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 1.36 2004/10/16 23:20:00 hsu 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 #include <netinet/ipprotosw.h>
 138
 139 #include <sys/socketvar.h>
 140
 141 #include <net/ipfw/ip_fw.h>
 142 #include <net/dummynet/ip_dummynet.h>
 143
 144 #ifdef IPSEC
 145 #include <netinet6/ipsec.h>
 146 #include <netproto/key/key.h>
 147 #endif
 148
 149 #ifdef FAST_IPSEC
 150 #include <netproto/ipsec/ipsec.h>
 151 #include <netproto/ipsec/key.h>
 152 #endif
 153
 154 int rsvp_on = 0;
 155 static int ip_rsvp_on;
 156 struct socket *ip_rsvpd;
 157
 158 int ipforwarding = 0;
 159 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 160     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 161
 162 static int ipsendredirects = 1; /* XXX */
 163 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 164     &ipsendredirects, 0, "Enable sending IP redirects");
 165
 166 int ip_defttl = IPDEFTTL;
 167 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 168     &ip_defttl, 0, "Maximum TTL on IP packets");
 169
 170 static int ip_dosourceroute = 0;
 171 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 172     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 173
 174 static int ip_acceptsourceroute = 0;
 175 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 176     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 177     "Enable accepting source routed IP packets");
 178
 179 static int ip_keepfaith = 0;
 180 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 181     &ip_keepfaith, 0,
 182     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
 183
 184 static int nipq = 0;    /* total # of reass queues */
 185 static int maxnipq;
 186 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 187     &maxnipq, 0,
 188     "Maximum number of IPv4 fragment reassembly queue entries");
 189
 190 static int maxfragsperpacket;
 191 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 192     &maxfragsperpacket, 0,
 193     "Maximum number of IPv4 fragments allowed per packet");
 194
 195 static int ip_sendsourcequench = 0;
 196 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 197     &ip_sendsourcequench, 0,
 198     "Enable the transmission of source quench packets");
 199
 200 /*
 201  * XXX - Setting ip_checkinterface mostly implements the receive side of
 202  * the Strong ES model described in RFC 1122, but since the routing table
 203  * and transmit implementation do not implement the Strong ES model,
 204  * setting this to 1 results in an odd hybrid.
 205  *
 206  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 207  * to translate the destination address to another local interface.
 208  *
 209  * XXX - ip_checkinterface must be disabled if you add IP aliases
 210  * to the loopback interface instead of the interface where the
 211  * packets for those addresses are received.
 212  */
 213 static int ip_checkinterface = 0;
 214 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 215     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 216
 217 #ifdef DIAGNOSTIC
 218 static int ipprintfs = 0;
 219 #endif
 220
 221 static struct ifqueue ipintrq;
 222 static int ipqmaxlen = IFQ_MAXLEN;
 223
 224 extern  struct domain inetdomain;
 225 extern  struct ipprotosw inetsw[];
 226 u_char  ip_protox[IPPROTO_MAX];
 227 struct  in_ifaddrhead in_ifaddrhead;            /* first inet address */
 228 struct  in_ifaddrhashhead *in_ifaddrhashtbl;    /* inet addr hash table */
 229 u_long  in_ifaddrhmask;                         /* mask for hash table */
 230
 231 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
 232     &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
 233 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
 234     &ipintrq.ifq_drops, 0, "Number of packets dropped from the IP input queue");
 235
 236 struct ip_stats ipstats_ary[MAXCPU];
 237 #ifdef SMP
 238 static int
 239 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 240 {
 241         int cpu, error = 0;
 242
 243         for (cpu = 0; cpu < ncpus; ++cpu) {
 244                 if ((error = SYSCTL_OUT(req, (void *)&ipstats_ary[cpu],
 245                                         sizeof(struct ip_stats))))
 246                         break;
 247                 if ((error = SYSCTL_IN(req, (void *)&ipstats_ary[cpu],
 248                                        sizeof(struct ip_stats))))
 249                         break;
 250         }
 251
 252         return (error);
 253 }
 254 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 255     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 256 #else
 257 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 258     &ipstat, ip_stats, "IP statistics");
 259 #endif
 260
 261 /* Packet reassembly stuff */
 262 #define IPREASS_NHASH_LOG2      6
 263 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 264 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 265 #define IPREASS_HASH(x,y)                                               \
 266     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 267
 268 static struct ipq ipq[IPREASS_NHASH];
 269 const  int    ipintrq_present = 1;
 270
 271 #ifdef IPCTL_DEFMTU
 272 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 273     &ip_mtu, 0, "Default MTU");
 274 #endif
 275
 276 #ifdef IPSTEALTH
 277 static int ipstealth = 0;
 278 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 279 #else
 280 static const int ipstealth = 0;
 281 #endif
 282
 283
 284 /* Firewall hooks */
 285 ip_fw_chk_t *ip_fw_chk_ptr;
 286 int fw_enable = 1;
 287 int fw_one_pass = 1;
 288
 289 /* Dummynet hooks */
 290 ip_dn_io_t *ip_dn_io_ptr;
 291
 292 struct pfil_head inet_pfil_hook;
 293
 294 /*
 295  * XXX this is ugly -- the following two global variables are
 296  * used to store packet state while it travels through the stack.
 297  * Note that the code even makes assumptions on the size and
 298  * alignment of fields inside struct ip_srcrt so e.g. adding some
 299  * fields will break the code. This needs to be fixed.
 300  *
 301  * We need to save the IP options in case a protocol wants to respond
 302  * to an incoming packet over the same route if the packet got here
 303  * using IP source routing.  This allows connection establishment and
 304  * maintenance when the remote end is on a network that is not known
 305  * to us.
 306  */
 307 static int ip_nhops = 0;
 308
 309 static  struct ip_srcrt {
 310         struct  in_addr dst;                    /* final destination */
 311         char    nop;                            /* one NOP to align */
 312         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 313         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 314 } ip_srcrt;
 315
 316 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 317 static struct malloc_pipe ipq_mpipe;
 318
 319 static void             save_rte (u_char *, struct in_addr);
 320 static int              ip_dooptions (struct mbuf *m, int,
 321                                         struct sockaddr_in *next_hop);
 322 static void             ip_forward (struct mbuf *m, int srcrt,
 323                                         struct sockaddr_in *next_hop);
 324 static void             ip_freef (struct ipq *);
 325 static int              ip_input_handler (struct netmsg *);
 326 static struct mbuf      *ip_reass (struct mbuf *, struct ipq *,
 327                                         struct ipq *, u_int32_t *, u_int16_t *);
 328
 329 /*
 330  * IP initialization: fill in IP protocol switch table.
 331  * All protocols not implemented in kernel go to raw IP protocol handler.
 332  */
 333 void
 334 ip_init(void)
 335 {
 336         struct ipprotosw *pr;
 337         int i;
 338 #ifdef SMP
 339         int cpu;
 340 #endif
 341
 342         /*
 343          * Make sure we can handle a reasonable number of fragments but
 344          * cap it at 4000 (XXX).
 345          */
 346         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 347                     IFQ_MAXLEN, 4000, 0, NULL);
 348         TAILQ_INIT(&in_ifaddrhead);
 349         in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 350         pr = (struct ipprotosw *)pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 351         if (pr == NULL)
 352                 panic("ip_init");
 353         for (i = 0; i < IPPROTO_MAX; i++)
 354                 ip_protox[i] = pr - inetsw;
 355         for (pr = (struct ipprotosw *)inetdomain.dom_protosw;
 356              pr < (struct ipprotosw *)inetdomain.dom_protoswNPROTOSW; pr++)
 357                 if (pr->pr_domain->dom_family == PF_INET &&
 358                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
 359                         ip_protox[pr->pr_protocol] = pr - inetsw;
 360
 361         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 362         inet_pfil_hook.ph_af = AF_INET;
 363         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 364                 printf("%s: WARNING: unable to register pfil hook, "
 365                         "error %d\n", __func__, i);
 366         }
 367
 368         for (i = 0; i < IPREASS_NHASH; i++)
 369             ipq[i].next = ipq[i].prev = &ipq[i];
 370
 371         maxnipq = nmbclusters / 32;
 372         maxfragsperpacket = 16;
 373
 374 #ifndef RANDOM_IP_ID
 375         ip_id = time_second & 0xffff;
 376 #endif
 377         ipintrq.ifq_maxlen = ipqmaxlen;
 378
 379         /*
 380          * Initialize IP statistics.
 381          *
 382          * It is layed out as an array which is has one element for UP,
 383          * and SMP_MAXCPU elements for SMP.  This allows us to retain
 384          * the access mechanism from userland for both UP and SMP.
 385          */
 386 #ifdef SMP
 387         for (cpu = 0; cpu < ncpus; ++cpu) {
 388                 bzero(&ipstats_ary[cpu], sizeof(struct ip_stats));
 389         }
 390 #else
 391         bzero(&ipstat, sizeof(struct ip_stats));
 392 #endif
 393
 394         netisr_register(NETISR_IP, ip_mport, ip_input_handler);
 395 }
 396
 397 /*
 398  * XXX watch out this one. It is perhaps used as a cache for
 399  * the most recently used route ? it is cleared in in_addroute()
 400  * when a new route is successfully created.
