sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.67 2008/10/28 03:07:28 sephe Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42 #include "opt_mpls.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/proc.h>
  53 #include <sys/priv.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/thread2.h>
  56 #include <sys/in_cksum.h>
  57 #include <sys/lock.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                                 backtrace();
 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 #ifdef INVARIANTS
 307         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 308                 /*
 309                  * XXX
 310                  * Multicast is not MPSAFE yet.  Caller must hold
 311                  * BGL when output a multicast IP packet.
 312                  */
 313                 ASSERT_MP_LOCK_HELD(curthread);
 314         }
 315 #endif
 316
 317         dst = (struct sockaddr_in *)&ro->ro_dst;
 318         /*
 319          * If there is a cached route,
 320          * check that it is to the same destination
 321          * and is still up.  If not, free it and try again.
 322          * The address family should also be checked in case of sharing the
 323          * cache with IPv6.
 324          */
 325         if (ro->ro_rt &&
 326             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 327              dst->sin_family != AF_INET ||
 328              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 329                 rtfree(ro->ro_rt);
 330                 ro->ro_rt = (struct rtentry *)NULL;
 331         }
 332         if (ro->ro_rt == NULL) {
 333                 bzero(dst, sizeof *dst);
 334                 dst->sin_family = AF_INET;
 335                 dst->sin_len = sizeof *dst;
 336                 dst->sin_addr = pkt_dst;
 337         }
 338         /*
 339          * If routing to interface only,
 340          * short circuit routing lookup.
 341          */
 342         if (flags & IP_ROUTETOIF) {
 343                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 344                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 345                         ipstat.ips_noroute++;
 346                         error = ENETUNREACH;
 347                         goto bad;
 348                 }
 349                 ifp = ia->ia_ifp;
 350                 ip->ip_ttl = 1;
 351                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 352         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 353                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 354                 /*
 355                  * Bypass the normal routing lookup for multicast
 356                  * packets if the interface is specified.
 357                  */
 358                 ifp = imo->imo_multicast_ifp;
 359                 ia = IFP_TO_IA(ifp);
 360                 isbroadcast = 0;        /* fool gcc */
 361         } else {
 362                 /*
 363                  * If this is the case, we probably don't want to allocate
 364                  * a protocol-cloned route since we didn't get one from the
 365                  * ULP.  This lets TCP do its thing, while not burdening
 366                  * forwarding or ICMP with the overhead of cloning a route.
 367                  * Of course, we still want to do any cloning requested by
 368                  * the link layer, as this is probably required in all cases
 369                  * for correct operation (as it is for ARP).
 370                  */
 371                 if (ro->ro_rt == NULL)
 372                         rtalloc_ign(ro, RTF_PRCLONING);
 373                 if (ro->ro_rt == NULL) {
 374                         ipstat.ips_noroute++;
 375                         error = EHOSTUNREACH;
 376                         goto bad;
 377                 }
 378                 ia = ifatoia(ro->ro_rt->rt_ifa);
 379                 ifp = ro->ro_rt->rt_ifp;
 380                 ro->ro_rt->rt_use++;
 381                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 382                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 383                 if (ro->ro_rt->rt_flags & RTF_HOST)
 384                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 385                 else
 386                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 387         }
 388         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 389                 struct in_multi *inm;
 390
 391                 m->m_flags |= M_MCAST;
 392                 /*
 393                  * IP destination address is multicast.  Make sure "dst"
 394                  * still points to the address in "ro".  (It may have been
 395                  * changed to point to a gateway address, above.)
 396                  */
 397                 dst = (struct sockaddr_in *)&ro->ro_dst;
 398                 /*
 399                  * See if the caller provided any multicast options
 400                  */
 401                 if (imo != NULL) {
 402                         ip->ip_ttl = imo->imo_multicast_ttl;
 403                         if (imo->imo_multicast_vif != -1) {
 404                                 ip->ip_src.s_addr =
 405                                     ip_mcast_src ?
 406                                     ip_mcast_src(imo->imo_multicast_vif) :
 407                                     INADDR_ANY;
 408                         }
 409                 } else {
 410                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 411                 }
 412                 /*
 413                  * Confirm that the outgoing interface supports multicast.
 414                  */
 415                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 416                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 417                                 ipstat.ips_noroute++;
 418                                 error = ENETUNREACH;
 419                                 goto bad;
 420                         }
 421                 }
 422                 /*
 423                  * If source address not specified yet, use address
 424                  * of outgoing interface.
 425                  */
 426                 if (ip->ip_src.s_addr == INADDR_ANY) {
 427                         /* Interface may have no addresses. */
 428                         if (ia != NULL)
 429                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 430                 }
 431
 432                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 433                 if (inm != NULL &&
 434                     (imo == NULL || imo->imo_multicast_loop)) {
 435                         /*
 436                          * If we belong to the destination multicast group
 437                          * on the outgoing interface, and the caller did not
 438                          * forbid loopback, loop back a copy.
 439                          */
 440                         ip_mloopback(ifp, m, dst, hlen);
 441                 } else {
 442                         /*
 443                          * If we are acting as a multicast router, perform
 444                          * multicast forwarding as if the packet had just
 445                          * arrived on the interface to which we are about
 446                          * to send.  The multicast forwarding function
 447                          * recursively calls this function, using the
 448                          * IP_FORWARDING flag to prevent infinite recursion.
 449                          *
 450                          * Multicasts that are looped back by ip_mloopback(),
 451                          * above, will be forwarded by the ip_input() routine,
 452                          * if necessary.
 453                          */
 454                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 455                                 /*
 456                                  * If rsvp daemon is not running, do not
 457                                  * set ip_moptions. This ensures that the packet
 458                                  * is multicast and not just sent down one link
 459                                  * as prescribed by rsvpd.
 460                                  */
 461                                 if (!rsvp_on)
 462                                         imo = NULL;
 463                                 if (ip_mforward &&
 464                                     ip_mforward(ip, ifp, m, imo) != 0) {
 465                                         m_freem(m);
 466                                         goto done;
 467                                 }
 468                         }
 469                 }
 470
 471                 /*
 472                  * Multicasts with a time-to-live of zero may be looped-
 473                  * back, above, but must not be transmitted on a network.
 474                  * Also, multicasts addressed to the loopback interface
 475                  * are not sent -- the above call to ip_mloopback() will
 476                  * loop back a copy if this host actually belongs to the
 477                  * destination group on the loopback interface.
 478                  */
 479                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 480                         m_freem(m);
 481                         goto done;
 482                 }
 483
 484                 goto sendit;
 485         } else {
 486                 m->m_flags &= ~M_MCAST;
 487         }
 488
 489         /*
 490          * If the source address is not specified yet, use the address
 491          * of the outoing interface. In case, keep note we did that, so
 492          * if the the firewall changes the next-hop causing the output
 493          * interface to change, we can fix that.
 494          */
 495         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 496                 /* Interface may have no addresses. */
 497                 if (ia != NULL) {
 498                         ip->ip_src = IA_SIN(ia)->sin_addr;
 499                         src_was_INADDR_ANY = 1;
 500                 }
 501         }
 502
 503 #ifdef ALTQ
 504         /*
 505          * Disable packet drop hack.
 506          * Packetdrop should be done by queueing.
