sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  */
  32
  33 #define _IP_VHL
  34
  35 #include "opt_ipdn.h"
  36 #include "opt_ipdivert.h"
  37 #include "opt_ipsec.h"
  38 #include "opt_mbuf_stress_test.h"
  39 #include "opt_mpls.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/kernel.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mbuf.h>
  46 #include <sys/protosw.h>
  47 #include <sys/socket.h>
  48 #include <sys/socketvar.h>
  49 #include <sys/proc.h>
  50 #include <sys/priv.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/in_cksum.h>
  53 #include <sys/lock.h>
  54
  55 #include <sys/thread2.h>
  56 #include <sys/mplock2.h>
  57 #include <sys/msgport2.h>
  58
  59 #include <net/if.h>
  60 #include <net/netisr.h>
  61 #include <net/pfil.h>
  62 #include <net/route.h>
  63
  64 #include <netinet/in.h>
  65 #include <netinet/in_systm.h>
  66 #include <netinet/ip.h>
  67 #include <netinet/in_pcb.h>
  68 #include <netinet/in_var.h>
  69 #include <netinet/ip_var.h>
  70
  71 #include <netproto/mpls/mpls_var.h>
  72
  73 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  74
  75 #ifdef IPSEC
  76 #include <netinet6/ipsec.h>
  77 #include <netproto/key/key.h>
  78 #ifdef IPSEC_DEBUG
  79 #include <netproto/key/key_debug.h>
  80 #else
  81 #define KEYDEBUG(lev,arg)
  82 #endif
  83 #endif /*IPSEC*/
  84
  85 #ifdef FAST_IPSEC
  86 #include <netproto/ipsec/ipsec.h>
  87 #include <netproto/ipsec/xform.h>
  88 #include <netproto/ipsec/key.h>
  89 #endif /*FAST_IPSEC*/
  90
  91 #include <net/ipfw/ip_fw.h>
  92 #include <net/dummynet/ip_dummynet.h>
  93
  94 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  95                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  96                                   (ntohl(a.s_addr)>>16)&0xFF,\
  97                                   (ntohl(a.s_addr)>>8)&0xFF,\
  98                                   (ntohl(a.s_addr))&0xFF, y);
  99
 100 u_short ip_id;
 101
 102 #ifdef MBUF_STRESS_TEST
 103 int mbuf_frag_size = 0;
 104 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 105         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 106 #endif
 107
 108 static int ip_do_rfc6864 = 1;
 109 SYSCTL_INT(_net_inet_ip, OID_AUTO, rfc6864, CTLFLAG_RW, &ip_do_rfc6864, 0,
 110     "Don't generate IP ID for DF IP datagrams");
 111
 112 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 113 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 114 static void     ip_mloopback
 115         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 116 static int      ip_getmoptions
 117         (struct sockopt *, struct ip_moptions *);
 118 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 119 static int      ip_setmoptions
 120         (struct sockopt *, struct ip_moptions **);
 121
 122 int     ip_optcopy(struct ip *, struct ip *);
 123
 124 extern  struct protosw inetsw[];
 125
 126 static int
 127 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
 128 {
 129         struct in_ifaddr_container *iac;
 130
 131         /*
 132          * We need to figure out if we have been forwarded to a local
 133          * socket.  If so, then we should somehow "loop back" to
 134          * ip_input(), and get directed to the PCB as if we had received
 135          * this packet.  This is because it may be difficult to identify
 136          * the packets you want to forward until they are being output
 137          * and have selected an interface (e.g. locally initiated
 138          * packets).  If we used the loopback inteface, we would not be
 139          * able to control what happens as the packet runs through
 140          * ip_input() as it is done through a ISR.
 141          */
 142         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
 143                 /*
 144                  * If the addr to forward to is one of ours, we pretend
 145                  * to be the destination for this packet.
 146                  */
 147                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
 148                         break;
 149         }
 150         if (iac != NULL) {
 151                 struct ip *ip;
 152
 153                 if (m->m_pkthdr.rcvif == NULL)
 154                         m->m_pkthdr.rcvif = ifunit_netisr("lo0");
 155                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 156                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 157                                                   CSUM_PSEUDO_HDR;
 158                         m->m_pkthdr.csum_data = 0xffff;
 159                 }
 160                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 161
 162                 /*
 163                  * Make sure that the IP header is in one mbuf,
 164                  * required by ip_input
 165                  */
 166                 if (m->m_len < hlen) {
 167                         m = m_pullup(m, hlen);
 168                         if (m == NULL) {
 169                                 /* The packet was freed; we are done */
 170                                 return 1;
 171                         }
 172                 }
 173                 ip = mtod(m, struct ip *);
 174
 175                 ip->ip_len = htons(ip->ip_len);
 176                 ip->ip_off = htons(ip->ip_off);
 177                 ip_input(m);
 178
 179                 return 1; /* The packet gets forwarded locally */
 180         }
 181         return 0;
 182 }
 183
 184 /*
 185  * IP output.  The packet in mbuf chain m contains a skeletal IP
 186  * header (with len, off, ttl, proto, tos, src, dst).
 187  * The mbuf chain containing the packet will be freed.
 188  * The mbuf opt, if present, will not be freed.
 189  */
 190 int
 191 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 192           int flags, struct ip_moptions *imo, struct inpcb *inp)
 193 {
 194         struct ip *ip;
 195         struct ifnet *ifp = NULL;       /* keep compiler happy */
 196         struct mbuf *m;
 197         int hlen = sizeof(struct ip);
 198         int len, error = 0;
 199         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 200         struct in_ifaddr *ia = NULL;
 201         int isbroadcast, sw_csum;
 202         struct in_addr pkt_dst;
 203         struct route iproute;
 204         struct m_tag *mtag;
 205 #ifdef IPSEC
 206         struct secpolicy *sp = NULL;
 207         struct socket *so = inp ? inp->inp_socket : NULL;
 208 #endif
 209 #ifdef FAST_IPSEC
 210         struct secpolicy *sp = NULL;
 211         struct tdb_ident *tdbi;
 212 #endif /* FAST_IPSEC */
 213         struct sockaddr_in *next_hop = NULL;
 214         int src_was_INADDR_ANY = 0;     /* as the name says... */
 215
 216         m = m0;
 217         M_ASSERTPKTHDR(m);
 218
 219         if (ro == NULL) {
 220                 ro = &iproute;
 221                 bzero(ro, sizeof *ro);
 222         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
 223                 if (flags & IP_DEBUGROUTE) {
 224                         panic("ip_output: rt rt_cpuid %d accessed on cpu %d\n",
 225                             ro->ro_rt->rt_cpuid, mycpuid);
 226                 }
 227
 228                 /*
 229                  * XXX
 230                  * If the cached rtentry's owner CPU is not the current CPU,
 231                  * then don't touch the cached rtentry (remote free is too
 232                  * expensive in this context); just relocate the route.
 233                  */
 234                 ro = &iproute;
 235                 bzero(ro, sizeof *ro);
 236         }
 237
 238         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 239                 /* Next hop */
 240                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 241                 KKASSERT(mtag != NULL);
 242                 next_hop = m_tag_data(mtag);
 243         }
 244
 245         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 246                 struct dn_pkt *dn_pkt;
 247
 248                 /* Extract info from dummynet tag */
 249                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 250                 KKASSERT(mtag != NULL);
 251                 dn_pkt = m_tag_data(mtag);
 252
 253                 /*
 254                  * The packet was already tagged, so part of the
 255                  * processing was already done, and we need to go down.
 256                  * Get the calculated parameters from the tag.
 257                  */
 258                 ifp = dn_pkt->ifp;
 259
 260                 KKASSERT(ro == &iproute);
 261                 *ro = dn_pkt->ro; /* structure copy */
 262                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
 263
 264                 dst = dn_pkt->dn_dst;
 265                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
 266                         /* If 'dst' points into dummynet tag, adjust it */
 267                         dst = (struct sockaddr_in *)&(ro->ro_dst);
 268                 }
 269
 270                 ip = mtod(m, struct ip *);
 271                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 272                 if (ro->ro_rt)
 273                         ia = ifatoia(ro->ro_rt->rt_ifa);
 274                 goto sendit;
 275         }
 276
 277         if (opt) {
 278                 len = 0;
 279                 m = ip_insertoptions(m, opt, &len);
 280                 if (len != 0)
 281                         hlen = len;
 282         }
 283         ip = mtod(m, struct ip *);
 284
 285         /*
 286          * Fill in IP header.
