sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.67 2008/10/28 03:07:28 sephe Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42 #include "opt_mpls.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/proc.h>
  53 #include <sys/priv.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/in_cksum.h>
  56 #include <sys/lock.h>
  57
  58 #include <sys/thread2.h>
  59 #include <sys/mplock2.h>
  60
  61 #include <net/if.h>
  62 #include <net/netisr.h>
  63 #include <net/pfil.h>
  64 #include <net/route.h>
  65
  66 #include <netinet/in.h>
  67 #include <netinet/in_systm.h>
  68 #include <netinet/ip.h>
  69 #include <netinet/in_pcb.h>
  70 #include <netinet/in_var.h>
  71 #include <netinet/ip_var.h>
  72
  73 #include <netproto/mpls/mpls_var.h>
  74
  75 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  76
  77 #ifdef IPSEC
  78 #include <netinet6/ipsec.h>
  79 #include <netproto/key/key.h>
  80 #ifdef IPSEC_DEBUG
  81 #include <netproto/key/key_debug.h>
  82 #else
  83 #define KEYDEBUG(lev,arg)
  84 #endif
  85 #endif /*IPSEC*/
  86
  87 #ifdef FAST_IPSEC
  88 #include <netproto/ipsec/ipsec.h>
  89 #include <netproto/ipsec/xform.h>
  90 #include <netproto/ipsec/key.h>
  91 #endif /*FAST_IPSEC*/
  92
  93 #include <net/ipfw/ip_fw.h>
  94 #include <net/dummynet/ip_dummynet.h>
  95
  96 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  97                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  98                                   (ntohl(a.s_addr)>>16)&0xFF,\
  99                                   (ntohl(a.s_addr)>>8)&0xFF,\
 100                                   (ntohl(a.s_addr))&0xFF, y);
 101
 102 u_short ip_id;
 103
 104 #ifdef MBUF_STRESS_TEST
 105 int mbuf_frag_size = 0;
 106 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 107         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 108 #endif
 109
 110 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 111 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 112 static void     ip_mloopback
 113         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 114 static int      ip_getmoptions
 115         (struct sockopt *, struct ip_moptions *);
 116 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 117 static int      ip_setmoptions
 118         (struct sockopt *, struct ip_moptions **);
 119
 120 int     ip_optcopy(struct ip *, struct ip *);
 121
 122 extern  int route_assert_owner_access;
 123
 124 extern  struct protosw inetsw[];
 125
 126 static int
 127 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
 128 {
 129         struct in_ifaddr_container *iac;
 130
 131         /*
 132          * We need to figure out if we have been forwarded to a local
 133          * socket.  If so, then we should somehow "loop back" to
 134          * ip_input(), and get directed to the PCB as if we had received
 135          * this packet.  This is because it may be difficult to identify
 136          * the packets you want to forward until they are being output
 137          * and have selected an interface (e.g. locally initiated
 138          * packets).  If we used the loopback inteface, we would not be
 139          * able to control what happens as the packet runs through
 140          * ip_input() as it is done through a ISR.
 141          */
 142         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 143                 /*
 144                  * If the addr to forward to is one of ours, we pretend
 145                  * to be the destination for this packet.
 146                  */
 147                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 148                         break;
 149         }
 150         if (iac != NULL) {
 151                 struct ip *ip;
 152
 153                 if (m->m_pkthdr.rcvif == NULL)
 154                         m->m_pkthdr.rcvif = ifunit("lo0");
 155                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 156                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 157                                                   CSUM_PSEUDO_HDR;
 158                         m->m_pkthdr.csum_data = 0xffff;
 159                 }
 160                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 161
 162                 /*
 163                  * Make sure that the IP header is in one mbuf,
 164                  * required by ip_input
 165                  */
 166                 if (m->m_len < hlen) {
 167                         m = m_pullup(m, hlen);
 168                         if (m == NULL) {
 169                                 /* The packet was freed; we are done */
 170                                 return 1;
 171                         }
 172                 }
 173                 ip = mtod(m, struct ip *);
 174
 175                 ip->ip_len = htons(ip->ip_len);
 176                 ip->ip_off = htons(ip->ip_off);
 177                 ip_input(m);
 178
 179                 return 1; /* The packet gets forwarded locally */
 180         }
 181         return 0;
 182 }
 183
 184 /*
 185  * IP output.  The packet in mbuf chain m contains a skeletal IP
 186  * header (with len, off, ttl, proto, tos, src, dst).
 187  * The mbuf chain containing the packet will be freed.
 188  * The mbuf opt, if present, will not be freed.
 189  */
 190 int
 191 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 192           int flags, struct ip_moptions *imo, struct inpcb *inp)
 193 {
 194         struct ip *ip;
 195         struct ifnet *ifp = NULL;       /* keep compiler happy */
 196         struct mbuf *m;
 197         int hlen = sizeof(struct ip);
 198         int len, error = 0;
 199         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 200         struct in_ifaddr *ia = NULL;
 201         int isbroadcast, sw_csum;
 202         struct in_addr pkt_dst;
 203         struct route iproute;
 204         struct m_tag *mtag;
 205 #ifdef IPSEC
 206         struct secpolicy *sp = NULL;
 207         struct socket *so = inp ? inp->inp_socket : NULL;
 208 #endif
 209 #ifdef FAST_IPSEC
 210         struct secpolicy *sp = NULL;
 211         struct tdb_ident *tdbi;
 212 #endif /* FAST_IPSEC */
 213         struct sockaddr_in *next_hop = NULL;
 214         int src_was_INADDR_ANY = 0;     /* as the name says... */
 215
 216         m = m0;
 217         M_ASSERTPKTHDR(m);
 218
 219         if (ro == NULL) {
 220                 ro = &iproute;
 221                 bzero(ro, sizeof *ro);
 222         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 223                 if (flags & IP_DEBUGROUTE) {
 224                         if (route_assert_owner_access) {
 225                                 panic("ip_output: "
 226                                       "rt rt_cpuid %d accessed on cpu %d\n",
 227                                       ro->ro_rt->rt_cpuid, mycpuid);
 228                         } else {
 229                                 kprintf("ip_output: "
 230                                         "rt rt_cpuid %d accessed on cpu %d\n",
 231                                         ro->ro_rt->rt_cpuid, mycpuid);
 232                                 print_backtrace(-1);
 233                         }
 234                 }
 235
 236                 /*
 237                  * XXX
 238                  * If the cached rtentry's owner CPU is not the current CPU,
 239                  * then don't touch the cached rtentry (remote free is too
 240                  * expensive in this context); just relocate the route.
 241                  */
 242                 ro = &iproute;
 243                 bzero(ro, sizeof *ro);
 244         }
 245
 246         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 247                 /* Next hop */
 248                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 249                 KKASSERT(mtag != NULL);
 250                 next_hop = m_tag_data(mtag);
 251         }
 252
 253         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 254                 struct dn_pkt *dn_pkt;
 255
 256                 /* Extract info from dummynet tag */
 257                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 258                 KKASSERT(mtag != NULL);
 259                 dn_pkt = m_tag_data(mtag);
 260
 261                 /*
 262                  * The packet was already tagged, so part of the
 263                  * processing was already done, and we need to go down.
 264                  * Get the calculated parameters from the tag.
 265                  */
 266                 ifp = dn_pkt->ifp;
 267
 268                 KKASSERT(ro == &iproute);
 269                 *ro = dn_pkt->ro; /* structure copy */
 270                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 271
 272                 dst = dn_pkt->dn_dst;
 273                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 274                         /* If 'dst' points into dummynet tag, adjust it */
 275                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 276                 }
 277
 278                 ip = mtod(m, struct ip *);
 279                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 280                 if (ro->ro_rt)
 281                         ia = ifatoia(ro->ro_rt->rt_ifa);
 282                 goto sendit;
 283         }
 284
 285         if (opt) {
 286                 len = 0;
 287                 m = ip_insertoptions(m, opt, &len);
 288                 if (len != 0)
 289                         hlen = len;
 290         }
 291         ip = mtod(m, struct ip *);
 292
 293         /*
 294          * Fill in IP header.
 295          */
 296         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 297                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 298                 ip->ip_off &= IP_DF;
 299                 ip->ip_id = ip_newid();
 300                 ipstat.ips_localout++;
 301         } else {
 302                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 303         }
 304
 305 reroute:
 306         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 307
 308 #ifdef INVARIANTS
 309         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 310                 /*
 311                  * XXX
 312                  * Multicast is not MPSAFE yet.  Caller must hold
 313                  * BGL when output a multicast IP packet.
