sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  */
  32
  33 #define _IP_VHL
  34
  35 #include "opt_ipdn.h"
  36 #include "opt_ipdivert.h"
  37 #include "opt_ipsec.h"
  38 #include "opt_mbuf_stress_test.h"
  39 #include "opt_mpls.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/kernel.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mbuf.h>
  46 #include <sys/protosw.h>
  47 #include <sys/socket.h>
  48 #include <sys/socketvar.h>
  49 #include <sys/proc.h>
  50 #include <sys/priv.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/in_cksum.h>
  53 #include <sys/lock.h>
  54
  55 #include <sys/thread2.h>
  56 #include <sys/mplock2.h>
  57 #include <sys/msgport2.h>
  58
  59 #include <net/if.h>
  60 #include <net/netisr.h>
  61 #include <net/pfil.h>
  62 #include <net/route.h>
  63
  64 #include <netinet/in.h>
  65 #include <netinet/in_systm.h>
  66 #include <netinet/ip.h>
  67 #include <netinet/in_pcb.h>
  68 #include <netinet/in_var.h>
  69 #include <netinet/ip_var.h>
  70
  71 #include <netproto/mpls/mpls_var.h>
  72
  73 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  74
  75 #ifdef IPSEC
  76 #include <netinet6/ipsec.h>
  77 #include <netproto/key/key.h>
  78 #ifdef IPSEC_DEBUG
  79 #include <netproto/key/key_debug.h>
  80 #else
  81 #define KEYDEBUG(lev,arg)
  82 #endif
  83 #endif /*IPSEC*/
  84
  85 #ifdef FAST_IPSEC
  86 #include <netproto/ipsec/ipsec.h>
  87 #include <netproto/ipsec/xform.h>
  88 #include <netproto/ipsec/key.h>
  89 #endif /*FAST_IPSEC*/
  90
  91 #include <net/ipfw/ip_fw.h>
  92 #include <net/dummynet/ip_dummynet.h>
  93
  94 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  95                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  96                                   (ntohl(a.s_addr)>>16)&0xFF,\
  97                                   (ntohl(a.s_addr)>>8)&0xFF,\
  98                                   (ntohl(a.s_addr))&0xFF, y);
  99
 100 u_short ip_id;
 101
 102 #ifdef MBUF_STRESS_TEST
 103 int mbuf_frag_size = 0;
 104 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 105         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 106 #endif
 107
 108 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 109 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 110 static void     ip_mloopback
 111         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 112 static int      ip_getmoptions
 113         (struct sockopt *, struct ip_moptions *);
 114 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 115 static int      ip_setmoptions
 116         (struct sockopt *, struct ip_moptions **);
 117
 118 int     ip_optcopy(struct ip *, struct ip *);
 119
 120 extern  int route_assert_owner_access;
 121
 122 extern  struct protosw inetsw[];
 123
 124 static int
 125 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
 126 {
 127         struct in_ifaddr_container *iac;
 128
 129         /*
 130          * We need to figure out if we have been forwarded to a local
 131          * socket.  If so, then we should somehow "loop back" to
 132          * ip_input(), and get directed to the PCB as if we had received
 133          * this packet.  This is because it may be difficult to identify
 134          * the packets you want to forward until they are being output
 135          * and have selected an interface (e.g. locally initiated
 136          * packets).  If we used the loopback inteface, we would not be
 137          * able to control what happens as the packet runs through
 138          * ip_input() as it is done through a ISR.
 139          */
 140         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 141                 /*
 142                  * If the addr to forward to is one of ours, we pretend
 143                  * to be the destination for this packet.
 144                  */
 145                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 146                         break;
 147         }
 148         if (iac != NULL) {
 149                 struct ip *ip;
 150
 151                 if (m->m_pkthdr.rcvif == NULL)
 152                         m->m_pkthdr.rcvif = ifunit("lo0");
 153                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 154                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 155                                                   CSUM_PSEUDO_HDR;
 156                         m->m_pkthdr.csum_data = 0xffff;
 157                 }
 158                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 159
 160                 /*
 161                  * Make sure that the IP header is in one mbuf,
 162                  * required by ip_input
 163                  */
 164                 if (m->m_len < hlen) {
 165                         m = m_pullup(m, hlen);
 166                         if (m == NULL) {
 167                                 /* The packet was freed; we are done */
 168                                 return 1;
 169                         }
 170                 }
 171                 ip = mtod(m, struct ip *);
 172
 173                 ip->ip_len = htons(ip->ip_len);
 174                 ip->ip_off = htons(ip->ip_off);
 175                 ip_input(m);
 176
 177                 return 1; /* The packet gets forwarded locally */
 178         }
 179         return 0;
 180 }
 181
 182 /*
 183  * IP output.  The packet in mbuf chain m contains a skeletal IP
 184  * header (with len, off, ttl, proto, tos, src, dst).
 185  * The mbuf chain containing the packet will be freed.
 186  * The mbuf opt, if present, will not be freed.
 187  */
 188 int
 189 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 190           int flags, struct ip_moptions *imo, struct inpcb *inp)
 191 {
 192         struct ip *ip;
 193         struct ifnet *ifp = NULL;       /* keep compiler happy */
 194         struct mbuf *m;
 195         int hlen = sizeof(struct ip);
 196         int len, error = 0;
 197         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 198         struct in_ifaddr *ia = NULL;
 199         int isbroadcast, sw_csum;
 200         struct in_addr pkt_dst;
 201         struct route iproute;
 202         struct m_tag *mtag;
 203 #ifdef IPSEC
 204         struct secpolicy *sp = NULL;
 205         struct socket *so = inp ? inp->inp_socket : NULL;
 206 #endif
 207 #ifdef FAST_IPSEC
 208         struct secpolicy *sp = NULL;
 209         struct tdb_ident *tdbi;
 210 #endif /* FAST_IPSEC */
 211         struct sockaddr_in *next_hop = NULL;
 212         int src_was_INADDR_ANY = 0;     /* as the name says... */
 213
 214         m = m0;
 215         M_ASSERTPKTHDR(m);
 216
 217         if (ro == NULL) {
 218                 ro = &iproute;
 219                 bzero(ro, sizeof *ro);
 220         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 221                 if (flags & IP_DEBUGROUTE) {
 222                         if (route_assert_owner_access) {
 223                                 panic("ip_output: "
 224                                       "rt rt_cpuid %d accessed on cpu %d\n",
 225                                       ro->ro_rt->rt_cpuid, mycpuid);
 226                         } else {
 227                                 kprintf("ip_output: "
 228                                         "rt rt_cpuid %d accessed on cpu %d\n",
 229                                         ro->ro_rt->rt_cpuid, mycpuid);
 230                                 print_backtrace(-1);
 231                         }
 232                 }
 233
 234                 /*
 235                  * XXX
 236                  * If the cached rtentry's owner CPU is not the current CPU,
 237                  * then don't touch the cached rtentry (remote free is too
 238                  * expensive in this context); just relocate the route.
 239                  */
 240                 ro = &iproute;
 241                 bzero(ro, sizeof *ro);
 242         }
 243
 244         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 245                 /* Next hop */
 246                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 247                 KKASSERT(mtag != NULL);
 248                 next_hop = m_tag_data(mtag);
 249         }
 250
 251         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 252                 struct dn_pkt *dn_pkt;
 253
 254                 /* Extract info from dummynet tag */
 255                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 256                 KKASSERT(mtag != NULL);
 257                 dn_pkt = m_tag_data(mtag);
 258
 259                 /*
 260                  * The packet was already tagged, so part of the
 261                  * processing was already done, and we need to go down.
 262                  * Get the calculated parameters from the tag.
 263                  */
 264                 ifp = dn_pkt->ifp;
 265
 266                 KKASSERT(ro == &iproute);
 267                 *ro = dn_pkt->ro; /* structure copy */
 268                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 269
 270                 dst = dn_pkt->dn_dst;
 271                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 272                         /* If 'dst' points into dummynet tag, adjust it */
 273                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 274                 }
 275
 276                 ip = mtod(m, struct ip *);
 277                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 278                 if (ro->ro_rt)
 279                         ia = ifatoia(ro->ro_rt->rt_ifa);
 280                 goto sendit;
 281         }
 282
 283         if (opt) {
 284                 len = 0;
 285                 m = ip_insertoptions(m, opt, &len);
 286                 if (len != 0)
 287                         hlen = len;
 288         }
 289         ip = mtod(m, struct ip *);
 290
 291         /*
 292          * Fill in IP header.
