sys/netinet/ip_input.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
  36  *
  37  * License terms: all terms for the DragonFly license above plus the following:
  38  *
  39  * 4. All advertising materials mentioning features or use of this software
  40  *    must display the following acknowledgement:
  41  *
  42  *      This product includes software developed by Jeffrey M. Hsu
  43  *      for the DragonFly Project.
  44  *
  45  *    This requirement may be waived with permission from Jeffrey Hsu.
  46  *    This requirement will sunset and may be removed on July 8 2005,
  47  *    after which the standard DragonFly license (as shown above) will
  48  *    apply.
  49  */
  50
  51 /*
  52  * Copyright (c) 1982, 1986, 1988, 1993
  53  *      The Regents of the University of California.  All rights reserved.
  54  *
  55  * Redistribution and use in source and binary forms, with or without
  56  * modification, are permitted provided that the following conditions
  57  * are met:
  58  * 1. Redistributions of source code must retain the above copyright
  59  *    notice, this list of conditions and the following disclaimer.
  60  * 2. Redistributions in binary form must reproduce the above copyright
  61  *    notice, this list of conditions and the following disclaimer in the
  62  *    documentation and/or other materials provided with the distribution.
  63  * 3. All advertising materials mentioning features or use of this software
  64  *    must display the following acknowledgement:
  65  *      This product includes software developed by the University of
  66  *      California, Berkeley and its contributors.
  67  * 4. Neither the name of the University nor the names of its contributors
  68  *    may be used to endorse or promote products derived from this software
  69  *    without specific prior written permission.
  70  *
  71  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  72  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  73  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  74  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  75  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  76  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  77  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  78  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  79  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  80  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  81  * SUCH DAMAGE.
  82  *
  83  *      @(#)ip_input.c  8.2 (Berkeley) 1/4/94
  84  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
  85  * $DragonFly: src/sys/netinet/ip_input.c,v 1.49 2005/04/18 14:26:57 joerg Exp $
  86  */
  87
  88 #define _IP_VHL
  89
  90 #include "opt_bootp.h"
  91 #include "opt_ipfw.h"
  92 #include "opt_ipdn.h"
  93 #include "opt_ipdivert.h"
  94 #include "opt_ipfilter.h"
  95 #include "opt_ipstealth.h"
  96 #include "opt_ipsec.h"
  97 #include "opt_random_ip_id.h"
  98
  99 #include <sys/param.h>
 100 #include <sys/systm.h>
 101 #include <sys/mbuf.h>
 102 #include <sys/malloc.h>
 103 #include <sys/mpipe.h>
 104 #include <sys/domain.h>
 105 #include <sys/protosw.h>
 106 #include <sys/socket.h>
 107 #include <sys/time.h>
 108 #include <sys/globaldata.h>
 109 #include <sys/thread.h>
 110 #include <sys/kernel.h>
 111 #include <sys/syslog.h>
 112 #include <sys/sysctl.h>
 113 #include <sys/in_cksum.h>
 114
 115 #include <sys/thread2.h>
 116 #include <sys/msgport2.h>
 117
 118 #include <machine/stdarg.h>
 119
 120 #include <net/if.h>
 121 #include <net/if_types.h>
 122 #include <net/if_var.h>
 123 #include <net/if_dl.h>
 124 #include <net/pfil.h>
 125 #include <net/route.h>
 126 #include <net/netisr.h>
 127 #include <net/intrq.h>
 128
 129 #include <netinet/in.h>
 130 #include <netinet/in_systm.h>
 131 #include <netinet/in_var.h>
 132 #include <netinet/ip.h>
 133 #include <netinet/in_pcb.h>
 134 #include <netinet/ip_var.h>
 135 #include <netinet/ip_icmp.h>
 136
 137
 138 #include <sys/socketvar.h>
 139
 140 #include <net/ipfw/ip_fw.h>
 141 #include <net/dummynet/ip_dummynet.h>
 142
 143 #ifdef IPSEC
 144 #include <netinet6/ipsec.h>
 145 #include <netproto/key/key.h>
 146 #endif
 147
 148 #ifdef FAST_IPSEC
 149 #include <netproto/ipsec/ipsec.h>
 150 #include <netproto/ipsec/key.h>
 151 #endif
 152
 153 int rsvp_on = 0;
 154 static int ip_rsvp_on;
 155 struct socket *ip_rsvpd;
 156
 157 int ipforwarding = 0;
 158 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 159     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 160
 161 static int ipsendredirects = 1; /* XXX */
 162 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 163     &ipsendredirects, 0, "Enable sending IP redirects");
 164
 165 int ip_defttl = IPDEFTTL;
 166 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 167     &ip_defttl, 0, "Maximum TTL on IP packets");
 168
 169 static int ip_dosourceroute = 0;
 170 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 171     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 172
 173 static int ip_acceptsourceroute = 0;
 174 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 175     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 176     "Enable accepting source routed IP packets");
 177
 178 static int ip_keepfaith = 0;
 179 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 180     &ip_keepfaith, 0,
 181     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
 182
 183 static int nipq = 0;    /* total # of reass queues */
 184 static int maxnipq;
 185 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 186     &maxnipq, 0,
 187     "Maximum number of IPv4 fragment reassembly queue entries");
 188
 189 static int maxfragsperpacket;
 190 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 191     &maxfragsperpacket, 0,
 192     "Maximum number of IPv4 fragments allowed per packet");
 193
 194 static int ip_sendsourcequench = 0;
 195 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 196     &ip_sendsourcequench, 0,
 197     "Enable the transmission of source quench packets");
 198
 199 /*
 200  * XXX - Setting ip_checkinterface mostly implements the receive side of
 201  * the Strong ES model described in RFC 1122, but since the routing table
 202  * and transmit implementation do not implement the Strong ES model,
 203  * setting this to 1 results in an odd hybrid.
 204  *
 205  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 206  * to translate the destination address to another local interface.
 207  *
 208  * XXX - ip_checkinterface must be disabled if you add IP aliases
 209  * to the loopback interface instead of the interface where the
 210  * packets for those addresses are received.
 211  */
 212 static int ip_checkinterface = 0;
 213 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 214     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 215
 216 #ifdef DIAGNOSTIC
 217 static int ipprintfs = 0;
 218 #endif
 219
 220 static struct ifqueue ipintrq;
 221 static int ipqmaxlen = IFQ_MAXLEN;
 222
 223 extern  struct domain inetdomain;
 224 extern  struct protosw inetsw[];
 225 u_char  ip_protox[IPPROTO_MAX];
 226 struct  in_ifaddrhead in_ifaddrhead;            /* first inet address */
 227 struct  in_ifaddrhashhead *in_ifaddrhashtbl;    /* inet addr hash table */
 228 u_long  in_ifaddrhmask;                         /* mask for hash table */
 229
 230 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
 231     &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
 232 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
 233     &ipintrq.ifq_drops, 0, "Number of packets dropped from the IP input queue");
 234
 235 struct ip_stats ipstats_percpu[MAXCPU];
 236 #ifdef SMP
 237 static int
 238 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 239 {
 240         int cpu, error = 0;
 241
 242         for (cpu = 0; cpu < ncpus; ++cpu) {
 243                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 244                                         sizeof(struct ip_stats))))
 245                         break;
 246                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 247                                        sizeof(struct ip_stats))))
 248                         break;
 249         }
 250
 251         return (error);
 252 }
 253 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 254     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 255 #else
 256 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 257     &ipstat, ip_stats, "IP statistics");
 258 #endif
 259
 260 /* Packet reassembly stuff */
 261 #define IPREASS_NHASH_LOG2      6
 262 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 263 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 264 #define IPREASS_HASH(x,y)                                               \
 265     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 266
 267 static struct ipq ipq[IPREASS_NHASH];
 268 const  int    ipintrq_present = 1;
 269
 270 #ifdef IPCTL_DEFMTU
 271 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 272     &ip_mtu, 0, "Default MTU");
 273 #endif
 274
 275 #ifdef IPSTEALTH
 276 static int ipstealth = 0;
 277 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 278 #else
 279 static const int ipstealth = 0;
 280 #endif
 281
 282
 283 /* Firewall hooks */
 284 ip_fw_chk_t *ip_fw_chk_ptr;
 285 int fw_enable = 1;
 286 int fw_one_pass = 1;
 287
 288 /* Dummynet hooks */
 289 ip_dn_io_t *ip_dn_io_ptr;
 290
 291 struct pfil_head inet_pfil_hook;
 292
 293 /*
 294  * XXX this is ugly -- the following two global variables are
 295  * used to store packet state while it travels through the stack.
 296  * Note that the code even makes assumptions on the size and
 297  * alignment of fields inside struct ip_srcrt so e.g. adding some
 298  * fields will break the code. This needs to be fixed.
 299  *
 300  * We need to save the IP options in case a protocol wants to respond
 301  * to an incoming packet over the same route if the packet got here
 302  * using IP source routing.  This allows connection establishment and
 303  * maintenance when the remote end is on a network that is not known
 304  * to us.
