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) 1982, 1986, 1988, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. All advertising materials mentioning features or use of this software
  47  *    must display the following acknowledgement:
  48  *      This product includes software developed by the University of
  49  *      California, Berkeley and its contributors.
  50  * 4. Neither the name of the University nor the names of its contributors
  51  *    may be used to endorse or promote products derived from this software
  52  *    without specific prior written permission.
  53  *
  54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  64  * SUCH DAMAGE.
  65  *
  66  *      @(#)ip_input.c  8.2 (Berkeley) 1/4/94
  67  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
  68  * $DragonFly: src/sys/netinet/ip_input.c,v 1.88 2008/08/22 11:58:49 sephe Exp $
  69  */
  70
  71 #define _IP_VHL
  72
  73 #include "opt_bootp.h"
  74 #include "opt_ipfw.h"
  75 #include "opt_ipdn.h"
  76 #include "opt_ipdivert.h"
  77 #include "opt_ipfilter.h"
  78 #include "opt_ipstealth.h"
  79 #include "opt_ipsec.h"
  80
  81 #include <sys/param.h>
  82 #include <sys/systm.h>
  83 #include <sys/mbuf.h>
  84 #include <sys/malloc.h>
  85 #include <sys/mpipe.h>
  86 #include <sys/domain.h>
  87 #include <sys/protosw.h>
  88 #include <sys/socket.h>
  89 #include <sys/time.h>
  90 #include <sys/globaldata.h>
  91 #include <sys/thread.h>
  92 #include <sys/kernel.h>
  93 #include <sys/syslog.h>
  94 #include <sys/sysctl.h>
  95 #include <sys/in_cksum.h>
  96
  97 #include <machine/stdarg.h>
  98
  99 #include <net/if.h>
 100 #include <net/if_types.h>
 101 #include <net/if_var.h>
 102 #include <net/if_dl.h>
 103 #include <net/pfil.h>
 104 #include <net/route.h>
 105 #include <net/netisr.h>
 106
 107 #include <netinet/in.h>
 108 #include <netinet/in_systm.h>
 109 #include <netinet/in_var.h>
 110 #include <netinet/ip.h>
 111 #include <netinet/in_pcb.h>
 112 #include <netinet/ip_var.h>
 113 #include <netinet/ip_icmp.h>
 114
 115 #include <sys/thread2.h>
 116 #include <sys/msgport2.h>
 117 #include <net/netmsg2.h>
 118
 119 #include <sys/socketvar.h>
 120
 121 #include <net/ipfw/ip_fw.h>
 122 #include <net/dummynet/ip_dummynet.h>
 123
 124 #ifdef IPSEC
 125 #include <netinet6/ipsec.h>
 126 #include <netproto/key/key.h>
 127 #endif
 128
 129 #ifdef FAST_IPSEC
 130 #include <netproto/ipsec/ipsec.h>
 131 #include <netproto/ipsec/key.h>
 132 #endif
 133
 134 int rsvp_on = 0;
 135 static int ip_rsvp_on;
 136 struct socket *ip_rsvpd;
 137
 138 int ipforwarding = 0;
 139 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 140     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 141
 142 static int ipsendredirects = 1; /* XXX */
 143 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 144     &ipsendredirects, 0, "Enable sending IP redirects");
 145
 146 int ip_defttl = IPDEFTTL;
 147 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 148     &ip_defttl, 0, "Maximum TTL on IP packets");
 149
 150 static int ip_dosourceroute = 0;
 151 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 152     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 153
 154 static int ip_acceptsourceroute = 0;
 155 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 156     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 157     "Enable accepting source routed IP packets");
 158
 159 static int ip_keepfaith = 0;
 160 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 161     &ip_keepfaith, 0,
 162     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 163
 164 static int nipq = 0;    /* total # of reass queues */
 165 static int maxnipq;
 166 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 167     &maxnipq, 0,
 168     "Maximum number of IPv4 fragment reassembly queue entries");
 169
 170 static int maxfragsperpacket;
 171 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 172     &maxfragsperpacket, 0,
 173     "Maximum number of IPv4 fragments allowed per packet");
 174
 175 static int ip_sendsourcequench = 0;
 176 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 177     &ip_sendsourcequench, 0,
 178     "Enable the transmission of source quench packets");
 179
 180 int ip_do_randomid = 1;
 181 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 182     &ip_do_randomid, 0,
 183     "Assign random ip_id values");
 184 /*
 185  * XXX - Setting ip_checkinterface mostly implements the receive side of
 186  * the Strong ES model described in RFC 1122, but since the routing table
 187  * and transmit implementation do not implement the Strong ES model,
 188  * setting this to 1 results in an odd hybrid.
 189  *
 190  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 191  * to translate the destination address to another local interface.
 192  *
 193  * XXX - ip_checkinterface must be disabled if you add IP aliases
 194  * to the loopback interface instead of the interface where the
 195  * packets for those addresses are received.
 196  */
 197 static int ip_checkinterface = 0;
 198 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 199     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 200
 201 #ifdef DIAGNOSTIC
 202 static int ipprintfs = 0;
 203 #endif
 204
 205 extern  struct domain inetdomain;
 206 extern  struct protosw inetsw[];
 207 u_char  ip_protox[IPPROTO_MAX];
 208 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 209 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 210                                                 /* inet addr hash table */
 211 u_long  in_ifaddrhmask;                         /* mask for hash table */
 212
 213 struct ip_stats ipstats_percpu[MAXCPU];
 214 #ifdef SMP
 215 static int
 216 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 217 {
 218         int cpu, error = 0;
 219
 220         for (cpu = 0; cpu < ncpus; ++cpu) {
 221                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 222                                         sizeof(struct ip_stats))))
 223                         break;
 224                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 225                                        sizeof(struct ip_stats))))
 226                         break;
 227         }
 228
 229         return (error);
 230 }
 231 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 232     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 233 #else
 234 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 235     &ipstat, ip_stats, "IP statistics");
 236 #endif
 237
 238 /* Packet reassembly stuff */
 239 #define IPREASS_NHASH_LOG2      6
 240 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 241 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 242 #define IPREASS_HASH(x,y)                                               \
 243     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 244
 245 static struct ipq ipq[IPREASS_NHASH];
 246
 247 #ifdef IPCTL_DEFMTU
 248 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 249     &ip_mtu, 0, "Default MTU");
 250 #endif
 251
 252 #ifdef IPSTEALTH
 253 static int ipstealth = 0;
 254 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 255 #else
 256 static const int ipstealth = 0;
 257 #endif
 258
 259
 260 /* Firewall hooks */
 261 ip_fw_chk_t *ip_fw_chk_ptr;
 262 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
 263 int ip_fw_loaded;
 264 int fw_enable = 1;
 265 int fw_one_pass = 1;
 266
 267 struct pfil_head inet_pfil_hook;
 268
 269 /*
 270  * XXX this is ugly -- the following two global variables are
 271  * used to store packet state while it travels through the stack.
 272  * Note that the code even makes assumptions on the size and
 273  * alignment of fields inside struct ip_srcrt so e.g. adding some
 274  * fields will break the code. This needs to be fixed.
 275  *
 276  * We need to save the IP options in case a protocol wants to respond
 277  * to an incoming packet over the same route if the packet got here
 278  * using IP source routing.  This allows connection establishment and
 279  * maintenance when the remote end is on a network that is not known
 280  * to us.
 281  */
 282 static int ip_nhops = 0;
 283
 284 static  struct ip_srcrt {
 285         struct  in_addr dst;                    /* final destination */
 286         char    nop;                            /* one NOP to align */
 287         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 288         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 289 } ip_srcrt;
 290
 291 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 292 static struct malloc_pipe ipq_mpipe;
 293
 294 static void             save_rte(u_char *, struct in_addr);
 295 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 296 static void             ip_freef(struct ipq *);
 297 static void             ip_input_handler(struct netmsg *);
 298 static struct mbuf      *ip_reass(struct mbuf *, struct ipq *, struct ipq *,
 299                                   u_int32_t *);
 300
 301 /*
 302  * IP initialization: fill in IP protocol switch table.
 303  * All protocols not implemented in kernel go to raw IP protocol handler.
 304  */
 305 void
 306 ip_init(void)
 307 {
 308         struct protosw *pr;
 309         int i;
 310 #ifdef SMP
 311         int cpu;
 312 #endif
 313
 314         /*
 315          * Make sure we can handle a reasonable number of fragments but
 316          * cap it at 4000 (XXX).
