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  struct protosw inetsw[];
 121
 122 static int
 123 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
 124 {
 125         struct in_ifaddr_container *iac;
 126
 127         /*
 128          * We need to figure out if we have been forwarded to a local
 129          * socket.  If so, then we should somehow "loop back" to
 130          * ip_input(), and get directed to the PCB as if we had received
 131          * this packet.  This is because it may be difficult to identify
 132          * the packets you want to forward until they are being output
 133          * and have selected an interface (e.g. locally initiated
 134          * packets).  If we used the loopback inteface, we would not be
 135          * able to control what happens as the packet runs through
 136          * ip_input() as it is done through a ISR.
 137          */
 138         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 139                 /*
 140                  * If the addr to forward to is one of ours, we pretend
 141                  * to be the destination for this packet.
 142                  */
 143                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 144                         break;
 145         }
 146         if (iac != NULL) {
 147                 struct ip *ip;
 148
 149                 if (m->m_pkthdr.rcvif == NULL)
 150                         m->m_pkthdr.rcvif = ifunit("lo0");
 151                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 152                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 153                                                   CSUM_PSEUDO_HDR;
 154                         m->m_pkthdr.csum_data = 0xffff;
 155                 }
 156                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 157
 158                 /*
 159                  * Make sure that the IP header is in one mbuf,
 160                  * required by ip_input
 161                  */
 162                 if (m->m_len < hlen) {
 163                         m = m_pullup(m, hlen);
 164                         if (m == NULL) {
 165                                 /* The packet was freed; we are done */
 166                                 return 1;
 167                         }
 168                 }
 169                 ip = mtod(m, struct ip *);
 170
 171                 ip->ip_len = htons(ip->ip_len);
 172                 ip->ip_off = htons(ip->ip_off);
 173                 ip_input(m);
 174
 175                 return 1; /* The packet gets forwarded locally */
 176         }
 177         return 0;
 178 }
 179
 180 /*
 181  * IP output.  The packet in mbuf chain m contains a skeletal IP
 182  * header (with len, off, ttl, proto, tos, src, dst).
 183  * The mbuf chain containing the packet will be freed.
 184  * The mbuf opt, if present, will not be freed.
 185  */
 186 int
 187 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 188           int flags, struct ip_moptions *imo, struct inpcb *inp)
 189 {
 190         struct ip *ip;
 191         struct ifnet *ifp = NULL;       /* keep compiler happy */
 192         struct mbuf *m;
 193         int hlen = sizeof(struct ip);
 194         int len, error = 0;
 195         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 196         struct in_ifaddr *ia = NULL;
 197         int isbroadcast, sw_csum;
 198         struct in_addr pkt_dst;
 199         struct route iproute;
 200         struct m_tag *mtag;
 201 #ifdef IPSEC
 202         struct secpolicy *sp = NULL;
 203         struct socket *so = inp ? inp->inp_socket : NULL;
 204 #endif
 205 #ifdef FAST_IPSEC
 206         struct secpolicy *sp = NULL;
 207         struct tdb_ident *tdbi;
 208 #endif /* FAST_IPSEC */
 209         struct sockaddr_in *next_hop = NULL;
 210         int src_was_INADDR_ANY = 0;     /* as the name says... */
 211
 212         m = m0;
 213         M_ASSERTPKTHDR(m);
 214
 215         if (ro == NULL) {
 216                 ro = &iproute;
 217                 bzero(ro, sizeof *ro);
 218         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 219                 if (flags & IP_DEBUGROUTE) {
 220                         panic("ip_output: rt rt_cpuid %d accessed on cpu %d\n",
 221                             ro->ro_rt->rt_cpuid, mycpuid);
 222                 }
 223
 224                 /*
 225                  * XXX
 226                  * If the cached rtentry's owner CPU is not the current CPU,
 227                  * then don't touch the cached rtentry (remote free is too
 228                  * expensive in this context); just relocate the route.
 229                  */
 230                 ro = &iproute;
 231                 bzero(ro, sizeof *ro);
 232         }
 233
 234         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 235                 /* Next hop */
 236                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 237                 KKASSERT(mtag != NULL);
 238                 next_hop = m_tag_data(mtag);
 239         }
 240
 241         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 242                 struct dn_pkt *dn_pkt;
 243
 244                 /* Extract info from dummynet tag */
 245                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 246                 KKASSERT(mtag != NULL);
 247                 dn_pkt = m_tag_data(mtag);
 248
 249                 /*
 250                  * The packet was already tagged, so part of the
 251                  * processing was already done, and we need to go down.
 252                  * Get the calculated parameters from the tag.
 253                  */
 254                 ifp = dn_pkt->ifp;
 255
 256                 KKASSERT(ro == &iproute);
 257                 *ro = dn_pkt->ro; /* structure copy */
 258                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 259
 260                 dst = dn_pkt->dn_dst;
 261                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 262                         /* If 'dst' points into dummynet tag, adjust it */
 263                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 264                 }
 265
 266                 ip = mtod(m, struct ip *);
 267                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 268                 if (ro->ro_rt)
 269                         ia = ifatoia(ro->ro_rt->rt_ifa);
 270                 goto sendit;
 271         }
 272
 273         if (opt) {
 274                 len = 0;
 275                 m = ip_insertoptions(m, opt, &len);
 276                 if (len != 0)
 277                         hlen = len;
 278         }
 279         ip = mtod(m, struct ip *);
 280
 281         /*
 282          * Fill in IP header.
 283          */
 284         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 285                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 286                 ip->ip_off &= IP_DF;
 287                 ip->ip_id = ip_newid();
 288                 ipstat.ips_localout++;
 289         } else {
 290                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 291         }
 292
 293 reroute:
 294         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 295
 296         dst = (struct sockaddr_in *)&ro->ro_dst;
 297         /*
 298          * If there is a cached route,
 299          * check that it is to the same destination
 300          * and is still up.  If not, free it and try again.
 301          * The address family should also be checked in case of sharing the
 302          * cache with IPv6.
 303          */
 304         if (ro->ro_rt &&
 305             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 306              dst->sin_family != AF_INET ||
 307              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 308                 rtfree(ro->ro_rt);
 309                 ro->ro_rt = NULL;
 310         }
 311         if (ro->ro_rt == NULL) {
 312                 bzero(dst, sizeof *dst);
 313                 dst->sin_family = AF_INET;
 314                 dst->sin_len = sizeof *dst;
 315                 dst->sin_addr = pkt_dst;
 316         }
 317         /*
 318          * If routing to interface only,
 319          * short circuit routing lookup.
 320          */
 321         if (flags & IP_ROUTETOIF) {
 322                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 323                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 324                         ipstat.ips_noroute++;
 325                         error = ENETUNREACH;
 326                         goto bad;
 327                 }
 328                 ifp = ia->ia_ifp;
 329                 ip->ip_ttl = 1;
 330                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 331         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 332                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 333                 /*
 334                  * Bypass the normal routing lookup for multicast
 335                  * packets if the interface is specified.
 336                  */
 337                 ifp = imo->imo_multicast_ifp;
 338                 ia = IFP_TO_IA(ifp);
 339                 isbroadcast = 0;        /* fool gcc */
 340         } else {
 341                 /*
 342                  * If this is the case, we probably don't want to allocate
 343                  * a protocol-cloned route since we didn't get one from the
 344                  * ULP.  This lets TCP do its thing, while not burdening
 345                  * forwarding or ICMP with the overhead of cloning a route.
 346                  * Of course, we still want to do any cloning requested by
 347                  * the link layer, as this is probably required in all cases
 348                  * for correct operation (as it is for ARP).
 349                  */
 350                 if (ro->ro_rt == NULL)
 351                         rtalloc_ign(ro, RTF_PRCLONING);
 352                 if (ro->ro_rt == NULL) {
 353                         ipstat.ips_noroute++;
 354                         error = EHOSTUNREACH;
 355                         goto bad;
 356                 }
 357                 ia = ifatoia(ro->ro_rt->rt_ifa);
 358                 ifp = ro->ro_rt->rt_ifp;
 359                 ro->ro_rt->rt_use++;
 360                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 361                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 362                 if (ro->ro_rt->rt_flags & RTF_HOST)
 363                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 364                 else
 365                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 366         }
 367         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 368                 m->m_flags |= M_MCAST;
 369                 /*
 370                  * IP destination address is multicast.  Make sure "dst"
 371                  * still points to the address in "ro".  (It may have been
 372                  * changed to point to a gateway address, above.)
