sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.60 2008/09/12 11:37:41 sephe Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42 #include "opt_mpls.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/proc.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/thread2.h>
  55 #include <sys/in_cksum.h>
  56
  57 #include <net/if.h>
  58 #include <net/netisr.h>
  59 #include <net/pfil.h>
  60 #include <net/route.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/in_systm.h>
  64 #include <netinet/ip.h>
  65 #include <netinet/in_pcb.h>
  66 #include <netinet/in_var.h>
  67 #include <netinet/ip_var.h>
  68 #ifdef IPDIVERT
  69 #include <netinet/ip_divert.h>
  70 #endif
  71
  72 #include <netproto/mpls/mpls_var.h>
  73
  74 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  75
  76 #ifdef IPSEC
  77 #include <netinet6/ipsec.h>
  78 #include <netproto/key/key.h>
  79 #ifdef IPSEC_DEBUG
  80 #include <netproto/key/key_debug.h>
  81 #else
  82 #define KEYDEBUG(lev,arg)
  83 #endif
  84 #endif /*IPSEC*/
  85
  86 #ifdef FAST_IPSEC
  87 #include <netproto/ipsec/ipsec.h>
  88 #include <netproto/ipsec/xform.h>
  89 #include <netproto/ipsec/key.h>
  90 #endif /*FAST_IPSEC*/
  91
  92 #include <net/ipfw/ip_fw.h>
  93 #include <net/dummynet/ip_dummynet.h>
  94
  95 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  96                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  97                                   (ntohl(a.s_addr)>>16)&0xFF,\
  98                                   (ntohl(a.s_addr)>>8)&0xFF,\
  99                                   (ntohl(a.s_addr))&0xFF, y);
 100
 101 u_short ip_id;
 102
 103 #ifdef MBUF_STRESS_TEST
 104 int mbuf_frag_size = 0;
 105 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 106         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 107 #endif
 108
 109 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 110 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 111 static void     ip_mloopback
 112         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 113 static int      ip_getmoptions
 114         (struct sockopt *, struct ip_moptions *);
 115 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 116 static int      ip_setmoptions
 117         (struct sockopt *, struct ip_moptions **);
 118
 119 int     ip_optcopy(struct ip *, struct ip *);
 120
 121 extern  int route_assert_owner_access;
 122 extern  void db_print_backtrace(void);
 123
 124 extern  struct protosw inetsw[];
 125
 126 #ifdef IPDIVERT
 127 static struct mbuf *
 128 ip_divert_out(struct mbuf *m, int tee)
 129 {
 130         struct mbuf *clone = NULL;
 131         struct ip *ip = mtod(m, struct ip *);
 132
 133         /* Clone packet if we're doing a 'tee' */
 134         if (tee)
 135                 clone = m_dup(m, MB_DONTWAIT);
 136
 137         /*
 138          * XXX
 139          * delayed checksums are not currently compatible
 140          * with divert sockets.
 141          */
 142         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 143                 in_delayed_cksum(m);
 144                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 145         }
 146
 147         /* Restore packet header fields to original values */
 148         ip->ip_len = htons(ip->ip_len);
 149         ip->ip_off = htons(ip->ip_off);
 150
 151         /* Deliver packet to divert input routine */
 152         divert_packet(m, 0);
 153
 154         /* If 'tee', continue with original packet */
 155         return clone;
 156 }
 157 #endif  /* IPDIVERT */
 158
 159 static int
 160 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst)
 161 {
 162         struct in_ifaddr_container *iac;
 163
 164         /*
 165          * We need to figure out if we have been forwarded to a local
 166          * socket.  If so, then we should somehow "loop back" to
 167          * ip_input(), and get directed to the PCB as if we had received
 168          * this packet.  This is because it may be difficult to identify
 169          * the packets you want to forward until they are being output
 170          * and have selected an interface (e.g. locally initiated
 171          * packets).  If we used the loopback inteface, we would not be
 172          * able to control what happens as the packet runs through
 173          * ip_input() as it is done through a ISR.
 174          */
 175         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 176                 /*
 177                  * If the addr to forward to is one of ours, we pretend
 178                  * to be the destination for this packet.
 179                  */
 180                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 181                         break;
 182         }
 183         if (iac != NULL) {
 184                 struct ip *ip = mtod(m, struct ip *);
 185
 186                 if (m->m_pkthdr.rcvif == NULL)
 187                         m->m_pkthdr.rcvif = ifunit("lo0");
 188                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 189                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 190                                                   CSUM_PSEUDO_HDR;
 191                         m->m_pkthdr.csum_data = 0xffff;
 192                 }
 193                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 194
 195                 ip->ip_len = htons(ip->ip_len);
 196                 ip->ip_off = htons(ip->ip_off);
 197                 ip_input(m);
 198
 199                 return 1; /* Packet gets forwarded locally */
 200         }
 201         return 0;
 202 }
 203
 204 /*
 205  * IP output.  The packet in mbuf chain m contains a skeletal IP
 206  * header (with len, off, ttl, proto, tos, src, dst).
 207  * The mbuf chain containing the packet will be freed.
 208  * The mbuf opt, if present, will not be freed.
 209  */
 210 int
 211 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 212           int flags, struct ip_moptions *imo, struct inpcb *inp)
 213 {
 214         struct ip *ip;
 215         struct ifnet *ifp = NULL;       /* keep compiler happy */
 216         struct mbuf *m;
 217         int hlen = sizeof(struct ip);
 218         int len, off, error = 0;
 219         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 220         struct in_ifaddr *ia = NULL;
 221         int isbroadcast, sw_csum;
 222         struct in_addr pkt_dst;
 223         struct route iproute;
 224         struct m_tag *mtag;
 225 #ifdef IPSEC
 226         struct secpolicy *sp = NULL;
 227         struct socket *so = inp ? inp->inp_socket : NULL;
 228 #endif
 229 #ifdef FAST_IPSEC
 230         struct secpolicy *sp = NULL;
 231         struct tdb_ident *tdbi;
 232 #endif /* FAST_IPSEC */
 233         struct sockaddr_in *next_hop = NULL;
 234         int src_was_INADDR_ANY = 0;     /* as the name says... */
 235
 236         m = m0;
 237         M_ASSERTPKTHDR(m);
 238
 239         if (ro == NULL) {
 240                 ro = &iproute;
 241                 bzero(ro, sizeof *ro);
 242         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 243                 if (flags & IP_DEBUGROUTE) {
 244                         if (route_assert_owner_access) {
 245                                 panic("ip_output: "
 246                                       "rt rt_cpuid %d accessed on cpu %d\n",
 247                                       ro->ro_rt->rt_cpuid, mycpuid);
 248                         } else {
 249                                 kprintf("ip_output: "
 250                                         "rt rt_cpuid %d accessed on cpu %d\n",
 251                                         ro->ro_rt->rt_cpuid, mycpuid);
 252                                 db_print_backtrace();
 253                         }
 254                 }
 255
 256                 /*
 257                  * XXX
 258                  * If the cached rtentry's owner CPU is not the current CPU,
 259                  * then don't touch the cached rtentry (remote free is too
 260                  * expensive in this context); just relocate the route.
 261                  */
 262                 ro = &iproute;
 263                 bzero(ro, sizeof *ro);
 264         }
 265
 266         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 267                 /* Next hop */
 268                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 269                 KKASSERT(mtag != NULL);
 270                 next_hop = m_tag_data(mtag);
 271         }
 272
 273         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 274                 struct dn_pkt *dn_pkt;
 275
 276                 /* Extract info from dummynet tag */
 277                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 278                 KKASSERT(mtag != NULL);
 279                 dn_pkt = m_tag_data(mtag);
 280
 281                 /*
 282                  * The packet was already tagged, so part of the
 283                  * processing was already done, and we need to go down.
 284                  * Get the calculated parameters from the tag.
 285                  */
 286                 ifp = dn_pkt->ifp;
 287
 288                 KKASSERT(ro == &iproute);
 289                 *ro = dn_pkt->ro; /* structure copy */
 290                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 291
 292                 dst = dn_pkt->dn_dst;
 293                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 294                         /* If 'dst' points into dummynet tag, adjust it */
 295                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 296                 }
 297
 298                 ip = mtod(m, struct ip *);
 299                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 300                 if (ro->ro_rt)
 301                         ia = ifatoia(ro->ro_rt->rt_ifa);
 302                 goto sendit;
 303         }
 304
 305         if (opt) {
 306                 len = 0;
 307                 m = ip_insertoptions(m, opt, &len);
 308                 if (len != 0)
 309                         hlen = len;
 310         }
 311         ip = mtod(m, struct ip *);
 312
 313         /*
 314          * Fill in IP header.