 293          */
 294         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 295                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 296                 ip->ip_off &= IP_DF;
 297                 ip->ip_id = ip_newid();
 298                 ipstat.ips_localout++;
 299         } else {
 300                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 301         }
 302
 303 reroute:
 304         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 305
 306         dst = (struct sockaddr_in *)&ro->ro_dst;
 307         /*
 308          * If there is a cached route,
 309          * check that it is to the same destination
 310          * and is still up.  If not, free it and try again.
 311          * The address family should also be checked in case of sharing the
 312          * cache with IPv6.
 313          */
 314         if (ro->ro_rt &&
 315             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 316              dst->sin_family != AF_INET ||
 317              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 318                 rtfree(ro->ro_rt);
 319                 ro->ro_rt = NULL;
 320         }
 321         if (ro->ro_rt == NULL) {
 322                 bzero(dst, sizeof *dst);
 323                 dst->sin_family = AF_INET;
 324                 dst->sin_len = sizeof *dst;
 325                 dst->sin_addr = pkt_dst;
 326         }
 327         /*
 328          * If routing to interface only,
 329          * short circuit routing lookup.
 330          */
 331         if (flags & IP_ROUTETOIF) {
 332                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 333                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 334                         ipstat.ips_noroute++;
 335                         error = ENETUNREACH;
 336                         goto bad;
 337                 }
 338                 ifp = ia->ia_ifp;
 339                 ip->ip_ttl = 1;
 340                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 341         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 342                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 343                 /*
 344                  * Bypass the normal routing lookup for multicast
 345                  * packets if the interface is specified.
 346                  */
 347                 ifp = imo->imo_multicast_ifp;
 348                 ia = IFP_TO_IA(ifp);
 349                 isbroadcast = 0;        /* fool gcc */
 350         } else {
 351                 /*
 352                  * If this is the case, we probably don't want to allocate
 353                  * a protocol-cloned route since we didn't get one from the
 354                  * ULP.  This lets TCP do its thing, while not burdening
 355                  * forwarding or ICMP with the overhead of cloning a route.
 356                  * Of course, we still want to do any cloning requested by
 357                  * the link layer, as this is probably required in all cases
 358                  * for correct operation (as it is for ARP).
 359                  */
 360                 if (ro->ro_rt == NULL)
 361                         rtalloc_ign(ro, RTF_PRCLONING);
 362                 if (ro->ro_rt == NULL) {
 363                         ipstat.ips_noroute++;
 364                         error = EHOSTUNREACH;
 365                         goto bad;
 366                 }
 367                 ia = ifatoia(ro->ro_rt->rt_ifa);
 368                 ifp = ro->ro_rt->rt_ifp;
 369                 ro->ro_rt->rt_use++;
 370                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 371                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 372                 if (ro->ro_rt->rt_flags & RTF_HOST)
 373                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 374                 else
 375                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 376         }
 377         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 378                 struct in_multi *inm;
 379
 380                 m->m_flags |= M_MCAST;
 381                 /*
 382                  * IP destination address is multicast.  Make sure "dst"
 383                  * still points to the address in "ro".  (It may have been
 384                  * changed to point to a gateway address, above.)
 385                  */
 386                 dst = (struct sockaddr_in *)&ro->ro_dst;
 387                 /*
 388                  * See if the caller provided any multicast options
 389                  */
 390                 if (imo != NULL) {
 391                         ip->ip_ttl = imo->imo_multicast_ttl;
 392                         if (imo->imo_multicast_vif != -1) {
 393                                 ip->ip_src.s_addr =
 394                                     ip_mcast_src ?
 395                                     ip_mcast_src(imo->imo_multicast_vif) :
 396                                     INADDR_ANY;
 397                         }
 398                 } else {
 399                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 400                 }
 401                 /*
 402                  * Confirm that the outgoing interface supports multicast.
 403                  */
 404                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 405                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 406                                 ipstat.ips_noroute++;
 407                                 error = ENETUNREACH;
 408                                 goto bad;
 409                         }
 410                 }
 411                 /*
 412                  * If source address not specified yet, use address
 413                  * of outgoing interface.
 414                  */
 415                 if (ip->ip_src.s_addr == INADDR_ANY) {
 416                         /* Interface may have no addresses. */
 417                         if (ia != NULL)
 418                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 419                 }
 420
 421                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 422                 if (inm != NULL &&
 423                     (imo == NULL || imo->imo_multicast_loop)) {
 424                         /*
 425                          * If we belong to the destination multicast group
 426                          * on the outgoing interface, and the caller did not
 427                          * forbid loopback, loop back a copy.
 428                          */
 429                         ip_mloopback(ifp, m, dst, hlen);
 430                 } else {
 431                         /*
 432                          * If we are acting as a multicast router, perform
 433                          * multicast forwarding as if the packet had just
 434                          * arrived on the interface to which we are about
 435                          * to send.  The multicast forwarding function
 436                          * recursively calls this function, using the
 437                          * IP_FORWARDING flag to prevent infinite recursion.
 438                          *
 439                          * Multicasts that are looped back by ip_mloopback(),
 440                          * above, will be forwarded by the ip_input() routine,
 441                          * if necessary.
 442                          */
 443                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 444                                 /*
 445                                  * If rsvp daemon is not running, do not
 446                                  * set ip_moptions. This ensures that the packet
 447                                  * is multicast and not just sent down one link
 448                                  * as prescribed by rsvpd.
 449                                  */
 450                                 if (!rsvp_on)
 451                                         imo = NULL;
 452                                 if (ip_mforward) {
 453                                         get_mplock();
 454                                         if (ip_mforward(ip, ifp, m, imo) != 0) {
 455                                                 m_freem(m);
 456                                                 rel_mplock();
 457                                                 goto done;
 458                                         }
 459                                         rel_mplock();
 460                                 }
 461                         }
 462                 }
 463
 464                 /*
 465                  * Multicasts with a time-to-live of zero may be looped-
 466                  * back, above, but must not be transmitted on a network.
 467                  * Also, multicasts addressed to the loopback interface
 468                  * are not sent -- the above call to ip_mloopback() will
 469                  * loop back a copy if this host actually belongs to the
 470                  * destination group on the loopback interface.
 471                  */
 472                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 473                         m_freem(m);
 474                         goto done;
 475                 }
 476
 477                 goto sendit;
 478         } else {
 479                 m->m_flags &= ~M_MCAST;
 480         }
 481
 482         /*
 483          * If the source address is not specified yet, use the address
 484          * of the outoing interface. In case, keep note we did that, so
 485          * if the the firewall changes the next-hop causing the output
 486          * interface to change, we can fix that.
 487          */
 488         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 489                 /* Interface may have no addresses. */
 490                 if (ia != NULL) {
 491                         ip->ip_src = IA_SIN(ia)->sin_addr;
 492                         src_was_INADDR_ANY = 1;
 493                 }
 494         }
 495
 496 #ifdef ALTQ
 497         /*
 498          * Disable packet drop hack.
 499          * Packetdrop should be done by queueing.
 500          */
 501 #else /* !ALTQ */
 502         /*
 503          * Verify that we have any chance at all of being able to queue
 504          *      the packet or packet fragments
 505          */
 506         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 507             ifp->if_snd.ifq_maxlen) {
 508                 error = ENOBUFS;
 509                 ipstat.ips_odropped++;
 510                 goto bad;
 511         }
 512 #endif /* !ALTQ */
 513
 514         /*
 515          * Look for broadcast address and
 516          * verify user is allowed to send
 517          * such a packet.
 518          */
 519         if (isbroadcast) {
 520                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 521                         error = EADDRNOTAVAIL;
 522                         goto bad;
 523                 }
 524                 if (!(flags & IP_ALLOWBROADCAST)) {
 525                         error = EACCES;
 526                         goto bad;
 527                 }
 528                 /* don't allow broadcast messages to be fragmented */
 529                 if (ip->ip_len > ifp->if_mtu) {
 530                         error = EMSGSIZE;
 531                         goto bad;
 532                 }
 533                 m->m_flags |= M_BCAST;
 534         } else {
 535                 m->m_flags &= ~M_BCAST;
 536         }
 537
 538 sendit:
 539 #ifdef IPSEC
 540         /* get SP for this packet */
 541         if (so == NULL)
 542                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 543         else
 544                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 545
 546         if (sp == NULL) {
 547                 ipsecstat.out_inval++;
 548                 goto bad;
 549         }
 550
 551         error = 0;
 552
 553         /* check policy */
 554         switch (sp->policy) {
 555         case IPSEC_POLICY_DISCARD:
 556                 /*
 557                  * This packet is just discarded.