 305  */
 306 static int ip_nhops = 0;
 307
 308 static  struct ip_srcrt {
 309         struct  in_addr dst;                    /* final destination */
 310         char    nop;                            /* one NOP to align */
 311         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 312         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 313 } ip_srcrt;
 314
 315 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 316 static struct malloc_pipe ipq_mpipe;
 317
 318 static void             save_rte (u_char *, struct in_addr);
 319 static int              ip_dooptions (struct mbuf *m, int,
 320                                         struct sockaddr_in *next_hop);
 321 static void             ip_forward (struct mbuf *m, int srcrt,
 322                                         struct sockaddr_in *next_hop);
 323 static void             ip_freef (struct ipq *);
 324 static int              ip_input_handler (struct netmsg *);
 325 static struct mbuf      *ip_reass (struct mbuf *, struct ipq *,
 326                                         struct ipq *, u_int32_t *);
 327
 328 /*
 329  * IP initialization: fill in IP protocol switch table.
 330  * All protocols not implemented in kernel go to raw IP protocol handler.
 331  */
 332 void
 333 ip_init(void)
 334 {
 335         struct protosw *pr;
 336         int i;
 337 #ifdef SMP
 338         int cpu;
 339 #endif
 340
 341         /*
 342          * Make sure we can handle a reasonable number of fragments but
 343          * cap it at 4000 (XXX).
 344          */
 345         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 346                     IFQ_MAXLEN, 4000, 0, NULL);
 347         TAILQ_INIT(&in_ifaddrhead);
 348         in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 349         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 350         if (pr == NULL)
 351                 panic("ip_init");
 352         for (i = 0; i < IPPROTO_MAX; i++)
 353                 ip_protox[i] = pr - inetsw;
 354         for (pr = inetdomain.dom_protosw;
 355              pr < inetdomain.dom_protoswNPROTOSW; pr++)
 356                 if (pr->pr_domain->dom_family == PF_INET &&
 357                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
 358                         ip_protox[pr->pr_protocol] = pr - inetsw;
 359
 360         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 361         inet_pfil_hook.ph_af = AF_INET;
 362         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 363                 printf("%s: WARNING: unable to register pfil hook, "
 364                         "error %d\n", __func__, i);
 365         }
 366
 367         for (i = 0; i < IPREASS_NHASH; i++)
 368             ipq[i].next = ipq[i].prev = &ipq[i];
 369
 370         maxnipq = nmbclusters / 32;
 371         maxfragsperpacket = 16;
 372
 373 #ifndef RANDOM_IP_ID
 374         ip_id = time_second & 0xffff;
 375 #endif
 376         ipintrq.ifq_maxlen = ipqmaxlen;
 377
 378         /*
 379          * Initialize IP statistics counters for each CPU.
 380          *
 381          */
 382 #ifdef SMP
 383         for (cpu = 0; cpu < ncpus; ++cpu) {
 384                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 385         }
 386 #else
 387         bzero(&ipstat, sizeof(struct ip_stats));
 388 #endif
 389
 390         netisr_register(NETISR_IP, ip_mport, ip_input_handler);
 391 }
 392
 393 /*
 394  * XXX watch out this one. It is perhaps used as a cache for
 395  * the most recently used route ? it is cleared in in_addroute()
 396  * when a new route is successfully created.
 397  */
 398 struct route ipforward_rt;
 399
 400 /* Do transport protocol processing. */
 401 static void
 402 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip,
 403                            struct sockaddr_in *nexthop)
 404 {
 405         /*
 406          * Switch out to protocol's input routine.
 407          */
 408         if (nexthop && ip->ip_p == IPPROTO_TCP) {
 409                 /* TCP needs IPFORWARD info if available */
 410                 struct m_hdr tag;
 411
 412                 tag.mh_type = MT_TAG;
 413                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 414                 tag.mh_data = (caddr_t)nexthop;
 415                 tag.mh_next = m;
 416
 417                 (*inetsw[ip_protox[ip->ip_p]].pr_input)
 418                     ((struct mbuf *)&tag, hlen, ip->ip_p);
 419         } else {
 420                 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
 421         }
 422 }
 423
 424 struct netmsg_transport_packet {
 425         struct lwkt_msg         nm_lmsg;
 426         struct mbuf             *nm_mbuf;
 427         int                     nm_hlen;
 428         boolean_t               nm_hasnexthop;
 429         struct sockaddr_in      nm_nexthop;
 430 };
 431
 432 static int
 433 transport_processing_handler(lwkt_msg_t lmsg)
 434 {
 435         struct netmsg_transport_packet *msg = (void *)lmsg;
 436         struct sockaddr_in *nexthop;
 437         struct ip *ip;
 438
 439         ip = mtod(msg->nm_mbuf, struct ip *);
 440         nexthop = msg->nm_hasnexthop ? &msg->nm_nexthop : NULL;
 441         transport_processing_oncpu(msg->nm_mbuf, msg->nm_hlen, ip, nexthop);
 442         lwkt_replymsg(lmsg, 0);
 443         return(EASYNC);
 444 }
 445
 446 static int
 447 ip_input_handler(struct netmsg *msg0)
 448 {
 449         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 450
 451         ip_input(m);
 452         lwkt_replymsg(&msg0->nm_lmsg, 0);
 453         return(EASYNC);
 454 }
 455
 456 /*
 457  * IP input routine.  Checksum and byte swap header.  If fragmented
 458  * try to reassemble.  Process options.  Pass to next level.
 459  */
 460 void
 461 ip_input(struct mbuf *m)
 462 {
 463         struct ip *ip;
 464         struct ipq *fp;
 465         struct in_ifaddr *ia = NULL;
 466         struct ifaddr *ifa;
 467         int i, hlen, checkif;
 468         u_short sum;
 469         struct in_addr pkt_dst;
 470         u_int32_t divert_info = 0;              /* packet divert/tee info */
 471         struct ip_fw_args args;
 472         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 473         boolean_t needredispatch = FALSE;
 474         struct in_addr odst;                    /* original dst address(NAT) */
 475 #if defined(FAST_IPSEC) || defined(IPDIVERT)
 476         struct m_tag *mtag;
 477 #endif
 478 #ifdef FAST_IPSEC
 479         struct tdb_ident *tdbi;
 480         struct secpolicy *sp;
 481         int s, error;
 482 #endif
 483
 484         args.eh = NULL;
 485         args.oif = NULL;
 486         args.rule = NULL;
 487         args.next_hop = NULL;
 488
 489         /* Grab info from MT_TAG mbufs prepended to the chain. */
 490         while (m != NULL && m->m_type == MT_TAG) {
 491                 switch(m->_m_tag_id) {
 492                 case PACKET_TAG_DUMMYNET:
 493                         args.rule = ((struct dn_pkt *)m)->rule;
 494                         break;
 495                 case PACKET_TAG_IPFORWARD:
 496                         args.next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
 497                         break;
 498                 default:
 499                         printf("ip_input: unrecognised MT_TAG tag %d\n",
 500                             m->_m_tag_id);
 501                         break;
 502                 }
 503                 m = m->m_next;
 504         }
 505         KASSERT(m != NULL && (m->m_flags & M_PKTHDR), ("ip_input: no HDR"));
 506
 507         if (args.rule != NULL) {        /* dummynet already filtered us */
 508                 ip = mtod(m, struct ip *);
 509                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 510                 goto iphack;
 511         }
 512
 513         ipstat.ips_total++;
 514
 515         /* length checks already done in ip_demux() */
 516         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 517
 518         ip = mtod(m, struct ip *);
 519
 520         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 521                 ipstat.ips_badvers++;
 522                 goto bad;
 523         }
 524
 525         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 526         /* length checks already done in ip_demux() */
 527         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 528         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 529
 530         /* 127/8 must not appear on wire - RFC1122 */
 531         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 532             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 533                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 534                         ipstat.ips_badaddr++;
 535                         goto bad;
 536                 }
 537         }
 538
 539         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 540                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 541         } else {
 542                 if (hlen == sizeof(struct ip)) {
 543                         sum = in_cksum_hdr(ip);
 544                 } else {
 545                         sum = in_cksum(m, hlen);
 546                 }
 547         }
 548         if (sum != 0) {
 549                 ipstat.ips_badsum++;
 550                 goto bad;
 551         }
 552
 553 #ifdef ALTQ
 554         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 555                 /* packet is dropped by traffic conditioner */
 556                 return;
 557         }
 558 #endif
 559         /*
 560          * Convert fields to host representation.
 561          */
 562         ip->ip_len = ntohs(ip->ip_len);
 563         if (ip->ip_len < hlen) {
 564                 ipstat.ips_badlen++;
 565                 goto bad;
 566         }
 567         ip->ip_off = ntohs(ip->ip_off);
 568
 569         /*
 570          * Check that the amount of data in the buffers
 571          * is as at least much as the IP header would have us expect.
 572          * Trim mbufs if longer than we expect.
 573          * Drop packet if shorter than we expect.
 574          */
 575         if (m->m_pkthdr.len < ip->ip_len) {
 576                 ipstat.ips_tooshort++;
 577                 goto bad;
 578         }
 579         if (m->m_pkthdr.len > ip->ip_len) {
 580                 if (m->m_len == m->m_pkthdr.len) {
 581                         m->m_len = ip->ip_len;
 582                         m->m_pkthdr.len = ip->ip_len;
 583                 } else
 584                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 585         }
 586 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 587         /*
 588          * Bypass packet filtering for packets from a tunnel (gif).
 589          */
 590         if (ipsec_gethist(m, NULL))
 591                 goto pass;
 592 #endif
 593
 594         /*
 595          * IpHack's section.
 596          * Right now when no processing on packet has done
 597          * and it is still fresh out of network we do our black
 598          * deals with it.
 599          * - Firewall: deny/allow/divert
 600          * - Xlate: translate packet's addr/port (NAT).
 601          * - Pipe: pass pkt through dummynet.