 287          */
 288         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 289                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 290                 ip->ip_off &= IP_DF;
 291                 if (ip_do_rfc6864 && (ip->ip_off & IP_DF))
 292                         ip->ip_id = 0;
 293                 else
 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 = 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                 m->m_flags |= M_MCAST;
 376                 /*
 377                  * IP destination address is multicast.  Make sure "dst"
 378                  * still points to the address in "ro".  (It may have been
 379                  * changed to point to a gateway address, above.)
 380                  */
 381                 dst = (struct sockaddr_in *)&ro->ro_dst;
 382                 /*
 383                  * See if the caller provided any multicast options
 384                  */
 385                 if (imo != NULL) {
 386                         ip->ip_ttl = imo->imo_multicast_ttl;
 387                         if (imo->imo_multicast_vif != -1) {
 388                                 ip->ip_src.s_addr =
 389                                     ip_mcast_src ?
 390                                     ip_mcast_src(imo->imo_multicast_vif) :
 391                                     INADDR_ANY;
 392                         }
 393                 } else {
 394                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 395                 }
 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 of the
 408                  * outgoing interface.  In case, keep note we did that, so
 409                  * if the the firewall changes the next-hop causing the
 410                  * output interface to change, we can fix that.
 411                  */
 412                 if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 413                         /* Interface may have no addresses. */
 414                         if (ia != NULL) {
 415                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 416                                 src_was_INADDR_ANY = 1;
 417                         }
 418                 }
 419
 420                 if (ip->ip_src.s_addr != INADDR_ANY) {
 421                         struct in_multi *inm;
 422
 423                         inm = IN_LOOKUP_MULTI(&pkt_dst, ifp);
 424                         if (inm != NULL &&
 425                             (imo == NULL || imo->imo_multicast_loop)) {
 426                                 /*
 427                                  * If we belong to the destination multicast
 428                                  * group on the outgoing interface, and the
 429                                  * caller did not forbid loopback, loop back
 430                                  * a copy.
 431                                  */
 432                                 ip_mloopback(ifp, m, dst, hlen);
 433                         } else {
 434                                 /*
 435                                  * If we are acting as a multicast router,
 436                                  * perform multicast forwarding as if the
 437                                  * packet had just arrived on the interface
 438                                  * to which we are about to send.  The
 439                                  * multicast forwarding function recursively
 440                                  * calls this function, using the IP_FORWARDING
 441                                  * flag to prevent infinite recursion.
 442                                  *
 443                                  * Multicasts that are looped back by
 444                                  * ip_mloopback(), above, will be forwarded by
 445                                  * the ip_input() routine, if necessary.
 446                                  */
 447                                 if (ip_mrouter && !(flags & IP_FORWARDING)) {
 448                                         /*
 449                                          * If rsvp daemon is not running, do
 450                                          * not set ip_moptions. This ensures
 451                                          * that the packet is multicast and
 452                                          * not just sent down one link as
 453                                          * prescribed by rsvpd.
 454                                          */
 455                                         if (!rsvp_on)
 456                                                 imo = NULL;
 457                                         if (ip_mforward) {
 458                                                 get_mplock();
 459                                                 if (ip_mforward(ip, ifp,
 460                                                     m, imo) != 0) {
 461                                                         m_freem(m);
 462                                                         rel_mplock();
 463                                                         goto done;
 464                                                 }
 465                                                 rel_mplock();
 466                                         }
 467                                 }
 468                         }
 469                 }
 470
 471                 /*
 472                  * Multicasts with a time-to-live of zero may be looped-
 473                  * back, above, but must not be transmitted on a network.
 474                  * Also, multicasts addressed to the loopback interface
 475                  * are not sent -- the above call to ip_mloopback() will
 476                  * loop back a copy if this host actually belongs to the
 477                  * destination group on the loopback interface.
 478                  */
 479                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 480                         m_freem(m);
 481                         goto done;
 482                 }
 483
 484                 goto sendit;
 485         } else {
 486                 m->m_flags &= ~M_MCAST;
 487         }
 488
 489         /*
 490          * If the source address is not specified yet, use the address
 491          * of the outgoing interface.  In case, keep note we did that,
 492          * so if the the firewall changes the next-hop causing the output
 493          * interface to change, we can fix that.
 494          */
 495         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
 496                 /* Interface may have no addresses. */
 497                 if (ia != NULL) {
 498                         ip->ip_src = IA_SIN(ia)->sin_addr;
 499                         src_was_INADDR_ANY = 1;
 500                 }
 501         }
 502
 503         /*
 504          * Look for broadcast address and
 505          * verify user is allowed to send
 506          * such a packet.
 507          */
 508         if (isbroadcast) {
 509                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 510                         error = EADDRNOTAVAIL;
 511                         goto bad;
 512                 }
 513                 if (!(flags & IP_ALLOWBROADCAST)) {
 514                         error = EACCES;
 515                         goto bad;
 516                 }
 517                 /* don't allow broadcast messages to be fragmented */
 518                 if (ip->ip_len > ifp->if_mtu) {
 519                         error = EMSGSIZE;
 520                         goto bad;
 521                 }
 522                 m->m_flags |= M_BCAST;
 523         } else {
 524                 m->m_flags &= ~M_BCAST;
 525         }
 526
 527 sendit:
 528 #ifdef IPSEC
 529         /* get SP for this packet */
 530         if (so == NULL)
 531                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 532         else
 533                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 534
 535         if (sp == NULL) {
 536                 ipsecstat.out_inval++;
 537                 goto bad;
 538         }
 539
 540         error = 0;
 541
 542         /* check policy */
 543         switch (sp->policy) {
 544         case IPSEC_POLICY_DISCARD:
 545                 /*
 546                  * This packet is just discarded.
 547                  */
 548                 ipsecstat.out_polvio++;
 549                 goto bad;
 550
 551         case IPSEC_POLICY_BYPASS:
 552         case IPSEC_POLICY_NONE:
 553         case IPSEC_POLICY_TCP:
 554                 /* no need to do IPsec. */
 555                 goto skip_ipsec;
 556
 557         case IPSEC_POLICY_IPSEC:
 558                 if (sp->req == NULL) {
 559                         /* acquire a policy */
 560                         error = key_spdacquire(sp);
 561                         goto bad;
 562                 }
 563                 break;
 564
 565         case IPSEC_POLICY_ENTRUST:
 566         default:
 567                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 568         }
 569     {
 570         struct ipsec_output_state state;
 571         bzero(&state, sizeof state);
 572         state.m = m;
 573         if (flags & IP_ROUTETOIF) {
 574                 state.ro = &iproute;
 575                 bzero(&iproute, sizeof iproute);
 576         } else
 577                 state.ro = ro;
 578         state.dst = (struct sockaddr *)dst;
 579
 580         ip->ip_sum = 0;
 581
 582         /*
 583          * XXX
 584          * delayed checksums are not currently compatible with IPsec
 585          */
 586         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 587                 in_delayed_cksum(m);
 588                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 589         }
 590
 591         ip->ip_len = htons(ip->ip_len);
 592         ip->ip_off = htons(ip->ip_off);
 593
 594         error = ipsec4_output(&state, sp, flags);
 595
 596         m = state.m;
 597         if (flags & IP_ROUTETOIF) {
 598                 /*
 599                  * if we have tunnel mode SA, we may need to ignore
 600                  * IP_ROUTETOIF.
