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 1.115 2008/10/28 07:09:26 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 #include <sys/lock.h>
  97
  98 #include <machine/stdarg.h>
  99
 100 #include <net/if.h>
 101 #include <net/if_types.h>
 102 #include <net/if_var.h>
 103 #include <net/if_dl.h>
 104 #include <net/pfil.h>
 105 #include <net/route.h>
 106 #include <net/netisr.h>
 107
 108 #include <netinet/in.h>
 109 #include <netinet/in_systm.h>
 110 #include <netinet/in_var.h>
 111 #include <netinet/ip.h>
 112 #include <netinet/in_pcb.h>
 113 #include <netinet/ip_var.h>
 114 #include <netinet/ip_icmp.h>
 115 #include <netinet/ip_divert.h>
 116 #include <netinet/ip_flow.h>
 117
 118 #include <sys/thread2.h>
 119 #include <sys/msgport2.h>
 120 #include <net/netmsg2.h>
 121
 122 #include <sys/socketvar.h>
 123
 124 #include <net/ipfw/ip_fw.h>
 125 #include <net/dummynet/ip_dummynet.h>
 126
 127 #ifdef IPSEC
 128 #include <netinet6/ipsec.h>
 129 #include <netproto/key/key.h>
 130 #endif
 131
 132 #ifdef FAST_IPSEC
 133 #include <netproto/ipsec/ipsec.h>
 134 #include <netproto/ipsec/key.h>
 135 #endif
 136
 137 int rsvp_on = 0;
 138 static int ip_rsvp_on;
 139 struct socket *ip_rsvpd;
 140
 141 int ip_mpsafe = 0;
 142 TUNABLE_INT("net.inet.ip.mpsafe", &ip_mpsafe);
 143
 144 int ipforwarding = 0;
 145 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 146     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 147
 148 static int ipsendredirects = 1; /* XXX */
 149 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 150     &ipsendredirects, 0, "Enable sending IP redirects");
 151
 152 int ip_defttl = IPDEFTTL;
 153 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 154     &ip_defttl, 0, "Maximum TTL on IP packets");
 155
 156 static int ip_dosourceroute = 0;
 157 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 158     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 159
 160 static int ip_acceptsourceroute = 0;
 161 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 162     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 163     "Enable accepting source routed IP packets");
 164
 165 static int ip_keepfaith = 0;
 166 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 167     &ip_keepfaith, 0,
 168     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 169
 170 static int nipq = 0;    /* total # of reass queues */
 171 static int maxnipq;
 172 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 173     &maxnipq, 0,
 174     "Maximum number of IPv4 fragment reassembly queue entries");
 175
 176 static int maxfragsperpacket;
 177 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 178     &maxfragsperpacket, 0,
 179     "Maximum number of IPv4 fragments allowed per packet");
 180
 181 static int ip_sendsourcequench = 0;
 182 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 183     &ip_sendsourcequench, 0,
 184     "Enable the transmission of source quench packets");
 185
 186 int ip_do_randomid = 1;
 187 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 188     &ip_do_randomid, 0,
 189     "Assign random ip_id values");
 190 /*
 191  * XXX - Setting ip_checkinterface mostly implements the receive side of
 192  * the Strong ES model described in RFC 1122, but since the routing table
 193  * and transmit implementation do not implement the Strong ES model,
 194  * setting this to 1 results in an odd hybrid.
 195  *
 196  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 197  * to translate the destination address to another local interface.
 198  *
 199  * XXX - ip_checkinterface must be disabled if you add IP aliases
 200  * to the loopback interface instead of the interface where the
 201  * packets for those addresses are received.
 202  */
 203 static int ip_checkinterface = 0;
 204 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 205     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 206
 207 #ifdef DIAGNOSTIC
 208 static int ipprintfs = 0;
 209 #endif
 210
 211 extern  int udp_mpsafe_proto;
 212 extern  int tcp_mpsafe_proto;
 213
 214 extern  struct domain inetdomain;
 215 extern  struct protosw inetsw[];
 216 u_char  ip_protox[IPPROTO_MAX];
 217 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 218 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 219                                                 /* inet addr hash table */
 220 u_long  in_ifaddrhmask;                         /* mask for hash table */
 221
 222 struct ip_stats ipstats_percpu[MAXCPU];
 223 #ifdef SMP
 224 static int
 225 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 226 {
 227         int cpu, error = 0;
 228
 229         for (cpu = 0; cpu < ncpus; ++cpu) {
 230                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 231                                         sizeof(struct ip_stats))))
 232                         break;
 233                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 234                                        sizeof(struct ip_stats))))
 235                         break;
 236         }
 237
 238         return (error);
 239 }
 240 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 241     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 242 #else
 243 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 244     &ipstat, ip_stats, "IP statistics");
 245 #endif
 246
 247 /* Packet reassembly stuff */
 248 #define IPREASS_NHASH_LOG2      6
 249 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 250 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 251 #define IPREASS_HASH(x,y)                                               \
 252     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 253
 254 static struct ipq ipq[IPREASS_NHASH];
 255
 256 #ifdef IPCTL_DEFMTU
 257 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 258     &ip_mtu, 0, "Default MTU");
 259 #endif
 260
 261 #ifdef IPSTEALTH
 262 static int ipstealth = 0;
 263 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 264 #else
 265 static const int ipstealth = 0;
 266 #endif
 267
 268 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
 269
 270 struct pfil_head inet_pfil_hook;
 271
 272 /*
 273  * struct ip_srcrt_opt is used to store packet state while it travels
 274  * through the stack.
 275  *
 276  * XXX Note that the code even makes assumptions on the size and
 277  * alignment of fields inside struct ip_srcrt so e.g. adding some
 278  * fields will break the code.  This needs to be fixed.
 279  *
 280  * We need to save the IP options in case a protocol wants to respond
 281  * to an incoming packet over the same route if the packet got here
 282  * using IP source routing.  This allows connection establishment and
 283  * maintenance when the remote end is on a network that is not known
 284  * to us.
 285  */
 286 struct ip_srcrt {
 287         struct  in_addr dst;                    /* final destination */
 288         char    nop;                            /* one NOP to align */
 289         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 290         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 291 };
 292
 293 struct ip_srcrt_opt {
 294         int             ip_nhops;
 295         struct ip_srcrt ip_srcrt;
 296 };
 297
 298 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 299 static struct malloc_pipe ipq_mpipe;
 300
 301 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
 302 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 303 static void             ip_freef(struct ipq *);
 304 static void             ip_input_handler(struct netmsg *);
 305
 306 /*
 307  * IP initialization: fill in IP protocol switch table.
 308  * All protocols not implemented in kernel go to raw IP protocol handler.
 309  */
 310 void
 311 ip_init(void)
 312 {
 313         struct protosw *pr;
 314         uint32_t flags;
 315         int i;
 316 #ifdef SMP
 317         int cpu;
 318 #endif
 319
 320         /*
 321          * Make sure we can handle a reasonable number of fragments but
 322          * cap it at 4000 (XXX).
 323          */
 324         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 325                     IFQ_MAXLEN, 4000, 0, NULL);
 326         for (i = 0; i < ncpus; ++i) {
 327                 TAILQ_INIT(&in_ifaddrheads[i]);
 328                 in_ifaddrhashtbls[i] =
 329                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 330         }
 331         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 332         if (pr == NULL)
 333                 panic("ip_init");
 334         for (i = 0; i < IPPROTO_MAX; i++)
 335                 ip_protox[i] = pr - inetsw;
 336         for (pr = inetdomain.dom_protosw;
 337              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
 338                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
 339                         if (pr->pr_protocol != IPPROTO_RAW)
 340                                 ip_protox[pr->pr_protocol] = pr - inetsw;
 341
 342                         /* XXX */
 343                         switch (pr->pr_protocol) {
 344                         case IPPROTO_TCP:
 345                                 if (tcp_mpsafe_proto)
 346                                         pr->pr_flags |= PR_MPSAFE;
 347                                 break;
 348
 349                         case IPPROTO_UDP:
 350                                 if (udp_mpsafe_proto)
 351                                         pr->pr_flags |= PR_MPSAFE;
 352                                 break;
 353                         }
 354                 }
 355         }
 356
 357         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 358         inet_pfil_hook.ph_af = AF_INET;
 359         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 360                 kprintf("%s: WARNING: unable to register pfil hook, "
 361                         "error %d\n", __func__, i);
 362         }
 363
 364         for (i = 0; i < IPREASS_NHASH; i++)
 365                 ipq[i].next = ipq[i].prev = &ipq[i];
 366
 367         maxnipq = nmbclusters / 32;
 368         maxfragsperpacket = 16;
 369
 370         ip_id = time_second & 0xffff;
 371
 372         /*
 373          * Initialize IP statistics counters for each CPU.
