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 1.58 2008/09/08 12:41:39 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
 122 extern  struct protosw inetsw[];
 123
 124 #ifdef IPDIVERT
 125 static struct mbuf *
 126 ip_divert_out(struct mbuf *m, int tee)
 127 {
 128         struct mbuf *clone = NULL;
 129         struct ip *ip = mtod(m, struct ip *);
 130
 131         /* Clone packet if we're doing a 'tee' */
 132         if (tee)
 133                 clone = m_dup(m, MB_DONTWAIT);
 134
 135         /*
 136          * XXX
 137          * delayed checksums are not currently compatible
 138          * with divert sockets.
 139          */
 140         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 141                 in_delayed_cksum(m);
 142                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 143         }
 144
 145         /* Restore packet header fields to original values */
 146         ip->ip_len = htons(ip->ip_len);
 147         ip->ip_off = htons(ip->ip_off);
 148
 149         /* Deliver packet to divert input routine */
 150         divert_packet(m, 0);
 151
 152         /* If 'tee', continue with original packet */
 153         return clone;
 154 }
 155 #endif  /* IPDIVERT */
 156
 157 static int
 158 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst)
 159 {
 160         struct in_ifaddr_container *iac;
 161
 162         /*
 163          * We need to figure out if we have been forwarded to a local
 164          * socket.  If so, then we should somehow "loop back" to
 165          * ip_input(), and get directed to the PCB as if we had received
 166          * this packet.  This is because it may be difficult to identify
 167          * the packets you want to forward until they are being output
 168          * and have selected an interface (e.g. locally initiated
 169          * packets).  If we used the loopback inteface, we would not be
 170          * able to control what happens as the packet runs through
 171          * ip_input() as it is done through a ISR.
 172          */
 173         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 174                 /*
 175                  * If the addr to forward to is one of ours, we pretend
 176                  * to be the destination for this packet.
 177                  */
 178                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 179                         break;
 180         }
 181         if (iac != NULL) {
 182                 struct ip *ip = mtod(m, struct ip *);
 183
 184                 if (m->m_pkthdr.rcvif == NULL)
 185                         m->m_pkthdr.rcvif = ifunit("lo0");
 186                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 187                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 188                                                   CSUM_PSEUDO_HDR;
 189                         m->m_pkthdr.csum_data = 0xffff;
 190                 }
 191                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 192
 193                 ip->ip_len = htons(ip->ip_len);
 194                 ip->ip_off = htons(ip->ip_off);
 195                 ip_input(m);
 196
 197                 return 1; /* Packet gets forwarded locally */
 198         }
 199         return 0;
 200 }
 201
 202 /*
 203  * IP output.  The packet in mbuf chain m contains a skeletal IP
 204  * header (with len, off, ttl, proto, tos, src, dst).
 205  * The mbuf chain containing the packet will be freed.
 206  * The mbuf opt, if present, will not be freed.
 207  */
 208 int
 209 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 210           int flags, struct ip_moptions *imo, struct inpcb *inp)
 211 {
 212         struct ip *ip;
 213         struct ifnet *ifp = NULL;       /* keep compiler happy */
 214         struct mbuf *m;
 215         int hlen = sizeof(struct ip);
 216         int len, off, error = 0;
 217         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 218         struct in_ifaddr *ia = NULL;
 219         int isbroadcast, sw_csum;
 220         struct in_addr pkt_dst;
 221         struct route iproute;
 222         struct m_tag *mtag;
 223 #ifdef IPSEC
 224         struct secpolicy *sp = NULL;
 225         struct socket *so = inp ? inp->inp_socket : NULL;
 226 #endif
 227 #ifdef FAST_IPSEC
 228         struct secpolicy *sp = NULL;
 229         struct tdb_ident *tdbi;
 230 #endif /* FAST_IPSEC */
 231         struct sockaddr_in *next_hop = NULL;
 232         int src_was_INADDR_ANY = 0;     /* as the name says... */
 233
 234         m = m0;
 235         M_ASSERTPKTHDR(m);
 236
 237         if (ro == NULL) {
 238                 ro = &iproute;
 239                 bzero(ro, sizeof *ro);
 240         }
 241
 242         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 243                 /* Next hop */
 244                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 245                 KKASSERT(mtag != NULL);
 246                 next_hop = m_tag_data(mtag);
 247         }
 248
 249         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 250                 struct dn_pkt *dn_pkt;
 251
 252                 /* Extract info from dummynet tag */
 253                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 254                 KKASSERT(mtag != NULL);
 255                 dn_pkt = m_tag_data(mtag);
 256
 257                 /*
 258                  * The packet was already tagged, so part of the
 259                  * processing was already done, and we need to go down.
 260                  * Get the calculated parameters from the tag.
 261                  */
 262                 ifp = dn_pkt->ifp;
 263
 264                 KKASSERT(ro == &iproute);
 265                 *ro = dn_pkt->ro; /* structure copy */
 266
 267                 dst = dn_pkt->dn_dst;
 268                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 269                         /* If 'dst' points into dummynet tag, adjust it */
 270                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 271                 }
 272
 273                 ip = mtod(m, struct ip *);
 274                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 275                 if (ro->ro_rt)
 276                         ia = ifatoia(ro->ro_rt->rt_ifa);
 277                 goto sendit;
 278         }
 279
 280         if (opt) {
 281                 len = 0;
 282                 m = ip_insertoptions(m, opt, &len);
 283                 if (len != 0)
 284                         hlen = len;
 285         }
 286         ip = mtod(m, struct ip *);
 287
 288         /*
 289          * Fill in IP header.
 290          */
 291         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 292                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 293                 ip->ip_off &= IP_DF;
 294                 ip->ip_id = ip_newid();
 295                 ipstat.ips_localout++;
 296         } else {
 297                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 298         }
 299
 300 reroute:
 301         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 302
 303         dst = (struct sockaddr_in *)&ro->ro_dst;
 304         /*
 305          * If there is a cached route,
 306          * check that it is to the same destination
 307          * and is still up.  If not, free it and try again.
 308          * The address family should also be checked in case of sharing the
 309          * cache with IPv6.
 310          */
 311         if (ro->ro_rt &&
 312             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 313              dst->sin_family != AF_INET ||
 314              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 315                 rtfree(ro->ro_rt);
 316                 ro->ro_rt = (struct rtentry *)NULL;
 317         }
 318         if (ro->ro_rt == NULL) {
 319                 bzero(dst, sizeof *dst);
 320                 dst->sin_family = AF_INET;
 321                 dst->sin_len = sizeof *dst;
 322                 dst->sin_addr = pkt_dst;
 323         }
 324         /*
 325          * If routing to interface only,
 326          * short circuit routing lookup.
 327          */
 328         if (flags & IP_ROUTETOIF) {
 329                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 330                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 331                         ipstat.ips_noroute++;
 332                         error = ENETUNREACH;
 333                         goto bad;
 334                 }
 335                 ifp = ia->ia_ifp;
 336                 ip->ip_ttl = 1;
 337                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 338         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
 339                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 340                 /*
 341                  * Bypass the normal routing lookup for multicast
 342                  * packets if the interface is specified.
 343                  */
 344                 ifp = imo->imo_multicast_ifp;
 345                 ia = IFP_TO_IA(ifp);
 346                 isbroadcast = 0;        /* fool gcc */
 347         } else {
 348                 /*
 349                  * If this is the case, we probably don't want to allocate
 350                  * a protocol-cloned route since we didn't get one from the
 351                  * ULP.  This lets TCP do its thing, while not burdening
 352                  * forwarding or ICMP with the overhead of cloning a route.
 353                  * Of course, we still want to do any cloning requested by
 354                  * the link layer, as this is probably required in all cases
 355                  * for correct operation (as it is for ARP).
 356                  */
 357                 if (ro->ro_rt == NULL)
 358                         rtalloc_ign(ro, RTF_PRCLONING);
 359                 if (ro->ro_rt == NULL) {
 360                         ipstat.ips_noroute++;
 361                         error = EHOSTUNREACH;
 362                         goto bad;
 363                 }
 364                 ia = ifatoia(ro->ro_rt->rt_ifa);
 365                 ifp = ro->ro_rt->rt_ifp;
 366                 ro->ro_rt->rt_use++;
 367                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 368                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 369                 if (ro->ro_rt->rt_flags & RTF_HOST)
 370                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 371                 else
 372                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 373         }
 374         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 375                 struct in_multi *inm;
 376
 377                 m->m_flags |= M_MCAST;
 378                 /*
 379                  * IP destination address is multicast.  Make sure "dst"
 380                  * still points to the address in "ro".  (It may have been
 381                  * changed to point to a gateway address, above.)
