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 = 1;
 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_mport_pktinfo,
 396                         ip_input_handler, flags);
 397 }
 398
 399 /*
 400  * XXX watch out this one. It is perhaps used as a cache for
 401  * the most recently used route ? it is cleared in in_addroute()
 402  * when a new route is successfully created.
 403  */
 404 struct route ipforward_rt[MAXCPU];
 405
 406 /* Do transport protocol processing. */
 407 static void
 408 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 409 {
 410         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 411
 412         /*
 413          * Switch out to protocol's input routine.
 414          */
 415         PR_GET_MPLOCK(pr);
 416         pr->pr_input(m, hlen, ip->ip_p);
 417         PR_REL_MPLOCK(pr);
 418 }
 419
 420 static void
 421 transport_processing_handler(netmsg_t netmsg)
 422 {
 423         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 424         struct ip *ip;
 425         int hlen;
 426
 427         ip = mtod(pmsg->nm_packet, struct ip *);
 428         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 429
 430         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 431         /* netmsg was embedded in the mbuf, do not reply! */
 432 }
 433
 434 static void
 435 ip_input_handler(struct netmsg *msg0)
 436 {
 437         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 438
 439         ip_input(m);
 440         /* msg0 was embedded in the mbuf, do not reply! */
 441 }
 442
 443 /*
 444  * IP input routine.  Checksum and byte swap header.  If fragmented
 445  * try to reassemble.  Process options.  Pass to next level.
 446  */
 447 void
 448 ip_input(struct mbuf *m)
 449 {
 450         struct ip *ip;
 451         struct in_ifaddr *ia = NULL;
 452         struct in_ifaddr_container *iac;
 453         int hlen, checkif;
 454         u_short sum;
 455         struct in_addr pkt_dst;
 456         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 457         boolean_t needredispatch = FALSE;
 458         struct in_addr odst;                    /* original dst address(NAT) */
 459         struct m_tag *mtag;
 460         struct sockaddr_in *next_hop = NULL;
 461 #ifdef FAST_IPSEC
 462         struct tdb_ident *tdbi;
 463         struct secpolicy *sp;
 464         int error;
 465 #endif
 466
 467         M_ASSERTPKTHDR(m);
 468
 469         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 470                 /* Next hop */
 471                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 472                 KKASSERT(mtag != NULL);
 473                 next_hop = m_tag_data(mtag);
 474         }
 475
 476         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 477                 /* dummynet already filtered us */
 478                 ip = mtod(m, struct ip *);
 479                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 480                 goto iphack;
 481         }
 482
 483         ipstat.ips_total++;
 484
 485         /* length checks already done in ip_mport() */
 486         KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
 487         ip = mtod(m, struct ip *);
 488
 489         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 490                 ipstat.ips_badvers++;
 491                 goto bad;
 492         }
 493
 494         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 495         /* length checks already done in ip_mport() */
 496         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 497         KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
 498
 499         /* 127/8 must not appear on wire - RFC1122 */
 500         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 501             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 502                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 503                         ipstat.ips_badaddr++;
 504                         goto bad;
 505                 }
 506         }
 507
 508         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 509                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 510         } else {
 511                 if (hlen == sizeof(struct ip))
 512                         sum = in_cksum_hdr(ip);
 513                 else
 514                         sum = in_cksum(m, hlen);
 515         }
 516         if (sum != 0) {
 517                 ipstat.ips_badsum++;
 518                 goto bad;
 519         }
 520
 521 #ifdef ALTQ
 522         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 523                 /* packet is dropped by traffic conditioner */
 524                 return;
 525         }
 526 #endif
 527         /*
 528          * Convert fields to host representation.
 529          */
 530         ip->ip_len = ntohs(ip->ip_len);
 531         ip->ip_off = ntohs(ip->ip_off);
 532
 533         /* length checks already done in ip_mport() */
 534         KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
 535         KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
 536
 537         /*
 538          * Trim mbufs if longer than the IP header would have us expect.
 539          */
 540         if (m->m_pkthdr.len > ip->ip_len) {
 541                 if (m->m_len == m->m_pkthdr.len) {
 542                         m->m_len = ip->ip_len;
 543                         m->m_pkthdr.len = ip->ip_len;
 544                 } else {
 545                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 546                 }
 547         }
 548 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 549         /*
 550          * Bypass packet filtering for packets from a tunnel (gif).
 551          */
 552         if (ipsec_gethist(m, NULL))
 553                 goto pass;
 554 #endif
 555
 556         /*
 557          * IpHack's section.
 558          * Right now when no processing on packet has done
 559          * and it is still fresh out of network we do our black
 560          * deals with it.
 561          * - Firewall: deny/allow/divert
 562          * - Xlate: translate packet's addr/port (NAT).
 563          * - Pipe: pass pkt through dummynet.
 564          * - Wrap: fake packet's addr/port <unimpl.>
 565          * - Encapsulate: put it in another IP and send out. <unimp.>
 566          */
 567
 568 iphack:
 569         /*
 570          * If we've been forwarded from the output side, then
 571          * skip the firewall a second time
 572          */
 573         if (next_hop != NULL)
 574                 goto ours;
 575
 576         /* No pfil hooks */
 577         if (!pfil_has_hooks(&inet_pfil_hook)) {
 578                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 579                         /*
 580                          * Strip dummynet tags from stranded packets
 581                          */
 582                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 583                         KKASSERT(mtag != NULL);
 584                         m_tag_delete(m, mtag);
 585                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 586                 }
 587                 goto pass;
 588         }
 589
 590         /*
 591          * Run through list of hooks for input packets.
 592          *
 593          * NB: Beware of the destination address changing (e.g.
 594          *     by NAT rewriting). When this happens, tell
 595          *     ip_forward to do the right thing.
 596          */
 597         odst = ip->ip_dst;
 598         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 599                 return;
 600         if (m == NULL)  /* consumed by filter */
 601                 return;
 602         ip = mtod(m, struct ip *);
 603         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 604         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 605
 606         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 607                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 608                 KKASSERT(mtag != NULL);
 609                 next_hop = m_tag_data(mtag);
 610         }
 611         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 612                 ip_dn_queue(m);
 613                 return;
 614         }
 615         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 616                 needredispatch = TRUE;
 617                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 618         }
 619 pass:
 620         /*
 621          * Process options and, if not destined for us,
 622          * ship it on.  ip_dooptions returns 1 when an
 623          * error was detected (causing an icmp message
 624          * to be sent and the original packet to be freed).