 601                  */
 602                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 603                         flags &= ~IP_ROUTETOIF;
 604                         ro = state.ro;
 605                 }
 606         } else
 607                 ro = state.ro;
 608         dst = (struct sockaddr_in *)state.dst;
 609         if (error) {
 610                 /* mbuf is already reclaimed in ipsec4_output. */
 611                 m0 = NULL;
 612                 switch (error) {
 613                 case EHOSTUNREACH:
 614                 case ENETUNREACH:
 615                 case EMSGSIZE:
 616                 case ENOBUFS:
 617                 case ENOMEM:
 618                         break;
 619                 default:
 620                         kprintf("ip4_output (ipsec): error code %d\n", error);
 621                         /*fall through*/
 622                 case ENOENT:
 623                         /* don't show these error codes to the user */
 624                         error = 0;
 625                         break;
 626                 }
 627                 goto bad;
 628         }
 629     }
 630
 631         /* be sure to update variables that are affected by ipsec4_output() */
 632         ip = mtod(m, struct ip *);
 633 #ifdef _IP_VHL
 634         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 635 #else
 636         hlen = ip->ip_hl << 2;
 637 #endif
 638         if (ro->ro_rt == NULL) {
 639                 if (!(flags & IP_ROUTETOIF)) {
 640                         kprintf("ip_output: "
 641                                 "can't update route after IPsec processing\n");
 642                         error = EHOSTUNREACH;   /*XXX*/
 643                         goto bad;
 644                 }
 645         } else {
 646                 ia = ifatoia(ro->ro_rt->rt_ifa);
 647                 ifp = ro->ro_rt->rt_ifp;
 648         }
 649
 650         /* make it flipped, again. */
 651         ip->ip_len = ntohs(ip->ip_len);
 652         ip->ip_off = ntohs(ip->ip_off);
 653 skip_ipsec:
 654 #endif /*IPSEC*/
 655 #ifdef FAST_IPSEC
 656         /*
 657          * Check the security policy (SP) for the packet and, if
 658          * required, do IPsec-related processing.  There are two
 659          * cases here; the first time a packet is sent through
 660          * it will be untagged and handled by ipsec4_checkpolicy.
 661          * If the packet is resubmitted to ip_output (e.g. after
 662          * AH, ESP, etc. processing), there will be a tag to bypass
 663          * the lookup and related policy checking.
 664          */
 665         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 666         crit_enter();
 667         if (mtag != NULL) {
 668                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 669                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 670                 if (sp == NULL)
 671                         error = -EINVAL;        /* force silent drop */
 672                 m_tag_delete(m, mtag);
 673         } else {
 674                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 675                                         &error, inp);
 676         }
 677         /*
 678          * There are four return cases:
 679          *    sp != NULL                    apply IPsec policy
 680          *    sp == NULL, error == 0        no IPsec handling needed
 681          *    sp == NULL, error == -EINVAL  discard packet w/o error
 682          *    sp == NULL, error != 0        discard packet, report error
 683          */
 684         if (sp != NULL) {
 685                 /* Loop detection, check if ipsec processing already done */
 686                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 687                 for (mtag = m_tag_first(m); mtag != NULL;
 688                      mtag = m_tag_next(m, mtag)) {
 689                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 690                                 continue;
 691                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 692                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 693                                 continue;
 694                         /*
 695                          * Check if policy has an SA associated with it.
 696                          * This can happen when an SP has yet to acquire
 697                          * an SA; e.g. on first reference.  If it occurs,
 698                          * then we let ipsec4_process_packet do its thing.
 699                          */
 700                         if (sp->req->sav == NULL)
 701                                 break;
 702                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 703                         if (tdbi->spi == sp->req->sav->spi &&
 704                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 705                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 706                                  sizeof(union sockaddr_union)) == 0) {
 707                                 /*
 708                                  * No IPsec processing is needed, free
 709                                  * reference to SP.
 710                                  *
 711                                  * NB: null pointer to avoid free at
 712                                  *     done: below.
 713                                  */
 714                                 KEY_FREESP(&sp), sp = NULL;
 715                                 crit_exit();
 716                                 goto spd_done;
 717                         }
 718                 }
 719
 720                 /*
 721                  * Do delayed checksums now because we send before
 722                  * this is done in the normal processing path.
 723                  */
 724                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 725                         in_delayed_cksum(m);
 726                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 727                 }
 728
 729                 ip->ip_len = htons(ip->ip_len);
 730                 ip->ip_off = htons(ip->ip_off);
 731
 732                 /* NB: callee frees mbuf */
 733                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 734                 /*
 735                  * Preserve KAME behaviour: ENOENT can be returned
 736                  * when an SA acquire is in progress.  Don't propagate
 737                  * this to user-level; it confuses applications.
 738                  *
 739                  * XXX this will go away when the SADB is redone.
 740                  */
 741                 if (error == ENOENT)
 742                         error = 0;
 743                 crit_exit();
 744                 goto done;
 745         } else {
 746                 crit_exit();
 747
 748                 if (error != 0) {
 749                         /*
 750                          * Hack: -EINVAL is used to signal that a packet
 751                          * should be silently discarded.  This is typically
 752                          * because we asked key management for an SA and
 753                          * it was delayed (e.g. kicked up to IKE).
 754                          */
 755                         if (error == -EINVAL)
 756                                 error = 0;
 757                         goto bad;
 758                 } else {
 759                         /* No IPsec processing for this packet. */
 760                 }
 761 #ifdef notyet
 762                 /*
 763                  * If deferred crypto processing is needed, check that
 764                  * the interface supports it.
 765                  */
 766                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 767                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 768                         /* notify IPsec to do its own crypto */
 769                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 770                         error = EHOSTUNREACH;
 771                         goto bad;
 772                 }
 773 #endif
 774         }
 775 spd_done:
 776 #endif /* FAST_IPSEC */
 777
 778         /* We are already being fwd'd from a firewall. */
 779         if (next_hop != NULL)
 780                 goto pass;
 781
 782         /* No pfil hooks */
 783         if (!pfil_has_hooks(&inet_pfil_hook)) {
 784                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 785                         /*
 786                          * Strip dummynet tags from stranded packets
 787                          */
 788                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 789                         KKASSERT(mtag != NULL);
 790                         m_tag_delete(m, mtag);
 791                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 792                 }
 793                 goto pass;
 794         }
 795
 796         /*
 797          * IpHack's section.
 798          * - Xlate: translate packet's addr/port (NAT).
 799          * - Firewall: deny/allow/etc.
 800          * - Wrap: fake packet's addr/port <unimpl.>
 801          * - Encapsulate: put it in another IP and send out. <unimp.>
 802          */
 803
 804         /*
 805          * Run through list of hooks for output packets.
 806          */
 807         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 808         if (error != 0 || m == NULL)
 809                 goto done;
 810         ip = mtod(m, struct ip *);
 811
 812         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 813                 /*
 814                  * Check dst to make sure it is directly reachable on the
 815                  * interface we previously thought it was.
 816                  * If it isn't (which may be likely in some situations) we have
 817                  * to re-route it (ie, find a route for the next-hop and the
 818                  * associated interface) and set them here. This is nested
 819                  * forwarding which in most cases is undesirable, except where
 820                  * such control is nigh impossible. So we do it here.
 821                  * And I'm babbling.
 822                  */
 823                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 824                 KKASSERT(mtag != NULL);
 825                 next_hop = m_tag_data(mtag);
 826
 827                 /*
 828                  * Try local forwarding first
 829                  */
 830                 if (ip_localforward(m, next_hop, hlen))
 831                         goto done;
 832
 833                 /*
 834                  * Relocate the route based on next_hop.
 835                  * If the current route is inp's cache, keep it untouched.
 836                  */
 837                 if (ro == &iproute && ro->ro_rt != NULL) {
 838                         RTFREE(ro->ro_rt);
 839                         ro->ro_rt = NULL;
 840                 }
 841                 ro = &iproute;
 842                 bzero(ro, sizeof *ro);
 843
 844                 /*
 845                  * Forwarding to broadcast address is not allowed.
 846                  * XXX Should we follow IP_ROUTETOIF?
 847                  */
 848                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
 849
 850                 /* We are doing forwarding now */
 851                 flags |= IP_FORWARDING;
 852
 853                 goto reroute;
 854         }
 855
 856         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 857                 struct dn_pkt *dn_pkt;
 858
 859                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 860                 KKASSERT(mtag != NULL);
 861                 dn_pkt = m_tag_data(mtag);
 862
 863                 /*
 864                  * Under certain cases it is not possible to recalculate
 865                  * 'ro' and 'dst', let alone 'flags', so just save them in
 866                  * dummynet tag and avoid the possible wrong reculcalation
 867                  * when we come back to ip_output() again.
 868                  *
 869                  * All other parameters have been already used and so they
 870                  * are not needed anymore.
 871                  * XXX if the ifp is deleted while a pkt is in dummynet,
 872                  * we are in trouble! (TODO use ifnet_detach_event)
 873                  *
 874                  * We need to copy *ro because for ICMP pkts (and maybe
 875                  * others) the caller passed a pointer into the stack;
 876                  * dst might also be a pointer into *ro so it needs to
 877                  * be updated.