 382                  */
 383                 dst = (struct sockaddr_in *)&ro->ro_dst;
 384                 /*
 385                  * See if the caller provided any multicast options
 386                  */
 387                 if (imo != NULL) {
 388                         ip->ip_ttl = imo->imo_multicast_ttl;
 389                         if (imo->imo_multicast_vif != -1)
 390                                 ip->ip_src.s_addr =
 391                                     ip_mcast_src ?
 392                                     ip_mcast_src(imo->imo_multicast_vif) :
 393                                     INADDR_ANY;
 394                 } else
 395                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 396                 /*
 397                  * Confirm that the outgoing interface supports multicast.
 398                  */
 399                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 400                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 401                                 ipstat.ips_noroute++;
 402                                 error = ENETUNREACH;
 403                                 goto bad;
 404                         }
 405                 }
 406                 /*
 407                  * If source address not specified yet, use address
 408                  * of outgoing interface.
 409                  */
 410                 if (ip->ip_src.s_addr == INADDR_ANY) {
 411                         /* Interface may have no addresses. */
 412                         if (ia != NULL)
 413                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 414                 }
 415
 416                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 417                 if (inm != NULL &&
 418                    (imo == NULL || imo->imo_multicast_loop)) {
 419                         /*
 420                          * If we belong to the destination multicast group
 421                          * on the outgoing interface, and the caller did not
 422                          * forbid loopback, loop back a copy.
 423                          */
 424                         ip_mloopback(ifp, m, dst, hlen);
 425                 }
 426                 else {
 427                         /*
 428                          * If we are acting as a multicast router, perform
 429                          * multicast forwarding as if the packet had just
 430                          * arrived on the interface to which we are about
 431                          * to send.  The multicast forwarding function
 432                          * recursively calls this function, using the
 433                          * IP_FORWARDING flag to prevent infinite recursion.
 434                          *
 435                          * Multicasts that are looped back by ip_mloopback(),
 436                          * above, will be forwarded by the ip_input() routine,
 437                          * if necessary.
 438                          */
 439                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 440                                 /*
 441                                  * If rsvp daemon is not running, do not
 442                                  * set ip_moptions. This ensures that the packet
 443                                  * is multicast and not just sent down one link
 444                                  * as prescribed by rsvpd.
 445                                  */
 446                                 if (!rsvp_on)
 447                                         imo = NULL;
 448                                 if (ip_mforward &&
 449                                     ip_mforward(ip, ifp, m, imo) != 0) {
 450                                         m_freem(m);
 451                                         goto done;
 452                                 }
 453                         }
 454                 }
 455
 456                 /*
 457                  * Multicasts with a time-to-live of zero may be looped-
 458                  * back, above, but must not be transmitted on a network.
 459                  * Also, multicasts addressed to the loopback interface
 460                  * are not sent -- the above call to ip_mloopback() will
 461                  * loop back a copy if this host actually belongs to the
 462                  * destination group on the loopback interface.
 463                  */
 464                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 465                         m_freem(m);
 466                         goto done;
 467                 }
 468
 469                 goto sendit;
 470         } else {
 471                 m->m_flags &= ~M_MCAST;
 472         }
 473
 474         /*
 475          * If the source address is not specified yet, use the address
 476          * of the outoing interface. In case, keep note we did that, so
 477          * if the the firewall changes the next-hop causing the output
 478          * interface to change, we can fix that.
 479          */
 480         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 481                 /* Interface may have no addresses. */
 482                 if (ia != NULL) {
 483                         ip->ip_src = IA_SIN(ia)->sin_addr;
 484                         src_was_INADDR_ANY = 1;
 485                 }
 486         }
 487
 488 #ifdef ALTQ
 489         /*
 490          * Disable packet drop hack.
 491          * Packetdrop should be done by queueing.
 492          */
 493 #else /* !ALTQ */
 494         /*
 495          * Verify that we have any chance at all of being able to queue
 496          *      the packet or packet fragments
 497          */
 498         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 499                 ifp->if_snd.ifq_maxlen) {
 500                         error = ENOBUFS;
 501                         ipstat.ips_odropped++;
 502                         goto bad;
 503         }
 504 #endif /* !ALTQ */
 505
 506         /*
 507          * Look for broadcast address and
 508          * verify user is allowed to send
 509          * such a packet.
 510          */
 511         if (isbroadcast) {
 512                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 513                         error = EADDRNOTAVAIL;
 514                         goto bad;
 515                 }
 516                 if (!(flags & IP_ALLOWBROADCAST)) {
 517                         error = EACCES;
 518                         goto bad;
 519                 }
 520                 /* don't allow broadcast messages to be fragmented */
 521                 if (ip->ip_len > ifp->if_mtu) {
 522                         error = EMSGSIZE;
 523                         goto bad;
 524                 }
 525                 m->m_flags |= M_BCAST;
 526         } else {
 527                 m->m_flags &= ~M_BCAST;
 528         }
 529
 530 sendit:
 531 #ifdef IPSEC
 532         /* get SP for this packet */
 533         if (so == NULL)
 534                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 535         else
 536                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 537
 538         if (sp == NULL) {
 539                 ipsecstat.out_inval++;
 540                 goto bad;
 541         }
 542
 543         error = 0;
 544
 545         /* check policy */
 546         switch (sp->policy) {
 547         case IPSEC_POLICY_DISCARD:
 548                 /*
 549                  * This packet is just discarded.
 550                  */
 551                 ipsecstat.out_polvio++;
 552                 goto bad;
 553
 554         case IPSEC_POLICY_BYPASS:
 555         case IPSEC_POLICY_NONE:
 556                 /* no need to do IPsec. */
 557                 goto skip_ipsec;
 558
 559         case IPSEC_POLICY_IPSEC:
 560                 if (sp->req == NULL) {
 561                         /* acquire a policy */
 562                         error = key_spdacquire(sp);
 563                         goto bad;
 564                 }
 565                 break;
 566
 567         case IPSEC_POLICY_ENTRUST:
 568         default:
 569                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 570         }
 571     {
 572         struct ipsec_output_state state;
 573         bzero(&state, sizeof state);
 574         state.m = m;
 575         if (flags & IP_ROUTETOIF) {
 576                 state.ro = &iproute;
 577                 bzero(&iproute, sizeof iproute);
 578         } else
 579                 state.ro = ro;
 580         state.dst = (struct sockaddr *)dst;
 581
 582         ip->ip_sum = 0;
 583
 584         /*
 585          * XXX
 586          * delayed checksums are not currently compatible with IPsec
 587          */
 588         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 589                 in_delayed_cksum(m);
 590                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 591         }
 592
 593         ip->ip_len = htons(ip->ip_len);
 594         ip->ip_off = htons(ip->ip_off);
 595
 596         error = ipsec4_output(&state, sp, flags);
 597
 598         m = state.m;
 599         if (flags & IP_ROUTETOIF) {
 600                 /*
 601                  * if we have tunnel mode SA, we may need to ignore
 602                  * IP_ROUTETOIF.