 625          */
 626         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 627                 return;
 628
 629         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 630          * matter if it is destined to another node, or whether it is
 631          * a multicast one, RSVP wants it! and prevents it from being forwarded
 632          * anywhere else. Also checks if the rsvp daemon is running before
 633          * grabbing the packet.
 634          */
 635         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 636                 goto ours;
 637
 638         /*
 639          * Check our list of addresses, to see if the packet is for us.
 640          * If we don't have any addresses, assume any unicast packet
 641          * we receive might be for us (and let the upper layers deal
 642          * with it).
 643          */
 644         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 645             !(m->m_flags & (M_MCAST | M_BCAST)))
 646                 goto ours;
 647
 648         /*
 649          * Cache the destination address of the packet; this may be
 650          * changed by use of 'ipfw fwd'.
 651          */
 652         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 653
 654         /*
 655          * Enable a consistency check between the destination address
 656          * and the arrival interface for a unicast packet (the RFC 1122
 657          * strong ES model) if IP forwarding is disabled and the packet
 658          * is not locally generated and the packet is not subject to
 659          * 'ipfw fwd'.
 660          *
 661          * XXX - Checking also should be disabled if the destination
 662          * address is ipnat'ed to a different interface.
 663          *
 664          * XXX - Checking is incompatible with IP aliases added
 665          * to the loopback interface instead of the interface where
 666          * the packets are received.
 667          */
 668         checkif = ip_checkinterface &&
 669                   !ipforwarding &&
 670                   m->m_pkthdr.rcvif != NULL &&
 671                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 672                   next_hop == NULL;
 673
 674         /*
 675          * Check for exact addresses in the hash bucket.
 676          */
 677         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 678                 ia = iac->ia;
 679
 680                 /*
 681                  * If the address matches, verify that the packet
 682                  * arrived via the correct interface if checking is
 683                  * enabled.
 684                  */
 685                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 686                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 687                         goto ours;
 688         }
 689         ia = NULL;
 690
 691         /*
 692          * Check for broadcast addresses.
 693          *
 694          * Only accept broadcast packets that arrive via the matching
 695          * interface.  Reception of forwarded directed broadcasts would
 696          * be handled via ip_forward() and ether_output() with the loopback
 697          * into the stack for SIMPLEX interfaces handled by ether_output().
 698          */
 699         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 700                 struct ifaddr_container *ifac;
 701
 702                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 703                               ifa_link) {
 704                         struct ifaddr *ifa = ifac->ifa;
 705
 706                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 707                                 continue;
 708                         if (ifa->ifa_addr->sa_family != AF_INET)
 709                                 continue;
 710                         ia = ifatoia(ifa);
 711                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 712                                                                 pkt_dst.s_addr)
 713                                 goto ours;
 714                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 715                                 goto ours;
 716 #ifdef BOOTP_COMPAT
 717                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 718                                 goto ours;
 719 #endif
 720                 }
 721         }
 722         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 723                 struct in_multi *inm;
 724
 725                 /* XXX Multicast is not MPSAFE yet */
 726                 get_mplock();
 727
 728                 if (ip_mrouter != NULL) {
 729                         /*
 730                          * If we are acting as a multicast router, all
 731                          * incoming multicast packets are passed to the
 732                          * kernel-level multicast forwarding function.
 733                          * The packet is returned (relatively) intact; if
 734                          * ip_mforward() returns a non-zero value, the packet
 735                          * must be discarded, else it may be accepted below.
 736                          */
 737                         if (ip_mforward != NULL &&
 738                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 739                                 rel_mplock();
 740                                 ipstat.ips_cantforward++;
 741                                 m_freem(m);
 742                                 return;
 743                         }
 744
 745                         /*
 746                          * The process-level routing daemon needs to receive
 747                          * all multicast IGMP packets, whether or not this
 748                          * host belongs to their destination groups.
 749                          */
 750                         if (ip->ip_p == IPPROTO_IGMP) {
 751                                 rel_mplock();
 752                                 goto ours;
 753                         }
 754                         ipstat.ips_forward++;
 755                 }
 756                 /*
 757                  * See if we belong to the destination multicast group on the
 758                  * arrival interface.
 759                  */
 760                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 761                 if (inm == NULL) {
 762                         rel_mplock();
 763                         ipstat.ips_notmember++;
 764                         m_freem(m);
 765                         return;
 766                 }
 767
 768                 rel_mplock();
 769                 goto ours;
 770         }
 771         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 772                 goto ours;
 773         if (ip->ip_dst.s_addr == INADDR_ANY)
 774                 goto ours;
 775
 776         /*
 777          * FAITH(Firewall Aided Internet Translator)
 778          */
 779         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 780                 if (ip_keepfaith) {
 781                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 782                                 goto ours;
 783                 }
 784                 m_freem(m);
 785                 return;
 786         }
 787
 788         /*
 789          * Not for us; forward if possible and desirable.
 790          */
 791         if (!ipforwarding) {
 792                 ipstat.ips_cantforward++;
 793                 m_freem(m);
 794         } else {
 795 #ifdef IPSEC
 796                 /*
 797                  * Enforce inbound IPsec SPD.
 798                  */
 799                 if (ipsec4_in_reject(m, NULL)) {
 800                         ipsecstat.in_polvio++;
 801                         goto bad;
 802                 }
 803 #endif
 804 #ifdef FAST_IPSEC
 805                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 806                 crit_enter();
 807                 if (mtag != NULL) {
 808                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 809                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 810                 } else {
 811                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 812                                                    IP_FORWARDING, &error);
 813                 }
 814                 if (sp == NULL) {       /* NB: can happen if error */
 815                         crit_exit();
 816                         /*XXX error stat???*/
 817                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 818                         goto bad;
 819                 }
 820
 821                 /*
 822                  * Check security policy against packet attributes.
 823                  */
 824                 error = ipsec_in_reject(sp, m);
 825                 KEY_FREESP(&sp);
 826                 crit_exit();
 827                 if (error) {
 828                         ipstat.ips_cantforward++;
 829                         goto bad;
 830                 }
 831 #endif
 832                 ip_forward(m, using_srcrt, next_hop);
 833         }
 834         return;
 835
 836 ours:
 837
 838         /*
 839          * IPSTEALTH: Process non-routing options only
 840          * if the packet is destined for us.
 841          */
 842         if (ipstealth &&
 843             hlen > sizeof(struct ip) &&
 844             ip_dooptions(m, 1, next_hop))
 845                 return;
 846
 847         /* Count the packet in the ip address stats */
 848         if (ia != NULL) {
 849                 ia->ia_ifa.if_ipackets++;
 850                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 851         }
 852
 853         /*
 854          * If offset or IP_MF are set, must reassemble.
