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                 m->m_flags |= M_MCAST;
 379                 /*
 380                  * IP destination address is multicast.  Make sure "dst"
 381                  * still points to the address in "ro".  (It may have been
 382                  * changed to point to a gateway address, above.)
 383                  */
 384                 dst = (struct sockaddr_in *)&ro->ro_dst;
 385                 /*
 386                  * See if the caller provided any multicast options
 387                  */
 388                 if (imo != NULL) {
 389                         ip->ip_ttl = imo->imo_multicast_ttl;
 390                         if (imo->imo_multicast_vif != -1) {
 391                                 ip->ip_src.s_addr =
 392                                     ip_mcast_src ?
 393                                     ip_mcast_src(imo->imo_multicast_vif) :
 394                                     INADDR_ANY;
 395                         }
 396                 } else {
 397                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 398                 }
 399                 /*
 400                  * Confirm that the outgoing interface supports multicast.
 401                  */
 402                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 403                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 404                                 ipstat.ips_noroute++;
 405                                 error = ENETUNREACH;
 406                                 goto bad;
 407                         }
 408                 }
 409                 /*
 410                  * If source address not specified yet, use address of the
 411                  * outgoing interface.  In case, keep note we did that, so
 412                  * if the the firewall changes the next-hop causing the
 413                  * output interface to change, we can fix that.
 414                  */
 415                 if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 416                         /* Interface may have no addresses. */
 417                         if (ia != NULL) {
 418                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 419                                 src_was_INADDR_ANY = 1;
 420                         }
 421                 }
 422
 423                 if (ip->ip_src.s_addr != INADDR_ANY) {
 424                         struct in_multi *inm;
 425
 426                         IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 427                         if (inm != NULL &&
 428                             (imo == NULL || imo->imo_multicast_loop)) {
 429                                 /*
 430                                  * If we belong to the destination multicast
 431                                  * group on the outgoing interface, and the
 432                                  * caller did not forbid loopback, loop back
 433                                  * a copy.
 434                                  */
 435                                 ip_mloopback(ifp, m, dst, hlen);
 436                         } else {
 437                                 /*
 438                                  * If we are acting as a multicast router,
 439                                  * perform multicast forwarding as if the
 440                                  * packet had just arrived on the interface
 441                                  * to which we are about to send.  The
 442                                  * multicast forwarding function recursively
 443                                  * calls this function, using the IP_FORWARDING
 444                                  * flag to prevent infinite recursion.
 445                                  *
 446                                  * Multicasts that are looped back by
 447                                  * ip_mloopback(), above, will be forwarded by
 448                                  * the ip_input() routine, if necessary.
 449                                  */
 450                                 if (ip_mrouter && !(flags & IP_FORWARDING)) {
 451                                         /*
 452                                          * If rsvp daemon is not running, do
 453                                          * not set ip_moptions. This ensures
 454                                          * that the packet is multicast and
 455                                          * not just sent down one link as
 456                                          * prescribed by rsvpd.
 457                                          */
 458                                         if (!rsvp_on)
 459                                                 imo = NULL;
 460                                         if (ip_mforward) {
 461                                                 get_mplock();
 462                                                 if (ip_mforward(ip, ifp,
 463                                                     m, imo) != 0) {
 464                                                         m_freem(m);
 465                                                         rel_mplock();
 466                                                         goto done;
 467                                                 }
 468                                                 rel_mplock();
 469                                         }
 470                                 }
 471                         }
 472                 }
 473
 474                 /*
 475                  * Multicasts with a time-to-live of zero may be looped-
 476                  * back, above, but must not be transmitted on a network.
 477                  * Also, multicasts addressed to the loopback interface
 478                  * are not sent -- the above call to ip_mloopback() will
 479                  * loop back a copy if this host actually belongs to the
 480                  * destination group on the loopback interface.
 481                  */
 482                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 483                         m_freem(m);
 484                         goto done;
 485                 }
 486
 487                 goto sendit;
 488         } else {
 489                 m->m_flags &= ~M_MCAST;
 490         }
 491
 492         /*
 493          * If the source address is not specified yet, use the address
 494          * of the outgoing interface.  In case, keep note we did that,
 495          * so if the the firewall changes the next-hop causing the output
 496          * interface to change, we can fix that.
 497          */
 498         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 499                 /* Interface may have no addresses. */
 500                 if (ia != NULL) {
 501                         ip->ip_src = IA_SIN(ia)->sin_addr;
 502                         src_was_INADDR_ANY = 1;
 503                 }
 504         }
 505
 506         /*
 507          * Look for broadcast address and
 508          * verify user is allowed to send
 509          * such a packet.
 510          */
 511         if (isbroadcast) {
 512                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 513                         error = EADDRNOTAVAIL;
 514                         goto bad;
 515                 }
 516                 if (!(flags & IP_ALLOWBROADCAST)) {
 517                         error = EACCES;
 518                         goto bad;
 519                 }
 520                 /* don't allow broadcast messages to be fragmented */
 521                 if (ip->ip_len > ifp->if_mtu) {
 522                         error = EMSGSIZE;
 523                         goto bad;
 524                 }
 525                 m->m_flags |= M_BCAST;
 526         } else {
 527                 m->m_flags &= ~M_BCAST;
 528         }
 529
 530 sendit:
 531 #ifdef IPSEC
 532         /* get SP for this packet */
 533         if (so == NULL)
 534                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 535         else
 536                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 537
 538         if (sp == NULL) {
 539                 ipsecstat.out_inval++;
 540                 goto bad;
 541         }
 542
 543         error = 0;
 544
 545         /* check policy */
 546         switch (sp->policy) {
 547         case IPSEC_POLICY_DISCARD:
 548                 /*
 549                  * This packet is just discarded.
 550                  */
 551                 ipsecstat.out_polvio++;
 552                 goto bad;
 553
 554         case IPSEC_POLICY_BYPASS:
 555         case IPSEC_POLICY_NONE:
 556         case IPSEC_POLICY_TCP:
 557                 /* no need to do IPsec. */
 558                 goto skip_ipsec;
 559
 560         case IPSEC_POLICY_IPSEC:
 561                 if (sp->req == NULL) {
 562                         /* acquire a policy */
 563                         error = key_spdacquire(sp);
 564                         goto bad;
 565                 }
 566                 break;
 567
 568         case IPSEC_POLICY_ENTRUST:
 569         default:
 570                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 571         }
 572     {
 573         struct ipsec_output_state state;
 574         bzero(&state, sizeof state);
 575         state.m = m;
 576         if (flags & IP_ROUTETOIF) {
 577                 state.ro = &iproute;
 578                 bzero(&iproute, sizeof iproute);
 579         } else
 580                 state.ro = ro;
 581         state.dst = (struct sockaddr *)dst;
 582
 583         ip->ip_sum = 0;
 584
 585         /*
 586          * XXX
 587          * delayed checksums are not currently compatible with IPsec
 588          */
 589         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 590                 in_delayed_cksum(m);
 591                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 592         }
 593
 594         ip->ip_len = htons(ip->ip_len);
 595         ip->ip_off = htons(ip->ip_off);
 596
 597         error = ipsec4_output(&state, sp, flags);
 598
 599         m = state.m;
 600         if (flags & IP_ROUTETOIF) {
 601                 /*
 602                  * if we have tunnel mode SA, we may need to ignore
 603                  * IP_ROUTETOIF.