 603                  */
 604                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 605                         flags &= ~IP_ROUTETOIF;
 606                         ro = state.ro;
 607                 }
 608         } else
 609                 ro = state.ro;
 610         dst = (struct sockaddr_in *)state.dst;
 611         if (error) {
 612                 /* mbuf is already reclaimed in ipsec4_output. */
 613                 m0 = NULL;
 614                 switch (error) {
 615                 case EHOSTUNREACH:
 616                 case ENETUNREACH:
 617                 case EMSGSIZE:
 618                 case ENOBUFS:
 619                 case ENOMEM:
 620                         break;
 621                 default:
 622                         kprintf("ip4_output (ipsec): error code %d\n", error);
 623                         /*fall through*/
 624                 case ENOENT:
 625                         /* don't show these error codes to the user */
 626                         error = 0;
 627                         break;
 628                 }
 629                 goto bad;
 630         }
 631     }
 632
 633         /* be sure to update variables that are affected by ipsec4_output() */
 634         ip = mtod(m, struct ip *);
 635 #ifdef _IP_VHL
 636         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 637 #else
 638         hlen = ip->ip_hl << 2;
 639 #endif
 640         if (ro->ro_rt == NULL) {
 641                 if (!(flags & IP_ROUTETOIF)) {
 642                         kprintf("ip_output: "
 643                                 "can't update route after IPsec processing\n");
 644                         error = EHOSTUNREACH;   /*XXX*/
 645                         goto bad;
 646                 }
 647         } else {
 648                 ia = ifatoia(ro->ro_rt->rt_ifa);
 649                 ifp = ro->ro_rt->rt_ifp;
 650         }
 651
 652         /* make it flipped, again. */
 653         ip->ip_len = ntohs(ip->ip_len);
 654         ip->ip_off = ntohs(ip->ip_off);
 655 skip_ipsec:
 656 #endif /*IPSEC*/
 657 #ifdef FAST_IPSEC
 658         /*
 659          * Check the security policy (SP) for the packet and, if
 660          * required, do IPsec-related processing.  There are two
 661          * cases here; the first time a packet is sent through
 662          * it will be untagged and handled by ipsec4_checkpolicy.
 663          * If the packet is resubmitted to ip_output (e.g. after
 664          * AH, ESP, etc. processing), there will be a tag to bypass
 665          * the lookup and related policy checking.
 666          */
 667         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 668         crit_enter();
 669         if (mtag != NULL) {
 670                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 671                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 672                 if (sp == NULL)
 673                         error = -EINVAL;        /* force silent drop */
 674                 m_tag_delete(m, mtag);
 675         } else {
 676                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 677                                         &error, inp);
 678         }
 679         /*
 680          * There are four return cases:
 681          *    sp != NULL                    apply IPsec policy
 682          *    sp == NULL, error == 0        no IPsec handling needed
 683          *    sp == NULL, error == -EINVAL  discard packet w/o error
 684          *    sp == NULL, error != 0        discard packet, report error
 685          */
 686         if (sp != NULL) {
 687                 /* Loop detection, check if ipsec processing already done */
 688                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 689                 for (mtag = m_tag_first(m); mtag != NULL;
 690                      mtag = m_tag_next(m, mtag)) {
 691                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 692                                 continue;
 693                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 694                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 695                                 continue;
 696                         /*
 697                          * Check if policy has an SA associated with it.
 698                          * This can happen when an SP has yet to acquire
 699                          * an SA; e.g. on first reference.  If it occurs,
 700                          * then we let ipsec4_process_packet do its thing.
 701                          */
 702                         if (sp->req->sav == NULL)
 703                                 break;
 704                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 705                         if (tdbi->spi == sp->req->sav->spi &&
 706                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 707                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 708                                  sizeof(union sockaddr_union)) == 0) {
 709                                 /*
 710                                  * No IPsec processing is needed, free
 711                                  * reference to SP.
 712                                  *
 713                                  * NB: null pointer to avoid free at
 714                                  *     done: below.
 715                                  */
 716                                 KEY_FREESP(&sp), sp = NULL;
 717                                 crit_exit();
 718                                 goto spd_done;
 719                         }
 720                 }
 721
 722                 /*
 723                  * Do delayed checksums now because we send before
 724                  * this is done in the normal processing path.
 725                  */
 726                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 727                         in_delayed_cksum(m);
 728                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 729                 }
 730
 731                 ip->ip_len = htons(ip->ip_len);
 732                 ip->ip_off = htons(ip->ip_off);
 733
 734                 /* NB: callee frees mbuf */
 735                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 736                 /*
 737                  * Preserve KAME behaviour: ENOENT can be returned
 738                  * when an SA acquire is in progress.  Don't propagate
 739                  * this to user-level; it confuses applications.
 740                  *
 741                  * XXX this will go away when the SADB is redone.
 742                  */
 743                 if (error == ENOENT)
 744                         error = 0;
 745                 crit_exit();
 746                 goto done;
 747         } else {
 748                 crit_exit();
 749
 750                 if (error != 0) {
 751                         /*
 752                          * Hack: -EINVAL is used to signal that a packet
 753                          * should be silently discarded.  This is typically
 754                          * because we asked key management for an SA and
 755                          * it was delayed (e.g. kicked up to IKE).
 756                          */
 757                         if (error == -EINVAL)
 758                                 error = 0;
 759                         goto bad;
 760                 } else {
 761                         /* No IPsec processing for this packet. */
 762                 }
 763 #ifdef notyet
 764                 /*
 765                  * If deferred crypto processing is needed, check that
 766                  * the interface supports it.
 767                  */
 768                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 769                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 770                         /* notify IPsec to do its own crypto */
 771                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 772                         error = EHOSTUNREACH;
 773                         goto bad;
 774                 }
 775 #endif
 776         }
 777 spd_done:
 778 #endif /* FAST_IPSEC */
 779
 780         /* We are already being fwd'd from a firewall. */
 781         if (next_hop != NULL)
 782                 goto pass;
 783
 784         /*
 785          * IpHack's section.
 786          * - Xlate: translate packet's addr/port (NAT).
 787          * - Firewall: deny/allow/etc.
 788          * - Wrap: fake packet's addr/port <unimpl.>
 789          * - Encapsulate: put it in another IP and send out. <unimp.>
 790          */
 791
 792         /*
 793          * Run through list of hooks for output packets.
 794          */
 795         if (pfil_has_hooks(&inet_pfil_hook)) {
 796                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 797                 if (error != 0 || m == NULL)
 798                         goto done;
 799                 ip = mtod(m, struct ip *);
 800         }
 801
 802         /*
 803          * Check with the firewall...
 804          */
 805         if (fw_enable && IPFW_LOADED) {
 806                 struct ip_fw_args args;
 807                 int tee = 0;
 808
 809                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 810                         /* Extract info from dummynet tag */
 811                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 812                         KKASSERT(mtag != NULL);
 813                         args.rule =
 814                         ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
 815                         KKASSERT(args.rule != NULL);
 816
 817                         m_tag_delete(m, mtag);
 818                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 819                 } else {
 820                         args.rule = NULL;
 821                 }
 822
 823                 args.eh = NULL;
 824                 args.m = m;
 825                 args.oif = ifp;
 826                 off = ip_fw_chk_ptr(&args);
 827                 m = args.m;
 828
 829                 if (m == NULL) {
 830                         error = EACCES;
 831                         goto done;
 832                 }
 833                 ip = mtod(m, struct ip *);
 834
 835                 switch (off) {
 836                 case IP_FW_PASS:
 837                         break;
 838
 839                 case IP_FW_DENY:
 840                         m_freem(m);
 841                         error = EACCES;
 842                         goto done;
 843
 844                 case IP_FW_DUMMYNET:
 845                         /*
 846                          * pass the pkt to dummynet. Need to include
 847                          * pipe number, m, ifp, ro, dst because these are
 848                          * not recomputed in the next pass.
 849                          * All other parameters have been already used and
 850                          * so they are not needed anymore.
 851                          * XXX note: if the ifp or ro entry are deleted
 852                          * while a pkt is in dummynet, we are in trouble!
 853                          */
 854                         args.ro = ro;
 855                         args.dst = dst;
 856                         args.flags = flags;
 857
 858                         error = 0;
 859                         ip_fw_dn_io_ptr(m, args.cookie, DN_TO_IP_OUT, &args);
 860                         goto done;
 861
 862                 case IP_FW_TEE:
 863                         tee = 1;
 864                         /* FALL THROUGH */
 865
 866                 case IP_FW_DIVERT:
 867 #ifdef IPDIVERT
 868                         m = ip_divert_out(m, tee);
 869                         if (m == NULL)
 870                                 goto done;
 871                         ip = mtod(m, struct ip *);
 872                         break;
 873 #else
 874                         m_freem(m);
 875                         /* not sure this is the right error msg */
 876                         error = EACCES;
 877                         goto done;
 878 #endif
 879
 880                 default:
 881                         panic("unknown ipfw return value: %d\n", off);
 882                 }
 883         }
 884
 885         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 886                 /*
 887                  * Check dst to make sure it is directly reachable on the
 888                  * interface we previously thought it was.