 855          * Otherwise, nothing need be done.
 856          * (We could look in the reassembly queue to see
 857          * if the packet was previously fragmented,
 858          * but it's not worth the time; just let them time out.)
 859          */
 860         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 861                 /*
 862                  * Attempt reassembly; if it succeeds, proceed.
 863                  * ip_reass() will return a different mbuf.
 864                  */
 865                 m = ip_reass(m);
 866                 if (m == NULL)
 867                         return;
 868                 ip = mtod(m, struct ip *);
 869
 870                 /* Get the header length of the reassembled packet */
 871                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 872
 873                 needredispatch = TRUE;
 874         } else {
 875                 ip->ip_len -= hlen;
 876         }
 877
 878 #ifdef IPSEC
 879         /*
 880          * enforce IPsec policy checking if we are seeing last header.
 881          * note that we do not visit this with protocols with pcb layer
 882          * code - like udp/tcp/raw ip.
 883          */
 884         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 885             ipsec4_in_reject(m, NULL)) {
 886                 ipsecstat.in_polvio++;
 887                 goto bad;
 888         }
 889 #endif
 890 #if FAST_IPSEC
 891         /*
 892          * enforce IPsec policy checking if we are seeing last header.
 893          * note that we do not visit this with protocols with pcb layer
 894          * code - like udp/tcp/raw ip.
 895          */
 896         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 897                 /*
 898                  * Check if the packet has already had IPsec processing
 899                  * done.  If so, then just pass it along.  This tag gets
 900                  * set during AH, ESP, etc. input handling, before the
 901                  * packet is returned to the ip input queue for delivery.
 902                  */
 903                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 904                 crit_enter();
 905                 if (mtag != NULL) {
 906                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 907                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 908                 } else {
 909                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 910                                                    IP_FORWARDING, &error);
 911                 }
 912                 if (sp != NULL) {
 913                         /*
 914                          * Check security policy against packet attributes.
 915                          */
 916                         error = ipsec_in_reject(sp, m);
 917                         KEY_FREESP(&sp);
 918                 } else {
 919                         /* XXX error stat??? */
 920                         error = EINVAL;
 921 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 922                         goto bad;
 923                 }
 924                 crit_exit();
 925                 if (error)
 926                         goto bad;
 927         }
 928 #endif /* FAST_IPSEC */
 929
 930         ipstat.ips_delivered++;
 931         if (needredispatch) {
 932                 struct netmsg_packet *pmsg;
 933                 lwkt_port_t port;
 934
 935                 ip->ip_off = htons(ip->ip_off);
 936                 ip->ip_len = htons(ip->ip_len);
 937                 port = ip_mport_in(&m);
 938                 if (port == NULL)
 939                         return;
 940
 941                 pmsg = &m->m_hdr.mh_netmsg;
 942                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, MSGF_MPSAFE,
 943                             transport_processing_handler);
 944                 pmsg->nm_packet = m;
 945                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
 946
 947                 ip = mtod(m, struct ip *);
 948                 ip->ip_len = ntohs(ip->ip_len);
 949                 ip->ip_off = ntohs(ip->ip_off);
 950                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
 951         } else {
 952                 transport_processing_oncpu(m, hlen, ip);
 953         }
 954         return;
 955
 956 bad:
 957         m_freem(m);
 958 }
 959
 960 /*
 961  * Take incoming datagram fragment and try to reassemble it into
 962  * whole datagram.  If a chain for reassembly of this datagram already
 963  * exists, then it is given as fp; otherwise have to make a chain.
 964  */
 965 struct mbuf *
 966 ip_reass(struct mbuf *m)
 967 {
 968         struct ip *ip = mtod(m, struct ip *);
 969         struct mbuf *p = NULL, *q, *nq;
 970         struct mbuf *n;
 971         struct ipq *fp = NULL;
 972         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 973         int i, next;
 974         u_short sum;
 975
 976         /* If maxnipq is 0, never accept fragments. */
 977         if (maxnipq == 0) {
 978                 ipstat.ips_fragments++;
 979                 ipstat.ips_fragdropped++;
 980                 m_freem(m);
 981                 return NULL;
 982         }
 983
 984         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 985         /*
 986          * Look for queue of fragments of this datagram.
 987          */
 988         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 989                 if (ip->ip_id == fp->ipq_id &&
 990                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 991                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 992                     ip->ip_p == fp->ipq_p)
 993                         goto found;
 994
 995         fp = NULL;
 996
 997         /*
 998          * Enforce upper bound on number of fragmented packets
 999          * for which we attempt reassembly;
1000          * If maxnipq is -1, accept all fragments without limitation.
1001          */
1002         if (nipq > maxnipq && maxnipq > 0) {
1003                 /*
1004                  * drop something from the tail of the current queue
1005                  * before proceeding further
1006                  */
1007                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1008                         for (i = 0; i < IPREASS_NHASH; i++) {
1009                                 if (ipq[i].prev != &ipq[i]) {
1010                                         ipstat.ips_fragtimeout +=
1011                                             ipq[i].prev->ipq_nfrags;
1012                                         ip_freef(ipq[i].prev);
1013                                         break;
1014                                 }
1015                         }
1016                 } else {
1017                         ipstat.ips_fragtimeout +=
1018                             ipq[sum].prev->ipq_nfrags;
1019                         ip_freef(ipq[sum].prev);
1020                 }
1021         }
1022 found:
1023         /*
1024          * Adjust ip_len to not reflect header,
1025          * convert offset of this to bytes.
1026          */
1027         ip->ip_len -= hlen;
1028         if (ip->ip_off & IP_MF) {
1029                 /*
1030                  * Make sure that fragments have a data length
1031                  * that's a non-zero multiple of 8 bytes.
1032                  */
1033                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1034                         ipstat.ips_toosmall++; /* XXX */
1035                         m_freem(m);
1036                         return NULL;
1037                 }
1038                 m->m_flags |= M_FRAG;
1039         } else
1040                 m->m_flags &= ~M_FRAG;
1041         ip->ip_off <<= 3;
1042
1043         ipstat.ips_fragments++;
1044         m->m_pkthdr.header = ip;
1045
1046         /*
1047          * If the hardware has not done csum over this fragment
1048          * then csum_data is not valid at all.
1049          */
1050         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1051             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1052                 m->m_pkthdr.csum_data = 0;
1053                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1054         }
1055
1056         /*
1057          * Presence of header sizes in mbufs
1058          * would confuse code below.