 878                  */
 879                 dn_pkt->ro = *ro;
 880                 if (ro->ro_rt)
 881                         ro->ro_rt->rt_refcnt++;
 882                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
 883                         /* 'dst' points into 'ro' */
 884                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
 885                 }
 886                 dn_pkt->dn_dst = dst;
 887                 dn_pkt->flags = flags;
 888
 889                 ip_dn_queue(m);
 890                 goto done;
 891         }
 892 pass:
 893         /* 127/8 must not appear on wire - RFC1122. */
 894         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 895             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 896                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 897                         ipstat.ips_badaddr++;
 898                         error = EADDRNOTAVAIL;
 899                         goto bad;
 900                 }
 901         }
 902         if (ip->ip_src.s_addr == INADDR_ANY ||
 903             IN_MULTICAST(ntohl(ip->ip_src.s_addr))) {
 904                 ipstat.ips_badaddr++;
 905                 error = EADDRNOTAVAIL;
 906                 goto bad;
 907         }
 908
 909         if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
 910                 m->m_pkthdr.csum_flags |= CSUM_IP;
 911                 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 912                 if (sw_csum & CSUM_DELAY_DATA) {
 913                         in_delayed_cksum(m);
 914                         sw_csum &= ~CSUM_DELAY_DATA;
 915                 }
 916                 m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 917         } else {
 918                 sw_csum = 0;
 919         }
 920         m->m_pkthdr.csum_iphlen = hlen;
 921
 922         /*
 923          * If small enough for interface, or the interface will take
 924          * care of the fragmentation or segmentation for us, can just
 925          * send directly.
 926          */
 927         if (ip->ip_len <= ifp->if_mtu ||
 928             ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
 929             (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 930                 ip->ip_len = htons(ip->ip_len);
 931                 ip->ip_off = htons(ip->ip_off);
 932                 ip->ip_sum = 0;
 933                 if (sw_csum & CSUM_DELAY_IP) {
 934                         if (ip->ip_vhl == IP_VHL_BORING)
 935                                 ip->ip_sum = in_cksum_hdr(ip);
 936                         else
 937                                 ip->ip_sum = in_cksum(m, hlen);
 938                 }
 939
 940                 /* Record statistics for this interface address. */
 941                 if (!(flags & IP_FORWARDING) && ia) {
 942                         IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
 943                         IFA_STAT_INC(&ia->ia_ifa, obytes, m->m_pkthdr.len);
 944                 }
 945
 946 #ifdef IPSEC
 947                 /* clean ipsec history once it goes out of the node */
 948                 ipsec_delaux(m);
 949 #endif
 950
 951 #ifdef MBUF_STRESS_TEST
 952                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
 953                         struct mbuf *m1, *m2;
 954                         int length, tmp;
 955
 956                         tmp = length = m->m_pkthdr.len;
 957
 958                         while ((length -= mbuf_frag_size) >= 1) {
 959                                 m1 = m_split(m, length, M_NOWAIT);
 960                                 if (m1 == NULL)
 961                                         break;
 962                                 m2 = m;
 963                                 while (m2->m_next != NULL)
 964                                         m2 = m2->m_next;
 965                                 m2->m_next = m1;
 966                         }
 967                         m->m_pkthdr.len = tmp;
 968                 }
 969 #endif
 970
 971 #ifdef MPLS
 972                 if (!mpls_output_process(m, ro->ro_rt))
 973                         goto done;
 974 #endif
 975                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
 976                                        ro->ro_rt);
 977                 goto done;
 978         }
 979
 980         if (ip->ip_off & IP_DF) {
 981                 error = EMSGSIZE;
 982                 /*
 983                  * This case can happen if the user changed the MTU
 984                  * of an interface after enabling IP on it.  Because
 985                  * most netifs don't keep track of routes pointing to
 986                  * them, there is no way for one to update all its
 987                  * routes when the MTU is changed.
 988                  */
 989                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
 990                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 991                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 992                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 993                 }
 994                 ipstat.ips_cantfrag++;
 995                 goto bad;
 996         }
 997
 998         /*
 999          * Too large for interface; fragment if possible. If successful,
1000          * on return, m will point to a list of packets to be sent.
1001          */
1002         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1003         if (error)
1004                 goto bad;
1005         for (; m; m = m0) {
1006                 m0 = m->m_nextpkt;
1007                 m->m_nextpkt = NULL;
1008 #ifdef IPSEC
1009                 /* clean ipsec history once it goes out of the node */
1010                 ipsec_delaux(m);
1011 #endif
1012                 if (error == 0) {
1013                         /* Record statistics for this interface address. */
1014                         if (ia != NULL) {
1015                                 IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
1016                                 IFA_STAT_INC(&ia->ia_ifa, obytes,
1017                                     m->m_pkthdr.len);
1018                         }
1019 #ifdef MPLS
1020                         if (!mpls_output_process(m, ro->ro_rt))
1021                                 continue;
1022 #endif
1023                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1024                                                ro->ro_rt);
1025                 } else {
1026                         m_freem(m);
1027                 }
1028         }
1029
1030         if (error == 0)
1031                 ipstat.ips_fragmented++;
1032
1033 done:
1034         if (ro == &iproute && ro->ro_rt != NULL) {
1035                 RTFREE(ro->ro_rt);
1036                 ro->ro_rt = NULL;
1037         }
1038 #ifdef IPSEC
1039         if (sp != NULL) {
1040                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1041                         kprintf("DP ip_output call free SP:%p\n", sp));
1042                 key_freesp(sp);
1043         }
1044 #endif
1045 #ifdef FAST_IPSEC
1046         if (sp != NULL)
1047                 KEY_FREESP(&sp);
1048 #endif
1049         return (error);
1050 bad:
1051         m_freem(m);
1052         goto done;
1053 }
1054
1055 /*
1056  * Create a chain of fragments which fit the given mtu. m_frag points to the
1057  * mbuf to be fragmented; on return it points to the chain with the fragments.
1058  * Return 0 if no error. If error, m_frag may contain a partially built
1059  * chain of fragments that should be freed by the caller.
1060  *
1061  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1062  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1063  */
1064 int
1065 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1066             u_long if_hwassist_flags, int sw_csum)
1067 {
1068         int error = 0;
1069         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1070         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1071         int off;
1072         struct mbuf *m0 = *m_frag;      /* the original packet          */
1073         int firstlen;
1074         struct mbuf **mnext;
1075         int nfrags;
1076
1077         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1078                 ipstat.ips_cantfrag++;
1079                 return EMSGSIZE;
1080         }
1081
1082         /*
1083          * Must be able to put at least 8 bytes per fragment.
1084          */
1085         if (len < 8)
1086                 return EMSGSIZE;
1087
1088         /*
1089          * If the interface will not calculate checksums on
1090          * fragmented packets, then do it here.
1091          */
1092         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1093             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1094                 in_delayed_cksum(m0);
1095                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1096         }
1097
1098         if (len > PAGE_SIZE) {
1099                 /*
1100                  * Fragment large datagrams such that each segment
1101                  * contains a multiple of PAGE_SIZE amount of data,
1102                  * plus headers. This enables a receiver to perform
1103                  * page-flipping zero-copy optimizations.
1104                  *
1105                  * XXX When does this help given that sender and receiver
1106                  * could have different page sizes, and also mtu could
1107                  * be less than the receiver's page size ?
1108                  */
1109                 int newlen;
1110                 struct mbuf *m;
1111
1112                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1113                         off += m->m_len;
1114
1115                 /*
1116                  * firstlen (off - hlen) must be aligned on an
1117                  * 8-byte boundary
1118                  */
1119                 if (off < hlen)
1120                         goto smart_frag_failure;
1121                 off = ((off - hlen) & ~7) + hlen;
1122                 newlen = (~PAGE_MASK) & mtu;
1123                 if ((newlen + sizeof(struct ip)) > mtu) {
1124                         /* we failed, go back the default */
1125 smart_frag_failure:
1126                         newlen = len;
1127                         off = hlen + len;
1128                 }
1129                 len = newlen;
1130
1131         } else {
1132                 off = hlen + len;
1133         }
1134
1135         firstlen = off - hlen;
1136         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1137
1138         /*
1139          * Loop through length of segment after first fragment,
1140          * make new header and copy data of each part and link onto chain.
1141          * Here, m0 is the original packet, m is the fragment being created.
1142          * The fragments are linked off the m_nextpkt of the original
1143          * packet, which after processing serves as the first fragment.