 507          */
 508 #else /* !ALTQ */
 509         /*
 510          * Verify that we have any chance at all of being able to queue
 511          *      the packet or packet fragments
 512          */
 513         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 514             ifp->if_snd.ifq_maxlen) {
 515                 error = ENOBUFS;
 516                 ipstat.ips_odropped++;
 517                 goto bad;
 518         }
 519 #endif /* !ALTQ */
 520
 521         /*
 522          * Look for broadcast address and
 523          * verify user is allowed to send
 524          * such a packet.
 525          */
 526         if (isbroadcast) {
 527                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 528                         error = EADDRNOTAVAIL;
 529                         goto bad;
 530                 }
 531                 if (!(flags & IP_ALLOWBROADCAST)) {
 532                         error = EACCES;
 533                         goto bad;
 534                 }
 535                 /* don't allow broadcast messages to be fragmented */
 536                 if (ip->ip_len > ifp->if_mtu) {
 537                         error = EMSGSIZE;
 538                         goto bad;
 539                 }
 540                 m->m_flags |= M_BCAST;
 541         } else {
 542                 m->m_flags &= ~M_BCAST;
 543         }
 544
 545 sendit:
 546 #ifdef IPSEC
 547         /* get SP for this packet */
 548         if (so == NULL)
 549                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 550         else
 551                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 552
 553         if (sp == NULL) {
 554                 ipsecstat.out_inval++;
 555                 goto bad;
 556         }
 557
 558         error = 0;
 559
 560         /* check policy */
 561         switch (sp->policy) {
 562         case IPSEC_POLICY_DISCARD:
 563                 /*
 564                  * This packet is just discarded.
 565                  */
 566                 ipsecstat.out_polvio++;
 567                 goto bad;
 568
 569         case IPSEC_POLICY_BYPASS:
 570         case IPSEC_POLICY_NONE:
 571                 /* no need to do IPsec. */
 572                 goto skip_ipsec;
 573
 574         case IPSEC_POLICY_IPSEC:
 575                 if (sp->req == NULL) {
 576                         /* acquire a policy */
 577                         error = key_spdacquire(sp);
 578                         goto bad;
 579                 }
 580                 break;
 581
 582         case IPSEC_POLICY_ENTRUST:
 583         default:
 584                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 585         }
 586     {
 587         struct ipsec_output_state state;
 588         bzero(&state, sizeof state);
 589         state.m = m;
 590         if (flags & IP_ROUTETOIF) {
 591                 state.ro = &iproute;
 592                 bzero(&iproute, sizeof iproute);
 593         } else
 594                 state.ro = ro;
 595         state.dst = (struct sockaddr *)dst;
 596
 597         ip->ip_sum = 0;
 598
 599         /*
 600          * XXX
 601          * delayed checksums are not currently compatible with IPsec
 602          */
 603         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 604                 in_delayed_cksum(m);
 605                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 606         }
 607
 608         ip->ip_len = htons(ip->ip_len);
 609         ip->ip_off = htons(ip->ip_off);
 610
 611         error = ipsec4_output(&state, sp, flags);
 612
 613         m = state.m;
 614         if (flags & IP_ROUTETOIF) {
 615                 /*
 616                  * if we have tunnel mode SA, we may need to ignore
 617                  * IP_ROUTETOIF.
 618                  */
 619                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 620                         flags &= ~IP_ROUTETOIF;
 621                         ro = state.ro;
 622                 }
 623         } else
 624                 ro = state.ro;
 625         dst = (struct sockaddr_in *)state.dst;
 626         if (error) {
 627                 /* mbuf is already reclaimed in ipsec4_output. */
 628                 m0 = NULL;
 629                 switch (error) {
 630                 case EHOSTUNREACH:
 631                 case ENETUNREACH:
 632                 case EMSGSIZE:
 633                 case ENOBUFS:
 634                 case ENOMEM:
 635                         break;
 636                 default:
 637                         kprintf("ip4_output (ipsec): error code %d\n", error);
 638                         /*fall through*/
 639                 case ENOENT:
 640                         /* don't show these error codes to the user */
 641                         error = 0;
 642                         break;
 643                 }
 644                 goto bad;
 645         }
 646     }
 647
 648         /* be sure to update variables that are affected by ipsec4_output() */
 649         ip = mtod(m, struct ip *);
 650 #ifdef _IP_VHL
 651         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 652 #else
 653         hlen = ip->ip_hl << 2;
 654 #endif
 655         if (ro->ro_rt == NULL) {
 656                 if (!(flags & IP_ROUTETOIF)) {
 657                         kprintf("ip_output: "
 658                                 "can't update route after IPsec processing\n");
 659                         error = EHOSTUNREACH;   /*XXX*/
 660                         goto bad;
 661                 }
 662         } else {
 663                 ia = ifatoia(ro->ro_rt->rt_ifa);
 664                 ifp = ro->ro_rt->rt_ifp;
 665         }
 666
 667         /* make it flipped, again. */
 668         ip->ip_len = ntohs(ip->ip_len);
 669         ip->ip_off = ntohs(ip->ip_off);
 670 skip_ipsec:
 671 #endif /*IPSEC*/
 672 #ifdef FAST_IPSEC
 673         /*
 674          * Check the security policy (SP) for the packet and, if
 675          * required, do IPsec-related processing.  There are two
 676          * cases here; the first time a packet is sent through
 677          * it will be untagged and handled by ipsec4_checkpolicy.
 678          * If the packet is resubmitted to ip_output (e.g. after
 679          * AH, ESP, etc. processing), there will be a tag to bypass
 680          * the lookup and related policy checking.
 681          */
 682         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 683         crit_enter();
 684         if (mtag != NULL) {
 685                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 686                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 687                 if (sp == NULL)
 688                         error = -EINVAL;        /* force silent drop */
 689                 m_tag_delete(m, mtag);
 690         } else {
 691                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 692                                         &error, inp);
 693         }
 694         /*
 695          * There are four return cases:
 696          *    sp != NULL                    apply IPsec policy
 697          *    sp == NULL, error == 0        no IPsec handling needed
 698          *    sp == NULL, error == -EINVAL  discard packet w/o error
 699          *    sp == NULL, error != 0        discard packet, report error
 700          */
 701         if (sp != NULL) {
 702                 /* Loop detection, check if ipsec processing already done */
 703                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 704                 for (mtag = m_tag_first(m); mtag != NULL;
 705                      mtag = m_tag_next(m, mtag)) {
 706                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 707                                 continue;
 708                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 709                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 710                                 continue;
 711                         /*
 712                          * Check if policy has an SA associated with it.
 713                          * This can happen when an SP has yet to acquire
 714                          * an SA; e.g. on first reference.  If it occurs,
 715                          * then we let ipsec4_process_packet do its thing.
 716                          */
 717                         if (sp->req->sav == NULL)
 718                                 break;
 719                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 720                         if (tdbi->spi == sp->req->sav->spi &&
 721                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 722                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 723                                  sizeof(union sockaddr_union)) == 0) {
 724                                 /*
 725                                  * No IPsec processing is needed, free
 726                                  * reference to SP.
 727                                  *
 728                                  * NB: null pointer to avoid free at
 729                                  *     done: below.
 730                                  */
 731                                 KEY_FREESP(&sp), sp = NULL;
 732                                 crit_exit();
 733                                 goto spd_done;
 734                         }
 735                 }
 736
 737                 /*
 738                  * Do delayed checksums now because we send before
 739                  * this is done in the normal processing path.