 315          */
 316         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 317                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 318                 ip->ip_off &= IP_DF;
 319                 ip->ip_id = ip_newid();
 320                 ipstat.ips_localout++;
 321         } else {
 322                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 323         }
 324
 325 reroute:
 326         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 327
 328         dst = (struct sockaddr_in *)&ro->ro_dst;
 329         /*
 330          * If there is a cached route,
 331          * check that it is to the same destination
 332          * and is still up.  If not, free it and try again.
 333          * The address family should also be checked in case of sharing the
 334          * cache with IPv6.
 335          */
 336         if (ro->ro_rt &&
 337             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 338              dst->sin_family != AF_INET ||
 339              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 340                 rtfree(ro->ro_rt);
 341                 ro->ro_rt = (struct rtentry *)NULL;
 342         }
 343         if (ro->ro_rt == NULL) {
 344                 bzero(dst, sizeof *dst);
 345                 dst->sin_family = AF_INET;
 346                 dst->sin_len = sizeof *dst;
 347                 dst->sin_addr = pkt_dst;
 348         }
 349         /*
 350          * If routing to interface only,
 351          * short circuit routing lookup.
 352          */
 353         if (flags & IP_ROUTETOIF) {
 354                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 355                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 356                         ipstat.ips_noroute++;
 357                         error = ENETUNREACH;
 358                         goto bad;
 359                 }
 360                 ifp = ia->ia_ifp;
 361                 ip->ip_ttl = 1;
 362                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 363         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 364                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 365                 /*
 366                  * Bypass the normal routing lookup for multicast
 367                  * packets if the interface is specified.
 368                  */
 369                 ifp = imo->imo_multicast_ifp;
 370                 ia = IFP_TO_IA(ifp);
 371                 isbroadcast = 0;        /* fool gcc */
 372         } else {
 373                 /*
 374                  * If this is the case, we probably don't want to allocate
 375                  * a protocol-cloned route since we didn't get one from the
 376                  * ULP.  This lets TCP do its thing, while not burdening
 377                  * forwarding or ICMP with the overhead of cloning a route.
 378                  * Of course, we still want to do any cloning requested by
 379                  * the link layer, as this is probably required in all cases
 380                  * for correct operation (as it is for ARP).
 381                  */
 382                 if (ro->ro_rt == NULL)
 383                         rtalloc_ign(ro, RTF_PRCLONING);
 384                 if (ro->ro_rt == NULL) {
 385                         ipstat.ips_noroute++;
 386                         error = EHOSTUNREACH;
 387                         goto bad;
 388                 }
 389                 ia = ifatoia(ro->ro_rt->rt_ifa);
 390                 ifp = ro->ro_rt->rt_ifp;
 391                 ro->ro_rt->rt_use++;
 392                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 393                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 394                 if (ro->ro_rt->rt_flags & RTF_HOST)
 395                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 396                 else
 397                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 398         }
 399         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 400                 struct in_multi *inm;
 401
 402                 m->m_flags |= M_MCAST;
 403                 /*
 404                  * IP destination address is multicast.  Make sure "dst"
 405                  * still points to the address in "ro".  (It may have been
 406                  * changed to point to a gateway address, above.)
 407                  */
 408                 dst = (struct sockaddr_in *)&ro->ro_dst;
 409                 /*
 410                  * See if the caller provided any multicast options
 411                  */
 412                 if (imo != NULL) {
 413                         ip->ip_ttl = imo->imo_multicast_ttl;
 414                         if (imo->imo_multicast_vif != -1)
 415                                 ip->ip_src.s_addr =
 416                                     ip_mcast_src ?
 417                                     ip_mcast_src(imo->imo_multicast_vif) :
 418                                     INADDR_ANY;
 419                 } else
 420                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 421                 /*
 422                  * Confirm that the outgoing interface supports multicast.
 423                  */
 424                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 425                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 426                                 ipstat.ips_noroute++;
 427                                 error = ENETUNREACH;
 428                                 goto bad;
 429                         }
 430                 }
 431                 /*
 432                  * If source address not specified yet, use address
 433                  * of outgoing interface.
 434                  */
 435                 if (ip->ip_src.s_addr == INADDR_ANY) {
 436                         /* Interface may have no addresses. */
 437                         if (ia != NULL)
 438                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 439                 }
 440
 441                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 442                 if (inm != NULL &&
 443                    (imo == NULL || imo->imo_multicast_loop)) {
 444                         /*
 445                          * If we belong to the destination multicast group
 446                          * on the outgoing interface, and the caller did not
 447                          * forbid loopback, loop back a copy.
 448                          */
 449                         ip_mloopback(ifp, m, dst, hlen);
 450                 }
 451                 else {
 452                         /*
 453                          * If we are acting as a multicast router, perform
 454                          * multicast forwarding as if the packet had just
 455                          * arrived on the interface to which we are about
 456                          * to send.  The multicast forwarding function
 457                          * recursively calls this function, using the
 458                          * IP_FORWARDING flag to prevent infinite recursion.
 459                          *
 460                          * Multicasts that are looped back by ip_mloopback(),
 461                          * above, will be forwarded by the ip_input() routine,
 462                          * if necessary.
 463                          */
 464                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 465                                 /*
 466                                  * If rsvp daemon is not running, do not
 467                                  * set ip_moptions. This ensures that the packet
 468                                  * is multicast and not just sent down one link
 469                                  * as prescribed by rsvpd.
 470                                  */
 471                                 if (!rsvp_on)
 472                                         imo = NULL;
 473                                 if (ip_mforward &&
 474                                     ip_mforward(ip, ifp, m, imo) != 0) {
 475                                         m_freem(m);
 476                                         goto done;
 477                                 }
 478                         }
 479                 }
 480
 481                 /*
 482                  * Multicasts with a time-to-live of zero may be looped-
 483                  * back, above, but must not be transmitted on a network.
 484                  * Also, multicasts addressed to the loopback interface
 485                  * are not sent -- the above call to ip_mloopback() will
 486                  * loop back a copy if this host actually belongs to the
 487                  * destination group on the loopback interface.
 488                  */
 489                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 490                         m_freem(m);
 491                         goto done;
 492                 }
 493
 494                 goto sendit;
 495         } else {
 496                 m->m_flags &= ~M_MCAST;
 497         }
 498
 499         /*
 500          * If the source address is not specified yet, use the address
 501          * of the outoing interface. In case, keep note we did that, so
 502          * if the the firewall changes the next-hop causing the output
 503          * interface to change, we can fix that.
 504          */
 505         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 506                 /* Interface may have no addresses. */
 507                 if (ia != NULL) {
 508                         ip->ip_src = IA_SIN(ia)->sin_addr;
 509                         src_was_INADDR_ANY = 1;
 510                 }
 511         }
 512
 513 #ifdef ALTQ
 514         /*
 515          * Disable packet drop hack.
 516          * Packetdrop should be done by queueing.
 517          */
 518 #else /* !ALTQ */
 519         /*
 520          * Verify that we have any chance at all of being able to queue
 521          *      the packet or packet fragments
 522          */
 523         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 524                 ifp->if_snd.ifq_maxlen) {
 525                         error = ENOBUFS;
 526                         ipstat.ips_odropped++;
 527                         goto bad;
 528         }
 529 #endif /* !ALTQ */
 530
 531         /*
 532          * Look for broadcast address and
 533          * verify user is allowed to send
 534          * such a packet.
 535          */
 536         if (isbroadcast) {
 537                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 538                         error = EADDRNOTAVAIL;
 539                         goto bad;
 540                 }
 541                 if (!(flags & IP_ALLOWBROADCAST)) {
 542                         error = EACCES;
 543                         goto bad;
 544                 }
 545                 /* don't allow broadcast messages to be fragmented */
 546                 if (ip->ip_len > ifp->if_mtu) {
 547                         error = EMSGSIZE;
 548                         goto bad;
 549                 }
 550                 m->m_flags |= M_BCAST;
 551         } else {
 552                 m->m_flags &= ~M_BCAST;
 553         }
 554
 555 sendit:
 556 #ifdef IPSEC
 557         /* get SP for this packet */
 558         if (so == NULL)
 559                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 560         else
 561                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 562
 563         if (sp == NULL) {
 564                 ipsecstat.out_inval++;
 565                 goto bad;
 566         }
 567
 568         error = 0;
 569
 570         /* check policy */
 571         switch (sp->policy) {
 572         case IPSEC_POLICY_DISCARD:
 573                 /*
 574                  * This packet is just discarded.