1144          */
1145         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1146                 struct ip *mhip;        /* ip header on the fragment */
1147                 struct mbuf *m;
1148                 int mhlen = sizeof(struct ip);
1149
1150                 MGETHDR(m, M_NOWAIT, MT_HEADER);
1151                 if (m == NULL) {
1152                         error = ENOBUFS;
1153                         ipstat.ips_odropped++;
1154                         goto done;
1155                 }
1156                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1157                 /*
1158                  * In the first mbuf, leave room for the link header, then
1159                  * copy the original IP header including options. The payload
1160                  * goes into an additional mbuf chain returned by m_copy().
1161                  */
1162                 m->m_data += max_linkhdr;
1163                 mhip = mtod(m, struct ip *);
1164                 *mhip = *ip;
1165                 if (hlen > sizeof(struct ip)) {
1166                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1167                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1168                 }
1169                 m->m_len = mhlen;
1170                 /* XXX do we need to add ip->ip_off below ? */
1171                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1172                 if (off + len >= ip->ip_len) {  /* last fragment */
1173                         len = ip->ip_len - off;
1174                         m->m_flags |= M_LASTFRAG;
1175                 } else
1176                         mhip->ip_off |= IP_MF;
1177                 mhip->ip_len = htons((u_short)(len + mhlen));
1178                 m->m_next = m_copy(m0, off, len);
1179                 if (m->m_next == NULL) {                /* copy failed */
1180                         m_free(m);
1181                         error = ENOBUFS;        /* ??? */
1182                         ipstat.ips_odropped++;
1183                         goto done;
1184                 }
1185                 m->m_pkthdr.len = mhlen + len;
1186                 m->m_pkthdr.rcvif = NULL;
1187                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1188                 m->m_pkthdr.csum_iphlen = mhlen;
1189                 mhip->ip_off = htons(mhip->ip_off);
1190                 mhip->ip_sum = 0;
1191                 if (sw_csum & CSUM_DELAY_IP)
1192                         mhip->ip_sum = in_cksum(m, mhlen);
1193                 *mnext = m;
1194                 mnext = &m->m_nextpkt;
1195         }
1196         ipstat.ips_ofragments += nfrags;
1197
1198         /* set first marker for fragment chain */
1199         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1200         m0->m_pkthdr.csum_data = nfrags;
1201
1202         /*
1203          * Update first fragment by trimming what's been copied out
1204          * and updating header.
1205          */
1206         m_adj(m0, hlen + firstlen - ip->ip_len);
1207         m0->m_pkthdr.len = hlen + firstlen;
1208         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1209         ip->ip_off |= IP_MF;
1210         ip->ip_off = htons(ip->ip_off);
1211         ip->ip_sum = 0;
1212         if (sw_csum & CSUM_DELAY_IP)
1213                 ip->ip_sum = in_cksum(m0, hlen);
1214
1215 done:
1216         *m_frag = m0;
1217         return error;
1218 }
1219
1220 void
1221 in_delayed_cksum(struct mbuf *m)
1222 {
1223         struct ip *ip;
1224         u_short csum, offset;
1225
1226         ip = mtod(m, struct ip *);
1227         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1228         csum = in_cksum_skip(m, ip->ip_len, offset);
1229         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1230                 csum = 0xffff;
1231         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1232
1233         if (offset + sizeof(u_short) > m->m_len) {
1234                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1235                     m->m_len, offset, ip->ip_p);
1236                 /*
1237                  * XXX
1238                  * this shouldn't happen, but if it does, the
1239                  * correct behavior may be to insert the checksum
1240                  * in the existing chain instead of rearranging it.
1241                  */
1242                 m = m_pullup(m, offset + sizeof(u_short));
1243         }
1244         *(u_short *)(m->m_data + offset) = csum;
1245 }
1246
1247 /*
1248  * Insert IP options into preformed packet.
1249  * Adjust IP destination as required for IP source routing,
1250  * as indicated by a non-zero in_addr at the start of the options.
1251  *
1252  * XXX This routine assumes that the packet has no options in place.
1253  */
1254 static struct mbuf *
1255 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1256 {
1257         struct ipoption *p = mtod(opt, struct ipoption *);
1258         struct mbuf *n;
1259         struct ip *ip = mtod(m, struct ip *);
1260         unsigned optlen;
1261
1262         optlen = opt->m_len - sizeof p->ipopt_dst;
1263         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1264                 *phlen = 0;
1265                 return (m);             /* XXX should fail */
1266         }
1267         if (p->ipopt_dst.s_addr)
1268                 ip->ip_dst = p->ipopt_dst;
1269         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1270                 MGETHDR(n, M_NOWAIT, MT_HEADER);
1271                 if (n == NULL) {
1272                         *phlen = 0;
1273                         return (m);
1274                 }
1275                 n->m_pkthdr.rcvif = NULL;
1276                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1277                 m->m_len -= sizeof(struct ip);
1278                 m->m_data += sizeof(struct ip);
1279                 n->m_next = m;
1280                 m = n;
1281                 m->m_len = optlen + sizeof(struct ip);
1282                 m->m_data += max_linkhdr;
1283                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1284         } else {
1285                 m->m_data -= optlen;
1286                 m->m_len += optlen;
1287                 m->m_pkthdr.len += optlen;
1288                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1289         }
1290         ip = mtod(m, struct ip *);
1291         bcopy(p->ipopt_list, ip + 1, optlen);
1292         *phlen = sizeof(struct ip) + optlen;
1293         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1294         ip->ip_len += optlen;
1295         return (m);
1296 }
1297
1298 /*
1299  * Copy options from ip to jp,
1300  * omitting those not copied during fragmentation.
1301  */
1302 int
1303 ip_optcopy(struct ip *ip, struct ip *jp)
1304 {
1305         u_char *cp, *dp;
1306         int opt, optlen, cnt;
1307
1308         cp = (u_char *)(ip + 1);
1309         dp = (u_char *)(jp + 1);
1310         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1311         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1312                 opt = cp[0];
1313                 if (opt == IPOPT_EOL)
1314                         break;
1315                 if (opt == IPOPT_NOP) {
1316                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1317                         *dp++ = IPOPT_NOP;
1318                         optlen = 1;
1319                         continue;
1320                 }
1321
1322                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1323                     ("ip_optcopy: malformed ipv4 option"));
1324                 optlen = cp[IPOPT_OLEN];
1325                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1326                     ("ip_optcopy: malformed ipv4 option"));
1327
1328                 /* bogus lengths should have been caught by ip_dooptions */
1329                 if (optlen > cnt)
1330                         optlen = cnt;
1331                 if (IPOPT_COPIED(opt)) {
1332                         bcopy(cp, dp, optlen);
1333                         dp += optlen;
1334                 }
1335         }
1336         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1337                 *dp++ = IPOPT_EOL;
1338         return (optlen);
1339 }
1340
1341 /*
1342  * IP socket option processing.