 314                  */
 315                 ASSERT_MP_LOCK_HELD(curthread);
 316         }
 317 #endif
 318
 319         dst = (struct sockaddr_in *)&ro->ro_dst;
 320         /*
 321          * If there is a cached route,
 322          * check that it is to the same destination
 323          * and is still up.  If not, free it and try again.
 324          * The address family should also be checked in case of sharing the
 325          * cache with IPv6.
 326          */
 327         if (ro->ro_rt &&
 328             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 329              dst->sin_family != AF_INET ||
 330              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 331                 rtfree(ro->ro_rt);
 332                 ro->ro_rt = NULL;
 333         }
 334         if (ro->ro_rt == NULL) {
 335                 bzero(dst, sizeof *dst);
 336                 dst->sin_family = AF_INET;
 337                 dst->sin_len = sizeof *dst;
 338                 dst->sin_addr = pkt_dst;
 339         }
 340         /*
 341          * If routing to interface only,
 342          * short circuit routing lookup.
 343          */
 344         if (flags & IP_ROUTETOIF) {
 345                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 346                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 347                         ipstat.ips_noroute++;
 348                         error = ENETUNREACH;
 349                         goto bad;
 350                 }
 351                 ifp = ia->ia_ifp;
 352                 ip->ip_ttl = 1;
 353                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 354         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 355                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 356                 /*
 357                  * Bypass the normal routing lookup for multicast
 358                  * packets if the interface is specified.
 359                  */
 360                 ifp = imo->imo_multicast_ifp;
 361                 ia = IFP_TO_IA(ifp);
 362                 isbroadcast = 0;        /* fool gcc */
 363         } else {
 364                 /*
 365                  * If this is the case, we probably don't want to allocate
 366                  * a protocol-cloned route since we didn't get one from the
 367                  * ULP.  This lets TCP do its thing, while not burdening
 368                  * forwarding or ICMP with the overhead of cloning a route.
 369                  * Of course, we still want to do any cloning requested by
 370                  * the link layer, as this is probably required in all cases
 371                  * for correct operation (as it is for ARP).
 372                  */
 373                 if (ro->ro_rt == NULL)
 374                         rtalloc_ign(ro, RTF_PRCLONING);
 375                 if (ro->ro_rt == NULL) {
 376                         ipstat.ips_noroute++;
 377                         error = EHOSTUNREACH;
 378                         goto bad;
 379                 }
 380                 ia = ifatoia(ro->ro_rt->rt_ifa);
 381                 ifp = ro->ro_rt->rt_ifp;
 382                 ro->ro_rt->rt_use++;
 383                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 384                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 385                 if (ro->ro_rt->rt_flags & RTF_HOST)
 386                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 387                 else
 388                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 389         }
 390         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 391                 struct in_multi *inm;
 392
 393                 m->m_flags |= M_MCAST;
 394                 /*
 395                  * IP destination address is multicast.  Make sure "dst"
 396                  * still points to the address in "ro".  (It may have been
 397                  * changed to point to a gateway address, above.)
 398                  */
 399                 dst = (struct sockaddr_in *)&ro->ro_dst;
 400                 /*
 401                  * See if the caller provided any multicast options
 402                  */
 403                 if (imo != NULL) {
 404                         ip->ip_ttl = imo->imo_multicast_ttl;
 405                         if (imo->imo_multicast_vif != -1) {
 406                                 ip->ip_src.s_addr =
 407                                     ip_mcast_src ?
 408                                     ip_mcast_src(imo->imo_multicast_vif) :
 409                                     INADDR_ANY;
 410                         }
 411                 } else {
 412                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 413                 }
 414                 /*
 415                  * Confirm that the outgoing interface supports multicast.
 416                  */
 417                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 418                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 419                                 ipstat.ips_noroute++;
 420                                 error = ENETUNREACH;
 421                                 goto bad;
 422                         }
 423                 }
 424                 /*
 425                  * If source address not specified yet, use address
 426                  * of outgoing interface.
 427                  */
 428                 if (ip->ip_src.s_addr == INADDR_ANY) {
 429                         /* Interface may have no addresses. */
 430                         if (ia != NULL)
 431                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 432                 }
 433
 434                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 435                 if (inm != NULL &&
 436                     (imo == NULL || imo->imo_multicast_loop)) {
 437                         /*
 438                          * If we belong to the destination multicast group
 439                          * on the outgoing interface, and the caller did not
 440                          * forbid loopback, loop back a copy.
 441                          */
 442                         ip_mloopback(ifp, m, dst, hlen);
 443                 } else {
 444                         /*
 445                          * If we are acting as a multicast router, perform
 446                          * multicast forwarding as if the packet had just
 447                          * arrived on the interface to which we are about
 448                          * to send.  The multicast forwarding function
 449                          * recursively calls this function, using the
 450                          * IP_FORWARDING flag to prevent infinite recursion.
 451                          *
 452                          * Multicasts that are looped back by ip_mloopback(),
 453                          * above, will be forwarded by the ip_input() routine,
 454                          * if necessary.
 455                          */
 456                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 457                                 /*
 458                                  * If rsvp daemon is not running, do not
 459                                  * set ip_moptions. This ensures that the packet
 460                                  * is multicast and not just sent down one link
 461                                  * as prescribed by rsvpd.
 462                                  */
 463                                 if (!rsvp_on)
 464                                         imo = NULL;
 465                                 if (ip_mforward &&
 466                                     ip_mforward(ip, ifp, m, imo) != 0) {
 467                                         m_freem(m);
 468                                         goto done;
 469                                 }
 470                         }
 471                 }
 472
 473                 /*
 474                  * Multicasts with a time-to-live of zero may be looped-
 475                  * back, above, but must not be transmitted on a network.
 476                  * Also, multicasts addressed to the loopback interface
 477                  * are not sent -- the above call to ip_mloopback() will
 478                  * loop back a copy if this host actually belongs to the
 479                  * destination group on the loopback interface.
 480                  */
 481                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 482                         m_freem(m);
 483                         goto done;
 484                 }
 485
 486                 goto sendit;
 487         } else {
 488                 m->m_flags &= ~M_MCAST;
 489         }
 490
 491         /*
 492          * If the source address is not specified yet, use the address
 493          * of the outoing interface. In case, keep note we did that, so
 494          * if the the firewall changes the next-hop causing the output
 495          * interface to change, we can fix that.
 496          */
 497         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 498                 /* Interface may have no addresses. */
 499                 if (ia != NULL) {
 500                         ip->ip_src = IA_SIN(ia)->sin_addr;
 501                         src_was_INADDR_ANY = 1;
 502                 }
 503         }
 504
 505 #ifdef ALTQ
 506         /*
 507          * Disable packet drop hack.
 508          * Packetdrop should be done by queueing.
 509          */
 510 #else /* !ALTQ */
 511         /*
 512          * Verify that we have any chance at all of being able to queue
 513          *      the packet or packet fragments
 514          */
 515         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 516             ifp->if_snd.ifq_maxlen) {
 517                 error = ENOBUFS;
 518                 ipstat.ips_odropped++;
 519                 goto bad;
 520         }
 521 #endif /* !ALTQ */
 522
 523         /*
 524          * Look for broadcast address and
 525          * verify user is allowed to send
 526          * such a packet.
 527          */
 528         if (isbroadcast) {
 529                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 530                         error = EADDRNOTAVAIL;
 531                         goto bad;
 532                 }
 533                 if (!(flags & IP_ALLOWBROADCAST)) {
 534                         error = EACCES;
 535                         goto bad;
 536                 }
 537                 /* don't allow broadcast messages to be fragmented */
 538                 if (ip->ip_len > ifp->if_mtu) {
 539                         error = EMSGSIZE;
 540                         goto bad;
 541                 }
 542                 m->m_flags |= M_BCAST;
 543         } else {
 544                 m->m_flags &= ~M_BCAST;
 545         }
 546
 547 sendit:
 548 #ifdef IPSEC
 549         /* get SP for this packet */
 550         if (so == NULL)
 551                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 552         else
 553                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 554
 555         if (sp == NULL) {
 556                 ipsecstat.out_inval++;
 557                 goto bad;
 558         }
 559
 560         error = 0;
 561
 562         /* check policy */
 563         switch (sp->policy) {
 564         case IPSEC_POLICY_DISCARD:
 565                 /*
 566                  * This packet is just discarded.