 374          *
 375          */
 376 #ifdef SMP
 377         for (cpu = 0; cpu < ncpus; ++cpu) {
 378                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 379         }
 380 #else
 381         bzero(&ipstat, sizeof(struct ip_stats));
 382 #endif
 383
 384 #if defined(IPSEC) || defined(FAST_IPSEC)
 385         /* XXX IPSEC is not MPSAFE yet */
 386         flags = NETISR_FLAG_NOTMPSAFE;
 387 #else
 388         if (ip_mpsafe) {
 389                 kprintf("ip: MPSAFE\n");
 390                 flags = NETISR_FLAG_MPSAFE;
 391         } else {
 392                 flags = NETISR_FLAG_NOTMPSAFE;
 393         }
 394 #endif
 395         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler, flags);
 396 }
 397
 398 /*
 399  * XXX watch out this one. It is perhaps used as a cache for
 400  * the most recently used route ? it is cleared in in_addroute()
 401  * when a new route is successfully created.
 402  */
 403 struct route ipforward_rt[MAXCPU];
 404
 405 /* Do transport protocol processing. */
 406 static void
 407 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 408 {
 409         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 410
 411         /*
 412          * Switch out to protocol's input routine.
 413          */
 414         PR_GET_MPLOCK(pr);
 415         pr->pr_input(m, hlen, ip->ip_p);
 416         PR_REL_MPLOCK(pr);
 417 }
 418
 419 static void
 420 transport_processing_handler(netmsg_t netmsg)
 421 {
 422         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 423         struct ip *ip;
 424         int hlen;
 425
 426         ip = mtod(pmsg->nm_packet, struct ip *);
 427         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 428
 429         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 430         /* netmsg was embedded in the mbuf, do not reply! */
 431 }
 432
 433 static void
 434 ip_input_handler(struct netmsg *msg0)
 435 {
 436         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 437
 438         ip_input(m);
 439         /* msg0 was embedded in the mbuf, do not reply! */
 440 }
 441
 442 /*
 443  * IP input routine.  Checksum and byte swap header.  If fragmented
 444  * try to reassemble.  Process options.  Pass to next level.
 445  */
 446 void
 447 ip_input(struct mbuf *m)
 448 {
 449         struct ip *ip;
 450         struct in_ifaddr *ia = NULL;
 451         struct in_ifaddr_container *iac;
 452         int hlen, checkif;
 453         u_short sum;
 454         struct in_addr pkt_dst;
 455         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 456         boolean_t needredispatch = FALSE;
 457         struct in_addr odst;                    /* original dst address(NAT) */
 458         struct m_tag *mtag;
 459         struct sockaddr_in *next_hop = NULL;
 460 #ifdef FAST_IPSEC
 461         struct tdb_ident *tdbi;
 462         struct secpolicy *sp;
 463         int error;
 464 #endif
 465
 466         M_ASSERTPKTHDR(m);
 467
 468         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 469                 /* Next hop */
 470                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 471                 KKASSERT(mtag != NULL);
 472                 next_hop = m_tag_data(mtag);
 473         }
 474
 475         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 476                 /* dummynet already filtered us */
 477                 ip = mtod(m, struct ip *);
 478                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 479                 goto iphack;
 480         }
 481
 482         ipstat.ips_total++;
 483
 484         /* length checks already done in ip_mport() */
 485         KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
 486         ip = mtod(m, struct ip *);
 487
 488         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 489                 ipstat.ips_badvers++;
 490                 goto bad;
 491         }
 492
 493         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 494         /* length checks already done in ip_mport() */
 495         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 496         KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
 497
 498         /* 127/8 must not appear on wire - RFC1122 */
 499         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 500             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 501                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 502                         ipstat.ips_badaddr++;
 503                         goto bad;
 504                 }
 505         }
 506
 507         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 508                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 509         } else {
 510                 if (hlen == sizeof(struct ip))
 511                         sum = in_cksum_hdr(ip);
 512                 else
 513                         sum = in_cksum(m, hlen);
 514         }
 515         if (sum != 0) {
 516                 ipstat.ips_badsum++;
 517                 goto bad;
 518         }
 519
 520 #ifdef ALTQ
 521         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 522                 /* packet is dropped by traffic conditioner */
 523                 return;
 524         }
 525 #endif
 526         /*
 527          * Convert fields to host representation.
 528          */
 529         ip->ip_len = ntohs(ip->ip_len);
 530         ip->ip_off = ntohs(ip->ip_off);
 531
 532         /* length checks already done in ip_mport() */
 533         KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
 534         KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
 535
 536         /*
 537          * Trim mbufs if longer than the IP header would have us expect.
 538          */
 539         if (m->m_pkthdr.len > ip->ip_len) {
 540                 if (m->m_len == m->m_pkthdr.len) {
 541                         m->m_len = ip->ip_len;
 542                         m->m_pkthdr.len = ip->ip_len;
 543                 } else {
 544                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 545                 }
 546         }
 547 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 548         /*
 549          * Bypass packet filtering for packets from a tunnel (gif).
 550          */
 551         if (ipsec_gethist(m, NULL))
 552                 goto pass;
 553 #endif
 554
 555         /*
 556          * IpHack's section.
 557          * Right now when no processing on packet has done
 558          * and it is still fresh out of network we do our black
 559          * deals with it.
 560          * - Firewall: deny/allow/divert
 561          * - Xlate: translate packet's addr/port (NAT).
 562          * - Pipe: pass pkt through dummynet.
 563          * - Wrap: fake packet's addr/port <unimpl.>
 564          * - Encapsulate: put it in another IP and send out. <unimp.>
 565          */
 566
 567 iphack:
 568         /*
 569          * If we've been forwarded from the output side, then
 570          * skip the firewall a second time
 571          */
 572         if (next_hop != NULL)
 573                 goto ours;
 574
 575         /* No pfil hooks */
 576         if (!pfil_has_hooks(&inet_pfil_hook)) {
 577                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 578                         /*
 579                          * Strip dummynet tags from stranded packets
 580                          */
 581                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 582                         KKASSERT(mtag != NULL);
 583                         m_tag_delete(m, mtag);
 584                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 585                 }
 586                 goto pass;
 587         }
 588
 589         /*
 590          * Run through list of hooks for input packets.
 591          *
 592          * NB: Beware of the destination address changing (e.g.
 593          *     by NAT rewriting). When this happens, tell
 594          *     ip_forward to do the right thing.
 595          */
 596         odst = ip->ip_dst;
 597         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 598                 return;
 599         if (m == NULL)  /* consumed by filter */
 600                 return;
 601         ip = mtod(m, struct ip *);
 602         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 603         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 604
 605         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 606                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 607                 KKASSERT(mtag != NULL);
 608                 next_hop = m_tag_data(mtag);
 609         }
 610         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 611                 ip_dn_queue(m);
 612                 return;
 613         }
 614         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 615                 needredispatch = TRUE;
 616                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 617         }
 618 pass:
 619         /*
 620          * Process options and, if not destined for us,
 621          * ship it on.  ip_dooptions returns 1 when an
 622          * error was detected (causing an icmp message
 623          * to be sent and the original packet to be freed).
 624          */
 625         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 626                 return;
 627
 628         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 629          * matter if it is destined to another node, or whether it is
 630          * a multicast one, RSVP wants it! and prevents it from being forwarded
 631          * anywhere else. Also checks if the rsvp daemon is running before
 632          * grabbing the packet.
 633          */
 634         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 635                 goto ours;
 636
 637         /*
 638          * Check our list of addresses, to see if the packet is for us.
 639          * If we don't have any addresses, assume any unicast packet
 640          * we receive might be for us (and let the upper layers deal
 641          * with it).