 558                  */
 559                 ipsecstat.out_polvio++;
 560                 goto bad;
 561
 562         case IPSEC_POLICY_BYPASS:
 563         case IPSEC_POLICY_NONE:
 564         case IPSEC_POLICY_TCP:
 565                 /* no need to do IPsec. */
 566                 goto skip_ipsec;
 567
 568         case IPSEC_POLICY_IPSEC:
 569                 if (sp->req == NULL) {
 570                         /* acquire a policy */
 571                         error = key_spdacquire(sp);
 572                         goto bad;
 573                 }
 574                 break;
 575
 576         case IPSEC_POLICY_ENTRUST:
 577         default:
 578                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 579         }
 580     {
 581         struct ipsec_output_state state;
 582         bzero(&state, sizeof state);
 583         state.m = m;
 584         if (flags & IP_ROUTETOIF) {
 585                 state.ro = &iproute;
 586                 bzero(&iproute, sizeof iproute);
 587         } else
 588                 state.ro = ro;
 589         state.dst = (struct sockaddr *)dst;
 590
 591         ip->ip_sum = 0;
 592
 593         /*
 594          * XXX
 595          * delayed checksums are not currently compatible with IPsec
 596          */
 597         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 598                 in_delayed_cksum(m);
 599                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 600         }
 601
 602         ip->ip_len = htons(ip->ip_len);
 603         ip->ip_off = htons(ip->ip_off);
 604
 605         error = ipsec4_output(&state, sp, flags);
 606
 607         m = state.m;
 608         if (flags & IP_ROUTETOIF) {
 609                 /*
 610                  * if we have tunnel mode SA, we may need to ignore
 611                  * IP_ROUTETOIF.
 612                  */
 613                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 614                         flags &= ~IP_ROUTETOIF;
 615                         ro = state.ro;
 616                 }
 617         } else
 618                 ro = state.ro;
 619         dst = (struct sockaddr_in *)state.dst;
 620         if (error) {
 621                 /* mbuf is already reclaimed in ipsec4_output. */
 622                 m0 = NULL;
 623                 switch (error) {
 624                 case EHOSTUNREACH:
 625                 case ENETUNREACH:
 626                 case EMSGSIZE:
 627                 case ENOBUFS:
 628                 case ENOMEM:
 629                         break;
 630                 default:
 631                         kprintf("ip4_output (ipsec): error code %d\n", error);
 632                         /*fall through*/
 633                 case ENOENT:
 634                         /* don't show these error codes to the user */
 635                         error = 0;
 636                         break;
 637                 }
 638                 goto bad;
 639         }
 640     }
 641
 642         /* be sure to update variables that are affected by ipsec4_output() */
 643         ip = mtod(m, struct ip *);
 644 #ifdef _IP_VHL
 645         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 646 #else
 647         hlen = ip->ip_hl << 2;
 648 #endif
 649         if (ro->ro_rt == NULL) {
 650                 if (!(flags & IP_ROUTETOIF)) {
 651                         kprintf("ip_output: "
 652                                 "can't update route after IPsec processing\n");
 653                         error = EHOSTUNREACH;   /*XXX*/
 654                         goto bad;
 655                 }
 656         } else {
 657                 ia = ifatoia(ro->ro_rt->rt_ifa);
 658                 ifp = ro->ro_rt->rt_ifp;
 659         }
 660
 661         /* make it flipped, again. */
 662         ip->ip_len = ntohs(ip->ip_len);
 663         ip->ip_off = ntohs(ip->ip_off);
 664 skip_ipsec:
 665 #endif /*IPSEC*/
 666 #ifdef FAST_IPSEC
 667         /*
 668          * Check the security policy (SP) for the packet and, if
 669          * required, do IPsec-related processing.  There are two
 670          * cases here; the first time a packet is sent through
 671          * it will be untagged and handled by ipsec4_checkpolicy.
 672          * If the packet is resubmitted to ip_output (e.g. after
 673          * AH, ESP, etc. processing), there will be a tag to bypass
 674          * the lookup and related policy checking.
 675          */
 676         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 677         crit_enter();
 678         if (mtag != NULL) {
 679                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 680                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 681                 if (sp == NULL)
 682                         error = -EINVAL;        /* force silent drop */
 683                 m_tag_delete(m, mtag);
 684         } else {
 685                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 686                                         &error, inp);
 687         }
 688         /*
 689          * There are four return cases:
 690          *    sp != NULL                    apply IPsec policy
 691          *    sp == NULL, error == 0        no IPsec handling needed
 692          *    sp == NULL, error == -EINVAL  discard packet w/o error
 693          *    sp == NULL, error != 0        discard packet, report error
 694          */
 695         if (sp != NULL) {
 696                 /* Loop detection, check if ipsec processing already done */
 697                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 698                 for (mtag = m_tag_first(m); mtag != NULL;
 699                      mtag = m_tag_next(m, mtag)) {
 700                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 701                                 continue;
 702                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 703                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 704                                 continue;
 705                         /*
 706                          * Check if policy has an SA associated with it.
 707                          * This can happen when an SP has yet to acquire
 708                          * an SA; e.g. on first reference.  If it occurs,
 709                          * then we let ipsec4_process_packet do its thing.
 710                          */
 711                         if (sp->req->sav == NULL)
 712                                 break;
 713                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 714                         if (tdbi->spi == sp->req->sav->spi &&
 715                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 716                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 717                                  sizeof(union sockaddr_union)) == 0) {
 718                                 /*
 719                                  * No IPsec processing is needed, free
 720                                  * reference to SP.
 721                                  *
 722                                  * NB: null pointer to avoid free at
 723                                  *     done: below.
 724                                  */
 725                                 KEY_FREESP(&sp), sp = NULL;
 726                                 crit_exit();
 727                                 goto spd_done;
 728                         }
 729                 }
 730
 731                 /*
 732                  * Do delayed checksums now because we send before
 733                  * this is done in the normal processing path.
 734                  */
 735                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 736                         in_delayed_cksum(m);
 737                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 738                 }
 739
 740                 ip->ip_len = htons(ip->ip_len);
 741                 ip->ip_off = htons(ip->ip_off);
 742
 743                 /* NB: callee frees mbuf */
 744                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 745                 /*
 746                  * Preserve KAME behaviour: ENOENT can be returned
 747                  * when an SA acquire is in progress.  Don't propagate
 748                  * this to user-level; it confuses applications.
 749                  *
 750                  * XXX this will go away when the SADB is redone.
 751                  */
 752                 if (error == ENOENT)
 753                         error = 0;
 754                 crit_exit();
 755                 goto done;
 756         } else {
 757                 crit_exit();
 758
 759                 if (error != 0) {
 760                         /*
 761                          * Hack: -EINVAL is used to signal that a packet
 762                          * should be silently discarded.  This is typically
 763                          * because we asked key management for an SA and
 764                          * it was delayed (e.g. kicked up to IKE).
 765                          */
 766                         if (error == -EINVAL)
 767                                 error = 0;
 768                         goto bad;
 769                 } else {
 770                         /* No IPsec processing for this packet. */
 771                 }
 772 #ifdef notyet
 773                 /*
 774                  * If deferred crypto processing is needed, check that
 775                  * the interface supports it.
 776                  */
 777                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 778                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 779                         /* notify IPsec to do its own crypto */
 780                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 781                         error = EHOSTUNREACH;
 782                         goto bad;
 783                 }
 784 #endif
 785         }
 786 spd_done:
 787 #endif /* FAST_IPSEC */
 788
 789         /* We are already being fwd'd from a firewall. */
 790         if (next_hop != NULL)
 791                 goto pass;
 792
 793         /* No pfil hooks */
 794         if (!pfil_has_hooks(&inet_pfil_hook)) {
 795                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 796                         /*
 797                          * Strip dummynet tags from stranded packets
 798                          */
 799                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 800                         KKASSERT(mtag != NULL);
 801                         m_tag_delete(m, mtag);
 802                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 803                 }
 804                 goto pass;
 805         }
 806
 807         /*
 808          * IpHack's section.
 809          * - Xlate: translate packet's addr/port (NAT).
 810          * - Firewall: deny/allow/etc.
 811          * - Wrap: fake packet's addr/port <unimpl.>
 812          * - Encapsulate: put it in another IP and send out. <unimp.>
 813          */
 814
 815         /*
 816          * Run through list of hooks for output packets.
 817          */
 818         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 819         if (error != 0 || m == NULL)
 820                 goto done;
 821         ip = mtod(m, struct ip *);
 822
 823         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 824                 /*
 825                  * Check dst to make sure it is directly reachable on the
 826                  * interface we previously thought it was.