 602          * - Wrap: fake packet's addr/port <unimpl.>
 603          * - Encapsulate: put it in another IP and send out. <unimp.>
 604          */
 605
 606 iphack:
 607
 608         /*
 609          * Run through list of hooks for input packets.
 610          *
 611          * NB: Beware of the destination address changing (e.g.
 612          *     by NAT rewriting). When this happens, tell
 613          *     ip_forward to do the right thing.
 614          */
 615         if (pfil_has_hooks(&inet_pfil_hook)) {
 616                 odst = ip->ip_dst;
 617                 if (pfil_run_hooks(&inet_pfil_hook, &m,
 618                     m->m_pkthdr.rcvif, PFIL_IN)) {
 619                         return;
 620                 }
 621                 if (m == NULL)                  /* consumed by filter */
 622                         return;
 623                 ip = mtod(m, struct ip *);
 624                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 625         }
 626
 627         if (fw_enable && IPFW_LOADED) {
 628                 /*
 629                  * If we've been forwarded from the output side, then
 630                  * skip the firewall a second time
 631                  */
 632                 if (args.next_hop != NULL)
 633                         goto ours;
 634
 635                 args.m = m;
 636                 i = ip_fw_chk_ptr(&args);
 637                 m = args.m;
 638
 639                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
 640                         if (m != NULL)
 641                                 m_freem(m);
 642                         return;
 643                 }
 644                 ip = mtod(m, struct ip *);      /* just in case m changed */
 645                 if (i == 0 && args.next_hop == NULL)    /* common case */
 646                         goto pass;
 647                 if (DUMMYNET_LOADED && (i & IP_FW_PORT_DYNT_FLAG)) {
 648                         /* Send packet to the appropriate pipe */
 649                         ip_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
 650                         return;
 651                 }
 652 #ifdef IPDIVERT
 653                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
 654                         /* Divert or tee packet */
 655                         divert_info = i;
 656                         goto ours;
 657                 }
 658 #endif
 659                 if (i == 0 && args.next_hop != NULL)
 660                         goto pass;
 661                 /*
 662                  * if we get here, the packet must be dropped
 663                  */
 664                 m_freem(m);
 665                 return;
 666         }
 667 pass:
 668
 669         /*
 670          * Process options and, if not destined for us,
 671          * ship it on.  ip_dooptions returns 1 when an
 672          * error was detected (causing an icmp message
 673          * to be sent and the original packet to be freed).
 674          */
 675         ip_nhops = 0;           /* for source routed packets */
 676         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, args.next_hop))
 677                 return;
 678
 679         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 680          * matter if it is destined to another node, or whether it is
 681          * a multicast one, RSVP wants it! and prevents it from being forwarded
 682          * anywhere else. Also checks if the rsvp daemon is running before
 683          * grabbing the packet.
 684          */
 685         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 686                 goto ours;
 687
 688         /*
 689          * Check our list of addresses, to see if the packet is for us.
 690          * If we don't have any addresses, assume any unicast packet
 691          * we receive might be for us (and let the upper layers deal
 692          * with it).
 693          */
 694         if (TAILQ_EMPTY(&in_ifaddrhead) && !(m->m_flags & (M_MCAST | M_BCAST)))
 695                 goto ours;
 696
 697         /*
 698          * Cache the destination address of the packet; this may be
 699          * changed by use of 'ipfw fwd'.
 700          */
 701         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 702
 703         /*
 704          * Enable a consistency check between the destination address
 705          * and the arrival interface for a unicast packet (the RFC 1122
 706          * strong ES model) if IP forwarding is disabled and the packet
 707          * is not locally generated and the packet is not subject to
 708          * 'ipfw fwd'.
 709          *
 710          * XXX - Checking also should be disabled if the destination
 711          * address is ipnat'ed to a different interface.
 712          *
 713          * XXX - Checking is incompatible with IP aliases added
 714          * to the loopback interface instead of the interface where
 715          * the packets are received.
 716          */
 717         checkif = ip_checkinterface &&
 718                   !ipforwarding &&
 719                   m->m_pkthdr.rcvif != NULL &&
 720                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 721                   (args.next_hop == NULL);
 722
 723         /*
 724          * Check for exact addresses in the hash bucket.
 725          */
 726         LIST_FOREACH(ia, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 727                 /*
 728                  * If the address matches, verify that the packet
 729                  * arrived via the correct interface if checking is
 730                  * enabled.
 731                  */
 732                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 733                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 734                         goto ours;
 735         }
 736         /*
 737          * Check for broadcast addresses.
 738          *
 739          * Only accept broadcast packets that arrive via the matching
 740          * interface.  Reception of forwarded directed broadcasts would
 741          * be handled via ip_forward() and ether_output() with the loopback
 742          * into the stack for SIMPLEX interfaces handled by ether_output().
 743          */
 744         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 745                 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
 746                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 747                                 continue;
 748                         if (ifa->ifa_addr->sa_family != AF_INET)
 749                                 continue;
 750                         ia = ifatoia(ifa);
 751                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 752                                                                 pkt_dst.s_addr)
 753                                 goto ours;
 754                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 755                                 goto ours;
 756 #ifdef BOOTP_COMPAT
 757                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 758                                 goto ours;
 759 #endif
 760                 }
 761         }
 762         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 763                 struct in_multi *inm;
 764
 765                 if (ip_mrouter != NULL) {
 766                         /*
 767                          * If we are acting as a multicast router, all
 768                          * incoming multicast packets are passed to the
 769                          * kernel-level multicast forwarding function.
 770                          * The packet is returned (relatively) intact; if
 771                          * ip_mforward() returns a non-zero value, the packet
 772                          * must be discarded, else it may be accepted below.
 773                          */
 774                         if (ip_mforward != NULL &&
 775                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 776                                 ipstat.ips_cantforward++;
 777                                 m_freem(m);
 778                                 return;
 779                         }
 780
 781                         /*
 782                          * The process-level routing daemon needs to receive
 783                          * all multicast IGMP packets, whether or not this
 784                          * host belongs to their destination groups.
 785                          */
 786                         if (ip->ip_p == IPPROTO_IGMP)
 787                                 goto ours;
 788                         ipstat.ips_forward++;
 789                 }
 790                 /*
 791                  * See if we belong to the destination multicast group on the
 792                  * arrival interface.
 793                  */
 794                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 795                 if (inm == NULL) {
 796                         ipstat.ips_notmember++;
 797                         m_freem(m);
 798                         return;
 799                 }
 800                 goto ours;
 801         }
 802         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 803                 goto ours;
 804         if (ip->ip_dst.s_addr == INADDR_ANY)
 805                 goto ours;
 806
 807         /*
 808          * FAITH(Firewall Aided Internet Translator)
 809          */
 810         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 811                 if (ip_keepfaith) {
 812                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 813                                 goto ours;
 814                 }
 815                 m_freem(m);
 816                 return;
 817         }
 818
 819         /*
 820          * Not for us; forward if possible and desirable.
 821          */
 822         if (!ipforwarding) {
 823                 ipstat.ips_cantforward++;
 824                 m_freem(m);
 825         } else {
 826 #ifdef IPSEC
 827                 /*
 828                  * Enforce inbound IPsec SPD.
 829                  */
 830                 if (ipsec4_in_reject(m, NULL)) {
 831                         ipsecstat.in_polvio++;
 832                         goto bad;
 833                 }
 834 #endif
 835 #ifdef FAST_IPSEC
 836                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 837                 s = splnet();
 838                 if (mtag != NULL) {
 839                         tdbi = (struct tdb_ident *)(mtag + 1);
 840                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 841                 } else {
 842                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 843                                                    IP_FORWARDING, &error);
 844                 }
 845                 if (sp == NULL) {       /* NB: can happen if error */
 846                         splx(s);
 847                         /*XXX error stat???*/
 848                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 849                         goto bad;
 850                 }
 851
 852                 /*
 853                  * Check security policy against packet attributes.
 854                  */
 855                 error = ipsec_in_reject(sp, m);
 856                 KEY_FREESP(&sp);
 857                 splx(s);
 858                 if (error) {
 859                         ipstat.ips_cantforward++;
 860                         goto bad;
 861                 }
 862 #endif
 863                 ip_forward(m, using_srcrt, args.next_hop);
 864         }
 865         return;
 866
 867 ours:
 868
 869         /*
 870          * IPSTEALTH: Process non-routing options only
 871          * if the packet is destined for us.
 872          */
 873         if (ipstealth &&
 874             hlen > sizeof(struct ip) &&
 875             ip_dooptions(m, 1, args.next_hop))
 876                 return;
 877
 878         /* Count the packet in the ip address stats */
 879         if (ia != NULL) {
 880                 ia->ia_ifa.if_ipackets++;
 881                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 882         }
 883
 884         /*
 885          * If offset or IP_MF are set, must reassemble.
 886          * Otherwise, nothing need be done.
 887          * (We could look in the reassembly queue to see
 888          * if the packet was previously fragmented,
 889          * but it's not worth the time; just let them time out.)
 890          */
 891         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 892
 893                 /* If maxnipq is 0, never accept fragments. */
 894                 if (maxnipq == 0) {
 895                         ipstat.ips_fragments++;
 896                         ipstat.ips_fragdropped++;
 897                         goto bad;
 898                 }
 899
 900                 sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 901                 /*
 902                  * Look for queue of fragments
 903                  * of this datagram.