 401  */
 402 struct route ipforward_rt;
 403 static struct sockaddr_in ipaddr = { sizeof(ipaddr), AF_INET };
 404
 405 /* Do transport protocol processing. */
 406 static void
 407 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip,
 408                            struct sockaddr_in *nexthop)
 409 {
 410         /*
 411          * Switch out to protocol's input routine.
 412          */
 413         if (nexthop && ip->ip_p == IPPROTO_TCP) {
 414                 /* TCP needs IPFORWARD info if available */
 415                 struct m_hdr tag;
 416
 417                 tag.mh_type = MT_TAG;
 418                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 419                 tag.mh_data = (caddr_t)nexthop;
 420                 tag.mh_next = m;
 421
 422                 (*inetsw[ip_protox[ip->ip_p]].pr_input)
 423                     ((struct mbuf *)&tag, hlen, ip->ip_p);
 424         } else {
 425                 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
 426         }
 427 }
 428
 429 struct netmsg_transport_packet {
 430         struct lwkt_msg         nm_lmsg;
 431         struct mbuf             *nm_mbuf;
 432         int                     nm_hlen;
 433         boolean_t               nm_hasnexthop;
 434         struct sockaddr_in      nm_nexthop;
 435 };
 436
 437 static int
 438 transport_processing_handler(lwkt_msg_t lmsg)
 439 {
 440         struct netmsg_transport_packet *msg = (void *)lmsg;
 441         struct sockaddr_in *nexthop;
 442         struct ip *ip;
 443
 444         ip = mtod(msg->nm_mbuf, struct ip *);
 445         nexthop = msg->nm_hasnexthop ? &msg->nm_nexthop : NULL;
 446         transport_processing_oncpu(msg->nm_mbuf, msg->nm_hlen, ip, nexthop);
 447         lwkt_replymsg(lmsg, 0);
 448         return(EASYNC);
 449 }
 450
 451 static int
 452 ip_input_handler(struct netmsg *msg0)
 453 {
 454         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 455
 456         ip_input(m);
 457         lwkt_replymsg(&msg0->nm_lmsg, 0);
 458         return(EASYNC);
 459 }
 460
 461 /*
 462  * Ip input routine.  Checksum and byte swap header.  If fragmented
 463  * try to reassemble.  Process options.  Pass to next level.
 464  */
 465 void
 466 ip_input(struct mbuf *m)
 467 {
 468         struct ip *ip;
 469         struct ipq *fp;
 470         struct in_ifaddr *ia = NULL;
 471         struct ifaddr *ifa;
 472         int i, hlen, checkif;
 473         u_short sum;
 474         struct in_addr pkt_dst;
 475         u_int32_t divert_info = 0;              /* packet divert/tee info */
 476         struct ip_fw_args args;
 477         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 478         boolean_t needredispatch = FALSE;
 479         struct in_addr odst;                    /* original dst address(NAT) */
 480 #ifdef FAST_IPSEC
 481         struct m_tag *mtag;
 482         struct tdb_ident *tdbi;
 483         struct secpolicy *sp;
 484         int s, error;
 485 #endif
 486
 487         args.eh = NULL;
 488         args.oif = NULL;
 489         args.rule = NULL;
 490         args.divert_rule = 0;                   /* divert cookie */
 491         args.next_hop = NULL;
 492
 493         /* Grab info from MT_TAG mbufs prepended to the chain. */
 494         for (; m && m->m_type == MT_TAG; m = m->m_next) {
 495                 switch(m->_m_tag_id) {
 496                 default:
 497                         printf("ip_input: unrecognised MT_TAG tag %d\n",
 498                             m->_m_tag_id);
 499                         break;
 500
 501                 case PACKET_TAG_DUMMYNET:
 502                         args.rule = ((struct dn_pkt *)m)->rule;
 503                         break;
 504
 505                 case PACKET_TAG_DIVERT:
 506                         args.divert_rule = (int)m->m_hdr.mh_data & 0xffff;
 507                         break;
 508
 509                 case PACKET_TAG_IPFORWARD:
 510                         args.next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
 511                         break;
 512                 }
 513         }
 514
 515         KASSERT(m != NULL && (m->m_flags & M_PKTHDR) != 0,
 516             ("ip_input: no HDR"));
 517
 518         if (args.rule) {        /* dummynet already filtered us */
 519                 ip = mtod(m, struct ip *);
 520                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 521                 goto iphack;
 522         }
 523
 524         ipstat.ips_total++;
 525
 526         /* length checks already done in ip_demux() */
 527         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 528
 529         ip = mtod(m, struct ip *);
 530
 531         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 532                 ipstat.ips_badvers++;
 533                 goto bad;
 534         }
 535
 536         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 537         /* length checks already done in ip_demux() */
 538         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 539         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 540
 541         /* 127/8 must not appear on wire - RFC1122 */
 542         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 543             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 544                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 545                         ipstat.ips_badaddr++;
 546                         goto bad;
 547                 }
 548         }
 549
 550         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 551                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 552         } else {
 553                 if (hlen == sizeof(struct ip)) {
 554                         sum = in_cksum_hdr(ip);
 555                 } else {
 556                         sum = in_cksum(m, hlen);
 557                 }
 558         }
 559         if (sum) {
 560                 ipstat.ips_badsum++;
 561                 goto bad;
 562         }
 563
 564         /*
 565          * Convert fields to host representation.
 566          */
 567         ip->ip_len = ntohs(ip->ip_len);
 568         if (ip->ip_len < hlen) {
 569                 ipstat.ips_badlen++;
 570                 goto bad;
 571         }
 572         ip->ip_off = ntohs(ip->ip_off);
 573
 574         /*
 575          * Check that the amount of data in the buffers
 576          * is as at least much as the IP header would have us expect.
 577          * Trim mbufs if longer than we expect.
 578          * Drop packet if shorter than we expect.
 579          */
 580         if (m->m_pkthdr.len < ip->ip_len) {
 581                 ipstat.ips_tooshort++;
 582                 goto bad;
 583         }
 584         if (m->m_pkthdr.len > ip->ip_len) {
 585                 if (m->m_len == m->m_pkthdr.len) {
 586                         m->m_len = ip->ip_len;
 587                         m->m_pkthdr.len = ip->ip_len;
 588                 } else
 589                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 590         }
 591 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 592         /*
 593          * Bypass packet filtering for packets from a tunnel (gif).
 594          */
 595         if (ipsec_gethist(m, NULL))
 596                 goto pass;
 597 #endif
 598
 599         /*
 600          * IpHack's section.
 601          * Right now when no processing on packet has done
 602          * and it is still fresh out of network we do our black
 603          * deals with it.
 604          * - Firewall: deny/allow/divert
 605          * - Xlate: translate packet's addr/port (NAT).
 606          * - Pipe: pass pkt through dummynet.
 607          * - Wrap: fake packet's addr/port <unimpl.>
 608          * - Encapsulate: put it in another IP and send out. <unimp.>
 609          */
 610
 611 iphack:
 612
 613         /*
 614          * Run through list of hooks for input packets.
 615          *
 616          * NB: Beware of the destination address changing (e.g.
 617          *     by NAT rewriting). When this happens, tell
 618          *     ip_forward to do the right thing.
 619          */
 620         if (pfil_has_hooks(&inet_pfil_hook)) {
 621                 odst = ip->ip_dst;
 622                 if (pfil_run_hooks(&inet_pfil_hook, &m,
 623                     m->m_pkthdr.rcvif, PFIL_IN)) {
 624                         return;
 625                 }
 626                 if (m == NULL)                  /* consumed by filter */
 627                         return;
 628                 ip = mtod(m, struct ip *);
 629                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 630         }
 631
 632         if (fw_enable && IPFW_LOADED) {
 633                 /*
 634                  * If we've been forwarded from the output side, then
 635                  * skip the firewall a second time
 636                  */
 637                 if (args.next_hop)
 638                         goto ours;
 639
 640                 args.m = m;
 641                 i = ip_fw_chk_ptr(&args);
 642                 m = args.m;
 643
 644                 if ( (i & IP_FW_PORT_DENY_FLAG) || m == NULL) { /* drop */
 645                         if (m)
 646                                 m_freem(m);
 647                         return;
 648                 }
 649                 ip = mtod(m, struct ip *); /* just in case m changed */
 650                 if (i == 0 && args.next_hop == NULL)    /* common case */
 651                         goto pass;
 652                 if (DUMMYNET_LOADED && (i & IP_FW_PORT_DYNT_FLAG)) {
 653                         /* Send packet to the appropriate pipe */
 654                         ip_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
 655                         return;
 656                 }
 657 #ifdef IPDIVERT
 658                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
 659                         /* Divert or tee packet */
 660                         divert_info = i;
 661                         goto ours;
 662                 }
 663 #endif
 664                 if (i == 0 && args.next_hop != NULL)
 665                         goto pass;
 666                 /*
 667                  * if we get here, the packet must be dropped
 668                  */
 669                 m_freem(m);
 670                 return;
 671         }
 672 pass:
 673
 674         /*
 675          * Process options and, if not destined for us,
 676          * ship it on.  ip_dooptions returns 1 when an
 677          * error was detected (causing an icmp message
 678          * to be sent and the original packet to be freed).