 604                  */
 605                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 606                         flags &= ~IP_ROUTETOIF;
 607                         ro = state.ro;
 608                 }
 609         } else
 610                 ro = state.ro;
 611         dst = (struct sockaddr_in *)state.dst;
 612         if (error) {
 613                 /* mbuf is already reclaimed in ipsec4_output. */
 614                 m0 = NULL;
 615                 switch (error) {
 616                 case EHOSTUNREACH:
 617                 case ENETUNREACH:
 618                 case EMSGSIZE:
 619                 case ENOBUFS:
 620                 case ENOMEM:
 621                         break;
 622                 default:
 623                         kprintf("ip4_output (ipsec): error code %d\n", error);
 624                         /*fall through*/
 625                 case ENOENT:
 626                         /* don't show these error codes to the user */
 627                         error = 0;
 628                         break;
 629                 }
 630                 goto bad;
 631         }
 632     }
 633
 634         /* be sure to update variables that are affected by ipsec4_output() */
 635         ip = mtod(m, struct ip *);
 636 #ifdef _IP_VHL
 637         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 638 #else
 639         hlen = ip->ip_hl << 2;
 640 #endif
 641         if (ro->ro_rt == NULL) {
 642                 if (!(flags & IP_ROUTETOIF)) {
 643                         kprintf("ip_output: "
 644                                 "can't update route after IPsec processing\n");
 645                         error = EHOSTUNREACH;   /*XXX*/
 646                         goto bad;
 647                 }
 648         } else {
 649                 ia = ifatoia(ro->ro_rt->rt_ifa);
 650                 ifp = ro->ro_rt->rt_ifp;
 651         }
 652
 653         /* make it flipped, again. */
 654         ip->ip_len = ntohs(ip->ip_len);
 655         ip->ip_off = ntohs(ip->ip_off);
 656 skip_ipsec:
 657 #endif /*IPSEC*/
 658 #ifdef FAST_IPSEC
 659         /*
 660          * Check the security policy (SP) for the packet and, if
 661          * required, do IPsec-related processing.  There are two
 662          * cases here; the first time a packet is sent through
 663          * it will be untagged and handled by ipsec4_checkpolicy.
 664          * If the packet is resubmitted to ip_output (e.g. after
 665          * AH, ESP, etc. processing), there will be a tag to bypass
 666          * the lookup and related policy checking.
 667          */
 668         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 669         crit_enter();
 670         if (mtag != NULL) {
 671                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 672                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 673                 if (sp == NULL)
 674                         error = -EINVAL;        /* force silent drop */
 675                 m_tag_delete(m, mtag);
 676         } else {
 677                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 678                                         &error, inp);
 679         }
 680         /*
 681          * There are four return cases:
 682          *    sp != NULL                    apply IPsec policy
 683          *    sp == NULL, error == 0        no IPsec handling needed
 684          *    sp == NULL, error == -EINVAL  discard packet w/o error
 685          *    sp == NULL, error != 0        discard packet, report error
 686          */
 687         if (sp != NULL) {
 688                 /* Loop detection, check if ipsec processing already done */
 689                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 690                 for (mtag = m_tag_first(m); mtag != NULL;
 691                      mtag = m_tag_next(m, mtag)) {
 692                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 693                                 continue;
 694                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 695                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 696                                 continue;
 697                         /*
 698                          * Check if policy has an SA associated with it.
 699                          * This can happen when an SP has yet to acquire
 700                          * an SA; e.g. on first reference.  If it occurs,
 701                          * then we let ipsec4_process_packet do its thing.
 702                          */
 703                         if (sp->req->sav == NULL)
 704                                 break;
 705                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 706                         if (tdbi->spi == sp->req->sav->spi &&
 707                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 708                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 709                                  sizeof(union sockaddr_union)) == 0) {
 710                                 /*
 711                                  * No IPsec processing is needed, free
 712                                  * reference to SP.
 713                                  *
 714                                  * NB: null pointer to avoid free at
 715                                  *     done: below.
 716                                  */
 717                                 KEY_FREESP(&sp), sp = NULL;
 718                                 crit_exit();
 719                                 goto spd_done;
 720                         }
 721                 }
 722
 723                 /*
 724                  * Do delayed checksums now because we send before
 725                  * this is done in the normal processing path.
 726                  */
 727                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 728                         in_delayed_cksum(m);
 729                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 730                 }
 731
 732                 ip->ip_len = htons(ip->ip_len);
 733                 ip->ip_off = htons(ip->ip_off);
 734
 735                 /* NB: callee frees mbuf */
 736                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 737                 /*
 738                  * Preserve KAME behaviour: ENOENT can be returned
 739                  * when an SA acquire is in progress.  Don't propagate
 740                  * this to user-level; it confuses applications.
 741                  *
 742                  * XXX this will go away when the SADB is redone.
 743                  */
 744                 if (error == ENOENT)
 745                         error = 0;
 746                 crit_exit();
 747                 goto done;
 748         } else {
 749                 crit_exit();
 750
 751                 if (error != 0) {
 752                         /*
 753                          * Hack: -EINVAL is used to signal that a packet
 754                          * should be silently discarded.  This is typically
 755                          * because we asked key management for an SA and
 756                          * it was delayed (e.g. kicked up to IKE).
 757                          */
 758                         if (error == -EINVAL)
 759                                 error = 0;
 760                         goto bad;
 761                 } else {
 762                         /* No IPsec processing for this packet. */
 763                 }
 764 #ifdef notyet
 765                 /*
 766                  * If deferred crypto processing is needed, check that
 767                  * the interface supports it.
 768                  */
 769                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 770                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 771                         /* notify IPsec to do its own crypto */
 772                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 773                         error = EHOSTUNREACH;
 774                         goto bad;
 775                 }
 776 #endif
 777         }
 778 spd_done:
 779 #endif /* FAST_IPSEC */
 780
 781         /* We are already being fwd'd from a firewall. */
 782         if (next_hop != NULL)
 783                 goto pass;
 784
 785         /* No pfil hooks */
 786         if (!pfil_has_hooks(&inet_pfil_hook)) {
 787                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 788                         /*
 789                          * Strip dummynet tags from stranded packets
 790                          */
 791                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 792                         KKASSERT(mtag != NULL);
 793                         m_tag_delete(m, mtag);
 794                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 795                 }
 796                 goto pass;
 797         }
 798
 799         /*
 800          * IpHack's section.
 801          * - Xlate: translate packet's addr/port (NAT).
 802          * - Firewall: deny/allow/etc.
 803          * - Wrap: fake packet's addr/port <unimpl.>
 804          * - Encapsulate: put it in another IP and send out. <unimp.>
 805          */
 806
 807         /*
 808          * Run through list of hooks for output packets.
 809          */
 810         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 811         if (error != 0 || m == NULL)
 812                 goto done;
 813         ip = mtod(m, struct ip *);
 814
 815         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 816                 /*
 817                  * Check dst to make sure it is directly reachable on the
 818                  * interface we previously thought it was.