 373                  */
 374                 dst = (struct sockaddr_in *)&ro->ro_dst;
 375                 /*
 376                  * See if the caller provided any multicast options
 377                  */
 378                 if (imo != NULL) {
 379                         ip->ip_ttl = imo->imo_multicast_ttl;
 380                         if (imo->imo_multicast_vif != -1) {
 381                                 ip->ip_src.s_addr =
 382                                     ip_mcast_src ?
 383                                     ip_mcast_src(imo->imo_multicast_vif) :
 384                                     INADDR_ANY;
 385                         }
 386                 } else {
 387                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 388                 }
 389                 /*
 390                  * Confirm that the outgoing interface supports multicast.
 391                  */
 392                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 393                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 394                                 ipstat.ips_noroute++;
 395                                 error = ENETUNREACH;
 396                                 goto bad;
 397                         }
 398                 }
 399                 /*
 400                  * If source address not specified yet, use address of the
 401                  * outgoing interface.  In case, keep note we did that, so
 402                  * if the the firewall changes the next-hop causing the
 403                  * output interface to change, we can fix that.
 404                  */
 405                 if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 406                         /* Interface may have no addresses. */
 407                         if (ia != NULL) {
 408                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 409                                 src_was_INADDR_ANY = 1;
 410                         }
 411                 }
 412
 413                 if (ip->ip_src.s_addr != INADDR_ANY) {
 414                         struct in_multi *inm;
 415
 416                         inm = IN_LOOKUP_MULTI(&pkt_dst, ifp);
 417                         if (inm != NULL &&
 418                             (imo == NULL || imo->imo_multicast_loop)) {
 419                                 /*
 420                                  * If we belong to the destination multicast
 421                                  * group on the outgoing interface, and the
 422                                  * caller did not forbid loopback, loop back
 423                                  * a copy.
 424                                  */
 425                                 ip_mloopback(ifp, m, dst, hlen);
 426                         } else {
 427                                 /*
 428                                  * If we are acting as a multicast router,
 429                                  * perform multicast forwarding as if the
 430                                  * packet had just arrived on the interface
 431                                  * to which we are about to send.  The
 432                                  * multicast forwarding function recursively
 433                                  * calls this function, using the IP_FORWARDING
 434                                  * flag to prevent infinite recursion.
 435                                  *
 436                                  * Multicasts that are looped back by
 437                                  * ip_mloopback(), above, will be forwarded by
 438                                  * the ip_input() routine, if necessary.
 439                                  */
 440                                 if (ip_mrouter && !(flags & IP_FORWARDING)) {
 441                                         /*
 442                                          * If rsvp daemon is not running, do
 443                                          * not set ip_moptions. This ensures
 444                                          * that the packet is multicast and
 445                                          * not just sent down one link as
 446                                          * prescribed by rsvpd.
 447                                          */
 448                                         if (!rsvp_on)
 449                                                 imo = NULL;
 450                                         if (ip_mforward) {
 451                                                 get_mplock();
 452                                                 if (ip_mforward(ip, ifp,
 453                                                     m, imo) != 0) {
 454                                                         m_freem(m);
 455                                                         rel_mplock();
 456                                                         goto done;
 457                                                 }
 458                                                 rel_mplock();
 459                                         }
 460                                 }
 461                         }
 462                 }
 463
 464                 /*
 465                  * Multicasts with a time-to-live of zero may be looped-
 466                  * back, above, but must not be transmitted on a network.
 467                  * Also, multicasts addressed to the loopback interface
 468                  * are not sent -- the above call to ip_mloopback() will
 469                  * loop back a copy if this host actually belongs to the
 470                  * destination group on the loopback interface.
 471                  */
 472                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 473                         m_freem(m);
 474                         goto done;
 475                 }
 476
 477                 goto sendit;
 478         } else {
 479                 m->m_flags &= ~M_MCAST;
 480         }
 481
 482         /*
 483          * If the source address is not specified yet, use the address
 484          * of the outgoing interface.  In case, keep note we did that,
 485          * so if the the firewall changes the next-hop causing the output
 486          * interface to change, we can fix that.
 487          */
 488         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 489                 /* Interface may have no addresses. */
 490                 if (ia != NULL) {
 491                         ip->ip_src = IA_SIN(ia)->sin_addr;
 492                         src_was_INADDR_ANY = 1;
 493                 }
 494         }
 495
 496         /*
 497          * Look for broadcast address and
 498          * verify user is allowed to send
 499          * such a packet.
 500          */
 501         if (isbroadcast) {
 502                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 503                         error = EADDRNOTAVAIL;
 504                         goto bad;
 505                 }
 506                 if (!(flags & IP_ALLOWBROADCAST)) {
 507                         error = EACCES;
 508                         goto bad;
 509                 }
 510                 /* don't allow broadcast messages to be fragmented */
 511                 if (ip->ip_len > ifp->if_mtu) {
 512                         error = EMSGSIZE;
 513                         goto bad;
 514                 }
 515                 m->m_flags |= M_BCAST;
 516         } else {
 517                 m->m_flags &= ~M_BCAST;
 518         }
 519
 520 sendit:
 521 #ifdef IPSEC
 522         /* get SP for this packet */
 523         if (so == NULL)
 524                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 525         else
 526                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 527
 528         if (sp == NULL) {
 529                 ipsecstat.out_inval++;
 530                 goto bad;
 531         }
 532
 533         error = 0;
 534
 535         /* check policy */
 536         switch (sp->policy) {
 537         case IPSEC_POLICY_DISCARD:
 538                 /*
 539                  * This packet is just discarded.
 540                  */
 541                 ipsecstat.out_polvio++;
 542                 goto bad;
 543
 544         case IPSEC_POLICY_BYPASS:
 545         case IPSEC_POLICY_NONE:
 546         case IPSEC_POLICY_TCP:
 547                 /* no need to do IPsec. */
 548                 goto skip_ipsec;
 549
 550         case IPSEC_POLICY_IPSEC:
 551                 if (sp->req == NULL) {
 552                         /* acquire a policy */
 553                         error = key_spdacquire(sp);
 554                         goto bad;
 555                 }
 556                 break;
 557
 558         case IPSEC_POLICY_ENTRUST:
 559         default:
 560                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 561         }
 562     {
 563         struct ipsec_output_state state;
 564         bzero(&state, sizeof state);
 565         state.m = m;
 566         if (flags & IP_ROUTETOIF) {
 567                 state.ro = &iproute;
 568                 bzero(&iproute, sizeof iproute);
 569         } else
 570                 state.ro = ro;
 571         state.dst = (struct sockaddr *)dst;
 572
 573         ip->ip_sum = 0;
 574
 575         /*
 576          * XXX
 577          * delayed checksums are not currently compatible with IPsec
 578          */
 579         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 580                 in_delayed_cksum(m);
 581                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 582         }
 583
 584         ip->ip_len = htons(ip->ip_len);
 585         ip->ip_off = htons(ip->ip_off);
 586
 587         error = ipsec4_output(&state, sp, flags);
 588
 589         m = state.m;
 590         if (flags & IP_ROUTETOIF) {
 591                 /*
 592                  * if we have tunnel mode SA, we may need to ignore
 593                  * IP_ROUTETOIF.
 594                  */
 595                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 596                         flags &= ~IP_ROUTETOIF;
 597                         ro = state.ro;
 598                 }
 599         } else
 600                 ro = state.ro;
 601         dst = (struct sockaddr_in *)state.dst;
 602         if (error) {
 603                 /* mbuf is already reclaimed in ipsec4_output. */
 604                 m0 = NULL;
 605                 switch (error) {
 606                 case EHOSTUNREACH:
 607                 case ENETUNREACH:
 608                 case EMSGSIZE:
 609                 case ENOBUFS:
 610                 case ENOMEM:
 611                         break;
 612                 default:
 613                         kprintf("ip4_output (ipsec): error code %d\n", error);
 614                         /*fall through*/
 615                 case ENOENT:
 616                         /* don't show these error codes to the user */
 617                         error = 0;
 618                         break;
 619                 }
 620                 goto bad;
 621         }
 622     }
 623
 624         /* be sure to update variables that are affected by ipsec4_output() */
 625         ip = mtod(m, struct ip *);
 626 #ifdef _IP_VHL
 627         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 628 #else
 629         hlen = ip->ip_hl << 2;
 630 #endif
 631         if (ro->ro_rt == NULL) {
 632                 if (!(flags & IP_ROUTETOIF)) {
 633                         kprintf("ip_output: "
 634                                 "can't update route after IPsec processing\n");
 635                         error = EHOSTUNREACH;   /*XXX*/
 636                         goto bad;
 637                 }
 638         } else {
 639                 ia = ifatoia(ro->ro_rt->rt_ifa);
 640                 ifp = ro->ro_rt->rt_ifp;
 641         }
 642
 643         /* make it flipped, again. */
 644         ip->ip_len = ntohs(ip->ip_len);
 645         ip->ip_off = ntohs(ip->ip_off);
 646 skip_ipsec:
 647 #endif /*IPSEC*/
 648 #ifdef FAST_IPSEC
 649         /*
 650          * Check the security policy (SP) for the packet and, if
 651          * required, do IPsec-related processing.  There are two
 652          * cases here; the first time a packet is sent through
 653          * it will be untagged and handled by ipsec4_checkpolicy.