 679          */
 680         ip_nhops = 0;           /* for source routed packets */
 681         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, args.next_hop))
 682                 return;
 683
 684         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 685          * matter if it is destined to another node, or whether it is
 686          * a multicast one, RSVP wants it! and prevents it from being forwarded
 687          * anywhere else. Also checks if the rsvp daemon is running before
 688          * grabbing the packet.
 689          */
 690         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 691                 goto ours;
 692
 693         /*
 694          * Check our list of addresses, to see if the packet is for us.
 695          * If we don't have any addresses, assume any unicast packet
 696          * we receive might be for us (and let the upper layers deal
 697          * with it).
 698          */
 699         if (TAILQ_EMPTY(&in_ifaddrhead) && !(m->m_flags & (M_MCAST | M_BCAST)))
 700                 goto ours;
 701
 702         /*
 703          * Cache the destination address of the packet; this may be
 704          * changed by use of 'ipfw fwd'.
 705          */
 706         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 707
 708         /*
 709          * Enable a consistency check between the destination address
 710          * and the arrival interface for a unicast packet (the RFC 1122
 711          * strong ES model) if IP forwarding is disabled and the packet
 712          * is not locally generated and the packet is not subject to
 713          * 'ipfw fwd'.
 714          *
 715          * XXX - Checking also should be disabled if the destination
 716          * address is ipnat'ed to a different interface.
 717          *
 718          * XXX - Checking is incompatible with IP aliases added
 719          * to the loopback interface instead of the interface where
 720          * the packets are received.
 721          */
 722         checkif = ip_checkinterface &&
 723                   !ipforwarding &&
 724                   m->m_pkthdr.rcvif != NULL &&
 725                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 726                   (args.next_hop == NULL);
 727
 728         /*
 729          * Check for exact addresses in the hash bucket.
 730          */
 731         LIST_FOREACH(ia, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 732                 /*
 733                  * If the address matches, verify that the packet
 734                  * arrived via the correct interface if checking is
 735                  * enabled.
 736                  */
 737                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 738                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 739                         goto ours;
 740         }
 741         /*
 742          * Check for broadcast addresses.
 743          *
 744          * Only accept broadcast packets that arrive via the matching
 745          * interface.  Reception of forwarded directed broadcasts would
 746          * be handled via ip_forward() and ether_output() with the loopback
 747          * into the stack for SIMPLEX interfaces handled by ether_output().
 748          */
 749         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 750                 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
 751                         if (ifa->ifa_addr->sa_family != AF_INET)
 752                                 continue;
 753                         ia = ifatoia(ifa);
 754                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 755                                                                 pkt_dst.s_addr)
 756                                 goto ours;
 757                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 758                                 goto ours;
 759 #ifdef BOOTP_COMPAT
 760                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 761                                 goto ours;
 762 #endif
 763                 }
 764         }
 765         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 766                 struct in_multi *inm;
 767                 if (ip_mrouter) {
 768                         /*
 769                          * If we are acting as a multicast router, all
 770                          * incoming multicast packets are passed to the
 771                          * kernel-level multicast forwarding function.
 772                          * The packet is returned (relatively) intact; if
 773                          * ip_mforward() returns a non-zero value, the packet
 774                          * must be discarded, else it may be accepted below.
 775                          */
 776                         if (ip_mforward &&
 777                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 778                                 ipstat.ips_cantforward++;
 779                                 m_freem(m);
 780                                 return;
 781                         }
 782
 783                         /*
 784                          * The process-level routing daemon needs to receive
 785                          * all multicast IGMP packets, whether or not this
 786                          * host belongs to their destination groups.
 787                          */
 788                         if (ip->ip_p == IPPROTO_IGMP)
 789                                 goto ours;
 790                         ipstat.ips_forward++;
 791                 }
 792                 /*
 793                  * See if we belong to the destination multicast group on the
 794                  * arrival interface.
 795                  */
 796                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 797                 if (inm == NULL) {
 798                         ipstat.ips_notmember++;
 799                         m_freem(m);
 800                         return;
 801                 }
 802                 goto ours;
 803         }
 804         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 805                 goto ours;
 806         if (ip->ip_dst.s_addr == INADDR_ANY)
 807                 goto ours;
 808
 809         /*
 810          * FAITH(Firewall Aided Internet Translator)
 811          */
 812         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 813                 if (ip_keepfaith) {
 814                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 815                                 goto ours;
 816                 }
 817                 m_freem(m);
 818                 return;
 819         }
 820
 821         /*
 822          * Not for us; forward if possible and desirable.
 823          */
 824         if (!ipforwarding) {
 825                 ipstat.ips_cantforward++;
 826                 m_freem(m);
 827         } else {
 828 #ifdef IPSEC
 829                 /*
 830                  * Enforce inbound IPsec SPD.
 831                  */
 832                 if (ipsec4_in_reject(m, NULL)) {
 833                         ipsecstat.in_polvio++;
 834                         goto bad;
 835                 }
 836 #endif
 837 #ifdef FAST_IPSEC
 838                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 839                 s = splnet();
 840                 if (mtag != NULL) {
 841                         tdbi = (struct tdb_ident *)(mtag + 1);
 842                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 843                 } else {
 844                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 845                                                    IP_FORWARDING, &error);
 846                 }
 847                 if (sp == NULL) {       /* NB: can happen if error */
 848                         splx(s);
 849                         /*XXX error stat???*/
 850                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 851                         goto bad;
 852                 }
 853
 854                 /*
 855                  * Check security policy against packet attributes.
 856                  */
 857                 error = ipsec_in_reject(sp, m);
 858                 KEY_FREESP(&sp);
 859                 splx(s);
 860                 if (error) {
 861                         ipstat.ips_cantforward++;
 862                         goto bad;
 863                 }
 864 #endif
 865                 ip_forward(m, using_srcrt, args.next_hop);
 866         }
 867         return;
 868
 869 ours:
 870
 871         /*
 872          * IPSTEALTH: Process non-routing options only
 873          * if the packet is destined for us.
 874          */
 875         if (ipstealth &&
 876             hlen > sizeof(struct ip) &&
 877             ip_dooptions(m, 1, args.next_hop))
 878                 return;
 879
 880         /* Count the packet in the ip address stats */
 881         if (ia != NULL) {
 882                 ia->ia_ifa.if_ipackets++;
 883                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 884         }
 885
 886         /*
 887          * If offset or IP_MF are set, must reassemble.
 888          * Otherwise, nothing need be done.
 889          * (We could look in the reassembly queue to see
 890          * if the packet was previously fragmented,
 891          * but it's not worth the time; just let them time out.)
 892          */
 893         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 894
 895                 /* If maxnipq is 0, never accept fragments. */
 896                 if (maxnipq == 0) {
 897                         ipstat.ips_fragments++;
 898                         ipstat.ips_fragdropped++;
 899                         goto bad;
 900                 }
 901
 902                 sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 903                 /*
 904                  * Look for queue of fragments
 905                  * of this datagram.
 906                  */
 907                 for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 908                         if (ip->ip_id == fp->ipq_id &&
 909                             ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 910                             ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 911                             ip->ip_p == fp->ipq_p)
 912                                 goto found;
 913
 914                 fp = NULL;
 915
 916                 /*
 917                  * Enforce upper bound on number of fragmented packets
 918                  * for which we attempt reassembly;
 919                  * If maxnipq is -1, accept all fragments without limitation.
 920                  */
 921                 if ((nipq > maxnipq) && (maxnipq > 0)) {
 922                         /*
 923                          * drop something from the tail of the current queue
 924                          * before proceeding further
 925                          */
 926                         if (ipq[sum].prev == &ipq[sum]) {   /* gak */
 927                                 for (i = 0; i < IPREASS_NHASH; i++) {
 928                                         if (ipq[i].prev != &ipq[i]) {
 929                                                 ipstat.ips_fragtimeout +=
 930                                                     ipq[i].prev->ipq_nfrags;
 931                                                 ip_freef(ipq[i].prev);
 932                                                 break;
 933                                         }
 934                                 }
 935                         } else {
 936                                 ipstat.ips_fragtimeout +=
 937                                     ipq[sum].prev->ipq_nfrags;
 938                                 ip_freef(ipq[sum].prev);
 939                         }
 940                 }
 941 found:
 942                 /*
 943                  * Adjust ip_len to not reflect header,
 944                  * convert offset of this to bytes.
 945                  */
 946                 ip->ip_len -= hlen;
 947                 if (ip->ip_off & IP_MF) {
 948                         /*
 949                          * Make sure that fragments have a data length
 950                          * that's a non-zero multiple of 8 bytes.
 951                          */
 952                         if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
 953                                 ipstat.ips_toosmall++; /* XXX */
 954                                 goto bad;
 955                         }
 956                         m->m_flags |= M_FRAG;
 957                 } else
 958                         m->m_flags &= ~M_FRAG;
 959                 ip->ip_off <<= 3;
 960
 961                 /*
 962                  * Attempt reassembly; if it succeeds, proceed.