 889                  * If it isn't (which may be likely in some situations) we have
 890                  * to re-route it (ie, find a route for the next-hop and the
 891                  * associated interface) and set them here. This is nested
 892                  * forwarding which in most cases is undesirable, except where
 893                  * such control is nigh impossible. So we do it here.
 894                  * And I'm babbling.
 895                  */
 896                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 897                 KKASSERT(mtag != NULL);
 898                 next_hop = m_tag_data(mtag);
 899
 900                 /*
 901                  * Try local forwarding first
 902                  */
 903                 if (ip_localforward(m, next_hop))
 904                         goto done;
 905
 906                 /*
 907                  * Relocate the route based on next_hop.
 908                  * If the current route is inp's cache, keep it untouched.
 909                  */
 910                 if (ro == &iproute && ro->ro_rt != NULL) {
 911                         RTFREE(ro->ro_rt);
 912                         ro->ro_rt = NULL;
 913                 }
 914                 ro = &iproute;
 915                 bzero(ro, sizeof *ro);
 916
 917                 /*
 918                  * Forwarding to broadcast address is not allowed.
 919                  * XXX Should we follow IP_ROUTETOIF?
 920                  */
 921                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 922
 923                 /* We are doing forwarding now */
 924                 flags |= IP_FORWARDING;
 925
 926                 goto reroute;
 927         }
 928
 929 pass:
 930         /* 127/8 must not appear on wire - RFC1122. */
 931         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 932             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 933                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 934                         ipstat.ips_badaddr++;
 935                         error = EADDRNOTAVAIL;
 936                         goto bad;
 937                 }
 938         }
 939
 940         m->m_pkthdr.csum_flags |= CSUM_IP;
 941         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 942         if (sw_csum & CSUM_DELAY_DATA) {
 943                 in_delayed_cksum(m);
 944                 sw_csum &= ~CSUM_DELAY_DATA;
 945         }
 946         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 947
 948         /*
 949          * If small enough for interface, or the interface will take
 950          * care of the fragmentation for us, can just send directly.
 951          */
 952         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 953             !(ip->ip_off & IP_DF))) {
 954                 ip->ip_len = htons(ip->ip_len);
 955                 ip->ip_off = htons(ip->ip_off);
 956                 ip->ip_sum = 0;
 957                 if (sw_csum & CSUM_DELAY_IP) {
 958                         if (ip->ip_vhl == IP_VHL_BORING) {
 959                                 ip->ip_sum = in_cksum_hdr(ip);
 960                         } else {
 961                                 ip->ip_sum = in_cksum(m, hlen);
 962                         }
 963                 }
 964
 965                 /* Record statistics for this interface address. */
 966                 if (!(flags & IP_FORWARDING) && ia) {
 967                         ia->ia_ifa.if_opackets++;
 968                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 969                 }
 970
 971 #ifdef IPSEC
 972                 /* clean ipsec history once it goes out of the node */
 973                 ipsec_delaux(m);
 974 #endif
 975
 976 #ifdef MBUF_STRESS_TEST
 977                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 978                         struct mbuf *m1, *m2;
 979                         int length, tmp;
 980
 981                         tmp = length = m->m_pkthdr.len;
 982
 983                         while ((length -= mbuf_frag_size) >= 1) {
 984                                 m1 = m_split(m, length, MB_DONTWAIT);
 985                                 if (m1 == NULL)
 986                                         break;
 987                                 m2 = m;
 988                                 while (m2->m_next != NULL)
 989                                         m2 = m2->m_next;
 990                                 m2->m_next = m1;
 991                         }
 992                         m->m_pkthdr.len = tmp;
 993                 }
 994 #endif
 995
 996 #ifdef MPLS
 997                 if (!mpls_output_process(m, ro->ro_rt))
 998                         goto done;
 999 #endif
1000                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1001                                        ro->ro_rt);
1002                 goto done;
1003         }
1004
1005         if (ip->ip_off & IP_DF) {
1006                 error = EMSGSIZE;
1007                 /*
1008                  * This case can happen if the user changed the MTU
1009                  * of an interface after enabling IP on it.  Because
1010                  * most netifs don't keep track of routes pointing to
1011                  * them, there is no way for one to update all its
1012                  * routes when the MTU is changed.
1013                  */
1014                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1015                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1016                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1017                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1018                 }
1019                 ipstat.ips_cantfrag++;
1020                 goto bad;
1021         }
1022
1023         /*
1024          * Too large for interface; fragment if possible. If successful,
1025          * on return, m will point to a list of packets to be sent.
1026          */
1027         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1028         if (error)
1029                 goto bad;
1030         for (; m; m = m0) {
1031                 m0 = m->m_nextpkt;
1032                 m->m_nextpkt = NULL;
1033 #ifdef IPSEC
1034                 /* clean ipsec history once it goes out of the node */
1035                 ipsec_delaux(m);
1036 #endif
1037                 if (error == 0) {
1038                         /* Record statistics for this interface address. */
1039                         if (ia != NULL) {
1040                                 ia->ia_ifa.if_opackets++;
1041                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1042                         }
1043 #ifdef MPLS
1044                         if (!mpls_output_process(m, ro->ro_rt))
1045                                 continue;
1046 #endif
1047                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1048                                                ro->ro_rt);
1049                 } else {
1050                         m_freem(m);
1051                 }
1052         }
1053
1054         if (error == 0)
1055                 ipstat.ips_fragmented++;
1056
1057 done:
1058         if (ro == &iproute && ro->ro_rt != NULL) {
1059                 RTFREE(ro->ro_rt);
1060                 ro->ro_rt = NULL;
1061         }
1062 #ifdef IPSEC
1063         if (sp != NULL) {
1064                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1065                         kprintf("DP ip_output call free SP:%p\n", sp));
1066                 key_freesp(sp);
1067         }
1068 #endif
1069 #ifdef FAST_IPSEC
1070         if (sp != NULL)
1071                 KEY_FREESP(&sp);
1072 #endif
1073         return (error);
1074 bad:
1075         m_freem(m);
1076         goto done;
1077 }
1078
1079 /*
1080  * Create a chain of fragments which fit the given mtu. m_frag points to the
1081  * mbuf to be fragmented; on return it points to the chain with the fragments.
1082  * Return 0 if no error. If error, m_frag may contain a partially built
1083  * chain of fragments that should be freed by the caller.
1084  *
1085  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1086  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1087  */
1088 int
1089 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1090             u_long if_hwassist_flags, int sw_csum)
1091 {
1092         int error = 0;
1093         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1094         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1095         int off;
1096         struct mbuf *m0 = *m_frag;      /* the original packet          */
1097         int firstlen;
1098         struct mbuf **mnext;
1099         int nfrags;
1100
1101         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1102                 ipstat.ips_cantfrag++;
1103                 return EMSGSIZE;
1104         }
1105
1106         /*
1107          * Must be able to put at least 8 bytes per fragment.
1108          */
1109         if (len < 8)
1110                 return EMSGSIZE;
1111
1112         /*
1113          * If the interface will not calculate checksums on
1114          * fragmented packets, then do it here.
1115          */
1116         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1117             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1118                 in_delayed_cksum(m0);
1119                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1120         }
1121
1122         if (len > PAGE_SIZE) {
1123                 /*
1124                  * Fragment large datagrams such that each segment
1125                  * contains a multiple of PAGE_SIZE amount of data,
1126                  * plus headers. This enables a receiver to perform
1127                  * page-flipping zero-copy optimizations.
1128                  *
1129                  * XXX When does this help given that sender and receiver
1130                  * could have different page sizes, and also mtu could
1131                  * be less than the receiver's page size ?
1132                  */
1133                 int newlen;
1134                 struct mbuf *m;
1135
1136                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1137                         off += m->m_len;
1138
1139                 /*
1140                  * firstlen (off - hlen) must be aligned on an
1141                  * 8-byte boundary
1142                  */
1143                 if (off < hlen)
1144                         goto smart_frag_failure;
1145                 off = ((off - hlen) & ~7) + hlen;
1146                 newlen = (~PAGE_MASK) & mtu;
1147                 if ((newlen + sizeof(struct ip)) > mtu) {
1148                         /* we failed, go back the default */
1149 smart_frag_failure:
1150                         newlen = len;
1151                         off = hlen + len;
1152                 }
1153                 len = newlen;
1154
1155         } else {
1156                 off = hlen + len;
1157         }
1158
1159         firstlen = off - hlen;
1160         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1161
1162         /*
1163          * Loop through length of segment after first fragment,
1164          * make new header and copy data of each part and link onto chain.