 654          * If the packet is resubmitted to ip_output (e.g. after
 655          * AH, ESP, etc. processing), there will be a tag to bypass
 656          * the lookup and related policy checking.
 657          */
 658         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 659         crit_enter();
 660         if (mtag != NULL) {
 661                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 662                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 663                 if (sp == NULL)
 664                         error = -EINVAL;        /* force silent drop */
 665                 m_tag_delete(m, mtag);
 666         } else {
 667                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 668                                         &error, inp);
 669         }
 670         /*
 671          * There are four return cases:
 672          *    sp != NULL                    apply IPsec policy
 673          *    sp == NULL, error == 0        no IPsec handling needed
 674          *    sp == NULL, error == -EINVAL  discard packet w/o error
 675          *    sp == NULL, error != 0        discard packet, report error
 676          */
 677         if (sp != NULL) {
 678                 /* Loop detection, check if ipsec processing already done */
 679                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 680                 for (mtag = m_tag_first(m); mtag != NULL;
 681                      mtag = m_tag_next(m, mtag)) {
 682                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 683                                 continue;
 684                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 685                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 686                                 continue;
 687                         /*
 688                          * Check if policy has an SA associated with it.
 689                          * This can happen when an SP has yet to acquire
 690                          * an SA; e.g. on first reference.  If it occurs,
 691                          * then we let ipsec4_process_packet do its thing.
 692                          */
 693                         if (sp->req->sav == NULL)
 694                                 break;
 695                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 696                         if (tdbi->spi == sp->req->sav->spi &&
 697                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 698                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 699                                  sizeof(union sockaddr_union)) == 0) {
 700                                 /*
 701                                  * No IPsec processing is needed, free
 702                                  * reference to SP.
 703                                  *
 704                                  * NB: null pointer to avoid free at
 705                                  *     done: below.
 706                                  */
 707                                 KEY_FREESP(&sp), sp = NULL;
 708                                 crit_exit();
 709                                 goto spd_done;
 710                         }
 711                 }
 712
 713                 /*
 714                  * Do delayed checksums now because we send before
 715                  * this is done in the normal processing path.
 716                  */
 717                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 718                         in_delayed_cksum(m);
 719                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 720                 }
 721
 722                 ip->ip_len = htons(ip->ip_len);
 723                 ip->ip_off = htons(ip->ip_off);
 724
 725                 /* NB: callee frees mbuf */
 726                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 727                 /*
 728                  * Preserve KAME behaviour: ENOENT can be returned
 729                  * when an SA acquire is in progress.  Don't propagate
 730                  * this to user-level; it confuses applications.
 731                  *
 732                  * XXX this will go away when the SADB is redone.
 733                  */
 734                 if (error == ENOENT)
 735                         error = 0;
 736                 crit_exit();
 737                 goto done;
 738         } else {
 739                 crit_exit();
 740
 741                 if (error != 0) {
 742                         /*
 743                          * Hack: -EINVAL is used to signal that a packet
 744                          * should be silently discarded.  This is typically
 745                          * because we asked key management for an SA and
 746                          * it was delayed (e.g. kicked up to IKE).
 747                          */
 748                         if (error == -EINVAL)
 749                                 error = 0;
 750                         goto bad;
 751                 } else {
 752                         /* No IPsec processing for this packet. */
 753                 }
 754 #ifdef notyet
 755                 /*
 756                  * If deferred crypto processing is needed, check that
 757                  * the interface supports it.
 758                  */
 759                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 760                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 761                         /* notify IPsec to do its own crypto */
 762                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 763                         error = EHOSTUNREACH;
 764                         goto bad;
 765                 }
 766 #endif
 767         }
 768 spd_done:
 769 #endif /* FAST_IPSEC */
 770
 771         /* We are already being fwd'd from a firewall. */
 772         if (next_hop != NULL)
 773                 goto pass;
 774
 775         /* No pfil hooks */
 776         if (!pfil_has_hooks(&inet_pfil_hook)) {
 777                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 778                         /*
 779                          * Strip dummynet tags from stranded packets
 780                          */
 781                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 782                         KKASSERT(mtag != NULL);
 783                         m_tag_delete(m, mtag);
 784                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 785                 }
 786                 goto pass;
 787         }
 788
 789         /*
 790          * IpHack's section.
 791          * - Xlate: translate packet's addr/port (NAT).
 792          * - Firewall: deny/allow/etc.
 793          * - Wrap: fake packet's addr/port <unimpl.>
 794          * - Encapsulate: put it in another IP and send out. <unimp.>
 795          */
 796
 797         /*
 798          * Run through list of hooks for output packets.
 799          */
 800         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 801         if (error != 0 || m == NULL)
 802                 goto done;
 803         ip = mtod(m, struct ip *);
 804
 805         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 806                 /*
 807                  * Check dst to make sure it is directly reachable on the
 808                  * interface we previously thought it was.
 809                  * If it isn't (which may be likely in some situations) we have
 810                  * to re-route it (ie, find a route for the next-hop and the
 811                  * associated interface) and set them here. This is nested
 812                  * forwarding which in most cases is undesirable, except where
 813                  * such control is nigh impossible. So we do it here.
 814                  * And I'm babbling.
 815                  */
 816                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 817                 KKASSERT(mtag != NULL);
 818                 next_hop = m_tag_data(mtag);
 819
 820                 /*
 821                  * Try local forwarding first
 822                  */
 823                 if (ip_localforward(m, next_hop, hlen))
 824                         goto done;
 825
 826                 /*
 827                  * Relocate the route based on next_hop.
 828                  * If the current route is inp's cache, keep it untouched.
 829                  */
 830                 if (ro == &iproute && ro->ro_rt != NULL) {
 831                         RTFREE(ro->ro_rt);
 832                         ro->ro_rt = NULL;
 833                 }
 834                 ro = &iproute;
 835                 bzero(ro, sizeof *ro);
 836
 837                 /*
 838                  * Forwarding to broadcast address is not allowed.
 839                  * XXX Should we follow IP_ROUTETOIF?
 840                  */
 841                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 842
 843                 /* We are doing forwarding now */
 844                 flags |= IP_FORWARDING;
 845
 846                 goto reroute;
 847         }
 848
 849         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 850                 struct dn_pkt *dn_pkt;
 851
 852                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 853                 KKASSERT(mtag != NULL);
 854                 dn_pkt = m_tag_data(mtag);
 855
 856                 /*
 857                  * Under certain cases it is not possible to recalculate
 858                  * 'ro' and 'dst', let alone 'flags', so just save them in
 859                  * dummynet tag and avoid the possible wrong reculcalation
 860                  * when we come back to ip_output() again.
 861                  *
 862                  * All other parameters have been already used and so they
 863                  * are not needed anymore.
 864                  * XXX if the ifp is deleted while a pkt is in dummynet,
 865                  * we are in trouble! (TODO use ifnet_detach_event)
 866                  *
 867                  * We need to copy *ro because for ICMP pkts (and maybe
 868                  * others) the caller passed a pointer into the stack;
 869                  * dst might also be a pointer into *ro so it needs to
 870                  * be updated.