 317          */
 318         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 319                     IFQ_MAXLEN, 4000, 0, NULL);
 320         for (i = 0; i < ncpus; ++i) {
 321                 TAILQ_INIT(&in_ifaddrheads[i]);
 322                 in_ifaddrhashtbls[i] =
 323                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 324         }
 325         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 326         if (pr == NULL)
 327                 panic("ip_init");
 328         for (i = 0; i < IPPROTO_MAX; i++)
 329                 ip_protox[i] = pr - inetsw;
 330         for (pr = inetdomain.dom_protosw;
 331              pr < inetdomain.dom_protoswNPROTOSW; pr++)
 332                 if (pr->pr_domain->dom_family == PF_INET &&
 333                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
 334                         ip_protox[pr->pr_protocol] = pr - inetsw;
 335
 336         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 337         inet_pfil_hook.ph_af = AF_INET;
 338         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 339                 kprintf("%s: WARNING: unable to register pfil hook, "
 340                         "error %d\n", __func__, i);
 341         }
 342
 343         for (i = 0; i < IPREASS_NHASH; i++)
 344             ipq[i].next = ipq[i].prev = &ipq[i];
 345
 346         maxnipq = nmbclusters / 32;
 347         maxfragsperpacket = 16;
 348
 349         ip_id = time_second & 0xffff;
 350
 351         /*
 352          * Initialize IP statistics counters for each CPU.
 353          *
 354          */
 355 #ifdef SMP
 356         for (cpu = 0; cpu < ncpus; ++cpu) {
 357                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 358         }
 359 #else
 360         bzero(&ipstat, sizeof(struct ip_stats));
 361 #endif
 362
 363         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
 364 }
 365
 366 /*
 367  * XXX watch out this one. It is perhaps used as a cache for
 368  * the most recently used route ? it is cleared in in_addroute()
 369  * when a new route is successfully created.
 370  */
 371 struct route ipforward_rt[MAXCPU];
 372
 373 /* Do transport protocol processing. */
 374 static void
 375 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 376 {
 377         /*
 378          * Switch out to protocol's input routine.
 379          */
 380         (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
 381 }
 382
 383 static void
 384 transport_processing_handler(netmsg_t netmsg)
 385 {
 386         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 387         struct ip *ip;
 388         int hlen;
 389
 390         ip = mtod(pmsg->nm_packet, struct ip *);
 391         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 392
 393         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 394         /* netmsg was embedded in the mbuf, do not reply! */
 395 }
 396
 397 static void
 398 ip_input_handler(struct netmsg *msg0)
 399 {
 400         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 401
 402         ip_input(m);
 403         /* msg0 was embedded in the mbuf, do not reply! */
 404 }
 405
 406 /*
 407  * IP input routine.  Checksum and byte swap header.  If fragmented
 408  * try to reassemble.  Process options.  Pass to next level.
 409  */
 410 void
 411 ip_input(struct mbuf *m)
 412 {
 413         struct ip *ip;
 414         struct ipq *fp;
 415         struct in_ifaddr *ia = NULL;
 416         struct in_ifaddr_container *iac;
 417         int i, hlen, checkif;
 418         u_short sum;
 419         struct in_addr pkt_dst;
 420         u_int32_t divert_info = 0;              /* packet divert/tee info */
 421         struct ip_fw_args args;
 422         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 423         boolean_t needredispatch = FALSE;
 424         struct in_addr odst;                    /* original dst address(NAT) */
 425         struct m_tag *mtag;
 426         struct sockaddr_in *next_hop = NULL;
 427 #ifdef FAST_IPSEC
 428         struct tdb_ident *tdbi;
 429         struct secpolicy *sp;
 430         int error;
 431 #endif
 432
 433         args.eh = NULL;
 434         args.oif = NULL;
 435         args.rule = NULL;
 436
 437         M_ASSERTPKTHDR(m);
 438
 439         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 440                 /* Next hop */
 441                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 442                 KKASSERT(mtag != NULL);
 443                 next_hop = m_tag_data(mtag);
 444         }
 445
 446         /* Extract info from dummynet tag */
 447         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 448         if (mtag != NULL) {
 449                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
 450                 m_tag_delete(m, mtag);
 451         }
 452
 453         if (args.rule != NULL) {        /* dummynet already filtered us */
 454                 ip = mtod(m, struct ip *);
 455                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 456                 goto iphack;
 457         }
 458
 459         ipstat.ips_total++;
 460
 461         /* length checks already done in ip_demux() */
 462         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 463
 464         ip = mtod(m, struct ip *);
 465
 466         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 467                 ipstat.ips_badvers++;
 468                 goto bad;
 469         }
 470
 471         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 472         /* length checks already done in ip_demux() */
 473         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 474         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 475
 476         /* 127/8 must not appear on wire - RFC1122 */
 477         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 478             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 479                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 480                         ipstat.ips_badaddr++;
 481                         goto bad;
 482                 }
 483         }
 484
 485         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 486                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 487         } else {
 488                 if (hlen == sizeof(struct ip)) {
 489                         sum = in_cksum_hdr(ip);
 490                 } else {
 491                         sum = in_cksum(m, hlen);
 492                 }
 493         }
 494         if (sum != 0) {
 495                 ipstat.ips_badsum++;
 496                 goto bad;
 497         }
 498
 499 #ifdef ALTQ
 500         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 501                 /* packet is dropped by traffic conditioner */
 502                 return;
 503         }
 504 #endif
 505         /*
 506          * Convert fields to host representation.
 507          */
 508         ip->ip_len = ntohs(ip->ip_len);
 509         if (ip->ip_len < hlen) {
 510                 ipstat.ips_badlen++;
 511                 goto bad;
 512         }
 513         ip->ip_off = ntohs(ip->ip_off);
 514
 515         /*
 516          * Check that the amount of data in the buffers
 517          * is as at least much as the IP header would have us expect.
 518          * Trim mbufs if longer than we expect.
 519          * Drop packet if shorter than we expect.
 520          */
 521         if (m->m_pkthdr.len < ip->ip_len) {
 522                 ipstat.ips_tooshort++;
 523                 goto bad;
 524         }
 525         if (m->m_pkthdr.len > ip->ip_len) {
 526                 if (m->m_len == m->m_pkthdr.len) {
 527                         m->m_len = ip->ip_len;
 528                         m->m_pkthdr.len = ip->ip_len;
 529                 } else
 530                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 531         }
 532 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 533         /*
 534          * Bypass packet filtering for packets from a tunnel (gif).
 535          */
 536         if (ipsec_gethist(m, NULL))
 537                 goto pass;
 538 #endif
 539
 540         /*
 541          * IpHack's section.
 542          * Right now when no processing on packet has done
 543          * and it is still fresh out of network we do our black
 544          * deals with it.
 545          * - Firewall: deny/allow/divert
 546          * - Xlate: translate packet's addr/port (NAT).
 547          * - Pipe: pass pkt through dummynet.
 548          * - Wrap: fake packet's addr/port <unimpl.>
 549          * - Encapsulate: put it in another IP and send out. <unimp.>
 550          */
 551
 552 iphack:
 553
 554         /*
 555          * Run through list of hooks for input packets.
 556          *
 557          * NB: Beware of the destination address changing (e.g.
 558          *     by NAT rewriting). When this happens, tell
 559          *     ip_forward to do the right thing.
 560          */
 561         if (pfil_has_hooks(&inet_pfil_hook)) {
 562                 odst = ip->ip_dst;
 563                 if (pfil_run_hooks(&inet_pfil_hook, &m,
 564                     m->m_pkthdr.rcvif, PFIL_IN)) {
 565                         return;
 566                 }
 567                 if (m == NULL)                  /* consumed by filter */
 568                         return;
 569                 ip = mtod(m, struct ip *);
 570                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 571         }
 572
 573         if (fw_enable && IPFW_LOADED) {
 574                 /*
 575                  * If we've been forwarded from the output side, then
 576                  * skip the firewall a second time
 577                  */
 578                 if (next_hop != NULL)
 579                         goto ours;
 580
 581                 args.m = m;
 582                 i = ip_fw_chk_ptr(&args);
 583                 m = args.m;
 584
 585                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
 586                         if (m != NULL)
 587                                 m_freem(m);
 588                         return;
 589                 }
 590                 ip = mtod(m, struct ip *);      /* just in case m changed */
 591
 592                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 593                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 594                         KKASSERT(mtag != NULL);
 595                         next_hop = m_tag_data(mtag);
 596                 }
 597
 598                 if (i == 0 && next_hop == NULL) /* common case */
 599                         goto pass;
 600                 if (i & IP_FW_PORT_DYNT_FLAG) {
 601                         /* Send packet to the appropriate pipe */
 602                         ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
 603                         return;
 604                 }
 605 #ifdef IPDIVERT
 606                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
 607                         /* Divert or tee packet */
 608                         divert_info = i;
 609                         goto ours;
 610                 }
 611 #endif
 612                 if (i == 0 && next_hop != NULL)
 613                         goto pass;
 614                 /*
 615                  * if we get here, the packet must be dropped
 616                  */
 617                 m_freem(m);
 618                 return;
 619         }
 620 pass:
 621
 622         /*
 623          * Process options and, if not destined for us,
 624          * ship it on.  ip_dooptions returns 1 when an
 625          * error was detected (causing an icmp message
 626          * to be sent and the original packet to be freed).
