
2162                 ip->ip_off = htons(ip->ip_off);
2163                 ip->ip_sum = 0;
2164                 if (ip->ip_vhl == IP_VHL_BORING) {
2165                         ip->ip_sum = in_cksum_hdr(ip);
2166                 } else {
2167                         ip->ip_sum = in_cksum(copym, hlen);
2168                 }
2169                 /*
2170                  * NB:
2171                  * It's not clear whether there are any lingering
2172                  * reentrancy problems in other areas which might
2173                  * be exposed by using ip_input directly (in
2174                  * particular, everything which modifies the packet
2175                  * in-place).  Yet another option is using the
2176                  * protosw directly to deliver the looped back
2177                  * packet.  For the moment, we'll err on the side
2178                  * of safety by using if_simloop().
2179                  */
2180 #if 1 /* XXX */
2181                 if (dst->sin_family != AF_INET) {
2182                         kprintf("ip_mloopback: bad address family %d\n",
2183                                                 dst->sin_family);
2184                         dst->sin_family = AF_INET;
2185                 }
2186 #endif
2187                 get_mplock();   /* is if_simloop() mpsafe yet? */
2188                 if_simloop(ifp, copym, dst->sin_family, 0);
2189                 rel_mplock();
2190         }
2191 }