1059          */
1060         m->m_data += hlen;
1061         m->m_len -= hlen;
1062
1063         /*
1064          * If first fragment to arrive, create a reassembly queue.
1065          */
1066         if (fp == NULL) {
1067                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1068                         goto dropfrag;
1069                 insque(fp, &ipq[sum]);
1070                 nipq++;
1071                 fp->ipq_nfrags = 1;
1072                 fp->ipq_ttl = IPFRAGTTL;
1073                 fp->ipq_p = ip->ip_p;
1074                 fp->ipq_id = ip->ip_id;
1075                 fp->ipq_src = ip->ip_src;
1076                 fp->ipq_dst = ip->ip_dst;
1077                 fp->ipq_frags = m;
1078                 m->m_nextpkt = NULL;
1079                 goto inserted;
1080         } else {
1081                 fp->ipq_nfrags++;
1082         }
1083
1084 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1085
1086         /*
1087          * Find a segment which begins after this one does.
1088          */
1089         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1090                 if (GETIP(q)->ip_off > ip->ip_off)
1091                         break;
1092
1093         /*
1094          * If there is a preceding segment, it may provide some of
1095          * our data already.  If so, drop the data from the incoming
1096          * segment.  If it provides all of our data, drop us, otherwise
1097          * stick new segment in the proper place.
1098          *
1099          * If some of the data is dropped from the the preceding
1100          * segment, then it's checksum is invalidated.
1101          */
1102         if (p) {
1103                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1104                 if (i > 0) {
1105                         if (i >= ip->ip_len)
1106                                 goto dropfrag;
1107                         m_adj(m, i);
1108                         m->m_pkthdr.csum_flags = 0;
1109                         ip->ip_off += i;
1110                         ip->ip_len -= i;
1111                 }
1112                 m->m_nextpkt = p->m_nextpkt;
1113                 p->m_nextpkt = m;
1114         } else {
1115                 m->m_nextpkt = fp->ipq_frags;
1116                 fp->ipq_frags = m;
1117         }
1118
1119         /*
1120          * While we overlap succeeding segments trim them or,
1121          * if they are completely covered, dequeue them.
1122          */
1123         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1124              q = nq) {
1125                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1126                 if (i < GETIP(q)->ip_len) {
1127                         GETIP(q)->ip_len -= i;
1128                         GETIP(q)->ip_off += i;
1129                         m_adj(q, i);
1130                         q->m_pkthdr.csum_flags = 0;
1131                         break;
1132                 }
1133                 nq = q->m_nextpkt;
1134                 m->m_nextpkt = nq;
1135                 ipstat.ips_fragdropped++;
1136                 fp->ipq_nfrags--;
1137                 q->m_nextpkt = NULL;
1138                 m_freem(q);
1139         }
1140
1141 inserted:
1142         /*
1143          * Check for complete reassembly and perform frag per packet
1144          * limiting.
1145          *
1146          * Frag limiting is performed here so that the nth frag has
1147          * a chance to complete the packet before we drop the packet.
1148          * As a result, n+1 frags are actually allowed per packet, but
1149          * only n will ever be stored. (n = maxfragsperpacket.)
1150          *
1151          */
1152         next = 0;
1153         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1154                 if (GETIP(q)->ip_off != next) {
1155                         if (fp->ipq_nfrags > maxfragsperpacket) {
1156                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1157                                 ip_freef(fp);
1158                         }
1159                         return (NULL);
1160                 }
1161                 next += GETIP(q)->ip_len;
1162         }
1163         /* Make sure the last packet didn't have the IP_MF flag */
1164         if (p->m_flags & M_FRAG) {
1165                 if (fp->ipq_nfrags > maxfragsperpacket) {
1166                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1167                         ip_freef(fp);
1168                 }
1169                 return (NULL);
1170         }
1171
1172         /*
1173          * Reassembly is complete.  Make sure the packet is a sane size.
1174          */
1175         q = fp->ipq_frags;
1176         ip = GETIP(q);
1177         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1178                 ipstat.ips_toolong++;
1179                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1180                 ip_freef(fp);
1181                 return (NULL);
1182         }
1183
1184         /*
1185          * Concatenate fragments.
1186          */
1187         m = q;
1188         n = m->m_next;
1189         m->m_next = NULL;
1190         m_cat(m, n);
1191         nq = q->m_nextpkt;
1192         q->m_nextpkt = NULL;
1193         for (q = nq; q != NULL; q = nq) {
1194                 nq = q->m_nextpkt;
1195                 q->m_nextpkt = NULL;
1196                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1197                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1198                 m_cat(m, q);
1199         }
1200
1201         /*
1202          * Clean up the 1's complement checksum.  Carry over 16 bits must
1203          * be added back.  This assumes no more then 65535 packet fragments
1204          * were reassembled.  A second carry can also occur (but not a third).
1205          */
1206         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1207                                 (m->m_pkthdr.csum_data >> 16);
1208         if (m->m_pkthdr.csum_data > 0xFFFF)
1209                 m->m_pkthdr.csum_data -= 0xFFFF;
1210
1211         /*
1212          * Create header for new ip packet by
1213          * modifying header of first packet;
1214          * dequeue and discard fragment reassembly header.
1215          * Make header visible.
1216          */
1217         ip->ip_len = next;
1218         ip->ip_src = fp->ipq_src;
1219         ip->ip_dst = fp->ipq_dst;
1220         remque(fp);
1221         nipq--;
1222         mpipe_free(&ipq_mpipe, fp);
1223         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1224         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1225         /* some debugging cruft by sklower, below, will go away soon */
1226         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1227                 int plen = 0;
1228
1229                 for (n = m; n; n = n->m_next)
1230                         plen += n->m_len;
1231                 m->m_pkthdr.len = plen;
1232         }
1233
1234         ipstat.ips_reassembled++;
1235         return (m);
1236
1237 dropfrag:
1238         ipstat.ips_fragdropped++;
1239         if (fp != NULL)
1240                 fp->ipq_nfrags--;
1241         m_freem(m);
1242         return (NULL);
1243
1244 #undef GETIP
1245 }
1246
1247 /*
1248  * Free a fragment reassembly header and all
1249  * associated datagrams.