 575                  */
 576                 ipsecstat.out_polvio++;
 577                 goto bad;
 578
 579         case IPSEC_POLICY_BYPASS:
 580         case IPSEC_POLICY_NONE:
 581                 /* no need to do IPsec. */
 582                 goto skip_ipsec;
 583
 584         case IPSEC_POLICY_IPSEC:
 585                 if (sp->req == NULL) {
 586                         /* acquire a policy */
 587                         error = key_spdacquire(sp);
 588                         goto bad;
 589                 }
 590                 break;
 591
 592         case IPSEC_POLICY_ENTRUST:
 593         default:
 594                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 595         }
 596     {
 597         struct ipsec_output_state state;
 598         bzero(&state, sizeof state);
 599         state.m = m;
 600         if (flags & IP_ROUTETOIF) {
 601                 state.ro = &iproute;
 602                 bzero(&iproute, sizeof iproute);
 603         } else
 604                 state.ro = ro;
 605         state.dst = (struct sockaddr *)dst;
 606
 607         ip->ip_sum = 0;
 608
 609         /*
 610          * XXX
 611          * delayed checksums are not currently compatible with IPsec
 612          */
 613         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 614                 in_delayed_cksum(m);
 615                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 616         }
 617
 618         ip->ip_len = htons(ip->ip_len);
 619         ip->ip_off = htons(ip->ip_off);
 620
 621         error = ipsec4_output(&state, sp, flags);
 622
 623         m = state.m;
 624         if (flags & IP_ROUTETOIF) {
 625                 /*
 626                  * if we have tunnel mode SA, we may need to ignore
 627                  * IP_ROUTETOIF.
 628                  */
 629                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 630                         flags &= ~IP_ROUTETOIF;
 631                         ro = state.ro;
 632                 }
 633         } else
 634                 ro = state.ro;
 635         dst = (struct sockaddr_in *)state.dst;
 636         if (error) {
 637                 /* mbuf is already reclaimed in ipsec4_output. */
 638                 m0 = NULL;
 639                 switch (error) {
 640                 case EHOSTUNREACH:
 641                 case ENETUNREACH:
 642                 case EMSGSIZE:
 643                 case ENOBUFS:
 644                 case ENOMEM:
 645                         break;
 646                 default:
 647                         kprintf("ip4_output (ipsec): error code %d\n", error);
 648                         /*fall through*/
 649                 case ENOENT:
 650                         /* don't show these error codes to the user */
 651                         error = 0;
 652                         break;
 653                 }
 654                 goto bad;
 655         }
 656     }
 657
 658         /* be sure to update variables that are affected by ipsec4_output() */
 659         ip = mtod(m, struct ip *);
 660 #ifdef _IP_VHL
 661         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 662 #else
 663         hlen = ip->ip_hl << 2;
 664 #endif
 665         if (ro->ro_rt == NULL) {
 666                 if (!(flags & IP_ROUTETOIF)) {
 667                         kprintf("ip_output: "
 668                                 "can't update route after IPsec processing\n");
 669                         error = EHOSTUNREACH;   /*XXX*/
 670                         goto bad;
 671                 }
 672         } else {
 673                 ia = ifatoia(ro->ro_rt->rt_ifa);
 674                 ifp = ro->ro_rt->rt_ifp;
 675         }
 676
 677         /* make it flipped, again. */
 678         ip->ip_len = ntohs(ip->ip_len);
 679         ip->ip_off = ntohs(ip->ip_off);
 680 skip_ipsec:
 681 #endif /*IPSEC*/
 682 #ifdef FAST_IPSEC
 683         /*
 684          * Check the security policy (SP) for the packet and, if
 685          * required, do IPsec-related processing.  There are two
 686          * cases here; the first time a packet is sent through
 687          * it will be untagged and handled by ipsec4_checkpolicy.
 688          * If the packet is resubmitted to ip_output (e.g. after
 689          * AH, ESP, etc. processing), there will be a tag to bypass
 690          * the lookup and related policy checking.
 691          */
 692         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 693         crit_enter();
 694         if (mtag != NULL) {
 695                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 696                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 697                 if (sp == NULL)
 698                         error = -EINVAL;        /* force silent drop */
 699                 m_tag_delete(m, mtag);
 700         } else {
 701                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 702                                         &error, inp);
 703         }
 704         /*
 705          * There are four return cases:
 706          *    sp != NULL                    apply IPsec policy
 707          *    sp == NULL, error == 0        no IPsec handling needed
 708          *    sp == NULL, error == -EINVAL  discard packet w/o error
 709          *    sp == NULL, error != 0        discard packet, report error
 710          */
 711         if (sp != NULL) {
 712                 /* Loop detection, check if ipsec processing already done */
 713                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 714                 for (mtag = m_tag_first(m); mtag != NULL;
 715                      mtag = m_tag_next(m, mtag)) {
 716                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 717                                 continue;
 718                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 719                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 720                                 continue;
 721                         /*
 722                          * Check if policy has an SA associated with it.
 723                          * This can happen when an SP has yet to acquire
 724                          * an SA; e.g. on first reference.  If it occurs,
 725                          * then we let ipsec4_process_packet do its thing.
 726                          */
 727                         if (sp->req->sav == NULL)
 728                                 break;
 729                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 730                         if (tdbi->spi == sp->req->sav->spi &&
 731                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 732                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 733                                  sizeof(union sockaddr_union)) == 0) {
 734                                 /*
 735                                  * No IPsec processing is needed, free
 736                                  * reference to SP.
 737                                  *
 738                                  * NB: null pointer to avoid free at
 739                                  *     done: below.
 740                                  */
 741                                 KEY_FREESP(&sp), sp = NULL;
 742                                 crit_exit();
 743                                 goto spd_done;
 744                         }
 745                 }
 746
 747                 /*
 748                  * Do delayed checksums now because we send before
 749                  * this is done in the normal processing path.
 750                  */
 751                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 752                         in_delayed_cksum(m);
 753                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 754                 }
 755
 756                 ip->ip_len = htons(ip->ip_len);
 757                 ip->ip_off = htons(ip->ip_off);
 758
 759                 /* NB: callee frees mbuf */
 760                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 761                 /*
 762                  * Preserve KAME behaviour: ENOENT can be returned
 763                  * when an SA acquire is in progress.  Don't propagate
 764                  * this to user-level; it confuses applications.
 765                  *
 766                  * XXX this will go away when the SADB is redone.
 767                  */
 768                 if (error == ENOENT)
 769                         error = 0;
 770                 crit_exit();
 771                 goto done;
 772         } else {
 773                 crit_exit();
 774
 775                 if (error != 0) {
 776                         /*
 777                          * Hack: -EINVAL is used to signal that a packet
 778                          * should be silently discarded.  This is typically
 779                          * because we asked key management for an SA and
 780                          * it was delayed (e.g. kicked up to IKE).
 781                          */
 782                         if (error == -EINVAL)
 783                                 error = 0;
 784                         goto bad;
 785                 } else {
 786                         /* No IPsec processing for this packet. */
 787                 }
 788 #ifdef notyet
 789                 /*
 790                  * If deferred crypto processing is needed, check that
 791                  * the interface supports it.
 792                  */
 793                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 794                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 795                         /* notify IPsec to do its own crypto */
 796                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 797                         error = EHOSTUNREACH;
 798                         goto bad;
 799                 }
 800 #endif
 801         }
 802 spd_done:
 803 #endif /* FAST_IPSEC */
 804
 805         /* We are already being fwd'd from a firewall. */
 806         if (next_hop != NULL)
 807                 goto pass;
 808
 809         /*
 810          * IpHack's section.
 811          * - Xlate: translate packet's addr/port (NAT).
 812          * - Firewall: deny/allow/etc.
 813          * - Wrap: fake packet's addr/port <unimpl.>
 814          * - Encapsulate: put it in another IP and send out. <unimp.>
 815          */
 816
 817         /*
 818          * Run through list of hooks for output packets.
 819          */
 820         if (pfil_has_hooks(&inet_pfil_hook)) {
 821                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 822                 if (error != 0 || m == NULL)
 823                         goto done;
 824                 ip = mtod(m, struct ip *);
 825         }
 826
 827         /*
 828          * Check with the firewall...