 904                  */
 905                 for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 906                         if (ip->ip_id == fp->ipq_id &&
 907                             ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 908                             ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 909                             ip->ip_p == fp->ipq_p)
 910                                 goto found;
 911
 912                 fp = NULL;
 913
 914                 /*
 915                  * Enforce upper bound on number of fragmented packets
 916                  * for which we attempt reassembly;
 917                  * If maxnipq is -1, accept all fragments without limitation.
 918                  */
 919                 if ((nipq > maxnipq) && (maxnipq > 0)) {
 920                         /*
 921                          * drop something from the tail of the current queue
 922                          * before proceeding further
 923                          */
 924                         if (ipq[sum].prev == &ipq[sum]) {   /* gak */
 925                                 for (i = 0; i < IPREASS_NHASH; i++) {
 926                                         if (ipq[i].prev != &ipq[i]) {
 927                                                 ipstat.ips_fragtimeout +=
 928                                                     ipq[i].prev->ipq_nfrags;
 929                                                 ip_freef(ipq[i].prev);
 930                                                 break;
 931                                         }
 932                                 }
 933                         } else {
 934                                 ipstat.ips_fragtimeout +=
 935                                     ipq[sum].prev->ipq_nfrags;
 936                                 ip_freef(ipq[sum].prev);
 937                         }
 938                 }
 939 found:
 940                 /*
 941                  * Adjust ip_len to not reflect header,
 942                  * convert offset of this to bytes.
 943                  */
 944                 ip->ip_len -= hlen;
 945                 if (ip->ip_off & IP_MF) {
 946                         /*
 947                          * Make sure that fragments have a data length
 948                          * that's a non-zero multiple of 8 bytes.
 949                          */
 950                         if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
 951                                 ipstat.ips_toosmall++; /* XXX */
 952                                 goto bad;
 953                         }
 954                         m->m_flags |= M_FRAG;
 955                 } else
 956                         m->m_flags &= ~M_FRAG;
 957                 ip->ip_off <<= 3;
 958
 959                 /*
 960                  * Attempt reassembly; if it succeeds, proceed.
 961                  * ip_reass() will return a different mbuf, and update
 962                  * the divert info in divert_info.
 963                  */
 964                 ipstat.ips_fragments++;
 965                 m->m_pkthdr.header = ip;
 966                 m = ip_reass(m, fp, &ipq[sum], &divert_info);
 967                 if (m == NULL)
 968                         return;
 969                 ipstat.ips_reassembled++;
 970                 needredispatch = TRUE;
 971                 ip = mtod(m, struct ip *);
 972                 /* Get the header length of the reassembled packet */
 973                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 974 #ifdef IPDIVERT
 975                 /* Restore original checksum before diverting packet */
 976                 if (divert_info != 0) {
 977                         ip->ip_len += hlen;
 978                         ip->ip_len = htons(ip->ip_len);
 979                         ip->ip_off = htons(ip->ip_off);
 980                         ip->ip_sum = 0;
 981                         if (hlen == sizeof(struct ip))
 982                                 ip->ip_sum = in_cksum_hdr(ip);
 983                         else
 984                                 ip->ip_sum = in_cksum(m, hlen);
 985                         ip->ip_off = ntohs(ip->ip_off);
 986                         ip->ip_len = ntohs(ip->ip_len);
 987                         ip->ip_len -= hlen;
 988                 }
 989 #endif
 990         } else {
 991                 ip->ip_len -= hlen;
 992         }
 993
 994 #ifdef IPDIVERT
 995         /*
 996          * Divert or tee packet to the divert protocol if required.
 997          */
 998         if (divert_info != 0) {
 999                 struct mbuf *clone = NULL;
1000
1001                 /* Clone packet if we're doing a 'tee' */
1002                 if ((divert_info & IP_FW_PORT_TEE_FLAG) != 0)
1003                         clone = m_dup(m, MB_DONTWAIT);
1004
1005                 /* Restore packet header fields to original values */
1006                 ip->ip_len += hlen;
1007                 ip->ip_len = htons(ip->ip_len);
1008                 ip->ip_off = htons(ip->ip_off);
1009
1010                 /* Deliver packet to divert input routine */
1011                 divert_packet(m, 1, divert_info & 0xffff);
1012                 ipstat.ips_delivered++;
1013
1014                 /* If 'tee', continue with original packet */
1015                 if (clone == NULL)
1016                         return;
1017                 m = clone;
1018                 ip = mtod(m, struct ip *);
1019                 ip->ip_len += hlen;
1020                 /*
1021                  * Jump backwards to complete processing of the
1022                  * packet. But first clear divert_info to avoid
1023                  * entering this block again.
1024                  * We do not need to clear args.divert_rule
1025                  * or args.next_hop as they will not be used.
1026                  *
1027                  * XXX Better safe than sorry, remove the DIVERT tag.
1028                  */
1029                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1030                 if (mtag != NULL)
1031                         m_tag_delete(m, mtag);
1032
1033                 divert_info = 0;
1034                 goto pass;
1035         }
1036 #endif
1037
1038 #ifdef IPSEC
1039         /*
1040          * enforce IPsec policy checking if we are seeing last header.
1041          * note that we do not visit this with protocols with pcb layer
1042          * code - like udp/tcp/raw ip.
1043          */
1044         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
1045             ipsec4_in_reject(m, NULL)) {
1046                 ipsecstat.in_polvio++;
1047                 goto bad;
1048         }
1049 #endif
1050 #if FAST_IPSEC
1051         /*
1052          * enforce IPsec policy checking if we are seeing last header.
1053          * note that we do not visit this with protocols with pcb layer
1054          * code - like udp/tcp/raw ip.
1055          */
1056         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1057                 /*
1058                  * Check if the packet has already had IPsec processing
1059                  * done.  If so, then just pass it along.  This tag gets
1060                  * set during AH, ESP, etc. input handling, before the
1061                  * packet is returned to the ip input queue for delivery.
1062                  */
1063                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1064                 s = splnet();
1065                 if (mtag != NULL) {
1066                         tdbi = (struct tdb_ident *)(mtag + 1);
1067                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1068                 } else {
1069                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1070                                                    IP_FORWARDING, &error);
1071                 }
1072                 if (sp != NULL) {
1073                         /*
1074                          * Check security policy against packet attributes.
1075                          */
1076                         error = ipsec_in_reject(sp, m);
1077                         KEY_FREESP(&sp);
1078                 } else {
1079                         /* XXX error stat??? */
1080                         error = EINVAL;
1081 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1082                         goto bad;
1083                 }
1084                 splx(s);
1085                 if (error)
1086                         goto bad;
1087         }
1088 #endif /* FAST_IPSEC */
1089
1090         ipstat.ips_delivered++;
1091         if (needredispatch) {
1092                 struct netmsg_transport_packet *msg;
1093                 lwkt_port_t port;
1094
1095                 msg = malloc(sizeof(struct netmsg_transport_packet),
1096                                 M_LWKTMSG, M_INTWAIT | M_NULLOK);
1097                 if (msg == NULL)
1098                         goto bad;
1099
1100                 lwkt_initmsg(&msg->nm_lmsg, &netisr_afree_rport, 0,
1101                         lwkt_cmd_func(transport_processing_handler),
1102                         lwkt_cmd_op_none);
1103                 msg->nm_hlen = hlen;
1104                 msg->nm_hasnexthop = (args.next_hop != NULL);
1105                 if (msg->nm_hasnexthop)
1106                         msg->nm_nexthop = *args.next_hop;  /* structure copy */
1107
1108                 ip->ip_off = htons(ip->ip_off);
1109                 ip->ip_len = htons(ip->ip_len);
1110                 port = ip_mport(&m);
1111                 if (port) {
1112                     msg->nm_mbuf = m;
1113                     ip = mtod(m, struct ip *);
1114                     ip->ip_len = ntohs(ip->ip_len);
1115                     ip->ip_off = ntohs(ip->ip_off);
1116                     lwkt_sendmsg(port, &msg->nm_lmsg);
1117                 }
1118         } else {
1119                 transport_processing_oncpu(m, hlen, ip, args.next_hop);
1120         }
1121         return;
1122
1123 bad:
1124         m_freem(m);
1125 }
1126
1127 /*
1128  * Take incoming datagram fragment and try to reassemble it into
1129  * whole datagram.  If a chain for reassembly of this datagram already
1130  * exists, then it is given as fp; otherwise have to make a chain.
1131  *
1132  * When IPDIVERT enabled, keep additional state with each packet that
1133  * tells us if we need to divert or tee the packet we're building.
1134  * In particular, *divinfo includes the port and TEE flag.
1135  */
1136
1137 static struct mbuf *
1138 ip_reass(struct mbuf *m, struct ipq *fp, struct ipq *where,
1139          u_int32_t *divinfo)
1140 {
1141         struct ip *ip = mtod(m, struct ip *);
1142         struct mbuf *p = NULL, *q, *nq;
1143         struct mbuf *n;
1144         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1145         int i, next;
1146 #ifdef IPDIVERT
1147         struct m_tag *mtag;
1148 #endif
1149
1150         /*
1151          * Presence of header sizes in mbufs
1152          * would confuse code below.
1153          */
1154         m->m_data += hlen;
1155         m->m_len -= hlen;
1156
1157         /*
1158          * If first fragment to arrive, create a reassembly queue.