 642          */
 643         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 644             !(m->m_flags & (M_MCAST | M_BCAST)))
 645                 goto ours;
 646
 647         /*
 648          * Cache the destination address of the packet; this may be
 649          * changed by use of 'ipfw fwd'.
 650          */
 651         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 652
 653         /*
 654          * Enable a consistency check between the destination address
 655          * and the arrival interface for a unicast packet (the RFC 1122
 656          * strong ES model) if IP forwarding is disabled and the packet
 657          * is not locally generated and the packet is not subject to
 658          * 'ipfw fwd'.
 659          *
 660          * XXX - Checking also should be disabled if the destination
 661          * address is ipnat'ed to a different interface.
 662          *
 663          * XXX - Checking is incompatible with IP aliases added
 664          * to the loopback interface instead of the interface where
 665          * the packets are received.
 666          */
 667         checkif = ip_checkinterface &&
 668                   !ipforwarding &&
 669                   m->m_pkthdr.rcvif != NULL &&
 670                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 671                   next_hop == NULL;
 672
 673         /*
 674          * Check for exact addresses in the hash bucket.
 675          */
 676         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 677                 ia = iac->ia;
 678
 679                 /*
 680                  * If the address matches, verify that the packet
 681                  * arrived via the correct interface if checking is
 682                  * enabled.
 683                  */
 684                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 685                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 686                         goto ours;
 687         }
 688         ia = NULL;
 689
 690         /*
 691          * Check for broadcast addresses.
 692          *
 693          * Only accept broadcast packets that arrive via the matching
 694          * interface.  Reception of forwarded directed broadcasts would
 695          * be handled via ip_forward() and ether_output() with the loopback
 696          * into the stack for SIMPLEX interfaces handled by ether_output().
 697          */
 698         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 699                 struct ifaddr_container *ifac;
 700
 701                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 702                               ifa_link) {
 703                         struct ifaddr *ifa = ifac->ifa;
 704
 705                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 706                                 continue;
 707                         if (ifa->ifa_addr->sa_family != AF_INET)
 708                                 continue;
 709                         ia = ifatoia(ifa);
 710                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 711                                                                 pkt_dst.s_addr)
 712                                 goto ours;
 713                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 714                                 goto ours;
 715 #ifdef BOOTP_COMPAT
 716                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 717                                 goto ours;
 718 #endif
 719                 }
 720         }
 721         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 722                 struct in_multi *inm;
 723
 724                 /* XXX Multicast is not MPSAFE yet */
 725                 get_mplock();
 726
 727                 if (ip_mrouter != NULL) {
 728                         /*
 729                          * If we are acting as a multicast router, all
 730                          * incoming multicast packets are passed to the
 731                          * kernel-level multicast forwarding function.
 732                          * The packet is returned (relatively) intact; if
 733                          * ip_mforward() returns a non-zero value, the packet
 734                          * must be discarded, else it may be accepted below.
 735                          */
 736                         if (ip_mforward != NULL &&
 737                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 738                                 rel_mplock();
 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                                 rel_mplock();
 751                                 goto ours;
 752                         }
 753                         ipstat.ips_forward++;
 754                 }
 755                 /*
 756                  * See if we belong to the destination multicast group on the
 757                  * arrival interface.
 758                  */
 759                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 760                 if (inm == NULL) {
 761                         rel_mplock();
 762                         ipstat.ips_notmember++;
 763                         m_freem(m);
 764                         return;
 765                 }
 766
 767                 rel_mplock();
 768                 goto ours;
 769         }
 770         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 771                 goto ours;
 772         if (ip->ip_dst.s_addr == INADDR_ANY)
 773                 goto ours;
 774
 775         /*
 776          * FAITH(Firewall Aided Internet Translator)
 777          */
 778         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 779                 if (ip_keepfaith) {
 780                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 781                                 goto ours;
 782                 }
 783                 m_freem(m);
 784                 return;
 785         }
 786
 787         /*
 788          * Not for us; forward if possible and desirable.
 789          */
 790         if (!ipforwarding) {
 791                 ipstat.ips_cantforward++;
 792                 m_freem(m);
 793         } else {
 794 #ifdef IPSEC
 795                 /*
 796                  * Enforce inbound IPsec SPD.
 797                  */
 798                 if (ipsec4_in_reject(m, NULL)) {
 799                         ipsecstat.in_polvio++;
 800                         goto bad;
 801                 }
 802 #endif
 803 #ifdef FAST_IPSEC
 804                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 805                 crit_enter();
 806                 if (mtag != NULL) {
 807                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 808                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 809                 } else {
 810                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 811                                                    IP_FORWARDING, &error);
 812                 }
 813                 if (sp == NULL) {       /* NB: can happen if error */
 814                         crit_exit();
 815                         /*XXX error stat???*/
 816                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 817                         goto bad;
 818                 }
 819
 820                 /*
 821                  * Check security policy against packet attributes.
 822                  */
 823                 error = ipsec_in_reject(sp, m);
 824                 KEY_FREESP(&sp);
 825                 crit_exit();
 826                 if (error) {
 827                         ipstat.ips_cantforward++;
 828                         goto bad;
 829                 }
 830 #endif
 831                 ip_forward(m, using_srcrt, next_hop);
 832         }
 833         return;
 834
 835 ours:
 836
 837         /*
 838          * IPSTEALTH: Process non-routing options only
 839          * if the packet is destined for us.
 840          */
 841         if (ipstealth &&
 842             hlen > sizeof(struct ip) &&
 843             ip_dooptions(m, 1, next_hop))
 844                 return;
 845
 846         /* Count the packet in the ip address stats */
 847         if (ia != NULL) {
 848                 ia->ia_ifa.if_ipackets++;
 849                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 850         }
 851
 852         /*
 853          * If offset or IP_MF are set, must reassemble.
 854          * Otherwise, nothing need be done.
 855          * (We could look in the reassembly queue to see
 856          * if the packet was previously fragmented,
 857          * but it's not worth the time; just let them time out.)
 858          */
 859         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 860                 /*
 861                  * Attempt reassembly; if it succeeds, proceed.
 862                  * ip_reass() will return a different mbuf.
 863                  */
 864                 m = ip_reass(m);
 865                 if (m == NULL)
 866                         return;
 867                 ip = mtod(m, struct ip *);
 868
 869                 /* Get the header length of the reassembled packet */
 870                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 871
 872                 needredispatch = TRUE;
 873         } else {
 874                 ip->ip_len -= hlen;
 875         }
 876
 877 #ifdef IPSEC
 878         /*
 879          * enforce IPsec policy checking if we are seeing last header.
 880          * note that we do not visit this with protocols with pcb layer
 881          * code - like udp/tcp/raw ip.
 882          */
 883         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 884             ipsec4_in_reject(m, NULL)) {
 885                 ipsecstat.in_polvio++;
 886                 goto bad;
 887         }
 888 #endif
 889 #if FAST_IPSEC
 890         /*
 891          * enforce IPsec policy checking if we are seeing last header.
 892          * note that we do not visit this with protocols with pcb layer
 893          * code - like udp/tcp/raw ip.
 894          */
 895         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 896                 /*
 897                  * Check if the packet has already had IPsec processing
 898                  * done.  If so, then just pass it along.  This tag gets
 899                  * set during AH, ESP, etc. input handling, before the
 900                  * packet is returned to the ip input queue for delivery.
 901                  */
 902                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 903                 crit_enter();
 904                 if (mtag != NULL) {
 905                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 906                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 907                 } else {
 908                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 909                                                    IP_FORWARDING, &error);
 910                 }
 911                 if (sp != NULL) {
 912                         /*
 913                          * Check security policy against packet attributes.