 740                  */
 741                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 742                         in_delayed_cksum(m);
 743                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 744                 }
 745
 746                 ip->ip_len = htons(ip->ip_len);
 747                 ip->ip_off = htons(ip->ip_off);
 748
 749                 /* NB: callee frees mbuf */
 750                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 751                 /*
 752                  * Preserve KAME behaviour: ENOENT can be returned
 753                  * when an SA acquire is in progress.  Don't propagate
 754                  * this to user-level; it confuses applications.
 755                  *
 756                  * XXX this will go away when the SADB is redone.
 757                  */
 758                 if (error == ENOENT)
 759                         error = 0;
 760                 crit_exit();
 761                 goto done;
 762         } else {
 763                 crit_exit();
 764
 765                 if (error != 0) {
 766                         /*
 767                          * Hack: -EINVAL is used to signal that a packet
 768                          * should be silently discarded.  This is typically
 769                          * because we asked key management for an SA and
 770                          * it was delayed (e.g. kicked up to IKE).
 771                          */
 772                         if (error == -EINVAL)
 773                                 error = 0;
 774                         goto bad;
 775                 } else {
 776                         /* No IPsec processing for this packet. */
 777                 }
 778 #ifdef notyet
 779                 /*
 780                  * If deferred crypto processing is needed, check that
 781                  * the interface supports it.
 782                  */
 783                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 784                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 785                         /* notify IPsec to do its own crypto */
 786                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 787                         error = EHOSTUNREACH;
 788                         goto bad;
 789                 }
 790 #endif
 791         }
 792 spd_done:
 793 #endif /* FAST_IPSEC */
 794
 795         /* We are already being fwd'd from a firewall. */
 796         if (next_hop != NULL)
 797                 goto pass;
 798
 799         /* No pfil hooks */
 800         if (!pfil_has_hooks(&inet_pfil_hook)) {
 801                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 802                         /*
 803                          * Strip dummynet tags from stranded packets
 804                          */
 805                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 806                         KKASSERT(mtag != NULL);
 807                         m_tag_delete(m, mtag);
 808                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 809                 }
 810                 goto pass;
 811         }
 812
 813         /*
 814          * IpHack's section.
 815          * - Xlate: translate packet's addr/port (NAT).
 816          * - Firewall: deny/allow/etc.
 817          * - Wrap: fake packet's addr/port <unimpl.>
 818          * - Encapsulate: put it in another IP and send out. <unimp.>
 819          */
 820
 821         /*
 822          * Run through list of hooks for output packets.
 823          */
 824         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 825         if (error != 0 || m == NULL)
 826                 goto done;
 827         ip = mtod(m, struct ip *);
 828
 829         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 830                 /*
 831                  * Check dst to make sure it is directly reachable on the
 832                  * interface we previously thought it was.
 833                  * If it isn't (which may be likely in some situations) we have
 834                  * to re-route it (ie, find a route for the next-hop and the
 835                  * associated interface) and set them here. This is nested
 836                  * forwarding which in most cases is undesirable, except where
 837                  * such control is nigh impossible. So we do it here.
 838                  * And I'm babbling.
 839                  */
 840                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 841                 KKASSERT(mtag != NULL);
 842                 next_hop = m_tag_data(mtag);
 843
 844                 /*
 845                  * Try local forwarding first
 846                  */
 847                 if (ip_localforward(m, next_hop, hlen))
 848                         goto done;
 849
 850                 /*
 851                  * Relocate the route based on next_hop.
 852                  * If the current route is inp's cache, keep it untouched.
 853                  */
 854                 if (ro == &iproute && ro->ro_rt != NULL) {
 855                         RTFREE(ro->ro_rt);
 856                         ro->ro_rt = NULL;
 857                 }
 858                 ro = &iproute;
 859                 bzero(ro, sizeof *ro);
 860
 861                 /*
 862                  * Forwarding to broadcast address is not allowed.
 863                  * XXX Should we follow IP_ROUTETOIF?
 864                  */
 865                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 866
 867                 /* We are doing forwarding now */
 868                 flags |= IP_FORWARDING;
 869
 870                 goto reroute;
 871         }
 872
 873         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 874                 struct dn_pkt *dn_pkt;
 875
 876                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 877                 KKASSERT(mtag != NULL);
 878                 dn_pkt = m_tag_data(mtag);
 879
 880                 /*
 881                  * Under certain cases it is not possible to recalculate
 882                  * 'ro' and 'dst', let alone 'flags', so just save them in
 883                  * dummynet tag and avoid the possible wrong reculcalation
 884                  * when we come back to ip_output() again.
 885                  *
 886                  * All other parameters have been already used and so they
 887                  * are not needed anymore.
 888                  * XXX if the ifp is deleted while a pkt is in dummynet,
 889                  * we are in trouble! (TODO use ifnet_detach_event)
 890                  *
 891                  * We need to copy *ro because for ICMP pkts (and maybe
 892                  * others) the caller passed a pointer into the stack;
 893                  * dst might also be a pointer into *ro so it needs to
 894                  * be updated.
 895                  */
 896                 dn_pkt->ro = *ro;
 897                 if (ro->ro_rt)
 898                         ro->ro_rt->rt_refcnt++;
 899                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 900                         /* 'dst' points into 'ro' */
 901                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 902                 }
 903                 dn_pkt->dn_dst = dst;
 904                 dn_pkt->flags = flags;
 905
 906                 ip_dn_queue(m);
 907                 goto done;
 908         }
 909 pass:
 910         /* 127/8 must not appear on wire - RFC1122. */
 911         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 912             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 913                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 914                         ipstat.ips_badaddr++;
 915                         error = EADDRNOTAVAIL;
 916                         goto bad;
 917                 }
 918         }
 919
 920         m->m_pkthdr.csum_flags |= CSUM_IP;
 921         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 922         if (sw_csum & CSUM_DELAY_DATA) {
 923                 in_delayed_cksum(m);
 924                 sw_csum &= ~CSUM_DELAY_DATA;
 925         }
 926         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 927
 928         /*
 929          * If small enough for interface, or the interface will take
 930          * care of the fragmentation for us, can just send directly.
 931          */
 932         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 933             !(ip->ip_off & IP_DF))) {
 934                 ip->ip_len = htons(ip->ip_len);
 935                 ip->ip_off = htons(ip->ip_off);
 936                 ip->ip_sum = 0;
 937                 if (sw_csum & CSUM_DELAY_IP) {
 938                         if (ip->ip_vhl == IP_VHL_BORING)
 939                                 ip->ip_sum = in_cksum_hdr(ip);
 940                         else
 941                                 ip->ip_sum = in_cksum(m, hlen);
 942                 }
 943
 944                 /* Record statistics for this interface address. */
 945                 if (!(flags & IP_FORWARDING) && ia) {
 946                         ia->ia_ifa.if_opackets++;
 947                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 948                 }
 949
 950 #ifdef IPSEC
 951                 /* clean ipsec history once it goes out of the node */
 952                 ipsec_delaux(m);
 953 #endif
 954
 955 #ifdef MBUF_STRESS_TEST
 956                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 957                         struct mbuf *m1, *m2;
 958                         int length, tmp;
 959
 960                         tmp = length = m->m_pkthdr.len;
 961
 962                         while ((length -= mbuf_frag_size) >= 1) {
 963                                 m1 = m_split(m, length, MB_DONTWAIT);
 964                                 if (m1 == NULL)
 965                                         break;
 966                                 m2 = m;
 967                                 while (m2->m_next != NULL)
 968                                         m2 = m2->m_next;
 969                                 m2->m_next = m1;
 970                         }
 971                         m->m_pkthdr.len = tmp;
 972                 }
 973 #endif
 974
 975 #ifdef MPLS
 976                 if (!mpls_output_process(m, ro->ro_rt))
 977                         goto done;
 978 #endif
 979                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 980                                        ro->ro_rt);
 981                 goto done;
 982         }
 983
 984         if (ip->ip_off & IP_DF) {
 985                 error = EMSGSIZE;
 986                 /*
 987                  * This case can happen if the user changed the MTU
 988                  * of an interface after enabling IP on it.  Because
 989                  * most netifs don't keep track of routes pointing to
 990                  * them, there is no way for one to update all its
 991                  * routes when the MTU is changed.