 819                  * If it isn't (which may be likely in some situations) we have
 820                  * to re-route it (ie, find a route for the next-hop and the
 821                  * associated interface) and set them here. This is nested
 822                  * forwarding which in most cases is undesirable, except where
 823                  * such control is nigh impossible. So we do it here.
 824                  * And I'm babbling.
 825                  */
 826                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 827                 KKASSERT(mtag != NULL);
 828                 next_hop = m_tag_data(mtag);
 829
 830                 /*
 831                  * Try local forwarding first
 832                  */
 833                 if (ip_localforward(m, next_hop, hlen))
 834                         goto done;
 835
 836                 /*
 837                  * Relocate the route based on next_hop.
 838                  * If the current route is inp's cache, keep it untouched.
 839                  */
 840                 if (ro == &iproute && ro->ro_rt != NULL) {
 841                         RTFREE(ro->ro_rt);
 842                         ro->ro_rt = NULL;
 843                 }
 844                 ro = &iproute;
 845                 bzero(ro, sizeof *ro);
 846
 847                 /*
 848                  * Forwarding to broadcast address is not allowed.
 849                  * XXX Should we follow IP_ROUTETOIF?
 850                  */
 851                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 852
 853                 /* We are doing forwarding now */
 854                 flags |= IP_FORWARDING;
 855
 856                 goto reroute;
 857         }
 858
 859         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 860                 struct dn_pkt *dn_pkt;
 861
 862                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 863                 KKASSERT(mtag != NULL);
 864                 dn_pkt = m_tag_data(mtag);
 865
 866                 /*
 867                  * Under certain cases it is not possible to recalculate
 868                  * 'ro' and 'dst', let alone 'flags', so just save them in
 869                  * dummynet tag and avoid the possible wrong reculcalation
 870                  * when we come back to ip_output() again.
 871                  *
 872                  * All other parameters have been already used and so they
 873                  * are not needed anymore.
 874                  * XXX if the ifp is deleted while a pkt is in dummynet,
 875                  * we are in trouble! (TODO use ifnet_detach_event)
 876                  *
 877                  * We need to copy *ro because for ICMP pkts (and maybe
 878                  * others) the caller passed a pointer into the stack;
 879                  * dst might also be a pointer into *ro so it needs to
 880                  * be updated.
 881                  */
 882                 dn_pkt->ro = *ro;
 883                 if (ro->ro_rt)
 884                         ro->ro_rt->rt_refcnt++;
 885                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 886                         /* 'dst' points into 'ro' */
 887                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 888                 }
 889                 dn_pkt->dn_dst = dst;
 890                 dn_pkt->flags = flags;
 891
 892                 ip_dn_queue(m);
 893                 goto done;
 894         }
 895 pass:
 896         /* 127/8 must not appear on wire - RFC1122. */
 897         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 898             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 899                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 900                         ipstat.ips_badaddr++;
 901                         error = EADDRNOTAVAIL;
 902                         goto bad;
 903                 }
 904         }
 905         if (ip->ip_src.s_addr == INADDR_ANY ||
 906             IN_MULTICAST(ntohl(ip->ip_src.s_addr))) {
 907                 ipstat.ips_badaddr++;
 908                 error = EADDRNOTAVAIL;
 909                 goto bad;
 910         }
 911
 912         if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
 913                 m->m_pkthdr.csum_flags |= CSUM_IP;
 914                 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 915                 if (sw_csum & CSUM_DELAY_DATA) {
 916                         in_delayed_cksum(m);
 917                         sw_csum &= ~CSUM_DELAY_DATA;
 918                 }
 919                 m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 920         } else {
 921                 sw_csum = 0;
 922         }
 923         m->m_pkthdr.csum_iphlen = hlen;
 924
 925         /*
 926          * If small enough for interface, or the interface will take
 927          * care of the fragmentation or segmentation for us, can just
 928          * send directly.
 929          */
 930         if (ip->ip_len <= ifp->if_mtu ||
 931             ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
 932             (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 933                 ip->ip_len = htons(ip->ip_len);
 934                 ip->ip_off = htons(ip->ip_off);
 935                 ip->ip_sum = 0;
 936                 if (sw_csum & CSUM_DELAY_IP) {
 937                         if (ip->ip_vhl == IP_VHL_BORING)
 938                                 ip->ip_sum = in_cksum_hdr(ip);
 939                         else
 940                                 ip->ip_sum = in_cksum(m, hlen);
 941                 }
 942
 943                 /* Record statistics for this interface address. */
 944                 if (!(flags & IP_FORWARDING) && ia) {
 945                         IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
 946                         IFA_STAT_INC(&ia->ia_ifa, obytes, m->m_pkthdr.len);
 947                 }
 948
 949 #ifdef IPSEC
 950                 /* clean ipsec history once it goes out of the node */
 951                 ipsec_delaux(m);
 952 #endif
 953
 954 #ifdef MBUF_STRESS_TEST
 955                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 956                         struct mbuf *m1, *m2;
 957                         int length, tmp;
 958
 959                         tmp = length = m->m_pkthdr.len;
 960
 961                         while ((length -= mbuf_frag_size) >= 1) {
 962                                 m1 = m_split(m, length, MB_DONTWAIT);
 963                                 if (m1 == NULL)
 964                                         break;
 965                                 m2 = m;
 966                                 while (m2->m_next != NULL)
 967                                         m2 = m2->m_next;
 968                                 m2->m_next = m1;
 969                         }
 970                         m->m_pkthdr.len = tmp;
 971                 }
 972 #endif
 973
 974 #ifdef MPLS
 975                 if (!mpls_output_process(m, ro->ro_rt))
 976                         goto done;
 977 #endif
 978                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 979                                        ro->ro_rt);
 980                 goto done;
 981         }
 982
 983         if (ip->ip_off & IP_DF) {
 984                 error = EMSGSIZE;
 985                 /*
 986                  * This case can happen if the user changed the MTU
 987                  * of an interface after enabling IP on it.  Because
 988                  * most netifs don't keep track of routes pointing to
 989                  * them, there is no way for one to update all its
 990                  * routes when the MTU is changed.
 991                  */
 992                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 993                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 994                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 995                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 996                 }
 997                 ipstat.ips_cantfrag++;
 998                 goto bad;
 999         }
1000
1001         /*
1002          * Too large for interface; fragment if possible. If successful,
1003          * on return, m will point to a list of packets to be sent.
1004          */
1005         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1006         if (error)
1007                 goto bad;
1008         for (; m; m = m0) {
1009                 m0 = m->m_nextpkt;
1010                 m->m_nextpkt = NULL;
1011 #ifdef IPSEC
1012                 /* clean ipsec history once it goes out of the node */
1013                 ipsec_delaux(m);
1014 #endif
1015                 if (error == 0) {
1016                         /* Record statistics for this interface address. */
1017                         if (ia != NULL) {
1018                                 IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
1019                                 IFA_STAT_INC(&ia->ia_ifa, obytes,
1020                                     m->m_pkthdr.len);
1021                         }
1022 #ifdef MPLS
1023                         if (!mpls_output_process(m, ro->ro_rt))
1024                                 continue;
1025 #endif
1026                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1027                                                ro->ro_rt);
1028                 } else {
1029                         m_freem(m);
1030                 }
1031         }
1032
1033         if (error == 0)
1034                 ipstat.ips_fragmented++;
1035
1036 done:
1037         if (ro == &iproute && ro->ro_rt != NULL) {
1038                 RTFREE(ro->ro_rt);
1039                 ro->ro_rt = NULL;
1040         }
1041 #ifdef IPSEC
1042         if (sp != NULL) {
1043                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1044                         kprintf("DP ip_output call free SP:%p\n", sp));
1045                 key_freesp(sp);
1046         }
1047 #endif
1048 #ifdef FAST_IPSEC
1049         if (sp != NULL)
1050                 KEY_FREESP(&sp);
1051 #endif
1052         return (error);
1053 bad:
1054         m_freem(m);
1055         goto done;
1056 }
1057
1058 /*
1059  * Create a chain of fragments which fit the given mtu. m_frag points to the
1060  * mbuf to be fragmented; on return it points to the chain with the fragments.