 914                          */
 915                         error = ipsec_in_reject(sp, m);
 916                         KEY_FREESP(&sp);
 917                 } else {
 918                         /* XXX error stat??? */
 919                         error = EINVAL;
 920 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 921                         goto bad;
 922                 }
 923                 crit_exit();
 924                 if (error)
 925                         goto bad;
 926         }
 927 #endif /* FAST_IPSEC */
 928
 929         ipstat.ips_delivered++;
 930         if (needredispatch) {
 931                 struct netmsg_packet *pmsg;
 932                 lwkt_port_t port;
 933
 934                 ip->ip_off = htons(ip->ip_off);
 935                 ip->ip_len = htons(ip->ip_len);
 936                 port = ip_mport_in(&m);
 937                 if (port == NULL)
 938                         return;
 939
 940                 pmsg = &m->m_hdr.mh_netmsg;
 941                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, MSGF_MPSAFE,
 942                             transport_processing_handler);
 943                 pmsg->nm_packet = m;
 944                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
 945
 946                 ip = mtod(m, struct ip *);
 947                 ip->ip_len = ntohs(ip->ip_len);
 948                 ip->ip_off = ntohs(ip->ip_off);
 949                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
 950         } else {
 951                 transport_processing_oncpu(m, hlen, ip);
 952         }
 953         return;
 954
 955 bad:
 956         m_freem(m);
 957 }
 958
 959 /*
 960  * Take incoming datagram fragment and try to reassemble it into
 961  * whole datagram.  If a chain for reassembly of this datagram already
 962  * exists, then it is given as fp; otherwise have to make a chain.
 963  */
 964 struct mbuf *
 965 ip_reass(struct mbuf *m)
 966 {
 967         struct ip *ip = mtod(m, struct ip *);
 968         struct mbuf *p = NULL, *q, *nq;
 969         struct mbuf *n;
 970         struct ipq *fp = NULL;
 971         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 972         int i, next;
 973         u_short sum;
 974
 975         /* If maxnipq is 0, never accept fragments. */
 976         if (maxnipq == 0) {
 977                 ipstat.ips_fragments++;
 978                 ipstat.ips_fragdropped++;
 979                 m_freem(m);
 980                 return NULL;
 981         }
 982
 983         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 984         /*
 985          * Look for queue of fragments of this datagram.
 986          */
 987         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 988                 if (ip->ip_id == fp->ipq_id &&
 989                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 990                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 991                     ip->ip_p == fp->ipq_p)
 992                         goto found;
 993
 994         fp = NULL;
 995
 996         /*
 997          * Enforce upper bound on number of fragmented packets
 998          * for which we attempt reassembly;
 999          * If maxnipq is -1, accept all fragments without limitation.
1000          */
1001         if (nipq > maxnipq && maxnipq > 0) {
1002                 /*
1003                  * drop something from the tail of the current queue
1004                  * before proceeding further
1005                  */
1006                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1007                         for (i = 0; i < IPREASS_NHASH; i++) {
1008                                 if (ipq[i].prev != &ipq[i]) {
1009                                         ipstat.ips_fragtimeout +=
1010                                             ipq[i].prev->ipq_nfrags;
1011                                         ip_freef(ipq[i].prev);
1012                                         break;
1013                                 }
1014                         }
1015                 } else {
1016                         ipstat.ips_fragtimeout +=
1017                             ipq[sum].prev->ipq_nfrags;
1018                         ip_freef(ipq[sum].prev);
1019                 }
1020         }
1021 found:
1022         /*
1023          * Adjust ip_len to not reflect header,
1024          * convert offset of this to bytes.
1025          */
1026         ip->ip_len -= hlen;
1027         if (ip->ip_off & IP_MF) {
1028                 /*
1029                  * Make sure that fragments have a data length
1030                  * that's a non-zero multiple of 8 bytes.
1031                  */
1032                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1033                         ipstat.ips_toosmall++; /* XXX */
1034                         m_freem(m);
1035                         return NULL;
1036                 }
1037                 m->m_flags |= M_FRAG;
1038         } else
1039                 m->m_flags &= ~M_FRAG;
1040         ip->ip_off <<= 3;
1041
1042         ipstat.ips_fragments++;
1043         m->m_pkthdr.header = ip;
1044
1045         /*
1046          * If the hardware has not done csum over this fragment
1047          * then csum_data is not valid at all.
1048          */
1049         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1050             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1051                 m->m_pkthdr.csum_data = 0;
1052                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1053         }
1054
1055         /*
1056          * Presence of header sizes in mbufs
1057          * would confuse code below.
1058          */
1059         m->m_data += hlen;
1060         m->m_len -= hlen;
1061
1062         /*
1063          * If first fragment to arrive, create a reassembly queue.
1064          */
1065         if (fp == NULL) {
1066                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1067                         goto dropfrag;
1068                 insque(fp, &ipq[sum]);
1069                 nipq++;
1070                 fp->ipq_nfrags = 1;
1071                 fp->ipq_ttl = IPFRAGTTL;
1072                 fp->ipq_p = ip->ip_p;
1073                 fp->ipq_id = ip->ip_id;
1074                 fp->ipq_src = ip->ip_src;
1075                 fp->ipq_dst = ip->ip_dst;
1076                 fp->ipq_frags = m;
1077                 m->m_nextpkt = NULL;
1078                 goto inserted;
1079         } else {
1080                 fp->ipq_nfrags++;
1081         }
1082
1083 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1084
1085         /*
1086          * Find a segment which begins after this one does.
1087          */
1088         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1089                 if (GETIP(q)->ip_off > ip->ip_off)
1090                         break;
1091
1092         /*
1093          * If there is a preceding segment, it may provide some of
1094          * our data already.  If so, drop the data from the incoming
1095          * segment.  If it provides all of our data, drop us, otherwise
1096          * stick new segment in the proper place.
1097          *
1098          * If some of the data is dropped from the the preceding
1099          * segment, then it's checksum is invalidated.
1100          */
1101         if (p) {
1102                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1103                 if (i > 0) {
1104                         if (i >= ip->ip_len)
1105                                 goto dropfrag;
1106                         m_adj(m, i);
1107                         m->m_pkthdr.csum_flags = 0;
1108                         ip->ip_off += i;
1109                         ip->ip_len -= i;
1110                 }
1111                 m->m_nextpkt = p->m_nextpkt;
1112                 p->m_nextpkt = m;
1113         } else {
1114                 m->m_nextpkt = fp->ipq_frags;
1115                 fp->ipq_frags = m;
1116         }
1117
1118         /*
1119          * While we overlap succeeding segments trim them or,
1120          * if they are completely covered, dequeue them.
1121          */
1122         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1123              q = nq) {
1124                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1125                 if (i < GETIP(q)->ip_len) {
1126                         GETIP(q)->ip_len -= i;
1127                         GETIP(q)->ip_off += i;
1128                         m_adj(q, i);
1129                         q->m_pkthdr.csum_flags = 0;
1130                         break;
1131                 }
1132                 nq = q->m_nextpkt;
1133                 m->m_nextpkt = nq;
1134                 ipstat.ips_fragdropped++;
1135                 fp->ipq_nfrags--;
1136                 q->m_nextpkt = NULL;
1137                 m_freem(q);
1138         }
1139
1140 inserted:
1141         /*
1142          * Check for complete reassembly and perform frag per packet
1143          * limiting.
1144          *
1145          * Frag limiting is performed here so that the nth frag has
1146          * a chance to complete the packet before we drop the packet.
1147          * As a result, n+1 frags are actually allowed per packet, but
1148          * only n will ever be stored. (n = maxfragsperpacket.)
1149          *
1150          */
1151         next = 0;
1152         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1153                 if (GETIP(q)->ip_off != next) {
1154                         if (fp->ipq_nfrags > maxfragsperpacket) {
1155                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1156                                 ip_freef(fp);
1157                         }
1158                         return (NULL);
1159                 }
1160                 next += GETIP(q)->ip_len;
1161         }
1162         /* Make sure the last packet didn't have the IP_MF flag */
1163         if (p->m_flags & M_FRAG) {
1164                 if (fp->ipq_nfrags > maxfragsperpacket) {
1165                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1166                         ip_freef(fp);
1167                 }
1168                 return (NULL);
1169         }
1170
1171         /*
1172          * Reassembly is complete.  Make sure the packet is a sane size.
1173          */
1174         q = fp->ipq_frags;
1175         ip = GETIP(q);
1176         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1177                 ipstat.ips_toolong++;
1178                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1179                 ip_freef(fp);
1180                 return (NULL);
1181         }
1182
1183         /*
1184          * Concatenate fragments.