 827                  * If it isn't (which may be likely in some situations) we have
 828                  * to re-route it (ie, find a route for the next-hop and the
 829                  * associated interface) and set them here. This is nested
 830                  * forwarding which in most cases is undesirable, except where
 831                  * such control is nigh impossible. So we do it here.
 832                  * And I'm babbling.
 833                  */
 834                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 835                 KKASSERT(mtag != NULL);
 836                 next_hop = m_tag_data(mtag);
 837
 838                 /*
 839                  * Try local forwarding first
 840                  */
 841                 if (ip_localforward(m, next_hop, hlen))
 842                         goto done;
 843
 844                 /*
 845                  * Relocate the route based on next_hop.
 846                  * If the current route is inp's cache, keep it untouched.
 847                  */
 848                 if (ro == &iproute && ro->ro_rt != NULL) {
 849                         RTFREE(ro->ro_rt);
 850                         ro->ro_rt = NULL;
 851                 }
 852                 ro = &iproute;
 853                 bzero(ro, sizeof *ro);
 854
 855                 /*
 856                  * Forwarding to broadcast address is not allowed.
 857                  * XXX Should we follow IP_ROUTETOIF?
 858                  */
 859                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 860
 861                 /* We are doing forwarding now */
 862                 flags |= IP_FORWARDING;
 863
 864                 goto reroute;
 865         }
 866
 867         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 868                 struct dn_pkt *dn_pkt;
 869
 870                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 871                 KKASSERT(mtag != NULL);
 872                 dn_pkt = m_tag_data(mtag);
 873
 874                 /*
 875                  * Under certain cases it is not possible to recalculate
 876                  * 'ro' and 'dst', let alone 'flags', so just save them in
 877                  * dummynet tag and avoid the possible wrong reculcalation
 878                  * when we come back to ip_output() again.
 879                  *
 880                  * All other parameters have been already used and so they
 881                  * are not needed anymore.
 882                  * XXX if the ifp is deleted while a pkt is in dummynet,
 883                  * we are in trouble! (TODO use ifnet_detach_event)
 884                  *
 885                  * We need to copy *ro because for ICMP pkts (and maybe
 886                  * others) the caller passed a pointer into the stack;
 887                  * dst might also be a pointer into *ro so it needs to
 888                  * be updated.
 889                  */
 890                 dn_pkt->ro = *ro;
 891                 if (ro->ro_rt)
 892                         ro->ro_rt->rt_refcnt++;
 893                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 894                         /* 'dst' points into 'ro' */
 895                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 896                 }
 897                 dn_pkt->dn_dst = dst;
 898                 dn_pkt->flags = flags;
 899
 900                 ip_dn_queue(m);
 901                 goto done;
 902         }
 903 pass:
 904         /* 127/8 must not appear on wire - RFC1122. */
 905         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 906             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 907                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 908                         ipstat.ips_badaddr++;
 909                         error = EADDRNOTAVAIL;
 910                         goto bad;
 911                 }
 912         }
 913
 914         m->m_pkthdr.csum_flags |= CSUM_IP;
 915         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 916         if (sw_csum & CSUM_DELAY_DATA) {
 917                 in_delayed_cksum(m);
 918                 sw_csum &= ~CSUM_DELAY_DATA;
 919         }
 920         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 921
 922         /*
 923          * If small enough for interface, or the interface will take
 924          * care of the fragmentation for us, can just send directly.
 925          */
 926         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 927             !(ip->ip_off & IP_DF))) {
 928                 ip->ip_len = htons(ip->ip_len);
 929                 ip->ip_off = htons(ip->ip_off);
 930                 ip->ip_sum = 0;
 931                 if (sw_csum & CSUM_DELAY_IP) {
 932                         if (ip->ip_vhl == IP_VHL_BORING)
 933                                 ip->ip_sum = in_cksum_hdr(ip);
 934                         else
 935                                 ip->ip_sum = in_cksum(m, hlen);
 936                 }
 937
 938                 /* Record statistics for this interface address. */
 939                 if (!(flags & IP_FORWARDING) && ia) {
 940                         ia->ia_ifa.if_opackets++;
 941                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 942                 }
 943
 944 #ifdef IPSEC
 945                 /* clean ipsec history once it goes out of the node */
 946                 ipsec_delaux(m);
 947 #endif
 948
 949 #ifdef MBUF_STRESS_TEST
 950                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 951                         struct mbuf *m1, *m2;
 952                         int length, tmp;
 953
 954                         tmp = length = m->m_pkthdr.len;
 955
 956                         while ((length -= mbuf_frag_size) >= 1) {
 957                                 m1 = m_split(m, length, MB_DONTWAIT);
 958                                 if (m1 == NULL)
 959                                         break;
 960                                 m2 = m;
 961                                 while (m2->m_next != NULL)
 962                                         m2 = m2->m_next;
 963                                 m2->m_next = m1;
 964                         }
 965                         m->m_pkthdr.len = tmp;
 966                 }
 967 #endif
 968
 969 #ifdef MPLS
 970                 if (!mpls_output_process(m, ro->ro_rt))
 971                         goto done;
 972 #endif
 973                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 974                                        ro->ro_rt);
 975                 goto done;
 976         }
 977
 978         if (ip->ip_off & IP_DF) {
 979                 error = EMSGSIZE;
 980                 /*
 981                  * This case can happen if the user changed the MTU
 982                  * of an interface after enabling IP on it.  Because
 983                  * most netifs don't keep track of routes pointing to
 984                  * them, there is no way for one to update all its
 985                  * routes when the MTU is changed.
 986                  */
 987                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 988                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 989                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 990                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 991                 }
 992                 ipstat.ips_cantfrag++;
 993                 goto bad;
 994         }
 995
 996         /*
 997          * Too large for interface; fragment if possible. If successful,
 998          * on return, m will point to a list of packets to be sent.
 999          */
1000         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1001         if (error)
1002                 goto bad;
1003         for (; m; m = m0) {
1004                 m0 = m->m_nextpkt;
1005                 m->m_nextpkt = NULL;
1006 #ifdef IPSEC
1007                 /* clean ipsec history once it goes out of the node */
1008                 ipsec_delaux(m);
1009 #endif
1010                 if (error == 0) {
1011                         /* Record statistics for this interface address. */
1012                         if (ia != NULL) {
1013                                 ia->ia_ifa.if_opackets++;
1014                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1015                         }
1016 #ifdef MPLS
1017                         if (!mpls_output_process(m, ro->ro_rt))
1018                                 continue;
1019 #endif
1020                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1021                                                ro->ro_rt);
1022                 } else {
1023                         m_freem(m);
1024                 }
1025         }
1026
1027         if (error == 0)
1028                 ipstat.ips_fragmented++;
1029
1030 done:
1031         if (ro == &iproute && ro->ro_rt != NULL) {
1032                 RTFREE(ro->ro_rt);
1033                 ro->ro_rt = NULL;
1034         }
1035 #ifdef IPSEC
1036         if (sp != NULL) {
1037                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1038                         kprintf("DP ip_output call free SP:%p\n", sp));
1039                 key_freesp(sp);
1040         }
1041 #endif
1042 #ifdef FAST_IPSEC
1043         if (sp != NULL)
1044                 KEY_FREESP(&sp);
1045 #endif
1046         return (error);
1047 bad:
1048         m_freem(m);
1049         goto done;
1050 }
1051
1052 /*
1053  * Create a chain of fragments which fit the given mtu. m_frag points to the
1054  * mbuf to be fragmented; on return it points to the chain with the fragments.
1055  * Return 0 if no error. If error, m_frag may contain a partially built
1056  * chain of fragments that should be freed by the caller.
1057  *
1058  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1059  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1060  */
1061 int
1062 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1063             u_long if_hwassist_flags, int sw_csum)
1064 {
1065         int error = 0;
1066         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1067         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1068         int off;
1069         struct mbuf *m0 = *m_frag;      /* the original packet          */
1070         int firstlen;
1071         struct mbuf **mnext;
1072         int nfrags;
1073
1074         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1075                 ipstat.ips_cantfrag++;
1076                 return EMSGSIZE;
1077         }
1078
1079         /*
1080          * Must be able to put at least 8 bytes per fragment.
1081          */
1082         if (len < 8)
1083                 return EMSGSIZE;
1084
1085         /*
1086          * If the interface will not calculate checksums on
1087          * fragmented packets, then do it here.