 627          */
 628         ip_nhops = 0;           /* for source routed packets */
 629         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 630                 return;
 631
 632         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 633          * matter if it is destined to another node, or whether it is
 634          * a multicast one, RSVP wants it! and prevents it from being forwarded
 635          * anywhere else. Also checks if the rsvp daemon is running before
 636          * grabbing the packet.
 637          */
 638         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 639                 goto ours;
 640
 641         /*
 642          * Check our list of addresses, to see if the packet is for us.
 643          * If we don't have any addresses, assume any unicast packet
 644          * we receive might be for us (and let the upper layers deal
 645          * with it).
 646          */
 647         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 648             !(m->m_flags & (M_MCAST | M_BCAST)))
 649                 goto ours;
 650
 651         /*
 652          * Cache the destination address of the packet; this may be
 653          * changed by use of 'ipfw fwd'.
 654          */
 655         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 656
 657         /*
 658          * Enable a consistency check between the destination address
 659          * and the arrival interface for a unicast packet (the RFC 1122
 660          * strong ES model) if IP forwarding is disabled and the packet
 661          * is not locally generated and the packet is not subject to
 662          * 'ipfw fwd'.
 663          *
 664          * XXX - Checking also should be disabled if the destination
 665          * address is ipnat'ed to a different interface.
 666          *
 667          * XXX - Checking is incompatible with IP aliases added
 668          * to the loopback interface instead of the interface where
 669          * the packets are received.
 670          */
 671         checkif = ip_checkinterface &&
 672                   !ipforwarding &&
 673                   m->m_pkthdr.rcvif != NULL &&
 674                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 675                   next_hop == NULL;
 676
 677         /*
 678          * Check for exact addresses in the hash bucket.
 679          */
 680         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 681                 ia = iac->ia;
 682
 683                 /*
 684                  * If the address matches, verify that the packet
 685                  * arrived via the correct interface if checking is
 686                  * enabled.
 687                  */
 688                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 689                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 690                         goto ours;
 691         }
 692         ia = NULL;
 693
 694         /*
 695          * Check for broadcast addresses.
 696          *
 697          * Only accept broadcast packets that arrive via the matching
 698          * interface.  Reception of forwarded directed broadcasts would
 699          * be handled via ip_forward() and ether_output() with the loopback
 700          * into the stack for SIMPLEX interfaces handled by ether_output().
 701          */
 702         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 703                 struct ifaddr_container *ifac;
 704
 705                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 706                               ifa_link) {
 707                         struct ifaddr *ifa = ifac->ifa;
 708
 709                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 710                                 continue;
 711                         if (ifa->ifa_addr->sa_family != AF_INET)
 712                                 continue;
 713                         ia = ifatoia(ifa);
 714                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 715                                                                 pkt_dst.s_addr)
 716                                 goto ours;
 717                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 718                                 goto ours;
 719 #ifdef BOOTP_COMPAT
 720                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 721                                 goto ours;
 722 #endif
 723                 }
 724         }
 725         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 726                 struct in_multi *inm;
 727
 728                 if (ip_mrouter != NULL) {
 729                         /*
 730                          * If we are acting as a multicast router, all
 731                          * incoming multicast packets are passed to the
 732                          * kernel-level multicast forwarding function.
 733                          * The packet is returned (relatively) intact; if
 734                          * ip_mforward() returns a non-zero value, the packet
 735                          * must be discarded, else it may be accepted below.
 736                          */
 737                         if (ip_mforward != NULL &&
 738                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 739                                 ipstat.ips_cantforward++;
 740                                 m_freem(m);
 741                                 return;
 742                         }
 743
 744                         /*
 745                          * The process-level routing daemon needs to receive
 746                          * all multicast IGMP packets, whether or not this
 747                          * host belongs to their destination groups.
 748                          */
 749                         if (ip->ip_p == IPPROTO_IGMP)
 750                                 goto ours;
 751                         ipstat.ips_forward++;
 752                 }
 753                 /*
 754                  * See if we belong to the destination multicast group on the
 755                  * arrival interface.
 756                  */
 757                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 758                 if (inm == NULL) {
 759                         ipstat.ips_notmember++;
 760                         m_freem(m);
 761                         return;
 762                 }
 763                 goto ours;
 764         }
 765         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 766                 goto ours;
 767         if (ip->ip_dst.s_addr == INADDR_ANY)
 768                 goto ours;
 769
 770         /*
 771          * FAITH(Firewall Aided Internet Translator)
 772          */
 773         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 774                 if (ip_keepfaith) {
 775                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 776                                 goto ours;
 777                 }
 778                 m_freem(m);
 779                 return;
 780         }
 781
 782         /*
 783          * Not for us; forward if possible and desirable.
 784          */
 785         if (!ipforwarding) {
 786                 ipstat.ips_cantforward++;
 787                 m_freem(m);
 788         } else {
 789 #ifdef IPSEC
 790                 /*
 791                  * Enforce inbound IPsec SPD.
 792                  */
 793                 if (ipsec4_in_reject(m, NULL)) {
 794                         ipsecstat.in_polvio++;
 795                         goto bad;
 796                 }
 797 #endif
 798 #ifdef FAST_IPSEC
 799                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 800                 crit_enter();
 801                 if (mtag != NULL) {
 802                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 803                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 804                 } else {
 805                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 806                                                    IP_FORWARDING, &error);
 807                 }
 808                 if (sp == NULL) {       /* NB: can happen if error */
 809                         crit_exit();
 810                         /*XXX error stat???*/
 811                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 812                         goto bad;
 813                 }
 814
 815                 /*
 816                  * Check security policy against packet attributes.
 817                  */
 818                 error = ipsec_in_reject(sp, m);
 819                 KEY_FREESP(&sp);
 820                 crit_exit();
 821                 if (error) {
 822                         ipstat.ips_cantforward++;
 823                         goto bad;
 824                 }
 825 #endif
 826                 ip_forward(m, using_srcrt, next_hop);
 827         }
 828         return;
 829
 830 ours:
 831
 832         /*
 833          * IPSTEALTH: Process non-routing options only
 834          * if the packet is destined for us.
 835          */
 836         if (ipstealth &&
 837             hlen > sizeof(struct ip) &&
 838             ip_dooptions(m, 1, next_hop))
 839                 return;
 840
 841         /* Count the packet in the ip address stats */
 842         if (ia != NULL) {
 843                 ia->ia_ifa.if_ipackets++;
 844                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 845         }
 846
 847         /*
 848          * If offset or IP_MF are set, must reassemble.
 849          * Otherwise, nothing need be done.
 850          * (We could look in the reassembly queue to see
 851          * if the packet was previously fragmented,
 852          * but it's not worth the time; just let them time out.)
 853          */
 854         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 855
 856                 /* If maxnipq is 0, never accept fragments. */
 857                 if (maxnipq == 0) {
 858                         ipstat.ips_fragments++;
 859                         ipstat.ips_fragdropped++;
 860                         goto bad;
 861                 }
 862
 863                 sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 864                 /*
 865                  * Look for queue of fragments
 866                  * of this datagram.
 867                  */
 868                 for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 869                         if (ip->ip_id == fp->ipq_id &&
 870                             ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 871                             ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 872                             ip->ip_p == fp->ipq_p)
 873                                 goto found;
 874
 875                 fp = NULL;
 876
 877                 /*
 878                  * Enforce upper bound on number of fragmented packets
 879                  * for which we attempt reassembly;
 880                  * If maxnipq is -1, accept all fragments without limitation.
 881                  */
 882                 if ((nipq > maxnipq) && (maxnipq > 0)) {
 883                         /*
 884                          * drop something from the tail of the current queue
 885                          * before proceeding further
 886                          */
 887                         if (ipq[sum].prev == &ipq[sum]) {   /* gak */
 888                                 for (i = 0; i < IPREASS_NHASH; i++) {
 889                                         if (ipq[i].prev != &ipq[i]) {
 890                                                 ipstat.ips_fragtimeout +=
 891                                                     ipq[i].prev->ipq_nfrags;
 892                                                 ip_freef(ipq[i].prev);
 893                                                 break;
 894                                         }
 895                                 }
 896                         } else {
 897                                 ipstat.ips_fragtimeout +=
 898                                     ipq[sum].prev->ipq_nfrags;
 899                                 ip_freef(ipq[sum].prev);
 900                         }
 901                 }
 902 found:
 903                 /*
 904                  * Adjust ip_len to not reflect header,
 905                  * convert offset of this to bytes.
 906                  */
 907                 ip->ip_len -= hlen;
 908                 if (ip->ip_off & IP_MF) {
 909                         /*
 910                          * Make sure that fragments have a data length
 911                          * that's a non-zero multiple of 8 bytes.