1343  */
1344 void
1345 ip_ctloutput(netmsg_t msg)
1346 {
1347         struct socket *so = msg->base.nm_so;
1348         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1349         struct  inpcb *inp = so->so_pcb;
1350         int     error, optval;
1351
1352         error = optval = 0;
1353
1354         /* Get socket's owner cpuid hint */
1355         if (sopt->sopt_level == SOL_SOCKET &&
1356             sopt->sopt_dir == SOPT_GET &&
1357             sopt->sopt_name == SO_CPUHINT) {
1358                 optval = mycpuid;
1359                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1360                 goto done;
1361         }
1362
1363         if (sopt->sopt_level != IPPROTO_IP) {
1364                 error = EINVAL;
1365                 goto done;
1366         }
1367
1368         switch (sopt->sopt_name) {
1369         case IP_MULTICAST_IF:
1370         case IP_MULTICAST_VIF:
1371         case IP_MULTICAST_TTL:
1372         case IP_MULTICAST_LOOP:
1373         case IP_ADD_MEMBERSHIP:
1374         case IP_DROP_MEMBERSHIP:
1375                 /*
1376                  * Handle multicast options in netisr0
1377                  */
1378                 if (&curthread->td_msgport != netisr_cpuport(0)) {
1379                         /* NOTE: so_port MUST NOT be checked in netisr0 */
1380                         msg->lmsg.ms_flags |= MSGF_IGNSOPORT;
1381                         lwkt_forwardmsg(netisr_cpuport(0), &msg->lmsg);
1382                         return;
1383                 }
1384                 break;
1385         }
1386
1387         switch (sopt->sopt_dir) {
1388         case SOPT_SET:
1389                 switch (sopt->sopt_name) {
1390                 case IP_OPTIONS:
1391 #ifdef notyet
1392                 case IP_RETOPTS:
1393 #endif
1394                 {
1395                         struct mbuf *m;
1396                         if (sopt->sopt_valsize > MLEN) {
1397                                 error = EMSGSIZE;
1398                                 break;
1399                         }
1400                         MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_HEADER);
1401                         if (m == NULL) {
1402                                 error = ENOBUFS;
1403                                 break;
1404                         }
1405                         m->m_len = sopt->sopt_valsize;
1406                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1407                                               m->m_len);
1408                         error = ip_pcbopts(sopt->sopt_name,
1409                                            &inp->inp_options, m);
1410                         goto done;
1411                 }
1412
1413                 case IP_TOS:
1414                 case IP_TTL:
1415                 case IP_MINTTL:
1416                 case IP_RECVOPTS:
1417                 case IP_RECVRETOPTS:
1418                 case IP_RECVDSTADDR:
1419                 case IP_RECVIF:
1420                 case IP_RECVTTL:
1421                 case IP_FAITH:
1422                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1423                                              sizeof optval);
1424                         if (error)
1425                                 break;
1426                         switch (sopt->sopt_name) {
1427                         case IP_TOS:
1428                                 inp->inp_ip_tos = optval;
1429                                 break;
1430
1431                         case IP_TTL:
1432                                 inp->inp_ip_ttl = optval;
1433                                 break;
1434                         case IP_MINTTL:
1435                                 if (optval >= 0 && optval <= MAXTTL)
1436                                         inp->inp_ip_minttl = optval;
1437                                 else
1438                                         error = EINVAL;
1439                                 break;
1440 #define OPTSET(bit) \
1441         if (optval) \
1442                 inp->inp_flags |= bit; \
1443         else \
1444                 inp->inp_flags &= ~bit;
1445
1446                         case IP_RECVOPTS:
1447                                 OPTSET(INP_RECVOPTS);
1448                                 break;
1449
1450                         case IP_RECVRETOPTS:
1451                                 OPTSET(INP_RECVRETOPTS);
1452                                 break;
1453
1454                         case IP_RECVDSTADDR:
1455                                 OPTSET(INP_RECVDSTADDR);
1456                                 break;
1457
1458                         case IP_RECVIF:
1459                                 OPTSET(INP_RECVIF);
1460                                 break;
1461
1462                         case IP_RECVTTL:
1463                                 OPTSET(INP_RECVTTL);
1464                                 break;
1465
1466                         case IP_FAITH:
1467                                 OPTSET(INP_FAITH);
1468                                 break;
1469                         }
1470                         break;
1471 #undef OPTSET
1472
1473                 case IP_MULTICAST_IF:
1474                 case IP_MULTICAST_VIF:
1475                 case IP_MULTICAST_TTL:
1476                 case IP_MULTICAST_LOOP:
1477                 case IP_ADD_MEMBERSHIP:
1478                 case IP_DROP_MEMBERSHIP:
1479                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1480                         break;
1481
1482                 case IP_PORTRANGE:
1483                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1484                                             sizeof optval);
1485                         if (error)
1486                                 break;
1487
1488                         switch (optval) {
1489                         case IP_PORTRANGE_DEFAULT:
1490                                 inp->inp_flags &= ~(INP_LOWPORT);
1491                                 inp->inp_flags &= ~(INP_HIGHPORT);
1492                                 break;
1493
1494                         case IP_PORTRANGE_HIGH:
1495                                 inp->inp_flags &= ~(INP_LOWPORT);
1496                                 inp->inp_flags |= INP_HIGHPORT;
1497                                 break;
1498
1499                         case IP_PORTRANGE_LOW:
1500                                 inp->inp_flags &= ~(INP_HIGHPORT);
1501                                 inp->inp_flags |= INP_LOWPORT;
1502                                 break;
1503
1504                         default:
1505                                 error = EINVAL;
1506                                 break;
1507                         }
1508                         break;
1509
1510 #if defined(IPSEC) || defined(FAST_IPSEC)
1511                 case IP_IPSEC_POLICY:
1512                 {
1513                         caddr_t req;
1514                         size_t len = 0;
1515                         int priv;
1516                         struct mbuf *m;
1517                         int optname;
1518
1519                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1520                                 break;
1521                         soopt_to_mbuf(sopt, m);
1522                         priv = (sopt->sopt_td != NULL &&
1523                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1524                         req = mtod(m, caddr_t);
1525                         len = m->m_len;
1526                         optname = sopt->sopt_name;
1527                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1528                         m_freem(m);
1529                         break;
1530                 }
1531 #endif /*IPSEC*/
1532
1533                 default:
1534                         error = ENOPROTOOPT;
1535                         break;
1536                 }
1537                 break;
1538
1539         case SOPT_GET:
1540                 switch (sopt->sopt_name) {
1541                 case IP_OPTIONS:
1542                 case IP_RETOPTS:
1543                         if (inp->inp_options)
1544                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1545                                                            char *),
1546                                                 inp->inp_options->m_len);
1547                         else
1548                                 sopt->sopt_valsize = 0;
1549                         break;
1550
1551                 case IP_TOS:
1552                 case IP_TTL:
1553                 case IP_MINTTL:
1554                 case IP_RECVOPTS:
1555                 case IP_RECVRETOPTS:
1556                 case IP_RECVDSTADDR:
1557                 case IP_RECVTTL:
1558                 case IP_RECVIF:
1559                 case IP_PORTRANGE:
1560                 case IP_FAITH:
1561                         switch (sopt->sopt_name) {
1562
1563                         case IP_TOS:
1564                                 optval = inp->inp_ip_tos;
1565                                 break;
1566
1567                         case IP_TTL:
1568                                 optval = inp->inp_ip_ttl;
1569                                 break;
1570                         case IP_MINTTL:
1571                                 optval = inp->inp_ip_minttl;
1572                                 break;
1573
1574 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1575
1576                         case IP_RECVOPTS:
1577                                 optval = OPTBIT(INP_RECVOPTS);
1578                                 break;
1579
1580                         case IP_RECVRETOPTS:
1581                                 optval = OPTBIT(INP_RECVRETOPTS);
1582                                 break;
1583
1584                         case IP_RECVDSTADDR:
1585                                 optval = OPTBIT(INP_RECVDSTADDR);
1586                                 break;
1587
1588                         case IP_RECVTTL:
1589                                 optval = OPTBIT(INP_RECVTTL);
1590                                 break;
1591
1592                         case IP_RECVIF:
1593                                 optval = OPTBIT(INP_RECVIF);
1594                                 break;
1595
1596                         case IP_PORTRANGE:
1597                                 if (inp->inp_flags & INP_HIGHPORT)
1598                                         optval = IP_PORTRANGE_HIGH;
1599                                 else if (inp->inp_flags & INP_LOWPORT)
1600                                         optval = IP_PORTRANGE_LOW;
1601                                 else
1602                                         optval = 0;
1603                                 break;
1604
1605                         case IP_FAITH:
1606                                 optval = OPTBIT(INP_FAITH);
1607                                 break;
1608                         }
1609                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1610                         break;
1611
1612                 case IP_MULTICAST_IF:
1613                 case IP_MULTICAST_VIF:
1614                 case IP_MULTICAST_TTL:
1615                 case IP_MULTICAST_LOOP:
1616                 case IP_ADD_MEMBERSHIP:
1617                 case IP_DROP_MEMBERSHIP:
1618                         error = ip_getmoptions(sopt, inp->inp_moptions);
1619                         break;
1620
1621 #if defined(IPSEC) || defined(FAST_IPSEC)
1622                 case IP_IPSEC_POLICY:
1623                 {
1624                         struct mbuf *m = NULL;
1625                         caddr_t req = NULL;
1626                         size_t len = 0;
1627
1628                         if (m != NULL) {
1629                                 req = mtod(m, caddr_t);
1630                                 len = m->m_len;
1631                         }
1632                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1633                         if (error == 0)
1634                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1635                         if (error == 0)
1636                                 m_freem(m);
1637                         break;
1638                 }
1639 #endif /*IPSEC*/
1640
1641                 default:
1642                         error = ENOPROTOOPT;
1643                         break;
1644                 }
1645                 break;
1646         }
1647 done:
1648         lwkt_replymsg(&msg->lmsg, error);
1649 }
1650
1651 /*
1652  * Set up IP options in pcb for insertion in output packets.