1061  * Return 0 if no error. If error, m_frag may contain a partially built
1062  * chain of fragments that should be freed by the caller.
1063  *
1064  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1065  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1066  */
1067 int
1068 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1069             u_long if_hwassist_flags, int sw_csum)
1070 {
1071         int error = 0;
1072         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1073         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1074         int off;
1075         struct mbuf *m0 = *m_frag;      /* the original packet          */
1076         int firstlen;
1077         struct mbuf **mnext;
1078         int nfrags;
1079
1080         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1081                 ipstat.ips_cantfrag++;
1082                 return EMSGSIZE;
1083         }
1084
1085         /*
1086          * Must be able to put at least 8 bytes per fragment.
1087          */
1088         if (len < 8)
1089                 return EMSGSIZE;
1090
1091         /*
1092          * If the interface will not calculate checksums on
1093          * fragmented packets, then do it here.
1094          */
1095         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1096             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1097                 in_delayed_cksum(m0);
1098                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1099         }
1100
1101         if (len > PAGE_SIZE) {
1102                 /*
1103                  * Fragment large datagrams such that each segment
1104                  * contains a multiple of PAGE_SIZE amount of data,
1105                  * plus headers. This enables a receiver to perform
1106                  * page-flipping zero-copy optimizations.
1107                  *
1108                  * XXX When does this help given that sender and receiver
1109                  * could have different page sizes, and also mtu could
1110                  * be less than the receiver's page size ?
1111                  */
1112                 int newlen;
1113                 struct mbuf *m;
1114
1115                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1116                         off += m->m_len;
1117
1118                 /*
1119                  * firstlen (off - hlen) must be aligned on an
1120                  * 8-byte boundary
1121                  */
1122                 if (off < hlen)
1123                         goto smart_frag_failure;
1124                 off = ((off - hlen) & ~7) + hlen;
1125                 newlen = (~PAGE_MASK) & mtu;
1126                 if ((newlen + sizeof(struct ip)) > mtu) {
1127                         /* we failed, go back the default */
1128 smart_frag_failure:
1129                         newlen = len;
1130                         off = hlen + len;
1131                 }
1132                 len = newlen;
1133
1134         } else {
1135                 off = hlen + len;
1136         }
1137
1138         firstlen = off - hlen;
1139         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1140
1141         /*
1142          * Loop through length of segment after first fragment,
1143          * make new header and copy data of each part and link onto chain.
1144          * Here, m0 is the original packet, m is the fragment being created.
1145          * The fragments are linked off the m_nextpkt of the original
1146          * packet, which after processing serves as the first fragment.
1147          */
1148         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1149                 struct ip *mhip;        /* ip header on the fragment */
1150                 struct mbuf *m;
1151                 int mhlen = sizeof(struct ip);
1152
1153                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1154                 if (m == NULL) {
1155                         error = ENOBUFS;
1156                         ipstat.ips_odropped++;
1157                         goto done;
1158                 }
1159                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1160                 /*
1161                  * In the first mbuf, leave room for the link header, then
1162                  * copy the original IP header including options. The payload
1163                  * goes into an additional mbuf chain returned by m_copy().
1164                  */
1165                 m->m_data += max_linkhdr;
1166                 mhip = mtod(m, struct ip *);
1167                 *mhip = *ip;
1168                 if (hlen > sizeof(struct ip)) {
1169                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1170                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1171                 }
1172                 m->m_len = mhlen;
1173                 /* XXX do we need to add ip->ip_off below ? */
1174                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1175                 if (off + len >= ip->ip_len) {  /* last fragment */
1176                         len = ip->ip_len - off;
1177                         m->m_flags |= M_LASTFRAG;
1178                 } else
1179                         mhip->ip_off |= IP_MF;
1180                 mhip->ip_len = htons((u_short)(len + mhlen));
1181                 m->m_next = m_copy(m0, off, len);
1182                 if (m->m_next == NULL) {                /* copy failed */
1183                         m_free(m);
1184                         error = ENOBUFS;        /* ??? */
1185                         ipstat.ips_odropped++;
1186                         goto done;
1187                 }
1188                 m->m_pkthdr.len = mhlen + len;
1189                 m->m_pkthdr.rcvif = NULL;
1190                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1191                 m->m_pkthdr.csum_iphlen = mhlen;
1192                 mhip->ip_off = htons(mhip->ip_off);
1193                 mhip->ip_sum = 0;
1194                 if (sw_csum & CSUM_DELAY_IP)
1195                         mhip->ip_sum = in_cksum(m, mhlen);
1196                 *mnext = m;
1197                 mnext = &m->m_nextpkt;
1198         }
1199         ipstat.ips_ofragments += nfrags;
1200
1201         /* set first marker for fragment chain */
1202         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1203         m0->m_pkthdr.csum_data = nfrags;
1204
1205         /*
1206          * Update first fragment by trimming what's been copied out
1207          * and updating header.
1208          */
1209         m_adj(m0, hlen + firstlen - ip->ip_len);
1210         m0->m_pkthdr.len = hlen + firstlen;
1211         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1212         ip->ip_off |= IP_MF;
1213         ip->ip_off = htons(ip->ip_off);
1214         ip->ip_sum = 0;
1215         if (sw_csum & CSUM_DELAY_IP)
1216                 ip->ip_sum = in_cksum(m0, hlen);
1217
1218 done:
1219         *m_frag = m0;
1220         return error;
1221 }
1222
1223 void
1224 in_delayed_cksum(struct mbuf *m)
1225 {
1226         struct ip *ip;
1227         u_short csum, offset;
1228
1229         ip = mtod(m, struct ip *);
1230         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1231         csum = in_cksum_skip(m, ip->ip_len, offset);
1232         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1233                 csum = 0xffff;
1234         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1235
1236         if (offset + sizeof(u_short) > m->m_len) {
1237                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1238                     m->m_len, offset, ip->ip_p);
1239                 /*
1240                  * XXX
1241                  * this shouldn't happen, but if it does, the
1242                  * correct behavior may be to insert the checksum
1243                  * in the existing chain instead of rearranging it.
1244                  */
1245                 m = m_pullup(m, offset + sizeof(u_short));
1246         }
1247         *(u_short *)(m->m_data + offset) = csum;
1248 }
1249
1250 /*
1251  * Insert IP options into preformed packet.
1252  * Adjust IP destination as required for IP source routing,
1253  * as indicated by a non-zero in_addr at the start of the options.