1088          */
1089         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1090             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1091                 in_delayed_cksum(m0);
1092                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1093         }
1094
1095         if (len > PAGE_SIZE) {
1096                 /*
1097                  * Fragment large datagrams such that each segment
1098                  * contains a multiple of PAGE_SIZE amount of data,
1099                  * plus headers. This enables a receiver to perform
1100                  * page-flipping zero-copy optimizations.
1101                  *
1102                  * XXX When does this help given that sender and receiver
1103                  * could have different page sizes, and also mtu could
1104                  * be less than the receiver's page size ?
1105                  */
1106                 int newlen;
1107                 struct mbuf *m;
1108
1109                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1110                         off += m->m_len;
1111
1112                 /*
1113                  * firstlen (off - hlen) must be aligned on an
1114                  * 8-byte boundary
1115                  */
1116                 if (off < hlen)
1117                         goto smart_frag_failure;
1118                 off = ((off - hlen) & ~7) + hlen;
1119                 newlen = (~PAGE_MASK) & mtu;
1120                 if ((newlen + sizeof(struct ip)) > mtu) {
1121                         /* we failed, go back the default */
1122 smart_frag_failure:
1123                         newlen = len;
1124                         off = hlen + len;
1125                 }
1126                 len = newlen;
1127
1128         } else {
1129                 off = hlen + len;
1130         }
1131
1132         firstlen = off - hlen;
1133         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1134
1135         /*
1136          * Loop through length of segment after first fragment,
1137          * make new header and copy data of each part and link onto chain.
1138          * Here, m0 is the original packet, m is the fragment being created.
1139          * The fragments are linked off the m_nextpkt of the original
1140          * packet, which after processing serves as the first fragment.
1141          */
1142         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1143                 struct ip *mhip;        /* ip header on the fragment */
1144                 struct mbuf *m;
1145                 int mhlen = sizeof(struct ip);
1146
1147                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1148                 if (m == NULL) {
1149                         error = ENOBUFS;
1150                         ipstat.ips_odropped++;
1151                         goto done;
1152                 }
1153                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1154                 /*
1155                  * In the first mbuf, leave room for the link header, then
1156                  * copy the original IP header including options. The payload
1157                  * goes into an additional mbuf chain returned by m_copy().
1158                  */
1159                 m->m_data += max_linkhdr;
1160                 mhip = mtod(m, struct ip *);
1161                 *mhip = *ip;
1162                 if (hlen > sizeof(struct ip)) {
1163                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1164                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1165                 }
1166                 m->m_len = mhlen;
1167                 /* XXX do we need to add ip->ip_off below ? */
1168                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1169                 if (off + len >= ip->ip_len) {  /* last fragment */
1170                         len = ip->ip_len - off;
1171                         m->m_flags |= M_LASTFRAG;
1172                 } else
1173                         mhip->ip_off |= IP_MF;
1174                 mhip->ip_len = htons((u_short)(len + mhlen));
1175                 m->m_next = m_copy(m0, off, len);
1176                 if (m->m_next == NULL) {                /* copy failed */
1177                         m_free(m);
1178                         error = ENOBUFS;        /* ??? */
1179                         ipstat.ips_odropped++;
1180                         goto done;
1181                 }
1182                 m->m_pkthdr.len = mhlen + len;
1183                 m->m_pkthdr.rcvif = NULL;
1184                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1185                 mhip->ip_off = htons(mhip->ip_off);
1186                 mhip->ip_sum = 0;
1187                 if (sw_csum & CSUM_DELAY_IP)
1188                         mhip->ip_sum = in_cksum(m, mhlen);
1189                 *mnext = m;
1190                 mnext = &m->m_nextpkt;
1191         }
1192         ipstat.ips_ofragments += nfrags;
1193
1194         /* set first marker for fragment chain */
1195         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1196         m0->m_pkthdr.csum_data = nfrags;
1197
1198         /*
1199          * Update first fragment by trimming what's been copied out
1200          * and updating header.
1201          */
1202         m_adj(m0, hlen + firstlen - ip->ip_len);
1203         m0->m_pkthdr.len = hlen + firstlen;
1204         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1205         ip->ip_off |= IP_MF;
1206         ip->ip_off = htons(ip->ip_off);
1207         ip->ip_sum = 0;
1208         if (sw_csum & CSUM_DELAY_IP)
1209                 ip->ip_sum = in_cksum(m0, hlen);
1210
1211 done:
1212         *m_frag = m0;
1213         return error;
1214 }
1215
1216 void
1217 in_delayed_cksum(struct mbuf *m)
1218 {
1219         struct ip *ip;
1220         u_short csum, offset;
1221
1222         ip = mtod(m, struct ip *);
1223         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1224         csum = in_cksum_skip(m, ip->ip_len, offset);
1225         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1226                 csum = 0xffff;
1227         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1228
1229         if (offset + sizeof(u_short) > m->m_len) {
1230                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1231                     m->m_len, offset, ip->ip_p);
1232                 /*
1233                  * XXX
1234                  * this shouldn't happen, but if it does, the
1235                  * correct behavior may be to insert the checksum
1236                  * in the existing chain instead of rearranging it.
1237                  */
1238                 m = m_pullup(m, offset + sizeof(u_short));
1239         }
1240         *(u_short *)(m->m_data + offset) = csum;
1241 }
1242
1243 /*
1244  * Insert IP options into preformed packet.
1245  * Adjust IP destination as required for IP source routing,
1246  * as indicated by a non-zero in_addr at the start of the options.
1247  *
1248  * XXX This routine assumes that the packet has no options in place.
1249  */
1250 static struct mbuf *
1251 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1252 {
1253         struct ipoption *p = mtod(opt, struct ipoption *);
1254         struct mbuf *n;
1255         struct ip *ip = mtod(m, struct ip *);
1256         unsigned optlen;
1257
1258         optlen = opt->m_len - sizeof p->ipopt_dst;
1259         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1260                 *phlen = 0;
1261                 return (m);             /* XXX should fail */
1262         }
1263         if (p->ipopt_dst.s_addr)
1264                 ip->ip_dst = p->ipopt_dst;
1265         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1266                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1267                 if (n == NULL) {
1268                         *phlen = 0;
1269                         return (m);
1270                 }
1271                 n->m_pkthdr.rcvif = NULL;
1272                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1273                 m->m_len -= sizeof(struct ip);
1274                 m->m_data += sizeof(struct ip);
1275                 n->m_next = m;
1276                 m = n;
1277                 m->m_len = optlen + sizeof(struct ip);
1278                 m->m_data += max_linkhdr;
1279                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1280         } else {
1281                 m->m_data -= optlen;
1282                 m->m_len += optlen;
1283                 m->m_pkthdr.len += optlen;
1284                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1285         }
1286         ip = mtod(m, struct ip *);
1287         bcopy(p->ipopt_list, ip + 1, optlen);
1288         *phlen = sizeof(struct ip) + optlen;
1289         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1290         ip->ip_len += optlen;
1291         return (m);
1292 }
1293
1294 /*
1295  * Copy options from ip to jp,
1296  * omitting those not copied during fragmentation.
1297  */
1298 int
1299 ip_optcopy(struct ip *ip, struct ip *jp)
1300 {
1301         u_char *cp, *dp;
1302         int opt, optlen, cnt;
1303
1304         cp = (u_char *)(ip + 1);
1305         dp = (u_char *)(jp + 1);
1306         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1307         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1308                 opt = cp[0];
1309                 if (opt == IPOPT_EOL)
1310                         break;
1311                 if (opt == IPOPT_NOP) {
1312                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1313                         *dp++ = IPOPT_NOP;
1314                         optlen = 1;
1315                         continue;
1316                 }
1317
1318                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1319                     ("ip_optcopy: malformed ipv4 option"));
1320                 optlen = cp[IPOPT_OLEN];
1321                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1322                     ("ip_optcopy: malformed ipv4 option"));
1323
1324                 /* bogus lengths should have been caught by ip_dooptions */
1325                 if (optlen > cnt)
1326                         optlen = cnt;
1327                 if (IPOPT_COPIED(opt)) {
1328                         bcopy(cp, dp, optlen);
1329                         dp += optlen;
1330                 }
1331         }
1332         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1333                 *dp++ = IPOPT_EOL;
1334         return (optlen);
1335 }
1336
1337 /*
1338  * IP socket option processing.