1185          */
1186         m = q;
1187         n = m->m_next;
1188         m->m_next = NULL;
1189         m_cat(m, n);
1190         nq = q->m_nextpkt;
1191         q->m_nextpkt = NULL;
1192         for (q = nq; q != NULL; q = nq) {
1193                 nq = q->m_nextpkt;
1194                 q->m_nextpkt = NULL;
1195                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1196                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1197                 m_cat(m, q);
1198         }
1199
1200         /*
1201          * Clean up the 1's complement checksum.  Carry over 16 bits must
1202          * be added back.  This assumes no more then 65535 packet fragments
1203          * were reassembled.  A second carry can also occur (but not a third).
1204          */
1205         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1206                                 (m->m_pkthdr.csum_data >> 16);
1207         if (m->m_pkthdr.csum_data > 0xFFFF)
1208                 m->m_pkthdr.csum_data -= 0xFFFF;
1209
1210         /*
1211          * Create header for new ip packet by
1212          * modifying header of first packet;
1213          * dequeue and discard fragment reassembly header.
1214          * Make header visible.
1215          */
1216         ip->ip_len = next;
1217         ip->ip_src = fp->ipq_src;
1218         ip->ip_dst = fp->ipq_dst;
1219         remque(fp);
1220         nipq--;
1221         mpipe_free(&ipq_mpipe, fp);
1222         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1223         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1224         /* some debugging cruft by sklower, below, will go away soon */
1225         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1226                 int plen = 0;
1227
1228                 for (n = m; n; n = n->m_next)
1229                         plen += n->m_len;
1230                 m->m_pkthdr.len = plen;
1231         }
1232
1233         ipstat.ips_reassembled++;
1234         return (m);
1235
1236 dropfrag:
1237         ipstat.ips_fragdropped++;
1238         if (fp != NULL)
1239                 fp->ipq_nfrags--;
1240         m_freem(m);
1241         return (NULL);
1242
1243 #undef GETIP
1244 }
1245
1246 /*
1247  * Free a fragment reassembly header and all
1248  * associated datagrams.
1249  */
1250 static void
1251 ip_freef(struct ipq *fp)
1252 {
1253         struct mbuf *q;
1254
1255         while (fp->ipq_frags) {
1256                 q = fp->ipq_frags;
1257                 fp->ipq_frags = q->m_nextpkt;
1258                 q->m_nextpkt = NULL;
1259                 m_freem(q);
1260         }
1261         remque(fp);
1262         mpipe_free(&ipq_mpipe, fp);
1263         nipq--;
1264 }
1265
1266 /*
1267  * IP timer processing;
1268  * if a timer expires on a reassembly
1269  * queue, discard it.
1270  */
1271 void
1272 ip_slowtimo(void)
1273 {
1274         struct ipq *fp;
1275         int i;
1276
1277         crit_enter();
1278         for (i = 0; i < IPREASS_NHASH; i++) {
1279                 fp = ipq[i].next;
1280                 if (fp == NULL)
1281                         continue;
1282                 while (fp != &ipq[i]) {
1283                         --fp->ipq_ttl;
1284                         fp = fp->next;
1285                         if (fp->prev->ipq_ttl == 0) {
1286                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1287                                 ip_freef(fp->prev);
1288                         }
1289                 }
1290         }
1291         /*
1292          * If we are over the maximum number of fragments
1293          * (due to the limit being lowered), drain off
1294          * enough to get down to the new limit.
1295          */
1296         if (maxnipq >= 0 && nipq > maxnipq) {
1297                 for (i = 0; i < IPREASS_NHASH; i++) {
1298                         while (nipq > maxnipq &&
1299                                 (ipq[i].next != &ipq[i])) {
1300                                 ipstat.ips_fragdropped +=
1301                                     ipq[i].next->ipq_nfrags;
1302                                 ip_freef(ipq[i].next);
1303                         }
1304                 }
1305         }
1306         ipflow_slowtimo();
1307         crit_exit();
1308 }
1309
1310 /*
1311  * Drain off all datagram fragments.
1312  */
1313 void
1314 ip_drain(void)
1315 {
1316         int i;
1317
1318         for (i = 0; i < IPREASS_NHASH; i++) {
1319                 while (ipq[i].next != &ipq[i]) {
1320                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1321                         ip_freef(ipq[i].next);
1322                 }
1323         }
1324         in_rtqdrain();
1325 }
1326
1327 /*
1328  * Do option processing on a datagram,
1329  * possibly discarding it if bad options are encountered,
1330  * or forwarding it if source-routed.
1331  * The pass argument is used when operating in the IPSTEALTH
1332  * mode to tell what options to process:
1333  * [LS]SRR (pass 0) or the others (pass 1).
1334  * The reason for as many as two passes is that when doing IPSTEALTH,
1335  * non-routing options should be processed only if the packet is for us.
1336  * Returns 1 if packet has been forwarded/freed,
1337  * 0 if the packet should be processed further.
1338  */
1339 static int
1340 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1341 {
1342         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1343         struct ip *ip = mtod(m, struct ip *);
1344         u_char *cp;
1345         struct in_ifaddr *ia;
1346         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1347         boolean_t forward = FALSE;
1348         struct in_addr *sin, dst;
1349         n_time ntime;
1350
1351         dst = ip->ip_dst;
1352         cp = (u_char *)(ip + 1);
1353         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1354         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1355                 opt = cp[IPOPT_OPTVAL];
1356                 if (opt == IPOPT_EOL)
1357                         break;
1358                 if (opt == IPOPT_NOP)
1359                         optlen = 1;
1360                 else {
1361                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1362                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1363                                 goto bad;
1364                         }
1365                         optlen = cp[IPOPT_OLEN];
1366                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1367                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1368                                 goto bad;
1369                         }
1370                 }
1371                 switch (opt) {
1372
1373                 default:
1374                         break;
1375
1376                 /*
1377                  * Source routing with record.
1378                  * Find interface with current destination address.
1379                  * If none on this machine then drop if strictly routed,
1380                  * or do nothing if loosely routed.
1381                  * Record interface address and bring up next address
1382                  * component.  If strictly routed make sure next
1383                  * address is on directly accessible net.
1384                  */
1385                 case IPOPT_LSRR:
1386                 case IPOPT_SSRR:
1387                         if (ipstealth && pass > 0)
1388                                 break;
1389                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1390                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1391                                 goto bad;
1392                         }
1393                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1394                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1395                                 goto bad;
1396                         }
1397                         ipaddr.sin_addr = ip->ip_dst;
1398                         ia = (struct in_ifaddr *)
1399                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1400                         if (ia == NULL) {
1401                                 if (opt == IPOPT_SSRR) {
1402                                         type = ICMP_UNREACH;
1403                                         code = ICMP_UNREACH_SRCFAIL;
1404                                         goto bad;
1405                                 }
1406                                 if (!ip_dosourceroute)
1407                                         goto nosourcerouting;
1408                                 /*
1409                                  * Loose routing, and not at next destination
1410                                  * yet; nothing to do except forward.
1411                                  */
1412                                 break;
1413                         }
1414                         off--;                  /* 0 origin */
1415                         if (off > optlen - (int)sizeof(struct in_addr)) {
1416                                 /*
1417                                  * End of source route.  Should be for us.
1418                                  */
1419                                 if (!ip_acceptsourceroute)
1420                                         goto nosourcerouting;
1421                                 save_rte(m, cp, ip->ip_src);
1422                                 break;
1423                         }
1424                         if (ipstealth)
1425                                 goto dropit;
1426                         if (!ip_dosourceroute) {
1427                                 if (ipforwarding) {
1428                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1429
1430                                         /*
1431                                          * Acting as a router, so generate ICMP
1432                                          */
1433 nosourcerouting:
1434                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1435                                         log(LOG_WARNING,
1436                                             "attempted source route from %s to %s\n",
1437                                             inet_ntoa(ip->ip_src), buf);
1438                                         type = ICMP_UNREACH;
1439                                         code = ICMP_UNREACH_SRCFAIL;
1440                                         goto bad;
1441                                 } else {
1442                                         /*
1443                                          * Not acting as a router,
1444                                          * so silently drop.