1254  *
1255  * XXX This routine assumes that the packet has no options in place.
1256  */
1257 static struct mbuf *
1258 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1259 {
1260         struct ipoption *p = mtod(opt, struct ipoption *);
1261         struct mbuf *n;
1262         struct ip *ip = mtod(m, struct ip *);
1263         unsigned optlen;
1264
1265         optlen = opt->m_len - sizeof p->ipopt_dst;
1266         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1267                 *phlen = 0;
1268                 return (m);             /* XXX should fail */
1269         }
1270         if (p->ipopt_dst.s_addr)
1271                 ip->ip_dst = p->ipopt_dst;
1272         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1273                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1274                 if (n == NULL) {
1275                         *phlen = 0;
1276                         return (m);
1277                 }
1278                 n->m_pkthdr.rcvif = NULL;
1279                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1280                 m->m_len -= sizeof(struct ip);
1281                 m->m_data += sizeof(struct ip);
1282                 n->m_next = m;
1283                 m = n;
1284                 m->m_len = optlen + sizeof(struct ip);
1285                 m->m_data += max_linkhdr;
1286                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1287         } else {
1288                 m->m_data -= optlen;
1289                 m->m_len += optlen;
1290                 m->m_pkthdr.len += optlen;
1291                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1292         }
1293         ip = mtod(m, struct ip *);
1294         bcopy(p->ipopt_list, ip + 1, optlen);
1295         *phlen = sizeof(struct ip) + optlen;
1296         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1297         ip->ip_len += optlen;
1298         return (m);
1299 }
1300
1301 /*
1302  * Copy options from ip to jp,
1303  * omitting those not copied during fragmentation.
1304  */
1305 int
1306 ip_optcopy(struct ip *ip, struct ip *jp)
1307 {
1308         u_char *cp, *dp;
1309         int opt, optlen, cnt;
1310
1311         cp = (u_char *)(ip + 1);
1312         dp = (u_char *)(jp + 1);
1313         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1314         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1315                 opt = cp[0];
1316                 if (opt == IPOPT_EOL)
1317                         break;
1318                 if (opt == IPOPT_NOP) {
1319                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1320                         *dp++ = IPOPT_NOP;
1321                         optlen = 1;
1322                         continue;
1323                 }
1324
1325                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1326                     ("ip_optcopy: malformed ipv4 option"));
1327                 optlen = cp[IPOPT_OLEN];
1328                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1329                     ("ip_optcopy: malformed ipv4 option"));
1330
1331                 /* bogus lengths should have been caught by ip_dooptions */
1332                 if (optlen > cnt)
1333                         optlen = cnt;
1334                 if (IPOPT_COPIED(opt)) {
1335                         bcopy(cp, dp, optlen);
1336                         dp += optlen;
1337                 }
1338         }
1339         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1340                 *dp++ = IPOPT_EOL;
1341         return (optlen);
1342 }
1343
1344 /*
1345  * IP socket option processing.
1346  */
1347 void
1348 ip_ctloutput(netmsg_t msg)
1349 {
1350         struct socket *so = msg->base.nm_so;
1351         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1352         struct  inpcb *inp = so->so_pcb;
1353         int     error, optval;
1354
1355         error = optval = 0;
1356         if (sopt->sopt_level != IPPROTO_IP) {
1357                 error = EINVAL;
1358                 goto done;
1359         }
1360
1361         switch (sopt->sopt_dir) {
1362         case SOPT_SET:
1363                 switch (sopt->sopt_name) {
1364                 case IP_OPTIONS:
1365 #ifdef notyet
1366                 case IP_RETOPTS:
1367 #endif
1368                 {
1369                         struct mbuf *m;
1370                         if (sopt->sopt_valsize > MLEN) {
1371                                 error = EMSGSIZE;
1372                                 break;
1373                         }
1374                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1375                         if (m == NULL) {
1376                                 error = ENOBUFS;
1377                                 break;
1378                         }
1379                         m->m_len = sopt->sopt_valsize;
1380                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1381                                               m->m_len);
1382                         error = ip_pcbopts(sopt->sopt_name,
1383                                            &inp->inp_options, m);
1384                         goto done;
1385                 }
1386
1387                 case IP_TOS:
1388                 case IP_TTL:
1389                 case IP_MINTTL:
1390                 case IP_RECVOPTS:
1391                 case IP_RECVRETOPTS:
1392                 case IP_RECVDSTADDR:
1393                 case IP_RECVIF:
1394                 case IP_RECVTTL:
1395                 case IP_FAITH:
1396                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1397                                              sizeof optval);
1398                         if (error)
1399                                 break;
1400                         switch (sopt->sopt_name) {
1401                         case IP_TOS:
1402                                 inp->inp_ip_tos = optval;
1403                                 break;
1404
1405                         case IP_TTL:
1406                                 inp->inp_ip_ttl = optval;
1407                                 break;
1408                         case IP_MINTTL:
1409                                 if (optval >= 0 && optval <= MAXTTL)
1410                                         inp->inp_ip_minttl = optval;
1411                                 else
1412                                         error = EINVAL;
1413                                 break;
1414 #define OPTSET(bit) \
1415         if (optval) \
1416                 inp->inp_flags |= bit; \
1417         else \
1418                 inp->inp_flags &= ~bit;
1419
1420                         case IP_RECVOPTS:
1421                                 OPTSET(INP_RECVOPTS);
1422                                 break;
1423
1424                         case IP_RECVRETOPTS:
1425                                 OPTSET(INP_RECVRETOPTS);
1426                                 break;
1427
1428                         case IP_RECVDSTADDR:
1429                                 OPTSET(INP_RECVDSTADDR);
1430                                 break;
1431
1432                         case IP_RECVIF:
1433                                 OPTSET(INP_RECVIF);
1434                                 break;
1435
1436                         case IP_RECVTTL:
1437                                 OPTSET(INP_RECVTTL);
1438                                 break;
1439
1440                         case IP_FAITH:
1441                                 OPTSET(INP_FAITH);
1442                                 break;
1443                         }
1444                         break;
1445 #undef OPTSET
1446
1447                 case IP_MULTICAST_IF:
1448                 case IP_MULTICAST_VIF:
1449                 case IP_MULTICAST_TTL:
1450                 case IP_MULTICAST_LOOP:
1451                 case IP_ADD_MEMBERSHIP:
1452                 case IP_DROP_MEMBERSHIP:
1453                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1454                         break;
1455
1456                 case IP_PORTRANGE:
1457                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1458                                             sizeof optval);
1459                         if (error)
1460                                 break;
1461
1462                         switch (optval) {
1463                         case IP_PORTRANGE_DEFAULT:
1464                                 inp->inp_flags &= ~(INP_LOWPORT);
1465                                 inp->inp_flags &= ~(INP_HIGHPORT);
1466                                 break;
1467
1468                         case IP_PORTRANGE_HIGH:
1469                                 inp->inp_flags &= ~(INP_LOWPORT);
1470                                 inp->inp_flags |= INP_HIGHPORT;
1471                                 break;
1472
1473                         case IP_PORTRANGE_LOW:
1474                                 inp->inp_flags &= ~(INP_HIGHPORT);
1475                                 inp->inp_flags |= INP_LOWPORT;
1476                                 break;
1477
1478                         default:
1479                                 error = EINVAL;
1480                                 break;
1481                         }
1482                         break;
1483