 829          */
 830         if (fw_enable && IPFW_LOADED) {
 831                 struct ip_fw_args args;
 832                 int tee = 0;
 833
 834                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 835                         /* Extract info from dummynet tag */
 836                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 837                         KKASSERT(mtag != NULL);
 838                         args.rule =
 839                         ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
 840                         KKASSERT(args.rule != NULL);
 841
 842                         m_tag_delete(m, mtag);
 843                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 844                 } else {
 845                         args.rule = NULL;
 846                 }
 847
 848                 args.eh = NULL;
 849                 args.m = m;
 850                 args.oif = ifp;
 851                 off = ip_fw_chk_ptr(&args);
 852                 m = args.m;
 853
 854                 if (m == NULL) {
 855                         error = EACCES;
 856                         goto done;
 857                 }
 858                 ip = mtod(m, struct ip *);
 859
 860                 switch (off) {
 861                 case IP_FW_PASS:
 862                         break;
 863
 864                 case IP_FW_DENY:
 865                         m_freem(m);
 866                         error = EACCES;
 867                         goto done;
 868
 869                 case IP_FW_DUMMYNET:
 870                         /*
 871                          * pass the pkt to dummynet. Need to include
 872                          * pipe number, m, ifp, ro, dst because these are
 873                          * not recomputed in the next pass.
 874                          * All other parameters have been already used and
 875                          * so they are not needed anymore.
 876                          * XXX note: if the ifp or ro entry are deleted
 877                          * while a pkt is in dummynet, we are in trouble!
 878                          */
 879                         args.ro = ro;
 880                         args.dst = dst;
 881                         args.flags = flags;
 882
 883                         error = 0;
 884                         ip_fw_dn_io_ptr(m, args.cookie, DN_TO_IP_OUT, &args);
 885                         goto done;
 886
 887                 case IP_FW_TEE:
 888                         tee = 1;
 889                         /* FALL THROUGH */
 890
 891                 case IP_FW_DIVERT:
 892 #ifdef IPDIVERT
 893                         m = ip_divert_out(m, tee);
 894                         if (m == NULL)
 895                                 goto done;
 896                         ip = mtod(m, struct ip *);
 897                         break;
 898 #else
 899                         m_freem(m);
 900                         /* not sure this is the right error msg */
 901                         error = EACCES;
 902                         goto done;
 903 #endif
 904
 905                 default:
 906                         panic("unknown ipfw return value: %d\n", off);
 907                 }
 908         }
 909
 910         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 911                 /*
 912                  * Check dst to make sure it is directly reachable on the
 913                  * interface we previously thought it was.
 914                  * If it isn't (which may be likely in some situations) we have
 915                  * to re-route it (ie, find a route for the next-hop and the
 916                  * associated interface) and set them here. This is nested
 917                  * forwarding which in most cases is undesirable, except where
 918                  * such control is nigh impossible. So we do it here.
 919                  * And I'm babbling.
 920                  */
 921                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 922                 KKASSERT(mtag != NULL);
 923                 next_hop = m_tag_data(mtag);
 924
 925                 /*
 926                  * Try local forwarding first
 927                  */
 928                 if (ip_localforward(m, next_hop))
 929                         goto done;
 930
 931                 /*
 932                  * Relocate the route based on next_hop.
 933                  * If the current route is inp's cache, keep it untouched.
 934                  */
 935                 if (ro == &iproute && ro->ro_rt != NULL) {
 936                         RTFREE(ro->ro_rt);
 937                         ro->ro_rt = NULL;
 938                 }
 939                 ro = &iproute;
 940                 bzero(ro, sizeof *ro);
 941
 942                 /*
 943                  * Forwarding to broadcast address is not allowed.
 944                  * XXX Should we follow IP_ROUTETOIF?
 945                  */
 946                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 947
 948                 /* We are doing forwarding now */
 949                 flags |= IP_FORWARDING;
 950
 951                 goto reroute;
 952         }
 953
 954 pass:
 955         /* 127/8 must not appear on wire - RFC1122. */
 956         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 957             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 958                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 959                         ipstat.ips_badaddr++;
 960                         error = EADDRNOTAVAIL;
 961                         goto bad;
 962                 }
 963         }
 964
 965         m->m_pkthdr.csum_flags |= CSUM_IP;
 966         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 967         if (sw_csum & CSUM_DELAY_DATA) {
 968                 in_delayed_cksum(m);
 969                 sw_csum &= ~CSUM_DELAY_DATA;
 970         }
 971         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 972
 973         /*
 974          * If small enough for interface, or the interface will take
 975          * care of the fragmentation for us, can just send directly.
 976          */
 977         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 978             !(ip->ip_off & IP_DF))) {
 979                 ip->ip_len = htons(ip->ip_len);
 980                 ip->ip_off = htons(ip->ip_off);
 981                 ip->ip_sum = 0;
 982                 if (sw_csum & CSUM_DELAY_IP) {
 983                         if (ip->ip_vhl == IP_VHL_BORING) {
 984                                 ip->ip_sum = in_cksum_hdr(ip);
 985                         } else {
 986                                 ip->ip_sum = in_cksum(m, hlen);
 987                         }
 988                 }
 989
 990                 /* Record statistics for this interface address. */
 991                 if (!(flags & IP_FORWARDING) && ia) {
 992                         ia->ia_ifa.if_opackets++;
 993                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 994                 }
 995
 996 #ifdef IPSEC
 997                 /* clean ipsec history once it goes out of the node */
 998                 ipsec_delaux(m);
 999 #endif
1000
1001 #ifdef MBUF_STRESS_TEST
1002                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1003                         struct mbuf *m1, *m2;
1004                         int length, tmp;
1005
1006                         tmp = length = m->m_pkthdr.len;
1007
1008                         while ((length -= mbuf_frag_size) >= 1) {
1009                                 m1 = m_split(m, length, MB_DONTWAIT);
1010                                 if (m1 == NULL)
1011                                         break;
1012                                 m2 = m;
1013                                 while (m2->m_next != NULL)
1014                                         m2 = m2->m_next;
1015                                 m2->m_next = m1;
1016                         }
1017                         m->m_pkthdr.len = tmp;
1018                 }
1019 #endif
1020
1021 #ifdef MPLS
1022                 if (!mpls_output_process(m, ro->ro_rt))
1023                         goto done;
1024 #endif
1025                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1026                                        ro->ro_rt);
1027                 goto done;
1028         }
1029
1030         if (ip->ip_off & IP_DF) {
1031                 error = EMSGSIZE;
1032                 /*
1033                  * This case can happen if the user changed the MTU
1034                  * of an interface after enabling IP on it.  Because
1035                  * most netifs don't keep track of routes pointing to
1036                  * them, there is no way for one to update all its
1037                  * routes when the MTU is changed.
1038                  */
1039                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1040                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1041                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1042                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1043                 }
1044                 ipstat.ips_cantfrag++;
1045                 goto bad;
1046         }
1047
1048         /*
1049          * Too large for interface; fragment if possible. If successful,
1050          * on return, m will point to a list of packets to be sent.
1051          */
1052         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1053         if (error)
1054                 goto bad;
1055         for (; m; m = m0) {
1056                 m0 = m->m_nextpkt;
1057                 m->m_nextpkt = NULL;
1058 #ifdef IPSEC
1059                 /* clean ipsec history once it goes out of the node */
1060                 ipsec_delaux(m);
1061 #endif
1062                 if (error == 0) {
1063                         /* Record statistics for this interface address. */
1064                         if (ia != NULL) {
1065                                 ia->ia_ifa.if_opackets++;
1066                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1067                         }
1068 #ifdef MPLS
1069                         if (!mpls_output_process(m, ro->ro_rt))
1070                                 continue;
1071 #endif
1072                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1073                                                ro->ro_rt);
1074                 } else {
1075                         m_freem(m);
1076                 }
1077         }
1078
1079         if (error == 0)
1080                 ipstat.ips_fragmented++;
1081
1082 done:
1083         if (ro == &iproute && ro->ro_rt != NULL) {
1084                 RTFREE(ro->ro_rt);
1085                 ro->ro_rt = NULL;
1086         }
1087 #ifdef IPSEC
1088         if (sp != NULL) {
1089                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1090                         kprintf("DP ip_output call free SP:%p\n", sp));
1091                 key_freesp(sp);
1092         }
1093 #endif
1094 #ifdef FAST_IPSEC
1095         if (sp != NULL)
1096                 KEY_FREESP(&sp);
1097 #endif
1098         return (error);
1099 bad:
1100         m_freem(m);
1101         goto done;
1102 }
1103
1104 /*
1105  * Create a chain of fragments which fit the given mtu. m_frag points to the
1106  * mbuf to be fragmented; on return it points to the chain with the fragments.
1107  * Return 0 if no error. If error, m_frag may contain a partially built
1108  * chain of fragments that should be freed by the caller.