 871                  */
 872                 dn_pkt->ro = *ro;
 873                 if (ro->ro_rt)
 874                         ro->ro_rt->rt_refcnt++;
 875                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 876                         /* 'dst' points into 'ro' */
 877                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 878                 }
 879                 dn_pkt->dn_dst = dst;
 880                 dn_pkt->flags = flags;
 881
 882                 ip_dn_queue(m);
 883                 goto done;
 884         }
 885 pass:
 886         /* 127/8 must not appear on wire - RFC1122. */
 887         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 888             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 889                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 890                         ipstat.ips_badaddr++;
 891                         error = EADDRNOTAVAIL;
 892                         goto bad;
 893                 }
 894         }
 895         if (ip->ip_src.s_addr == INADDR_ANY ||
 896             IN_MULTICAST(ntohl(ip->ip_src.s_addr))) {
 897                 ipstat.ips_badaddr++;
 898                 error = EADDRNOTAVAIL;
 899                 goto bad;
 900         }
 901
 902         if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
 903                 m->m_pkthdr.csum_flags |= CSUM_IP;
 904                 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 905                 if (sw_csum & CSUM_DELAY_DATA) {
 906                         in_delayed_cksum(m);
 907                         sw_csum &= ~CSUM_DELAY_DATA;
 908                 }
 909                 m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 910         } else {
 911                 sw_csum = 0;
 912         }
 913         m->m_pkthdr.csum_iphlen = hlen;
 914
 915         /*
 916          * If small enough for interface, or the interface will take
 917          * care of the fragmentation or segmentation for us, can just
 918          * send directly.
 919          */
 920         if (ip->ip_len <= ifp->if_mtu ||
 921             ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
 922             (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 923                 ip->ip_len = htons(ip->ip_len);
 924                 ip->ip_off = htons(ip->ip_off);
 925                 ip->ip_sum = 0;
 926                 if (sw_csum & CSUM_DELAY_IP) {
 927                         if (ip->ip_vhl == IP_VHL_BORING)
 928                                 ip->ip_sum = in_cksum_hdr(ip);
 929                         else
 930                                 ip->ip_sum = in_cksum(m, hlen);
 931                 }
 932
 933                 /* Record statistics for this interface address. */
 934                 if (!(flags & IP_FORWARDING) && ia) {
 935                         IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
 936                         IFA_STAT_INC(&ia->ia_ifa, obytes, m->m_pkthdr.len);
 937                 }
 938
 939 #ifdef IPSEC
 940                 /* clean ipsec history once it goes out of the node */
 941                 ipsec_delaux(m);
 942 #endif
 943
 944 #ifdef MBUF_STRESS_TEST
 945                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 946                         struct mbuf *m1, *m2;
 947                         int length, tmp;
 948
 949                         tmp = length = m->m_pkthdr.len;
 950
 951                         while ((length -= mbuf_frag_size) >= 1) {
 952                                 m1 = m_split(m, length, MB_DONTWAIT);
 953                                 if (m1 == NULL)
 954                                         break;
 955                                 m2 = m;
 956                                 while (m2->m_next != NULL)
 957                                         m2 = m2->m_next;
 958                                 m2->m_next = m1;
 959                         }
 960                         m->m_pkthdr.len = tmp;
 961                 }
 962 #endif
 963
 964 #ifdef MPLS
 965                 if (!mpls_output_process(m, ro->ro_rt))
 966                         goto done;
 967 #endif
 968                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 969                                        ro->ro_rt);
 970                 goto done;
 971         }
 972
 973         if (ip->ip_off & IP_DF) {
 974                 error = EMSGSIZE;
 975                 /*
 976                  * This case can happen if the user changed the MTU
 977                  * of an interface after enabling IP on it.  Because
 978                  * most netifs don't keep track of routes pointing to
 979                  * them, there is no way for one to update all its
 980                  * routes when the MTU is changed.
 981                  */
 982                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 983                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 984                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 985                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 986                 }
 987                 ipstat.ips_cantfrag++;
 988                 goto bad;
 989         }
 990
 991         /*
 992          * Too large for interface; fragment if possible. If successful,
 993          * on return, m will point to a list of packets to be sent.
 994          */
 995         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
 996         if (error)
 997                 goto bad;
 998         for (; m; m = m0) {
 999                 m0 = m->m_nextpkt;
1000                 m->m_nextpkt = NULL;
1001 #ifdef IPSEC
1002                 /* clean ipsec history once it goes out of the node */
1003                 ipsec_delaux(m);
1004 #endif
1005                 if (error == 0) {
1006                         /* Record statistics for this interface address. */
1007                         if (ia != NULL) {
1008                                 IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
1009                                 IFA_STAT_INC(&ia->ia_ifa, obytes,
1010                                     m->m_pkthdr.len);
1011                         }
1012 #ifdef MPLS
1013                         if (!mpls_output_process(m, ro->ro_rt))
1014                                 continue;
1015 #endif
1016                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1017                                                ro->ro_rt);
1018                 } else {
1019                         m_freem(m);
1020                 }
1021         }
1022
1023         if (error == 0)
1024                 ipstat.ips_fragmented++;
1025
1026 done:
1027         if (ro == &iproute && ro->ro_rt != NULL) {
1028                 RTFREE(ro->ro_rt);
1029                 ro->ro_rt = NULL;
1030         }
1031 #ifdef IPSEC
1032         if (sp != NULL) {
1033                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1034                         kprintf("DP ip_output call free SP:%p\n", sp));
1035                 key_freesp(sp);
1036         }
1037 #endif
1038 #ifdef FAST_IPSEC
1039         if (sp != NULL)
1040                 KEY_FREESP(&sp);
1041 #endif
1042         return (error);
1043 bad:
1044         m_freem(m);
1045         goto done;
1046 }
1047
1048 /*
1049  * Create a chain of fragments which fit the given mtu. m_frag points to the
1050  * mbuf to be fragmented; on return it points to the chain with the fragments.
1051  * Return 0 if no error. If error, m_frag may contain a partially built
1052  * chain of fragments that should be freed by the caller.
1053  *
1054  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1055  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1056  */
1057 int
1058 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1059             u_long if_hwassist_flags, int sw_csum)
1060 {
1061         int error = 0;
1062         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1063         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1064         int off;
1065         struct mbuf *m0 = *m_frag;      /* the original packet          */
1066         int firstlen;
1067         struct mbuf **mnext;
1068         int nfrags;
1069
1070         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1071                 ipstat.ips_cantfrag++;
1072                 return EMSGSIZE;
1073         }
1074
1075         /*
1076          * Must be able to put at least 8 bytes per fragment.
1077          */
1078         if (len < 8)
1079                 return EMSGSIZE;
1080
1081         /*
1082          * If the interface will not calculate checksums on
1083          * fragmented packets, then do it here.
1084          */
1085         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1086             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1087                 in_delayed_cksum(m0);
1088                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1089         }
1090
1091         if (len > PAGE_SIZE) {
1092                 /*
1093                  * Fragment large datagrams such that each segment
1094                  * contains a multiple of PAGE_SIZE amount of data,
1095                  * plus headers. This enables a receiver to perform
1096                  * page-flipping zero-copy optimizations.
1097                  *
1098                  * XXX When does this help given that sender and receiver
1099                  * could have different page sizes, and also mtu could
1100                  * be less than the receiver's page size ?
1101                  */
1102                 int newlen;
1103                 struct mbuf *m;
1104
1105                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1106                         off += m->m_len;
1107
1108                 /*
1109                  * firstlen (off - hlen) must be aligned on an
1110                  * 8-byte boundary
1111                  */
1112                 if (off < hlen)
1113                         goto smart_frag_failure;
1114                 off = ((off - hlen) & ~7) + hlen;
1115                 newlen = (~PAGE_MASK) & mtu;
1116                 if ((newlen + sizeof(struct ip)) > mtu) {
1117                         /* we failed, go back the default */
1118 smart_frag_failure:
1119                         newlen = len;
1120                         off = hlen + len;
1121                 }
1122                 len = newlen;
1123
1124         } else {
1125                 off = hlen + len;
1126         }
1127
1128         firstlen = off - hlen;
1129         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1130
1131         /*
1132          * Loop through length of segment after first fragment,
1133          * make new header and copy data of each part and link onto chain.
1134          * Here, m0 is the original packet, m is the fragment being created.
1135          * The fragments are linked off the m_nextpkt of the original
1136          * packet, which after processing serves as the first fragment.