 992                  */
 993                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 994                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 995                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 996                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 997                 }
 998                 ipstat.ips_cantfrag++;
 999                 goto bad;
1000         }
1001
1002         /*
1003          * Too large for interface; fragment if possible. If successful,
1004          * on return, m will point to a list of packets to be sent.
1005          */
1006         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1007         if (error)
1008                 goto bad;
1009         for (; m; m = m0) {
1010                 m0 = m->m_nextpkt;
1011                 m->m_nextpkt = NULL;
1012 #ifdef IPSEC
1013                 /* clean ipsec history once it goes out of the node */
1014                 ipsec_delaux(m);
1015 #endif
1016                 if (error == 0) {
1017                         /* Record statistics for this interface address. */
1018                         if (ia != NULL) {
1019                                 ia->ia_ifa.if_opackets++;
1020                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1021                         }
1022 #ifdef MPLS
1023                         if (!mpls_output_process(m, ro->ro_rt))
1024                                 continue;
1025 #endif
1026                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1027                                                ro->ro_rt);
1028                 } else {
1029                         m_freem(m);
1030                 }
1031         }
1032
1033         if (error == 0)
1034                 ipstat.ips_fragmented++;
1035
1036 done:
1037         if (ro == &iproute && ro->ro_rt != NULL) {
1038                 RTFREE(ro->ro_rt);
1039                 ro->ro_rt = NULL;
1040         }
1041 #ifdef IPSEC
1042         if (sp != NULL) {
1043                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1044                         kprintf("DP ip_output call free SP:%p\n", sp));
1045                 key_freesp(sp);
1046         }
1047 #endif
1048 #ifdef FAST_IPSEC
1049         if (sp != NULL)
1050                 KEY_FREESP(&sp);
1051 #endif
1052         return (error);
1053 bad:
1054         m_freem(m);
1055         goto done;
1056 }
1057
1058 /*
1059  * Create a chain of fragments which fit the given mtu. m_frag points to the
1060  * mbuf to be fragmented; on return it points to the chain with the fragments.
1061  * Return 0 if no error. If error, m_frag may contain a partially built
1062  * chain of fragments that should be freed by the caller.
1063  *
1064  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1065  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1066  */
1067 int
1068 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1069             u_long if_hwassist_flags, int sw_csum)
1070 {
1071         int error = 0;
1072         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1073         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1074         int off;
1075         struct mbuf *m0 = *m_frag;      /* the original packet          */
1076         int firstlen;
1077         struct mbuf **mnext;
1078         int nfrags;
1079
1080         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1081                 ipstat.ips_cantfrag++;
1082                 return EMSGSIZE;
1083         }
1084
1085         /*
1086          * Must be able to put at least 8 bytes per fragment.
1087          */
1088         if (len < 8)
1089                 return EMSGSIZE;
1090
1091         /*
1092          * If the interface will not calculate checksums on
1093          * fragmented packets, then do it here.
1094          */
1095         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1096             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1097                 in_delayed_cksum(m0);
1098                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1099         }
1100
1101         if (len > PAGE_SIZE) {
1102                 /*
1103                  * Fragment large datagrams such that each segment
1104                  * contains a multiple of PAGE_SIZE amount of data,
1105                  * plus headers. This enables a receiver to perform
1106                  * page-flipping zero-copy optimizations.
1107                  *
1108                  * XXX When does this help given that sender and receiver
1109                  * could have different page sizes, and also mtu could
1110                  * be less than the receiver's page size ?
1111                  */
1112                 int newlen;
1113                 struct mbuf *m;
1114
1115                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1116                         off += m->m_len;
1117
1118                 /*
1119                  * firstlen (off - hlen) must be aligned on an
1120                  * 8-byte boundary
1121                  */
1122                 if (off < hlen)
1123                         goto smart_frag_failure;
1124                 off = ((off - hlen) & ~7) + hlen;
1125                 newlen = (~PAGE_MASK) & mtu;
1126                 if ((newlen + sizeof(struct ip)) > mtu) {
1127                         /* we failed, go back the default */
1128 smart_frag_failure:
1129                         newlen = len;
1130                         off = hlen + len;
1131                 }
1132                 len = newlen;
1133
1134         } else {
1135                 off = hlen + len;
1136         }
1137
1138         firstlen = off - hlen;
1139         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1140
1141         /*
1142          * Loop through length of segment after first fragment,
1143          * make new header and copy data of each part and link onto chain.
1144          * Here, m0 is the original packet, m is the fragment being created.
1145          * The fragments are linked off the m_nextpkt of the original
1146          * packet, which after processing serves as the first fragment.
1147          */
1148         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1149                 struct ip *mhip;        /* ip header on the fragment */
1150                 struct mbuf *m;
1151                 int mhlen = sizeof(struct ip);
1152
1153                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1154                 if (m == NULL) {
1155                         error = ENOBUFS;
1156                         ipstat.ips_odropped++;
1157                         goto done;
1158                 }
1159                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1160                 /*
1161                  * In the first mbuf, leave room for the link header, then
1162                  * copy the original IP header including options. The payload
1163                  * goes into an additional mbuf chain returned by m_copy().
1164                  */
1165                 m->m_data += max_linkhdr;
1166                 mhip = mtod(m, struct ip *);
1167                 *mhip = *ip;
1168                 if (hlen > sizeof(struct ip)) {
1169                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1170                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1171                 }
1172                 m->m_len = mhlen;
1173                 /* XXX do we need to add ip->ip_off below ? */
1174                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1175                 if (off + len >= ip->ip_len) {  /* last fragment */
1176                         len = ip->ip_len - off;
1177                         m->m_flags |= M_LASTFRAG;
1178                 } else
1179                         mhip->ip_off |= IP_MF;
1180                 mhip->ip_len = htons((u_short)(len + mhlen));
1181                 m->m_next = m_copy(m0, off, len);
1182                 if (m->m_next == NULL) {                /* copy failed */
1183                         m_free(m);
1184                         error = ENOBUFS;        /* ??? */
1185                         ipstat.ips_odropped++;
1186                         goto done;
1187                 }
1188                 m->m_pkthdr.len = mhlen + len;
1189                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1190                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1191                 mhip->ip_off = htons(mhip->ip_off);
1192                 mhip->ip_sum = 0;
1193                 if (sw_csum & CSUM_DELAY_IP)
1194                         mhip->ip_sum = in_cksum(m, mhlen);
1195                 *mnext = m;
1196                 mnext = &m->m_nextpkt;
1197         }
1198         ipstat.ips_ofragments += nfrags;
1199
1200         /* set first marker for fragment chain */
1201         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1202         m0->m_pkthdr.csum_data = nfrags;
1203
1204         /*
1205          * Update first fragment by trimming what's been copied out
1206          * and updating header.