1339  */
1340 void
1341 ip_ctloutput(netmsg_t msg)
1342 {
1343         struct socket *so = msg->base.nm_so;
1344         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1345         struct  inpcb *inp = so->so_pcb;
1346         int     error, optval;
1347
1348         error = optval = 0;
1349         if (sopt->sopt_level != IPPROTO_IP) {
1350                 error = EINVAL;
1351                 goto done;
1352         }
1353
1354         switch (sopt->sopt_dir) {
1355         case SOPT_SET:
1356                 switch (sopt->sopt_name) {
1357                 case IP_OPTIONS:
1358 #ifdef notyet
1359                 case IP_RETOPTS:
1360 #endif
1361                 {
1362                         struct mbuf *m;
1363                         if (sopt->sopt_valsize > MLEN) {
1364                                 error = EMSGSIZE;
1365                                 break;
1366                         }
1367                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1368                         if (m == NULL) {
1369                                 error = ENOBUFS;
1370                                 break;
1371                         }
1372                         m->m_len = sopt->sopt_valsize;
1373                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1374                                               m->m_len);
1375                         error = ip_pcbopts(sopt->sopt_name,
1376                                            &inp->inp_options, m);
1377                         goto done;
1378                 }
1379
1380                 case IP_TOS:
1381                 case IP_TTL:
1382                 case IP_MINTTL:
1383                 case IP_RECVOPTS:
1384                 case IP_RECVRETOPTS:
1385                 case IP_RECVDSTADDR:
1386                 case IP_RECVIF:
1387                 case IP_RECVTTL:
1388                 case IP_FAITH:
1389                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1390                                              sizeof optval);
1391                         if (error)
1392                                 break;
1393                         switch (sopt->sopt_name) {
1394                         case IP_TOS:
1395                                 inp->inp_ip_tos = optval;
1396                                 break;
1397
1398                         case IP_TTL:
1399                                 inp->inp_ip_ttl = optval;
1400                                 break;
1401                         case IP_MINTTL:
1402                                 if (optval >= 0 && optval <= MAXTTL)
1403                                         inp->inp_ip_minttl = optval;
1404                                 else
1405                                         error = EINVAL;
1406                                 break;
1407 #define OPTSET(bit) \
1408         if (optval) \
1409                 inp->inp_flags |= bit; \
1410         else \
1411                 inp->inp_flags &= ~bit;
1412
1413                         case IP_RECVOPTS:
1414                                 OPTSET(INP_RECVOPTS);
1415                                 break;
1416
1417                         case IP_RECVRETOPTS:
1418                                 OPTSET(INP_RECVRETOPTS);
1419                                 break;
1420
1421                         case IP_RECVDSTADDR:
1422                                 OPTSET(INP_RECVDSTADDR);
1423                                 break;
1424
1425                         case IP_RECVIF:
1426                                 OPTSET(INP_RECVIF);
1427                                 break;
1428
1429                         case IP_RECVTTL:
1430                                 OPTSET(INP_RECVTTL);
1431                                 break;
1432
1433                         case IP_FAITH:
1434                                 OPTSET(INP_FAITH);
1435                                 break;
1436                         }
1437                         break;
1438 #undef OPTSET
1439
1440                 case IP_MULTICAST_IF:
1441                 case IP_MULTICAST_VIF:
1442                 case IP_MULTICAST_TTL:
1443                 case IP_MULTICAST_LOOP:
1444                 case IP_ADD_MEMBERSHIP:
1445                 case IP_DROP_MEMBERSHIP:
1446                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1447                         break;
1448
1449                 case IP_PORTRANGE:
1450                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1451                                             sizeof optval);
1452                         if (error)
1453                                 break;
1454
1455                         switch (optval) {
1456                         case IP_PORTRANGE_DEFAULT:
1457                                 inp->inp_flags &= ~(INP_LOWPORT);
1458                                 inp->inp_flags &= ~(INP_HIGHPORT);
1459                                 break;
1460
1461                         case IP_PORTRANGE_HIGH:
1462                                 inp->inp_flags &= ~(INP_LOWPORT);
1463                                 inp->inp_flags |= INP_HIGHPORT;
1464                                 break;
1465
1466                         case IP_PORTRANGE_LOW:
1467                                 inp->inp_flags &= ~(INP_HIGHPORT);
1468                                 inp->inp_flags |= INP_LOWPORT;
1469                                 break;
1470
1471                         default:
1472                                 error = EINVAL;
1473                                 break;
1474                         }
1475                         break;
1476
1477 #if defined(IPSEC) || defined(FAST_IPSEC)
1478                 case IP_IPSEC_POLICY:
1479                 {
1480                         caddr_t req;
1481                         size_t len = 0;
1482                         int priv;
1483                         struct mbuf *m;
1484                         int optname;
1485
1486                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1487                                 break;
1488                         soopt_to_mbuf(sopt, m);
1489                         priv = (sopt->sopt_td != NULL &&
1490                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1491                         req = mtod(m, caddr_t);
1492                         len = m->m_len;
1493                         optname = sopt->sopt_name;
1494                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1495                         m_freem(m);
1496                         break;
1497                 }
1498 #endif /*IPSEC*/
1499
1500                 default:
1501                         error = ENOPROTOOPT;
1502                         break;
1503                 }
1504                 break;
1505
1506         case SOPT_GET:
1507                 switch (sopt->sopt_name) {
1508                 case IP_OPTIONS:
1509                 case IP_RETOPTS:
1510                         if (inp->inp_options)
1511                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1512                                                            char *),
1513                                                 inp->inp_options->m_len);
1514                         else
1515                                 sopt->sopt_valsize = 0;
1516                         break;
1517
1518                 case IP_TOS:
1519                 case IP_TTL:
1520                 case IP_MINTTL:
1521                 case IP_RECVOPTS:
1522                 case IP_RECVRETOPTS:
1523                 case IP_RECVDSTADDR:
1524                 case IP_RECVTTL:
1525                 case IP_RECVIF:
1526                 case IP_PORTRANGE:
1527                 case IP_FAITH:
1528                         switch (sopt->sopt_name) {
1529
1530                         case IP_TOS:
1531                                 optval = inp->inp_ip_tos;
1532                                 break;
1533
1534                         case IP_TTL:
1535                                 optval = inp->inp_ip_ttl;
1536                                 break;
1537                         case IP_MINTTL:
1538                                 optval = inp->inp_ip_minttl;
1539                                 break;
1540
1541 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1542
1543                         case IP_RECVOPTS:
1544                                 optval = OPTBIT(INP_RECVOPTS);
1545                                 break;
1546
1547                         case IP_RECVRETOPTS:
1548                                 optval = OPTBIT(INP_RECVRETOPTS);
1549                                 break;
1550
1551                         case IP_RECVDSTADDR:
1552                                 optval = OPTBIT(INP_RECVDSTADDR);
1553                                 break;
1554
1555                         case IP_RECVTTL:
1556                                 optval = OPTBIT(INP_RECVTTL);
1557                                 break;
1558
1559                         case IP_RECVIF:
1560                                 optval = OPTBIT(INP_RECVIF);
1561                                 break;
1562
1563                         case IP_PORTRANGE:
1564                                 if (inp->inp_flags & INP_HIGHPORT)
1565                                         optval = IP_PORTRANGE_HIGH;
1566                                 else if (inp->inp_flags & INP_LOWPORT)
1567                                         optval = IP_PORTRANGE_LOW;
1568                                 else
1569                                         optval = 0;
1570                                 break;
1571
1572                         case IP_FAITH:
1573                                 optval = OPTBIT(INP_FAITH);
1574                                 break;
1575                         }
1576                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1577                         break;
1578
1579                 case IP_MULTICAST_IF:
1580                 case IP_MULTICAST_VIF:
1581                 case IP_MULTICAST_TTL:
1582                 case IP_MULTICAST_LOOP:
1583                 case IP_ADD_MEMBERSHIP:
1584                 case IP_DROP_MEMBERSHIP:
1585                         error = ip_getmoptions(sopt, inp->inp_moptions);
1586                         break;
1587
1588 #if defined(IPSEC) || defined(FAST_IPSEC)
1589                 case IP_IPSEC_POLICY:
1590                 {
1591                         struct mbuf *m = NULL;
1592                         caddr_t req = NULL;
1593                         size_t len = 0;
1594
1595                         if (m != NULL) {
1596                                 req = mtod(m, caddr_t);
1597                                 len = m->m_len;
1598                         }
1599                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1600                         if (error == 0)
1601                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1602                         if (error == 0)
1603                                 m_freem(m);
1604                         break;
1605                 }
1606 #endif /*IPSEC*/
1607
1608                 default:
1609                         error = ENOPROTOOPT;
1610                         break;
1611                 }
1612                 break;
1613         }
1614 done:
1615         lwkt_replymsg(&msg->lmsg, error);
1616 }
1617
1618 /*
1619  * Set up IP options in pcb for insertion in output packets.