1250  */
1251 static void
1252 ip_freef(struct ipq *fp)
1253 {
1254         struct mbuf *q;
1255
1256         while (fp->ipq_frags) {
1257                 q = fp->ipq_frags;
1258                 fp->ipq_frags = q->m_nextpkt;
1259                 q->m_nextpkt = NULL;
1260                 m_freem(q);
1261         }
1262         remque(fp);
1263         mpipe_free(&ipq_mpipe, fp);
1264         nipq--;
1265 }
1266
1267 /*
1268  * IP timer processing;
1269  * if a timer expires on a reassembly
1270  * queue, discard it.
1271  */
1272 void
1273 ip_slowtimo(void)
1274 {
1275         struct ipq *fp;
1276         int i;
1277
1278         crit_enter();
1279         for (i = 0; i < IPREASS_NHASH; i++) {
1280                 fp = ipq[i].next;
1281                 if (fp == NULL)
1282                         continue;
1283                 while (fp != &ipq[i]) {
1284                         --fp->ipq_ttl;
1285                         fp = fp->next;
1286                         if (fp->prev->ipq_ttl == 0) {
1287                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1288                                 ip_freef(fp->prev);
1289                         }
1290                 }
1291         }
1292         /*
1293          * If we are over the maximum number of fragments
1294          * (due to the limit being lowered), drain off
1295          * enough to get down to the new limit.
1296          */
1297         if (maxnipq >= 0 && nipq > maxnipq) {
1298                 for (i = 0; i < IPREASS_NHASH; i++) {
1299                         while (nipq > maxnipq &&
1300                                 (ipq[i].next != &ipq[i])) {
1301                                 ipstat.ips_fragdropped +=
1302                                     ipq[i].next->ipq_nfrags;
1303                                 ip_freef(ipq[i].next);
1304                         }
1305                 }
1306         }
1307         ipflow_slowtimo();
1308         crit_exit();
1309 }
1310
1311 /*
1312  * Drain off all datagram fragments.
1313  */
1314 void
1315 ip_drain(void)
1316 {
1317         int i;
1318
1319         for (i = 0; i < IPREASS_NHASH; i++) {
1320                 while (ipq[i].next != &ipq[i]) {
1321                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1322                         ip_freef(ipq[i].next);
1323                 }
1324         }
1325         in_rtqdrain();
1326 }
1327
1328 /*
1329  * Do option processing on a datagram,
1330  * possibly discarding it if bad options are encountered,
1331  * or forwarding it if source-routed.
1332  * The pass argument is used when operating in the IPSTEALTH
1333  * mode to tell what options to process:
1334  * [LS]SRR (pass 0) or the others (pass 1).
1335  * The reason for as many as two passes is that when doing IPSTEALTH,
1336  * non-routing options should be processed only if the packet is for us.
1337  * Returns 1 if packet has been forwarded/freed,
1338  * 0 if the packet should be processed further.
1339  */
1340 static int
1341 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1342 {
1343         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1344         struct ip *ip = mtod(m, struct ip *);
1345         u_char *cp;
1346         struct in_ifaddr *ia;
1347         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1348         boolean_t forward = FALSE;
1349         struct in_addr *sin, dst;
1350         n_time ntime;
1351
1352         dst = ip->ip_dst;
1353         cp = (u_char *)(ip + 1);
1354         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1355         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1356                 opt = cp[IPOPT_OPTVAL];
1357                 if (opt == IPOPT_EOL)
1358                         break;
1359                 if (opt == IPOPT_NOP)
1360                         optlen = 1;
1361                 else {
1362                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1363                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1364                                 goto bad;
1365                         }
1366                         optlen = cp[IPOPT_OLEN];
1367                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1368                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1369                                 goto bad;
1370                         }
1371                 }
1372                 switch (opt) {
1373
1374                 default:
1375                         break;
1376
1377                 /*
1378                  * Source routing with record.
1379                  * Find interface with current destination address.
1380                  * If none on this machine then drop if strictly routed,
1381                  * or do nothing if loosely routed.
1382                  * Record interface address and bring up next address
1383                  * component.  If strictly routed make sure next
1384                  * address is on directly accessible net.
1385                  */
1386                 case IPOPT_LSRR:
1387                 case IPOPT_SSRR:
1388                         if (ipstealth && pass > 0)
1389                                 break;
1390                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1391                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1392                                 goto bad;
1393                         }
1394                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1395                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1396                                 goto bad;
1397                         }
1398                         ipaddr.sin_addr = ip->ip_dst;
1399                         ia = (struct in_ifaddr *)
1400                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1401                         if (ia == NULL) {
1402                                 if (opt == IPOPT_SSRR) {
1403                                         type = ICMP_UNREACH;
1404                                         code = ICMP_UNREACH_SRCFAIL;
1405                                         goto bad;
1406                                 }
1407                                 if (!ip_dosourceroute)
1408                                         goto nosourcerouting;
1409                                 /*
1410                                  * Loose routing, and not at next destination
1411                                  * yet; nothing to do except forward.
1412                                  */
1413                                 break;
1414                         }
1415                         off--;                  /* 0 origin */
1416                         if (off > optlen - (int)sizeof(struct in_addr)) {
1417                                 /*
1418                                  * End of source route.  Should be for us.
1419                                  */
1420                                 if (!ip_acceptsourceroute)
1421                                         goto nosourcerouting;
1422                                 save_rte(m, cp, ip->ip_src);
1423                                 break;
1424                         }
1425                         if (ipstealth)
1426                                 goto dropit;
1427                         if (!ip_dosourceroute) {
1428                                 if (ipforwarding) {
1429                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1430
1431                                         /*
1432                                          * Acting as a router, so generate ICMP
1433                                          */
1434 nosourcerouting:
1435                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1436                                         log(LOG_WARNING,
1437                                             "attempted source route from %s to %s\n",
1438                                             inet_ntoa(ip->ip_src), buf);
1439                                         type = ICMP_UNREACH;
1440                                         code = ICMP_UNREACH_SRCFAIL;
1441                                         goto bad;
1442                                 } else {
1443                                         /*
1444                                          * Not acting as a router,
1445                                          * so silently drop.