1159          */
1160         if (fp == NULL) {
1161                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1162                         goto dropfrag;
1163                 insque(fp, where);
1164                 nipq++;
1165                 fp->ipq_nfrags = 1;
1166                 fp->ipq_ttl = IPFRAGTTL;
1167                 fp->ipq_p = ip->ip_p;
1168                 fp->ipq_id = ip->ip_id;
1169                 fp->ipq_src = ip->ip_src;
1170                 fp->ipq_dst = ip->ip_dst;
1171                 fp->ipq_frags = m;
1172                 m->m_nextpkt = NULL;
1173 #ifdef IPDIVERT
1174                 fp->ipq_div_info = 0;
1175 #endif
1176                 goto inserted;
1177         } else {
1178                 fp->ipq_nfrags++;
1179         }
1180
1181 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1182
1183         /*
1184          * Find a segment which begins after this one does.
1185          */
1186         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1187                 if (GETIP(q)->ip_off > ip->ip_off)
1188                         break;
1189
1190         /*
1191          * If there is a preceding segment, it may provide some of
1192          * our data already.  If so, drop the data from the incoming
1193          * segment.  If it provides all of our data, drop us, otherwise
1194          * stick new segment in the proper place.
1195          *
1196          * If some of the data is dropped from the the preceding
1197          * segment, then it's checksum is invalidated.
1198          */
1199         if (p) {
1200                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1201                 if (i > 0) {
1202                         if (i >= ip->ip_len)
1203                                 goto dropfrag;
1204                         m_adj(m, i);
1205                         m->m_pkthdr.csum_flags = 0;
1206                         ip->ip_off += i;
1207                         ip->ip_len -= i;
1208                 }
1209                 m->m_nextpkt = p->m_nextpkt;
1210                 p->m_nextpkt = m;
1211         } else {
1212                 m->m_nextpkt = fp->ipq_frags;
1213                 fp->ipq_frags = m;
1214         }
1215
1216         /*
1217          * While we overlap succeeding segments trim them or,
1218          * if they are completely covered, dequeue them.
1219          */
1220         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1221              q = nq) {
1222                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1223                 if (i < GETIP(q)->ip_len) {
1224                         GETIP(q)->ip_len -= i;
1225                         GETIP(q)->ip_off += i;
1226                         m_adj(q, i);
1227                         q->m_pkthdr.csum_flags = 0;
1228                         break;
1229                 }
1230                 nq = q->m_nextpkt;
1231                 m->m_nextpkt = nq;
1232                 ipstat.ips_fragdropped++;
1233                 fp->ipq_nfrags--;
1234                 m_freem(q);
1235         }
1236
1237 inserted:
1238
1239 #ifdef IPDIVERT
1240         /*
1241          * Transfer firewall instructions to the fragment structure.
1242          * Only trust info in the fragment at offset 0.
1243          */
1244         if (ip->ip_off == 0) {
1245                 fp->ipq_div_info = *divinfo;
1246         } else {
1247                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1248                 if (mtag != NULL)
1249                         m_tag_delete(m, mtag);
1250         }
1251         *divinfo = 0;
1252 #endif
1253
1254         /*
1255          * Check for complete reassembly and perform frag per packet
1256          * limiting.
1257          *
1258          * Frag limiting is performed here so that the nth frag has
1259          * a chance to complete the packet before we drop the packet.
1260          * As a result, n+1 frags are actually allowed per packet, but
1261          * only n will ever be stored. (n = maxfragsperpacket.)
1262          *
1263          */
1264         next = 0;
1265         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1266                 if (GETIP(q)->ip_off != next) {
1267                         if (fp->ipq_nfrags > maxfragsperpacket) {
1268                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1269                                 ip_freef(fp);
1270                         }
1271                         return (NULL);
1272                 }
1273                 next += GETIP(q)->ip_len;
1274         }
1275         /* Make sure the last packet didn't have the IP_MF flag */
1276         if (p->m_flags & M_FRAG) {
1277                 if (fp->ipq_nfrags > maxfragsperpacket) {
1278                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1279                         ip_freef(fp);
1280                 }
1281                 return (NULL);
1282         }
1283
1284         /*
1285          * Reassembly is complete.  Make sure the packet is a sane size.
1286          */
1287         q = fp->ipq_frags;
1288         ip = GETIP(q);
1289         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1290                 ipstat.ips_toolong++;
1291                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1292                 ip_freef(fp);
1293                 return (NULL);
1294         }
1295
1296         /*
1297          * Concatenate fragments.
1298          */
1299         m = q;
1300         n = m->m_next;
1301         m->m_next = NULL;
1302         m_cat(m, n);
1303         nq = q->m_nextpkt;
1304         q->m_nextpkt = NULL;
1305         for (q = nq; q != NULL; q = nq) {
1306                 nq = q->m_nextpkt;
1307                 q->m_nextpkt = NULL;
1308                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1309                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1310                 m_cat(m, q);
1311         }
1312
1313 #ifdef IPDIVERT
1314         /*
1315          * Extract firewall instructions from the fragment structure.
1316          */
1317         *divinfo = fp->ipq_div_info;
1318 #endif
1319
1320         /*
1321          * Create header for new ip packet by
1322          * modifying header of first packet;
1323          * dequeue and discard fragment reassembly header.
1324          * Make header visible.
1325          */
1326         ip->ip_len = next;
1327         ip->ip_src = fp->ipq_src;
1328         ip->ip_dst = fp->ipq_dst;
1329         remque(fp);
1330         nipq--;
1331         mpipe_free(&ipq_mpipe, fp);
1332         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1333         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1334         /* some debugging cruft by sklower, below, will go away soon */
1335         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1336                 int plen = 0;
1337
1338                 for (n = m; n; n = n->m_next)
1339                         plen += n->m_len;
1340                 m->m_pkthdr.len = plen;
1341         }
1342         return (m);
1343
1344 dropfrag:
1345 #ifdef IPDIVERT
1346         *divinfo = 0;
1347 #endif
1348         ipstat.ips_fragdropped++;
1349         if (fp != NULL)
1350                 fp->ipq_nfrags--;
1351         m_freem(m);
1352         return (NULL);
1353
1354 #undef GETIP
1355 }
1356
1357 /*
1358  * Free a fragment reassembly header and all
1359  * associated datagrams.
1360  */
1361 static void
1362 ip_freef(struct ipq *fp)
1363 {
1364         struct mbuf *q;
1365
1366         while (fp->ipq_frags) {
1367                 q = fp->ipq_frags;
1368                 fp->ipq_frags = q->m_nextpkt;
1369                 m_freem(q);
1370         }
1371         remque(fp);
1372         mpipe_free(&ipq_mpipe, fp);
1373         nipq--;
1374 }
1375
1376 /*
1377  * IP timer processing;
1378  * if a timer expires on a reassembly
1379  * queue, discard it.
1380  */
1381 void
1382 ip_slowtimo(void)
1383 {
1384         struct ipq *fp;
1385         int s = splnet();
1386         int i;
1387
1388         for (i = 0; i < IPREASS_NHASH; i++) {
1389                 fp = ipq[i].next;
1390                 if (fp == NULL)
1391                         continue;
1392                 while (fp != &ipq[i]) {
1393                         --fp->ipq_ttl;
1394                         fp = fp->next;
1395                         if (fp->prev->ipq_ttl == 0) {
1396                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1397                                 ip_freef(fp->prev);
1398                         }
1399                 }
1400         }
1401         /*
1402          * If we are over the maximum number of fragments
1403          * (due to the limit being lowered), drain off
1404          * enough to get down to the new limit.
1405          */
1406         if (maxnipq >= 0 && nipq > maxnipq) {
1407                 for (i = 0; i < IPREASS_NHASH; i++) {
1408                         while (nipq > maxnipq &&
1409                                 (ipq[i].next != &ipq[i])) {
1410                                 ipstat.ips_fragdropped +=
1411                                     ipq[i].next->ipq_nfrags;
1412                                 ip_freef(ipq[i].next);
1413                         }
1414                 }
1415         }
1416         ipflow_slowtimo();
1417         splx(s);
1418 }
1419
1420 /*
1421  * Drain off all datagram fragments.
1422  */
1423 void
1424 ip_drain(void)
1425 {
1426         int i;
1427
1428         for (i = 0; i < IPREASS_NHASH; i++) {
1429                 while (ipq[i].next != &ipq[i]) {
1430                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1431                         ip_freef(ipq[i].next);
1432                 }
1433         }
1434         in_rtqdrain();
1435 }
1436
1437 /*
1438  * Do option processing on a datagram,
1439  * possibly discarding it if bad options are encountered,
1440  * or forwarding it if source-routed.
1441  * The pass argument is used when operating in the IPSTEALTH
1442  * mode to tell what options to process:
1443  * [LS]SRR (pass 0) or the others (pass 1).
1444  * The reason for as many as two passes is that when doing IPSTEALTH,
1445  * non-routing options should be processed only if the packet is for us.
1446  * Returns 1 if packet has been forwarded/freed,
1447  * 0 if the packet should be processed further.
1448  */
1449 static int
1450 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1451 {
1452         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1453         struct ip *ip = mtod(m, struct ip *);
1454         u_char *cp;
1455         struct in_ifaddr *ia;
1456         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1457         boolean_t forward = FALSE;
1458         struct in_addr *sin, dst;
1459         n_time ntime;
1460
1461         dst = ip->ip_dst;
1462         cp = (u_char *)(ip + 1);
1463         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1464         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1465                 opt = cp[IPOPT_OPTVAL];
1466                 if (opt == IPOPT_EOL)
1467                         break;
1468                 if (opt == IPOPT_NOP)
1469                         optlen = 1;
1470                 else {
1471                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1472                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1473                                 goto bad;
1474                         }
1475                         optlen = cp[IPOPT_OLEN];
1476                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1477                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1478                                 goto bad;
1479                         }
1480                 }
1481                 switch (opt) {
1482
1483                 default:
1484                         break;
1485
1486                 /*
1487                  * Source routing with record.