1137          */
1138         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1139                 struct ip *mhip;        /* ip header on the fragment */
1140                 struct mbuf *m;
1141                 int mhlen = sizeof(struct ip);
1142
1143                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1144                 if (m == NULL) {
1145                         error = ENOBUFS;
1146                         ipstat.ips_odropped++;
1147                         goto done;
1148                 }
1149                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1150                 /*
1151                  * In the first mbuf, leave room for the link header, then
1152                  * copy the original IP header including options. The payload
1153                  * goes into an additional mbuf chain returned by m_copy().
1154                  */
1155                 m->m_data += max_linkhdr;
1156                 mhip = mtod(m, struct ip *);
1157                 *mhip = *ip;
1158                 if (hlen > sizeof(struct ip)) {
1159                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1160                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1161                 }
1162                 m->m_len = mhlen;
1163                 /* XXX do we need to add ip->ip_off below ? */
1164                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1165                 if (off + len >= ip->ip_len) {  /* last fragment */
1166                         len = ip->ip_len - off;
1167                         m->m_flags |= M_LASTFRAG;
1168                 } else
1169                         mhip->ip_off |= IP_MF;
1170                 mhip->ip_len = htons((u_short)(len + mhlen));
1171                 m->m_next = m_copy(m0, off, len);
1172                 if (m->m_next == NULL) {                /* copy failed */
1173                         m_free(m);
1174                         error = ENOBUFS;        /* ??? */
1175                         ipstat.ips_odropped++;
1176                         goto done;
1177                 }
1178                 m->m_pkthdr.len = mhlen + len;
1179                 m->m_pkthdr.rcvif = NULL;
1180                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1181                 m->m_pkthdr.csum_iphlen = mhlen;
1182                 mhip->ip_off = htons(mhip->ip_off);
1183                 mhip->ip_sum = 0;
1184                 if (sw_csum & CSUM_DELAY_IP)
1185                         mhip->ip_sum = in_cksum(m, mhlen);
1186                 *mnext = m;
1187                 mnext = &m->m_nextpkt;
1188         }
1189         ipstat.ips_ofragments += nfrags;
1190
1191         /* set first marker for fragment chain */
1192         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1193         m0->m_pkthdr.csum_data = nfrags;
1194
1195         /*
1196          * Update first fragment by trimming what's been copied out
1197          * and updating header.
1198          */
1199         m_adj(m0, hlen + firstlen - ip->ip_len);
1200         m0->m_pkthdr.len = hlen + firstlen;
1201         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1202         ip->ip_off |= IP_MF;
1203         ip->ip_off = htons(ip->ip_off);
1204         ip->ip_sum = 0;
1205         if (sw_csum & CSUM_DELAY_IP)
1206                 ip->ip_sum = in_cksum(m0, hlen);
1207
1208 done:
1209         *m_frag = m0;
1210         return error;
1211 }
1212
1213 void
1214 in_delayed_cksum(struct mbuf *m)
1215 {
1216         struct ip *ip;
1217         u_short csum, offset;
1218
1219         ip = mtod(m, struct ip *);
1220         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1221         csum = in_cksum_skip(m, ip->ip_len, offset);
1222         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1223                 csum = 0xffff;
1224         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1225
1226         if (offset + sizeof(u_short) > m->m_len) {
1227                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1228                     m->m_len, offset, ip->ip_p);
1229                 /*
1230                  * XXX
1231                  * this shouldn't happen, but if it does, the
1232                  * correct behavior may be to insert the checksum
1233                  * in the existing chain instead of rearranging it.
1234                  */
1235                 m = m_pullup(m, offset + sizeof(u_short));
1236         }
1237         *(u_short *)(m->m_data + offset) = csum;
1238 }
1239
1240 /*
1241  * Insert IP options into preformed packet.
1242  * Adjust IP destination as required for IP source routing,
1243  * as indicated by a non-zero in_addr at the start of the options.
1244  *
1245  * XXX This routine assumes that the packet has no options in place.
1246  */
1247 static struct mbuf *
1248 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1249 {
1250         struct ipoption *p = mtod(opt, struct ipoption *);
1251         struct mbuf *n;
1252         struct ip *ip = mtod(m, struct ip *);
1253         unsigned optlen;
1254
1255         optlen = opt->m_len - sizeof p->ipopt_dst;
1256         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1257                 *phlen = 0;
1258                 return (m);             /* XXX should fail */
1259         }
1260         if (p->ipopt_dst.s_addr)
1261                 ip->ip_dst = p->ipopt_dst;
1262         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1263                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1264                 if (n == NULL) {
1265                         *phlen = 0;
1266                         return (m);
1267                 }
1268                 n->m_pkthdr.rcvif = NULL;
1269                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1270                 m->m_len -= sizeof(struct ip);
1271                 m->m_data += sizeof(struct ip);
1272                 n->m_next = m;
1273                 m = n;
1274                 m->m_len = optlen + sizeof(struct ip);
1275                 m->m_data += max_linkhdr;
1276                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1277         } else {
1278                 m->m_data -= optlen;
1279                 m->m_len += optlen;
1280                 m->m_pkthdr.len += optlen;
1281                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1282         }
1283         ip = mtod(m, struct ip *);
1284         bcopy(p->ipopt_list, ip + 1, optlen);
1285         *phlen = sizeof(struct ip) + optlen;
1286         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1287         ip->ip_len += optlen;
1288         return (m);
1289 }
1290
1291 /*
1292  * Copy options from ip to jp,
1293  * omitting those not copied during fragmentation.
1294  */
1295 int
1296 ip_optcopy(struct ip *ip, struct ip *jp)
1297 {
1298         u_char *cp, *dp;
1299         int opt, optlen, cnt;
1300
1301         cp = (u_char *)(ip + 1);
1302         dp = (u_char *)(jp + 1);
1303         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1304         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1305                 opt = cp[0];
1306                 if (opt == IPOPT_EOL)
1307                         break;
1308                 if (opt == IPOPT_NOP) {
1309                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1310                         *dp++ = IPOPT_NOP;
1311                         optlen = 1;
1312                         continue;
1313                 }
1314
1315                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1316                     ("ip_optcopy: malformed ipv4 option"));
1317                 optlen = cp[IPOPT_OLEN];
1318                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1319                     ("ip_optcopy: malformed ipv4 option"));
1320
1321                 /* bogus lengths should have been caught by ip_dooptions */
1322                 if (optlen > cnt)
1323                         optlen = cnt;
1324                 if (IPOPT_COPIED(opt)) {
1325                         bcopy(cp, dp, optlen);
1326                         dp += optlen;
1327                 }
1328         }
1329         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1330                 *dp++ = IPOPT_EOL;
1331         return (optlen);
1332 }
1333
1334 /*
1335  * IP socket option processing.