1207          */
1208         m_adj(m0, hlen + firstlen - ip->ip_len);
1209         m0->m_pkthdr.len = hlen + firstlen;
1210         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1211         ip->ip_off |= IP_MF;
1212         ip->ip_off = htons(ip->ip_off);
1213         ip->ip_sum = 0;
1214         if (sw_csum & CSUM_DELAY_IP)
1215                 ip->ip_sum = in_cksum(m0, hlen);
1216
1217 done:
1218         *m_frag = m0;
1219         return error;
1220 }
1221
1222 void
1223 in_delayed_cksum(struct mbuf *m)
1224 {
1225         struct ip *ip;
1226         u_short csum, offset;
1227
1228         ip = mtod(m, struct ip *);
1229         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1230         csum = in_cksum_skip(m, ip->ip_len, offset);
1231         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1232                 csum = 0xffff;
1233         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1234
1235         if (offset + sizeof(u_short) > m->m_len) {
1236                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1237                     m->m_len, offset, ip->ip_p);
1238                 /*
1239                  * XXX
1240                  * this shouldn't happen, but if it does, the
1241                  * correct behavior may be to insert the checksum
1242                  * in the existing chain instead of rearranging it.
1243                  */
1244                 m = m_pullup(m, offset + sizeof(u_short));
1245         }
1246         *(u_short *)(m->m_data + offset) = csum;
1247 }
1248
1249 /*
1250  * Insert IP options into preformed packet.
1251  * Adjust IP destination as required for IP source routing,
1252  * as indicated by a non-zero in_addr at the start of the options.
1253  *
1254  * XXX This routine assumes that the packet has no options in place.
1255  */
1256 static struct mbuf *
1257 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1258 {
1259         struct ipoption *p = mtod(opt, struct ipoption *);
1260         struct mbuf *n;
1261         struct ip *ip = mtod(m, struct ip *);
1262         unsigned optlen;
1263
1264         optlen = opt->m_len - sizeof p->ipopt_dst;
1265         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1266                 *phlen = 0;
1267                 return (m);             /* XXX should fail */
1268         }
1269         if (p->ipopt_dst.s_addr)
1270                 ip->ip_dst = p->ipopt_dst;
1271         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1272                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1273                 if (n == NULL) {
1274                         *phlen = 0;
1275                         return (m);
1276                 }
1277                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1278                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1279                 m->m_len -= sizeof(struct ip);
1280                 m->m_data += sizeof(struct ip);
1281                 n->m_next = m;
1282                 m = n;
1283                 m->m_len = optlen + sizeof(struct ip);
1284                 m->m_data += max_linkhdr;
1285                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1286         } else {
1287                 m->m_data -= optlen;
1288                 m->m_len += optlen;
1289                 m->m_pkthdr.len += optlen;
1290                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1291         }
1292         ip = mtod(m, struct ip *);
1293         bcopy(p->ipopt_list, ip + 1, optlen);
1294         *phlen = sizeof(struct ip) + optlen;
1295         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1296         ip->ip_len += optlen;
1297         return (m);
1298 }
1299
1300 /*
1301  * Copy options from ip to jp,
1302  * omitting those not copied during fragmentation.
1303  */
1304 int
1305 ip_optcopy(struct ip *ip, struct ip *jp)
1306 {
1307         u_char *cp, *dp;
1308         int opt, optlen, cnt;
1309
1310         cp = (u_char *)(ip + 1);
1311         dp = (u_char *)(jp + 1);
1312         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1313         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1314                 opt = cp[0];
1315                 if (opt == IPOPT_EOL)
1316                         break;
1317                 if (opt == IPOPT_NOP) {
1318                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1319                         *dp++ = IPOPT_NOP;
1320                         optlen = 1;
1321                         continue;
1322                 }
1323
1324                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1325                     ("ip_optcopy: malformed ipv4 option"));
1326                 optlen = cp[IPOPT_OLEN];
1327                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1328                     ("ip_optcopy: malformed ipv4 option"));
1329
1330                 /* bogus lengths should have been caught by ip_dooptions */
1331                 if (optlen > cnt)
1332                         optlen = cnt;
1333                 if (IPOPT_COPIED(opt)) {
1334                         bcopy(cp, dp, optlen);
1335                         dp += optlen;
1336                 }
1337         }
1338         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1339                 *dp++ = IPOPT_EOL;
1340         return (optlen);
1341 }
1342
1343 /*
1344  * IP socket option processing.
1345  */
1346 int
1347 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1348 {
1349         struct  inpcb *inp = so->so_pcb;
1350         int     error, optval;
1351
1352         error = optval = 0;
1353         if (sopt->sopt_level != IPPROTO_IP) {
1354                 return (EINVAL);
1355         }
1356
1357         switch (sopt->sopt_dir) {
1358         case SOPT_SET:
1359                 switch (sopt->sopt_name) {
1360                 case IP_OPTIONS:
1361 #ifdef notyet
1362                 case IP_RETOPTS:
1363 #endif
1364                 {
1365                         struct mbuf *m;
1366                         if (sopt->sopt_valsize > MLEN) {
1367                                 error = EMSGSIZE;
1368                                 break;
1369                         }
1370                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1371                         if (m == NULL) {
1372                                 error = ENOBUFS;
1373                                 break;
1374                         }
1375                         m->m_len = sopt->sopt_valsize;
1376                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1377                                               m->m_len);
1378                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1379                                            m));
1380                 }
1381
1382                 case IP_TOS:
1383                 case IP_TTL:
1384                 case IP_MINTTL:
1385                 case IP_RECVOPTS:
1386                 case IP_RECVRETOPTS:
1387                 case IP_RECVDSTADDR:
1388                 case IP_RECVIF:
1389                 case IP_RECVTTL:
1390                 case IP_FAITH:
1391                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1392                                              sizeof optval);
1393                         if (error)
1394                                 break;
1395                         switch (sopt->sopt_name) {
1396                         case IP_TOS:
1397                                 inp->inp_ip_tos = optval;
1398                                 break;
1399
1400                         case IP_TTL:
1401                                 inp->inp_ip_ttl = optval;
1402                                 break;
1403                         case IP_MINTTL:
1404                                 if (optval > 0 && optval <= MAXTTL)
1405                                         inp->inp_ip_minttl = optval;
1406                                 else
1407                                         error = EINVAL;
1408                                 break;
1409 #define OPTSET(bit) \
1410         if (optval) \
1411                 inp->inp_flags |= bit; \
1412         else \
1413                 inp->inp_flags &= ~bit;
1414
1415                         case IP_RECVOPTS:
1416                                 OPTSET(INP_RECVOPTS);
1417                                 break;
1418
1419                         case IP_RECVRETOPTS:
1420                                 OPTSET(INP_RECVRETOPTS);
1421                                 break;
1422
1423                         case IP_RECVDSTADDR:
1424                                 OPTSET(INP_RECVDSTADDR);
1425                                 break;
1426
1427                         case IP_RECVIF:
1428                                 OPTSET(INP_RECVIF);
1429                                 break;
1430
1431                         case IP_RECVTTL:
1432                                 OPTSET(INP_RECVTTL);
1433                                 break;
1434
1435                         case IP_FAITH:
1436                                 OPTSET(INP_FAITH);
1437                                 break;
1438                         }
1439                         break;
1440 #undef OPTSET
1441
1442                 case IP_MULTICAST_IF:
1443                 case IP_MULTICAST_VIF:
1444                 case IP_MULTICAST_TTL:
1445                 case IP_MULTICAST_LOOP:
1446                 case IP_ADD_MEMBERSHIP:
1447                 case IP_DROP_MEMBERSHIP:
1448                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1449                         break;
1450
1451                 case IP_PORTRANGE:
1452                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1453                                             sizeof optval);
1454                         if (error)
1455                                 break;
1456
1457                         switch (optval) {
1458                         case IP_PORTRANGE_DEFAULT:
1459                                 inp->inp_flags &= ~(INP_LOWPORT);
1460                                 inp->inp_flags &= ~(INP_HIGHPORT);
1461                                 break;
1462
1463                         case IP_PORTRANGE_HIGH:
1464                                 inp->inp_flags &= ~(INP_LOWPORT);
1465                                 inp->inp_flags |= INP_HIGHPORT;
1466                                 break;
1467
1468                         case IP_PORTRANGE_LOW:
1469                                 inp->inp_flags &= ~(INP_HIGHPORT);
1470                                 inp->inp_flags |= INP_LOWPORT;
1471                                 break;
1472
1473                         default:
1474                                 error = EINVAL;
1475                                 break;
1476                         }
1477                         break;
1478
1479 #if defined(IPSEC) || defined(FAST_IPSEC)
1480                 case IP_IPSEC_POLICY:
1481                 {
1482                         caddr_t req;
1483                         size_t len = 0;
1484                         int priv;
1485                         struct mbuf *m;
1486                         int optname;
1487
1488                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1489                                 break;
1490                         soopt_to_mbuf(sopt, m);
1491                         priv = (sopt->sopt_td != NULL &&
1492                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1493                         req = mtod(m, caddr_t);
1494                         len = m->m_len;
1495                         optname = sopt->sopt_name;
1496                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1497                         m_freem(m);
1498                         break;
1499                 }
1500 #endif /*IPSEC*/
1501
1502                 default:
1503                         error = ENOPROTOOPT;
1504                         break;
1505                 }
1506                 break;
1507
1508         case SOPT_GET:
1509                 switch (sopt->sopt_name) {
1510                 case IP_OPTIONS:
1511                 case IP_RETOPTS:
1512                         if (inp->inp_options)
1513                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1514                                                            char *),
1515                                                 inp->inp_options->m_len);
1516                         else
1517                                 sopt->sopt_valsize = 0;
1518                         break;
1519
1520                 case IP_TOS:
1521                 case IP_TTL:
1522                 case IP_MINTTL:
1523                 case IP_RECVOPTS:
1524                 case IP_RECVRETOPTS:
1525                 case IP_RECVDSTADDR:
1526                 case IP_RECVTTL:
1527                 case IP_RECVIF:
1528                 case IP_PORTRANGE:
1529                 case IP_FAITH:
1530                         switch (sopt->sopt_name) {
1531
1532                         case IP_TOS:
1533                                 optval = inp->inp_ip_tos;
1534                                 break;
1535
1536                         case IP_TTL:
1537                                 optval = inp->inp_ip_ttl;
1538                                 break;
1539                         case IP_MINTTL:
1540                                 optval = inp->inp_ip_minttl;
1541                                 break;
1542
1543 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1544
1545                         case IP_RECVOPTS:
1546                                 optval = OPTBIT(INP_RECVOPTS);
1547                                 break;
1548
1549                         case IP_RECVRETOPTS:
1550                                 optval = OPTBIT(INP_RECVRETOPTS);
1551                                 break;
1552
1553                         case IP_RECVDSTADDR:
1554                                 optval = OPTBIT(INP_RECVDSTADDR);
1555                                 break;
1556
1557                         case IP_RECVTTL:
1558                                 optval = OPTBIT(INP_RECVTTL);
1559                                 break;
1560
1561                         case IP_RECVIF:
1562                                 optval = OPTBIT(INP_RECVIF);
1563                                 break;
1564
1565                         case IP_PORTRANGE:
1566                                 if (inp->inp_flags & INP_HIGHPORT)
1567                                         optval = IP_PORTRANGE_HIGH;
1568                                 else if (inp->inp_flags & INP_LOWPORT)
1569                                         optval = IP_PORTRANGE_LOW;
1570                                 else
1571                                         optval = 0;
1572                                 break;
1573
1574                         case IP_FAITH:
1575                                 optval = OPTBIT(INP_FAITH);
1576                                 break;
1577                         }
1578                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1579                         break;
1580
1581                 case IP_MULTICAST_IF:
1582                 case IP_MULTICAST_VIF:
1583                 case IP_MULTICAST_TTL:
1584                 case IP_MULTICAST_LOOP:
1585                 case IP_ADD_MEMBERSHIP:
1586                 case IP_DROP_MEMBERSHIP:
1587                         error = ip_getmoptions(sopt, inp->inp_moptions);
1588                         break;
1589
1590 #if defined(IPSEC) || defined(FAST_IPSEC)
1591                 case IP_IPSEC_POLICY:
1592                 {
1593                         struct mbuf *m = NULL;
1594                         caddr_t req = NULL;
1595                         size_t len = 0;
1596
1597                         if (m != NULL) {
1598                                 req = mtod(m, caddr_t);
1599                                 len = m->m_len;
1600                         }
1601                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1602                         if (error == 0)
1603                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1604                         if (error == 0)
1605                                 m_freem(m);
1606                         break;
1607                 }
1608 #endif /*IPSEC*/
1609
1610                 default:
1611                         error = ENOPROTOOPT;
1612                         break;
1613                 }
1614                 break;
1615         }
1616         return (error);
1617 }
1618
1619 /*
1620  * Set up IP options in pcb for insertion in output packets.
1621  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1622  * with destination address if source routed.
1623  */
1624 static int
1625 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1626 {
1627         int cnt, optlen;
1628         u_char *cp;
1629         u_char opt;
1630
1631         /* turn off any old options */
1632         if (*pcbopt)
1633                 m_free(*pcbopt);
1634         *pcbopt = 0;
1635         if (m == NULL || m->m_len == 0) {
1636                 /*
1637                  * Only turning off any previous options.
1638                  */
1639                 if (m != NULL)
1640                         m_free(m);
1641                 return (0);
1642         }
1643
1644         if (m->m_len % sizeof(int32_t))
1645                 goto bad;
1646         /*
1647          * IP first-hop destination address will be stored before
1648          * actual options; move other options back
1649          * and clear it when none present.
1650          */
1651         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1652                 goto bad;
1653         cnt = m->m_len;
1654         m->m_len += sizeof(struct in_addr);
1655         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1656         ovbcopy(mtod(m, caddr_t), cp, cnt);
1657         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1658
1659         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1660                 opt = cp[IPOPT_OPTVAL];
1661                 if (opt == IPOPT_EOL)
1662                         break;
1663                 if (opt == IPOPT_NOP)
1664                         optlen = 1;
1665                 else {
1666                         if (cnt < IPOPT_OLEN + sizeof *cp)
1667                                 goto bad;
1668                         optlen = cp[IPOPT_OLEN];
1669                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1670                                 goto bad;
1671                 }
1672                 switch (opt) {
1673
1674                 default:
1675                         break;
1676
1677                 case IPOPT_LSRR:
1678                 case IPOPT_SSRR:
1679                         /*
1680                          * user process specifies route as:
1681                          *      ->A->B->C->D
1682                          * D must be our final destination (but we can't
1683                          * check that since we may not have connected yet).
1684                          * A is first hop destination, which doesn't appear in
1685                          * actual IP option, but is stored before the options.