1446                                          */
1447 dropit:
1448                                         ipstat.ips_cantforward++;
1449                                         m_freem(m);
1450                                         return (1);
1451                                 }
1452                         }
1453
1454                         /*
1455                          * locate outgoing interface
1456                          */
1457                         memcpy(&ipaddr.sin_addr, cp + off,
1458                             sizeof ipaddr.sin_addr);
1459
1460                         if (opt == IPOPT_SSRR) {
1461 #define INA     struct in_ifaddr *
1462 #define SA      struct sockaddr *
1463                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1464                                                                         == NULL)
1465                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1466                         } else
1467                                 ia = ip_rtaddr(ipaddr.sin_addr,
1468                                                &ipforward_rt[mycpuid]);
1469                         if (ia == NULL) {
1470                                 type = ICMP_UNREACH;
1471                                 code = ICMP_UNREACH_SRCFAIL;
1472                                 goto bad;
1473                         }
1474                         ip->ip_dst = ipaddr.sin_addr;
1475                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1476                             sizeof(struct in_addr));
1477                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1478                         /*
1479                          * Let ip_intr's mcast routing check handle mcast pkts
1480                          */
1481                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1482                         break;
1483
1484                 case IPOPT_RR:
1485                         if (ipstealth && pass == 0)
1486                                 break;
1487                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1488                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1489                                 goto bad;
1490                         }
1491                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1492                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1493                                 goto bad;
1494                         }
1495                         /*
1496                          * If no space remains, ignore.
1497                          */
1498                         off--;                  /* 0 origin */
1499                         if (off > optlen - (int)sizeof(struct in_addr))
1500                                 break;
1501                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1502                             sizeof ipaddr.sin_addr);
1503                         /*
1504                          * locate outgoing interface; if we're the destination,
1505                          * use the incoming interface (should be same).
1506                          */
1507                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1508                             (ia = ip_rtaddr(ipaddr.sin_addr,
1509                                             &ipforward_rt[mycpuid]))
1510                                                                      == NULL) {
1511                                 type = ICMP_UNREACH;
1512                                 code = ICMP_UNREACH_HOST;
1513                                 goto bad;
1514                         }
1515                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1516                             sizeof(struct in_addr));
1517                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1518                         break;
1519
1520                 case IPOPT_TS:
1521                         if (ipstealth && pass == 0)
1522                                 break;
1523                         code = cp - (u_char *)ip;
1524                         if (optlen < 4 || optlen > 40) {
1525                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1526                                 goto bad;
1527                         }
1528                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1529                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1530                                 goto bad;
1531                         }
1532                         if (off > optlen - (int)sizeof(int32_t)) {
1533                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1534                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1535                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1536                                         goto bad;
1537                                 }
1538                                 break;
1539                         }
1540                         off--;                          /* 0 origin */
1541                         sin = (struct in_addr *)(cp + off);
1542                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1543
1544                         case IPOPT_TS_TSONLY:
1545                                 break;
1546
1547                         case IPOPT_TS_TSANDADDR:
1548                                 if (off + sizeof(n_time) +
1549                                     sizeof(struct in_addr) > optlen) {
1550                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1551                                         goto bad;
1552                                 }
1553                                 ipaddr.sin_addr = dst;
1554                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1555                                                             m->m_pkthdr.rcvif);
1556                                 if (ia == NULL)
1557                                         continue;
1558                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1559                                     sizeof(struct in_addr));
1560                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1561                                 off += sizeof(struct in_addr);
1562                                 break;
1563
1564                         case IPOPT_TS_PRESPEC:
1565                                 if (off + sizeof(n_time) +
1566                                     sizeof(struct in_addr) > optlen) {
1567                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1568                                         goto bad;
1569                                 }
1570                                 memcpy(&ipaddr.sin_addr, sin,
1571                                     sizeof(struct in_addr));
1572                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1573                                         continue;
1574                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1575                                 off += sizeof(struct in_addr);
1576                                 break;
1577
1578                         default:
1579                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1580                                 goto bad;
1581                         }
1582                         ntime = iptime();
1583                         memcpy(cp + off, &ntime, sizeof(n_time));
1584                         cp[IPOPT_OFFSET] += sizeof(n_time);
1585                 }
1586         }
1587         if (forward && ipforwarding) {
1588                 ip_forward(m, TRUE, next_hop);
1589                 return (1);
1590         }
1591         return (0);
1592 bad:
1593         icmp_error(m, type, code, 0, 0);
1594         ipstat.ips_badoptions++;
1595         return (1);
1596 }
1597
1598 /*
1599  * Given address of next destination (final or next hop),
1600  * return internet address info of interface to be used to get there.
1601  */
1602 struct in_ifaddr *
1603 ip_rtaddr(struct in_addr dst, struct route *ro)
1604 {
1605         struct sockaddr_in *sin;
1606
1607         sin = (struct sockaddr_in *)&ro->ro_dst;
1608
1609         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1610                 if (ro->ro_rt != NULL) {
1611                         RTFREE(ro->ro_rt);
1612                         ro->ro_rt = NULL;
1613                 }
1614                 sin->sin_family = AF_INET;
1615                 sin->sin_len = sizeof *sin;
1616                 sin->sin_addr = dst;
1617                 rtalloc_ign(ro, RTF_PRCLONING);
1618         }
1619
1620         if (ro->ro_rt == NULL)
1621                 return (NULL);
1622
1623         return (ifatoia(ro->ro_rt->rt_ifa));
1624 }
1625
1626 /*
1627  * Save incoming source route for use in replies,
1628  * to be picked up later by ip_srcroute if the receiver is interested.
1629  */
1630 static void
1631 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1632 {
1633         struct m_tag *mtag;
1634         struct ip_srcrt_opt *opt;
1635         unsigned olen;
1636
1637         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1638         if (mtag == NULL)
1639                 return;
1640         opt = m_tag_data(mtag);
1641
1642         olen = option[IPOPT_OLEN];
1643 #ifdef DIAGNOSTIC
1644         if (ipprintfs)
1645                 kprintf("save_rte: olen %d\n", olen);
1646 #endif
1647         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1648                 m_tag_free(mtag);
1649                 return;
1650         }
1651         bcopy(option, opt->ip_srcrt.srcopt, olen);
1652         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1653         opt->ip_srcrt.dst = dst;
1654         m_tag_prepend(m, mtag);
1655 }
1656
1657 /*
1658  * Retrieve incoming source route for use in replies,
1659  * in the same form used by setsockopt.
1660  * The first hop is placed before the options, will be removed later.