1109  *
1110  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1111  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1112  */
1113 int
1114 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1115             u_long if_hwassist_flags, int sw_csum)
1116 {
1117         int error = 0;
1118         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1119         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1120         int off;
1121         struct mbuf *m0 = *m_frag;      /* the original packet          */
1122         int firstlen;
1123         struct mbuf **mnext;
1124         int nfrags;
1125
1126         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1127                 ipstat.ips_cantfrag++;
1128                 return EMSGSIZE;
1129         }
1130
1131         /*
1132          * Must be able to put at least 8 bytes per fragment.
1133          */
1134         if (len < 8)
1135                 return EMSGSIZE;
1136
1137         /*
1138          * If the interface will not calculate checksums on
1139          * fragmented packets, then do it here.
1140          */
1141         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1142             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1143                 in_delayed_cksum(m0);
1144                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1145         }
1146
1147         if (len > PAGE_SIZE) {
1148                 /*
1149                  * Fragment large datagrams such that each segment
1150                  * contains a multiple of PAGE_SIZE amount of data,
1151                  * plus headers. This enables a receiver to perform
1152                  * page-flipping zero-copy optimizations.
1153                  *
1154                  * XXX When does this help given that sender and receiver
1155                  * could have different page sizes, and also mtu could
1156                  * be less than the receiver's page size ?
1157                  */
1158                 int newlen;
1159                 struct mbuf *m;
1160
1161                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1162                         off += m->m_len;
1163
1164                 /*
1165                  * firstlen (off - hlen) must be aligned on an
1166                  * 8-byte boundary
1167                  */
1168                 if (off < hlen)
1169                         goto smart_frag_failure;
1170                 off = ((off - hlen) & ~7) + hlen;
1171                 newlen = (~PAGE_MASK) & mtu;
1172                 if ((newlen + sizeof(struct ip)) > mtu) {
1173                         /* we failed, go back the default */
1174 smart_frag_failure:
1175                         newlen = len;
1176                         off = hlen + len;
1177                 }
1178                 len = newlen;
1179
1180         } else {
1181                 off = hlen + len;
1182         }
1183
1184         firstlen = off - hlen;
1185         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1186
1187         /*
1188          * Loop through length of segment after first fragment,
1189          * make new header and copy data of each part and link onto chain.
1190          * Here, m0 is the original packet, m is the fragment being created.
1191          * The fragments are linked off the m_nextpkt of the original
1192          * packet, which after processing serves as the first fragment.
1193          */
1194         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1195                 struct ip *mhip;        /* ip header on the fragment */
1196                 struct mbuf *m;
1197                 int mhlen = sizeof(struct ip);
1198
1199                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1200                 if (m == NULL) {
1201                         error = ENOBUFS;
1202                         ipstat.ips_odropped++;
1203                         goto done;
1204                 }
1205                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1206                 /*
1207                  * In the first mbuf, leave room for the link header, then
1208                  * copy the original IP header including options. The payload
1209                  * goes into an additional mbuf chain returned by m_copy().
1210                  */
1211                 m->m_data += max_linkhdr;
1212                 mhip = mtod(m, struct ip *);
1213                 *mhip = *ip;
1214                 if (hlen > sizeof(struct ip)) {
1215                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1216                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1217                 }
1218                 m->m_len = mhlen;
1219                 /* XXX do we need to add ip->ip_off below ? */
1220                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1221                 if (off + len >= ip->ip_len) {  /* last fragment */
1222                         len = ip->ip_len - off;
1223                         m->m_flags |= M_LASTFRAG;
1224                 } else
1225                         mhip->ip_off |= IP_MF;
1226                 mhip->ip_len = htons((u_short)(len + mhlen));
1227                 m->m_next = m_copy(m0, off, len);
1228                 if (m->m_next == NULL) {                /* copy failed */
1229                         m_free(m);
1230                         error = ENOBUFS;        /* ??? */
1231                         ipstat.ips_odropped++;
1232                         goto done;
1233                 }
1234                 m->m_pkthdr.len = mhlen + len;
1235                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1236                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1237                 mhip->ip_off = htons(mhip->ip_off);
1238                 mhip->ip_sum = 0;
1239                 if (sw_csum & CSUM_DELAY_IP)
1240                         mhip->ip_sum = in_cksum(m, mhlen);
1241                 *mnext = m;
1242                 mnext = &m->m_nextpkt;
1243         }
1244         ipstat.ips_ofragments += nfrags;
1245
1246         /* set first marker for fragment chain */
1247         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1248         m0->m_pkthdr.csum_data = nfrags;
1249
1250         /*
1251          * Update first fragment by trimming what's been copied out
1252          * and updating header.
1253          */
1254         m_adj(m0, hlen + firstlen - ip->ip_len);
1255         m0->m_pkthdr.len = hlen + firstlen;
1256         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1257         ip->ip_off |= IP_MF;
1258         ip->ip_off = htons(ip->ip_off);
1259         ip->ip_sum = 0;
1260         if (sw_csum & CSUM_DELAY_IP)
1261                 ip->ip_sum = in_cksum(m0, hlen);
1262
1263 done:
1264         *m_frag = m0;
1265         return error;
1266 }
1267
1268 void
1269 in_delayed_cksum(struct mbuf *m)
1270 {
1271         struct ip *ip;
1272         u_short csum, offset;
1273
1274         ip = mtod(m, struct ip *);
1275         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1276         csum = in_cksum_skip(m, ip->ip_len, offset);
1277         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1278                 csum = 0xffff;
1279         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1280
1281         if (offset + sizeof(u_short) > m->m_len) {
1282                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1283                     m->m_len, offset, ip->ip_p);
1284                 /*
1285                  * XXX
1286                  * this shouldn't happen, but if it does, the
1287                  * correct behavior may be to insert the checksum
1288                  * in the existing chain instead of rearranging it.
1289                  */
1290                 m = m_pullup(m, offset + sizeof(u_short));
1291         }
1292         *(u_short *)(m->m_data + offset) = csum;
1293 }
1294
1295 /*
1296  * Insert IP options into preformed packet.
1297  * Adjust IP destination as required for IP source routing,
1298  * as indicated by a non-zero in_addr at the start of the options.
1299  *
1300  * XXX This routine assumes that the packet has no options in place.
1301  */
1302 static struct mbuf *
1303 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1304 {
1305         struct ipoption *p = mtod(opt, struct ipoption *);
1306         struct mbuf *n;
1307         struct ip *ip = mtod(m, struct ip *);
1308         unsigned optlen;
1309
1310         optlen = opt->m_len - sizeof p->ipopt_dst;
1311         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1312                 *phlen = 0;
1313                 return (m);             /* XXX should fail */
1314         }
1315         if (p->ipopt_dst.s_addr)
1316                 ip->ip_dst = p->ipopt_dst;
1317         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1318                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1319                 if (n == NULL) {
1320                         *phlen = 0;
1321                         return (m);
1322                 }
1323                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1324                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1325                 m->m_len -= sizeof(struct ip);
1326                 m->m_data += sizeof(struct ip);
1327                 n->m_next = m;
1328                 m = n;
1329                 m->m_len = optlen + sizeof(struct ip);
1330                 m->m_data += max_linkhdr;
1331                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1332         } else {
1333                 m->m_data -= optlen;
1334                 m->m_len += optlen;
1335                 m->m_pkthdr.len += optlen;
1336                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1337         }
1338         ip = mtod(m, struct ip *);
1339         bcopy(p->ipopt_list, ip + 1, optlen);
1340         *phlen = sizeof(struct ip) + optlen;
1341         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1342         ip->ip_len += optlen;
1343         return (m);
1344 }
1345
1346 /*
1347  * Copy options from ip to jp,
1348  * omitting those not copied during fragmentation.
1349  */
1350 int
1351 ip_optcopy(struct ip *ip, struct ip *jp)
1352 {
1353         u_char *cp, *dp;
1354         int opt, optlen, cnt;
1355
1356         cp = (u_char *)(ip + 1);
1357         dp = (u_char *)(jp + 1);
1358         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1359         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1360                 opt = cp[0];
1361                 if (opt == IPOPT_EOL)
1362                         break;
1363                 if (opt == IPOPT_NOP) {
1364                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1365                         *dp++ = IPOPT_NOP;
1366                         optlen = 1;
1367                         continue;
1368                 }
1369
1370                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1371                     ("ip_optcopy: malformed ipv4 option"));
1372                 optlen = cp[IPOPT_OLEN];
1373                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1374                     ("ip_optcopy: malformed ipv4 option"));
1375
1376                 /* bogus lengths should have been caught by ip_dooptions */
1377                 if (optlen > cnt)
1378                         optlen = cnt;
1379                 if (IPOPT_COPIED(opt)) {
1380                         bcopy(cp, dp, optlen);
1381                         dp += optlen;
1382                 }
1383         }
1384         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1385                 *dp++ = IPOPT_EOL;
1386         return (optlen);
1387 }
1388
1389 /*
1390  * IP socket option processing.