1484 #if defined(IPSEC) || defined(FAST_IPSEC)
1485                 case IP_IPSEC_POLICY:
1486                 {
1487                         caddr_t req;
1488                         size_t len = 0;
1489                         int priv;
1490                         struct mbuf *m;
1491                         int optname;
1492
1493                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1494                                 break;
1495                         soopt_to_mbuf(sopt, m);
1496                         priv = (sopt->sopt_td != NULL &&
1497                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1498                         req = mtod(m, caddr_t);
1499                         len = m->m_len;
1500                         optname = sopt->sopt_name;
1501                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1502                         m_freem(m);
1503                         break;
1504                 }
1505 #endif /*IPSEC*/
1506
1507                 default:
1508                         error = ENOPROTOOPT;
1509                         break;
1510                 }
1511                 break;
1512
1513         case SOPT_GET:
1514                 switch (sopt->sopt_name) {
1515                 case IP_OPTIONS:
1516                 case IP_RETOPTS:
1517                         if (inp->inp_options)
1518                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1519                                                            char *),
1520                                                 inp->inp_options->m_len);
1521                         else
1522                                 sopt->sopt_valsize = 0;
1523                         break;
1524
1525                 case IP_TOS:
1526                 case IP_TTL:
1527                 case IP_MINTTL:
1528                 case IP_RECVOPTS:
1529                 case IP_RECVRETOPTS:
1530                 case IP_RECVDSTADDR:
1531                 case IP_RECVTTL:
1532                 case IP_RECVIF:
1533                 case IP_PORTRANGE:
1534                 case IP_FAITH:
1535                         switch (sopt->sopt_name) {
1536
1537                         case IP_TOS:
1538                                 optval = inp->inp_ip_tos;
1539                                 break;
1540
1541                         case IP_TTL:
1542                                 optval = inp->inp_ip_ttl;
1543                                 break;
1544                         case IP_MINTTL:
1545                                 optval = inp->inp_ip_minttl;
1546                                 break;
1547
1548 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1549
1550                         case IP_RECVOPTS:
1551                                 optval = OPTBIT(INP_RECVOPTS);
1552                                 break;
1553
1554                         case IP_RECVRETOPTS:
1555                                 optval = OPTBIT(INP_RECVRETOPTS);
1556                                 break;
1557
1558                         case IP_RECVDSTADDR:
1559                                 optval = OPTBIT(INP_RECVDSTADDR);
1560                                 break;
1561
1562                         case IP_RECVTTL:
1563                                 optval = OPTBIT(INP_RECVTTL);
1564                                 break;
1565
1566                         case IP_RECVIF:
1567                                 optval = OPTBIT(INP_RECVIF);
1568                                 break;
1569
1570                         case IP_PORTRANGE:
1571                                 if (inp->inp_flags & INP_HIGHPORT)
1572                                         optval = IP_PORTRANGE_HIGH;
1573                                 else if (inp->inp_flags & INP_LOWPORT)
1574                                         optval = IP_PORTRANGE_LOW;
1575                                 else
1576                                         optval = 0;
1577                                 break;
1578
1579                         case IP_FAITH:
1580                                 optval = OPTBIT(INP_FAITH);
1581                                 break;
1582                         }
1583                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1584                         break;
1585
1586                 case IP_MULTICAST_IF:
1587                 case IP_MULTICAST_VIF:
1588                 case IP_MULTICAST_TTL:
1589                 case IP_MULTICAST_LOOP:
1590                 case IP_ADD_MEMBERSHIP:
1591                 case IP_DROP_MEMBERSHIP:
1592                         error = ip_getmoptions(sopt, inp->inp_moptions);
1593                         break;
1594
1595 #if defined(IPSEC) || defined(FAST_IPSEC)
1596                 case IP_IPSEC_POLICY:
1597                 {
1598                         struct mbuf *m = NULL;
1599                         caddr_t req = NULL;
1600                         size_t len = 0;
1601
1602                         if (m != NULL) {
1603                                 req = mtod(m, caddr_t);
1604                                 len = m->m_len;
1605                         }
1606                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1607                         if (error == 0)
1608                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1609                         if (error == 0)
1610                                 m_freem(m);
1611                         break;
1612                 }
1613 #endif /*IPSEC*/
1614
1615                 default:
1616                         error = ENOPROTOOPT;
1617                         break;
1618                 }
1619                 break;
1620         }
1621 done:
1622         lwkt_replymsg(&msg->lmsg, error);
1623 }
1624
1625 /*
1626  * Set up IP options in pcb for insertion in output packets.
1627  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1628  * with destination address if source routed.
1629  */
1630 static int
1631 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1632 {
1633         int cnt, optlen;
1634         u_char *cp;
1635         u_char opt;
1636
1637         /* turn off any old options */
1638         if (*pcbopt)
1639                 m_free(*pcbopt);
1640         *pcbopt = NULL;
1641         if (m == NULL || m->m_len == 0) {
1642                 /*
1643                  * Only turning off any previous options.
1644                  */
1645                 if (m != NULL)
1646                         m_free(m);
1647                 return (0);
1648         }
1649
1650         if (m->m_len % sizeof(int32_t))
1651                 goto bad;
1652         /*
1653          * IP first-hop destination address will be stored before
1654          * actual options; move other options back
1655          * and clear it when none present.
1656          */
1657         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1658                 goto bad;
1659         cnt = m->m_len;
1660         m->m_len += sizeof(struct in_addr);
1661         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1662         ovbcopy(mtod(m, caddr_t), cp, cnt);
1663         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1664
1665         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1666                 opt = cp[IPOPT_OPTVAL];
1667                 if (opt == IPOPT_EOL)
1668                         break;
1669                 if (opt == IPOPT_NOP)
1670                         optlen = 1;
1671                 else {
1672                         if (cnt < IPOPT_OLEN + sizeof *cp)
1673                                 goto bad;
1674                         optlen = cp[IPOPT_OLEN];
1675                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1676                                 goto bad;
1677                 }
1678                 switch (opt) {
1679
1680                 default:
1681                         break;
1682
1683                 case IPOPT_LSRR:
1684                 case IPOPT_SSRR:
1685                         /*
1686                          * user process specifies route as:
1687                          *      ->A->B->C->D
1688                          * D must be our final destination (but we can't
1689                          * check that since we may not have connected yet).
1690                          * A is first hop destination, which doesn't appear in
1691                          * actual IP option, but is stored before the options.
1692                          */
1693                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1694                                 goto bad;
1695                         m->m_len -= sizeof(struct in_addr);
1696                         cnt -= sizeof(struct in_addr);
1697                         optlen -= sizeof(struct in_addr);
1698                         cp[IPOPT_OLEN] = optlen;
1699                         /*
1700                          * Move first hop before start of options.
1701                          */
1702                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1703                               sizeof(struct in_addr));
1704                         /*
1705                          * Then copy rest of options back
1706                          * to close up the deleted entry.