 567                  */
 568                 ipsecstat.out_polvio++;
 569                 goto bad;
 570
 571         case IPSEC_POLICY_BYPASS:
 572         case IPSEC_POLICY_NONE:
 573         case IPSEC_POLICY_TCP:
 574                 /* no need to do IPsec. */
 575                 goto skip_ipsec;
 576
 577         case IPSEC_POLICY_IPSEC:
 578                 if (sp->req == NULL) {
 579                         /* acquire a policy */
 580                         error = key_spdacquire(sp);
 581                         goto bad;
 582                 }
 583                 break;
 584
 585         case IPSEC_POLICY_ENTRUST:
 586         default:
 587                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 588         }
 589     {
 590         struct ipsec_output_state state;
 591         bzero(&state, sizeof state);
 592         state.m = m;
 593         if (flags & IP_ROUTETOIF) {
 594                 state.ro = &iproute;
 595                 bzero(&iproute, sizeof iproute);
 596         } else
 597                 state.ro = ro;
 598         state.dst = (struct sockaddr *)dst;
 599
 600         ip->ip_sum = 0;
 601
 602         /*
 603          * XXX
 604          * delayed checksums are not currently compatible with IPsec
 605          */
 606         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 607                 in_delayed_cksum(m);
 608                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 609         }
 610
 611         ip->ip_len = htons(ip->ip_len);
 612         ip->ip_off = htons(ip->ip_off);
 613
 614         error = ipsec4_output(&state, sp, flags);
 615
 616         m = state.m;
 617         if (flags & IP_ROUTETOIF) {
 618                 /*
 619                  * if we have tunnel mode SA, we may need to ignore
 620                  * IP_ROUTETOIF.
 621                  */
 622                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 623                         flags &= ~IP_ROUTETOIF;
 624                         ro = state.ro;
 625                 }
 626         } else
 627                 ro = state.ro;
 628         dst = (struct sockaddr_in *)state.dst;
 629         if (error) {
 630                 /* mbuf is already reclaimed in ipsec4_output. */
 631                 m0 = NULL;
 632                 switch (error) {
 633                 case EHOSTUNREACH:
 634                 case ENETUNREACH:
 635                 case EMSGSIZE:
 636                 case ENOBUFS:
 637                 case ENOMEM:
 638                         break;
 639                 default:
 640                         kprintf("ip4_output (ipsec): error code %d\n", error);
 641                         /*fall through*/
 642                 case ENOENT:
 643                         /* don't show these error codes to the user */
 644                         error = 0;
 645                         break;
 646                 }
 647                 goto bad;
 648         }
 649     }
 650
 651         /* be sure to update variables that are affected by ipsec4_output() */
 652         ip = mtod(m, struct ip *);
 653 #ifdef _IP_VHL
 654         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 655 #else
 656         hlen = ip->ip_hl << 2;
 657 #endif
 658         if (ro->ro_rt == NULL) {
 659                 if (!(flags & IP_ROUTETOIF)) {
 660                         kprintf("ip_output: "
 661                                 "can't update route after IPsec processing\n");
 662                         error = EHOSTUNREACH;   /*XXX*/
 663                         goto bad;
 664                 }
 665         } else {
 666                 ia = ifatoia(ro->ro_rt->rt_ifa);
 667                 ifp = ro->ro_rt->rt_ifp;
 668         }
 669
 670         /* make it flipped, again. */
 671         ip->ip_len = ntohs(ip->ip_len);
 672         ip->ip_off = ntohs(ip->ip_off);
 673 skip_ipsec:
 674 #endif /*IPSEC*/
 675 #ifdef FAST_IPSEC
 676         /*
 677          * Check the security policy (SP) for the packet and, if
 678          * required, do IPsec-related processing.  There are two
 679          * cases here; the first time a packet is sent through
 680          * it will be untagged and handled by ipsec4_checkpolicy.
 681          * If the packet is resubmitted to ip_output (e.g. after
 682          * AH, ESP, etc. processing), there will be a tag to bypass
 683          * the lookup and related policy checking.
 684          */
 685         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 686         crit_enter();
 687         if (mtag != NULL) {
 688                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 689                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 690                 if (sp == NULL)
 691                         error = -EINVAL;        /* force silent drop */
 692                 m_tag_delete(m, mtag);
 693         } else {
 694                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 695                                         &error, inp);
 696         }
 697         /*
 698          * There are four return cases:
 699          *    sp != NULL                    apply IPsec policy
 700          *    sp == NULL, error == 0        no IPsec handling needed
 701          *    sp == NULL, error == -EINVAL  discard packet w/o error
 702          *    sp == NULL, error != 0        discard packet, report error
 703          */
 704         if (sp != NULL) {
 705                 /* Loop detection, check if ipsec processing already done */
 706                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 707                 for (mtag = m_tag_first(m); mtag != NULL;
 708                      mtag = m_tag_next(m, mtag)) {
 709                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 710                                 continue;
 711                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 712                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 713                                 continue;
 714                         /*
 715                          * Check if policy has an SA associated with it.
 716                          * This can happen when an SP has yet to acquire
 717                          * an SA; e.g. on first reference.  If it occurs,
 718                          * then we let ipsec4_process_packet do its thing.
 719                          */
 720                         if (sp->req->sav == NULL)
 721                                 break;
 722                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 723                         if (tdbi->spi == sp->req->sav->spi &&
 724                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 725                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 726                                  sizeof(union sockaddr_union)) == 0) {
 727                                 /*
 728                                  * No IPsec processing is needed, free
 729                                  * reference to SP.
 730                                  *
 731                                  * NB: null pointer to avoid free at
 732                                  *     done: below.
 733                                  */
 734                                 KEY_FREESP(&sp), sp = NULL;
 735                                 crit_exit();
 736                                 goto spd_done;
 737                         }
 738                 }
 739
 740                 /*
 741                  * Do delayed checksums now because we send before
 742                  * this is done in the normal processing path.
 743                  */
 744                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 745                         in_delayed_cksum(m);
 746                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 747                 }
 748
 749                 ip->ip_len = htons(ip->ip_len);
 750                 ip->ip_off = htons(ip->ip_off);
 751
 752                 /* NB: callee frees mbuf */
 753                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 754                 /*
 755                  * Preserve KAME behaviour: ENOENT can be returned
 756                  * when an SA acquire is in progress.  Don't propagate
 757                  * this to user-level; it confuses applications.
 758                  *
 759                  * XXX this will go away when the SADB is redone.
 760                  */
 761                 if (error == ENOENT)
 762                         error = 0;
 763                 crit_exit();
 764                 goto done;
 765         } else {
 766                 crit_exit();
 767
 768                 if (error != 0) {
 769                         /*
 770                          * Hack: -EINVAL is used to signal that a packet
 771                          * should be silently discarded.  This is typically
 772                          * because we asked key management for an SA and
 773                          * it was delayed (e.g. kicked up to IKE).
 774                          */
 775                         if (error == -EINVAL)
 776                                 error = 0;
 777                         goto bad;
 778                 } else {
 779                         /* No IPsec processing for this packet. */
 780                 }
 781 #ifdef notyet
 782                 /*
 783                  * If deferred crypto processing is needed, check that
 784                  * the interface supports it.
 785                  */
 786                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 787                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 788                         /* notify IPsec to do its own crypto */
 789                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 790                         error = EHOSTUNREACH;
 791                         goto bad;
 792                 }
 793 #endif
 794         }
 795 spd_done:
 796 #endif /* FAST_IPSEC */
 797
 798         /* We are already being fwd'd from a firewall. */
 799         if (next_hop != NULL)
 800                 goto pass;
 801
 802         /* No pfil hooks */
 803         if (!pfil_has_hooks(&inet_pfil_hook)) {
 804                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 805                         /*
 806                          * Strip dummynet tags from stranded packets
 807                          */
 808                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 809                         KKASSERT(mtag != NULL);
 810                         m_tag_delete(m, mtag);
 811                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 812                 }
 813                 goto pass;
 814         }
 815
 816         /*
 817          * IpHack's section.
 818          * - Xlate: translate packet's addr/port (NAT).
 819          * - Firewall: deny/allow/etc.
 820          * - Wrap: fake packet's addr/port <unimpl.>
 821          * - Encapsulate: put it in another IP and send out. <unimp.>
 822          */
 823
 824         /*
 825          * Run through list of hooks for output packets.
 826          */
 827         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 828         if (error != 0 || m == NULL)
 829                 goto done;
 830         ip = mtod(m, struct ip *);
 831
 832         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 833                 /*
 834                  * Check dst to make sure it is directly reachable on the
 835                  * interface we previously thought it was.