1391  */
1392 int
1393 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1394 {
1395         struct  inpcb *inp = so->so_pcb;
1396         int     error, optval;
1397
1398         error = optval = 0;
1399         if (sopt->sopt_level != IPPROTO_IP) {
1400                 return (EINVAL);
1401         }
1402
1403         switch (sopt->sopt_dir) {
1404         case SOPT_SET:
1405                 switch (sopt->sopt_name) {
1406                 case IP_OPTIONS:
1407 #ifdef notyet
1408                 case IP_RETOPTS:
1409 #endif
1410                 {
1411                         struct mbuf *m;
1412                         if (sopt->sopt_valsize > MLEN) {
1413                                 error = EMSGSIZE;
1414                                 break;
1415                         }
1416                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1417                         if (m == NULL) {
1418                                 error = ENOBUFS;
1419                                 break;
1420                         }
1421                         m->m_len = sopt->sopt_valsize;
1422                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1423                                               m->m_len);
1424                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1425                                            m));
1426                 }
1427
1428                 case IP_TOS:
1429                 case IP_TTL:
1430                 case IP_MINTTL:
1431                 case IP_RECVOPTS:
1432                 case IP_RECVRETOPTS:
1433                 case IP_RECVDSTADDR:
1434                 case IP_RECVIF:
1435                 case IP_RECVTTL:
1436                 case IP_FAITH:
1437                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1438                                              sizeof optval);
1439                         if (error)
1440                                 break;
1441                         switch (sopt->sopt_name) {
1442                         case IP_TOS:
1443                                 inp->inp_ip_tos = optval;
1444                                 break;
1445
1446                         case IP_TTL:
1447                                 inp->inp_ip_ttl = optval;
1448                                 break;
1449                         case IP_MINTTL:
1450                                 if (optval > 0 && optval <= MAXTTL)
1451                                         inp->inp_ip_minttl = optval;
1452                                 else
1453                                         error = EINVAL;
1454                                 break;
1455 #define OPTSET(bit) \
1456         if (optval) \
1457                 inp->inp_flags |= bit; \
1458         else \
1459                 inp->inp_flags &= ~bit;
1460
1461                         case IP_RECVOPTS:
1462                                 OPTSET(INP_RECVOPTS);
1463                                 break;
1464
1465                         case IP_RECVRETOPTS:
1466                                 OPTSET(INP_RECVRETOPTS);
1467                                 break;
1468
1469                         case IP_RECVDSTADDR:
1470                                 OPTSET(INP_RECVDSTADDR);
1471                                 break;
1472
1473                         case IP_RECVIF:
1474                                 OPTSET(INP_RECVIF);
1475                                 break;
1476
1477                         case IP_RECVTTL:
1478                                 OPTSET(INP_RECVTTL);
1479                                 break;
1480
1481                         case IP_FAITH:
1482                                 OPTSET(INP_FAITH);
1483                                 break;
1484                         }
1485                         break;
1486 #undef OPTSET
1487
1488                 case IP_MULTICAST_IF:
1489                 case IP_MULTICAST_VIF:
1490                 case IP_MULTICAST_TTL:
1491                 case IP_MULTICAST_LOOP:
1492                 case IP_ADD_MEMBERSHIP:
1493                 case IP_DROP_MEMBERSHIP:
1494                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1495                         break;
1496
1497                 case IP_PORTRANGE:
1498                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1499                                             sizeof optval);
1500                         if (error)
1501                                 break;
1502
1503                         switch (optval) {
1504                         case IP_PORTRANGE_DEFAULT:
1505                                 inp->inp_flags &= ~(INP_LOWPORT);
1506                                 inp->inp_flags &= ~(INP_HIGHPORT);
1507                                 break;
1508
1509                         case IP_PORTRANGE_HIGH:
1510                                 inp->inp_flags &= ~(INP_LOWPORT);
1511                                 inp->inp_flags |= INP_HIGHPORT;
1512                                 break;
1513
1514                         case IP_PORTRANGE_LOW:
1515                                 inp->inp_flags &= ~(INP_HIGHPORT);
1516                                 inp->inp_flags |= INP_LOWPORT;
1517                                 break;
1518
1519                         default:
1520                                 error = EINVAL;
1521                                 break;
1522                         }
1523                         break;
1524
1525 #if defined(IPSEC) || defined(FAST_IPSEC)
1526                 case IP_IPSEC_POLICY:
1527                 {
1528                         caddr_t req;
1529                         size_t len = 0;
1530                         int priv;
1531                         struct mbuf *m;
1532                         int optname;
1533
1534                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1535                                 break;
1536                         soopt_to_mbuf(sopt, m);
1537                         priv = (sopt->sopt_td != NULL &&
1538                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1539                         req = mtod(m, caddr_t);
1540                         len = m->m_len;
1541                         optname = sopt->sopt_name;
1542                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1543                         m_freem(m);
1544                         break;
1545                 }
1546 #endif /*IPSEC*/
1547
1548                 default:
1549                         error = ENOPROTOOPT;
1550                         break;
1551                 }
1552                 break;
1553
1554         case SOPT_GET:
1555                 switch (sopt->sopt_name) {
1556                 case IP_OPTIONS:
1557                 case IP_RETOPTS:
1558                         if (inp->inp_options)
1559                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1560                                                            char *),
1561                                                 inp->inp_options->m_len);
1562                         else
1563                                 sopt->sopt_valsize = 0;
1564                         break;
1565
1566                 case IP_TOS:
1567                 case IP_TTL:
1568                 case IP_MINTTL:
1569                 case IP_RECVOPTS:
1570                 case IP_RECVRETOPTS:
1571                 case IP_RECVDSTADDR:
1572                 case IP_RECVTTL:
1573                 case IP_RECVIF:
1574                 case IP_PORTRANGE:
1575                 case IP_FAITH:
1576                         switch (sopt->sopt_name) {
1577
1578                         case IP_TOS:
1579                                 optval = inp->inp_ip_tos;
1580                                 break;
1581
1582                         case IP_TTL:
1583                                 optval = inp->inp_ip_ttl;
1584                                 break;
1585                         case IP_MINTTL:
1586                                 optval = inp->inp_ip_minttl;
1587                                 break;
1588
1589 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1590
1591                         case IP_RECVOPTS:
1592                                 optval = OPTBIT(INP_RECVOPTS);
1593                                 break;
1594
1595                         case IP_RECVRETOPTS:
1596                                 optval = OPTBIT(INP_RECVRETOPTS);
1597                                 break;
1598
1599                         case IP_RECVDSTADDR:
1600                                 optval = OPTBIT(INP_RECVDSTADDR);
1601                                 break;
1602
1603                         case IP_RECVTTL:
1604                                 optval = OPTBIT(INP_RECVTTL);
1605                                 break;
1606
1607                         case IP_RECVIF:
1608                                 optval = OPTBIT(INP_RECVIF);
1609                                 break;
1610
1611                         case IP_PORTRANGE:
1612                                 if (inp->inp_flags & INP_HIGHPORT)
1613                                         optval = IP_PORTRANGE_HIGH;
1614                                 else if (inp->inp_flags & INP_LOWPORT)
1615                                         optval = IP_PORTRANGE_LOW;
1616                                 else
1617                                         optval = 0;
1618                                 break;
1619
1620                         case IP_FAITH:
1621                                 optval = OPTBIT(INP_FAITH);
1622                                 break;
1623                         }
1624                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1625                         break;
1626
1627                 case IP_MULTICAST_IF:
1628                 case IP_MULTICAST_VIF:
1629                 case IP_MULTICAST_TTL:
1630                 case IP_MULTICAST_LOOP:
1631                 case IP_ADD_MEMBERSHIP:
1632                 case IP_DROP_MEMBERSHIP:
1633                         error = ip_getmoptions(sopt, inp->inp_moptions);
1634                         break;
1635
1636 #if defined(IPSEC) || defined(FAST_IPSEC)
1637                 case IP_IPSEC_POLICY:
1638                 {
1639                         struct mbuf *m = NULL;
1640                         caddr_t req = NULL;
1641                         size_t len = 0;
1642
1643                         if (m != NULL) {
1644                                 req = mtod(m, caddr_t);
1645                                 len = m->m_len;
1646                         }
1647                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1648                         if (error == 0)
1649                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1650                         if (error == 0)
1651                                 m_freem(m);
1652                         break;
1653                 }
1654 #endif /*IPSEC*/
1655
1656                 default:
1657                         error = ENOPROTOOPT;
1658                         break;
1659                 }
1660                 break;
1661         }
1662         return (error);
1663 }
1664
1665 /*
1666  * Set up IP options in pcb for insertion in output packets.