1445                                          */
1446 dropit:
1447                                         ipstat.ips_cantforward++;
1448                                         m_freem(m);
1449                                         return (1);
1450                                 }
1451                         }
1452
1453                         /*
1454                          * locate outgoing interface
1455                          */
1456                         memcpy(&ipaddr.sin_addr, cp + off,
1457                             sizeof ipaddr.sin_addr);
1458
1459                         if (opt == IPOPT_SSRR) {
1460 #define INA     struct in_ifaddr *
1461 #define SA      struct sockaddr *
1462                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1463                                                                         == NULL)
1464                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1465                         } else
1466                                 ia = ip_rtaddr(ipaddr.sin_addr,
1467                                                &ipforward_rt[mycpuid]);
1468                         if (ia == NULL) {
1469                                 type = ICMP_UNREACH;
1470                                 code = ICMP_UNREACH_SRCFAIL;
1471                                 goto bad;
1472                         }
1473                         ip->ip_dst = ipaddr.sin_addr;
1474                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1475                             sizeof(struct in_addr));
1476                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1477                         /*
1478                          * Let ip_intr's mcast routing check handle mcast pkts
1479                          */
1480                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1481                         break;
1482
1483                 case IPOPT_RR:
1484                         if (ipstealth && pass == 0)
1485                                 break;
1486                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1487                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1488                                 goto bad;
1489                         }
1490                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1491                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1492                                 goto bad;
1493                         }
1494                         /*
1495                          * If no space remains, ignore.
1496                          */
1497                         off--;                  /* 0 origin */
1498                         if (off > optlen - (int)sizeof(struct in_addr))
1499                                 break;
1500                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1501                             sizeof ipaddr.sin_addr);
1502                         /*
1503                          * locate outgoing interface; if we're the destination,
1504                          * use the incoming interface (should be same).
1505                          */
1506                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1507                             (ia = ip_rtaddr(ipaddr.sin_addr,
1508                                             &ipforward_rt[mycpuid]))
1509                                                                      == NULL) {
1510                                 type = ICMP_UNREACH;
1511                                 code = ICMP_UNREACH_HOST;
1512                                 goto bad;
1513                         }
1514                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1515                             sizeof(struct in_addr));
1516                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1517                         break;
1518
1519                 case IPOPT_TS:
1520                         if (ipstealth && pass == 0)
1521                                 break;
1522                         code = cp - (u_char *)ip;
1523                         if (optlen < 4 || optlen > 40) {
1524                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1525                                 goto bad;
1526                         }
1527                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1528                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1529                                 goto bad;
1530                         }
1531                         if (off > optlen - (int)sizeof(int32_t)) {
1532                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1533                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1534                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1535                                         goto bad;
1536                                 }
1537                                 break;
1538                         }
1539                         off--;                          /* 0 origin */
1540                         sin = (struct in_addr *)(cp + off);
1541                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1542
1543                         case IPOPT_TS_TSONLY:
1544                                 break;
1545
1546                         case IPOPT_TS_TSANDADDR:
1547                                 if (off + sizeof(n_time) +
1548                                     sizeof(struct in_addr) > optlen) {
1549                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1550                                         goto bad;
1551                                 }
1552                                 ipaddr.sin_addr = dst;
1553                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1554                                                             m->m_pkthdr.rcvif);
1555                                 if (ia == NULL)
1556                                         continue;
1557                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1558                                     sizeof(struct in_addr));
1559                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1560                                 off += sizeof(struct in_addr);
1561                                 break;
1562
1563                         case IPOPT_TS_PRESPEC:
1564                                 if (off + sizeof(n_time) +
1565                                     sizeof(struct in_addr) > optlen) {
1566                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1567                                         goto bad;
1568                                 }
1569                                 memcpy(&ipaddr.sin_addr, sin,
1570                                     sizeof(struct in_addr));
1571                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1572                                         continue;
1573                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1574                                 off += sizeof(struct in_addr);
1575                                 break;
1576
1577                         default:
1578                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1579                                 goto bad;
1580                         }
1581                         ntime = iptime();
1582                         memcpy(cp + off, &ntime, sizeof(n_time));
1583                         cp[IPOPT_OFFSET] += sizeof(n_time);
1584                 }
1585         }
1586         if (forward && ipforwarding) {
1587                 ip_forward(m, TRUE, next_hop);
1588                 return (1);
1589         }
1590         return (0);
1591 bad:
1592         icmp_error(m, type, code, 0, 0);
1593         ipstat.ips_badoptions++;
1594         return (1);
1595 }
1596
1597 /*
1598  * Given address of next destination (final or next hop),
1599  * return internet address info of interface to be used to get there.
1600  */
1601 struct in_ifaddr *
1602 ip_rtaddr(struct in_addr dst, struct route *ro)
1603 {
1604         struct sockaddr_in *sin;
1605
1606         sin = (struct sockaddr_in *)&ro->ro_dst;
1607
1608         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1609                 if (ro->ro_rt != NULL) {
1610                         RTFREE(ro->ro_rt);
1611                         ro->ro_rt = NULL;
1612                 }
1613                 sin->sin_family = AF_INET;
1614                 sin->sin_len = sizeof *sin;
1615                 sin->sin_addr = dst;
1616                 rtalloc_ign(ro, RTF_PRCLONING);
1617         }
1618
1619         if (ro->ro_rt == NULL)
1620                 return (NULL);
1621
1622         return (ifatoia(ro->ro_rt->rt_ifa));
1623 }
1624
1625 /*
1626  * Save incoming source route for use in replies,
1627  * to be picked up later by ip_srcroute if the receiver is interested.
1628  */
1629 static void
1630 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1631 {
1632         struct m_tag *mtag;
1633         struct ip_srcrt_opt *opt;
1634         unsigned olen;
1635
1636         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1637         if (mtag == NULL)
1638                 return;
1639         opt = m_tag_data(mtag);
1640
1641         olen = option[IPOPT_OLEN];
1642 #ifdef DIAGNOSTIC
1643         if (ipprintfs)
1644                 kprintf("save_rte: olen %d\n", olen);
1645 #endif
1646         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1647                 m_tag_free(mtag);
1648                 return;
1649         }
1650         bcopy(option, opt->ip_srcrt.srcopt, olen);
1651         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1652         opt->ip_srcrt.dst = dst;
1653         m_tag_prepend(m, mtag);
1654 }
1655
1656 /*
1657  * Retrieve incoming source route for use in replies,
1658  * in the same form used by setsockopt.
1659  * The first hop is placed before the options, will be removed later.
1660  */
1661 struct mbuf *
1662 ip_srcroute(struct mbuf *m0)
1663 {
1664         struct in_addr *p, *q;
1665         struct mbuf *m;
1666         struct m_tag *mtag;
1667         struct ip_srcrt_opt *opt;
1668
1669         if (m0 == NULL)
1670                 return NULL;
1671
1672         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1673         if (mtag == NULL)
1674                 return NULL;
1675         opt = m_tag_data(mtag);
1676
1677         if (opt->ip_nhops == 0)
1678                 return (NULL);
1679         m = m_get(MB_DONTWAIT, MT_HEADER);
1680         if (m == NULL)
1681                 return (NULL);
1682
1683 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1684
1685         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1686         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1687                    sizeof(struct in_addr) + OPTSIZ;
1688 #ifdef DIAGNOSTIC
1689         if (ipprintfs) {
1690                 kprintf("ip_srcroute: nhops %d mlen %d",
1691                         opt->ip_nhops, m->m_len);
1692         }
1693 #endif
1694
1695         /*
1696          * First save first hop for return route
1697          */
1698         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1699         *(mtod(m, struct in_addr *)) = *p--;
1700 #ifdef DIAGNOSTIC
1701         if (ipprintfs)
1702                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1703 #endif
1704
1705         /*
1706          * Copy option fields and padding (nop) to mbuf.
1707          */
1708         opt->ip_srcrt.nop = IPOPT_NOP;
1709         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1710         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1711             OPTSIZ);
1712         q = (struct in_addr *)(mtod(m, caddr_t) +
1713             sizeof(struct in_addr) + OPTSIZ);
1714 #undef OPTSIZ
1715         /*
1716          * Record return path as an IP source route,
1717          * reversing the path (pointers are now aligned).