1653  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1654  * with destination address if source routed.
1655  */
1656 static int
1657 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1658 {
1659         int cnt, optlen;
1660         u_char *cp;
1661         u_char opt;
1662
1663         /* turn off any old options */
1664         if (*pcbopt)
1665                 m_free(*pcbopt);
1666         *pcbopt = NULL;
1667         if (m == NULL || m->m_len == 0) {
1668                 /*
1669                  * Only turning off any previous options.
1670                  */
1671                 if (m != NULL)
1672                         m_free(m);
1673                 return (0);
1674         }
1675
1676         if (m->m_len % sizeof(int32_t))
1677                 goto bad;
1678         /*
1679          * IP first-hop destination address will be stored before
1680          * actual options; move other options back
1681          * and clear it when none present.
1682          */
1683         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1684                 goto bad;
1685         cnt = m->m_len;
1686         m->m_len += sizeof(struct in_addr);
1687         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1688         ovbcopy(mtod(m, caddr_t), cp, cnt);
1689         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1690
1691         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1692                 opt = cp[IPOPT_OPTVAL];
1693                 if (opt == IPOPT_EOL)
1694                         break;
1695                 if (opt == IPOPT_NOP)
1696                         optlen = 1;
1697                 else {
1698                         if (cnt < IPOPT_OLEN + sizeof *cp)
1699                                 goto bad;
1700                         optlen = cp[IPOPT_OLEN];
1701                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1702                                 goto bad;
1703                 }
1704                 switch (opt) {
1705
1706                 default:
1707                         break;
1708
1709                 case IPOPT_LSRR:
1710                 case IPOPT_SSRR:
1711                         /*
1712                          * user process specifies route as:
1713                          *      ->A->B->C->D
1714                          * D must be our final destination (but we can't
1715                          * check that since we may not have connected yet).
1716                          * A is first hop destination, which doesn't appear in
1717                          * actual IP option, but is stored before the options.
1718                          */
1719                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1720                                 goto bad;
1721                         m->m_len -= sizeof(struct in_addr);
1722                         cnt -= sizeof(struct in_addr);
1723                         optlen -= sizeof(struct in_addr);
1724                         cp[IPOPT_OLEN] = optlen;
1725                         /*
1726                          * Move first hop before start of options.
1727                          */
1728                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1729                               sizeof(struct in_addr));
1730                         /*
1731                          * Then copy rest of options back
1732                          * to close up the deleted entry.
1733                          */
1734                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1735                                 &cp[IPOPT_OFFSET+1],
1736                                 cnt - (IPOPT_MINOFF - 1));
1737                         break;
1738                 }
1739         }
1740         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1741                 goto bad;
1742         *pcbopt = m;
1743         return (0);
1744
1745 bad:
1746         m_free(m);
1747         return (EINVAL);
1748 }
1749
1750 /*
1751  * XXX
1752  * The whole multicast option thing needs to be re-thought.
1753  * Several of these options are equally applicable to non-multicast
1754  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1755  * standard option (IP_TTL).
1756  */
1757
1758 /*
1759  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1760  */
1761 static struct ifnet *
1762 ip_multicast_if(struct in_addr *a, int *ifindexp)
1763 {
1764         int ifindex;
1765         struct ifnet *ifp;
1766
1767         if (ifindexp)
1768                 *ifindexp = 0;
1769         if (ntohl(a->s_addr) >> 24 == 0) {
1770                 ifindex = ntohl(a->s_addr) & 0xffffff;
1771                 if (ifindex < 0 || if_index < ifindex)
1772                         return NULL;
1773                 ifp = ifindex2ifnet[ifindex];
1774                 if (ifindexp)
1775                         *ifindexp = ifindex;
1776         } else {
1777                 ifp = INADDR_TO_IFP(a);
1778         }
1779         return ifp;
1780 }
1781
1782 /*
1783  * Set the IP multicast options in response to user setsockopt().
1784  */
1785 static int
1786 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1787 {
1788         int error = 0;
1789         int i;
1790         struct in_addr addr;
1791         struct ip_mreq mreq;
1792         struct ifnet *ifp;
1793         struct ip_moptions *imo = *imop;
1794         int ifindex;
1795
1796         if (imo == NULL) {
1797                 /*
1798                  * No multicast option buffer attached to the pcb;
1799                  * allocate one and initialize to default values.
1800                  */
1801                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1802
1803                 imo->imo_multicast_ifp = NULL;
1804                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1805                 imo->imo_multicast_vif = -1;
1806                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1807                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1808                 imo->imo_num_memberships = 0;
1809                 /* Assign imo to imop after all fields are setup */
1810                 cpu_sfence();
1811                 *imop = imo;
1812         }
1813         switch (sopt->sopt_name) {
1814         /* store an index number for the vif you wanna use in the send */
1815         case IP_MULTICAST_VIF:
1816                 if (legal_vif_num == 0) {
1817                         error = EOPNOTSUPP;
1818                         break;
1819                 }
1820                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1821                 if (error)
1822                         break;
1823                 if (!legal_vif_num(i) && (i != -1)) {
1824                         error = EINVAL;
1825                         break;
1826                 }
1827                 imo->imo_multicast_vif = i;
1828                 break;
1829
1830         case IP_MULTICAST_IF:
1831                 /*
1832                  * Select the interface for outgoing multicast packets.
1833                  */
1834                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1835                 if (error)
1836                         break;
1837
1838                 /*
1839                  * INADDR_ANY is used to remove a previous selection.
1840                  * When no interface is selected, a default one is
1841                  * chosen every time a multicast packet is sent.
1842                  */
1843                 if (addr.s_addr == INADDR_ANY) {
1844                         imo->imo_multicast_ifp = NULL;
1845                         break;
1846                 }
1847                 /*
1848                  * The selected interface is identified by its local
1849                  * IP address.  Find the interface and confirm that
1850                  * it supports multicasting.
1851                  */
1852                 crit_enter();
1853                 ifp = ip_multicast_if(&addr, &ifindex);
1854                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1855                         crit_exit();
1856                         error = EADDRNOTAVAIL;
1857                         break;
1858                 }
1859                 imo->imo_multicast_ifp = ifp;
1860                 if (ifindex)
1861                         imo->imo_multicast_addr = addr;
1862                 else
1863                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1864                 crit_exit();
1865                 break;
1866
1867         case IP_MULTICAST_TTL:
1868                 /*
1869                  * Set the IP time-to-live for outgoing multicast packets.
1870                  * The original multicast API required a char argument,
1871                  * which is inconsistent with the rest of the socket API.
1872                  * We allow either a char or an int.
1873                  */
1874                 if (sopt->sopt_valsize == 1) {
1875                         u_char ttl;
1876                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1877                         if (error)
1878                                 break;
1879                         imo->imo_multicast_ttl = ttl;
1880                 } else {
1881                         u_int ttl;
1882                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1883                         if (error)
1884                                 break;
1885                         if (ttl > 255)
1886                                 error = EINVAL;
1887                         else
1888                                 imo->imo_multicast_ttl = ttl;
1889                 }
1890                 break;
1891
1892         case IP_MULTICAST_LOOP:
1893                 /*
1894                  * Set the loopback flag for outgoing multicast packets.
1895                  * Must be zero or one.  The original multicast API required a
1896                  * char argument, which is inconsistent with the rest
1897                  * of the socket API.  We allow either a char or an int.
1898                  */
1899                 if (sopt->sopt_valsize == 1) {
1900                         u_char loop;
1901
1902                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1903                         if (error)
1904                                 break;
1905                         imo->imo_multicast_loop = !!loop;
1906                 } else {
1907                         u_int loop;
1908
1909                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1910                                             sizeof loop);
1911                         if (error)
1912                                 break;
1913                         imo->imo_multicast_loop = !!loop;
1914                 }
1915                 break;
1916
1917         case IP_ADD_MEMBERSHIP:
1918                 /*
1919                  * Add a multicast group membership.
1920                  * Group must be a valid IP multicast address.
1921                  */
1922                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1923                 if (error)
1924                         break;
1925
1926                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1927                         error = EINVAL;
1928                         break;
1929                 }
1930                 crit_enter();
1931                 /*
1932                  * If no interface address was provided, use the interface of
1933                  * the route to the given multicast address.