 912                          */
 913                         if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
 914                                 ipstat.ips_toosmall++; /* XXX */
 915                                 goto bad;
 916                         }
 917                         m->m_flags |= M_FRAG;
 918                 } else
 919                         m->m_flags &= ~M_FRAG;
 920                 ip->ip_off <<= 3;
 921
 922                 /*
 923                  * Attempt reassembly; if it succeeds, proceed.
 924                  * ip_reass() will return a different mbuf, and update
 925                  * the divert info in divert_info.
 926                  */
 927                 ipstat.ips_fragments++;
 928                 m->m_pkthdr.header = ip;
 929                 m = ip_reass(m, fp, &ipq[sum], &divert_info);
 930                 if (m == NULL)
 931                         return;
 932
 933                 needredispatch = TRUE;
 934                 ip = mtod(m, struct ip *);
 935                 /* Get the header length of the reassembled packet */
 936                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 937 #ifdef IPDIVERT
 938                 /* Restore original checksum before diverting packet */
 939                 if (divert_info != 0) {
 940                         ip->ip_len += hlen;
 941                         ip->ip_len = htons(ip->ip_len);
 942                         ip->ip_off = htons(ip->ip_off);
 943                         ip->ip_sum = 0;
 944                         if (hlen == sizeof(struct ip))
 945                                 ip->ip_sum = in_cksum_hdr(ip);
 946                         else
 947                                 ip->ip_sum = in_cksum(m, hlen);
 948                         ip->ip_off = ntohs(ip->ip_off);
 949                         ip->ip_len = ntohs(ip->ip_len);
 950                         ip->ip_len -= hlen;
 951                 }
 952 #endif
 953         } else {
 954                 ip->ip_len -= hlen;
 955         }
 956
 957 #ifdef IPDIVERT
 958         /*
 959          * Divert or tee packet to the divert protocol if required.
 960          */
 961         if (divert_info != 0) {
 962                 struct mbuf *clone = NULL;
 963
 964                 /* Clone packet if we're doing a 'tee' */
 965                 if ((divert_info & IP_FW_PORT_TEE_FLAG) != 0)
 966                         clone = m_dup(m, MB_DONTWAIT);
 967
 968                 /* Restore packet header fields to original values */
 969                 ip->ip_len += hlen;
 970                 ip->ip_len = htons(ip->ip_len);
 971                 ip->ip_off = htons(ip->ip_off);
 972
 973                 /* Deliver packet to divert input routine */
 974                 divert_packet(m, 1, divert_info & 0xffff);
 975                 ipstat.ips_delivered++;
 976
 977                 /* If 'tee', continue with original packet */
 978                 if (clone == NULL)
 979                         return;
 980                 m = clone;
 981                 ip = mtod(m, struct ip *);
 982                 ip->ip_len += hlen;
 983                 /*
 984                  * Jump backwards to complete processing of the
 985                  * packet. But first clear divert_info to avoid
 986                  * entering this block again.
 987                  * We do not need to clear args.divert_rule as
 988                  * it will not be used.
 989                  *
 990                  * XXX Better safe than sorry, remove the DIVERT tag.
 991                  */
 992                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
 993                 if (mtag != NULL)
 994                         m_tag_delete(m, mtag);
 995
 996                 divert_info = 0;
 997                 goto pass;
 998         }
 999 #endif
1000
1001 #ifdef IPSEC
1002         /*
1003          * enforce IPsec policy checking if we are seeing last header.
1004          * note that we do not visit this with protocols with pcb layer
1005          * code - like udp/tcp/raw ip.
1006          */
1007         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
1008             ipsec4_in_reject(m, NULL)) {
1009                 ipsecstat.in_polvio++;
1010                 goto bad;
1011         }
1012 #endif
1013 #if FAST_IPSEC
1014         /*
1015          * enforce IPsec policy checking if we are seeing last header.
1016          * note that we do not visit this with protocols with pcb layer
1017          * code - like udp/tcp/raw ip.
1018          */
1019         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1020                 /*
1021                  * Check if the packet has already had IPsec processing
1022                  * done.  If so, then just pass it along.  This tag gets
1023                  * set during AH, ESP, etc. input handling, before the
1024                  * packet is returned to the ip input queue for delivery.
1025                  */
1026                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1027                 crit_enter();
1028                 if (mtag != NULL) {
1029                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
1030                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1031                 } else {
1032                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1033                                                    IP_FORWARDING, &error);
1034                 }
1035                 if (sp != NULL) {
1036                         /*
1037                          * Check security policy against packet attributes.
1038                          */
1039                         error = ipsec_in_reject(sp, m);
1040                         KEY_FREESP(&sp);
1041                 } else {
1042                         /* XXX error stat??? */
1043                         error = EINVAL;
1044 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1045                         goto bad;
1046                 }
1047                 crit_exit();
1048                 if (error)
1049                         goto bad;
1050         }
1051 #endif /* FAST_IPSEC */
1052
1053         ipstat.ips_delivered++;
1054         if (needredispatch) {
1055                 struct netmsg_packet *pmsg;
1056                 lwkt_port_t port;
1057
1058                 ip->ip_off = htons(ip->ip_off);
1059                 ip->ip_len = htons(ip->ip_len);
1060                 port = ip_mport_in(&m);
1061                 if (port == NULL)
1062                         return;
1063
1064                 pmsg = &m->m_hdr.mh_netmsg;
1065                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
1066                             transport_processing_handler);
1067                 pmsg->nm_packet = m;
1068                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
1069
1070                 ip = mtod(m, struct ip *);
1071                 ip->ip_len = ntohs(ip->ip_len);
1072                 ip->ip_off = ntohs(ip->ip_off);
1073                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
1074         } else {
1075                 transport_processing_oncpu(m, hlen, ip);
1076         }
1077         return;
1078
1079 bad:
1080         m_freem(m);
1081 }
1082
1083 /*
1084  * Take incoming datagram fragment and try to reassemble it into
1085  * whole datagram.  If a chain for reassembly of this datagram already
1086  * exists, then it is given as fp; otherwise have to make a chain.
1087  *
1088  * When IPDIVERT enabled, keep additional state with each packet that
1089  * tells us if we need to divert or tee the packet we're building.
1090  * In particular, *divinfo includes the port and TEE flag.
1091  */
1092
1093 static struct mbuf *
1094 ip_reass(struct mbuf *m, struct ipq *fp, struct ipq *where,
1095          u_int32_t *divinfo)
1096 {
1097         struct ip *ip = mtod(m, struct ip *);
1098         struct mbuf *p = NULL, *q, *nq;
1099         struct mbuf *n;
1100         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1101         int i, next;
1102 #ifdef IPDIVERT
1103         struct m_tag *mtag;
1104 #endif
1105
1106         /*
1107          * If the hardware has not done csum over this fragment
1108          * then csum_data is not valid at all.
1109          */
1110         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1111             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1112                 m->m_pkthdr.csum_data = 0;
1113                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1114         }
1115
1116         /*
1117          * Presence of header sizes in mbufs
1118          * would confuse code below.
1119          */
1120         m->m_data += hlen;
1121         m->m_len -= hlen;
1122
1123         /*
1124          * If first fragment to arrive, create a reassembly queue.
1125          */
1126         if (fp == NULL) {
1127                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1128                         goto dropfrag;
1129                 insque(fp, where);
1130                 nipq++;
1131                 fp->ipq_nfrags = 1;
1132                 fp->ipq_ttl = IPFRAGTTL;
1133                 fp->ipq_p = ip->ip_p;
1134                 fp->ipq_id = ip->ip_id;
1135                 fp->ipq_src = ip->ip_src;
1136                 fp->ipq_dst = ip->ip_dst;
1137                 fp->ipq_frags = m;
1138                 m->m_nextpkt = NULL;
1139 #ifdef IPDIVERT
1140                 fp->ipq_div_info = 0;
1141 #endif
1142                 goto inserted;
1143         } else {
1144                 fp->ipq_nfrags++;
1145         }
1146
1147 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1148
1149         /*
1150          * Find a segment which begins after this one does.
1151          */
1152         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1153                 if (GETIP(q)->ip_off > ip->ip_off)
1154                         break;
1155
1156         /*
1157          * If there is a preceding segment, it may provide some of
1158          * our data already.  If so, drop the data from the incoming
1159          * segment.  If it provides all of our data, drop us, otherwise
1160          * stick new segment in the proper place.
1161          *
1162          * If some of the data is dropped from the the preceding
1163          * segment, then it's checksum is invalidated.
1164          */
1165         if (p) {
1166                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1167                 if (i > 0) {
1168                         if (i >= ip->ip_len)
1169                                 goto dropfrag;
1170                         m_adj(m, i);
1171                         m->m_pkthdr.csum_flags = 0;
1172                         ip->ip_off += i;
1173                         ip->ip_len -= i;
1174                 }
1175                 m->m_nextpkt = p->m_nextpkt;
1176                 p->m_nextpkt = m;
1177         } else {
1178                 m->m_nextpkt = fp->ipq_frags;
1179                 fp->ipq_frags = m;
1180         }
1181
1182         /*
1183          * While we overlap succeeding segments trim them or,
1184          * if they are completely covered, dequeue them.