 836                  * If it isn't (which may be likely in some situations) we have
 837                  * to re-route it (ie, find a route for the next-hop and the
 838                  * associated interface) and set them here. This is nested
 839                  * forwarding which in most cases is undesirable, except where
 840                  * such control is nigh impossible. So we do it here.
 841                  * And I'm babbling.
 842                  */
 843                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 844                 KKASSERT(mtag != NULL);
 845                 next_hop = m_tag_data(mtag);
 846
 847                 /*
 848                  * Try local forwarding first
 849                  */
 850                 if (ip_localforward(m, next_hop, hlen))
 851                         goto done;
 852
 853                 /*
 854                  * Relocate the route based on next_hop.
 855                  * If the current route is inp's cache, keep it untouched.
 856                  */
 857                 if (ro == &iproute && ro->ro_rt != NULL) {
 858                         RTFREE(ro->ro_rt);
 859                         ro->ro_rt = NULL;
 860                 }
 861                 ro = &iproute;
 862                 bzero(ro, sizeof *ro);
 863
 864                 /*
 865                  * Forwarding to broadcast address is not allowed.
 866                  * XXX Should we follow IP_ROUTETOIF?
 867                  */
 868                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 869
 870                 /* We are doing forwarding now */
 871                 flags |= IP_FORWARDING;
 872
 873                 goto reroute;
 874         }
 875
 876         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 877                 struct dn_pkt *dn_pkt;
 878
 879                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 880                 KKASSERT(mtag != NULL);
 881                 dn_pkt = m_tag_data(mtag);
 882
 883                 /*
 884                  * Under certain cases it is not possible to recalculate
 885                  * 'ro' and 'dst', let alone 'flags', so just save them in
 886                  * dummynet tag and avoid the possible wrong reculcalation
 887                  * when we come back to ip_output() again.
 888                  *
 889                  * All other parameters have been already used and so they
 890                  * are not needed anymore.
 891                  * XXX if the ifp is deleted while a pkt is in dummynet,
 892                  * we are in trouble! (TODO use ifnet_detach_event)
 893                  *
 894                  * We need to copy *ro because for ICMP pkts (and maybe
 895                  * others) the caller passed a pointer into the stack;
 896                  * dst might also be a pointer into *ro so it needs to
 897                  * be updated.
 898                  */
 899                 dn_pkt->ro = *ro;
 900                 if (ro->ro_rt)
 901                         ro->ro_rt->rt_refcnt++;
 902                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 903                         /* 'dst' points into 'ro' */
 904                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 905                 }
 906                 dn_pkt->dn_dst = dst;
 907                 dn_pkt->flags = flags;
 908
 909                 ip_dn_queue(m);
 910                 goto done;
 911         }
 912 pass:
 913         /* 127/8 must not appear on wire - RFC1122. */
 914         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 915             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 916                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 917                         ipstat.ips_badaddr++;
 918                         error = EADDRNOTAVAIL;
 919                         goto bad;
 920                 }
 921         }
 922
 923         m->m_pkthdr.csum_flags |= CSUM_IP;
 924         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 925         if (sw_csum & CSUM_DELAY_DATA) {
 926                 in_delayed_cksum(m);
 927                 sw_csum &= ~CSUM_DELAY_DATA;
 928         }
 929         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 930
 931         /*
 932          * If small enough for interface, or the interface will take
 933          * care of the fragmentation for us, can just send directly.
 934          */
 935         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 936             !(ip->ip_off & IP_DF))) {
 937                 ip->ip_len = htons(ip->ip_len);
 938                 ip->ip_off = htons(ip->ip_off);
 939                 ip->ip_sum = 0;
 940                 if (sw_csum & CSUM_DELAY_IP) {
 941                         if (ip->ip_vhl == IP_VHL_BORING)
 942                                 ip->ip_sum = in_cksum_hdr(ip);
 943                         else
 944                                 ip->ip_sum = in_cksum(m, hlen);
 945                 }
 946
 947                 /* Record statistics for this interface address. */
 948                 if (!(flags & IP_FORWARDING) && ia) {
 949                         ia->ia_ifa.if_opackets++;
 950                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 951                 }
 952
 953 #ifdef IPSEC
 954                 /* clean ipsec history once it goes out of the node */
 955                 ipsec_delaux(m);
 956 #endif
 957
 958 #ifdef MBUF_STRESS_TEST
 959                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 960                         struct mbuf *m1, *m2;
 961                         int length, tmp;
 962
 963                         tmp = length = m->m_pkthdr.len;
 964
 965                         while ((length -= mbuf_frag_size) >= 1) {
 966                                 m1 = m_split(m, length, MB_DONTWAIT);
 967                                 if (m1 == NULL)
 968                                         break;
 969                                 m2 = m;
 970                                 while (m2->m_next != NULL)
 971                                         m2 = m2->m_next;
 972                                 m2->m_next = m1;
 973                         }
 974                         m->m_pkthdr.len = tmp;
 975                 }
 976 #endif
 977
 978 #ifdef MPLS
 979                 if (!mpls_output_process(m, ro->ro_rt))
 980                         goto done;
 981 #endif
 982                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 983                                        ro->ro_rt);
 984                 goto done;
 985         }
 986
 987         if (ip->ip_off & IP_DF) {
 988                 error = EMSGSIZE;
 989                 /*
 990                  * This case can happen if the user changed the MTU
 991                  * of an interface after enabling IP on it.  Because
 992                  * most netifs don't keep track of routes pointing to
 993                  * them, there is no way for one to update all its
 994                  * routes when the MTU is changed.
 995                  */
 996                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 997                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 998                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 999                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1000                 }
1001                 ipstat.ips_cantfrag++;
1002                 goto bad;
1003         }
1004
1005         /*
1006          * Too large for interface; fragment if possible. If successful,
1007          * on return, m will point to a list of packets to be sent.
1008          */
1009         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1010         if (error)
1011                 goto bad;
1012         for (; m; m = m0) {
1013                 m0 = m->m_nextpkt;
1014                 m->m_nextpkt = NULL;
1015 #ifdef IPSEC
1016                 /* clean ipsec history once it goes out of the node */
1017                 ipsec_delaux(m);
1018 #endif
1019                 if (error == 0) {
1020                         /* Record statistics for this interface address. */
1021                         if (ia != NULL) {
1022                                 ia->ia_ifa.if_opackets++;
1023                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1024                         }
1025 #ifdef MPLS
1026                         if (!mpls_output_process(m, ro->ro_rt))
1027                                 continue;
1028 #endif
1029                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1030                                                ro->ro_rt);
1031                 } else {
1032                         m_freem(m);
1033                 }
1034         }
1035
1036         if (error == 0)
1037                 ipstat.ips_fragmented++;
1038
1039 done:
1040         if (ro == &iproute && ro->ro_rt != NULL) {
1041                 RTFREE(ro->ro_rt);
1042                 ro->ro_rt = NULL;
1043         }
1044 #ifdef IPSEC
1045         if (sp != NULL) {
1046                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1047                         kprintf("DP ip_output call free SP:%p\n", sp));
1048                 key_freesp(sp);
1049         }
1050 #endif
1051 #ifdef FAST_IPSEC
1052         if (sp != NULL)
1053                 KEY_FREESP(&sp);
1054 #endif
1055         return (error);
1056 bad:
1057         m_freem(m);
1058         goto done;
1059 }
1060
1061 /*
1062  * Create a chain of fragments which fit the given mtu. m_frag points to the
1063  * mbuf to be fragmented; on return it points to the chain with the fragments.
1064  * Return 0 if no error. If error, m_frag may contain a partially built
1065  * chain of fragments that should be freed by the caller.
1066  *
1067  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1068  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1069  */
1070 int
1071 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1072             u_long if_hwassist_flags, int sw_csum)
1073 {
1074         int error = 0;
1075         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1076         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1077         int off;
1078         struct mbuf *m0 = *m_frag;      /* the original packet          */
1079         int firstlen;
1080         struct mbuf **mnext;
1081         int nfrags;
1082
1083         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1084                 ipstat.ips_cantfrag++;
1085                 return EMSGSIZE;
1086         }
1087
1088         /*
1089          * Must be able to put at least 8 bytes per fragment.
1090          */
1091         if (len < 8)
1092                 return EMSGSIZE;
1093
1094         /*
1095          * If the interface will not calculate checksums on
1096          * fragmented packets, then do it here.