1165          * Here, m0 is the original packet, m is the fragment being created.
1166          * The fragments are linked off the m_nextpkt of the original
1167          * packet, which after processing serves as the first fragment.
1168          */
1169         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1170                 struct ip *mhip;        /* ip header on the fragment */
1171                 struct mbuf *m;
1172                 int mhlen = sizeof(struct ip);
1173
1174                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1175                 if (m == NULL) {
1176                         error = ENOBUFS;
1177                         ipstat.ips_odropped++;
1178                         goto done;
1179                 }
1180                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1181                 /*
1182                  * In the first mbuf, leave room for the link header, then
1183                  * copy the original IP header including options. The payload
1184                  * goes into an additional mbuf chain returned by m_copy().
1185                  */
1186                 m->m_data += max_linkhdr;
1187                 mhip = mtod(m, struct ip *);
1188                 *mhip = *ip;
1189                 if (hlen > sizeof(struct ip)) {
1190                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1191                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1192                 }
1193                 m->m_len = mhlen;
1194                 /* XXX do we need to add ip->ip_off below ? */
1195                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1196                 if (off + len >= ip->ip_len) {  /* last fragment */
1197                         len = ip->ip_len - off;
1198                         m->m_flags |= M_LASTFRAG;
1199                 } else
1200                         mhip->ip_off |= IP_MF;
1201                 mhip->ip_len = htons((u_short)(len + mhlen));
1202                 m->m_next = m_copy(m0, off, len);
1203                 if (m->m_next == NULL) {                /* copy failed */
1204                         m_free(m);
1205                         error = ENOBUFS;        /* ??? */
1206                         ipstat.ips_odropped++;
1207                         goto done;
1208                 }
1209                 m->m_pkthdr.len = mhlen + len;
1210                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1211                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1212                 mhip->ip_off = htons(mhip->ip_off);
1213                 mhip->ip_sum = 0;
1214                 if (sw_csum & CSUM_DELAY_IP)
1215                         mhip->ip_sum = in_cksum(m, mhlen);
1216                 *mnext = m;
1217                 mnext = &m->m_nextpkt;
1218         }
1219         ipstat.ips_ofragments += nfrags;
1220
1221         /* set first marker for fragment chain */
1222         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1223         m0->m_pkthdr.csum_data = nfrags;
1224
1225         /*
1226          * Update first fragment by trimming what's been copied out
1227          * and updating header.
1228          */
1229         m_adj(m0, hlen + firstlen - ip->ip_len);
1230         m0->m_pkthdr.len = hlen + firstlen;
1231         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1232         ip->ip_off |= IP_MF;
1233         ip->ip_off = htons(ip->ip_off);
1234         ip->ip_sum = 0;
1235         if (sw_csum & CSUM_DELAY_IP)
1236                 ip->ip_sum = in_cksum(m0, hlen);
1237
1238 done:
1239         *m_frag = m0;
1240         return error;
1241 }
1242
1243 void
1244 in_delayed_cksum(struct mbuf *m)
1245 {
1246         struct ip *ip;
1247         u_short csum, offset;
1248
1249         ip = mtod(m, struct ip *);
1250         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1251         csum = in_cksum_skip(m, ip->ip_len, offset);
1252         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1253                 csum = 0xffff;
1254         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1255
1256         if (offset + sizeof(u_short) > m->m_len) {
1257                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1258                     m->m_len, offset, ip->ip_p);
1259                 /*
1260                  * XXX
1261                  * this shouldn't happen, but if it does, the
1262                  * correct behavior may be to insert the checksum
1263                  * in the existing chain instead of rearranging it.
1264                  */
1265                 m = m_pullup(m, offset + sizeof(u_short));
1266         }
1267         *(u_short *)(m->m_data + offset) = csum;
1268 }
1269
1270 /*
1271  * Insert IP options into preformed packet.
1272  * Adjust IP destination as required for IP source routing,
1273  * as indicated by a non-zero in_addr at the start of the options.
1274  *
1275  * XXX This routine assumes that the packet has no options in place.
1276  */
1277 static struct mbuf *
1278 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1279 {
1280         struct ipoption *p = mtod(opt, struct ipoption *);
1281         struct mbuf *n;
1282         struct ip *ip = mtod(m, struct ip *);
1283         unsigned optlen;
1284
1285         optlen = opt->m_len - sizeof p->ipopt_dst;
1286         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1287                 *phlen = 0;
1288                 return (m);             /* XXX should fail */
1289         }
1290         if (p->ipopt_dst.s_addr)
1291                 ip->ip_dst = p->ipopt_dst;
1292         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1293                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1294                 if (n == NULL) {
1295                         *phlen = 0;
1296                         return (m);
1297                 }
1298                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1299                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1300                 m->m_len -= sizeof(struct ip);
1301                 m->m_data += sizeof(struct ip);
1302                 n->m_next = m;
1303                 m = n;
1304                 m->m_len = optlen + sizeof(struct ip);
1305                 m->m_data += max_linkhdr;
1306                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1307         } else {
1308                 m->m_data -= optlen;
1309                 m->m_len += optlen;
1310                 m->m_pkthdr.len += optlen;
1311                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1312         }
1313         ip = mtod(m, struct ip *);
1314         bcopy(p->ipopt_list, ip + 1, optlen);
1315         *phlen = sizeof(struct ip) + optlen;
1316         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1317         ip->ip_len += optlen;
1318         return (m);
1319 }
1320
1321 /*
1322  * Copy options from ip to jp,
1323  * omitting those not copied during fragmentation.
1324  */
1325 int
1326 ip_optcopy(struct ip *ip, struct ip *jp)
1327 {
1328         u_char *cp, *dp;
1329         int opt, optlen, cnt;
1330
1331         cp = (u_char *)(ip + 1);
1332         dp = (u_char *)(jp + 1);
1333         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1334         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1335                 opt = cp[0];
1336                 if (opt == IPOPT_EOL)
1337                         break;
1338                 if (opt == IPOPT_NOP) {
1339                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1340                         *dp++ = IPOPT_NOP;
1341                         optlen = 1;
1342                         continue;
1343                 }
1344
1345                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1346                     ("ip_optcopy: malformed ipv4 option"));
1347                 optlen = cp[IPOPT_OLEN];
1348                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1349                     ("ip_optcopy: malformed ipv4 option"));
1350
1351                 /* bogus lengths should have been caught by ip_dooptions */
1352                 if (optlen > cnt)
1353                         optlen = cnt;
1354                 if (IPOPT_COPIED(opt)) {
1355                         bcopy(cp, dp, optlen);
1356                         dp += optlen;
1357                 }
1358         }
1359         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1360                 *dp++ = IPOPT_EOL;
1361         return (optlen);
1362 }
1363
1364 /*
1365  * IP socket option processing.