1185          */
1186         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1187              q = nq) {
1188                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1189                 if (i < GETIP(q)->ip_len) {
1190                         GETIP(q)->ip_len -= i;
1191                         GETIP(q)->ip_off += i;
1192                         m_adj(q, i);
1193                         q->m_pkthdr.csum_flags = 0;
1194                         break;
1195                 }
1196                 nq = q->m_nextpkt;
1197                 m->m_nextpkt = nq;
1198                 ipstat.ips_fragdropped++;
1199                 fp->ipq_nfrags--;
1200                 q->m_nextpkt = NULL;
1201                 m_freem(q);
1202         }
1203
1204 inserted:
1205
1206 #ifdef IPDIVERT
1207         /*
1208          * Transfer firewall instructions to the fragment structure.
1209          * Only trust info in the fragment at offset 0.
1210          */
1211         if (ip->ip_off == 0) {
1212                 fp->ipq_div_info = *divinfo;
1213         } else {
1214                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1215                 if (mtag != NULL)
1216                         m_tag_delete(m, mtag);
1217         }
1218         *divinfo = 0;
1219 #endif
1220
1221         /*
1222          * Check for complete reassembly and perform frag per packet
1223          * limiting.
1224          *
1225          * Frag limiting is performed here so that the nth frag has
1226          * a chance to complete the packet before we drop the packet.
1227          * As a result, n+1 frags are actually allowed per packet, but
1228          * only n will ever be stored. (n = maxfragsperpacket.)
1229          *
1230          */
1231         next = 0;
1232         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1233                 if (GETIP(q)->ip_off != next) {
1234                         if (fp->ipq_nfrags > maxfragsperpacket) {
1235                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1236                                 ip_freef(fp);
1237                         }
1238                         return (NULL);
1239                 }
1240                 next += GETIP(q)->ip_len;
1241         }
1242         /* Make sure the last packet didn't have the IP_MF flag */
1243         if (p->m_flags & M_FRAG) {
1244                 if (fp->ipq_nfrags > maxfragsperpacket) {
1245                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1246                         ip_freef(fp);
1247                 }
1248                 return (NULL);
1249         }
1250
1251         /*
1252          * Reassembly is complete.  Make sure the packet is a sane size.
1253          */
1254         q = fp->ipq_frags;
1255         ip = GETIP(q);
1256         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1257                 ipstat.ips_toolong++;
1258                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1259                 ip_freef(fp);
1260                 return (NULL);
1261         }
1262
1263         /*
1264          * Concatenate fragments.
1265          */
1266         m = q;
1267         n = m->m_next;
1268         m->m_next = NULL;
1269         m_cat(m, n);
1270         nq = q->m_nextpkt;
1271         q->m_nextpkt = NULL;
1272         for (q = nq; q != NULL; q = nq) {
1273                 nq = q->m_nextpkt;
1274                 q->m_nextpkt = NULL;
1275                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1276                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1277                 m_cat(m, q);
1278         }
1279
1280         /*
1281          * Clean up the 1's complement checksum.  Carry over 16 bits must
1282          * be added back.  This assumes no more then 65535 packet fragments
1283          * were reassembled.  A second carry can also occur (but not a third).
1284          */
1285         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1286                                 (m->m_pkthdr.csum_data >> 16);
1287         if (m->m_pkthdr.csum_data > 0xFFFF)
1288                 m->m_pkthdr.csum_data -= 0xFFFF;
1289
1290
1291 #ifdef IPDIVERT
1292         /*
1293          * Extract firewall instructions from the fragment structure.
1294          */
1295         *divinfo = fp->ipq_div_info;
1296 #endif
1297
1298         /*
1299          * Create header for new ip packet by
1300          * modifying header of first packet;
1301          * dequeue and discard fragment reassembly header.
1302          * Make header visible.
1303          */
1304         ip->ip_len = next;
1305         ip->ip_src = fp->ipq_src;
1306         ip->ip_dst = fp->ipq_dst;
1307         remque(fp);
1308         nipq--;
1309         mpipe_free(&ipq_mpipe, fp);
1310         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1311         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1312         /* some debugging cruft by sklower, below, will go away soon */
1313         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1314                 int plen = 0;
1315
1316                 for (n = m; n; n = n->m_next)
1317                         plen += n->m_len;
1318                 m->m_pkthdr.len = plen;
1319         }
1320
1321         ipstat.ips_reassembled++;
1322         return (m);
1323
1324 dropfrag:
1325 #ifdef IPDIVERT
1326         *divinfo = 0;
1327 #endif
1328         ipstat.ips_fragdropped++;
1329         if (fp != NULL)
1330                 fp->ipq_nfrags--;
1331         m_freem(m);
1332         return (NULL);
1333
1334 #undef GETIP
1335 }
1336
1337 /*
1338  * Free a fragment reassembly header and all
1339  * associated datagrams.
1340  */
1341 static void
1342 ip_freef(struct ipq *fp)
1343 {
1344         struct mbuf *q;
1345
1346         while (fp->ipq_frags) {
1347                 q = fp->ipq_frags;
1348                 fp->ipq_frags = q->m_nextpkt;
1349                 q->m_nextpkt = NULL;
1350                 m_freem(q);
1351         }
1352         remque(fp);
1353         mpipe_free(&ipq_mpipe, fp);
1354         nipq--;
1355 }
1356
1357 /*
1358  * IP timer processing;
1359  * if a timer expires on a reassembly
1360  * queue, discard it.
1361  */
1362 void
1363 ip_slowtimo(void)
1364 {
1365         struct ipq *fp;
1366         int i;
1367
1368         crit_enter();
1369         for (i = 0; i < IPREASS_NHASH; i++) {
1370                 fp = ipq[i].next;
1371                 if (fp == NULL)
1372                         continue;
1373                 while (fp != &ipq[i]) {
1374                         --fp->ipq_ttl;
1375                         fp = fp->next;
1376                         if (fp->prev->ipq_ttl == 0) {
1377                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1378                                 ip_freef(fp->prev);
1379                         }
1380                 }
1381         }
1382         /*
1383          * If we are over the maximum number of fragments
1384          * (due to the limit being lowered), drain off
1385          * enough to get down to the new limit.
1386          */
1387         if (maxnipq >= 0 && nipq > maxnipq) {
1388                 for (i = 0; i < IPREASS_NHASH; i++) {
1389                         while (nipq > maxnipq &&
1390                                 (ipq[i].next != &ipq[i])) {
1391                                 ipstat.ips_fragdropped +=
1392                                     ipq[i].next->ipq_nfrags;
1393                                 ip_freef(ipq[i].next);
1394                         }
1395                 }
1396         }
1397         ipflow_slowtimo();
1398         crit_exit();
1399 }
1400
1401 /*
1402  * Drain off all datagram fragments.
1403  */
1404 void
1405 ip_drain(void)
1406 {
1407         int i;
1408
1409         for (i = 0; i < IPREASS_NHASH; i++) {
1410                 while (ipq[i].next != &ipq[i]) {
1411                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1412                         ip_freef(ipq[i].next);
1413                 }
1414         }
1415         in_rtqdrain();
1416 }
1417
1418 /*
1419  * Do option processing on a datagram,
1420  * possibly discarding it if bad options are encountered,
1421  * or forwarding it if source-routed.
1422  * The pass argument is used when operating in the IPSTEALTH
1423  * mode to tell what options to process:
1424  * [LS]SRR (pass 0) or the others (pass 1).
1425  * The reason for as many as two passes is that when doing IPSTEALTH,
1426  * non-routing options should be processed only if the packet is for us.
1427  * Returns 1 if packet has been forwarded/freed,
1428  * 0 if the packet should be processed further.
1429  */
1430 static int
1431 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1432 {
1433         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1434         struct ip *ip = mtod(m, struct ip *);
1435         u_char *cp;
1436         struct in_ifaddr *ia;
1437         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1438         boolean_t forward = FALSE;
1439         struct in_addr *sin, dst;
1440         n_time ntime;
1441
1442         dst = ip->ip_dst;
1443         cp = (u_char *)(ip + 1);
1444         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1445         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1446                 opt = cp[IPOPT_OPTVAL];
1447                 if (opt == IPOPT_EOL)
1448                         break;
1449                 if (opt == IPOPT_NOP)
1450                         optlen = 1;
1451                 else {
1452                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1453                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1454                                 goto bad;
1455                         }
1456                         optlen = cp[IPOPT_OLEN];
1457                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1458                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1459                                 goto bad;
1460                         }
1461                 }
1462                 switch (opt) {
1463
1464                 default:
1465                         break;
1466
1467                 /*
1468                  * Source routing with record.