1488                  * Find interface with current destination address.
1489                  * If none on this machine then drop if strictly routed,
1490                  * or do nothing if loosely routed.
1491                  * Record interface address and bring up next address
1492                  * component.  If strictly routed make sure next
1493                  * address is on directly accessible net.
1494                  */
1495                 case IPOPT_LSRR:
1496                 case IPOPT_SSRR:
1497                         if (ipstealth && pass > 0)
1498                                 break;
1499                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1500                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1501                                 goto bad;
1502                         }
1503                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1504                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1505                                 goto bad;
1506                         }
1507                         ipaddr.sin_addr = ip->ip_dst;
1508                         ia = (struct in_ifaddr *)
1509                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1510                         if (ia == NULL) {
1511                                 if (opt == IPOPT_SSRR) {
1512                                         type = ICMP_UNREACH;
1513                                         code = ICMP_UNREACH_SRCFAIL;
1514                                         goto bad;
1515                                 }
1516                                 if (!ip_dosourceroute)
1517                                         goto nosourcerouting;
1518                                 /*
1519                                  * Loose routing, and not at next destination
1520                                  * yet; nothing to do except forward.
1521                                  */
1522                                 break;
1523                         }
1524                         off--;                  /* 0 origin */
1525                         if (off > optlen - (int)sizeof(struct in_addr)) {
1526                                 /*
1527                                  * End of source route.  Should be for us.
1528                                  */
1529                                 if (!ip_acceptsourceroute)
1530                                         goto nosourcerouting;
1531                                 save_rte(cp, ip->ip_src);
1532                                 break;
1533                         }
1534                         if (ipstealth)
1535                                 goto dropit;
1536                         if (!ip_dosourceroute) {
1537                                 if (ipforwarding) {
1538                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1539
1540                                         /*
1541                                          * Acting as a router, so generate ICMP
1542                                          */
1543 nosourcerouting:
1544                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1545                                         log(LOG_WARNING,
1546                                             "attempted source route from %s to %s\n",
1547                                             inet_ntoa(ip->ip_src), buf);
1548                                         type = ICMP_UNREACH;
1549                                         code = ICMP_UNREACH_SRCFAIL;
1550                                         goto bad;
1551                                 } else {
1552                                         /*
1553                                          * Not acting as a router,
1554                                          * so silently drop.
1555                                          */
1556 dropit:
1557                                         ipstat.ips_cantforward++;
1558                                         m_freem(m);
1559                                         return (1);
1560                                 }
1561                         }
1562
1563                         /*
1564                          * locate outgoing interface
1565                          */
1566                         memcpy(&ipaddr.sin_addr, cp + off,
1567                             sizeof ipaddr.sin_addr);
1568
1569                         if (opt == IPOPT_SSRR) {
1570 #define INA     struct in_ifaddr *
1571 #define SA      struct sockaddr *
1572                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1573                                                                         == NULL)
1574                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1575                         } else
1576                                 ia = ip_rtaddr(ipaddr.sin_addr, &ipforward_rt);
1577                         if (ia == NULL) {
1578                                 type = ICMP_UNREACH;
1579                                 code = ICMP_UNREACH_SRCFAIL;
1580                                 goto bad;
1581                         }
1582                         ip->ip_dst = ipaddr.sin_addr;
1583                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1584                             sizeof(struct in_addr));
1585                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1586                         /*
1587                          * Let ip_intr's mcast routing check handle mcast pkts
1588                          */
1589                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1590                         break;
1591
1592                 case IPOPT_RR:
1593                         if (ipstealth && pass == 0)
1594                                 break;
1595                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1596                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1597                                 goto bad;
1598                         }
1599                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1600                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1601                                 goto bad;
1602                         }
1603                         /*
1604                          * If no space remains, ignore.
1605                          */
1606                         off--;                  /* 0 origin */
1607                         if (off > optlen - (int)sizeof(struct in_addr))
1608                                 break;
1609                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1610                             sizeof ipaddr.sin_addr);
1611                         /*
1612                          * locate outgoing interface; if we're the destination,
1613                          * use the incoming interface (should be same).
1614                          */
1615                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1616                             (ia = ip_rtaddr(ipaddr.sin_addr, &ipforward_rt))
1617                                                                      == NULL) {
1618                                 type = ICMP_UNREACH;
1619                                 code = ICMP_UNREACH_HOST;
1620                                 goto bad;
1621                         }
1622                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1623                             sizeof(struct in_addr));
1624                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1625                         break;
1626
1627                 case IPOPT_TS:
1628                         if (ipstealth && pass == 0)
1629                                 break;
1630                         code = cp - (u_char *)ip;
1631                         if (optlen < 4 || optlen > 40) {
1632                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1633                                 goto bad;
1634                         }
1635                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1636                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1637                                 goto bad;
1638                         }
1639                         if (off > optlen - (int)sizeof(int32_t)) {
1640                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1641                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1642                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1643                                         goto bad;
1644                                 }
1645                                 break;
1646                         }
1647                         off--;                          /* 0 origin */
1648                         sin = (struct in_addr *)(cp + off);
1649                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1650
1651                         case IPOPT_TS_TSONLY:
1652                                 break;
1653
1654                         case IPOPT_TS_TSANDADDR:
1655                                 if (off + sizeof(n_time) +
1656                                     sizeof(struct in_addr) > optlen) {
1657                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1658                                         goto bad;
1659                                 }
1660                                 ipaddr.sin_addr = dst;
1661                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1662                                                             m->m_pkthdr.rcvif);
1663                                 if (ia == NULL)
1664                                         continue;
1665                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1666                                     sizeof(struct in_addr));
1667                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1668                                 off += sizeof(struct in_addr);
1669                                 break;
1670
1671                         case IPOPT_TS_PRESPEC:
1672                                 if (off + sizeof(n_time) +
1673                                     sizeof(struct in_addr) > optlen) {
1674                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1675                                         goto bad;
1676                                 }
1677                                 memcpy(&ipaddr.sin_addr, sin,
1678                                     sizeof(struct in_addr));
1679                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1680                                         continue;
1681                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1682                                 off += sizeof(struct in_addr);
1683                                 break;
1684
1685                         default:
1686                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1687                                 goto bad;
1688                         }
1689                         ntime = iptime();
1690                         memcpy(cp + off, &ntime, sizeof(n_time));
1691                         cp[IPOPT_OFFSET] += sizeof(n_time);
1692                 }
1693         }
1694         if (forward && ipforwarding) {
1695                 ip_forward(m, 1, next_hop);
1696                 return (1);
1697         }
1698         return (0);
1699 bad:
1700         icmp_error(m, type, code, 0, NULL);
1701         ipstat.ips_badoptions++;
1702         return (1);
1703 }
1704
1705 /*
1706  * Given address of next destination (final or next hop),
1707  * return internet address info of interface to be used to get there.
1708  */
1709 struct in_ifaddr *
1710 ip_rtaddr(struct in_addr dst, struct route *rt)
1711 {
1712         struct sockaddr_in *sin;
1713
1714         sin = (struct sockaddr_in *)&rt->ro_dst;
1715
1716         if (rt->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1717                 if (rt->ro_rt != NULL) {
1718                         RTFREE(rt->ro_rt);
1719                         rt->ro_rt = NULL;
1720                 }
1721                 sin->sin_family = AF_INET;
1722                 sin->sin_len = sizeof *sin;
1723                 sin->sin_addr = dst;
1724                 rtalloc_ign(rt, RTF_PRCLONING);
1725         }
1726
1727         if (rt->ro_rt == NULL)
1728                 return (NULL);
1729
1730         return (ifatoia(rt->ro_rt->rt_ifa));
1731 }
1732
1733 /*
1734  * Save incoming source route for use in replies,
1735  * to be picked up later by ip_srcroute if the receiver is interested.
1736  */
1737 void
1738 save_rte(u_char *option, struct in_addr dst)
1739 {
1740         unsigned olen;
1741
1742         olen = option[IPOPT_OLEN];
1743 #ifdef DIAGNOSTIC
1744         if (ipprintfs)
1745                 printf("save_rte: olen %d\n", olen);
1746 #endif
1747         if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst)))
1748                 return;
1749         bcopy(option, ip_srcrt.srcopt, olen);
1750         ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1751         ip_srcrt.dst = dst;
1752 }
1753
1754 /*
1755  * Retrieve incoming source route for use in replies,
1756  * in the same form used by setsockopt.
1757  * The first hop is placed before the options, will be removed later.
1758  */
1759 struct mbuf *
1760 ip_srcroute(void)
1761 {
1762         struct in_addr *p, *q;
1763         struct mbuf *m;
1764
1765         if (ip_nhops == 0)
1766                 return (NULL);
1767         m = m_get(MB_DONTWAIT, MT_HEADER);
1768         if (m == NULL)
1769                 return (NULL);
1770
1771 #define OPTSIZ  (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt))
1772
1773         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1774         m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) +
1775             OPTSIZ;
1776 #ifdef DIAGNOSTIC
1777         if (ipprintfs)
1778                 printf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
1779 #endif
1780
1781         /*
1782          * First save first hop for return route
1783          */
1784         p = &ip_srcrt.route[ip_nhops - 1];
1785         *(mtod(m, struct in_addr *)) = *p--;
1786 #ifdef DIAGNOSTIC
1787         if (ipprintfs)
1788                 printf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1789 #endif
1790
1791         /*
1792          * Copy option fields and padding (nop) to mbuf.