1686                          */
1687                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1688                                 goto bad;
1689                         m->m_len -= sizeof(struct in_addr);
1690                         cnt -= sizeof(struct in_addr);
1691                         optlen -= sizeof(struct in_addr);
1692                         cp[IPOPT_OLEN] = optlen;
1693                         /*
1694                          * Move first hop before start of options.
1695                          */
1696                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1697                               sizeof(struct in_addr));
1698                         /*
1699                          * Then copy rest of options back
1700                          * to close up the deleted entry.
1701                          */
1702                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1703                                 &cp[IPOPT_OFFSET+1],
1704                                 cnt - (IPOPT_MINOFF - 1));
1705                         break;
1706                 }
1707         }
1708         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1709                 goto bad;
1710         *pcbopt = m;
1711         return (0);
1712
1713 bad:
1714         m_free(m);
1715         return (EINVAL);
1716 }
1717
1718 /*
1719  * XXX
1720  * The whole multicast option thing needs to be re-thought.
1721  * Several of these options are equally applicable to non-multicast
1722  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1723  * standard option (IP_TTL).
1724  */
1725
1726 /*
1727  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1728  */
1729 static struct ifnet *
1730 ip_multicast_if(struct in_addr *a, int *ifindexp)
1731 {
1732         int ifindex;
1733         struct ifnet *ifp;
1734
1735         if (ifindexp)
1736                 *ifindexp = 0;
1737         if (ntohl(a->s_addr) >> 24 == 0) {
1738                 ifindex = ntohl(a->s_addr) & 0xffffff;
1739                 if (ifindex < 0 || if_index < ifindex)
1740                         return NULL;
1741                 ifp = ifindex2ifnet[ifindex];
1742                 if (ifindexp)
1743                         *ifindexp = ifindex;
1744         } else {
1745                 ifp = INADDR_TO_IFP(a);
1746         }
1747         return ifp;
1748 }
1749
1750 /*
1751  * Set the IP multicast options in response to user setsockopt().
1752  */
1753 static int
1754 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1755 {
1756         int error = 0;
1757         int i;
1758         struct in_addr addr;
1759         struct ip_mreq mreq;
1760         struct ifnet *ifp;
1761         struct ip_moptions *imo = *imop;
1762         int ifindex;
1763
1764         if (imo == NULL) {
1765                 /*
1766                  * No multicast option buffer attached to the pcb;
1767                  * allocate one and initialize to default values.
1768                  */
1769                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1770
1771                 *imop = imo;
1772                 imo->imo_multicast_ifp = NULL;
1773                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1774                 imo->imo_multicast_vif = -1;
1775                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1776                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1777                 imo->imo_num_memberships = 0;
1778         }
1779         switch (sopt->sopt_name) {
1780         /* store an index number for the vif you wanna use in the send */
1781         case IP_MULTICAST_VIF:
1782                 if (legal_vif_num == 0) {
1783                         error = EOPNOTSUPP;
1784                         break;
1785                 }
1786                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1787                 if (error)
1788                         break;
1789                 if (!legal_vif_num(i) && (i != -1)) {
1790                         error = EINVAL;
1791                         break;
1792                 }
1793                 imo->imo_multicast_vif = i;
1794                 break;
1795
1796         case IP_MULTICAST_IF:
1797                 /*
1798                  * Select the interface for outgoing multicast packets.
1799                  */
1800                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1801                 if (error)
1802                         break;
1803
1804                 /*
1805                  * INADDR_ANY is used to remove a previous selection.
1806                  * When no interface is selected, a default one is
1807                  * chosen every time a multicast packet is sent.
1808                  */
1809                 if (addr.s_addr == INADDR_ANY) {
1810                         imo->imo_multicast_ifp = NULL;
1811                         break;
1812                 }
1813                 /*
1814                  * The selected interface is identified by its local
1815                  * IP address.  Find the interface and confirm that
1816                  * it supports multicasting.
1817                  */
1818                 crit_enter();
1819                 ifp = ip_multicast_if(&addr, &ifindex);
1820                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1821                         crit_exit();
1822                         error = EADDRNOTAVAIL;
1823                         break;
1824                 }
1825                 imo->imo_multicast_ifp = ifp;
1826                 if (ifindex)
1827                         imo->imo_multicast_addr = addr;
1828                 else
1829                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1830                 crit_exit();
1831                 break;
1832
1833         case IP_MULTICAST_TTL:
1834                 /*
1835                  * Set the IP time-to-live for outgoing multicast packets.
1836                  * The original multicast API required a char argument,
1837                  * which is inconsistent with the rest of the socket API.
1838                  * We allow either a char or an int.
1839                  */
1840                 if (sopt->sopt_valsize == 1) {
1841                         u_char ttl;
1842                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1843                         if (error)
1844                                 break;
1845                         imo->imo_multicast_ttl = ttl;
1846                 } else {
1847                         u_int ttl;
1848                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1849                         if (error)
1850                                 break;
1851                         if (ttl > 255)
1852                                 error = EINVAL;
1853                         else
1854                                 imo->imo_multicast_ttl = ttl;
1855                 }
1856                 break;
1857
1858         case IP_MULTICAST_LOOP:
1859                 /*
1860                  * Set the loopback flag for outgoing multicast packets.
1861                  * Must be zero or one.  The original multicast API required a
1862                  * char argument, which is inconsistent with the rest
1863                  * of the socket API.  We allow either a char or an int.
1864                  */
1865                 if (sopt->sopt_valsize == 1) {
1866                         u_char loop;
1867
1868                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1869                         if (error)
1870                                 break;
1871                         imo->imo_multicast_loop = !!loop;
1872                 } else {
1873                         u_int loop;
1874
1875                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1876                                             sizeof loop);
1877                         if (error)
1878                                 break;
1879                         imo->imo_multicast_loop = !!loop;
1880                 }
1881                 break;
1882
1883         case IP_ADD_MEMBERSHIP:
1884                 /*
1885                  * Add a multicast group membership.
1886                  * Group must be a valid IP multicast address.
1887                  */
1888                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1889                 if (error)
1890                         break;
1891
1892                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1893                         error = EINVAL;
1894                         break;
1895                 }
1896                 crit_enter();
1897                 /*
1898                  * If no interface address was provided, use the interface of
1899                  * the route to the given multicast address.
1900                  */
1901                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1902                         struct sockaddr_in dst;
1903                         struct rtentry *rt;
1904
1905                         bzero(&dst, sizeof(struct sockaddr_in));
1906                         dst.sin_len = sizeof(struct sockaddr_in);
1907                         dst.sin_family = AF_INET;
1908                         dst.sin_addr = mreq.imr_multiaddr;
1909                         rt = rtlookup((struct sockaddr *)&dst);
1910                         if (rt == NULL) {
1911                                 error = EADDRNOTAVAIL;
1912                                 crit_exit();
1913                                 break;
1914                         }
1915                         --rt->rt_refcnt;
1916                         ifp = rt->rt_ifp;
1917                 } else {
1918                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1919                 }
1920
1921                 /*
1922                  * See if we found an interface, and confirm that it
1923                  * supports multicast.
1924                  */
1925                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1926                         error = EADDRNOTAVAIL;
1927                         crit_exit();
1928                         break;
1929                 }
1930                 /*
1931                  * See if the membership already exists or if all the
1932                  * membership slots are full.