1469                  * Find interface with current destination address.
1470                  * If none on this machine then drop if strictly routed,
1471                  * or do nothing if loosely routed.
1472                  * Record interface address and bring up next address
1473                  * component.  If strictly routed make sure next
1474                  * address is on directly accessible net.
1475                  */
1476                 case IPOPT_LSRR:
1477                 case IPOPT_SSRR:
1478                         if (ipstealth && pass > 0)
1479                                 break;
1480                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1481                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1482                                 goto bad;
1483                         }
1484                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1485                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1486                                 goto bad;
1487                         }
1488                         ipaddr.sin_addr = ip->ip_dst;
1489                         ia = (struct in_ifaddr *)
1490                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1491                         if (ia == NULL) {
1492                                 if (opt == IPOPT_SSRR) {
1493                                         type = ICMP_UNREACH;
1494                                         code = ICMP_UNREACH_SRCFAIL;
1495                                         goto bad;
1496                                 }
1497                                 if (!ip_dosourceroute)
1498                                         goto nosourcerouting;
1499                                 /*
1500                                  * Loose routing, and not at next destination
1501                                  * yet; nothing to do except forward.
1502                                  */
1503                                 break;
1504                         }
1505                         off--;                  /* 0 origin */
1506                         if (off > optlen - (int)sizeof(struct in_addr)) {
1507                                 /*
1508                                  * End of source route.  Should be for us.
1509                                  */
1510                                 if (!ip_acceptsourceroute)
1511                                         goto nosourcerouting;
1512                                 save_rte(cp, ip->ip_src);
1513                                 break;
1514                         }
1515                         if (ipstealth)
1516                                 goto dropit;
1517                         if (!ip_dosourceroute) {
1518                                 if (ipforwarding) {
1519                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1520
1521                                         /*
1522                                          * Acting as a router, so generate ICMP
1523                                          */
1524 nosourcerouting:
1525                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1526                                         log(LOG_WARNING,
1527                                             "attempted source route from %s to %s\n",
1528                                             inet_ntoa(ip->ip_src), buf);
1529                                         type = ICMP_UNREACH;
1530                                         code = ICMP_UNREACH_SRCFAIL;
1531                                         goto bad;
1532                                 } else {
1533                                         /*
1534                                          * Not acting as a router,
1535                                          * so silently drop.
1536                                          */
1537 dropit:
1538                                         ipstat.ips_cantforward++;
1539                                         m_freem(m);
1540                                         return (1);
1541                                 }
1542                         }
1543
1544                         /*
1545                          * locate outgoing interface
1546                          */
1547                         memcpy(&ipaddr.sin_addr, cp + off,
1548                             sizeof ipaddr.sin_addr);
1549
1550                         if (opt == IPOPT_SSRR) {
1551 #define INA     struct in_ifaddr *
1552 #define SA      struct sockaddr *
1553                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1554                                                                         == NULL)
1555                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1556                         } else
1557                                 ia = ip_rtaddr(ipaddr.sin_addr,
1558                                                &ipforward_rt[mycpuid]);
1559                         if (ia == NULL) {
1560                                 type = ICMP_UNREACH;
1561                                 code = ICMP_UNREACH_SRCFAIL;
1562                                 goto bad;
1563                         }
1564                         ip->ip_dst = ipaddr.sin_addr;
1565                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1566                             sizeof(struct in_addr));
1567                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1568                         /*
1569                          * Let ip_intr's mcast routing check handle mcast pkts
1570                          */
1571                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1572                         break;
1573
1574                 case IPOPT_RR:
1575                         if (ipstealth && pass == 0)
1576                                 break;
1577                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1578                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1579                                 goto bad;
1580                         }
1581                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1582                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1583                                 goto bad;
1584                         }
1585                         /*
1586                          * If no space remains, ignore.
1587                          */
1588                         off--;                  /* 0 origin */
1589                         if (off > optlen - (int)sizeof(struct in_addr))
1590                                 break;
1591                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1592                             sizeof ipaddr.sin_addr);
1593                         /*
1594                          * locate outgoing interface; if we're the destination,
1595                          * use the incoming interface (should be same).
1596                          */
1597                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1598                             (ia = ip_rtaddr(ipaddr.sin_addr,
1599                                             &ipforward_rt[mycpuid]))
1600                                                                      == NULL) {
1601                                 type = ICMP_UNREACH;
1602                                 code = ICMP_UNREACH_HOST;
1603                                 goto bad;
1604                         }
1605                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1606                             sizeof(struct in_addr));
1607                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1608                         break;
1609
1610                 case IPOPT_TS:
1611                         if (ipstealth && pass == 0)
1612                                 break;
1613                         code = cp - (u_char *)ip;
1614                         if (optlen < 4 || optlen > 40) {
1615                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1616                                 goto bad;
1617                         }
1618                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1619                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1620                                 goto bad;
1621                         }
1622                         if (off > optlen - (int)sizeof(int32_t)) {
1623                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1624                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1625                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1626                                         goto bad;
1627                                 }
1628                                 break;
1629                         }
1630                         off--;                          /* 0 origin */
1631                         sin = (struct in_addr *)(cp + off);
1632                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1633
1634                         case IPOPT_TS_TSONLY:
1635                                 break;
1636
1637                         case IPOPT_TS_TSANDADDR:
1638                                 if (off + sizeof(n_time) +
1639                                     sizeof(struct in_addr) > optlen) {
1640                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1641                                         goto bad;
1642                                 }
1643                                 ipaddr.sin_addr = dst;
1644                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1645                                                             m->m_pkthdr.rcvif);
1646                                 if (ia == NULL)
1647                                         continue;
1648                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1649                                     sizeof(struct in_addr));
1650                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1651                                 off += sizeof(struct in_addr);
1652                                 break;
1653
1654                         case IPOPT_TS_PRESPEC:
1655                                 if (off + sizeof(n_time) +
1656                                     sizeof(struct in_addr) > optlen) {
1657                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1658                                         goto bad;
1659                                 }
1660                                 memcpy(&ipaddr.sin_addr, sin,
1661                                     sizeof(struct in_addr));
1662                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1663                                         continue;
1664                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1665                                 off += sizeof(struct in_addr);
1666                                 break;
1667
1668                         default:
1669                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1670                                 goto bad;
1671                         }
1672                         ntime = iptime();
1673                         memcpy(cp + off, &ntime, sizeof(n_time));
1674                         cp[IPOPT_OFFSET] += sizeof(n_time);
1675                 }
1676         }
1677         if (forward && ipforwarding) {
1678                 ip_forward(m, TRUE, next_hop);
1679                 return (1);
1680         }
1681         return (0);
1682 bad:
1683         icmp_error(m, type, code, 0, 0);
1684         ipstat.ips_badoptions++;
1685         return (1);
1686 }
1687
1688 /*
1689  * Given address of next destination (final or next hop),
1690  * return internet address info of interface to be used to get there.
1691  */
1692 struct in_ifaddr *
1693 ip_rtaddr(struct in_addr dst, struct route *ro)
1694 {
1695         struct sockaddr_in *sin;
1696
1697         sin = (struct sockaddr_in *)&ro->ro_dst;
1698
1699         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1700                 if (ro->ro_rt != NULL) {
1701                         RTFREE(ro->ro_rt);
1702                         ro->ro_rt = NULL;
1703                 }
1704                 sin->sin_family = AF_INET;
1705                 sin->sin_len = sizeof *sin;
1706                 sin->sin_addr = dst;
1707                 rtalloc_ign(ro, RTF_PRCLONING);
1708         }
1709
1710         if (ro->ro_rt == NULL)
1711                 return (NULL);
1712
1713         return (ifatoia(ro->ro_rt->rt_ifa));
1714 }
1715
1716 /*
1717  * Save incoming source route for use in replies,
1718  * to be picked up later by ip_srcroute if the receiver is interested.
1719  */
1720 void
1721 save_rte(u_char *option, struct in_addr dst)
1722 {
1723         unsigned olen;
1724
1725         olen = option[IPOPT_OLEN];
1726 #ifdef DIAGNOSTIC
1727         if (ipprintfs)
1728                 kprintf("save_rte: olen %d\n", olen);
1729 #endif
1730         if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst)))
1731                 return;
1732         bcopy(option, ip_srcrt.srcopt, olen);
1733         ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1734         ip_srcrt.dst = dst;
1735 }
1736
1737 /*
1738  * Retrieve incoming source route for use in replies,
1739  * in the same form used by setsockopt.
1740  * The first hop is placed before the options, will be removed later.