1097          */
1098         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1099             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1100                 in_delayed_cksum(m0);
1101                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1102         }
1103
1104         if (len > PAGE_SIZE) {
1105                 /*
1106                  * Fragment large datagrams such that each segment
1107                  * contains a multiple of PAGE_SIZE amount of data,
1108                  * plus headers. This enables a receiver to perform
1109                  * page-flipping zero-copy optimizations.
1110                  *
1111                  * XXX When does this help given that sender and receiver
1112                  * could have different page sizes, and also mtu could
1113                  * be less than the receiver's page size ?
1114                  */
1115                 int newlen;
1116                 struct mbuf *m;
1117
1118                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1119                         off += m->m_len;
1120
1121                 /*
1122                  * firstlen (off - hlen) must be aligned on an
1123                  * 8-byte boundary
1124                  */
1125                 if (off < hlen)
1126                         goto smart_frag_failure;
1127                 off = ((off - hlen) & ~7) + hlen;
1128                 newlen = (~PAGE_MASK) & mtu;
1129                 if ((newlen + sizeof(struct ip)) > mtu) {
1130                         /* we failed, go back the default */
1131 smart_frag_failure:
1132                         newlen = len;
1133                         off = hlen + len;
1134                 }
1135                 len = newlen;
1136
1137         } else {
1138                 off = hlen + len;
1139         }
1140
1141         firstlen = off - hlen;
1142         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1143
1144         /*
1145          * Loop through length of segment after first fragment,
1146          * make new header and copy data of each part and link onto chain.
1147          * Here, m0 is the original packet, m is the fragment being created.
1148          * The fragments are linked off the m_nextpkt of the original
1149          * packet, which after processing serves as the first fragment.
1150          */
1151         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1152                 struct ip *mhip;        /* ip header on the fragment */
1153                 struct mbuf *m;
1154                 int mhlen = sizeof(struct ip);
1155
1156                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1157                 if (m == NULL) {
1158                         error = ENOBUFS;
1159                         ipstat.ips_odropped++;
1160                         goto done;
1161                 }
1162                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1163                 /*
1164                  * In the first mbuf, leave room for the link header, then
1165                  * copy the original IP header including options. The payload
1166                  * goes into an additional mbuf chain returned by m_copy().
1167                  */
1168                 m->m_data += max_linkhdr;
1169                 mhip = mtod(m, struct ip *);
1170                 *mhip = *ip;
1171                 if (hlen > sizeof(struct ip)) {
1172                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1173                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1174                 }
1175                 m->m_len = mhlen;
1176                 /* XXX do we need to add ip->ip_off below ? */
1177                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1178                 if (off + len >= ip->ip_len) {  /* last fragment */
1179                         len = ip->ip_len - off;
1180                         m->m_flags |= M_LASTFRAG;
1181                 } else
1182                         mhip->ip_off |= IP_MF;
1183                 mhip->ip_len = htons((u_short)(len + mhlen));
1184                 m->m_next = m_copy(m0, off, len);
1185                 if (m->m_next == NULL) {                /* copy failed */
1186                         m_free(m);
1187                         error = ENOBUFS;        /* ??? */
1188                         ipstat.ips_odropped++;
1189                         goto done;
1190                 }
1191                 m->m_pkthdr.len = mhlen + len;
1192                 m->m_pkthdr.rcvif = NULL;
1193                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1194                 mhip->ip_off = htons(mhip->ip_off);
1195                 mhip->ip_sum = 0;
1196                 if (sw_csum & CSUM_DELAY_IP)
1197                         mhip->ip_sum = in_cksum(m, mhlen);
1198                 *mnext = m;
1199                 mnext = &m->m_nextpkt;
1200         }
1201         ipstat.ips_ofragments += nfrags;
1202
1203         /* set first marker for fragment chain */
1204         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1205         m0->m_pkthdr.csum_data = nfrags;
1206
1207         /*
1208          * Update first fragment by trimming what's been copied out
1209          * and updating header.
1210          */
1211         m_adj(m0, hlen + firstlen - ip->ip_len);
1212         m0->m_pkthdr.len = hlen + firstlen;
1213         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1214         ip->ip_off |= IP_MF;
1215         ip->ip_off = htons(ip->ip_off);
1216         ip->ip_sum = 0;
1217         if (sw_csum & CSUM_DELAY_IP)
1218                 ip->ip_sum = in_cksum(m0, hlen);
1219
1220 done:
1221         *m_frag = m0;
1222         return error;
1223 }
1224
1225 void
1226 in_delayed_cksum(struct mbuf *m)
1227 {
1228         struct ip *ip;
1229         u_short csum, offset;
1230
1231         ip = mtod(m, struct ip *);
1232         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1233         csum = in_cksum_skip(m, ip->ip_len, offset);
1234         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1235                 csum = 0xffff;
1236         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1237
1238         if (offset + sizeof(u_short) > m->m_len) {
1239                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1240                     m->m_len, offset, ip->ip_p);
1241                 /*
1242                  * XXX
1243                  * this shouldn't happen, but if it does, the
1244                  * correct behavior may be to insert the checksum
1245                  * in the existing chain instead of rearranging it.
1246                  */
1247                 m = m_pullup(m, offset + sizeof(u_short));
1248         }
1249         *(u_short *)(m->m_data + offset) = csum;
1250 }
1251
1252 /*
1253  * Insert IP options into preformed packet.
1254  * Adjust IP destination as required for IP source routing,
1255  * as indicated by a non-zero in_addr at the start of the options.
1256  *
1257  * XXX This routine assumes that the packet has no options in place.
1258  */
1259 static struct mbuf *
1260 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1261 {
1262         struct ipoption *p = mtod(opt, struct ipoption *);
1263         struct mbuf *n;
1264         struct ip *ip = mtod(m, struct ip *);
1265         unsigned optlen;
1266
1267         optlen = opt->m_len - sizeof p->ipopt_dst;
1268         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1269                 *phlen = 0;
1270                 return (m);             /* XXX should fail */
1271         }
1272         if (p->ipopt_dst.s_addr)
1273                 ip->ip_dst = p->ipopt_dst;
1274         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1275                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1276                 if (n == NULL) {
1277                         *phlen = 0;
1278                         return (m);
1279                 }
1280                 n->m_pkthdr.rcvif = NULL;
1281                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1282                 m->m_len -= sizeof(struct ip);
1283                 m->m_data += sizeof(struct ip);
1284                 n->m_next = m;
1285                 m = n;
1286                 m->m_len = optlen + sizeof(struct ip);
1287                 m->m_data += max_linkhdr;
1288                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1289         } else {
1290                 m->m_data -= optlen;
1291                 m->m_len += optlen;
1292                 m->m_pkthdr.len += optlen;
1293                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1294         }
1295         ip = mtod(m, struct ip *);
1296         bcopy(p->ipopt_list, ip + 1, optlen);
1297         *phlen = sizeof(struct ip) + optlen;
1298         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1299         ip->ip_len += optlen;
1300         return (m);
1301 }
1302
1303 /*
1304  * Copy options from ip to jp,
1305  * omitting those not copied during fragmentation.
1306  */
1307 int
1308 ip_optcopy(struct ip *ip, struct ip *jp)
1309 {
1310         u_char *cp, *dp;
1311         int opt, optlen, cnt;
1312
1313         cp = (u_char *)(ip + 1);
1314         dp = (u_char *)(jp + 1);
1315         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1316         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1317                 opt = cp[0];
1318                 if (opt == IPOPT_EOL)
1319                         break;
1320                 if (opt == IPOPT_NOP) {
1321                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1322                         *dp++ = IPOPT_NOP;
1323                         optlen = 1;
1324                         continue;
1325                 }
1326
1327                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1328                     ("ip_optcopy: malformed ipv4 option"));
1329                 optlen = cp[IPOPT_OLEN];
1330                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1331                     ("ip_optcopy: malformed ipv4 option"));
1332
1333                 /* bogus lengths should have been caught by ip_dooptions */
1334                 if (optlen > cnt)
1335                         optlen = cnt;
1336                 if (IPOPT_COPIED(opt)) {
1337                         bcopy(cp, dp, optlen);
1338                         dp += optlen;
1339                 }
1340         }
1341         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1342                 *dp++ = IPOPT_EOL;
1343         return (optlen);
1344 }
1345
1346 /*
1347  * IP socket option processing.