1933                  */
1934                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1935                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1936                             imo->imo_membership[i]->inm_addr.s_addr
1937                                                 == mreq.imr_multiaddr.s_addr)
1938                                 break;
1939                 }
1940                 if (i < imo->imo_num_memberships) {
1941                         error = EADDRINUSE;
1942                         crit_exit();
1943                         break;
1944                 }
1945                 if (i == IP_MAX_MEMBERSHIPS) {
1946                         error = ETOOMANYREFS;
1947                         crit_exit();
1948                         break;
1949                 }
1950                 /*
1951                  * Everything looks good; add a new record to the multicast
1952                  * address list for the given interface.
1953                  */
1954                 if ((imo->imo_membership[i] =
1955                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1956                         error = ENOBUFS;
1957                         crit_exit();
1958                         break;
1959                 }
1960                 ++imo->imo_num_memberships;
1961                 crit_exit();
1962                 break;
1963
1964         case IP_DROP_MEMBERSHIP:
1965                 /*
1966                  * Drop a multicast group membership.
1967                  * Group must be a valid IP multicast address.
1968                  */
1969                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1970                 if (error)
1971                         break;
1972
1973                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1974                         error = EINVAL;
1975                         break;
1976                 }
1977
1978                 crit_enter();
1979                 /*
1980                  * If an interface address was specified, get a pointer
1981                  * to its ifnet structure.
1982                  */
1983                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1984                         ifp = NULL;
1985                 else {
1986                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1987                         if (ifp == NULL) {
1988                                 error = EADDRNOTAVAIL;
1989                                 crit_exit();
1990                                 break;
1991                         }
1992                 }
1993                 /*
1994                  * Find the membership in the membership array.
1995                  */
1996                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1997                         if ((ifp == NULL ||
1998                              imo->imo_membership[i]->inm_ifp == ifp) &&
1999                             imo->imo_membership[i]->inm_addr.s_addr ==
2000                             mreq.imr_multiaddr.s_addr)
2001                                 break;
2002                 }
2003                 if (i == imo->imo_num_memberships) {
2004                         error = EADDRNOTAVAIL;
2005                         crit_exit();
2006                         break;
2007                 }
2008                 /*
2009                  * Give up the multicast address record to which the
2010                  * membership points.
2011                  */
2012                 in_delmulti(imo->imo_membership[i]);
2013                 /*
2014                  * Remove the gap in the membership array.
2015                  */
2016                 for (++i; i < imo->imo_num_memberships; ++i)
2017                         imo->imo_membership[i-1] = imo->imo_membership[i];
2018                 --imo->imo_num_memberships;
2019                 crit_exit();
2020                 break;
2021
2022         default:
2023                 error = EOPNOTSUPP;
2024                 break;
2025         }
2026
2027         /*
2028          * If all options have default values, no need to keep the mbuf.
2029          */
2030         if (imo->imo_multicast_ifp == NULL &&
2031             imo->imo_multicast_vif == -1 &&
2032             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2033             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2034             imo->imo_num_memberships == 0) {
2035                 kfree(*imop, M_IPMOPTS);
2036                 *imop = NULL;
2037         }
2038
2039         return (error);
2040 }
2041
2042 /*
2043  * Return the IP multicast options in response to user getsockopt().
2044  */
2045 static int
2046 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2047 {
2048         struct in_addr addr;
2049         struct in_ifaddr *ia;
2050         int error, optval;
2051         u_char coptval;
2052
2053         error = 0;
2054         switch (sopt->sopt_name) {
2055         case IP_MULTICAST_VIF:
2056                 if (imo != NULL)
2057                         optval = imo->imo_multicast_vif;
2058                 else
2059                         optval = -1;
2060                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2061                 break;
2062
2063         case IP_MULTICAST_IF:
2064                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2065                         addr.s_addr = INADDR_ANY;
2066                 else if (imo->imo_multicast_addr.s_addr) {
2067                         /* return the value user has set */
2068                         addr = imo->imo_multicast_addr;
2069                 } else {
2070                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2071                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2072                                 : IA_SIN(ia)->sin_addr.s_addr;
2073                 }
2074                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2075                 break;
2076
2077         case IP_MULTICAST_TTL:
2078                 if (imo == NULL)
2079                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2080                 else
2081                         optval = coptval = imo->imo_multicast_ttl;
2082                 if (sopt->sopt_valsize == 1)
2083                         soopt_from_kbuf(sopt, &coptval, 1);
2084                 else
2085                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2086                 break;
2087
2088         case IP_MULTICAST_LOOP:
2089                 if (imo == NULL)
2090                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2091                 else
2092                         optval = coptval = imo->imo_multicast_loop;
2093                 if (sopt->sopt_valsize == 1)
2094                         soopt_from_kbuf(sopt, &coptval, 1);
2095                 else
2096                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2097                 break;
2098
2099         default:
2100                 error = ENOPROTOOPT;
2101                 break;
2102         }
2103         return (error);
2104 }
2105
2106 /*
2107  * Discard the IP multicast options.
2108  */
2109 void
2110 ip_freemoptions(struct ip_moptions *imo)
2111 {
2112         int i;
2113
2114         if (imo != NULL) {
2115                 for (i = 0; i < imo->imo_num_memberships; ++i)
2116                         in_delmulti(imo->imo_membership[i]);
2117                 kfree(imo, M_IPMOPTS);
2118         }
2119 }
2120
2121 /*
2122  * Routine called from ip_output() to loop back a copy of an IP multicast
2123  * packet to the input queue of a specified interface.  Note that this
2124  * calls the output routine of the loopback "driver", but with an interface
2125  * pointer that might NOT be a loopback interface -- evil, but easier than
2126  * replicating that code here.
2127  */
2128 static void
2129 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2130              int hlen)
2131 {
2132         struct ip *ip;
2133         struct mbuf *copym;
2134
2135         copym = m_copypacket(m, MB_DONTWAIT);
2136         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2137                 copym = m_pullup(copym, hlen);
2138         if (copym != NULL) {
2139                 /*
2140                  * if the checksum hasn't been computed, mark it as valid
2141                  */
2142                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2143                         in_delayed_cksum(copym);
2144                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2145                         copym->m_pkthdr.csum_flags |=
2146                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2147                         copym->m_pkthdr.csum_data = 0xffff;
2148                 }
2149                 /*
2150                  * We don't bother to fragment if the IP length is greater
2151                  * than the interface's MTU.  Can this possibly matter?
2152                  */
2153                 ip = mtod(copym, struct ip *);
2154                 ip->ip_len = htons(ip->ip_len);
2155                 ip->ip_off = htons(ip->ip_off);
2156                 ip->ip_sum = 0;
2157                 if (ip->ip_vhl == IP_VHL_BORING) {
2158                         ip->ip_sum = in_cksum_hdr(ip);
2159                 } else {
2160                         ip->ip_sum = in_cksum(copym, hlen);
2161                 }
2162                 /*
2163                  * NB:
2164                  * It's not clear whether there are any lingering
2165                  * reentrancy problems in other areas which might
2166                  * be exposed by using ip_input directly (in
2167                  * particular, everything which modifies the packet
2168                  * in-place).  Yet another option is using the
2169                  * protosw directly to deliver the looped back
2170                  * packet.  For the moment, we'll err on the side
2171                  * of safety by using if_simloop().
2172                  */
2173 #if 1 /* XXX */
2174                 if (dst->sin_family != AF_INET) {
2175                         kprintf("ip_mloopback: bad address family %d\n",
2176                                                 dst->sin_family);
2177                         dst->sin_family = AF_INET;
2178                 }
2179 #endif
2180                 if_simloop(ifp, copym, dst->sin_family, 0);
2181         }
2182 }