1741  */
1742 struct mbuf *
1743 ip_srcroute(void)
1744 {
1745         struct in_addr *p, *q;
1746         struct mbuf *m;
1747
1748         if (ip_nhops == 0)
1749                 return (NULL);
1750         m = m_get(MB_DONTWAIT, MT_HEADER);
1751         if (m == NULL)
1752                 return (NULL);
1753
1754 #define OPTSIZ  (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt))
1755
1756         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1757         m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) +
1758             OPTSIZ;
1759 #ifdef DIAGNOSTIC
1760         if (ipprintfs)
1761                 kprintf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
1762 #endif
1763
1764         /*
1765          * First save first hop for return route
1766          */
1767         p = &ip_srcrt.route[ip_nhops - 1];
1768         *(mtod(m, struct in_addr *)) = *p--;
1769 #ifdef DIAGNOSTIC
1770         if (ipprintfs)
1771                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1772 #endif
1773
1774         /*
1775          * Copy option fields and padding (nop) to mbuf.
1776          */
1777         ip_srcrt.nop = IPOPT_NOP;
1778         ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1779         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &ip_srcrt.nop,
1780             OPTSIZ);
1781         q = (struct in_addr *)(mtod(m, caddr_t) +
1782             sizeof(struct in_addr) + OPTSIZ);
1783 #undef OPTSIZ
1784         /*
1785          * Record return path as an IP source route,
1786          * reversing the path (pointers are now aligned).
1787          */
1788         while (p >= ip_srcrt.route) {
1789 #ifdef DIAGNOSTIC
1790                 if (ipprintfs)
1791                         kprintf(" %x", ntohl(q->s_addr));
1792 #endif
1793                 *q++ = *p--;
1794         }
1795         /*
1796          * Last hop goes to final destination.
1797          */
1798         *q = ip_srcrt.dst;
1799 #ifdef DIAGNOSTIC
1800         if (ipprintfs)
1801                 kprintf(" %x\n", ntohl(q->s_addr));
1802 #endif
1803         return (m);
1804 }
1805
1806 /*
1807  * Strip out IP options.
1808  */
1809 void
1810 ip_stripoptions(struct mbuf *m)
1811 {
1812         int datalen;
1813         struct ip *ip = mtod(m, struct ip *);
1814         caddr_t opts;
1815         int optlen;
1816
1817         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1818         opts = (caddr_t)(ip + 1);
1819         datalen = m->m_len - (sizeof(struct ip) + optlen);
1820         bcopy(opts + optlen, opts, datalen);
1821         m->m_len -= optlen;
1822         if (m->m_flags & M_PKTHDR)
1823                 m->m_pkthdr.len -= optlen;
1824         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1825 }
1826
1827 u_char inetctlerrmap[PRC_NCMDS] = {
1828         0,              0,              0,              0,
1829         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1830         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1831         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1832         0,              0,              0,              0,
1833         ENOPROTOOPT,    ECONNREFUSED
1834 };
1835
1836 /*
1837  * Forward a packet.  If some error occurs return the sender
1838  * an icmp packet.  Note we can't always generate a meaningful
1839  * icmp message because icmp doesn't have a large enough repertoire
1840  * of codes and types.
1841  *
1842  * If not forwarding, just drop the packet.  This could be confusing
1843  * if ipforwarding was zero but some routing protocol was advancing
1844  * us as a gateway to somewhere.  However, we must let the routing
1845  * protocol deal with that.
1846  *
1847  * The using_srcrt parameter indicates whether the packet is being forwarded
1848  * via a source route.
1849  */
1850 void
1851 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1852 {
1853         struct ip *ip = mtod(m, struct ip *);
1854         struct sockaddr_in *ipforward_rtaddr;
1855         struct rtentry *rt;
1856         int error, type = 0, code = 0, destmtu = 0;
1857         struct mbuf *mcopy;
1858         n_long dest;
1859         struct in_addr pkt_dst;
1860         struct route *cache_rt = &ipforward_rt[mycpuid];
1861
1862         dest = INADDR_ANY;
1863         /*
1864          * Cache the destination address of the packet; this may be
1865          * changed by use of 'ipfw fwd'.
1866          */
1867         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1868
1869 #ifdef DIAGNOSTIC
1870         if (ipprintfs)
1871                 kprintf("forward: src %x dst %x ttl %x\n",
1872                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1873 #endif
1874
1875         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1876                 ipstat.ips_cantforward++;
1877                 m_freem(m);
1878                 return;
1879         }
1880         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1881                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1882                 return;
1883         }
1884
1885         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1886         if (cache_rt->ro_rt == NULL ||
1887             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1888                 if (cache_rt->ro_rt != NULL) {
1889                         RTFREE(cache_rt->ro_rt);
1890                         cache_rt->ro_rt = NULL;
1891                 }
1892                 ipforward_rtaddr->sin_family = AF_INET;
1893                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1894                 ipforward_rtaddr->sin_addr = pkt_dst;
1895                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1896                 if (cache_rt->ro_rt == NULL) {
1897                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1898                         return;
1899                 }
1900         }
1901         rt = cache_rt->ro_rt;
1902
1903         /*
1904          * Save the IP header and at most 8 bytes of the payload,
1905          * in case we need to generate an ICMP message to the src.
1906          *
1907          * XXX this can be optimized a lot by saving the data in a local
1908          * buffer on the stack (72 bytes at most), and only allocating the
1909          * mbuf if really necessary. The vast majority of the packets
1910          * are forwarded without having to send an ICMP back (either
1911          * because unnecessary, or because rate limited), so we are
1912          * really we are wasting a lot of work here.
1913          *
1914          * We don't use m_copy() because it might return a reference
1915          * to a shared cluster. Both this function and ip_output()
1916          * assume exclusive access to the IP header in `m', so any
1917          * data in a cluster may change before we reach icmp_error().
1918          */
1919         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1920         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1921                 /*
1922                  * It's probably ok if the pkthdr dup fails (because
1923                  * the deep copy of the tag chain failed), but for now
1924                  * be conservative and just discard the copy since
1925                  * code below may some day want the tags.
1926                  */
1927                 m_free(mcopy);
1928                 mcopy = NULL;
1929         }
1930         if (mcopy != NULL) {
1931                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1932                     (int)ip->ip_len);
1933                 mcopy->m_pkthdr.len = mcopy->m_len;
1934                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1935         }
1936
1937         if (!ipstealth)
1938                 ip->ip_ttl -= IPTTLDEC;
1939
1940         /*
1941          * If forwarding packet using same interface that it came in on,
1942          * perhaps should send a redirect to sender to shortcut a hop.
1943          * Only send redirect if source is sending directly to us,
1944          * and if packet was not source routed (or has any options).
1945          * Also, don't send redirect if forwarding using a default route
1946          * or a route modified by a redirect.
1947          */
1948         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1949             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1950             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1951             ipsendredirects && !using_srcrt && next_hop == NULL) {
1952                 u_long src = ntohl(ip->ip_src.s_addr);
1953                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1954
1955                 if (rt_ifa != NULL &&
1956                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1957                         if (rt->rt_flags & RTF_GATEWAY)
1958                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1959                         else
1960                                 dest = pkt_dst.s_addr;
1961                         /*
1962                          * Router requirements says to only send
1963                          * host redirects.
1964                          */
1965                         type = ICMP_REDIRECT;
1966                         code = ICMP_REDIRECT_HOST;
1967 #ifdef DIAGNOSTIC
1968                         if (ipprintfs)
1969                                 kprintf("redirect (%d) to %x\n", code, dest);
1970 #endif
1971                 }
1972         }
1973
1974         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1975         if (error == 0) {
1976                 ipstat.ips_forward++;
1977                 if (type == 0) {
1978                         if (mcopy) {
1979                                 ipflow_create(cache_rt, mcopy);
1980                                 m_freem(mcopy);
1981                         }
1982                         return;         /* most common case */
1983                 } else {
1984                         ipstat.ips_redirectsent++;
1985                 }
1986         } else {
1987                 ipstat.ips_cantforward++;
1988         }
1989
1990         if (mcopy == NULL)
1991                 return;
1992
1993         /*
1994          * Send ICMP message.
1995          */
1996
1997         switch (error) {
1998
1999         case 0:                         /* forwarded, but need redirect */
2000                 /* type, code set above */
2001                 break;
2002
2003         case ENETUNREACH:               /* shouldn't happen, checked above */
2004         case EHOSTUNREACH:
2005         case ENETDOWN:
2006         case EHOSTDOWN:
2007         default:
2008                 type = ICMP_UNREACH;
2009                 code = ICMP_UNREACH_HOST;
2010                 break;
2011
2012         case EMSGSIZE:
2013                 type = ICMP_UNREACH;
2014                 code = ICMP_UNREACH_NEEDFRAG;
2015 #ifdef IPSEC
2016                 /*
2017                  * If the packet is routed over IPsec tunnel, tell the
2018                  * originator the tunnel MTU.
2019                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2020                  * XXX quickhack!!!