1661  */
1662 struct mbuf *
1663 ip_srcroute(struct mbuf *m0)
1664 {
1665         struct in_addr *p, *q;
1666         struct mbuf *m;
1667         struct m_tag *mtag;
1668         struct ip_srcrt_opt *opt;
1669
1670         if (m0 == NULL)
1671                 return NULL;
1672
1673         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1674         if (mtag == NULL)
1675                 return NULL;
1676         opt = m_tag_data(mtag);
1677
1678         if (opt->ip_nhops == 0)
1679                 return (NULL);
1680         m = m_get(MB_DONTWAIT, MT_HEADER);
1681         if (m == NULL)
1682                 return (NULL);
1683
1684 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1685
1686         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1687         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1688                    sizeof(struct in_addr) + OPTSIZ;
1689 #ifdef DIAGNOSTIC
1690         if (ipprintfs) {
1691                 kprintf("ip_srcroute: nhops %d mlen %d",
1692                         opt->ip_nhops, m->m_len);
1693         }
1694 #endif
1695
1696         /*
1697          * First save first hop for return route
1698          */
1699         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1700         *(mtod(m, struct in_addr *)) = *p--;
1701 #ifdef DIAGNOSTIC
1702         if (ipprintfs)
1703                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1704 #endif
1705
1706         /*
1707          * Copy option fields and padding (nop) to mbuf.
1708          */
1709         opt->ip_srcrt.nop = IPOPT_NOP;
1710         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1711         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1712             OPTSIZ);
1713         q = (struct in_addr *)(mtod(m, caddr_t) +
1714             sizeof(struct in_addr) + OPTSIZ);
1715 #undef OPTSIZ
1716         /*
1717          * Record return path as an IP source route,
1718          * reversing the path (pointers are now aligned).
1719          */
1720         while (p >= opt->ip_srcrt.route) {
1721 #ifdef DIAGNOSTIC
1722                 if (ipprintfs)
1723                         kprintf(" %x", ntohl(q->s_addr));
1724 #endif
1725                 *q++ = *p--;
1726         }
1727         /*
1728          * Last hop goes to final destination.
1729          */
1730         *q = opt->ip_srcrt.dst;
1731         m_tag_delete(m0, mtag);
1732 #ifdef DIAGNOSTIC
1733         if (ipprintfs)
1734                 kprintf(" %x\n", ntohl(q->s_addr));
1735 #endif
1736         return (m);
1737 }
1738
1739 /*
1740  * Strip out IP options.
1741  */
1742 void
1743 ip_stripoptions(struct mbuf *m)
1744 {
1745         int datalen;
1746         struct ip *ip = mtod(m, struct ip *);
1747         caddr_t opts;
1748         int optlen;
1749
1750         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1751         opts = (caddr_t)(ip + 1);
1752         datalen = m->m_len - (sizeof(struct ip) + optlen);
1753         bcopy(opts + optlen, opts, datalen);
1754         m->m_len -= optlen;
1755         if (m->m_flags & M_PKTHDR)
1756                 m->m_pkthdr.len -= optlen;
1757         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1758 }
1759
1760 u_char inetctlerrmap[PRC_NCMDS] = {
1761         0,              0,              0,              0,
1762         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1763         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1764         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1765         0,              0,              0,              0,
1766         ENOPROTOOPT,    ECONNREFUSED
1767 };
1768
1769 /*
1770  * Forward a packet.  If some error occurs return the sender
1771  * an icmp packet.  Note we can't always generate a meaningful
1772  * icmp message because icmp doesn't have a large enough repertoire
1773  * of codes and types.
1774  *
1775  * If not forwarding, just drop the packet.  This could be confusing
1776  * if ipforwarding was zero but some routing protocol was advancing
1777  * us as a gateway to somewhere.  However, we must let the routing
1778  * protocol deal with that.
1779  *
1780  * The using_srcrt parameter indicates whether the packet is being forwarded
1781  * via a source route.
1782  */
1783 void
1784 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1785 {
1786         struct ip *ip = mtod(m, struct ip *);
1787         struct sockaddr_in *ipforward_rtaddr;
1788         struct rtentry *rt;
1789         int error, type = 0, code = 0, destmtu = 0;
1790         struct mbuf *mcopy;
1791         n_long dest;
1792         struct in_addr pkt_dst;
1793         struct route *cache_rt = &ipforward_rt[mycpuid];
1794
1795         dest = INADDR_ANY;
1796         /*
1797          * Cache the destination address of the packet; this may be
1798          * changed by use of 'ipfw fwd'.
1799          */
1800         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1801
1802 #ifdef DIAGNOSTIC
1803         if (ipprintfs)
1804                 kprintf("forward: src %x dst %x ttl %x\n",
1805                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1806 #endif
1807
1808         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1809                 ipstat.ips_cantforward++;
1810                 m_freem(m);
1811                 return;
1812         }
1813         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1814                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1815                 return;
1816         }
1817
1818         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1819         if (cache_rt->ro_rt == NULL ||
1820             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1821                 if (cache_rt->ro_rt != NULL) {
1822                         RTFREE(cache_rt->ro_rt);
1823                         cache_rt->ro_rt = NULL;
1824                 }
1825                 ipforward_rtaddr->sin_family = AF_INET;
1826                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1827                 ipforward_rtaddr->sin_addr = pkt_dst;
1828                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1829                 if (cache_rt->ro_rt == NULL) {
1830                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1831                         return;
1832                 }
1833         }
1834         rt = cache_rt->ro_rt;
1835
1836         /*
1837          * Save the IP header and at most 8 bytes of the payload,
1838          * in case we need to generate an ICMP message to the src.
1839          *
1840          * XXX this can be optimized a lot by saving the data in a local
1841          * buffer on the stack (72 bytes at most), and only allocating the
1842          * mbuf if really necessary. The vast majority of the packets
1843          * are forwarded without having to send an ICMP back (either
1844          * because unnecessary, or because rate limited), so we are
1845          * really we are wasting a lot of work here.
1846          *
1847          * We don't use m_copy() because it might return a reference
1848          * to a shared cluster. Both this function and ip_output()
1849          * assume exclusive access to the IP header in `m', so any
1850          * data in a cluster may change before we reach icmp_error().
1851          */
1852         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1853         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1854                 /*
1855                  * It's probably ok if the pkthdr dup fails (because
1856                  * the deep copy of the tag chain failed), but for now
1857                  * be conservative and just discard the copy since
1858                  * code below may some day want the tags.
1859                  */
1860                 m_free(mcopy);
1861                 mcopy = NULL;
1862         }
1863         if (mcopy != NULL) {
1864                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1865                     (int)ip->ip_len);
1866                 mcopy->m_pkthdr.len = mcopy->m_len;
1867                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1868         }
1869
1870         if (!ipstealth)
1871                 ip->ip_ttl -= IPTTLDEC;
1872
1873         /*
1874          * If forwarding packet using same interface that it came in on,
1875          * perhaps should send a redirect to sender to shortcut a hop.
1876          * Only send redirect if source is sending directly to us,
1877          * and if packet was not source routed (or has any options).