1707                          */
1708                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1709                                 &cp[IPOPT_OFFSET+1],
1710                                 cnt - (IPOPT_MINOFF - 1));
1711                         break;
1712                 }
1713         }
1714         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1715                 goto bad;
1716         *pcbopt = m;
1717         return (0);
1718
1719 bad:
1720         m_free(m);
1721         return (EINVAL);
1722 }
1723
1724 /*
1725  * XXX
1726  * The whole multicast option thing needs to be re-thought.
1727  * Several of these options are equally applicable to non-multicast
1728  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1729  * standard option (IP_TTL).
1730  */
1731
1732 /*
1733  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1734  */
1735 static struct ifnet *
1736 ip_multicast_if(struct in_addr *a, int *ifindexp)
1737 {
1738         int ifindex;
1739         struct ifnet *ifp;
1740
1741         if (ifindexp)
1742                 *ifindexp = 0;
1743         if (ntohl(a->s_addr) >> 24 == 0) {
1744                 ifindex = ntohl(a->s_addr) & 0xffffff;
1745                 if (ifindex < 0 || if_index < ifindex)
1746                         return NULL;
1747                 ifp = ifindex2ifnet[ifindex];
1748                 if (ifindexp)
1749                         *ifindexp = ifindex;
1750         } else {
1751                 ifp = INADDR_TO_IFP(a);
1752         }
1753         return ifp;
1754 }
1755
1756 /*
1757  * Set the IP multicast options in response to user setsockopt().
1758  */
1759 static int
1760 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1761 {
1762         int error = 0;
1763         int i;
1764         struct in_addr addr;
1765         struct ip_mreq mreq;
1766         struct ifnet *ifp;
1767         struct ip_moptions *imo = *imop;
1768         int ifindex;
1769
1770         if (imo == NULL) {
1771                 /*
1772                  * No multicast option buffer attached to the pcb;
1773                  * allocate one and initialize to default values.
1774                  */
1775                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1776
1777                 *imop = imo;
1778                 imo->imo_multicast_ifp = NULL;
1779                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1780                 imo->imo_multicast_vif = -1;
1781                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1782                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1783                 imo->imo_num_memberships = 0;
1784         }
1785         switch (sopt->sopt_name) {
1786         /* store an index number for the vif you wanna use in the send */
1787         case IP_MULTICAST_VIF:
1788                 if (legal_vif_num == 0) {
1789                         error = EOPNOTSUPP;
1790                         break;
1791                 }
1792                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1793                 if (error)
1794                         break;
1795                 if (!legal_vif_num(i) && (i != -1)) {
1796                         error = EINVAL;
1797                         break;
1798                 }
1799                 imo->imo_multicast_vif = i;
1800                 break;
1801
1802         case IP_MULTICAST_IF:
1803                 /*
1804                  * Select the interface for outgoing multicast packets.
1805                  */
1806                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1807                 if (error)
1808                         break;
1809
1810                 /*
1811                  * INADDR_ANY is used to remove a previous selection.
1812                  * When no interface is selected, a default one is
1813                  * chosen every time a multicast packet is sent.
1814                  */
1815                 if (addr.s_addr == INADDR_ANY) {
1816                         imo->imo_multicast_ifp = NULL;
1817                         break;
1818                 }
1819                 /*
1820                  * The selected interface is identified by its local
1821                  * IP address.  Find the interface and confirm that
1822                  * it supports multicasting.
1823                  */
1824                 crit_enter();
1825                 ifp = ip_multicast_if(&addr, &ifindex);
1826                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1827                         crit_exit();
1828                         error = EADDRNOTAVAIL;
1829                         break;
1830                 }
1831                 imo->imo_multicast_ifp = ifp;
1832                 if (ifindex)
1833                         imo->imo_multicast_addr = addr;
1834                 else
1835                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1836                 crit_exit();
1837                 break;
1838
1839         case IP_MULTICAST_TTL:
1840                 /*
1841                  * Set the IP time-to-live for outgoing multicast packets.
1842                  * The original multicast API required a char argument,
1843                  * which is inconsistent with the rest of the socket API.
1844                  * We allow either a char or an int.
1845                  */
1846                 if (sopt->sopt_valsize == 1) {
1847                         u_char ttl;
1848                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1849                         if (error)
1850                                 break;
1851                         imo->imo_multicast_ttl = ttl;
1852                 } else {
1853                         u_int ttl;
1854                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1855                         if (error)
1856                                 break;
1857                         if (ttl > 255)
1858                                 error = EINVAL;
1859                         else
1860                                 imo->imo_multicast_ttl = ttl;
1861                 }
1862                 break;
1863
1864         case IP_MULTICAST_LOOP:
1865                 /*
1866                  * Set the loopback flag for outgoing multicast packets.
1867                  * Must be zero or one.  The original multicast API required a
1868                  * char argument, which is inconsistent with the rest
1869                  * of the socket API.  We allow either a char or an int.
1870                  */
1871                 if (sopt->sopt_valsize == 1) {
1872                         u_char loop;
1873
1874                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1875                         if (error)
1876                                 break;
1877                         imo->imo_multicast_loop = !!loop;
1878                 } else {
1879                         u_int loop;
1880
1881                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1882                                             sizeof loop);
1883                         if (error)
1884                                 break;
1885                         imo->imo_multicast_loop = !!loop;
1886                 }
1887                 break;
1888
1889         case IP_ADD_MEMBERSHIP:
1890                 /*
1891                  * Add a multicast group membership.
1892                  * Group must be a valid IP multicast address.
1893                  */
1894                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1895                 if (error)
1896                         break;
1897
1898                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1899                         error = EINVAL;
1900                         break;
1901                 }
1902                 crit_enter();
1903                 /*
1904                  * If no interface address was provided, use the interface of
1905                  * the route to the given multicast address.
1906                  */
1907                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1908                         struct sockaddr_in dst;
1909                         struct rtentry *rt;
1910
1911                         bzero(&dst, sizeof(struct sockaddr_in));
1912                         dst.sin_len = sizeof(struct sockaddr_in);
1913                         dst.sin_family = AF_INET;
1914                         dst.sin_addr = mreq.imr_multiaddr;
1915                         rt = rtlookup((struct sockaddr *)&dst);
1916                         if (rt == NULL) {
1917                                 error = EADDRNOTAVAIL;
1918                                 crit_exit();
1919                                 break;
1920                         }
1921                         --rt->rt_refcnt;
1922                         ifp = rt->rt_ifp;
1923                 } else {
1924                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1925                 }
1926
1927                 /*
1928                  * See if we found an interface, and confirm that it
1929                  * supports multicast.
1930                  */
1931                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1932                         error = EADDRNOTAVAIL;
1933                         crit_exit();
1934                         break;
1935                 }
1936                 /*
1937                  * See if the membership already exists or if all the
1938                  * membership slots are full.
1939                  */
1940                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1941                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1942                             imo->imo_membership[i]->inm_addr.s_addr
1943                                                 == mreq.imr_multiaddr.s_addr)
1944                                 break;
1945                 }
1946                 if (i < imo->imo_num_memberships) {
1947                         error = EADDRINUSE;
1948                         crit_exit();
1949                         break;
1950                 }
1951                 if (i == IP_MAX_MEMBERSHIPS) {
1952                         error = ETOOMANYREFS;
1953                         crit_exit();
1954                         break;
1955                 }
1956                 /*
1957                  * Everything looks good; add a new record to the multicast
1958                  * address list for the given interface.