 963                  * ip_reass() will return a different mbuf, and update
 964                  * the divert info in divert_info and args.divert_rule.
 965                  */
 966                 ipstat.ips_fragments++;
 967                 m->m_pkthdr.header = ip;
 968                 m = ip_reass(m, fp, &ipq[sum], &divert_info, &args.divert_rule);
 969                 if (m == NULL)
 970                         return;
 971                 ipstat.ips_reassembled++;
 972                 needredispatch = TRUE;
 973                 ip = mtod(m, struct ip *);
 974                 /* Get the header length of the reassembled packet */
 975                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 976 #ifdef IPDIVERT
 977                 /* Restore original checksum before diverting packet */
 978                 if (divert_info != 0) {
 979                         ip->ip_len += hlen;
 980                         ip->ip_len = htons(ip->ip_len);
 981                         ip->ip_off = htons(ip->ip_off);
 982                         ip->ip_sum = 0;
 983                         if (hlen == sizeof(struct ip))
 984                                 ip->ip_sum = in_cksum_hdr(ip);
 985                         else
 986                                 ip->ip_sum = in_cksum(m, hlen);
 987                         ip->ip_off = ntohs(ip->ip_off);
 988                         ip->ip_len = ntohs(ip->ip_len);
 989                         ip->ip_len -= hlen;
 990                 }
 991 #endif
 992         } else {
 993                 ip->ip_len -= hlen;
 994         }
 995
 996 #ifdef IPDIVERT
 997         /*
 998          * Divert or tee packet to the divert protocol if required.
 999          */
1000         if (divert_info != 0) {
1001                 struct mbuf *clone = NULL;
1002
1003                 /* Clone packet if we're doing a 'tee' */
1004                 if ((divert_info & IP_FW_PORT_TEE_FLAG) != 0)
1005                         clone = m_dup(m, MB_DONTWAIT);
1006
1007                 /* Restore packet header fields to original values */
1008                 ip->ip_len += hlen;
1009                 ip->ip_len = htons(ip->ip_len);
1010                 ip->ip_off = htons(ip->ip_off);
1011
1012                 /* Deliver packet to divert input routine */
1013                 divert_packet(m, 1, divert_info & 0xffff, args.divert_rule);
1014                 ipstat.ips_delivered++;
1015
1016                 /* If 'tee', continue with original packet */
1017                 if (clone == NULL)
1018                         return;
1019                 m = clone;
1020                 ip = mtod(m, struct ip *);
1021                 ip->ip_len += hlen;
1022                 /*
1023                  * Jump backwards to complete processing of the
1024                  * packet. But first clear divert_info to avoid
1025                  * entering this block again.
1026                  * We do not need to clear args.divert_rule
1027                  * or args.next_hop as they will not be used.
1028                  */
1029                 divert_info = 0;
1030                 goto pass;
1031         }
1032 #endif
1033
1034 #ifdef IPSEC
1035         /*
1036          * enforce IPsec policy checking if we are seeing last header.
1037          * note that we do not visit this with protocols with pcb layer
1038          * code - like udp/tcp/raw ip.
1039          */
1040         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
1041             ipsec4_in_reject(m, NULL)) {
1042                 ipsecstat.in_polvio++;
1043                 goto bad;
1044         }
1045 #endif
1046 #if FAST_IPSEC
1047         /*
1048          * enforce IPsec policy checking if we are seeing last header.
1049          * note that we do not visit this with protocols with pcb layer
1050          * code - like udp/tcp/raw ip.
1051          */
1052         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1053                 /*
1054                  * Check if the packet has already had IPsec processing
1055                  * done.  If so, then just pass it along.  This tag gets
1056                  * set during AH, ESP, etc. input handling, before the
1057                  * packet is returned to the ip input queue for delivery.
1058                  */
1059                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1060                 s = splnet();
1061                 if (mtag != NULL) {
1062                         tdbi = (struct tdb_ident *)(mtag + 1);
1063                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1064                 } else {
1065                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1066                                                    IP_FORWARDING, &error);
1067                 }
1068                 if (sp != NULL) {
1069                         /*
1070                          * Check security policy against packet attributes.
1071                          */
1072                         error = ipsec_in_reject(sp, m);
1073                         KEY_FREESP(&sp);
1074                 } else {
1075                         /* XXX error stat??? */
1076                         error = EINVAL;
1077 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1078                         goto bad;
1079                 }
1080                 splx(s);
1081                 if (error)
1082                         goto bad;
1083         }
1084 #endif /* FAST_IPSEC */
1085
1086         ipstat.ips_delivered++;
1087         if (needredispatch) {
1088                 struct netmsg_transport_packet *msg;
1089                 lwkt_port_t port;
1090
1091                 msg = malloc(sizeof(struct netmsg_transport_packet),
1092                                 M_LWKTMSG, M_INTWAIT | M_NULLOK);
1093                 if (msg == NULL)
1094                         goto bad;
1095
1096                 lwkt_initmsg(&msg->nm_lmsg, &netisr_afree_rport, 0,
1097                         lwkt_cmd_func(transport_processing_handler),
1098                         lwkt_cmd_op_none);
1099                 msg->nm_hlen = hlen;
1100                 msg->nm_hasnexthop = (args.next_hop != NULL);
1101                 if (msg->nm_hasnexthop)
1102                         msg->nm_nexthop = *args.next_hop;  /* structure copy */
1103
1104                 ip->ip_off = htons(ip->ip_off);
1105                 ip->ip_len = htons(ip->ip_len);
1106                 port = ip_mport(&m);
1107                 if (port) {
1108                     msg->nm_mbuf = m;
1109                     ip = mtod(m, struct ip *);
1110                     ip->ip_len = ntohs(ip->ip_len);
1111                     ip->ip_off = ntohs(ip->ip_off);
1112                     lwkt_sendmsg(port, &msg->nm_lmsg);
1113                 }
1114         } else {
1115                 transport_processing_oncpu(m, hlen, ip, args.next_hop);
1116         }
1117         return;
1118
1119 bad:
1120         m_freem(m);
1121 }
1122
1123 /*
1124  * Take incoming datagram fragment and try to reassemble it into
1125  * whole datagram.  If a chain for reassembly of this datagram already
1126  * exists, then it is given as fp; otherwise have to make a chain.
1127  *
1128  * When IPDIVERT enabled, keep additional state with each packet that
1129  * tells us if we need to divert or tee the packet we're building.
1130  * In particular, *divinfo includes the port and TEE flag,
1131  * *divert_rule is the number of the matching rule.
1132  */
1133
1134 static struct mbuf *
1135 ip_reass(struct mbuf *m, struct ipq *fp, struct ipq *where,
1136          u_int32_t *divinfo, u_int16_t *divert_rule)
1137 {
1138         struct ip *ip = mtod(m, struct ip *);
1139         struct mbuf *p = NULL, *q, *nq;
1140         struct mbuf *n;
1141         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1142         int i, next;
1143
1144         /*
1145          * Presence of header sizes in mbufs
1146          * would confuse code below.
1147          */
1148         m->m_data += hlen;
1149         m->m_len -= hlen;
1150
1151         /*
1152          * If first fragment to arrive, create a reassembly queue.
1153          */
1154         if (fp == NULL) {
1155                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1156                         goto dropfrag;
1157                 insque(fp, where);
1158                 nipq++;
1159                 fp->ipq_nfrags = 1;
1160                 fp->ipq_ttl = IPFRAGTTL;
1161                 fp->ipq_p = ip->ip_p;
1162                 fp->ipq_id = ip->ip_id;
1163                 fp->ipq_src = ip->ip_src;
1164                 fp->ipq_dst = ip->ip_dst;
1165                 fp->ipq_frags = m;
1166                 m->m_nextpkt = NULL;
1167 #ifdef IPDIVERT
1168                 fp->ipq_div_info = 0;
1169                 fp->ipq_div_cookie = 0;
1170 #endif
1171                 goto inserted;
1172         } else {
1173                 fp->ipq_nfrags++;
1174         }
1175
1176 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1177
1178         /*
1179          * Find a segment which begins after this one does.
1180          */
1181         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1182                 if (GETIP(q)->ip_off > ip->ip_off)
1183                         break;
1184
1185         /*
1186          * If there is a preceding segment, it may provide some of
1187          * our data already.  If so, drop the data from the incoming
1188          * segment.  If it provides all of our data, drop us, otherwise
1189          * stick new segment in the proper place.
1190          *
1191          * If some of the data is dropped from the the preceding
1192          * segment, then it's checksum is invalidated.
1193          */
1194         if (p) {
1195                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1196                 if (i > 0) {
1197                         if (i >= ip->ip_len)
1198                                 goto dropfrag;
1199                         m_adj(m, i);
1200                         m->m_pkthdr.csum_flags = 0;
1201                         ip->ip_off += i;
1202                         ip->ip_len -= i;
1203                 }
1204                 m->m_nextpkt = p->m_nextpkt;
1205                 p->m_nextpkt = m;
1206         } else {
1207                 m->m_nextpkt = fp->ipq_frags;
1208                 fp->ipq_frags = m;
1209         }
1210
1211         /*
1212          * While we overlap succeeding segments trim them or,
1213          * if they are completely covered, dequeue them.