1667  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1668  * with destination address if source routed.
1669  */
1670 static int
1671 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1672 {
1673         int cnt, optlen;
1674         u_char *cp;
1675         u_char opt;
1676
1677         /* turn off any old options */
1678         if (*pcbopt)
1679                 m_free(*pcbopt);
1680         *pcbopt = 0;
1681         if (m == NULL || m->m_len == 0) {
1682                 /*
1683                  * Only turning off any previous options.
1684                  */
1685                 if (m != NULL)
1686                         m_free(m);
1687                 return (0);
1688         }
1689
1690         if (m->m_len % sizeof(int32_t))
1691                 goto bad;
1692         /*
1693          * IP first-hop destination address will be stored before
1694          * actual options; move other options back
1695          * and clear it when none present.
1696          */
1697         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1698                 goto bad;
1699         cnt = m->m_len;
1700         m->m_len += sizeof(struct in_addr);
1701         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1702         ovbcopy(mtod(m, caddr_t), cp, cnt);
1703         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1704
1705         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1706                 opt = cp[IPOPT_OPTVAL];
1707                 if (opt == IPOPT_EOL)
1708                         break;
1709                 if (opt == IPOPT_NOP)
1710                         optlen = 1;
1711                 else {
1712                         if (cnt < IPOPT_OLEN + sizeof *cp)
1713                                 goto bad;
1714                         optlen = cp[IPOPT_OLEN];
1715                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1716                                 goto bad;
1717                 }
1718                 switch (opt) {
1719
1720                 default:
1721                         break;
1722
1723                 case IPOPT_LSRR:
1724                 case IPOPT_SSRR:
1725                         /*
1726                          * user process specifies route as:
1727                          *      ->A->B->C->D
1728                          * D must be our final destination (but we can't
1729                          * check that since we may not have connected yet).
1730                          * A is first hop destination, which doesn't appear in
1731                          * actual IP option, but is stored before the options.
1732                          */
1733                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1734                                 goto bad;
1735                         m->m_len -= sizeof(struct in_addr);
1736                         cnt -= sizeof(struct in_addr);
1737                         optlen -= sizeof(struct in_addr);
1738                         cp[IPOPT_OLEN] = optlen;
1739                         /*
1740                          * Move first hop before start of options.
1741                          */
1742                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1743                               sizeof(struct in_addr));
1744                         /*
1745                          * Then copy rest of options back
1746                          * to close up the deleted entry.
1747                          */
1748                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1749                                 &cp[IPOPT_OFFSET+1],
1750                                 cnt - (IPOPT_MINOFF - 1));
1751                         break;
1752                 }
1753         }
1754         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1755                 goto bad;
1756         *pcbopt = m;
1757         return (0);
1758
1759 bad:
1760         m_free(m);
1761         return (EINVAL);
1762 }
1763
1764 /*
1765  * XXX
1766  * The whole multicast option thing needs to be re-thought.
1767  * Several of these options are equally applicable to non-multicast
1768  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1769  * standard option (IP_TTL).
1770  */
1771
1772 /*
1773  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1774  */
1775 static struct ifnet *
1776 ip_multicast_if(struct in_addr *a, int *ifindexp)
1777 {
1778         int ifindex;
1779         struct ifnet *ifp;
1780
1781         if (ifindexp)
1782                 *ifindexp = 0;
1783         if (ntohl(a->s_addr) >> 24 == 0) {
1784                 ifindex = ntohl(a->s_addr) & 0xffffff;
1785                 if (ifindex < 0 || if_index < ifindex)
1786                         return NULL;
1787                 ifp = ifindex2ifnet[ifindex];
1788                 if (ifindexp)
1789                         *ifindexp = ifindex;
1790         } else {
1791                 ifp = INADDR_TO_IFP(a);
1792         }
1793         return ifp;
1794 }
1795
1796 /*
1797  * Set the IP multicast options in response to user setsockopt().
1798  */
1799 static int
1800 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1801 {
1802         int error = 0;
1803         int i;
1804         struct in_addr addr;
1805         struct ip_mreq mreq;
1806         struct ifnet *ifp;
1807         struct ip_moptions *imo = *imop;
1808         int ifindex;
1809
1810         if (imo == NULL) {
1811                 /*
1812                  * No multicast option buffer attached to the pcb;
1813                  * allocate one and initialize to default values.
1814                  */
1815                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1816
1817                 *imop = imo;
1818                 imo->imo_multicast_ifp = NULL;
1819                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1820                 imo->imo_multicast_vif = -1;
1821                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1822                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1823                 imo->imo_num_memberships = 0;
1824         }
1825         switch (sopt->sopt_name) {
1826         /* store an index number for the vif you wanna use in the send */
1827         case IP_MULTICAST_VIF:
1828                 if (legal_vif_num == 0) {
1829                         error = EOPNOTSUPP;
1830                         break;
1831                 }
1832                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1833                 if (error)
1834                         break;
1835                 if (!legal_vif_num(i) && (i != -1)) {
1836                         error = EINVAL;
1837                         break;
1838                 }
1839                 imo->imo_multicast_vif = i;
1840                 break;
1841
1842         case IP_MULTICAST_IF:
1843                 /*
1844                  * Select the interface for outgoing multicast packets.
1845                  */
1846                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1847                 if (error)
1848                         break;
1849
1850                 /*
1851                  * INADDR_ANY is used to remove a previous selection.
1852                  * When no interface is selected, a default one is
1853                  * chosen every time a multicast packet is sent.
1854                  */
1855                 if (addr.s_addr == INADDR_ANY) {
1856                         imo->imo_multicast_ifp = NULL;
1857                         break;
1858                 }
1859                 /*
1860                  * The selected interface is identified by its local
1861                  * IP address.  Find the interface and confirm that
1862                  * it supports multicasting.
1863                  */
1864                 crit_enter();
1865                 ifp = ip_multicast_if(&addr, &ifindex);
1866                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1867                         crit_exit();
1868                         error = EADDRNOTAVAIL;
1869                         break;
1870                 }
1871                 imo->imo_multicast_ifp = ifp;
1872                 if (ifindex)
1873                         imo->imo_multicast_addr = addr;
1874                 else
1875                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1876                 crit_exit();
1877                 break;
1878
1879         case IP_MULTICAST_TTL:
1880                 /*
1881                  * Set the IP time-to-live for outgoing multicast packets.
1882                  * The original multicast API required a char argument,
1883                  * which is inconsistent with the rest of the socket API.
1884                  * We allow either a char or an int.
1885                  */
1886                 if (sopt->sopt_valsize == 1) {
1887                         u_char ttl;
1888                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1889                         if (error)
1890                                 break;
1891                         imo->imo_multicast_ttl = ttl;
1892                 } else {
1893                         u_int ttl;
1894                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1895                         if (error)
1896                                 break;
1897                         if (ttl > 255)
1898                                 error = EINVAL;
1899                         else
1900                                 imo->imo_multicast_ttl = ttl;
1901                 }
1902                 break;
1903
1904         case IP_MULTICAST_LOOP:
1905                 /*
1906                  * Set the loopback flag for outgoing multicast packets.
1907                  * Must be zero or one.  The original multicast API required a
1908                  * char argument, which is inconsistent with the rest
1909                  * of the socket API.  We allow either a char or an int.
1910                  */
1911                 if (sopt->sopt_valsize == 1) {
1912                         u_char loop;
1913
1914                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1915                         if (error)
1916                                 break;
1917                         imo->imo_multicast_loop = !!loop;
1918                 } else {
1919                         u_int loop;
1920
1921                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1922                                             sizeof loop);
1923                         if (error)
1924                                 break;
1925                         imo->imo_multicast_loop = !!loop;
1926                 }
1927                 break;
1928
1929         case IP_ADD_MEMBERSHIP:
1930                 /*
1931                  * Add a multicast group membership.
1932                  * Group must be a valid IP multicast address.
1933                  */
1934                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1935                 if (error)
1936                         break;
1937
1938                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1939                         error = EINVAL;
1940                         break;
1941                 }
1942                 crit_enter();
1943                 /*
1944                  * If no interface address was provided, use the interface of
1945                  * the route to the given multicast address.