1366  */
1367 int
1368 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1369 {
1370         struct  inpcb *inp = so->so_pcb;
1371         int     error, optval;
1372
1373         error = optval = 0;
1374         if (sopt->sopt_level != IPPROTO_IP) {
1375                 return (EINVAL);
1376         }
1377
1378         switch (sopt->sopt_dir) {
1379         case SOPT_SET:
1380                 switch (sopt->sopt_name) {
1381                 case IP_OPTIONS:
1382 #ifdef notyet
1383                 case IP_RETOPTS:
1384 #endif
1385                 {
1386                         struct mbuf *m;
1387                         if (sopt->sopt_valsize > MLEN) {
1388                                 error = EMSGSIZE;
1389                                 break;
1390                         }
1391                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1392                         if (m == NULL) {
1393                                 error = ENOBUFS;
1394                                 break;
1395                         }
1396                         m->m_len = sopt->sopt_valsize;
1397                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1398                                               m->m_len);
1399                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1400                                            m));
1401                 }
1402
1403                 case IP_TOS:
1404                 case IP_TTL:
1405                 case IP_MINTTL:
1406                 case IP_RECVOPTS:
1407                 case IP_RECVRETOPTS:
1408                 case IP_RECVDSTADDR:
1409                 case IP_RECVIF:
1410                 case IP_RECVTTL:
1411                 case IP_FAITH:
1412                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1413                                              sizeof optval);
1414                         if (error)
1415                                 break;
1416                         switch (sopt->sopt_name) {
1417                         case IP_TOS:
1418                                 inp->inp_ip_tos = optval;
1419                                 break;
1420
1421                         case IP_TTL:
1422                                 inp->inp_ip_ttl = optval;
1423                                 break;
1424                         case IP_MINTTL:
1425                                 if (optval > 0 && optval <= MAXTTL)
1426                                         inp->inp_ip_minttl = optval;
1427                                 else
1428                                         error = EINVAL;
1429                                 break;
1430 #define OPTSET(bit) \
1431         if (optval) \
1432                 inp->inp_flags |= bit; \
1433         else \
1434                 inp->inp_flags &= ~bit;
1435
1436                         case IP_RECVOPTS:
1437                                 OPTSET(INP_RECVOPTS);
1438                                 break;
1439
1440                         case IP_RECVRETOPTS:
1441                                 OPTSET(INP_RECVRETOPTS);
1442                                 break;
1443
1444                         case IP_RECVDSTADDR:
1445                                 OPTSET(INP_RECVDSTADDR);
1446                                 break;
1447
1448                         case IP_RECVIF:
1449                                 OPTSET(INP_RECVIF);
1450                                 break;
1451
1452                         case IP_RECVTTL:
1453                                 OPTSET(INP_RECVTTL);
1454                                 break;
1455
1456                         case IP_FAITH:
1457                                 OPTSET(INP_FAITH);
1458                                 break;
1459                         }
1460                         break;
1461 #undef OPTSET
1462
1463                 case IP_MULTICAST_IF:
1464                 case IP_MULTICAST_VIF:
1465                 case IP_MULTICAST_TTL:
1466                 case IP_MULTICAST_LOOP:
1467                 case IP_ADD_MEMBERSHIP:
1468                 case IP_DROP_MEMBERSHIP:
1469                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1470                         break;
1471
1472                 case IP_PORTRANGE:
1473                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1474                                             sizeof optval);
1475                         if (error)
1476                                 break;
1477
1478                         switch (optval) {
1479                         case IP_PORTRANGE_DEFAULT:
1480                                 inp->inp_flags &= ~(INP_LOWPORT);
1481                                 inp->inp_flags &= ~(INP_HIGHPORT);
1482                                 break;
1483
1484                         case IP_PORTRANGE_HIGH:
1485                                 inp->inp_flags &= ~(INP_LOWPORT);
1486                                 inp->inp_flags |= INP_HIGHPORT;
1487                                 break;
1488
1489                         case IP_PORTRANGE_LOW:
1490                                 inp->inp_flags &= ~(INP_HIGHPORT);
1491                                 inp->inp_flags |= INP_LOWPORT;
1492                                 break;
1493
1494                         default:
1495                                 error = EINVAL;
1496                                 break;
1497                         }
1498                         break;
1499
1500 #if defined(IPSEC) || defined(FAST_IPSEC)
1501                 case IP_IPSEC_POLICY:
1502                 {
1503                         caddr_t req;
1504                         size_t len = 0;
1505                         int priv;
1506                         struct mbuf *m;
1507                         int optname;
1508
1509                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1510                                 break;
1511                         soopt_to_mbuf(sopt, m);
1512                         priv = (sopt->sopt_td != NULL &&
1513                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1514                         req = mtod(m, caddr_t);
1515                         len = m->m_len;
1516                         optname = sopt->sopt_name;
1517                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1518                         m_freem(m);
1519                         break;
1520                 }
1521 #endif /*IPSEC*/
1522
1523                 default:
1524                         error = ENOPROTOOPT;
1525                         break;
1526                 }
1527                 break;
1528
1529         case SOPT_GET:
1530                 switch (sopt->sopt_name) {
1531                 case IP_OPTIONS:
1532                 case IP_RETOPTS:
1533                         if (inp->inp_options)
1534                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1535                                                            char *),
1536                                                 inp->inp_options->m_len);
1537                         else
1538                                 sopt->sopt_valsize = 0;
1539                         break;
1540
1541                 case IP_TOS:
1542                 case IP_TTL:
1543                 case IP_MINTTL:
1544                 case IP_RECVOPTS:
1545                 case IP_RECVRETOPTS:
1546                 case IP_RECVDSTADDR:
1547                 case IP_RECVTTL:
1548                 case IP_RECVIF:
1549                 case IP_PORTRANGE:
1550                 case IP_FAITH:
1551                         switch (sopt->sopt_name) {
1552
1553                         case IP_TOS:
1554                                 optval = inp->inp_ip_tos;
1555                                 break;
1556
1557                         case IP_TTL:
1558                                 optval = inp->inp_ip_ttl;
1559                                 break;
1560                         case IP_MINTTL:
1561                                 optval = inp->inp_ip_minttl;
1562                                 break;
1563
1564 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1565
1566                         case IP_RECVOPTS:
1567                                 optval = OPTBIT(INP_RECVOPTS);
1568                                 break;
1569
1570                         case IP_RECVRETOPTS:
1571                                 optval = OPTBIT(INP_RECVRETOPTS);
1572                                 break;
1573
1574                         case IP_RECVDSTADDR:
1575                                 optval = OPTBIT(INP_RECVDSTADDR);
1576                                 break;
1577
1578                         case IP_RECVTTL:
1579                                 optval = OPTBIT(INP_RECVTTL);
1580                                 break;
1581
1582                         case IP_RECVIF:
1583                                 optval = OPTBIT(INP_RECVIF);
1584                                 break;
1585
1586                         case IP_PORTRANGE:
1587                                 if (inp->inp_flags & INP_HIGHPORT)
1588                                         optval = IP_PORTRANGE_HIGH;
1589                                 else if (inp->inp_flags & INP_LOWPORT)
1590                                         optval = IP_PORTRANGE_LOW;
1591                                 else
1592                                         optval = 0;
1593                                 break;
1594
1595                         case IP_FAITH:
1596                                 optval = OPTBIT(INP_FAITH);
1597                                 break;
1598                         }
1599                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1600                         break;
1601
1602                 case IP_MULTICAST_IF:
1603                 case IP_MULTICAST_VIF:
1604                 case IP_MULTICAST_TTL:
1605                 case IP_MULTICAST_LOOP:
1606                 case IP_ADD_MEMBERSHIP:
1607                 case IP_DROP_MEMBERSHIP:
1608                         error = ip_getmoptions(sopt, inp->inp_moptions);
1609                         break;
1610
1611 #if defined(IPSEC) || defined(FAST_IPSEC)
1612                 case IP_IPSEC_POLICY:
1613                 {
1614                         struct mbuf *m = NULL;
1615                         caddr_t req = NULL;
1616                         size_t len = 0;
1617
1618                         if (m != NULL) {
1619                                 req = mtod(m, caddr_t);
1620                                 len = m->m_len;
1621                         }
1622                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1623                         if (error == 0)
1624                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1625                         if (error == 0)
1626                                 m_freem(m);
1627                         break;
1628                 }
1629 #endif /*IPSEC*/
1630
1631                 default:
1632                         error = ENOPROTOOPT;
1633                         break;
1634                 }
1635                 break;
1636         }
1637         return (error);
1638 }
1639
1640 /*
1641  * Set up IP options in pcb for insertion in output packets.
1642  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1643  * with destination address if source routed.
1644  */
1645 static int
1646 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1647 {
1648         int cnt, optlen;
1649         u_char *cp;
1650         u_char opt;
1651
1652         /* turn off any old options */
1653         if (*pcbopt)
1654                 m_free(*pcbopt);
1655         *pcbopt = 0;
1656         if (m == NULL || m->m_len == 0) {
1657                 /*
1658                  * Only turning off any previous options.