1214          */
1215         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1216              q = nq) {
1217                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1218                 if (i < GETIP(q)->ip_len) {
1219                         GETIP(q)->ip_len -= i;
1220                         GETIP(q)->ip_off += i;
1221                         m_adj(q, i);
1222                         q->m_pkthdr.csum_flags = 0;
1223                         break;
1224                 }
1225                 nq = q->m_nextpkt;
1226                 m->m_nextpkt = nq;
1227                 ipstat.ips_fragdropped++;
1228                 fp->ipq_nfrags--;
1229                 m_freem(q);
1230         }
1231
1232 inserted:
1233
1234 #ifdef IPDIVERT
1235         /*
1236          * Transfer firewall instructions to the fragment structure.
1237          * Only trust info in the fragment at offset 0.
1238          */
1239         if (ip->ip_off == 0) {
1240                 fp->ipq_div_info = *divinfo;
1241                 fp->ipq_div_cookie = *divert_rule;
1242         }
1243         *divinfo = 0;
1244         *divert_rule = 0;
1245 #endif
1246
1247         /*
1248          * Check for complete reassembly and perform frag per packet
1249          * limiting.
1250          *
1251          * Frag limiting is performed here so that the nth frag has
1252          * a chance to complete the packet before we drop the packet.
1253          * As a result, n+1 frags are actually allowed per packet, but
1254          * only n will ever be stored. (n = maxfragsperpacket.)
1255          *
1256          */
1257         next = 0;
1258         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1259                 if (GETIP(q)->ip_off != next) {
1260                         if (fp->ipq_nfrags > maxfragsperpacket) {
1261                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1262                                 ip_freef(fp);
1263                         }
1264                         return (NULL);
1265                 }
1266                 next += GETIP(q)->ip_len;
1267         }
1268         /* Make sure the last packet didn't have the IP_MF flag */
1269         if (p->m_flags & M_FRAG) {
1270                 if (fp->ipq_nfrags > maxfragsperpacket) {
1271                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1272                         ip_freef(fp);
1273                 }
1274                 return (NULL);
1275         }
1276
1277         /*
1278          * Reassembly is complete.  Make sure the packet is a sane size.
1279          */
1280         q = fp->ipq_frags;
1281         ip = GETIP(q);
1282         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1283                 ipstat.ips_toolong++;
1284                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1285                 ip_freef(fp);
1286                 return (NULL);
1287         }
1288
1289         /*
1290          * Concatenate fragments.
1291          */
1292         m = q;
1293         n = m->m_next;
1294         m->m_next = NULL;
1295         m_cat(m, n);
1296         nq = q->m_nextpkt;
1297         q->m_nextpkt = NULL;
1298         for (q = nq; q != NULL; q = nq) {
1299                 nq = q->m_nextpkt;
1300                 q->m_nextpkt = NULL;
1301                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1302                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1303                 m_cat(m, q);
1304         }
1305
1306 #ifdef IPDIVERT
1307         /*
1308          * Extract firewall instructions from the fragment structure.
1309          */
1310         *divinfo = fp->ipq_div_info;
1311         *divert_rule = fp->ipq_div_cookie;
1312 #endif
1313
1314         /*
1315          * Create header for new ip packet by
1316          * modifying header of first packet;
1317          * dequeue and discard fragment reassembly header.
1318          * Make header visible.
1319          */
1320         ip->ip_len = next;
1321         ip->ip_src = fp->ipq_src;
1322         ip->ip_dst = fp->ipq_dst;
1323         remque(fp);
1324         nipq--;
1325         mpipe_free(&ipq_mpipe, fp);
1326         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1327         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1328         /* some debugging cruft by sklower, below, will go away soon */
1329         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1330                 int plen = 0;
1331
1332                 for (n = m; n; n = n->m_next)
1333                         plen += n->m_len;
1334                 m->m_pkthdr.len = plen;
1335         }
1336         return (m);
1337
1338 dropfrag:
1339 #ifdef IPDIVERT
1340         *divinfo = 0;
1341         *divert_rule = 0;
1342 #endif
1343         ipstat.ips_fragdropped++;
1344         if (fp != NULL)
1345                 fp->ipq_nfrags--;
1346         m_freem(m);
1347         return (NULL);
1348
1349 #undef GETIP
1350 }
1351
1352 /*
1353  * Free a fragment reassembly header and all
1354  * associated datagrams.
1355  */
1356 static void
1357 ip_freef(struct ipq *fp)
1358 {
1359         struct mbuf *q;
1360
1361         while (fp->ipq_frags) {
1362                 q = fp->ipq_frags;
1363                 fp->ipq_frags = q->m_nextpkt;
1364                 m_freem(q);
1365         }
1366         remque(fp);
1367         mpipe_free(&ipq_mpipe, fp);
1368         nipq--;
1369 }
1370
1371 /*
1372  * IP timer processing;
1373  * if a timer expires on a reassembly
1374  * queue, discard it.
1375  */
1376 void
1377 ip_slowtimo(void)
1378 {
1379         struct ipq *fp;
1380         int s = splnet();
1381         int i;
1382
1383         for (i = 0; i < IPREASS_NHASH; i++) {
1384                 fp = ipq[i].next;
1385                 if (fp == NULL)
1386                         continue;
1387                 while (fp != &ipq[i]) {
1388                         --fp->ipq_ttl;
1389                         fp = fp->next;
1390                         if (fp->prev->ipq_ttl == 0) {
1391                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1392                                 ip_freef(fp->prev);
1393                         }
1394                 }
1395         }
1396         /*
1397          * If we are over the maximum number of fragments
1398          * (due to the limit being lowered), drain off
1399          * enough to get down to the new limit.
1400          */
1401         if (maxnipq >= 0 && nipq > maxnipq) {
1402                 for (i = 0; i < IPREASS_NHASH; i++) {
1403                         while (nipq > maxnipq &&
1404                                 (ipq[i].next != &ipq[i])) {
1405                                 ipstat.ips_fragdropped +=
1406                                     ipq[i].next->ipq_nfrags;
1407                                 ip_freef(ipq[i].next);
1408                         }
1409                 }
1410         }
1411         ipflow_slowtimo();
1412         splx(s);
1413 }
1414
1415 /*
1416  * Drain off all datagram fragments.
1417  */
1418 void
1419 ip_drain(void)
1420 {
1421         int i;
1422
1423         for (i = 0; i < IPREASS_NHASH; i++) {
1424                 while (ipq[i].next != &ipq[i]) {
1425                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1426                         ip_freef(ipq[i].next);
1427                 }
1428         }
1429         in_rtqdrain();
1430 }
1431
1432 /*
1433  * Do option processing on a datagram,
1434  * possibly discarding it if bad options are encountered,
1435  * or forwarding it if source-routed.
1436  * The pass argument is used when operating in the IPSTEALTH
1437  * mode to tell what options to process:
1438  * [LS]SRR (pass 0) or the others (pass 1).
1439  * The reason for as many as two passes is that when doing IPSTEALTH,
1440  * non-routing options should be processed only if the packet is for us.
1441  * Returns 1 if packet has been forwarded/freed,
1442  * 0 if the packet should be processed further.
1443  */
1444 static int
1445 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1446 {
1447         struct ip *ip = mtod(m, struct ip *);
1448         u_char *cp;
1449         struct in_ifaddr *ia;
1450         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1451         boolean_t forward = FALSE;
1452         struct in_addr *sin, dst;
1453         n_time ntime;
1454
1455         dst = ip->ip_dst;
1456         cp = (u_char *)(ip + 1);
1457         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1458         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1459                 opt = cp[IPOPT_OPTVAL];
1460                 if (opt == IPOPT_EOL)
1461                         break;
1462                 if (opt == IPOPT_NOP)
1463                         optlen = 1;
1464                 else {
1465                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1466                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1467                                 goto bad;
1468                         }
1469                         optlen = cp[IPOPT_OLEN];
1470                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1471                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1472                                 goto bad;
1473                         }
1474                 }
1475                 switch (opt) {
1476
1477                 default:
1478                         break;
1479
1480                 /*
1481                  * Source routing with record.
1482                  * Find interface with current destination address.
1483                  * If none on this machine then drop if strictly routed,
1484                  * or do nothing if loosely routed.
1485                  * Record interface address and bring up next address
1486                  * component.  If strictly routed make sure next
1487                  * address is on directly accessible net.
1488                  */
1489                 case IPOPT_LSRR:
1490                 case IPOPT_SSRR:
1491                         if (ipstealth && pass > 0)
1492                                 break;
1493                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1494                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1495                                 goto bad;
1496                         }
1497                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1498                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1499                                 goto bad;
1500                         }
1501                         ipaddr.sin_addr = ip->ip_dst;
1502                         ia = (struct in_ifaddr *)
1503                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1504                         if (ia == NULL) {
1505                                 if (opt == IPOPT_SSRR) {
1506                                         type = ICMP_UNREACH;
1507                                         code = ICMP_UNREACH_SRCFAIL;
1508                                         goto bad;
1509                                 }
1510                                 if (!ip_dosourceroute)
1511                                         goto nosourcerouting;
1512                                 /*
1513                                  * Loose routing, and not at next destination
1514                                  * yet; nothing to do except forward.