1620  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1621  * with destination address if source routed.
1622  */
1623 static int
1624 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1625 {
1626         int cnt, optlen;
1627         u_char *cp;
1628         u_char opt;
1629
1630         /* turn off any old options */
1631         if (*pcbopt)
1632                 m_free(*pcbopt);
1633         *pcbopt = NULL;
1634         if (m == NULL || m->m_len == 0) {
1635                 /*
1636                  * Only turning off any previous options.
1637                  */
1638                 if (m != NULL)
1639                         m_free(m);
1640                 return (0);
1641         }
1642
1643         if (m->m_len % sizeof(int32_t))
1644                 goto bad;
1645         /*
1646          * IP first-hop destination address will be stored before
1647          * actual options; move other options back
1648          * and clear it when none present.
1649          */
1650         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1651                 goto bad;
1652         cnt = m->m_len;
1653         m->m_len += sizeof(struct in_addr);
1654         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1655         ovbcopy(mtod(m, caddr_t), cp, cnt);
1656         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1657
1658         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1659                 opt = cp[IPOPT_OPTVAL];
1660                 if (opt == IPOPT_EOL)
1661                         break;
1662                 if (opt == IPOPT_NOP)
1663                         optlen = 1;
1664                 else {
1665                         if (cnt < IPOPT_OLEN + sizeof *cp)
1666                                 goto bad;
1667                         optlen = cp[IPOPT_OLEN];
1668                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1669                                 goto bad;
1670                 }
1671                 switch (opt) {
1672
1673                 default:
1674                         break;
1675
1676                 case IPOPT_LSRR:
1677                 case IPOPT_SSRR:
1678                         /*
1679                          * user process specifies route as:
1680                          *      ->A->B->C->D
1681                          * D must be our final destination (but we can't
1682                          * check that since we may not have connected yet).
1683                          * A is first hop destination, which doesn't appear in
1684                          * actual IP option, but is stored before the options.
1685                          */
1686                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1687                                 goto bad;
1688                         m->m_len -= sizeof(struct in_addr);
1689                         cnt -= sizeof(struct in_addr);
1690                         optlen -= sizeof(struct in_addr);
1691                         cp[IPOPT_OLEN] = optlen;
1692                         /*
1693                          * Move first hop before start of options.
1694                          */
1695                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1696                               sizeof(struct in_addr));
1697                         /*
1698                          * Then copy rest of options back
1699                          * to close up the deleted entry.
1700                          */
1701                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1702                                 &cp[IPOPT_OFFSET+1],
1703                                 cnt - (IPOPT_MINOFF - 1));
1704                         break;
1705                 }
1706         }
1707         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1708                 goto bad;
1709         *pcbopt = m;
1710         return (0);
1711
1712 bad:
1713         m_free(m);
1714         return (EINVAL);
1715 }
1716
1717 /*
1718  * XXX
1719  * The whole multicast option thing needs to be re-thought.
1720  * Several of these options are equally applicable to non-multicast
1721  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1722  * standard option (IP_TTL).
1723  */
1724
1725 /*
1726  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1727  */
1728 static struct ifnet *
1729 ip_multicast_if(struct in_addr *a, int *ifindexp)
1730 {
1731         int ifindex;
1732         struct ifnet *ifp;
1733
1734         if (ifindexp)
1735                 *ifindexp = 0;
1736         if (ntohl(a->s_addr) >> 24 == 0) {
1737                 ifindex = ntohl(a->s_addr) & 0xffffff;
1738                 if (ifindex < 0 || if_index < ifindex)
1739                         return NULL;
1740                 ifp = ifindex2ifnet[ifindex];
1741                 if (ifindexp)
1742                         *ifindexp = ifindex;
1743         } else {
1744                 ifp = INADDR_TO_IFP(a);
1745         }
1746         return ifp;
1747 }
1748
1749 /*
1750  * Set the IP multicast options in response to user setsockopt().
1751  */
1752 static int
1753 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1754 {
1755         int error = 0;
1756         int i;
1757         struct in_addr addr;
1758         struct ip_mreq mreq;
1759         struct ifnet *ifp;
1760         struct ip_moptions *imo = *imop;
1761         int ifindex;
1762
1763         if (imo == NULL) {
1764                 /*
1765                  * No multicast option buffer attached to the pcb;
1766                  * allocate one and initialize to default values.
1767                  */
1768                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1769
1770                 *imop = imo;
1771                 imo->imo_multicast_ifp = NULL;
1772                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1773                 imo->imo_multicast_vif = -1;
1774                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1775                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1776                 imo->imo_num_memberships = 0;
1777         }
1778         switch (sopt->sopt_name) {
1779         /* store an index number for the vif you wanna use in the send */
1780         case IP_MULTICAST_VIF:
1781                 if (legal_vif_num == 0) {
1782                         error = EOPNOTSUPP;
1783                         break;
1784                 }
1785                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1786                 if (error)
1787                         break;
1788                 if (!legal_vif_num(i) && (i != -1)) {
1789                         error = EINVAL;
1790                         break;
1791                 }
1792                 imo->imo_multicast_vif = i;
1793                 break;
1794
1795         case IP_MULTICAST_IF:
1796                 /*
1797                  * Select the interface for outgoing multicast packets.
1798                  */
1799                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1800                 if (error)
1801                         break;
1802
1803                 /*
1804                  * INADDR_ANY is used to remove a previous selection.
1805                  * When no interface is selected, a default one is
1806                  * chosen every time a multicast packet is sent.
1807                  */
1808                 if (addr.s_addr == INADDR_ANY) {
1809                         imo->imo_multicast_ifp = NULL;
1810                         break;
1811                 }
1812                 /*
1813                  * The selected interface is identified by its local
1814                  * IP address.  Find the interface and confirm that
1815                  * it supports multicasting.
1816                  */
1817                 crit_enter();
1818                 ifp = ip_multicast_if(&addr, &ifindex);
1819                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1820                         crit_exit();
1821                         error = EADDRNOTAVAIL;
1822                         break;
1823                 }
1824                 imo->imo_multicast_ifp = ifp;
1825                 if (ifindex)
1826                         imo->imo_multicast_addr = addr;
1827                 else
1828                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1829                 crit_exit();
1830                 break;
1831
1832         case IP_MULTICAST_TTL:
1833                 /*
1834                  * Set the IP time-to-live for outgoing multicast packets.
1835                  * The original multicast API required a char argument,
1836                  * which is inconsistent with the rest of the socket API.
1837                  * We allow either a char or an int.
1838                  */
1839                 if (sopt->sopt_valsize == 1) {
1840                         u_char ttl;
1841                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1842                         if (error)
1843                                 break;
1844                         imo->imo_multicast_ttl = ttl;
1845                 } else {
1846                         u_int ttl;
1847                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1848                         if (error)
1849                                 break;
1850                         if (ttl > 255)
1851                                 error = EINVAL;
1852                         else
1853                                 imo->imo_multicast_ttl = ttl;
1854                 }
1855                 break;
1856
1857         case IP_MULTICAST_LOOP:
1858                 /*
1859                  * Set the loopback flag for outgoing multicast packets.
1860                  * Must be zero or one.  The original multicast API required a
1861                  * char argument, which is inconsistent with the rest
1862                  * of the socket API.  We allow either a char or an int.
1863                  */
1864                 if (sopt->sopt_valsize == 1) {
1865                         u_char loop;
1866
1867                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1868                         if (error)
1869                                 break;
1870                         imo->imo_multicast_loop = !!loop;
1871                 } else {
1872                         u_int loop;
1873
1874                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1875                                             sizeof loop);
1876                         if (error)
1877                                 break;
1878                         imo->imo_multicast_loop = !!loop;
1879                 }
1880                 break;
1881
1882         case IP_ADD_MEMBERSHIP:
1883                 /*
1884                  * Add a multicast group membership.
1885                  * Group must be a valid IP multicast address.
1886                  */
1887                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1888                 if (error)
1889                         break;
1890
1891                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1892                         error = EINVAL;
1893                         break;
1894                 }
1895                 crit_enter();
1896                 /*
1897                  * If no interface address was provided, use the interface of
1898                  * the route to the given multicast address.