1348  */
1349 int
1350 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1351 {
1352         struct  inpcb *inp = so->so_pcb;
1353         int     error, optval;
1354
1355         error = optval = 0;
1356         if (sopt->sopt_level != IPPROTO_IP) {
1357                 return (EINVAL);
1358         }
1359
1360         switch (sopt->sopt_dir) {
1361         case SOPT_SET:
1362                 switch (sopt->sopt_name) {
1363                 case IP_OPTIONS:
1364 #ifdef notyet
1365                 case IP_RETOPTS:
1366 #endif
1367                 {
1368                         struct mbuf *m;
1369                         if (sopt->sopt_valsize > MLEN) {
1370                                 error = EMSGSIZE;
1371                                 break;
1372                         }
1373                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1374                         if (m == NULL) {
1375                                 error = ENOBUFS;
1376                                 break;
1377                         }
1378                         m->m_len = sopt->sopt_valsize;
1379                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1380                                               m->m_len);
1381                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1382                                            m));
1383                 }
1384
1385                 case IP_TOS:
1386                 case IP_TTL:
1387                 case IP_MINTTL:
1388                 case IP_RECVOPTS:
1389                 case IP_RECVRETOPTS:
1390                 case IP_RECVDSTADDR:
1391                 case IP_RECVIF:
1392                 case IP_RECVTTL:
1393                 case IP_FAITH:
1394                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1395                                              sizeof optval);
1396                         if (error)
1397                                 break;
1398                         switch (sopt->sopt_name) {
1399                         case IP_TOS:
1400                                 inp->inp_ip_tos = optval;
1401                                 break;
1402
1403                         case IP_TTL:
1404                                 inp->inp_ip_ttl = optval;
1405                                 break;
1406                         case IP_MINTTL:
1407                                 if (optval >= 0 && optval <= MAXTTL)
1408                                         inp->inp_ip_minttl = optval;
1409                                 else
1410                                         error = EINVAL;
1411                                 break;
1412 #define OPTSET(bit) \
1413         if (optval) \
1414                 inp->inp_flags |= bit; \
1415         else \
1416                 inp->inp_flags &= ~bit;
1417
1418                         case IP_RECVOPTS:
1419                                 OPTSET(INP_RECVOPTS);
1420                                 break;
1421
1422                         case IP_RECVRETOPTS:
1423                                 OPTSET(INP_RECVRETOPTS);
1424                                 break;
1425
1426                         case IP_RECVDSTADDR:
1427                                 OPTSET(INP_RECVDSTADDR);
1428                                 break;
1429
1430                         case IP_RECVIF:
1431                                 OPTSET(INP_RECVIF);
1432                                 break;
1433
1434                         case IP_RECVTTL:
1435                                 OPTSET(INP_RECVTTL);
1436                                 break;
1437
1438                         case IP_FAITH:
1439                                 OPTSET(INP_FAITH);
1440                                 break;
1441                         }
1442                         break;
1443 #undef OPTSET
1444
1445                 case IP_MULTICAST_IF:
1446                 case IP_MULTICAST_VIF:
1447                 case IP_MULTICAST_TTL:
1448                 case IP_MULTICAST_LOOP:
1449                 case IP_ADD_MEMBERSHIP:
1450                 case IP_DROP_MEMBERSHIP:
1451                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1452                         break;
1453
1454                 case IP_PORTRANGE:
1455                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1456                                             sizeof optval);
1457                         if (error)
1458                                 break;
1459
1460                         switch (optval) {
1461                         case IP_PORTRANGE_DEFAULT:
1462                                 inp->inp_flags &= ~(INP_LOWPORT);
1463                                 inp->inp_flags &= ~(INP_HIGHPORT);
1464                                 break;
1465
1466                         case IP_PORTRANGE_HIGH:
1467                                 inp->inp_flags &= ~(INP_LOWPORT);
1468                                 inp->inp_flags |= INP_HIGHPORT;
1469                                 break;
1470
1471                         case IP_PORTRANGE_LOW:
1472                                 inp->inp_flags &= ~(INP_HIGHPORT);
1473                                 inp->inp_flags |= INP_LOWPORT;
1474                                 break;
1475
1476                         default:
1477                                 error = EINVAL;
1478                                 break;
1479                         }
1480                         break;
1481
1482 #if defined(IPSEC) || defined(FAST_IPSEC)
1483                 case IP_IPSEC_POLICY:
1484                 {
1485                         caddr_t req;
1486                         size_t len = 0;
1487                         int priv;
1488                         struct mbuf *m;
1489                         int optname;
1490
1491                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1492                                 break;
1493                         soopt_to_mbuf(sopt, m);
1494                         priv = (sopt->sopt_td != NULL &&
1495                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1496                         req = mtod(m, caddr_t);
1497                         len = m->m_len;
1498                         optname = sopt->sopt_name;
1499                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1500                         m_freem(m);
1501                         break;
1502                 }
1503 #endif /*IPSEC*/
1504
1505                 default:
1506                         error = ENOPROTOOPT;
1507                         break;
1508                 }
1509                 break;
1510
1511         case SOPT_GET:
1512                 switch (sopt->sopt_name) {
1513                 case IP_OPTIONS:
1514                 case IP_RETOPTS:
1515                         if (inp->inp_options)
1516                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1517                                                            char *),
1518                                                 inp->inp_options->m_len);
1519                         else
1520                                 sopt->sopt_valsize = 0;
1521                         break;
1522
1523                 case IP_TOS:
1524                 case IP_TTL:
1525                 case IP_MINTTL:
1526                 case IP_RECVOPTS:
1527                 case IP_RECVRETOPTS:
1528                 case IP_RECVDSTADDR:
1529                 case IP_RECVTTL:
1530                 case IP_RECVIF:
1531                 case IP_PORTRANGE:
1532                 case IP_FAITH:
1533                         switch (sopt->sopt_name) {
1534
1535                         case IP_TOS:
1536                                 optval = inp->inp_ip_tos;
1537                                 break;
1538
1539                         case IP_TTL:
1540                                 optval = inp->inp_ip_ttl;
1541                                 break;
1542                         case IP_MINTTL:
1543                                 optval = inp->inp_ip_minttl;
1544                                 break;
1545
1546 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1547
1548                         case IP_RECVOPTS:
1549                                 optval = OPTBIT(INP_RECVOPTS);
1550                                 break;
1551
1552                         case IP_RECVRETOPTS:
1553                                 optval = OPTBIT(INP_RECVRETOPTS);
1554                                 break;
1555
1556                         case IP_RECVDSTADDR:
1557                                 optval = OPTBIT(INP_RECVDSTADDR);
1558                                 break;
1559
1560                         case IP_RECVTTL:
1561                                 optval = OPTBIT(INP_RECVTTL);
1562                                 break;
1563
1564                         case IP_RECVIF:
1565                                 optval = OPTBIT(INP_RECVIF);
1566                                 break;
1567
1568                         case IP_PORTRANGE:
1569                                 if (inp->inp_flags & INP_HIGHPORT)
1570                                         optval = IP_PORTRANGE_HIGH;
1571                                 else if (inp->inp_flags & INP_LOWPORT)
1572                                         optval = IP_PORTRANGE_LOW;
1573                                 else
1574                                         optval = 0;
1575                                 break;
1576
1577                         case IP_FAITH:
1578                                 optval = OPTBIT(INP_FAITH);
1579                                 break;
1580                         }
1581                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1582                         break;
1583
1584                 case IP_MULTICAST_IF:
1585                 case IP_MULTICAST_VIF:
1586                 case IP_MULTICAST_TTL:
1587                 case IP_MULTICAST_LOOP:
1588                 case IP_ADD_MEMBERSHIP:
1589                 case IP_DROP_MEMBERSHIP:
1590                         error = ip_getmoptions(sopt, inp->inp_moptions);
1591                         break;
1592
1593 #if defined(IPSEC) || defined(FAST_IPSEC)
1594                 case IP_IPSEC_POLICY:
1595                 {
1596                         struct mbuf *m = NULL;
1597                         caddr_t req = NULL;
1598                         size_t len = 0;
1599
1600                         if (m != NULL) {
1601                                 req = mtod(m, caddr_t);
1602                                 len = m->m_len;
1603                         }
1604                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1605                         if (error == 0)
1606                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1607                         if (error == 0)
1608                                 m_freem(m);
1609                         break;
1610                 }
1611 #endif /*IPSEC*/
1612
1613                 default:
1614                         error = ENOPROTOOPT;
1615                         break;
1616                 }
1617                 break;
1618         }
1619         return (error);
1620 }
1621
1622 /*
1623  * Set up IP options in pcb for insertion in output packets.