1659                  */
1660                 if (m != NULL)
1661                         m_free(m);
1662                 return (0);
1663         }
1664
1665         if (m->m_len % sizeof(int32_t))
1666                 goto bad;
1667         /*
1668          * IP first-hop destination address will be stored before
1669          * actual options; move other options back
1670          * and clear it when none present.
1671          */
1672         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1673                 goto bad;
1674         cnt = m->m_len;
1675         m->m_len += sizeof(struct in_addr);
1676         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1677         ovbcopy(mtod(m, caddr_t), cp, cnt);
1678         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1679
1680         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1681                 opt = cp[IPOPT_OPTVAL];
1682                 if (opt == IPOPT_EOL)
1683                         break;
1684                 if (opt == IPOPT_NOP)
1685                         optlen = 1;
1686                 else {
1687                         if (cnt < IPOPT_OLEN + sizeof *cp)
1688                                 goto bad;
1689                         optlen = cp[IPOPT_OLEN];
1690                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1691                                 goto bad;
1692                 }
1693                 switch (opt) {
1694
1695                 default:
1696                         break;
1697
1698                 case IPOPT_LSRR:
1699                 case IPOPT_SSRR:
1700                         /*
1701                          * user process specifies route as:
1702                          *      ->A->B->C->D
1703                          * D must be our final destination (but we can't
1704                          * check that since we may not have connected yet).
1705                          * A is first hop destination, which doesn't appear in
1706                          * actual IP option, but is stored before the options.
1707                          */
1708                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1709                                 goto bad;
1710                         m->m_len -= sizeof(struct in_addr);
1711                         cnt -= sizeof(struct in_addr);
1712                         optlen -= sizeof(struct in_addr);
1713                         cp[IPOPT_OLEN] = optlen;
1714                         /*
1715                          * Move first hop before start of options.
1716                          */
1717                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1718                               sizeof(struct in_addr));
1719                         /*
1720                          * Then copy rest of options back
1721                          * to close up the deleted entry.
1722                          */
1723                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1724                                 &cp[IPOPT_OFFSET+1],
1725                                 cnt - (IPOPT_MINOFF - 1));
1726                         break;
1727                 }
1728         }
1729         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1730                 goto bad;
1731         *pcbopt = m;
1732         return (0);
1733
1734 bad:
1735         m_free(m);
1736         return (EINVAL);
1737 }
1738
1739 /*
1740  * XXX
1741  * The whole multicast option thing needs to be re-thought.
1742  * Several of these options are equally applicable to non-multicast
1743  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1744  * standard option (IP_TTL).
1745  */
1746
1747 /*
1748  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1749  */
1750 static struct ifnet *
1751 ip_multicast_if(struct in_addr *a, int *ifindexp)
1752 {
1753         int ifindex;
1754         struct ifnet *ifp;
1755
1756         if (ifindexp)
1757                 *ifindexp = 0;
1758         if (ntohl(a->s_addr) >> 24 == 0) {
1759                 ifindex = ntohl(a->s_addr) & 0xffffff;
1760                 if (ifindex < 0 || if_index < ifindex)
1761                         return NULL;
1762                 ifp = ifindex2ifnet[ifindex];
1763                 if (ifindexp)
1764                         *ifindexp = ifindex;
1765         } else {
1766                 ifp = INADDR_TO_IFP(a);
1767         }
1768         return ifp;
1769 }
1770
1771 /*
1772  * Set the IP multicast options in response to user setsockopt().
1773  */
1774 static int
1775 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1776 {
1777         int error = 0;
1778         int i;
1779         struct in_addr addr;
1780         struct ip_mreq mreq;
1781         struct ifnet *ifp;
1782         struct ip_moptions *imo = *imop;
1783         int ifindex;
1784
1785         if (imo == NULL) {
1786                 /*
1787                  * No multicast option buffer attached to the pcb;
1788                  * allocate one and initialize to default values.
1789                  */
1790                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1791
1792                 *imop = imo;
1793                 imo->imo_multicast_ifp = NULL;
1794                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1795                 imo->imo_multicast_vif = -1;
1796                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1797                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1798                 imo->imo_num_memberships = 0;
1799         }
1800         switch (sopt->sopt_name) {
1801         /* store an index number for the vif you wanna use in the send */
1802         case IP_MULTICAST_VIF:
1803                 if (legal_vif_num == 0) {
1804                         error = EOPNOTSUPP;
1805                         break;
1806                 }
1807                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1808                 if (error)
1809                         break;
1810                 if (!legal_vif_num(i) && (i != -1)) {
1811                         error = EINVAL;
1812                         break;
1813                 }
1814                 imo->imo_multicast_vif = i;
1815                 break;
1816
1817         case IP_MULTICAST_IF:
1818                 /*
1819                  * Select the interface for outgoing multicast packets.
1820                  */
1821                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1822                 if (error)
1823                         break;
1824
1825                 /*
1826                  * INADDR_ANY is used to remove a previous selection.
1827                  * When no interface is selected, a default one is
1828                  * chosen every time a multicast packet is sent.
1829                  */
1830                 if (addr.s_addr == INADDR_ANY) {
1831                         imo->imo_multicast_ifp = NULL;
1832                         break;
1833                 }
1834                 /*
1835                  * The selected interface is identified by its local
1836                  * IP address.  Find the interface and confirm that
1837                  * it supports multicasting.
1838                  */
1839                 crit_enter();
1840                 ifp = ip_multicast_if(&addr, &ifindex);
1841                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1842                         crit_exit();
1843                         error = EADDRNOTAVAIL;
1844                         break;
1845                 }
1846                 imo->imo_multicast_ifp = ifp;
1847                 if (ifindex)
1848                         imo->imo_multicast_addr = addr;
1849                 else
1850                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1851                 crit_exit();
1852                 break;
1853
1854         case IP_MULTICAST_TTL:
1855                 /*
1856                  * Set the IP time-to-live for outgoing multicast packets.
1857                  * The original multicast API required a char argument,
1858                  * which is inconsistent with the rest of the socket API.
1859                  * We allow either a char or an int.
1860                  */
1861                 if (sopt->sopt_valsize == 1) {
1862                         u_char ttl;
1863                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1864                         if (error)
1865                                 break;
1866                         imo->imo_multicast_ttl = ttl;
1867                 } else {
1868                         u_int ttl;
1869                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1870                         if (error)
1871                                 break;
1872                         if (ttl > 255)
1873                                 error = EINVAL;
1874                         else
1875                                 imo->imo_multicast_ttl = ttl;
1876                 }
1877                 break;
1878
1879         case IP_MULTICAST_LOOP:
1880                 /*
1881                  * Set the loopback flag for outgoing multicast packets.
1882                  * Must be zero or one.  The original multicast API required a
1883                  * char argument, which is inconsistent with the rest
1884                  * of the socket API.  We allow either a char or an int.
1885                  */
1886                 if (sopt->sopt_valsize == 1) {
1887                         u_char loop;
1888
1889                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1890                         if (error)
1891                                 break;
1892                         imo->imo_multicast_loop = !!loop;
1893                 } else {
1894                         u_int loop;
1895
1896                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1897                                             sizeof loop);
1898                         if (error)
1899                                 break;
1900                         imo->imo_multicast_loop = !!loop;
1901                 }
1902                 break;
1903
1904         case IP_ADD_MEMBERSHIP:
1905                 /*
1906                  * Add a multicast group membership.
1907                  * Group must be a valid IP multicast address.
1908                  */
1909                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1910                 if (error)
1911                         break;
1912
1913                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1914                         error = EINVAL;
1915                         break;
1916                 }
1917                 crit_enter();
1918                 /*
1919                  * If no interface address was provided, use the interface of
1920                  * the route to the given multicast address.
1921                  */
1922                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1923                         struct sockaddr_in dst;
1924                         struct rtentry *rt;
1925
1926                         bzero(&dst, sizeof(struct sockaddr_in));
1927                         dst.sin_len = sizeof(struct sockaddr_in);
1928                         dst.sin_family = AF_INET;
1929                         dst.sin_addr = mreq.imr_multiaddr;
1930                         rt = rtlookup((struct sockaddr *)&dst);
1931                         if (rt == NULL) {
1932                                 error = EADDRNOTAVAIL;
1933                                 crit_exit();
1934                                 break;
1935                         }
1936                         --rt->rt_refcnt;
1937                         ifp = rt->rt_ifp;
1938                 } else {
1939                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1940                 }
1941
1942                 /*
1943                  * See if we found an interface, and confirm that it
1944                  * supports multicast.