1793          */
1794         ip_srcrt.nop = IPOPT_NOP;
1795         ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1796         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &ip_srcrt.nop,
1797             OPTSIZ);
1798         q = (struct in_addr *)(mtod(m, caddr_t) +
1799             sizeof(struct in_addr) + OPTSIZ);
1800 #undef OPTSIZ
1801         /*
1802          * Record return path as an IP source route,
1803          * reversing the path (pointers are now aligned).
1804          */
1805         while (p >= ip_srcrt.route) {
1806 #ifdef DIAGNOSTIC
1807                 if (ipprintfs)
1808                         printf(" %x", ntohl(q->s_addr));
1809 #endif
1810                 *q++ = *p--;
1811         }
1812         /*
1813          * Last hop goes to final destination.
1814          */
1815         *q = ip_srcrt.dst;
1816 #ifdef DIAGNOSTIC
1817         if (ipprintfs)
1818                 printf(" %x\n", ntohl(q->s_addr));
1819 #endif
1820         return (m);
1821 }
1822
1823 /*
1824  * Strip out IP options.
1825  */
1826 void
1827 ip_stripoptions(struct mbuf *m)
1828 {
1829         int datalen;
1830         struct ip *ip = mtod(m, struct ip *);
1831         caddr_t opts;
1832         int optlen;
1833
1834         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1835         opts = (caddr_t)(ip + 1);
1836         datalen = m->m_len - (sizeof(struct ip) + optlen);
1837         bcopy(opts + optlen, opts, datalen);
1838         m->m_len -= optlen;
1839         if (m->m_flags & M_PKTHDR)
1840                 m->m_pkthdr.len -= optlen;
1841         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1842 }
1843
1844 u_char inetctlerrmap[PRC_NCMDS] = {
1845         0,              0,              0,              0,
1846         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1847         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1848         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1849         0,              0,              0,              0,
1850         ENOPROTOOPT,    ECONNREFUSED
1851 };
1852
1853 /*
1854  * Forward a packet.  If some error occurs return the sender
1855  * an icmp packet.  Note we can't always generate a meaningful
1856  * icmp message because icmp doesn't have a large enough repertoire
1857  * of codes and types.
1858  *
1859  * If not forwarding, just drop the packet.  This could be confusing
1860  * if ipforwarding was zero but some routing protocol was advancing
1861  * us as a gateway to somewhere.  However, we must let the routing
1862  * protocol deal with that.
1863  *
1864  * The using_srcrt parameter indicates whether the packet is being forwarded
1865  * via a source route.
1866  */
1867 static void
1868 ip_forward(struct mbuf *m, int using_srcrt, struct sockaddr_in *next_hop)
1869 {
1870         struct ip *ip = mtod(m, struct ip *);
1871         struct sockaddr_in *sin;
1872         struct rtentry *rt;
1873         int error, type = 0, code = 0;
1874         struct mbuf *mcopy;
1875         n_long dest;
1876         struct in_addr pkt_dst;
1877         struct ifnet *destifp;
1878         struct m_hdr tag;
1879 #if defined(IPSEC) || defined(FAST_IPSEC)
1880         struct ifnet dummyifp;
1881 #endif
1882
1883         dest = INADDR_ANY;
1884         /*
1885          * Cache the destination address of the packet; this may be
1886          * changed by use of 'ipfw fwd'.
1887          */
1888         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1889
1890 #ifdef DIAGNOSTIC
1891         if (ipprintfs)
1892                 printf("forward: src %x dst %x ttl %x\n",
1893                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1894 #endif
1895
1896         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1897                 ipstat.ips_cantforward++;
1898                 m_freem(m);
1899                 return;
1900         }
1901         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1902                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, NULL);
1903                 return;
1904         }
1905
1906         sin = (struct sockaddr_in *)&ipforward_rt.ro_dst;
1907         if ((rt = ipforward_rt.ro_rt) == NULL ||
1908             pkt_dst.s_addr != sin->sin_addr.s_addr) {
1909                 if (ipforward_rt.ro_rt != NULL) {
1910                         RTFREE(ipforward_rt.ro_rt);
1911                         ipforward_rt.ro_rt = NULL;
1912                 }
1913                 sin->sin_family = AF_INET;
1914                 sin->sin_len = sizeof(*sin);
1915                 sin->sin_addr = pkt_dst;
1916
1917                 rtalloc_ign(&ipforward_rt, RTF_PRCLONING);
1918                 if (ipforward_rt.ro_rt == NULL) {
1919                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest,
1920                                    NULL);
1921                         return;
1922                 }
1923                 rt = ipforward_rt.ro_rt;
1924         }
1925
1926         /*
1927          * Save the IP header and at most 8 bytes of the payload,
1928          * in case we need to generate an ICMP message to the src.
1929          *
1930          * XXX this can be optimized a lot by saving the data in a local
1931          * buffer on the stack (72 bytes at most), and only allocating the
1932          * mbuf if really necessary. The vast majority of the packets
1933          * are forwarded without having to send an ICMP back (either
1934          * because unnecessary, or because rate limited), so we are
1935          * really we are wasting a lot of work here.
1936          *
1937          * We don't use m_copy() because it might return a reference
1938          * to a shared cluster. Both this function and ip_output()
1939          * assume exclusive access to the IP header in `m', so any
1940          * data in a cluster may change before we reach icmp_error().
1941          */
1942         MGET(mcopy, MB_DONTWAIT, m->m_type);
1943         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1944                 /*
1945                  * It's probably ok if the pkthdr dup fails (because
1946                  * the deep copy of the tag chain failed), but for now
1947                  * be conservative and just discard the copy since
1948                  * code below may some day want the tags.
1949                  */
1950                 m_free(mcopy);
1951                 mcopy = NULL;
1952         }
1953         if (mcopy != NULL) {
1954                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1955                     (int)ip->ip_len);
1956                 mcopy->m_pkthdr.len = mcopy->m_len;
1957                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1958         }
1959
1960         if (!ipstealth)
1961                 ip->ip_ttl -= IPTTLDEC;
1962
1963         /*
1964          * If forwarding packet using same interface that it came in on,
1965          * perhaps should send a redirect to sender to shortcut a hop.
1966          * Only send redirect if source is sending directly to us,
1967          * and if packet was not source routed (or has any options).
1968          * Also, don't send redirect if forwarding using a default route
1969          * or a route modified by a redirect.
1970          */
1971         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1972             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1973             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1974             ipsendredirects && !using_srcrt && next_hop == NULL) {
1975                 u_long src = ntohl(ip->ip_src.s_addr);
1976
1977 #define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa))
1978                 if (RTA(rt) != NULL &&
1979                     (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1980                         if (rt->rt_flags & RTF_GATEWAY)
1981                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1982                         else
1983                                 dest = pkt_dst.s_addr;
1984                         /*
1985                          * Router requirements says to only send
1986                          * host redirects.
1987                          */
1988                         type = ICMP_REDIRECT;
1989                         code = ICMP_REDIRECT_HOST;
1990 #ifdef DIAGNOSTIC
1991                         if (ipprintfs)
1992                                 printf("redirect (%d) to %x\n", code, dest);
1993 #endif
1994                 }
1995         }
1996
1997         if (next_hop != NULL) {
1998                 /* Pass IPFORWARD info if available */
1999                 tag.mh_type = MT_TAG;
2000                 tag.mh_flags = PACKET_TAG_IPFORWARD;
2001                 tag.mh_data = (caddr_t)next_hop;
2002                 tag.mh_next = m;
2003                 m = (struct mbuf *)&tag;
2004         }
2005
2006         error = ip_output(m, NULL, &ipforward_rt, IP_FORWARDING, NULL, NULL);
2007
2008         if (error)
2009                 ipstat.ips_cantforward++;
2010         else {
2011                 ipstat.ips_forward++;
2012                 if (type)
2013                         ipstat.ips_redirectsent++;
2014                 else {
2015                         if (mcopy) {
2016                                 ipflow_create(&ipforward_rt, mcopy);
2017                                 m_freem(mcopy);
2018                         }
2019                         return;
2020                 }
2021         }
2022         if (mcopy == NULL)
2023                 return;
2024         destifp = NULL;
2025
2026         switch (error) {
2027
2028         case 0:                         /* forwarded, but need redirect */
2029                 /* type, code set above */
2030                 break;
2031
2032         case ENETUNREACH:               /* shouldn't happen, checked above */
2033         case EHOSTUNREACH:
2034         case ENETDOWN:
2035         case EHOSTDOWN:
2036         default:
2037                 type = ICMP_UNREACH;
2038                 code = ICMP_UNREACH_HOST;
2039                 break;
2040
2041         case EMSGSIZE:
2042                 type = ICMP_UNREACH;
2043                 code = ICMP_UNREACH_NEEDFRAG;
2044 #ifdef IPSEC
2045                 /*
2046                  * If the packet is routed over IPsec tunnel, tell the
2047                  * originator the tunnel MTU.
2048                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2049                  * XXX quickhack!!!