1624  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1625  * with destination address if source routed.
1626  */
1627 static int
1628 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1629 {
1630         int cnt, optlen;
1631         u_char *cp;
1632         u_char opt;
1633
1634         /* turn off any old options */
1635         if (*pcbopt)
1636                 m_free(*pcbopt);
1637         *pcbopt = 0;
1638         if (m == NULL || m->m_len == 0) {
1639                 /*
1640                  * Only turning off any previous options.
1641                  */
1642                 if (m != NULL)
1643                         m_free(m);
1644                 return (0);
1645         }
1646
1647         if (m->m_len % sizeof(int32_t))
1648                 goto bad;
1649         /*
1650          * IP first-hop destination address will be stored before
1651          * actual options; move other options back
1652          * and clear it when none present.
1653          */
1654         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1655                 goto bad;
1656         cnt = m->m_len;
1657         m->m_len += sizeof(struct in_addr);
1658         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1659         ovbcopy(mtod(m, caddr_t), cp, cnt);
1660         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1661
1662         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1663                 opt = cp[IPOPT_OPTVAL];
1664                 if (opt == IPOPT_EOL)
1665                         break;
1666                 if (opt == IPOPT_NOP)
1667                         optlen = 1;
1668                 else {
1669                         if (cnt < IPOPT_OLEN + sizeof *cp)
1670                                 goto bad;
1671                         optlen = cp[IPOPT_OLEN];
1672                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1673                                 goto bad;
1674                 }
1675                 switch (opt) {
1676
1677                 default:
1678                         break;
1679
1680                 case IPOPT_LSRR:
1681                 case IPOPT_SSRR:
1682                         /*
1683                          * user process specifies route as:
1684                          *      ->A->B->C->D
1685                          * D must be our final destination (but we can't
1686                          * check that since we may not have connected yet).
1687                          * A is first hop destination, which doesn't appear in
1688                          * actual IP option, but is stored before the options.
1689                          */
1690                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1691                                 goto bad;
1692                         m->m_len -= sizeof(struct in_addr);
1693                         cnt -= sizeof(struct in_addr);
1694                         optlen -= sizeof(struct in_addr);
1695                         cp[IPOPT_OLEN] = optlen;
1696                         /*
1697                          * Move first hop before start of options.
1698                          */
1699                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1700                               sizeof(struct in_addr));
1701                         /*
1702                          * Then copy rest of options back
1703                          * to close up the deleted entry.
1704                          */
1705                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1706                                 &cp[IPOPT_OFFSET+1],
1707                                 cnt - (IPOPT_MINOFF - 1));
1708                         break;
1709                 }
1710         }
1711         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1712                 goto bad;
1713         *pcbopt = m;
1714         return (0);
1715
1716 bad:
1717         m_free(m);
1718         return (EINVAL);
1719 }
1720
1721 /*
1722  * XXX
1723  * The whole multicast option thing needs to be re-thought.
1724  * Several of these options are equally applicable to non-multicast
1725  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1726  * standard option (IP_TTL).
1727  */
1728
1729 /*
1730  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1731  */
1732 static struct ifnet *
1733 ip_multicast_if(struct in_addr *a, int *ifindexp)
1734 {
1735         int ifindex;
1736         struct ifnet *ifp;
1737
1738         if (ifindexp)
1739                 *ifindexp = 0;
1740         if (ntohl(a->s_addr) >> 24 == 0) {
1741                 ifindex = ntohl(a->s_addr) & 0xffffff;
1742                 if (ifindex < 0 || if_index < ifindex)
1743                         return NULL;
1744                 ifp = ifindex2ifnet[ifindex];
1745                 if (ifindexp)
1746                         *ifindexp = ifindex;
1747         } else {
1748                 ifp = INADDR_TO_IFP(a);
1749         }
1750         return ifp;
1751 }
1752
1753 /*
1754  * Set the IP multicast options in response to user setsockopt().
1755  */
1756 static int
1757 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1758 {
1759         int error = 0;
1760         int i;
1761         struct in_addr addr;
1762         struct ip_mreq mreq;
1763         struct ifnet *ifp;
1764         struct ip_moptions *imo = *imop;
1765         int ifindex;
1766
1767         if (imo == NULL) {
1768                 /*
1769                  * No multicast option buffer attached to the pcb;
1770                  * allocate one and initialize to default values.
1771                  */
1772                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1773
1774                 *imop = imo;
1775                 imo->imo_multicast_ifp = NULL;
1776                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1777                 imo->imo_multicast_vif = -1;
1778                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1779                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1780                 imo->imo_num_memberships = 0;
1781         }
1782         switch (sopt->sopt_name) {
1783         /* store an index number for the vif you wanna use in the send */
1784         case IP_MULTICAST_VIF:
1785                 if (legal_vif_num == 0) {
1786                         error = EOPNOTSUPP;
1787                         break;
1788                 }
1789                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1790                 if (error)
1791                         break;
1792                 if (!legal_vif_num(i) && (i != -1)) {
1793                         error = EINVAL;
1794                         break;
1795                 }
1796                 imo->imo_multicast_vif = i;
1797                 break;
1798
1799         case IP_MULTICAST_IF:
1800                 /*
1801                  * Select the interface for outgoing multicast packets.
1802                  */
1803                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1804                 if (error)
1805                         break;
1806
1807                 /*
1808                  * INADDR_ANY is used to remove a previous selection.
1809                  * When no interface is selected, a default one is
1810                  * chosen every time a multicast packet is sent.
1811                  */
1812                 if (addr.s_addr == INADDR_ANY) {
1813                         imo->imo_multicast_ifp = NULL;
1814                         break;
1815                 }
1816                 /*
1817                  * The selected interface is identified by its local
1818                  * IP address.  Find the interface and confirm that
1819                  * it supports multicasting.
1820                  */
1821                 crit_enter();
1822                 ifp = ip_multicast_if(&addr, &ifindex);
1823                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1824                         crit_exit();
1825                         error = EADDRNOTAVAIL;
1826                         break;
1827                 }
1828                 imo->imo_multicast_ifp = ifp;
1829                 if (ifindex)
1830                         imo->imo_multicast_addr = addr;
1831                 else
1832                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1833                 crit_exit();
1834                 break;
1835
1836         case IP_MULTICAST_TTL:
1837                 /*
1838                  * Set the IP time-to-live for outgoing multicast packets.
1839                  * The original multicast API required a char argument,
1840                  * which is inconsistent with the rest of the socket API.
1841                  * We allow either a char or an int.
1842                  */
1843                 if (sopt->sopt_valsize == 1) {
1844                         u_char ttl;
1845                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1846                         if (error)
1847                                 break;
1848                         imo->imo_multicast_ttl = ttl;
1849                 } else {
1850                         u_int ttl;
1851                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1852                         if (error)
1853                                 break;
1854                         if (ttl > 255)
1855                                 error = EINVAL;
1856                         else
1857                                 imo->imo_multicast_ttl = ttl;
1858                 }
1859                 break;
1860
1861         case IP_MULTICAST_LOOP:
1862                 /*
1863                  * Set the loopback flag for outgoing multicast packets.
1864                  * Must be zero or one.  The original multicast API required a
1865                  * char argument, which is inconsistent with the rest
1866                  * of the socket API.  We allow either a char or an int.
1867                  */
1868                 if (sopt->sopt_valsize == 1) {
1869                         u_char loop;
1870
1871                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1872                         if (error)
1873                                 break;
1874                         imo->imo_multicast_loop = !!loop;
1875                 } else {
1876                         u_int loop;
1877
1878                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1879                                             sizeof loop);
1880                         if (error)
1881                                 break;
1882                         imo->imo_multicast_loop = !!loop;
1883                 }
1884                 break;
1885
1886         case IP_ADD_MEMBERSHIP:
1887                 /*
1888                  * Add a multicast group membership.
1889                  * Group must be a valid IP multicast address.
1890                  */
1891                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1892                 if (error)
1893                         break;
1894
1895                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1896                         error = EINVAL;
1897                         break;
1898                 }
1899                 crit_enter();
1900                 /*
1901                  * If no interface address was provided, use the interface of
1902                  * the route to the given multicast address.