1515                                  */
1516                                 break;
1517                         }
1518                         off--;                  /* 0 origin */
1519                         if (off > optlen - (int)sizeof(struct in_addr)) {
1520                                 /*
1521                                  * End of source route.  Should be for us.
1522                                  */
1523                                 if (!ip_acceptsourceroute)
1524                                         goto nosourcerouting;
1525                                 save_rte(cp, ip->ip_src);
1526                                 break;
1527                         }
1528                         if (ipstealth)
1529                                 goto dropit;
1530                         if (!ip_dosourceroute) {
1531                                 if (ipforwarding) {
1532                                         char buf[16]; /* aaa.bbb.ccc.ddd\0 */
1533                                         /*
1534                                          * Acting as a router, so generate ICMP
1535                                          */
1536 nosourcerouting:
1537                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1538                                         log(LOG_WARNING,
1539                                             "attempted source route from %s to %s\n",
1540                                             inet_ntoa(ip->ip_src), buf);
1541                                         type = ICMP_UNREACH;
1542                                         code = ICMP_UNREACH_SRCFAIL;
1543                                         goto bad;
1544                                 } else {
1545                                         /*
1546                                          * Not acting as a router,
1547                                          * so silently drop.
1548                                          */
1549 dropit:
1550                                         ipstat.ips_cantforward++;
1551                                         m_freem(m);
1552                                         return (1);
1553                                 }
1554                         }
1555
1556                         /*
1557                          * locate outgoing interface
1558                          */
1559                         (void)memcpy(&ipaddr.sin_addr, cp + off,
1560                             sizeof(ipaddr.sin_addr));
1561
1562                         if (opt == IPOPT_SSRR) {
1563 #define INA     struct in_ifaddr *
1564 #define SA      struct sockaddr *
1565                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1566                                                                         == NULL)
1567                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1568                         } else
1569                                 ia = ip_rtaddr(ipaddr.sin_addr, &ipforward_rt);
1570                         if (ia == NULL) {
1571                                 type = ICMP_UNREACH;
1572                                 code = ICMP_UNREACH_SRCFAIL;
1573                                 goto bad;
1574                         }
1575                         ip->ip_dst = ipaddr.sin_addr;
1576                         (void)memcpy(cp + off, &(IA_SIN(ia)->sin_addr),
1577                             sizeof(struct in_addr));
1578                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1579                         /*
1580                          * Let ip_intr's mcast routing check handle mcast pkts
1581                          */
1582                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1583                         break;
1584
1585                 case IPOPT_RR:
1586                         if (ipstealth && pass == 0)
1587                                 break;
1588                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1589                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1590                                 goto bad;
1591                         }
1592                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1593                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1594                                 goto bad;
1595                         }
1596                         /*
1597                          * If no space remains, ignore.
1598                          */
1599                         off--;                  /* 0 origin */
1600                         if (off > optlen - (int)sizeof(struct in_addr))
1601                                 break;
1602                         (void)memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1603                             sizeof(ipaddr.sin_addr));
1604                         /*
1605                          * locate outgoing interface; if we're the destination,
1606                          * use the incoming interface (should be same).
1607                          */
1608                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1609                             (ia = ip_rtaddr(ipaddr.sin_addr, &ipforward_rt))
1610                                                                      == NULL) {
1611                                 type = ICMP_UNREACH;
1612                                 code = ICMP_UNREACH_HOST;
1613                                 goto bad;
1614                         }
1615                         (void)memcpy(cp + off, &(IA_SIN(ia)->sin_addr),
1616                             sizeof(struct in_addr));
1617                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1618                         break;
1619
1620                 case IPOPT_TS:
1621                         if (ipstealth && pass == 0)
1622                                 break;
1623                         code = cp - (u_char *)ip;
1624                         if (optlen < 4 || optlen > 40) {
1625                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1626                                 goto bad;
1627                         }
1628                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1629                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1630                                 goto bad;
1631                         }
1632                         if (off > optlen - (int)sizeof(int32_t)) {
1633                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1634                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1635                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1636                                         goto bad;
1637                                 }
1638                                 break;
1639                         }
1640                         off--;                          /* 0 origin */
1641                         sin = (struct in_addr *)(cp + off);
1642                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1643
1644                         case IPOPT_TS_TSONLY:
1645                                 break;
1646
1647                         case IPOPT_TS_TSANDADDR:
1648                                 if (off + sizeof(n_time) +
1649                                     sizeof(struct in_addr) > optlen) {
1650                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1651                                         goto bad;
1652                                 }
1653                                 ipaddr.sin_addr = dst;
1654                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1655                                                             m->m_pkthdr.rcvif);
1656                                 if (ia == NULL)
1657                                         continue;
1658                                 (void)memcpy(sin, &IA_SIN(ia)->sin_addr,
1659                                     sizeof(struct in_addr));
1660                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1661                                 off += sizeof(struct in_addr);
1662                                 break;
1663
1664                         case IPOPT_TS_PRESPEC:
1665                                 if (off + sizeof(n_time) +
1666                                     sizeof(struct in_addr) > optlen) {
1667                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1668                                         goto bad;
1669                                 }
1670                                 (void)memcpy(&ipaddr.sin_addr, sin,
1671                                     sizeof(struct in_addr));
1672                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1673                                         continue;
1674                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1675                                 off += sizeof(struct in_addr);
1676                                 break;
1677
1678                         default:
1679                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1680                                 goto bad;
1681                         }
1682                         ntime = iptime();
1683                         (void)memcpy(cp + off, &ntime, sizeof(n_time));
1684                         cp[IPOPT_OFFSET] += sizeof(n_time);
1685                 }
1686         }
1687         if (forward && ipforwarding) {
1688                 ip_forward(m, 1, next_hop);
1689                 return (1);
1690         }
1691         return (0);
1692 bad:
1693         icmp_error(m, type, code, 0, NULL);
1694         ipstat.ips_badoptions++;
1695         return (1);
1696 }
1697
1698 /*
1699  * Given address of next destination (final or next hop),
1700  * return internet address info of interface to be used to get there.
1701  */
1702 struct in_ifaddr *
1703 ip_rtaddr(struct in_addr dst, struct route *rt)
1704 {
1705         struct sockaddr_in *sin;
1706
1707         sin = (struct sockaddr_in *)&rt->ro_dst;
1708
1709         if (rt->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1710                 if (rt->ro_rt != NULL) {
1711                         RTFREE(rt->ro_rt);
1712                         rt->ro_rt = NULL;
1713                 }
1714                 sin->sin_family = AF_INET;
1715                 sin->sin_len = sizeof(*sin);
1716                 sin->sin_addr = dst;
1717                 rtalloc_ign(rt, RTF_PRCLONING);
1718         }
1719
1720         if (rt->ro_rt == NULL)
1721                 return (NULL);
1722
1723         return (ifatoia(rt->ro_rt->rt_ifa));
1724 }
1725
1726 /*
1727  * Save incoming source route for use in replies,
1728  * to be picked up later by ip_srcroute if the receiver is interested.
1729  */
1730 void
1731 save_rte(u_char *option, struct in_addr dst)
1732 {
1733         unsigned olen;
1734
1735         olen = option[IPOPT_OLEN];
1736 #ifdef DIAGNOSTIC
1737         if (ipprintfs)
1738                 printf("save_rte: olen %d\n", olen);
1739 #endif
1740         if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst)))
1741                 return;
1742         bcopy(option, ip_srcrt.srcopt, olen);
1743         ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1744         ip_srcrt.dst = dst;
1745 }
1746
1747 /*
1748  * Retrieve incoming source route for use in replies,
1749  * in the same form used by setsockopt.
1750  * The first hop is placed before the options, will be removed later.
1751  */
1752 struct mbuf *
1753 ip_srcroute(void)
1754 {
1755         struct in_addr *p, *q;
1756         struct mbuf *m;
1757
1758         if (ip_nhops == 0)
1759                 return (NULL);
1760         m = m_get(MB_DONTWAIT, MT_HEADER);
1761         if (m == NULL)
1762                 return (NULL);
1763
1764 #define OPTSIZ  (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt))
1765
1766         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1767         m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) +
1768             OPTSIZ;
1769 #ifdef DIAGNOSTIC
1770         if (ipprintfs)
1771                 printf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
1772 #endif
1773
1774         /*
1775          * First save first hop for return route
1776          */
1777         p = &ip_srcrt.route[ip_nhops - 1];
1778         *(mtod(m, struct in_addr *)) = *p--;
1779 #ifdef DIAGNOSTIC
1780         if (ipprintfs)
1781                 printf(" hops %lx", ntohl(mtod(m, struct in_addr *)->s_addr));
1782 #endif
1783
1784         /*
1785          * Copy option fields and padding (nop) to mbuf.
1786          */
1787         ip_srcrt.nop = IPOPT_NOP;
1788         ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1789         (void)memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &ip_srcrt.nop,
1790             OPTSIZ);
1791         q = (struct in_addr *)(mtod(m, caddr_t) +
1792             sizeof(struct in_addr) + OPTSIZ);
1793 #undef OPTSIZ
1794         /*
1795          * Record return path as an IP source route,
1796          * reversing the path (pointers are now aligned).