1934                  */
1935                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1936                         struct sockaddr_in dst;
1937                         struct rtentry *rt;
1938
1939                         bzero(&dst, sizeof(struct sockaddr_in));
1940                         dst.sin_len = sizeof(struct sockaddr_in);
1941                         dst.sin_family = AF_INET;
1942                         dst.sin_addr = mreq.imr_multiaddr;
1943                         rt = rtlookup((struct sockaddr *)&dst);
1944                         if (rt == NULL) {
1945                                 error = EADDRNOTAVAIL;
1946                                 crit_exit();
1947                                 break;
1948                         }
1949                         --rt->rt_refcnt;
1950                         ifp = rt->rt_ifp;
1951                 } else {
1952                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1953                 }
1954
1955                 /*
1956                  * See if we found an interface, and confirm that it
1957                  * supports multicast.
1958                  */
1959                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1960                         error = EADDRNOTAVAIL;
1961                         crit_exit();
1962                         break;
1963                 }
1964                 /*
1965                  * See if the membership already exists or if all the
1966                  * membership slots are full.
1967                  */
1968                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1969                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1970                             imo->imo_membership[i]->inm_addr.s_addr
1971                                                 == mreq.imr_multiaddr.s_addr)
1972                                 break;
1973                 }
1974                 if (i < imo->imo_num_memberships) {
1975                         error = EADDRINUSE;
1976                         crit_exit();
1977                         break;
1978                 }
1979                 if (i == IP_MAX_MEMBERSHIPS) {
1980                         error = ETOOMANYREFS;
1981                         crit_exit();
1982                         break;
1983                 }
1984                 /*
1985                  * Everything looks good; add a new record to the multicast
1986                  * address list for the given interface.
1987                  */
1988                 if ((imo->imo_membership[i] =
1989                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1990                         error = ENOBUFS;
1991                         crit_exit();
1992                         break;
1993                 }
1994                 ++imo->imo_num_memberships;
1995                 crit_exit();
1996                 break;
1997
1998         case IP_DROP_MEMBERSHIP:
1999                 /*
2000                  * Drop a multicast group membership.
2001                  * Group must be a valid IP multicast address.
2002                  */
2003                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2004                 if (error)
2005                         break;
2006
2007                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2008                         error = EINVAL;
2009                         break;
2010                 }
2011
2012                 crit_enter();
2013                 /*
2014                  * If an interface address was specified, get a pointer
2015                  * to its ifnet structure.
2016                  */
2017                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2018                         ifp = NULL;
2019                 else {
2020                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2021                         if (ifp == NULL) {
2022                                 error = EADDRNOTAVAIL;
2023                                 crit_exit();
2024                                 break;
2025                         }
2026                 }
2027                 /*
2028                  * Find the membership in the membership array.
2029                  */
2030                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2031                         if ((ifp == NULL ||
2032                              imo->imo_membership[i]->inm_ifp == ifp) &&
2033                             imo->imo_membership[i]->inm_addr.s_addr ==
2034                             mreq.imr_multiaddr.s_addr)
2035                                 break;
2036                 }
2037                 if (i == imo->imo_num_memberships) {
2038                         error = EADDRNOTAVAIL;
2039                         crit_exit();
2040                         break;
2041                 }
2042                 /*
2043                  * Give up the multicast address record to which the
2044                  * membership points.
2045                  */
2046                 in_delmulti(imo->imo_membership[i]);
2047                 /*
2048                  * Remove the gap in the membership array.
2049                  */
2050                 for (++i; i < imo->imo_num_memberships; ++i)
2051                         imo->imo_membership[i-1] = imo->imo_membership[i];
2052                 --imo->imo_num_memberships;
2053                 crit_exit();
2054                 break;
2055
2056         default:
2057                 error = EOPNOTSUPP;
2058                 break;
2059         }
2060
2061         return (error);
2062 }
2063
2064 /*
2065  * Return the IP multicast options in response to user getsockopt().
2066  */
2067 static int
2068 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2069 {
2070         struct in_addr addr;
2071         struct in_ifaddr *ia;
2072         int error, optval;
2073         u_char coptval;
2074
2075         error = 0;
2076         switch (sopt->sopt_name) {
2077         case IP_MULTICAST_VIF:
2078                 if (imo != NULL)
2079                         optval = imo->imo_multicast_vif;
2080                 else
2081                         optval = -1;
2082                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2083                 break;
2084
2085         case IP_MULTICAST_IF:
2086                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2087                         addr.s_addr = INADDR_ANY;
2088                 else if (imo->imo_multicast_addr.s_addr) {
2089                         /* return the value user has set */
2090                         addr = imo->imo_multicast_addr;
2091                 } else {
2092                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2093                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2094                                 : IA_SIN(ia)->sin_addr.s_addr;
2095                 }
2096                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2097                 break;
2098
2099         case IP_MULTICAST_TTL:
2100                 if (imo == NULL)
2101                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2102                 else
2103                         optval = coptval = imo->imo_multicast_ttl;
2104                 if (sopt->sopt_valsize == 1)
2105                         soopt_from_kbuf(sopt, &coptval, 1);
2106                 else
2107                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2108                 break;
2109
2110         case IP_MULTICAST_LOOP:
2111                 if (imo == NULL)
2112                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2113                 else
2114                         optval = coptval = imo->imo_multicast_loop;
2115                 if (sopt->sopt_valsize == 1)
2116                         soopt_from_kbuf(sopt, &coptval, 1);
2117                 else
2118                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2119                 break;
2120
2121         default:
2122                 error = ENOPROTOOPT;
2123                 break;
2124         }
2125         return (error);
2126 }
2127
2128 /*
2129  * Discard the IP multicast options.
2130  */
2131 void
2132 ip_freemoptions(struct ip_moptions *imo)
2133 {
2134         int i;
2135
2136         if (imo != NULL) {
2137                 for (i = 0; i < imo->imo_num_memberships; ++i)
2138                         in_delmulti(imo->imo_membership[i]);
2139                 kfree(imo, M_IPMOPTS);
2140         }
2141 }
2142
2143 /*
2144  * Routine called from ip_output() to loop back a copy of an IP multicast
2145  * packet to the input queue of a specified interface.  Note that this
2146  * calls the output routine of the loopback "driver", but with an interface
2147  * pointer that might NOT be a loopback interface -- evil, but easier than
2148  * replicating that code here.
2149  */
2150 static void
2151 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2152              int hlen)
2153 {
2154         struct ip *ip;
2155         struct mbuf *copym;
2156
2157         copym = m_copypacket(m, M_NOWAIT);
2158         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2159                 copym = m_pullup(copym, hlen);
2160         if (copym != NULL) {
2161                 /*
2162                  * if the checksum hasn't been computed, mark it as valid
2163                  */
2164                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2165                         in_delayed_cksum(copym);
2166                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2167                         copym->m_pkthdr.csum_flags |=
2168                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2169                         copym->m_pkthdr.csum_data = 0xffff;
2170                 }
2171                 /*
2172                  * We don't bother to fragment if the IP length is greater
2173                  * than the interface's MTU.  Can this possibly matter?
2174                  */
2175                 ip = mtod(copym, struct ip *);
2176                 ip->ip_len = htons(ip->ip_len);
2177                 ip->ip_off = htons(ip->ip_off);
2178                 ip->ip_sum = 0;
2179                 if (ip->ip_vhl == IP_VHL_BORING) {
2180                         ip->ip_sum = in_cksum_hdr(ip);
2181                 } else {
2182                         ip->ip_sum = in_cksum(copym, hlen);
2183                 }
2184                 /*
2185                  * NB:
2186                  * It's not clear whether there are any lingering
2187                  * reentrancy problems in other areas which might
2188                  * be exposed by using ip_input directly (in
2189                  * particular, everything which modifies the packet
2190                  * in-place).  Yet another option is using the
2191                  * protosw directly to deliver the looped back
2192                  * packet.  For the moment, we'll err on the side
2193                  * of safety by using if_simloop().
2194                  */
2195 #if 1 /* XXX */
2196                 if (dst->sin_family != AF_INET) {
2197                         kprintf("ip_mloopback: bad address family %d\n",
2198                                                 dst->sin_family);
2199                         dst->sin_family = AF_INET;
2200                 }
2201 #endif
2202                 if_simloop(ifp, copym, dst->sin_family, 0);
2203         }
2204 }