1336  */
1337 void
1338 ip_ctloutput(netmsg_t msg)
1339 {
1340         struct socket *so = msg->base.nm_so;
1341         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1342         struct  inpcb *inp = so->so_pcb;
1343         int     error, optval;
1344
1345         error = optval = 0;
1346
1347         /* Get socket's owner cpuid hint */
1348         if (sopt->sopt_level == SOL_SOCKET &&
1349             sopt->sopt_dir == SOPT_GET &&
1350             sopt->sopt_name == SO_CPUHINT) {
1351                 optval = mycpuid;
1352                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1353                 goto done;
1354         }
1355
1356         if (sopt->sopt_level != IPPROTO_IP) {
1357                 error = EINVAL;
1358                 goto done;
1359         }
1360
1361         switch (sopt->sopt_name) {
1362         case IP_MULTICAST_IF:
1363         case IP_MULTICAST_VIF:
1364         case IP_MULTICAST_TTL:
1365         case IP_MULTICAST_LOOP:
1366         case IP_ADD_MEMBERSHIP:
1367         case IP_DROP_MEMBERSHIP:
1368                 /*
1369                  * Handle multicast options in netisr0
1370                  */
1371                 if (&curthread->td_msgport != netisr_cpuport(0)) {
1372                         /* NOTE: so_port MUST NOT be checked in netisr0 */
1373                         msg->lmsg.ms_flags |= MSGF_IGNSOPORT;
1374                         lwkt_forwardmsg(netisr_cpuport(0), &msg->lmsg);
1375                         return;
1376                 }
1377                 break;
1378         }
1379
1380         switch (sopt->sopt_dir) {
1381         case SOPT_SET:
1382                 switch (sopt->sopt_name) {
1383                 case IP_OPTIONS:
1384 #ifdef notyet
1385                 case IP_RETOPTS:
1386 #endif
1387                 {
1388                         struct mbuf *m;
1389                         if (sopt->sopt_valsize > MLEN) {
1390                                 error = EMSGSIZE;
1391                                 break;
1392                         }
1393                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1394                         if (m == NULL) {
1395                                 error = ENOBUFS;
1396                                 break;
1397                         }
1398                         m->m_len = sopt->sopt_valsize;
1399                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1400                                               m->m_len);
1401                         error = ip_pcbopts(sopt->sopt_name,
1402                                            &inp->inp_options, m);
1403                         goto done;
1404                 }
1405
1406                 case IP_TOS:
1407                 case IP_TTL:
1408                 case IP_MINTTL:
1409                 case IP_RECVOPTS:
1410                 case IP_RECVRETOPTS:
1411                 case IP_RECVDSTADDR:
1412                 case IP_RECVIF:
1413                 case IP_RECVTTL:
1414                 case IP_FAITH:
1415                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1416                                              sizeof optval);
1417                         if (error)
1418                                 break;
1419                         switch (sopt->sopt_name) {
1420                         case IP_TOS:
1421                                 inp->inp_ip_tos = optval;
1422                                 break;
1423
1424                         case IP_TTL:
1425                                 inp->inp_ip_ttl = optval;
1426                                 break;
1427                         case IP_MINTTL:
1428                                 if (optval >= 0 && optval <= MAXTTL)
1429                                         inp->inp_ip_minttl = optval;
1430                                 else
1431                                         error = EINVAL;
1432                                 break;
1433 #define OPTSET(bit) \
1434         if (optval) \
1435                 inp->inp_flags |= bit; \
1436         else \
1437                 inp->inp_flags &= ~bit;
1438
1439                         case IP_RECVOPTS:
1440                                 OPTSET(INP_RECVOPTS);
1441                                 break;
1442
1443                         case IP_RECVRETOPTS:
1444                                 OPTSET(INP_RECVRETOPTS);
1445                                 break;
1446
1447                         case IP_RECVDSTADDR:
1448                                 OPTSET(INP_RECVDSTADDR);
1449                                 break;
1450
1451                         case IP_RECVIF:
1452                                 OPTSET(INP_RECVIF);
1453                                 break;
1454
1455                         case IP_RECVTTL:
1456                                 OPTSET(INP_RECVTTL);
1457                                 break;
1458
1459                         case IP_FAITH:
1460                                 OPTSET(INP_FAITH);
1461                                 break;
1462                         }
1463                         break;
1464 #undef OPTSET
1465
1466                 case IP_MULTICAST_IF:
1467                 case IP_MULTICAST_VIF:
1468                 case IP_MULTICAST_TTL:
1469                 case IP_MULTICAST_LOOP:
1470                 case IP_ADD_MEMBERSHIP:
1471                 case IP_DROP_MEMBERSHIP:
1472                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1473                         break;
1474
1475                 case IP_PORTRANGE:
1476                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1477                                             sizeof optval);
1478                         if (error)
1479                                 break;
1480
1481                         switch (optval) {
1482                         case IP_PORTRANGE_DEFAULT:
1483                                 inp->inp_flags &= ~(INP_LOWPORT);
1484                                 inp->inp_flags &= ~(INP_HIGHPORT);
1485                                 break;
1486
1487                         case IP_PORTRANGE_HIGH:
1488                                 inp->inp_flags &= ~(INP_LOWPORT);
1489                                 inp->inp_flags |= INP_HIGHPORT;
1490                                 break;
1491
1492                         case IP_PORTRANGE_LOW:
1493                                 inp->inp_flags &= ~(INP_HIGHPORT);
1494                                 inp->inp_flags |= INP_LOWPORT;
1495                                 break;
1496
1497                         default:
1498                                 error = EINVAL;
1499                                 break;
1500                         }
1501                         break;
1502
1503 #if defined(IPSEC) || defined(FAST_IPSEC)
1504                 case IP_IPSEC_POLICY:
1505                 {
1506                         caddr_t req;
1507                         size_t len = 0;
1508                         int priv;
1509                         struct mbuf *m;
1510                         int optname;
1511
1512                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1513                                 break;
1514                         soopt_to_mbuf(sopt, m);
1515                         priv = (sopt->sopt_td != NULL &&
1516                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1517                         req = mtod(m, caddr_t);
1518                         len = m->m_len;
1519                         optname = sopt->sopt_name;
1520                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1521                         m_freem(m);
1522                         break;
1523                 }
1524 #endif /*IPSEC*/
1525
1526                 default:
1527                         error = ENOPROTOOPT;
1528                         break;
1529                 }
1530                 break;
1531
1532         case SOPT_GET:
1533                 switch (sopt->sopt_name) {
1534                 case IP_OPTIONS:
1535                 case IP_RETOPTS:
1536                         if (inp->inp_options)
1537                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1538                                                            char *),
1539                                                 inp->inp_options->m_len);
1540                         else
1541                                 sopt->sopt_valsize = 0;
1542                         break;
1543
1544                 case IP_TOS:
1545                 case IP_TTL:
1546                 case IP_MINTTL:
1547                 case IP_RECVOPTS:
1548                 case IP_RECVRETOPTS:
1549                 case IP_RECVDSTADDR:
1550                 case IP_RECVTTL:
1551                 case IP_RECVIF:
1552                 case IP_PORTRANGE:
1553                 case IP_FAITH:
1554                         switch (sopt->sopt_name) {
1555
1556                         case IP_TOS:
1557                                 optval = inp->inp_ip_tos;
1558                                 break;
1559
1560                         case IP_TTL:
1561                                 optval = inp->inp_ip_ttl;
1562                                 break;
1563                         case IP_MINTTL:
1564                                 optval = inp->inp_ip_minttl;
1565                                 break;
1566
1567 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1568
1569                         case IP_RECVOPTS:
1570                                 optval = OPTBIT(INP_RECVOPTS);
1571                                 break;
1572
1573                         case IP_RECVRETOPTS:
1574                                 optval = OPTBIT(INP_RECVRETOPTS);
1575                                 break;
1576
1577                         case IP_RECVDSTADDR:
1578                                 optval = OPTBIT(INP_RECVDSTADDR);
1579                                 break;
1580
1581                         case IP_RECVTTL:
1582                                 optval = OPTBIT(INP_RECVTTL);
1583                                 break;
1584
1585                         case IP_RECVIF:
1586                                 optval = OPTBIT(INP_RECVIF);
1587                                 break;
1588
1589                         case IP_PORTRANGE:
1590                                 if (inp->inp_flags & INP_HIGHPORT)
1591                                         optval = IP_PORTRANGE_HIGH;
1592                                 else if (inp->inp_flags & INP_LOWPORT)
1593                                         optval = IP_PORTRANGE_LOW;
1594                                 else
1595                                         optval = 0;
1596                                 break;
1597
1598                         case IP_FAITH:
1599                                 optval = OPTBIT(INP_FAITH);
1600                                 break;
1601                         }
1602                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1603                         break;
1604
1605                 case IP_MULTICAST_IF:
1606                 case IP_MULTICAST_VIF:
1607                 case IP_MULTICAST_TTL:
1608                 case IP_MULTICAST_LOOP:
1609                 case IP_ADD_MEMBERSHIP:
1610                 case IP_DROP_MEMBERSHIP:
1611                         error = ip_getmoptions(sopt, inp->inp_moptions);
1612                         break;
1613
1614 #if defined(IPSEC) || defined(FAST_IPSEC)
1615                 case IP_IPSEC_POLICY:
1616                 {
1617                         struct mbuf *m = NULL;
1618                         caddr_t req = NULL;
1619                         size_t len = 0;
1620
1621                         if (m != NULL) {
1622                                 req = mtod(m, caddr_t);
1623                                 len = m->m_len;
1624                         }
1625                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1626                         if (error == 0)
1627                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1628                         if (error == 0)
1629                                 m_freem(m);
1630                         break;
1631                 }
1632 #endif /*IPSEC*/
1633
1634                 default:
1635                         error = ENOPROTOOPT;
1636                         break;
1637                 }
1638                 break;
1639         }
1640 done:
1641         lwkt_replymsg(&msg->lmsg, error);
1642 }
1643
1644 /*
1645  * Set up IP options in pcb for insertion in output packets.
1646  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1647  * with destination address if source routed.