1718          */
1719         while (p >= opt->ip_srcrt.route) {
1720 #ifdef DIAGNOSTIC
1721                 if (ipprintfs)
1722                         kprintf(" %x", ntohl(q->s_addr));
1723 #endif
1724                 *q++ = *p--;
1725         }
1726         /*
1727          * Last hop goes to final destination.
1728          */
1729         *q = opt->ip_srcrt.dst;
1730         m_tag_delete(m0, mtag);
1731 #ifdef DIAGNOSTIC
1732         if (ipprintfs)
1733                 kprintf(" %x\n", ntohl(q->s_addr));
1734 #endif
1735         return (m);
1736 }
1737
1738 /*
1739  * Strip out IP options.
1740  */
1741 void
1742 ip_stripoptions(struct mbuf *m)
1743 {
1744         int datalen;
1745         struct ip *ip = mtod(m, struct ip *);
1746         caddr_t opts;
1747         int optlen;
1748
1749         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1750         opts = (caddr_t)(ip + 1);
1751         datalen = m->m_len - (sizeof(struct ip) + optlen);
1752         bcopy(opts + optlen, opts, datalen);
1753         m->m_len -= optlen;
1754         if (m->m_flags & M_PKTHDR)
1755                 m->m_pkthdr.len -= optlen;
1756         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1757 }
1758
1759 u_char inetctlerrmap[PRC_NCMDS] = {
1760         0,              0,              0,              0,
1761         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1762         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1763         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1764         0,              0,              0,              0,
1765         ENOPROTOOPT,    ECONNREFUSED
1766 };
1767
1768 /*
1769  * Forward a packet.  If some error occurs return the sender
1770  * an icmp packet.  Note we can't always generate a meaningful
1771  * icmp message because icmp doesn't have a large enough repertoire
1772  * of codes and types.
1773  *
1774  * If not forwarding, just drop the packet.  This could be confusing
1775  * if ipforwarding was zero but some routing protocol was advancing
1776  * us as a gateway to somewhere.  However, we must let the routing
1777  * protocol deal with that.
1778  *
1779  * The using_srcrt parameter indicates whether the packet is being forwarded
1780  * via a source route.
1781  */
1782 void
1783 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1784 {
1785         struct ip *ip = mtod(m, struct ip *);
1786         struct sockaddr_in *ipforward_rtaddr;
1787         struct rtentry *rt;
1788         int error, type = 0, code = 0, destmtu = 0;
1789         struct mbuf *mcopy;
1790         n_long dest;
1791         struct in_addr pkt_dst;
1792         struct route *cache_rt = &ipforward_rt[mycpuid];
1793
1794         dest = INADDR_ANY;
1795         /*
1796          * Cache the destination address of the packet; this may be
1797          * changed by use of 'ipfw fwd'.
1798          */
1799         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1800
1801 #ifdef DIAGNOSTIC
1802         if (ipprintfs)
1803                 kprintf("forward: src %x dst %x ttl %x\n",
1804                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1805 #endif
1806
1807         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1808                 ipstat.ips_cantforward++;
1809                 m_freem(m);
1810                 return;
1811         }
1812         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1813                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1814                 return;
1815         }
1816
1817         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1818         if (cache_rt->ro_rt == NULL ||
1819             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1820                 if (cache_rt->ro_rt != NULL) {
1821                         RTFREE(cache_rt->ro_rt);
1822                         cache_rt->ro_rt = NULL;
1823                 }
1824                 ipforward_rtaddr->sin_family = AF_INET;
1825                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1826                 ipforward_rtaddr->sin_addr = pkt_dst;
1827                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1828                 if (cache_rt->ro_rt == NULL) {
1829                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1830                         return;
1831                 }
1832         }
1833         rt = cache_rt->ro_rt;
1834
1835         /*
1836          * Save the IP header and at most 8 bytes of the payload,
1837          * in case we need to generate an ICMP message to the src.
1838          *
1839          * XXX this can be optimized a lot by saving the data in a local
1840          * buffer on the stack (72 bytes at most), and only allocating the
1841          * mbuf if really necessary. The vast majority of the packets
1842          * are forwarded without having to send an ICMP back (either
1843          * because unnecessary, or because rate limited), so we are
1844          * really we are wasting a lot of work here.
1845          *
1846          * We don't use m_copy() because it might return a reference
1847          * to a shared cluster. Both this function and ip_output()
1848          * assume exclusive access to the IP header in `m', so any
1849          * data in a cluster may change before we reach icmp_error().
1850          */
1851         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1852         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1853                 /*
1854                  * It's probably ok if the pkthdr dup fails (because
1855                  * the deep copy of the tag chain failed), but for now
1856                  * be conservative and just discard the copy since
1857                  * code below may some day want the tags.
1858                  */
1859                 m_free(mcopy);
1860                 mcopy = NULL;
1861         }
1862         if (mcopy != NULL) {
1863                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1864                     (int)ip->ip_len);
1865                 mcopy->m_pkthdr.len = mcopy->m_len;
1866                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1867         }
1868
1869         if (!ipstealth)
1870                 ip->ip_ttl -= IPTTLDEC;
1871
1872         /*
1873          * If forwarding packet using same interface that it came in on,
1874          * perhaps should send a redirect to sender to shortcut a hop.
1875          * Only send redirect if source is sending directly to us,
1876          * and if packet was not source routed (or has any options).
1877          * Also, don't send redirect if forwarding using a default route
1878          * or a route modified by a redirect.
1879          */
1880         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1881             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1882             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1883             ipsendredirects && !using_srcrt && next_hop == NULL) {
1884                 u_long src = ntohl(ip->ip_src.s_addr);
1885                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1886
1887                 if (rt_ifa != NULL &&
1888                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1889                         if (rt->rt_flags & RTF_GATEWAY)
1890                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1891                         else
1892                                 dest = pkt_dst.s_addr;
1893                         /*
1894                          * Router requirements says to only send
1895                          * host redirects.
1896                          */
1897                         type = ICMP_REDIRECT;
1898                         code = ICMP_REDIRECT_HOST;
1899 #ifdef DIAGNOSTIC
1900                         if (ipprintfs)
1901                                 kprintf("redirect (%d) to %x\n", code, dest);
1902 #endif
1903                 }
1904         }
1905
1906         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1907         if (error == 0) {
1908                 ipstat.ips_forward++;
1909                 if (type == 0) {
1910                         if (mcopy) {
1911                                 ipflow_create(cache_rt, mcopy);
1912                                 m_freem(mcopy);
1913                         }
1914                         return;         /* most common case */
1915                 } else {
1916                         ipstat.ips_redirectsent++;
1917                 }
1918         } else {
1919                 ipstat.ips_cantforward++;
1920         }
1921
1922         if (mcopy == NULL)
1923                 return;
1924
1925         /*
1926          * Send ICMP message.
1927          */
1928
1929         switch (error) {
1930
1931         case 0:                         /* forwarded, but need redirect */
1932                 /* type, code set above */
1933                 break;
1934
1935         case ENETUNREACH:               /* shouldn't happen, checked above */
1936         case EHOSTUNREACH:
1937         case ENETDOWN:
1938         case EHOSTDOWN:
1939         default:
1940                 type = ICMP_UNREACH;
1941                 code = ICMP_UNREACH_HOST;
1942                 break;
1943
1944         case EMSGSIZE:
1945                 type = ICMP_UNREACH;
1946                 code = ICMP_UNREACH_NEEDFRAG;
1947 #ifdef IPSEC
1948                 /*
1949                  * If the packet is routed over IPsec tunnel, tell the
1950                  * originator the tunnel MTU.
1951                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1952                  * XXX quickhack!!!
1953                  */
1954                 if (cache_rt->ro_rt != NULL) {
1955                         struct secpolicy *sp = NULL;
1956                         int ipsecerror;
1957                         int ipsechdr;
1958                         struct route *ro;
1959
1960                         sp = ipsec4_getpolicybyaddr(mcopy,
1961                                                     IPSEC_DIR_OUTBOUND,
1962                                                     IP_FORWARDING,
1963                                                     &ipsecerror);
1964
1965                         if (sp == NULL)
1966                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1967                         else {
1968                                 /* count IPsec header size */
1969                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1970                                                          IPSEC_DIR_OUTBOUND,
1971                                                          NULL);
1972
1973                                 /*
1974                                  * find the correct route for outer IPv4
1975                                  * header, compute tunnel MTU.