1899                  */
1900                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1901                         struct sockaddr_in dst;
1902                         struct rtentry *rt;
1903
1904                         bzero(&dst, sizeof(struct sockaddr_in));
1905                         dst.sin_len = sizeof(struct sockaddr_in);
1906                         dst.sin_family = AF_INET;
1907                         dst.sin_addr = mreq.imr_multiaddr;
1908                         rt = rtlookup((struct sockaddr *)&dst);
1909                         if (rt == NULL) {
1910                                 error = EADDRNOTAVAIL;
1911                                 crit_exit();
1912                                 break;
1913                         }
1914                         --rt->rt_refcnt;
1915                         ifp = rt->rt_ifp;
1916                 } else {
1917                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1918                 }
1919
1920                 /*
1921                  * See if we found an interface, and confirm that it
1922                  * supports multicast.
1923                  */
1924                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1925                         error = EADDRNOTAVAIL;
1926                         crit_exit();
1927                         break;
1928                 }
1929                 /*
1930                  * See if the membership already exists or if all the
1931                  * membership slots are full.
1932                  */
1933                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1934                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1935                             imo->imo_membership[i]->inm_addr.s_addr
1936                                                 == mreq.imr_multiaddr.s_addr)
1937                                 break;
1938                 }
1939                 if (i < imo->imo_num_memberships) {
1940                         error = EADDRINUSE;
1941                         crit_exit();
1942                         break;
1943                 }
1944                 if (i == IP_MAX_MEMBERSHIPS) {
1945                         error = ETOOMANYREFS;
1946                         crit_exit();
1947                         break;
1948                 }
1949                 /*
1950                  * Everything looks good; add a new record to the multicast
1951                  * address list for the given interface.
1952                  */
1953                 if ((imo->imo_membership[i] =
1954                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1955                         error = ENOBUFS;
1956                         crit_exit();
1957                         break;
1958                 }
1959                 ++imo->imo_num_memberships;
1960                 crit_exit();
1961                 break;
1962
1963         case IP_DROP_MEMBERSHIP:
1964                 /*
1965                  * Drop a multicast group membership.
1966                  * Group must be a valid IP multicast address.
1967                  */
1968                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1969                 if (error)
1970                         break;
1971
1972                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1973                         error = EINVAL;
1974                         break;
1975                 }
1976
1977                 crit_enter();
1978                 /*
1979                  * If an interface address was specified, get a pointer
1980                  * to its ifnet structure.
1981                  */
1982                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1983                         ifp = NULL;
1984                 else {
1985                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1986                         if (ifp == NULL) {
1987                                 error = EADDRNOTAVAIL;
1988                                 crit_exit();
1989                                 break;
1990                         }
1991                 }
1992                 /*
1993                  * Find the membership in the membership array.
1994                  */
1995                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1996                         if ((ifp == NULL ||
1997                              imo->imo_membership[i]->inm_ifp == ifp) &&
1998                             imo->imo_membership[i]->inm_addr.s_addr ==
1999                             mreq.imr_multiaddr.s_addr)
2000                                 break;
2001                 }
2002                 if (i == imo->imo_num_memberships) {
2003                         error = EADDRNOTAVAIL;
2004                         crit_exit();
2005                         break;
2006                 }
2007                 /*
2008                  * Give up the multicast address record to which the
2009                  * membership points.
2010                  */
2011                 in_delmulti(imo->imo_membership[i]);
2012                 /*
2013                  * Remove the gap in the membership array.
2014                  */
2015                 for (++i; i < imo->imo_num_memberships; ++i)
2016                         imo->imo_membership[i-1] = imo->imo_membership[i];
2017                 --imo->imo_num_memberships;
2018                 crit_exit();
2019                 break;
2020
2021         default:
2022                 error = EOPNOTSUPP;
2023                 break;
2024         }
2025
2026         /*
2027          * If all options have default values, no need to keep the mbuf.
2028          */
2029         if (imo->imo_multicast_ifp == NULL &&
2030             imo->imo_multicast_vif == -1 &&
2031             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2032             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2033             imo->imo_num_memberships == 0) {
2034                 kfree(*imop, M_IPMOPTS);
2035                 *imop = NULL;
2036         }
2037
2038         return (error);
2039 }
2040
2041 /*
2042  * Return the IP multicast options in response to user getsockopt().
2043  */
2044 static int
2045 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2046 {
2047         struct in_addr addr;
2048         struct in_ifaddr *ia;
2049         int error, optval;
2050         u_char coptval;
2051
2052         error = 0;
2053         switch (sopt->sopt_name) {
2054         case IP_MULTICAST_VIF:
2055                 if (imo != NULL)
2056                         optval = imo->imo_multicast_vif;
2057                 else
2058                         optval = -1;
2059                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2060                 break;
2061
2062         case IP_MULTICAST_IF:
2063                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2064                         addr.s_addr = INADDR_ANY;
2065                 else if (imo->imo_multicast_addr.s_addr) {
2066                         /* return the value user has set */
2067                         addr = imo->imo_multicast_addr;
2068                 } else {
2069                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2070                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2071                                 : IA_SIN(ia)->sin_addr.s_addr;
2072                 }
2073                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2074                 break;
2075
2076         case IP_MULTICAST_TTL:
2077                 if (imo == NULL)
2078                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2079                 else
2080                         optval = coptval = imo->imo_multicast_ttl;
2081                 if (sopt->sopt_valsize == 1)
2082                         soopt_from_kbuf(sopt, &coptval, 1);
2083                 else
2084                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2085                 break;
2086
2087         case IP_MULTICAST_LOOP:
2088                 if (imo == NULL)
2089                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2090                 else
2091                         optval = coptval = imo->imo_multicast_loop;
2092                 if (sopt->sopt_valsize == 1)
2093                         soopt_from_kbuf(sopt, &coptval, 1);
2094                 else
2095                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2096                 break;
2097
2098         default:
2099                 error = ENOPROTOOPT;
2100                 break;
2101         }
2102         return (error);
2103 }
2104
2105 /*
2106  * Discard the IP multicast options.
2107  */
2108 void
2109 ip_freemoptions(struct ip_moptions *imo)
2110 {
2111         int i;
2112
2113         if (imo != NULL) {
2114                 for (i = 0; i < imo->imo_num_memberships; ++i)
2115                         in_delmulti(imo->imo_membership[i]);
2116                 kfree(imo, M_IPMOPTS);
2117         }
2118 }
2119
2120 /*
2121  * Routine called from ip_output() to loop back a copy of an IP multicast
2122  * packet to the input queue of a specified interface.  Note that this
2123  * calls the output routine of the loopback "driver", but with an interface
2124  * pointer that might NOT be a loopback interface -- evil, but easier than
2125  * replicating that code here.
2126  */
2127 static void
2128 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2129              int hlen)
2130 {
2131         struct ip *ip;
2132         struct mbuf *copym;
2133
2134         copym = m_copypacket(m, MB_DONTWAIT);
2135         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2136                 copym = m_pullup(copym, hlen);
2137         if (copym != NULL) {
2138                 /*
2139                  * if the checksum hasn't been computed, mark it as valid
2140                  */
2141                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2142                         in_delayed_cksum(copym);
2143                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2144                         copym->m_pkthdr.csum_flags |=
2145                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2146                         copym->m_pkthdr.csum_data = 0xffff;
2147                 }
2148                 /*
2149                  * We don't bother to fragment if the IP length is greater
2150                  * than the interface's MTU.  Can this possibly matter?
2151                  */
2152                 ip = mtod(copym, struct ip *);
2153                 ip->ip_len = htons(ip->ip_len);
2154                 ip->ip_off = htons(ip->ip_off);
2155                 ip->ip_sum = 0;
2156                 if (ip->ip_vhl == IP_VHL_BORING) {
2157                         ip->ip_sum = in_cksum_hdr(ip);
2158                 } else {
2159                         ip->ip_sum = in_cksum(copym, hlen);
2160                 }
2161                 /*
2162                  * NB:
2163                  * It's not clear whether there are any lingering
2164                  * reentrancy problems in other areas which might
2165                  * be exposed by using ip_input directly (in
2166                  * particular, everything which modifies the packet
2167                  * in-place).  Yet another option is using the
2168                  * protosw directly to deliver the looped back
2169                  * packet.  For the moment, we'll err on the side
2170                  * of safety by using if_simloop().
2171                  */
2172 #if 1 /* XXX */
2173                 if (dst->sin_family != AF_INET) {
2174                         kprintf("ip_mloopback: bad address family %d\n",
2175                                                 dst->sin_family);
2176                         dst->sin_family = AF_INET;
2177                 }
2178 #endif
2179                 get_mplock();   /* is if_simloop() mpsafe yet? */
2180                 if_simloop(ifp, copym, dst->sin_family, 0);
2181                 rel_mplock();
2182         }
2183 }