1959                  */
1960                 if ((imo->imo_membership[i] =
1961                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1962                         error = ENOBUFS;
1963                         crit_exit();
1964                         break;
1965                 }
1966                 ++imo->imo_num_memberships;
1967                 crit_exit();
1968                 break;
1969
1970         case IP_DROP_MEMBERSHIP:
1971                 /*
1972                  * Drop a multicast group membership.
1973                  * Group must be a valid IP multicast address.
1974                  */
1975                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1976                 if (error)
1977                         break;
1978
1979                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1980                         error = EINVAL;
1981                         break;
1982                 }
1983
1984                 crit_enter();
1985                 /*
1986                  * If an interface address was specified, get a pointer
1987                  * to its ifnet structure.
1988                  */
1989                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1990                         ifp = NULL;
1991                 else {
1992                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1993                         if (ifp == NULL) {
1994                                 error = EADDRNOTAVAIL;
1995                                 crit_exit();
1996                                 break;
1997                         }
1998                 }
1999                 /*
2000                  * Find the membership in the membership array.
2001                  */
2002                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2003                         if ((ifp == NULL ||
2004                              imo->imo_membership[i]->inm_ifp == ifp) &&
2005                             imo->imo_membership[i]->inm_addr.s_addr ==
2006                             mreq.imr_multiaddr.s_addr)
2007                                 break;
2008                 }
2009                 if (i == imo->imo_num_memberships) {
2010                         error = EADDRNOTAVAIL;
2011                         crit_exit();
2012                         break;
2013                 }
2014                 /*
2015                  * Give up the multicast address record to which the
2016                  * membership points.
2017                  */
2018                 in_delmulti(imo->imo_membership[i]);
2019                 /*
2020                  * Remove the gap in the membership array.
2021                  */
2022                 for (++i; i < imo->imo_num_memberships; ++i)
2023                         imo->imo_membership[i-1] = imo->imo_membership[i];
2024                 --imo->imo_num_memberships;
2025                 crit_exit();
2026                 break;
2027
2028         default:
2029                 error = EOPNOTSUPP;
2030                 break;
2031         }
2032
2033         /*
2034          * If all options have default values, no need to keep the mbuf.
2035          */
2036         if (imo->imo_multicast_ifp == NULL &&
2037             imo->imo_multicast_vif == -1 &&
2038             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2039             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2040             imo->imo_num_memberships == 0) {
2041                 kfree(*imop, M_IPMOPTS);
2042                 *imop = NULL;
2043         }
2044
2045         return (error);
2046 }
2047
2048 /*
2049  * Return the IP multicast options in response to user getsockopt().
2050  */
2051 static int
2052 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2053 {
2054         struct in_addr addr;
2055         struct in_ifaddr *ia;
2056         int error, optval;
2057         u_char coptval;
2058
2059         error = 0;
2060         switch (sopt->sopt_name) {
2061         case IP_MULTICAST_VIF:
2062                 if (imo != NULL)
2063                         optval = imo->imo_multicast_vif;
2064                 else
2065                         optval = -1;
2066                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2067                 break;
2068
2069         case IP_MULTICAST_IF:
2070                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2071                         addr.s_addr = INADDR_ANY;
2072                 else if (imo->imo_multicast_addr.s_addr) {
2073                         /* return the value user has set */
2074                         addr = imo->imo_multicast_addr;
2075                 } else {
2076                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2077                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2078                                 : IA_SIN(ia)->sin_addr.s_addr;
2079                 }
2080                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2081                 break;
2082
2083         case IP_MULTICAST_TTL:
2084                 if (imo == NULL)
2085                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2086                 else
2087                         optval = coptval = imo->imo_multicast_ttl;
2088                 if (sopt->sopt_valsize == 1)
2089                         soopt_from_kbuf(sopt, &coptval, 1);
2090                 else
2091                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2092                 break;
2093
2094         case IP_MULTICAST_LOOP:
2095                 if (imo == NULL)
2096                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2097                 else
2098                         optval = coptval = imo->imo_multicast_loop;
2099                 if (sopt->sopt_valsize == 1)
2100                         soopt_from_kbuf(sopt, &coptval, 1);
2101                 else
2102                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2103                 break;
2104
2105         default:
2106                 error = ENOPROTOOPT;
2107                 break;
2108         }
2109         return (error);
2110 }
2111
2112 /*
2113  * Discard the IP multicast options.
2114  */
2115 void
2116 ip_freemoptions(struct ip_moptions *imo)
2117 {
2118         int i;
2119
2120         if (imo != NULL) {
2121                 for (i = 0; i < imo->imo_num_memberships; ++i)
2122                         in_delmulti(imo->imo_membership[i]);
2123                 kfree(imo, M_IPMOPTS);
2124         }
2125 }
2126
2127 /*
2128  * Routine called from ip_output() to loop back a copy of an IP multicast
2129  * packet to the input queue of a specified interface.  Note that this
2130  * calls the output routine of the loopback "driver", but with an interface
2131  * pointer that might NOT be a loopback interface -- evil, but easier than
2132  * replicating that code here.
2133  */
2134 static void
2135 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2136              int hlen)
2137 {
2138         struct ip *ip;
2139         struct mbuf *copym;
2140
2141         copym = m_copypacket(m, MB_DONTWAIT);
2142         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2143                 copym = m_pullup(copym, hlen);
2144         if (copym != NULL) {
2145                 /*
2146                  * if the checksum hasn't been computed, mark it as valid
2147                  */
2148                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2149                         in_delayed_cksum(copym);
2150                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2151                         copym->m_pkthdr.csum_flags |=
2152                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2153                         copym->m_pkthdr.csum_data = 0xffff;
2154                 }
2155                 /*
2156                  * We don't bother to fragment if the IP length is greater
2157                  * than the interface's MTU.  Can this possibly matter?
2158                  */
2159                 ip = mtod(copym, struct ip *);
2160                 ip->ip_len = htons(ip->ip_len);
2161                 ip->ip_off = htons(ip->ip_off);
2162                 ip->ip_sum = 0;
2163                 if (ip->ip_vhl == IP_VHL_BORING) {
2164                         ip->ip_sum = in_cksum_hdr(ip);
2165                 } else {
2166                         ip->ip_sum = in_cksum(copym, hlen);
2167                 }
2168                 /*
2169                  * NB:
2170                  * It's not clear whether there are any lingering
2171                  * reentrancy problems in other areas which might
2172                  * be exposed by using ip_input directly (in
2173                  * particular, everything which modifies the packet
2174                  * in-place).  Yet another option is using the
2175                  * protosw directly to deliver the looped back
2176                  * packet.  For the moment, we'll err on the side
2177                  * of safety by using if_simloop().
2178                  */
2179 #if 1 /* XXX */
2180                 if (dst->sin_family != AF_INET) {
2181                         kprintf("ip_mloopback: bad address family %d\n",
2182                                                 dst->sin_family);
2183                         dst->sin_family = AF_INET;
2184                 }
2185 #endif
2186                 get_mplock();   /* is if_simloop() mpsafe yet? */
2187                 if_simloop(ifp, copym, dst->sin_family, 0);
2188                 rel_mplock();
2189         }
2190 }