2021                  */
2022                 if (cache_rt->ro_rt != NULL) {
2023                         struct secpolicy *sp = NULL;
2024                         int ipsecerror;
2025                         int ipsechdr;
2026                         struct route *ro;
2027
2028                         sp = ipsec4_getpolicybyaddr(mcopy,
2029                                                     IPSEC_DIR_OUTBOUND,
2030                                                     IP_FORWARDING,
2031                                                     &ipsecerror);
2032
2033                         if (sp == NULL)
2034                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2035                         else {
2036                                 /* count IPsec header size */
2037                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2038                                                          IPSEC_DIR_OUTBOUND,
2039                                                          NULL);
2040
2041                                 /*
2042                                  * find the correct route for outer IPv4
2043                                  * header, compute tunnel MTU.
2044                                  *
2045                                  */
2046                                 if (sp->req != NULL && sp->req->sav != NULL &&
2047                                     sp->req->sav->sah != NULL) {
2048                                         ro = &sp->req->sav->sah->sa_route;
2049                                         if (ro->ro_rt != NULL &&
2050                                             ro->ro_rt->rt_ifp != NULL) {
2051                                                 destmtu =
2052                                                     ro->ro_rt->rt_ifp->if_mtu;
2053                                                 destmtu -= ipsechdr;
2054                                         }
2055                                 }
2056
2057                                 key_freesp(sp);
2058                         }
2059                 }
2060 #elif FAST_IPSEC
2061                 /*
2062                  * If the packet is routed over IPsec tunnel, tell the
2063                  * originator the tunnel MTU.
2064                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2065                  * XXX quickhack!!!
2066                  */
2067                 if (cache_rt->ro_rt != NULL) {
2068                         struct secpolicy *sp = NULL;
2069                         int ipsecerror;
2070                         int ipsechdr;
2071                         struct route *ro;
2072
2073                         sp = ipsec_getpolicybyaddr(mcopy,
2074                                                    IPSEC_DIR_OUTBOUND,
2075                                                    IP_FORWARDING,
2076                                                    &ipsecerror);
2077
2078                         if (sp == NULL)
2079                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2080                         else {
2081                                 /* count IPsec header size */
2082                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2083                                                          IPSEC_DIR_OUTBOUND,
2084                                                          NULL);
2085
2086                                 /*
2087                                  * find the correct route for outer IPv4
2088                                  * header, compute tunnel MTU.
2089                                  */
2090
2091                                 if (sp->req != NULL &&
2092                                     sp->req->sav != NULL &&
2093                                     sp->req->sav->sah != NULL) {
2094                                         ro = &sp->req->sav->sah->sa_route;
2095                                         if (ro->ro_rt != NULL &&
2096                                             ro->ro_rt->rt_ifp != NULL) {
2097                                                 destmtu =
2098                                                     ro->ro_rt->rt_ifp->if_mtu;
2099                                                 destmtu -= ipsechdr;
2100                                         }
2101                                 }
2102
2103                                 KEY_FREESP(&sp);
2104                         }
2105                 }
2106 #else /* !IPSEC && !FAST_IPSEC */
2107                 if (cache_rt->ro_rt != NULL)
2108                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2109 #endif /*IPSEC*/
2110                 ipstat.ips_cantfrag++;
2111                 break;
2112
2113         case ENOBUFS:
2114                 /*
2115                  * A router should not generate ICMP_SOURCEQUENCH as
2116                  * required in RFC1812 Requirements for IP Version 4 Routers.
2117                  * Source quench could be a big problem under DoS attacks,
2118                  * or if the underlying interface is rate-limited.
2119                  * Those who need source quench packets may re-enable them
2120                  * via the net.inet.ip.sendsourcequench sysctl.
2121                  */
2122                 if (!ip_sendsourcequench) {
2123                         m_freem(mcopy);
2124                         return;
2125                 } else {
2126                         type = ICMP_SOURCEQUENCH;
2127                         code = 0;
2128                 }
2129                 break;
2130
2131         case EACCES:                    /* ipfw denied packet */
2132                 m_freem(mcopy);
2133                 return;
2134         }
2135         icmp_error(mcopy, type, code, dest, destmtu);
2136 }
2137
2138 void
2139 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2140                struct mbuf *m)
2141 {
2142         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2143                 struct timeval tv;
2144
2145                 microtime(&tv);
2146                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2147                     SCM_TIMESTAMP, SOL_SOCKET);
2148                 if (*mp)
2149                         mp = &(*mp)->m_next;
2150         }
2151         if (inp->inp_flags & INP_RECVDSTADDR) {
2152                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2153                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2154                 if (*mp)
2155                         mp = &(*mp)->m_next;
2156         }
2157         if (inp->inp_flags & INP_RECVTTL) {
2158                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2159                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2160                 if (*mp)
2161                         mp = &(*mp)->m_next;
2162         }
2163 #ifdef notyet
2164         /* XXX
2165          * Moving these out of udp_input() made them even more broken
2166          * than they already were.
2167          */
2168         /* options were tossed already */
2169         if (inp->inp_flags & INP_RECVOPTS) {
2170                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2171                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2172                 if (*mp)
2173                         mp = &(*mp)->m_next;
2174         }
2175         /* ip_srcroute doesn't do what we want here, need to fix */
2176         if (inp->inp_flags & INP_RECVRETOPTS) {
2177                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(),
2178                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2179                 if (*mp)
2180                         mp = &(*mp)->m_next;
2181         }
2182 #endif
2183         if (inp->inp_flags & INP_RECVIF) {
2184                 struct ifnet *ifp;
2185                 struct sdlbuf {
2186                         struct sockaddr_dl sdl;
2187                         u_char  pad[32];
2188                 } sdlbuf;
2189                 struct sockaddr_dl *sdp;
2190                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2191
2192                 if (((ifp = m->m_pkthdr.rcvif)) &&
2193                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2194                         sdp = IF_LLSOCKADDR(ifp);
2195                         /*
2196                          * Change our mind and don't try copy.
2197                          */
2198                         if ((sdp->sdl_family != AF_LINK) ||
2199                             (sdp->sdl_len > sizeof(sdlbuf))) {
2200                                 goto makedummy;
2201                         }
2202                         bcopy(sdp, sdl2, sdp->sdl_len);
2203                 } else {
2204 makedummy:
2205                         sdl2->sdl_len =
2206                             offsetof(struct sockaddr_dl, sdl_data[0]);
2207                         sdl2->sdl_family = AF_LINK;
2208                         sdl2->sdl_index = 0;
2209                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2210                 }
2211                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2212                         IP_RECVIF, IPPROTO_IP);
2213                 if (*mp)
2214                         mp = &(*mp)->m_next;
2215         }
2216 }
2217
2218 /*
2219  * XXX these routines are called from the upper part of the kernel.
2220  *
2221  * They could also be moved to ip_mroute.c, since all the RSVP
2222  *  handling is done there already.
2223  */
2224 int
2225 ip_rsvp_init(struct socket *so)
2226 {
2227         if (so->so_type != SOCK_RAW ||
2228             so->so_proto->pr_protocol != IPPROTO_RSVP)
2229                 return EOPNOTSUPP;
2230
2231         if (ip_rsvpd != NULL)
2232                 return EADDRINUSE;
2233
2234         ip_rsvpd = so;
2235         /*
2236          * This may seem silly, but we need to be sure we don't over-increment
2237          * the RSVP counter, in case something slips up.
2238          */
2239         if (!ip_rsvp_on) {
2240                 ip_rsvp_on = 1;
2241                 rsvp_on++;
2242         }
2243
2244         return 0;
2245 }
2246
2247 int
2248 ip_rsvp_done(void)
2249 {
2250         ip_rsvpd = NULL;
2251         /*
2252          * This may seem silly, but we need to be sure we don't over-decrement
2253          * the RSVP counter, in case something slips up.
2254          */
2255         if (ip_rsvp_on) {
2256                 ip_rsvp_on = 0;
2257                 rsvp_on--;
2258         }
2259         return 0;
2260 }
2261
2262 void
2263 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2264 {
2265         int off, proto;
2266         __va_list ap;
2267
2268         __va_start(ap, m);
2269         off = __va_arg(ap, int);
2270         proto = __va_arg(ap, int);
2271         __va_end(ap);
2272
2273         if (rsvp_input_p) { /* call the real one if loaded */
2274                 rsvp_input_p(m, off, proto);
2275                 return;
2276         }
2277
2278         /* Can still get packets with rsvp_on = 0 if there is a local member
2279          * of the group to which the RSVP packet is addressed.  But in this
2280          * case we want to throw the packet away.
2281          */
2282
2283         if (!rsvp_on) {
2284                 m_freem(m);
2285                 return;
2286         }
2287
2288         if (ip_rsvpd != NULL) {
2289                 rip_input(m, off, proto);
2290                 return;
2291         }
2292         /* Drop the packet */
2293         m_freem(m);
2294 }