1976                                  *
1977                                  */
1978                                 if (sp->req != NULL && sp->req->sav != NULL &&
1979                                     sp->req->sav->sah != NULL) {
1980                                         ro = &sp->req->sav->sah->sa_route;
1981                                         if (ro->ro_rt != NULL &&
1982                                             ro->ro_rt->rt_ifp != NULL) {
1983                                                 destmtu =
1984                                                     ro->ro_rt->rt_ifp->if_mtu;
1985                                                 destmtu -= ipsechdr;
1986                                         }
1987                                 }
1988
1989                                 key_freesp(sp);
1990                         }
1991                 }
1992 #elif FAST_IPSEC
1993                 /*
1994                  * If the packet is routed over IPsec tunnel, tell the
1995                  * originator the tunnel MTU.
1996                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1997                  * XXX quickhack!!!
1998                  */
1999                 if (cache_rt->ro_rt != NULL) {
2000                         struct secpolicy *sp = NULL;
2001                         int ipsecerror;
2002                         int ipsechdr;
2003                         struct route *ro;
2004
2005                         sp = ipsec_getpolicybyaddr(mcopy,
2006                                                    IPSEC_DIR_OUTBOUND,
2007                                                    IP_FORWARDING,
2008                                                    &ipsecerror);
2009
2010                         if (sp == NULL)
2011                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2012                         else {
2013                                 /* count IPsec header size */
2014                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2015                                                          IPSEC_DIR_OUTBOUND,
2016                                                          NULL);
2017
2018                                 /*
2019                                  * find the correct route for outer IPv4
2020                                  * header, compute tunnel MTU.
2021                                  */
2022
2023                                 if (sp->req != NULL &&
2024                                     sp->req->sav != NULL &&
2025                                     sp->req->sav->sah != NULL) {
2026                                         ro = &sp->req->sav->sah->sa_route;
2027                                         if (ro->ro_rt != NULL &&
2028                                             ro->ro_rt->rt_ifp != NULL) {
2029                                                 destmtu =
2030                                                     ro->ro_rt->rt_ifp->if_mtu;
2031                                                 destmtu -= ipsechdr;
2032                                         }
2033                                 }
2034
2035                                 KEY_FREESP(&sp);
2036                         }
2037                 }
2038 #else /* !IPSEC && !FAST_IPSEC */
2039                 if (cache_rt->ro_rt != NULL)
2040                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2041 #endif /*IPSEC*/
2042                 ipstat.ips_cantfrag++;
2043                 break;
2044
2045         case ENOBUFS:
2046                 /*
2047                  * A router should not generate ICMP_SOURCEQUENCH as
2048                  * required in RFC1812 Requirements for IP Version 4 Routers.
2049                  * Source quench could be a big problem under DoS attacks,
2050                  * or if the underlying interface is rate-limited.
2051                  * Those who need source quench packets may re-enable them
2052                  * via the net.inet.ip.sendsourcequench sysctl.
2053                  */
2054                 if (!ip_sendsourcequench) {
2055                         m_freem(mcopy);
2056                         return;
2057                 } else {
2058                         type = ICMP_SOURCEQUENCH;
2059                         code = 0;
2060                 }
2061                 break;
2062
2063         case EACCES:                    /* ipfw denied packet */
2064                 m_freem(mcopy);
2065                 return;
2066         }
2067         icmp_error(mcopy, type, code, dest, destmtu);
2068 }
2069
2070 void
2071 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2072                struct mbuf *m)
2073 {
2074         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2075                 struct timeval tv;
2076
2077                 microtime(&tv);
2078                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2079                     SCM_TIMESTAMP, SOL_SOCKET);
2080                 if (*mp)
2081                         mp = &(*mp)->m_next;
2082         }
2083         if (inp->inp_flags & INP_RECVDSTADDR) {
2084                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2085                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2086                 if (*mp)
2087                         mp = &(*mp)->m_next;
2088         }
2089         if (inp->inp_flags & INP_RECVTTL) {
2090                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2091                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2092                 if (*mp)
2093                         mp = &(*mp)->m_next;
2094         }
2095 #ifdef notyet
2096         /* XXX
2097          * Moving these out of udp_input() made them even more broken
2098          * than they already were.
2099          */
2100         /* options were tossed already */
2101         if (inp->inp_flags & INP_RECVOPTS) {
2102                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2103                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2104                 if (*mp)
2105                         mp = &(*mp)->m_next;
2106         }
2107         /* ip_srcroute doesn't do what we want here, need to fix */
2108         if (inp->inp_flags & INP_RECVRETOPTS) {
2109                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2110                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2111                 if (*mp)
2112                         mp = &(*mp)->m_next;
2113         }
2114 #endif
2115         if (inp->inp_flags & INP_RECVIF) {
2116                 struct ifnet *ifp;
2117                 struct sdlbuf {
2118                         struct sockaddr_dl sdl;
2119                         u_char  pad[32];
2120                 } sdlbuf;
2121                 struct sockaddr_dl *sdp;
2122                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2123
2124                 if (((ifp = m->m_pkthdr.rcvif)) &&
2125                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2126                         sdp = IF_LLSOCKADDR(ifp);
2127                         /*
2128                          * Change our mind and don't try copy.
2129                          */
2130                         if ((sdp->sdl_family != AF_LINK) ||
2131                             (sdp->sdl_len > sizeof(sdlbuf))) {
2132                                 goto makedummy;
2133                         }
2134                         bcopy(sdp, sdl2, sdp->sdl_len);
2135                 } else {
2136 makedummy:
2137                         sdl2->sdl_len =
2138                             offsetof(struct sockaddr_dl, sdl_data[0]);
2139                         sdl2->sdl_family = AF_LINK;
2140                         sdl2->sdl_index = 0;
2141                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2142                 }
2143                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2144                         IP_RECVIF, IPPROTO_IP);
2145                 if (*mp)
2146                         mp = &(*mp)->m_next;
2147         }
2148 }
2149
2150 /*
2151  * XXX these routines are called from the upper part of the kernel.
2152  *
2153  * They could also be moved to ip_mroute.c, since all the RSVP
2154  *  handling is done there already.
2155  */
2156 int
2157 ip_rsvp_init(struct socket *so)
2158 {
2159         if (so->so_type != SOCK_RAW ||
2160             so->so_proto->pr_protocol != IPPROTO_RSVP)
2161                 return EOPNOTSUPP;
2162
2163         if (ip_rsvpd != NULL)
2164                 return EADDRINUSE;
2165
2166         ip_rsvpd = so;
2167         /*
2168          * This may seem silly, but we need to be sure we don't over-increment
2169          * the RSVP counter, in case something slips up.
2170          */
2171         if (!ip_rsvp_on) {
2172                 ip_rsvp_on = 1;
2173                 rsvp_on++;
2174         }
2175
2176         return 0;
2177 }
2178
2179 int
2180 ip_rsvp_done(void)
2181 {
2182         ip_rsvpd = NULL;
2183         /*
2184          * This may seem silly, but we need to be sure we don't over-decrement
2185          * the RSVP counter, in case something slips up.
2186          */
2187         if (ip_rsvp_on) {
2188                 ip_rsvp_on = 0;
2189                 rsvp_on--;
2190         }
2191         return 0;
2192 }
2193
2194 void
2195 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2196 {
2197         int off, proto;
2198         __va_list ap;
2199
2200         __va_start(ap, m);
2201         off = __va_arg(ap, int);
2202         proto = __va_arg(ap, int);
2203         __va_end(ap);
2204
2205         if (rsvp_input_p) { /* call the real one if loaded */
2206                 rsvp_input_p(m, off, proto);
2207                 return;
2208         }
2209
2210         /* Can still get packets with rsvp_on = 0 if there is a local member
2211          * of the group to which the RSVP packet is addressed.  But in this
2212          * case we want to throw the packet away.
2213          */
2214
2215         if (!rsvp_on) {
2216                 m_freem(m);
2217                 return;
2218         }
2219
2220         if (ip_rsvpd != NULL) {
2221                 rip_input(m, off, proto);
2222                 return;
2223         }
2224         /* Drop the packet */
2225         m_freem(m);
2226 }