1946                  */
1947                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1948                         struct sockaddr_in dst;
1949                         struct rtentry *rt;
1950
1951                         bzero(&dst, sizeof(struct sockaddr_in));
1952                         dst.sin_len = sizeof(struct sockaddr_in);
1953                         dst.sin_family = AF_INET;
1954                         dst.sin_addr = mreq.imr_multiaddr;
1955                         rt = rtlookup((struct sockaddr *)&dst);
1956                         if (rt == NULL) {
1957                                 error = EADDRNOTAVAIL;
1958                                 crit_exit();
1959                                 break;
1960                         }
1961                         --rt->rt_refcnt;
1962                         ifp = rt->rt_ifp;
1963                 } else {
1964                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1965                 }
1966
1967                 /*
1968                  * See if we found an interface, and confirm that it
1969                  * supports multicast.
1970                  */
1971                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1972                         error = EADDRNOTAVAIL;
1973                         crit_exit();
1974                         break;
1975                 }
1976                 /*
1977                  * See if the membership already exists or if all the
1978                  * membership slots are full.
1979                  */
1980                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1981                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1982                             imo->imo_membership[i]->inm_addr.s_addr
1983                                                 == mreq.imr_multiaddr.s_addr)
1984                                 break;
1985                 }
1986                 if (i < imo->imo_num_memberships) {
1987                         error = EADDRINUSE;
1988                         crit_exit();
1989                         break;
1990                 }
1991                 if (i == IP_MAX_MEMBERSHIPS) {
1992                         error = ETOOMANYREFS;
1993                         crit_exit();
1994                         break;
1995                 }
1996                 /*
1997                  * Everything looks good; add a new record to the multicast
1998                  * address list for the given interface.
1999                  */
2000                 if ((imo->imo_membership[i] =
2001                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2002                         error = ENOBUFS;
2003                         crit_exit();
2004                         break;
2005                 }
2006                 ++imo->imo_num_memberships;
2007                 crit_exit();
2008                 break;
2009
2010         case IP_DROP_MEMBERSHIP:
2011                 /*
2012                  * Drop a multicast group membership.
2013                  * Group must be a valid IP multicast address.
2014                  */
2015                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2016                 if (error)
2017                         break;
2018
2019                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2020                         error = EINVAL;
2021                         break;
2022                 }
2023
2024                 crit_enter();
2025                 /*
2026                  * If an interface address was specified, get a pointer
2027                  * to its ifnet structure.
2028                  */
2029                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2030                         ifp = NULL;
2031                 else {
2032                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2033                         if (ifp == NULL) {
2034                                 error = EADDRNOTAVAIL;
2035                                 crit_exit();
2036                                 break;
2037                         }
2038                 }
2039                 /*
2040                  * Find the membership in the membership array.
2041                  */
2042                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2043                         if ((ifp == NULL ||
2044                              imo->imo_membership[i]->inm_ifp == ifp) &&
2045                             imo->imo_membership[i]->inm_addr.s_addr ==
2046                             mreq.imr_multiaddr.s_addr)
2047                                 break;
2048                 }
2049                 if (i == imo->imo_num_memberships) {
2050                         error = EADDRNOTAVAIL;
2051                         crit_exit();
2052                         break;
2053                 }
2054                 /*
2055                  * Give up the multicast address record to which the
2056                  * membership points.
2057                  */
2058                 in_delmulti(imo->imo_membership[i]);
2059                 /*
2060                  * Remove the gap in the membership array.
2061                  */
2062                 for (++i; i < imo->imo_num_memberships; ++i)
2063                         imo->imo_membership[i-1] = imo->imo_membership[i];
2064                 --imo->imo_num_memberships;
2065                 crit_exit();
2066                 break;
2067
2068         default:
2069                 error = EOPNOTSUPP;
2070                 break;
2071         }
2072
2073         /*
2074          * If all options have default values, no need to keep the mbuf.
2075          */
2076         if (imo->imo_multicast_ifp == NULL &&
2077             imo->imo_multicast_vif == -1 &&
2078             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2079             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2080             imo->imo_num_memberships == 0) {
2081                 kfree(*imop, M_IPMOPTS);
2082                 *imop = NULL;
2083         }
2084
2085         return (error);
2086 }
2087
2088 /*
2089  * Return the IP multicast options in response to user getsockopt().
2090  */
2091 static int
2092 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2093 {
2094         struct in_addr addr;
2095         struct in_ifaddr *ia;
2096         int error, optval;
2097         u_char coptval;
2098
2099         error = 0;
2100         switch (sopt->sopt_name) {
2101         case IP_MULTICAST_VIF:
2102                 if (imo != NULL)
2103                         optval = imo->imo_multicast_vif;
2104                 else
2105                         optval = -1;
2106                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2107                 break;
2108
2109         case IP_MULTICAST_IF:
2110                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2111                         addr.s_addr = INADDR_ANY;
2112                 else if (imo->imo_multicast_addr.s_addr) {
2113                         /* return the value user has set */
2114                         addr = imo->imo_multicast_addr;
2115                 } else {
2116                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2117                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2118                                 : IA_SIN(ia)->sin_addr.s_addr;
2119                 }
2120                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2121                 break;
2122
2123         case IP_MULTICAST_TTL:
2124                 if (imo == NULL)
2125                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2126                 else
2127                         optval = coptval = imo->imo_multicast_ttl;
2128                 if (sopt->sopt_valsize == 1)
2129                         soopt_from_kbuf(sopt, &coptval, 1);
2130                 else
2131                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2132                 break;
2133
2134         case IP_MULTICAST_LOOP:
2135                 if (imo == NULL)
2136                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2137                 else
2138                         optval = coptval = imo->imo_multicast_loop;
2139                 if (sopt->sopt_valsize == 1)
2140                         soopt_from_kbuf(sopt, &coptval, 1);
2141                 else
2142                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2143                 break;
2144
2145         default:
2146                 error = ENOPROTOOPT;
2147                 break;
2148         }
2149         return (error);
2150 }
2151
2152 /*
2153  * Discard the IP multicast options.
2154  */
2155 void
2156 ip_freemoptions(struct ip_moptions *imo)
2157 {
2158         int i;
2159
2160         if (imo != NULL) {
2161                 for (i = 0; i < imo->imo_num_memberships; ++i)
2162                         in_delmulti(imo->imo_membership[i]);
2163                 kfree(imo, M_IPMOPTS);
2164         }
2165 }
2166
2167 /*
2168  * Routine called from ip_output() to loop back a copy of an IP multicast
2169  * packet to the input queue of a specified interface.  Note that this
2170  * calls the output routine of the loopback "driver", but with an interface
2171  * pointer that might NOT be a loopback interface -- evil, but easier than
2172  * replicating that code here.
2173  */
2174 static void
2175 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2176              int hlen)
2177 {
2178         struct ip *ip;
2179         struct mbuf *copym;
2180
2181         copym = m_copypacket(m, MB_DONTWAIT);
2182         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2183                 copym = m_pullup(copym, hlen);
2184         if (copym != NULL) {
2185                 /*
2186                  * if the checksum hasn't been computed, mark it as valid
2187                  */
2188                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2189                         in_delayed_cksum(copym);
2190                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2191                         copym->m_pkthdr.csum_flags |=
2192                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2193                         copym->m_pkthdr.csum_data = 0xffff;
2194                 }
2195                 /*
2196                  * We don't bother to fragment if the IP length is greater
2197                  * than the interface's MTU.  Can this possibly matter?
2198                  */
2199                 ip = mtod(copym, struct ip *);
2200                 ip->ip_len = htons(ip->ip_len);
2201                 ip->ip_off = htons(ip->ip_off);
2202                 ip->ip_sum = 0;
2203                 if (ip->ip_vhl == IP_VHL_BORING) {
2204                         ip->ip_sum = in_cksum_hdr(ip);
2205                 } else {
2206                         ip->ip_sum = in_cksum(copym, hlen);
2207                 }
2208                 /*
2209                  * NB:
2210                  * It's not clear whether there are any lingering
2211                  * reentrancy problems in other areas which might
2212                  * be exposed by using ip_input directly (in
2213                  * particular, everything which modifies the packet
2214                  * in-place).  Yet another option is using the
2215                  * protosw directly to deliver the looped back
2216                  * packet.  For the moment, we'll err on the side
2217                  * of safety by using if_simloop().
2218                  */
2219 #if 1 /* XXX */
2220                 if (dst->sin_family != AF_INET) {
2221                         kprintf("ip_mloopback: bad address family %d\n",
2222                                                 dst->sin_family);
2223                         dst->sin_family = AF_INET;
2224                 }
2225 #endif
2226
2227 #ifdef notdef
2228                 copym->m_pkthdr.rcvif = ifp;
2229                 ip_input(copym);
2230 #else
2231                 if_simloop(ifp, copym, dst->sin_family, 0);
2232 #endif
2233         }
2234 }