1797          */
1798         while (p >= ip_srcrt.route) {
1799 #ifdef DIAGNOSTIC
1800                 if (ipprintfs)
1801                         printf(" %lx", ntohl(q->s_addr));
1802 #endif
1803                 *q++ = *p--;
1804         }
1805         /*
1806          * Last hop goes to final destination.
1807          */
1808         *q = ip_srcrt.dst;
1809 #ifdef DIAGNOSTIC
1810         if (ipprintfs)
1811                 printf(" %lx\n", ntohl(q->s_addr));
1812 #endif
1813         return (m);
1814 }
1815
1816 /*
1817  * Strip out IP options.
1818  */
1819 void
1820 ip_stripoptions(struct mbuf *m)
1821 {
1822         int datalen;
1823         struct ip *ip = mtod(m, struct ip *);
1824         caddr_t opts;
1825         int optlen;
1826
1827         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1828         opts = (caddr_t)(ip + 1);
1829         datalen = m->m_len - (sizeof(struct ip) + optlen);
1830         bcopy(opts + optlen, opts, datalen);
1831         m->m_len -= optlen;
1832         if (m->m_flags & M_PKTHDR)
1833                 m->m_pkthdr.len -= optlen;
1834         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1835 }
1836
1837 u_char inetctlerrmap[PRC_NCMDS] = {
1838         0,              0,              0,              0,
1839         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1840         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1841         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1842         0,              0,              0,              0,
1843         ENOPROTOOPT,    ECONNREFUSED
1844 };
1845
1846 /*
1847  * Forward a packet.  If some error occurs return the sender
1848  * an icmp packet.  Note we can't always generate a meaningful
1849  * icmp message because icmp doesn't have a large enough repertoire
1850  * of codes and types.
1851  *
1852  * If not forwarding, just drop the packet.  This could be confusing
1853  * if ipforwarding was zero but some routing protocol was advancing
1854  * us as a gateway to somewhere.  However, we must let the routing
1855  * protocol deal with that.
1856  *
1857  * The using_srcrt parameter indicates whether the packet is being forwarded
1858  * via a source route.
1859  */
1860 static void
1861 ip_forward(struct mbuf *m, int using_srcrt, struct sockaddr_in *next_hop)
1862 {
1863         struct ip *ip = mtod(m, struct ip *);
1864         struct sockaddr_in *sin;
1865         struct rtentry *rt;
1866         int error, type = 0, code = 0;
1867         struct mbuf *mcopy;
1868         n_long dest;
1869         struct in_addr pkt_dst;
1870         struct ifnet *destifp;
1871         struct m_hdr tag;
1872 #if defined(IPSEC) || defined(FAST_IPSEC)
1873         struct ifnet dummyifp;
1874 #endif
1875
1876         dest = 0;
1877         /*
1878          * Cache the destination address of the packet; this may be
1879          * changed by use of 'ipfw fwd'.
1880          */
1881         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
1882
1883 #ifdef DIAGNOSTIC
1884         if (ipprintfs)
1885                 printf("forward: src %x dst %x ttl %x\n",
1886                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1887 #endif
1888
1889         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1890                 ipstat.ips_cantforward++;
1891                 m_freem(m);
1892                 return;
1893         }
1894         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1895                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, NULL);
1896                 return;
1897         }
1898
1899         sin = (struct sockaddr_in *)&ipforward_rt.ro_dst;
1900         if ((rt = ipforward_rt.ro_rt) == NULL ||
1901             pkt_dst.s_addr != sin->sin_addr.s_addr) {
1902                 if (ipforward_rt.ro_rt != NULL) {
1903                         RTFREE(ipforward_rt.ro_rt);
1904                         ipforward_rt.ro_rt = NULL;
1905                 }
1906                 sin->sin_family = AF_INET;
1907                 sin->sin_len = sizeof(*sin);
1908                 sin->sin_addr = pkt_dst;
1909
1910                 rtalloc_ign(&ipforward_rt, RTF_PRCLONING);
1911                 if (ipforward_rt.ro_rt == NULL) {
1912                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest,
1913                                    NULL);
1914                         return;
1915                 }
1916                 rt = ipforward_rt.ro_rt;
1917         }
1918
1919         /*
1920          * Save the IP header and at most 8 bytes of the payload,
1921          * in case we need to generate an ICMP message to the src.
1922          *
1923          * XXX this can be optimized a lot by saving the data in a local
1924          * buffer on the stack (72 bytes at most), and only allocating the
1925          * mbuf if really necessary. The vast majority of the packets
1926          * are forwarded without having to send an ICMP back (either
1927          * because unnecessary, or because rate limited), so we are
1928          * really we are wasting a lot of work here.
1929          *
1930          * We don't use m_copy() because it might return a reference
1931          * to a shared cluster. Both this function and ip_output()
1932          * assume exclusive access to the IP header in `m', so any
1933          * data in a cluster may change before we reach icmp_error().
1934          */
1935         MGET(mcopy, MB_DONTWAIT, m->m_type);
1936         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1937                 /*
1938                  * It's probably ok if the pkthdr dup fails (because
1939                  * the deep copy of the tag chain failed), but for now
1940                  * be conservative and just discard the copy since
1941                  * code below may some day want the tags.
1942                  */
1943                 m_free(mcopy);
1944                 mcopy = NULL;
1945         }
1946         if (mcopy != NULL) {
1947                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1948                     (int)ip->ip_len);
1949                 mcopy->m_pkthdr.len = mcopy->m_len;
1950                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1951         }
1952
1953         if (!ipstealth)
1954                 ip->ip_ttl -= IPTTLDEC;
1955
1956         /*
1957          * If forwarding packet using same interface that it came in on,
1958          * perhaps should send a redirect to sender to shortcut a hop.
1959          * Only send redirect if source is sending directly to us,
1960          * and if packet was not source routed (or has any options).
1961          * Also, don't send redirect if forwarding using a default route
1962          * or a route modified by a redirect.
1963          */
1964         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1965             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1966             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1967             ipsendredirects && !using_srcrt && next_hop != NULL) {
1968                 u_long src = ntohl(ip->ip_src.s_addr);
1969
1970 #define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa))
1971                 if (RTA(rt) != NULL &&
1972                     (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1973                         if (rt->rt_flags & RTF_GATEWAY)
1974                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1975                         else
1976                                 dest = pkt_dst.s_addr;
1977                         /*
1978                          * Router requirements says to only send
1979                          * host redirects.
1980                          */
1981                         type = ICMP_REDIRECT;
1982                         code = ICMP_REDIRECT_HOST;
1983 #ifdef DIAGNOSTIC
1984                         if (ipprintfs)
1985                                 printf("redirect (%d) to %x\n", code, dest);
1986 #endif
1987                 }
1988         }
1989
1990         if (next_hop) {
1991                 /* Pass IPFORWARD info if available */
1992                 tag.mh_type = MT_TAG;
1993                 tag.mh_flags = PACKET_TAG_IPFORWARD;
1994                 tag.mh_data = (caddr_t)next_hop;
1995                 tag.mh_next = m;
1996                 m = (struct mbuf *)&tag;
1997         }
1998
1999         error = ip_output(m, NULL, &ipforward_rt, IP_FORWARDING, NULL, NULL);
2000
2001         if (error)
2002                 ipstat.ips_cantforward++;
2003         else {
2004                 ipstat.ips_forward++;
2005                 if (type)
2006                         ipstat.ips_redirectsent++;
2007                 else {
2008                         if (mcopy) {
2009                                 ipflow_create(&ipforward_rt, mcopy);
2010                                 m_freem(mcopy);
2011                         }
2012                         return;
2013                 }
2014         }
2015         if (mcopy == NULL)
2016                 return;
2017         destifp = NULL;
2018
2019         switch (error) {
2020
2021         case 0:                         /* forwarded, but need redirect */
2022                 /* type, code set above */
2023                 break;
2024
2025         case ENETUNREACH:               /* shouldn't happen, checked above */
2026         case EHOSTUNREACH:
2027         case ENETDOWN:
2028         case EHOSTDOWN:
2029         default:
2030                 type = ICMP_UNREACH;
2031                 code = ICMP_UNREACH_HOST;
2032                 break;
2033
2034         case EMSGSIZE:
2035                 type = ICMP_UNREACH;
2036                 code = ICMP_UNREACH_NEEDFRAG;
2037 #ifdef IPSEC
2038                 /*
2039                  * If the packet is routed over IPsec tunnel, tell the
2040                  * originator the tunnel MTU.
2041                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2042                  * XXX quickhack!!!
2043                  */
2044                 if (ipforward_rt.ro_rt != NULL) {
2045                         struct secpolicy *sp = NULL;
2046                         int ipsecerror;
2047                         int ipsechdr;
2048                         struct route *ro;
2049
2050                         sp = ipsec4_getpolicybyaddr(mcopy,
2051                                                     IPSEC_DIR_OUTBOUND,
2052                                                     IP_FORWARDING,
2053                                                     &ipsecerror);
2054
2055                         if (sp == NULL)
2056                                 destifp = ipforward_rt.ro_rt->rt_ifp;
2057                         else {
2058                                 /* count IPsec header size */
2059                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2060                                                          IPSEC_DIR_OUTBOUND,
2061                                                          NULL);
2062
2063                                 /*
2064                                  * find the correct route for outer IPv4
2065                                  * header, compute tunnel MTU.