1903                  */
1904                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1905                         struct sockaddr_in dst;
1906                         struct rtentry *rt;
1907
1908                         bzero(&dst, sizeof(struct sockaddr_in));
1909                         dst.sin_len = sizeof(struct sockaddr_in);
1910                         dst.sin_family = AF_INET;
1911                         dst.sin_addr = mreq.imr_multiaddr;
1912                         rt = rtlookup((struct sockaddr *)&dst);
1913                         if (rt == NULL) {
1914                                 error = EADDRNOTAVAIL;
1915                                 crit_exit();
1916                                 break;
1917                         }
1918                         --rt->rt_refcnt;
1919                         ifp = rt->rt_ifp;
1920                 } else {
1921                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1922                 }
1923
1924                 /*
1925                  * See if we found an interface, and confirm that it
1926                  * supports multicast.
1927                  */
1928                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1929                         error = EADDRNOTAVAIL;
1930                         crit_exit();
1931                         break;
1932                 }
1933                 /*
1934                  * See if the membership already exists or if all the
1935                  * membership slots are full.
1936                  */
1937                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1938                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1939                             imo->imo_membership[i]->inm_addr.s_addr
1940                                                 == mreq.imr_multiaddr.s_addr)
1941                                 break;
1942                 }
1943                 if (i < imo->imo_num_memberships) {
1944                         error = EADDRINUSE;
1945                         crit_exit();
1946                         break;
1947                 }
1948                 if (i == IP_MAX_MEMBERSHIPS) {
1949                         error = ETOOMANYREFS;
1950                         crit_exit();
1951                         break;
1952                 }
1953                 /*
1954                  * Everything looks good; add a new record to the multicast
1955                  * address list for the given interface.
1956                  */
1957                 if ((imo->imo_membership[i] =
1958                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1959                         error = ENOBUFS;
1960                         crit_exit();
1961                         break;
1962                 }
1963                 ++imo->imo_num_memberships;
1964                 crit_exit();
1965                 break;
1966
1967         case IP_DROP_MEMBERSHIP:
1968                 /*
1969                  * Drop a multicast group membership.
1970                  * Group must be a valid IP multicast address.
1971                  */
1972                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1973                 if (error)
1974                         break;
1975
1976                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1977                         error = EINVAL;
1978                         break;
1979                 }
1980
1981                 crit_enter();
1982                 /*
1983                  * If an interface address was specified, get a pointer
1984                  * to its ifnet structure.
1985                  */
1986                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1987                         ifp = NULL;
1988                 else {
1989                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1990                         if (ifp == NULL) {
1991                                 error = EADDRNOTAVAIL;
1992                                 crit_exit();
1993                                 break;
1994                         }
1995                 }
1996                 /*
1997                  * Find the membership in the membership array.
1998                  */
1999                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2000                         if ((ifp == NULL ||
2001                              imo->imo_membership[i]->inm_ifp == ifp) &&
2002                             imo->imo_membership[i]->inm_addr.s_addr ==
2003                             mreq.imr_multiaddr.s_addr)
2004                                 break;
2005                 }
2006                 if (i == imo->imo_num_memberships) {
2007                         error = EADDRNOTAVAIL;
2008                         crit_exit();
2009                         break;
2010                 }
2011                 /*
2012                  * Give up the multicast address record to which the
2013                  * membership points.
2014                  */
2015                 in_delmulti(imo->imo_membership[i]);
2016                 /*
2017                  * Remove the gap in the membership array.
2018                  */
2019                 for (++i; i < imo->imo_num_memberships; ++i)
2020                         imo->imo_membership[i-1] = imo->imo_membership[i];
2021                 --imo->imo_num_memberships;
2022                 crit_exit();
2023                 break;
2024
2025         default:
2026                 error = EOPNOTSUPP;
2027                 break;
2028         }
2029
2030         /*
2031          * If all options have default values, no need to keep the mbuf.
2032          */
2033         if (imo->imo_multicast_ifp == NULL &&
2034             imo->imo_multicast_vif == -1 &&
2035             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2036             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2037             imo->imo_num_memberships == 0) {
2038                 kfree(*imop, M_IPMOPTS);
2039                 *imop = NULL;
2040         }
2041
2042         return (error);
2043 }
2044
2045 /*
2046  * Return the IP multicast options in response to user getsockopt().
2047  */
2048 static int
2049 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2050 {
2051         struct in_addr addr;
2052         struct in_ifaddr *ia;
2053         int error, optval;
2054         u_char coptval;
2055
2056         error = 0;
2057         switch (sopt->sopt_name) {
2058         case IP_MULTICAST_VIF:
2059                 if (imo != NULL)
2060                         optval = imo->imo_multicast_vif;
2061                 else
2062                         optval = -1;
2063                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2064                 break;
2065
2066         case IP_MULTICAST_IF:
2067                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2068                         addr.s_addr = INADDR_ANY;
2069                 else if (imo->imo_multicast_addr.s_addr) {
2070                         /* return the value user has set */
2071                         addr = imo->imo_multicast_addr;
2072                 } else {
2073                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2074                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2075                                 : IA_SIN(ia)->sin_addr.s_addr;
2076                 }
2077                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2078                 break;
2079
2080         case IP_MULTICAST_TTL:
2081                 if (imo == NULL)
2082                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2083                 else
2084                         optval = coptval = imo->imo_multicast_ttl;
2085                 if (sopt->sopt_valsize == 1)
2086                         soopt_from_kbuf(sopt, &coptval, 1);
2087                 else
2088                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2089                 break;
2090
2091         case IP_MULTICAST_LOOP:
2092                 if (imo == NULL)
2093                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2094                 else
2095                         optval = coptval = imo->imo_multicast_loop;
2096                 if (sopt->sopt_valsize == 1)
2097                         soopt_from_kbuf(sopt, &coptval, 1);
2098                 else
2099                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2100                 break;
2101
2102         default:
2103                 error = ENOPROTOOPT;
2104                 break;
2105         }
2106         return (error);
2107 }
2108
2109 /*
2110  * Discard the IP multicast options.
2111  */
2112 void
2113 ip_freemoptions(struct ip_moptions *imo)
2114 {
2115         int i;
2116
2117         if (imo != NULL) {
2118                 for (i = 0; i < imo->imo_num_memberships; ++i)
2119                         in_delmulti(imo->imo_membership[i]);
2120                 kfree(imo, M_IPMOPTS);
2121         }
2122 }
2123
2124 /*
2125  * Routine called from ip_output() to loop back a copy of an IP multicast
2126  * packet to the input queue of a specified interface.  Note that this
2127  * calls the output routine of the loopback "driver", but with an interface
2128  * pointer that might NOT be a loopback interface -- evil, but easier than
2129  * replicating that code here.
2130  */
2131 static void
2132 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2133              int hlen)
2134 {
2135         struct ip *ip;
2136         struct mbuf *copym;
2137
2138         copym = m_copypacket(m, MB_DONTWAIT);
2139         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2140                 copym = m_pullup(copym, hlen);
2141         if (copym != NULL) {
2142                 /*
2143                  * if the checksum hasn't been computed, mark it as valid
2144                  */
2145                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2146                         in_delayed_cksum(copym);
2147                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2148                         copym->m_pkthdr.csum_flags |=
2149                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2150                         copym->m_pkthdr.csum_data = 0xffff;
2151                 }
2152                 /*
2153                  * We don't bother to fragment if the IP length is greater
2154                  * than the interface's MTU.  Can this possibly matter?
2155                  */
2156                 ip = mtod(copym, struct ip *);
2157                 ip->ip_len = htons(ip->ip_len);
2158                 ip->ip_off = htons(ip->ip_off);
2159                 ip->ip_sum = 0;
2160                 if (ip->ip_vhl == IP_VHL_BORING) {
2161                         ip->ip_sum = in_cksum_hdr(ip);
2162                 } else {
2163                         ip->ip_sum = in_cksum(copym, hlen);
2164                 }
2165                 /*
2166                  * NB:
2167                  * It's not clear whether there are any lingering
2168                  * reentrancy problems in other areas which might
2169                  * be exposed by using ip_input directly (in
2170                  * particular, everything which modifies the packet
2171                  * in-place).  Yet another option is using the
2172                  * protosw directly to deliver the looped back
2173                  * packet.  For the moment, we'll err on the side
2174                  * of safety by using if_simloop().
2175                  */
2176 #if 1 /* XXX */
2177                 if (dst->sin_family != AF_INET) {
2178                         kprintf("ip_mloopback: bad address family %d\n",
2179                                                 dst->sin_family);
2180                         dst->sin_family = AF_INET;
2181                 }
2182 #endif
2183                 if_simloop(ifp, copym, dst->sin_family, 0);
2184         }
2185 }