2050                  */
2051                 if (ipforward_rt.ro_rt != NULL) {
2052                         struct secpolicy *sp = NULL;
2053                         int ipsecerror;
2054                         int ipsechdr;
2055                         struct route *ro;
2056
2057                         sp = ipsec4_getpolicybyaddr(mcopy,
2058                                                     IPSEC_DIR_OUTBOUND,
2059                                                     IP_FORWARDING,
2060                                                     &ipsecerror);
2061
2062                         if (sp == NULL)
2063                                 destifp = ipforward_rt.ro_rt->rt_ifp;
2064                         else {
2065                                 /* count IPsec header size */
2066                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2067                                                          IPSEC_DIR_OUTBOUND,
2068                                                          NULL);
2069
2070                                 /*
2071                                  * find the correct route for outer IPv4
2072                                  * header, compute tunnel MTU.
2073                                  *
2074                                  * XXX BUG ALERT
2075                                  * The "dummyifp" code relies upon the fact
2076                                  * that icmp_error() touches only ifp->if_mtu.
2077                                  */
2078                                 /*XXX*/
2079                                 destifp = NULL;
2080                                 if (sp->req != NULL && sp->req->sav != NULL &&
2081                                     sp->req->sav->sah != NULL) {
2082                                         ro = &sp->req->sav->sah->sa_route;
2083                                         if (ro->ro_rt != NULL &&
2084                                             ro->ro_rt->rt_ifp != NULL) {
2085                                                 dummyifp.if_mtu =
2086                                                     ro->ro_rt->rt_ifp->if_mtu;
2087                                                 dummyifp.if_mtu -= ipsechdr;
2088                                                 destifp = &dummyifp;
2089                                         }
2090                                 }
2091
2092                                 key_freesp(sp);
2093                         }
2094                 }
2095 #elif FAST_IPSEC
2096                 /*
2097                  * If the packet is routed over IPsec tunnel, tell the
2098                  * originator the tunnel MTU.
2099                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2100                  * XXX quickhack!!!
2101                  */
2102                 if (ipforward_rt.ro_rt != NULL) {
2103                         struct secpolicy *sp = NULL;
2104                         int ipsecerror;
2105                         int ipsechdr;
2106                         struct route *ro;
2107
2108                         sp = ipsec_getpolicybyaddr(mcopy,
2109                                                    IPSEC_DIR_OUTBOUND,
2110                                                    IP_FORWARDING,
2111                                                    &ipsecerror);
2112
2113                         if (sp == NULL)
2114                                 destifp = ipforward_rt.ro_rt->rt_ifp;
2115                         else {
2116                                 /* count IPsec header size */
2117                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2118                                                          IPSEC_DIR_OUTBOUND,
2119                                                          NULL);
2120
2121                                 /*
2122                                  * find the correct route for outer IPv4
2123                                  * header, compute tunnel MTU.
2124                                  *
2125                                  * XXX BUG ALERT
2126                                  * The "dummyifp" code relies upon the fact
2127                                  * that icmp_error() touches only ifp->if_mtu.
2128                                  */
2129                                 /*XXX*/
2130                                 destifp = NULL;
2131                                 if (sp->req != NULL &&
2132                                     sp->req->sav != NULL &&
2133                                     sp->req->sav->sah != NULL) {
2134                                         ro = &sp->req->sav->sah->sa_route;
2135                                         if (ro->ro_rt != NULL &&
2136                                             ro->ro_rt->rt_ifp != NULL) {
2137                                                 dummyifp.if_mtu =
2138                                                     ro->ro_rt->rt_ifp->if_mtu;
2139                                                 dummyifp.if_mtu -= ipsechdr;
2140                                                 destifp = &dummyifp;
2141                                         }
2142                                 }
2143
2144                                 KEY_FREESP(&sp);
2145                         }
2146                 }
2147 #else /* !IPSEC && !FAST_IPSEC */
2148                 if (ipforward_rt.ro_rt != NULL)
2149                         destifp = ipforward_rt.ro_rt->rt_ifp;
2150 #endif /*IPSEC*/
2151                 ipstat.ips_cantfrag++;
2152                 break;
2153
2154         case ENOBUFS:
2155                 /*
2156                  * A router should not generate ICMP_SOURCEQUENCH as
2157                  * required in RFC1812 Requirements for IP Version 4 Routers.
2158                  * Source quench could be a big problem under DoS attacks,
2159                  * or if the underlying interface is rate-limited.
2160                  * Those who need source quench packets may re-enable them
2161                  * via the net.inet.ip.sendsourcequench sysctl.
2162                  */
2163                 if (!ip_sendsourcequench) {
2164                         m_freem(mcopy);
2165                         return;
2166                 } else {
2167                         type = ICMP_SOURCEQUENCH;
2168                         code = 0;
2169                 }
2170                 break;
2171
2172         case EACCES:                    /* ipfw denied packet */
2173                 m_freem(mcopy);
2174                 return;
2175         }
2176         icmp_error(mcopy, type, code, dest, destifp);
2177 }
2178
2179 void
2180 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2181                struct mbuf *m)
2182 {
2183         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2184                 struct timeval tv;
2185
2186                 microtime(&tv);
2187                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2188                     SCM_TIMESTAMP, SOL_SOCKET);
2189                 if (*mp)
2190                         mp = &(*mp)->m_next;
2191         }
2192         if (inp->inp_flags & INP_RECVDSTADDR) {
2193                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2194                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2195                 if (*mp)
2196                         mp = &(*mp)->m_next;
2197         }
2198 #ifdef notyet
2199         /* XXX
2200          * Moving these out of udp_input() made them even more broken
2201          * than they already were.
2202          */
2203         /* options were tossed already */
2204         if (inp->inp_flags & INP_RECVOPTS) {
2205                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2206                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2207                 if (*mp)
2208                         mp = &(*mp)->m_next;
2209         }
2210         /* ip_srcroute doesn't do what we want here, need to fix */
2211         if (inp->inp_flags & INP_RECVRETOPTS) {
2212                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(),
2213                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2214                 if (*mp)
2215                         mp = &(*mp)->m_next;
2216         }
2217 #endif
2218         if (inp->inp_flags & INP_RECVIF) {
2219                 struct ifnet *ifp;
2220                 struct sdlbuf {
2221                         struct sockaddr_dl sdl;
2222                         u_char  pad[32];
2223                 } sdlbuf;
2224                 struct sockaddr_dl *sdp;
2225                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2226
2227                 if (((ifp = m->m_pkthdr.rcvif)) &&
2228                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2229                         sdp = (struct sockaddr_dl *)
2230                             ifnet_addrs[ifp->if_index - 1]->ifa_addr;
2231                         /*
2232                          * Change our mind and don't try copy.
2233                          */
2234                         if ((sdp->sdl_family != AF_LINK) ||
2235                             (sdp->sdl_len > sizeof(sdlbuf))) {
2236                                 goto makedummy;
2237                         }
2238                         bcopy(sdp, sdl2, sdp->sdl_len);
2239                 } else {
2240 makedummy:
2241                         sdl2->sdl_len =
2242                             offsetof(struct sockaddr_dl, sdl_data[0]);
2243                         sdl2->sdl_family = AF_LINK;
2244                         sdl2->sdl_index = 0;
2245                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2246                 }
2247                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2248                         IP_RECVIF, IPPROTO_IP);
2249                 if (*mp)
2250                         mp = &(*mp)->m_next;
2251         }
2252 }
2253
2254 /*
2255  * XXX these routines are called from the upper part of the kernel.
2256  *
2257  * They could also be moved to ip_mroute.c, since all the RSVP
2258  *  handling is done there already.
2259  */
2260 int
2261 ip_rsvp_init(struct socket *so)
2262 {
2263         if (so->so_type != SOCK_RAW ||
2264             so->so_proto->pr_protocol != IPPROTO_RSVP)
2265                 return EOPNOTSUPP;
2266
2267         if (ip_rsvpd != NULL)
2268                 return EADDRINUSE;
2269
2270         ip_rsvpd = so;
2271         /*
2272          * This may seem silly, but we need to be sure we don't over-increment
2273          * the RSVP counter, in case something slips up.
2274          */
2275         if (!ip_rsvp_on) {
2276                 ip_rsvp_on = 1;
2277                 rsvp_on++;
2278         }
2279
2280         return 0;
2281 }
2282
2283 int
2284 ip_rsvp_done(void)
2285 {
2286         ip_rsvpd = NULL;
2287         /*
2288          * This may seem silly, but we need to be sure we don't over-decrement
2289          * the RSVP counter, in case something slips up.
2290          */
2291         if (ip_rsvp_on) {
2292                 ip_rsvp_on = 0;
2293                 rsvp_on--;
2294         }
2295         return 0;
2296 }
2297
2298 void
2299 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2300 {
2301         int off, proto;
2302         __va_list ap;
2303
2304         __va_start(ap, m);
2305         off = __va_arg(ap, int);
2306         proto = __va_arg(ap, int);
2307         __va_end(ap);
2308
2309         if (rsvp_input_p) { /* call the real one if loaded */
2310                 rsvp_input_p(m, off, proto);
2311                 return;
2312         }
2313
2314         /* Can still get packets with rsvp_on = 0 if there is a local member
2315          * of the group to which the RSVP packet is addressed.  But in this
2316          * case we want to throw the packet away.
2317          */
2318
2319         if (!rsvp_on) {
2320                 m_freem(m);
2321                 return;
2322         }
2323
2324         if (ip_rsvpd != NULL) {
2325                 rip_input(m, off, proto);
2326                 return;
2327         }
2328         /* Drop the packet */
2329         m_freem(m);
2330 }