2066                                  *
2067                                  * XXX BUG ALERT
2068                                  * The "dummyifp" code relies upon the fact
2069                                  * that icmp_error() touches only ifp->if_mtu.
2070                                  */
2071                                 /*XXX*/
2072                                 destifp = NULL;
2073                                 if (sp->req != NULL && sp->req->sav != NULL &&
2074                                     sp->req->sav->sah != NULL) {
2075                                         ro = &sp->req->sav->sah->sa_route;
2076                                         if (ro->ro_rt != NULL &&
2077                                             ro->ro_rt->rt_ifp != NULL) {
2078                                                 dummyifp.if_mtu =
2079                                                     ro->ro_rt->rt_ifp->if_mtu;
2080                                                 dummyifp.if_mtu -= ipsechdr;
2081                                                 destifp = &dummyifp;
2082                                         }
2083                                 }
2084
2085                                 key_freesp(sp);
2086                         }
2087                 }
2088 #elif FAST_IPSEC
2089                 /*
2090                  * If the packet is routed over IPsec tunnel, tell the
2091                  * originator the tunnel MTU.
2092                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2093                  * XXX quickhack!!!
2094                  */
2095                 if (ipforward_rt.ro_rt != NULL) {
2096                         struct secpolicy *sp = NULL;
2097                         int ipsecerror;
2098                         int ipsechdr;
2099                         struct route *ro;
2100
2101                         sp = ipsec_getpolicybyaddr(mcopy,
2102                                                    IPSEC_DIR_OUTBOUND,
2103                                                    IP_FORWARDING,
2104                                                    &ipsecerror);
2105
2106                         if (sp == NULL)
2107                                 destifp = ipforward_rt.ro_rt->rt_ifp;
2108                         else {
2109                                 /* count IPsec header size */
2110                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2111                                                          IPSEC_DIR_OUTBOUND,
2112                                                          NULL);
2113
2114                                 /*
2115                                  * find the correct route for outer IPv4
2116                                  * header, compute tunnel MTU.
2117                                  *
2118                                  * XXX BUG ALERT
2119                                  * The "dummyifp" code relies upon the fact
2120                                  * that icmp_error() touches only ifp->if_mtu.
2121                                  */
2122                                 /*XXX*/
2123                                 destifp = NULL;
2124                                 if (sp->req != NULL &&
2125                                     sp->req->sav != NULL &&
2126                                     sp->req->sav->sah != NULL) {
2127                                         ro = &sp->req->sav->sah->sa_route;
2128                                         if (ro->ro_rt != NULL &&
2129                                             ro->ro_rt->rt_ifp != NULL) {
2130                                                 dummyifp.if_mtu =
2131                                                     ro->ro_rt->rt_ifp->if_mtu;
2132                                                 dummyifp.if_mtu -= ipsechdr;
2133                                                 destifp = &dummyifp;
2134                                         }
2135                                 }
2136
2137                                 KEY_FREESP(&sp);
2138                         }
2139                 }
2140 #else /* !IPSEC && !FAST_IPSEC */
2141                 if (ipforward_rt.ro_rt != NULL)
2142                         destifp = ipforward_rt.ro_rt->rt_ifp;
2143 #endif /*IPSEC*/
2144                 ipstat.ips_cantfrag++;
2145                 break;
2146
2147         case ENOBUFS:
2148                 /*
2149                  * A router should not generate ICMP_SOURCEQUENCH as
2150                  * required in RFC1812 Requirements for IP Version 4 Routers.
2151                  * Source quench could be a big problem under DoS attacks,
2152                  * or if the underlying interface is rate-limited.
2153                  * Those who need source quench packets may re-enable them
2154                  * via the net.inet.ip.sendsourcequench sysctl.
2155                  */
2156                 if (!ip_sendsourcequench) {
2157                         m_freem(mcopy);
2158                         return;
2159                 } else {
2160                         type = ICMP_SOURCEQUENCH;
2161                         code = 0;
2162                 }
2163                 break;
2164
2165         case EACCES:                    /* ipfw denied packet */
2166                 m_freem(mcopy);
2167                 return;
2168         }
2169         icmp_error(mcopy, type, code, dest, destifp);
2170 }
2171
2172 void
2173 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2174                struct mbuf *m)
2175 {
2176         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2177                 struct timeval tv;
2178
2179                 microtime(&tv);
2180                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2181                     SCM_TIMESTAMP, SOL_SOCKET);
2182                 if (*mp)
2183                         mp = &(*mp)->m_next;
2184         }
2185         if (inp->inp_flags & INP_RECVDSTADDR) {
2186                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2187                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2188                 if (*mp)
2189                         mp = &(*mp)->m_next;
2190         }
2191 #ifdef notyet
2192         /* XXX
2193          * Moving these out of udp_input() made them even more broken
2194          * than they already were.
2195          */
2196         /* options were tossed already */
2197         if (inp->inp_flags & INP_RECVOPTS) {
2198                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2199                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2200                 if (*mp)
2201                         mp = &(*mp)->m_next;
2202         }
2203         /* ip_srcroute doesn't do what we want here, need to fix */
2204         if (inp->inp_flags & INP_RECVRETOPTS) {
2205                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(),
2206                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2207                 if (*mp)
2208                         mp = &(*mp)->m_next;
2209         }
2210 #endif
2211         if (inp->inp_flags & INP_RECVIF) {
2212                 struct ifnet *ifp;
2213                 struct sdlbuf {
2214                         struct sockaddr_dl sdl;
2215                         u_char  pad[32];
2216                 } sdlbuf;
2217                 struct sockaddr_dl *sdp;
2218                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2219
2220                 if (((ifp = m->m_pkthdr.rcvif)) &&
2221                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2222                         sdp = (struct sockaddr_dl *)
2223                             ifnet_addrs[ifp->if_index - 1]->ifa_addr;
2224                         /*
2225                          * Change our mind and don't try copy.
2226                          */
2227                         if ((sdp->sdl_family != AF_LINK) ||
2228                             (sdp->sdl_len > sizeof(sdlbuf))) {
2229                                 goto makedummy;
2230                         }
2231                         bcopy(sdp, sdl2, sdp->sdl_len);
2232                 } else {
2233 makedummy:
2234                         sdl2->sdl_len =
2235                             offsetof(struct sockaddr_dl, sdl_data[0]);
2236                         sdl2->sdl_family = AF_LINK;
2237                         sdl2->sdl_index = 0;
2238                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2239                 }
2240                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2241                         IP_RECVIF, IPPROTO_IP);
2242                 if (*mp)
2243                         mp = &(*mp)->m_next;
2244         }
2245 }
2246
2247 /*
2248  * XXX these routines are called from the upper part of the kernel.
2249  *
2250  * They could also be moved to ip_mroute.c, since all the RSVP
2251  *  handling is done there already.
2252  */
2253 int
2254 ip_rsvp_init(struct socket *so)
2255 {
2256         if (so->so_type != SOCK_RAW ||
2257             so->so_proto->pr_protocol != IPPROTO_RSVP)
2258                 return EOPNOTSUPP;
2259
2260         if (ip_rsvpd != NULL)
2261                 return EADDRINUSE;
2262
2263         ip_rsvpd = so;
2264         /*
2265          * This may seem silly, but we need to be sure we don't over-increment
2266          * the RSVP counter, in case something slips up.
2267          */
2268         if (!ip_rsvp_on) {
2269                 ip_rsvp_on = 1;
2270                 rsvp_on++;
2271         }
2272
2273         return 0;
2274 }
2275
2276 int
2277 ip_rsvp_done(void)
2278 {
2279         ip_rsvpd = NULL;
2280         /*
2281          * This may seem silly, but we need to be sure we don't over-decrement
2282          * the RSVP counter, in case something slips up.
2283          */
2284         if (ip_rsvp_on) {
2285                 ip_rsvp_on = 0;
2286                 rsvp_on--;
2287         }
2288         return 0;
2289 }
2290
2291 void
2292 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2293 {
2294         int off, proto;
2295         __va_list ap;
2296
2297         __va_start(ap, m);
2298         off = __va_arg(ap, int);
2299         proto = __va_arg(ap, int);
2300         __va_end(ap);
2301
2302         if (rsvp_input_p) { /* call the real one if loaded */
2303                 rsvp_input_p(m, off, proto);
2304                 return;
2305         }
2306
2307         /* Can still get packets with rsvp_on = 0 if there is a local member
2308          * of the group to which the RSVP packet is addressed.  But in this
2309          * case we want to throw the packet away.
2310          */
2311
2312         if (!rsvp_on) {
2313                 m_freem(m);
2314                 return;
2315         }
2316
2317         if (ip_rsvpd != NULL) {
2318                 rip_input(m, off, proto);
2319                 return;
2320         }
2321         /* Drop the packet */
2322         m_freem(m);
2323 }