1945                  */
1946                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1947                         error = EADDRNOTAVAIL;
1948                         crit_exit();
1949                         break;
1950                 }
1951                 /*
1952                  * See if the membership already exists or if all the
1953                  * membership slots are full.
1954                  */
1955                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1956                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1957                             imo->imo_membership[i]->inm_addr.s_addr
1958                                                 == mreq.imr_multiaddr.s_addr)
1959                                 break;
1960                 }
1961                 if (i < imo->imo_num_memberships) {
1962                         error = EADDRINUSE;
1963                         crit_exit();
1964                         break;
1965                 }
1966                 if (i == IP_MAX_MEMBERSHIPS) {
1967                         error = ETOOMANYREFS;
1968                         crit_exit();
1969                         break;
1970                 }
1971                 /*
1972                  * Everything looks good; add a new record to the multicast
1973                  * address list for the given interface.
1974                  */
1975                 if ((imo->imo_membership[i] =
1976                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1977                         error = ENOBUFS;
1978                         crit_exit();
1979                         break;
1980                 }
1981                 ++imo->imo_num_memberships;
1982                 crit_exit();
1983                 break;
1984
1985         case IP_DROP_MEMBERSHIP:
1986                 /*
1987                  * Drop a multicast group membership.
1988                  * Group must be a valid IP multicast address.
1989                  */
1990                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1991                 if (error)
1992                         break;
1993
1994                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1995                         error = EINVAL;
1996                         break;
1997                 }
1998
1999                 crit_enter();
2000                 /*
2001                  * If an interface address was specified, get a pointer
2002                  * to its ifnet structure.
2003                  */
2004                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2005                         ifp = NULL;
2006                 else {
2007                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2008                         if (ifp == NULL) {
2009                                 error = EADDRNOTAVAIL;
2010                                 crit_exit();
2011                                 break;
2012                         }
2013                 }
2014                 /*
2015                  * Find the membership in the membership array.
2016                  */
2017                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2018                         if ((ifp == NULL ||
2019                              imo->imo_membership[i]->inm_ifp == ifp) &&
2020                             imo->imo_membership[i]->inm_addr.s_addr ==
2021                             mreq.imr_multiaddr.s_addr)
2022                                 break;
2023                 }
2024                 if (i == imo->imo_num_memberships) {
2025                         error = EADDRNOTAVAIL;
2026                         crit_exit();
2027                         break;
2028                 }
2029                 /*
2030                  * Give up the multicast address record to which the
2031                  * membership points.
2032                  */
2033                 in_delmulti(imo->imo_membership[i]);
2034                 /*
2035                  * Remove the gap in the membership array.
2036                  */
2037                 for (++i; i < imo->imo_num_memberships; ++i)
2038                         imo->imo_membership[i-1] = imo->imo_membership[i];
2039                 --imo->imo_num_memberships;
2040                 crit_exit();
2041                 break;
2042
2043         default:
2044                 error = EOPNOTSUPP;
2045                 break;
2046         }
2047
2048         /*
2049          * If all options have default values, no need to keep the mbuf.
2050          */
2051         if (imo->imo_multicast_ifp == NULL &&
2052             imo->imo_multicast_vif == -1 &&
2053             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2054             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2055             imo->imo_num_memberships == 0) {
2056                 kfree(*imop, M_IPMOPTS);
2057                 *imop = NULL;
2058         }
2059
2060         return (error);
2061 }
2062
2063 /*
2064  * Return the IP multicast options in response to user getsockopt().
2065  */
2066 static int
2067 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2068 {
2069         struct in_addr addr;
2070         struct in_ifaddr *ia;
2071         int error, optval;
2072         u_char coptval;
2073
2074         error = 0;
2075         switch (sopt->sopt_name) {
2076         case IP_MULTICAST_VIF:
2077                 if (imo != NULL)
2078                         optval = imo->imo_multicast_vif;
2079                 else
2080                         optval = -1;
2081                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2082                 break;
2083
2084         case IP_MULTICAST_IF:
2085                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2086                         addr.s_addr = INADDR_ANY;
2087                 else if (imo->imo_multicast_addr.s_addr) {
2088                         /* return the value user has set */
2089                         addr = imo->imo_multicast_addr;
2090                 } else {
2091                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2092                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2093                                 : IA_SIN(ia)->sin_addr.s_addr;
2094                 }
2095                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2096                 break;
2097
2098         case IP_MULTICAST_TTL:
2099                 if (imo == NULL)
2100                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2101                 else
2102                         optval = coptval = imo->imo_multicast_ttl;
2103                 if (sopt->sopt_valsize == 1)
2104                         soopt_from_kbuf(sopt, &coptval, 1);
2105                 else
2106                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2107                 break;
2108
2109         case IP_MULTICAST_LOOP:
2110                 if (imo == NULL)
2111                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2112                 else
2113                         optval = coptval = imo->imo_multicast_loop;
2114                 if (sopt->sopt_valsize == 1)
2115                         soopt_from_kbuf(sopt, &coptval, 1);
2116                 else
2117                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2118                 break;
2119
2120         default:
2121                 error = ENOPROTOOPT;
2122                 break;
2123         }
2124         return (error);
2125 }
2126
2127 /*
2128  * Discard the IP multicast options.
2129  */
2130 void
2131 ip_freemoptions(struct ip_moptions *imo)
2132 {
2133         int i;
2134
2135         if (imo != NULL) {
2136                 for (i = 0; i < imo->imo_num_memberships; ++i)
2137                         in_delmulti(imo->imo_membership[i]);
2138                 kfree(imo, M_IPMOPTS);
2139         }
2140 }
2141
2142 /*
2143  * Routine called from ip_output() to loop back a copy of an IP multicast
2144  * packet to the input queue of a specified interface.  Note that this
2145  * calls the output routine of the loopback "driver", but with an interface
2146  * pointer that might NOT be a loopback interface -- evil, but easier than
2147  * replicating that code here.
2148  */
2149 static void
2150 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2151              int hlen)
2152 {
2153         struct ip *ip;
2154         struct mbuf *copym;
2155
2156         copym = m_copypacket(m, MB_DONTWAIT);
2157         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2158                 copym = m_pullup(copym, hlen);
2159         if (copym != NULL) {
2160                 /*
2161                  * if the checksum hasn't been computed, mark it as valid
2162                  */
2163                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2164                         in_delayed_cksum(copym);
2165                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2166                         copym->m_pkthdr.csum_flags |=
2167                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2168                         copym->m_pkthdr.csum_data = 0xffff;
2169                 }
2170                 /*
2171                  * We don't bother to fragment if the IP length is greater
2172                  * than the interface's MTU.  Can this possibly matter?
2173                  */
2174                 ip = mtod(copym, struct ip *);
2175                 ip->ip_len = htons(ip->ip_len);
2176                 ip->ip_off = htons(ip->ip_off);
2177                 ip->ip_sum = 0;
2178                 if (ip->ip_vhl == IP_VHL_BORING) {
2179                         ip->ip_sum = in_cksum_hdr(ip);
2180                 } else {
2181                         ip->ip_sum = in_cksum(copym, hlen);
2182                 }
2183                 /*
2184                  * NB:
2185                  * It's not clear whether there are any lingering
2186                  * reentrancy problems in other areas which might
2187                  * be exposed by using ip_input directly (in
2188                  * particular, everything which modifies the packet
2189                  * in-place).  Yet another option is using the
2190                  * protosw directly to deliver the looped back
2191                  * packet.  For the moment, we'll err on the side
2192                  * of safety by using if_simloop().
2193                  */
2194 #if 1 /* XXX */
2195                 if (dst->sin_family != AF_INET) {
2196                         kprintf("ip_mloopback: bad address family %d\n",
2197                                                 dst->sin_family);
2198                         dst->sin_family = AF_INET;
2199                 }
2200 #endif
2201
2202 #ifdef notdef
2203                 copym->m_pkthdr.rcvif = ifp;
2204                 ip_input(copym);
2205 #else
2206                 if_simloop(ifp, copym, dst->sin_family, 0);
2207 #endif
2208         }
2209 }