1648  */
1649 static int
1650 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1651 {
1652         int cnt, optlen;
1653         u_char *cp;
1654         u_char opt;
1655
1656         /* turn off any old options */
1657         if (*pcbopt)
1658                 m_free(*pcbopt);
1659         *pcbopt = NULL;
1660         if (m == NULL || m->m_len == 0) {
1661                 /*
1662                  * Only turning off any previous options.
1663                  */
1664                 if (m != NULL)
1665                         m_free(m);
1666                 return (0);
1667         }
1668
1669         if (m->m_len % sizeof(int32_t))
1670                 goto bad;
1671         /*
1672          * IP first-hop destination address will be stored before
1673          * actual options; move other options back
1674          * and clear it when none present.
1675          */
1676         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1677                 goto bad;
1678         cnt = m->m_len;
1679         m->m_len += sizeof(struct in_addr);
1680         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1681         ovbcopy(mtod(m, caddr_t), cp, cnt);
1682         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1683
1684         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1685                 opt = cp[IPOPT_OPTVAL];
1686                 if (opt == IPOPT_EOL)
1687                         break;
1688                 if (opt == IPOPT_NOP)
1689                         optlen = 1;
1690                 else {
1691                         if (cnt < IPOPT_OLEN + sizeof *cp)
1692                                 goto bad;
1693                         optlen = cp[IPOPT_OLEN];
1694                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1695                                 goto bad;
1696                 }
1697                 switch (opt) {
1698
1699                 default:
1700                         break;
1701
1702                 case IPOPT_LSRR:
1703                 case IPOPT_SSRR:
1704                         /*
1705                          * user process specifies route as:
1706                          *      ->A->B->C->D
1707                          * D must be our final destination (but we can't
1708                          * check that since we may not have connected yet).
1709                          * A is first hop destination, which doesn't appear in
1710                          * actual IP option, but is stored before the options.
1711                          */
1712                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1713                                 goto bad;
1714                         m->m_len -= sizeof(struct in_addr);
1715                         cnt -= sizeof(struct in_addr);
1716                         optlen -= sizeof(struct in_addr);
1717                         cp[IPOPT_OLEN] = optlen;
1718                         /*
1719                          * Move first hop before start of options.
1720                          */
1721                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1722                               sizeof(struct in_addr));
1723                         /*
1724                          * Then copy rest of options back
1725                          * to close up the deleted entry.
1726                          */
1727                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1728                                 &cp[IPOPT_OFFSET+1],
1729                                 cnt - (IPOPT_MINOFF - 1));
1730                         break;
1731                 }
1732         }
1733         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1734                 goto bad;
1735         *pcbopt = m;
1736         return (0);
1737
1738 bad:
1739         m_free(m);
1740         return (EINVAL);
1741 }
1742
1743 /*
1744  * XXX
1745  * The whole multicast option thing needs to be re-thought.
1746  * Several of these options are equally applicable to non-multicast
1747  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1748  * standard option (IP_TTL).
1749  */
1750
1751 /*
1752  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1753  */
1754 static struct ifnet *
1755 ip_multicast_if(struct in_addr *a, int *ifindexp)
1756 {
1757         int ifindex;
1758         struct ifnet *ifp;
1759
1760         if (ifindexp)
1761                 *ifindexp = 0;
1762         if (ntohl(a->s_addr) >> 24 == 0) {
1763                 ifindex = ntohl(a->s_addr) & 0xffffff;
1764                 if (ifindex < 0 || if_index < ifindex)
1765                         return NULL;
1766                 ifp = ifindex2ifnet[ifindex];
1767                 if (ifindexp)
1768                         *ifindexp = ifindex;
1769         } else {
1770                 ifp = INADDR_TO_IFP(a);
1771         }
1772         return ifp;
1773 }
1774
1775 /*
1776  * Set the IP multicast options in response to user setsockopt().
1777  */
1778 static int
1779 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1780 {
1781         int error = 0;
1782         int i;
1783         struct in_addr addr;
1784         struct ip_mreq mreq;
1785         struct ifnet *ifp;
1786         struct ip_moptions *imo = *imop;
1787         int ifindex;
1788
1789         if (imo == NULL) {
1790                 /*
1791                  * No multicast option buffer attached to the pcb;
1792                  * allocate one and initialize to default values.
1793                  */
1794                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1795
1796                 *imop = imo;
1797                 imo->imo_multicast_ifp = NULL;
1798                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1799                 imo->imo_multicast_vif = -1;
1800                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1801                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1802                 imo->imo_num_memberships = 0;
1803         }
1804         switch (sopt->sopt_name) {
1805         /* store an index number for the vif you wanna use in the send */
1806         case IP_MULTICAST_VIF:
1807                 if (legal_vif_num == 0) {
1808                         error = EOPNOTSUPP;
1809                         break;
1810                 }
1811                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1812                 if (error)
1813                         break;
1814                 if (!legal_vif_num(i) && (i != -1)) {
1815                         error = EINVAL;
1816                         break;
1817                 }
1818                 imo->imo_multicast_vif = i;
1819                 break;
1820
1821         case IP_MULTICAST_IF:
1822                 /*
1823                  * Select the interface for outgoing multicast packets.
1824                  */
1825                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1826                 if (error)
1827                         break;
1828
1829                 /*
1830                  * INADDR_ANY is used to remove a previous selection.
1831                  * When no interface is selected, a default one is
1832                  * chosen every time a multicast packet is sent.
1833                  */
1834                 if (addr.s_addr == INADDR_ANY) {
1835                         imo->imo_multicast_ifp = NULL;
1836                         break;
1837                 }
1838                 /*
1839                  * The selected interface is identified by its local
1840                  * IP address.  Find the interface and confirm that
1841                  * it supports multicasting.
1842                  */
1843                 crit_enter();
1844                 ifp = ip_multicast_if(&addr, &ifindex);
1845                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1846                         crit_exit();
1847                         error = EADDRNOTAVAIL;
1848                         break;
1849                 }
1850                 imo->imo_multicast_ifp = ifp;
1851                 if (ifindex)
1852                         imo->imo_multicast_addr = addr;
1853                 else
1854                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1855                 crit_exit();
1856                 break;
1857
1858         case IP_MULTICAST_TTL:
1859                 /*
1860                  * Set the IP time-to-live for outgoing multicast packets.
1861                  * The original multicast API required a char argument,
1862                  * which is inconsistent with the rest of the socket API.
1863                  * We allow either a char or an int.
1864                  */
1865                 if (sopt->sopt_valsize == 1) {
1866                         u_char ttl;
1867                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1868                         if (error)
1869                                 break;
1870                         imo->imo_multicast_ttl = ttl;
1871                 } else {
1872                         u_int ttl;
1873                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1874                         if (error)
1875                                 break;
1876                         if (ttl > 255)
1877                                 error = EINVAL;
1878                         else
1879                                 imo->imo_multicast_ttl = ttl;
1880                 }
1881                 break;
1882
1883         case IP_MULTICAST_LOOP:
1884                 /*
1885                  * Set the loopback flag for outgoing multicast packets.
1886                  * Must be zero or one.  The original multicast API required a
1887                  * char argument, which is inconsistent with the rest
1888                  * of the socket API.  We allow either a char or an int.
1889                  */
1890                 if (sopt->sopt_valsize == 1) {
1891                         u_char loop;
1892
1893                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1894                         if (error)
1895                                 break;
1896                         imo->imo_multicast_loop = !!loop;
1897                 } else {
1898                         u_int loop;
1899
1900                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1901                                             sizeof loop);
1902                         if (error)
1903                                 break;
1904                         imo->imo_multicast_loop = !!loop;
1905                 }
1906                 break;
1907
1908         case IP_ADD_MEMBERSHIP:
1909                 /*
1910                  * Add a multicast group membership.
1911                  * Group must be a valid IP multicast address.
1912                  */
1913                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1914                 if (error)
1915                         break;
1916
1917                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1918                         error = EINVAL;
1919                         break;
1920                 }
1921                 crit_enter();
1922                 /*
1923                  * If no interface address was provided, use the interface of
1924                  * the route to the given multicast address.
1925                  */
1926                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1927                         struct sockaddr_in dst;
1928                         struct rtentry *rt;
1929
1930                         bzero(&dst, sizeof(struct sockaddr_in));
1931                         dst.sin_len = sizeof(struct sockaddr_in);
1932                         dst.sin_family = AF_INET;
1933                         dst.sin_addr = mreq.imr_multiaddr;
1934                         rt = rtlookup((struct sockaddr *)&dst);
1935                         if (rt == NULL) {
1936                                 error = EADDRNOTAVAIL;
1937                                 crit_exit();
1938                                 break;
1939                         }
1940                         --rt->rt_refcnt;
1941                         ifp = rt->rt_ifp;
1942                 } else {
1943                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1944                 }
1945
1946                 /*
1947                  * See if we found an interface, and confirm that it
1948                  * supports multicast.
1949                  */
1950                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1951                         error = EADDRNOTAVAIL;
1952                         crit_exit();
1953                         break;
1954                 }
1955                 /*
1956                  * See if the membership already exists or if all the
1957                  * membership slots are full.
1958                  */
1959                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1960                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1961                             imo->imo_membership[i]->inm_addr.s_addr
1962                                                 == mreq.imr_multiaddr.s_addr)
1963                                 break;
1964                 }
1965                 if (i < imo->imo_num_memberships) {
1966                         error = EADDRINUSE;
1967                         crit_exit();
1968                         break;
1969                 }
1970                 if (i == IP_MAX_MEMBERSHIPS) {
1971                         error = ETOOMANYREFS;
1972                         crit_exit();
1973                         break;
1974                 }
1975                 /*
1976                  * Everything looks good; add a new record to the multicast
1977                  * address list for the given interface.
1978                  */
1979                 if ((imo->imo_membership[i] =
1980                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1981                         error = ENOBUFS;
1982                         crit_exit();
1983                         break;
1984                 }
1985                 ++imo->imo_num_memberships;
1986                 crit_exit();
1987                 break;
1988
1989         case IP_DROP_MEMBERSHIP:
1990                 /*
1991                  * Drop a multicast group membership.
1992                  * Group must be a valid IP multicast address.
1993                  */
1994                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1995                 if (error)
1996                         break;
1997
1998                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1999                         error = EINVAL;
2000                         break;
2001                 }
2002
2003                 crit_enter();
2004                 /*
2005                  * If an interface address was specified, get a pointer
2006                  * to its ifnet structure.
2007                  */
2008                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2009                         ifp = NULL;
2010                 else {
2011                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2012                         if (ifp == NULL) {
2013                                 error = EADDRNOTAVAIL;
2014                                 crit_exit();
2015                                 break;
2016                         }
2017                 }
2018                 /*
2019                  * Find the membership in the membership array.
2020                  */
2021                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2022                         if ((ifp == NULL ||
2023                              imo->imo_membership[i]->inm_ifp == ifp) &&
2024                             imo->imo_membership[i]->inm_addr.s_addr ==
2025                             mreq.imr_multiaddr.s_addr)
2026                                 break;
2027                 }
2028                 if (i == imo->imo_num_memberships) {
2029                         error = EADDRNOTAVAIL;
2030                         crit_exit();
2031                         break;
2032                 }
2033                 /*
2034                  * Give up the multicast address record to which the
2035                  * membership points.
2036                  */
2037                 in_delmulti(imo->imo_membership[i]);
2038                 /*
2039                  * Remove the gap in the membership array.
2040                  */
2041                 for (++i; i < imo->imo_num_memberships; ++i)
2042                         imo->imo_membership[i-1] = imo->imo_membership[i];
2043                 --imo->imo_num_memberships;
2044                 crit_exit();
2045                 break;
2046
2047         default:
2048                 error = EOPNOTSUPP;
2049                 break;
2050         }
2051
2052         /*
2053          * If all options have default values, no need to keep the mbuf.
2054          */
2055         if (imo->imo_multicast_ifp == NULL &&
2056             imo->imo_multicast_vif == -1 &&
2057             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2058             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2059             imo->imo_num_memberships == 0) {
2060                 kfree(*imop, M_IPMOPTS);
2061                 *imop = NULL;
2062         }
2063
2064         return (error);
2065 }
2066
2067 /*
2068  * Return the IP multicast options in response to user getsockopt().
2069  */
2070 static int
2071 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2072 {
2073         struct in_addr addr;
2074         struct in_ifaddr *ia;
2075         int error, optval;
2076         u_char coptval;
2077
2078         error = 0;
2079         switch (sopt->sopt_name) {
2080         case IP_MULTICAST_VIF:
2081                 if (imo != NULL)
2082                         optval = imo->imo_multicast_vif;
2083                 else
2084                         optval = -1;
2085                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2086                 break;
2087
2088         case IP_MULTICAST_IF:
2089                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2090                         addr.s_addr = INADDR_ANY;
2091                 else if (imo->imo_multicast_addr.s_addr) {
2092                         /* return the value user has set */
2093                         addr = imo->imo_multicast_addr;
2094                 } else {
2095                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2096                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2097                                 : IA_SIN(ia)->sin_addr.s_addr;
2098                 }
2099                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2100                 break;
2101
2102         case IP_MULTICAST_TTL:
2103                 if (imo == NULL)
2104                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2105                 else
2106                         optval = coptval = imo->imo_multicast_ttl;
2107                 if (sopt->sopt_valsize == 1)
2108                         soopt_from_kbuf(sopt, &coptval, 1);
2109                 else
2110                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2111                 break;
2112
2113         case IP_MULTICAST_LOOP:
2114                 if (imo == NULL)
2115                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2116                 else
2117                         optval = coptval = imo->imo_multicast_loop;
2118                 if (sopt->sopt_valsize == 1)
2119                         soopt_from_kbuf(sopt, &coptval, 1);
2120                 else
2121                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2122                 break;
2123
2124         default:
2125                 error = ENOPROTOOPT;
2126                 break;
2127         }
2128         return (error);
2129 }
2130
2131 /*
2132  * Discard the IP multicast options.
2133  */
2134 void
2135 ip_freemoptions(struct ip_moptions *imo)
2136 {
2137         int i;
2138
2139         if (imo != NULL) {
2140                 for (i = 0; i < imo->imo_num_memberships; ++i)
2141                         in_delmulti(imo->imo_membership[i]);
2142                 kfree(imo, M_IPMOPTS);
2143         }
2144 }
2145
2146 /*
2147  * Routine called from ip_output() to loop back a copy of an IP multicast
2148  * packet to the input queue of a specified interface.  Note that this
2149  * calls the output routine of the loopback "driver", but with an interface
2150  * pointer that might NOT be a loopback interface -- evil, but easier than
2151  * replicating that code here.
2152  */
2153 static void
2154 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2155              int hlen)
2156 {
2157         struct ip *ip;
2158         struct mbuf *copym;
2159
2160         copym = m_copypacket(m, MB_DONTWAIT);
2161         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2162                 copym = m_pullup(copym, hlen);
2163         if (copym != NULL) {
2164                 /*
2165                  * if the checksum hasn't been computed, mark it as valid
2166                  */
2167                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2168                         in_delayed_cksum(copym);
2169                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2170                         copym->m_pkthdr.csum_flags |=
2171                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2172                         copym->m_pkthdr.csum_data = 0xffff;
2173                 }
2174                 /*
2175                  * We don't bother to fragment if the IP length is greater
2176                  * than the interface's MTU.  Can this possibly matter?
2177                  */
2178                 ip = mtod(copym, struct ip *);
2179                 ip->ip_len = htons(ip->ip_len);
2180                 ip->ip_off = htons(ip->ip_off);
2181                 ip->ip_sum = 0;
2182                 if (ip->ip_vhl == IP_VHL_BORING) {
2183                         ip->ip_sum = in_cksum_hdr(ip);
2184                 } else {
2185                         ip->ip_sum = in_cksum(copym, hlen);
2186                 }
2187                 /*
2188                  * NB:
2189                  * It's not clear whether there are any lingering
2190                  * reentrancy problems in other areas which might
2191                  * be exposed by using ip_input directly (in
2192                  * particular, everything which modifies the packet
2193                  * in-place).  Yet another option is using the
2194                  * protosw directly to deliver the looped back
2195                  * packet.  For the moment, we'll err on the side
2196                  * of safety by using if_simloop().
2197                  */
2198 #if 1 /* XXX */
2199                 if (dst->sin_family != AF_INET) {
2200                         kprintf("ip_mloopback: bad address family %d\n",
2201                                                 dst->sin_family);
2202                         dst->sin_family = AF_INET;
2203                 }
2204 #endif
2205                 if_simloop(ifp, copym, dst->sin_family, 0);
2206         }
2207 }