1878          * Also, don't send redirect if forwarding using a default route
1879          * or a route modified by a redirect.
1880          */
1881         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1882             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1883             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1884             ipsendredirects && !using_srcrt && next_hop == NULL) {
1885                 u_long src = ntohl(ip->ip_src.s_addr);
1886                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1887
1888                 if (rt_ifa != NULL &&
1889                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1890                         if (rt->rt_flags & RTF_GATEWAY)
1891                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1892                         else
1893                                 dest = pkt_dst.s_addr;
1894                         /*
1895                          * Router requirements says to only send
1896                          * host redirects.
1897                          */
1898                         type = ICMP_REDIRECT;
1899                         code = ICMP_REDIRECT_HOST;
1900 #ifdef DIAGNOSTIC
1901                         if (ipprintfs)
1902                                 kprintf("redirect (%d) to %x\n", code, dest);
1903 #endif
1904                 }
1905         }
1906
1907         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1908         if (error == 0) {
1909                 ipstat.ips_forward++;
1910                 if (type == 0) {
1911                         if (mcopy) {
1912                                 ipflow_create(cache_rt, mcopy);
1913                                 m_freem(mcopy);
1914                         }
1915                         return;         /* most common case */
1916                 } else {
1917                         ipstat.ips_redirectsent++;
1918                 }
1919         } else {
1920                 ipstat.ips_cantforward++;
1921         }
1922
1923         if (mcopy == NULL)
1924                 return;
1925
1926         /*
1927          * Send ICMP message.
1928          */
1929
1930         switch (error) {
1931
1932         case 0:                         /* forwarded, but need redirect */
1933                 /* type, code set above */
1934                 break;
1935
1936         case ENETUNREACH:               /* shouldn't happen, checked above */
1937         case EHOSTUNREACH:
1938         case ENETDOWN:
1939         case EHOSTDOWN:
1940         default:
1941                 type = ICMP_UNREACH;
1942                 code = ICMP_UNREACH_HOST;
1943                 break;
1944
1945         case EMSGSIZE:
1946                 type = ICMP_UNREACH;
1947                 code = ICMP_UNREACH_NEEDFRAG;
1948 #ifdef IPSEC
1949                 /*
1950                  * If the packet is routed over IPsec tunnel, tell the
1951                  * originator the tunnel MTU.
1952                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1953                  * XXX quickhack!!!
1954                  */
1955                 if (cache_rt->ro_rt != NULL) {
1956                         struct secpolicy *sp = NULL;
1957                         int ipsecerror;
1958                         int ipsechdr;
1959                         struct route *ro;
1960
1961                         sp = ipsec4_getpolicybyaddr(mcopy,
1962                                                     IPSEC_DIR_OUTBOUND,
1963                                                     IP_FORWARDING,
1964                                                     &ipsecerror);
1965
1966                         if (sp == NULL)
1967                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1968                         else {
1969                                 /* count IPsec header size */
1970                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1971                                                          IPSEC_DIR_OUTBOUND,
1972                                                          NULL);
1973
1974                                 /*
1975                                  * find the correct route for outer IPv4
1976                                  * header, compute tunnel MTU.
1977                                  *
1978                                  */
1979                                 if (sp->req != NULL && sp->req->sav != NULL &&
1980                                     sp->req->sav->sah != NULL) {
1981                                         ro = &sp->req->sav->sah->sa_route;
1982                                         if (ro->ro_rt != NULL &&
1983                                             ro->ro_rt->rt_ifp != NULL) {
1984                                                 destmtu =
1985                                                     ro->ro_rt->rt_ifp->if_mtu;
1986                                                 destmtu -= ipsechdr;
1987                                         }
1988                                 }
1989
1990                                 key_freesp(sp);
1991                         }
1992                 }
1993 #elif FAST_IPSEC
1994                 /*
1995                  * If the packet is routed over IPsec tunnel, tell the
1996                  * originator the tunnel MTU.
1997                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1998                  * XXX quickhack!!!
1999                  */
2000                 if (cache_rt->ro_rt != NULL) {
2001                         struct secpolicy *sp = NULL;
2002                         int ipsecerror;
2003                         int ipsechdr;
2004                         struct route *ro;
2005
2006                         sp = ipsec_getpolicybyaddr(mcopy,
2007                                                    IPSEC_DIR_OUTBOUND,
2008                                                    IP_FORWARDING,
2009                                                    &ipsecerror);
2010
2011                         if (sp == NULL)
2012                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2013                         else {
2014                                 /* count IPsec header size */
2015                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2016                                                          IPSEC_DIR_OUTBOUND,
2017                                                          NULL);
2018
2019                                 /*
2020                                  * find the correct route for outer IPv4
2021                                  * header, compute tunnel MTU.
2022                                  */
2023
2024                                 if (sp->req != NULL &&
2025                                     sp->req->sav != NULL &&
2026                                     sp->req->sav->sah != NULL) {
2027                                         ro = &sp->req->sav->sah->sa_route;
2028                                         if (ro->ro_rt != NULL &&
2029                                             ro->ro_rt->rt_ifp != NULL) {
2030                                                 destmtu =
2031                                                     ro->ro_rt->rt_ifp->if_mtu;
2032                                                 destmtu -= ipsechdr;
2033                                         }
2034                                 }
2035
2036                                 KEY_FREESP(&sp);
2037                         }
2038                 }
2039 #else /* !IPSEC && !FAST_IPSEC */
2040                 if (cache_rt->ro_rt != NULL)
2041                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2042 #endif /*IPSEC*/
2043                 ipstat.ips_cantfrag++;
2044                 break;
2045
2046         case ENOBUFS:
2047                 /*
2048                  * A router should not generate ICMP_SOURCEQUENCH as
2049                  * required in RFC1812 Requirements for IP Version 4 Routers.
2050                  * Source quench could be a big problem under DoS attacks,
2051                  * or if the underlying interface is rate-limited.
2052                  * Those who need source quench packets may re-enable them
2053                  * via the net.inet.ip.sendsourcequench sysctl.
2054                  */
2055                 if (!ip_sendsourcequench) {
2056                         m_freem(mcopy);
2057                         return;
2058                 } else {
2059                         type = ICMP_SOURCEQUENCH;
2060                         code = 0;
2061                 }
2062                 break;
2063
2064         case EACCES:                    /* ipfw denied packet */
2065                 m_freem(mcopy);
2066                 return;
2067         }
2068         icmp_error(mcopy, type, code, dest, destmtu);
2069 }
2070
2071 void
2072 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2073                struct mbuf *m)
2074 {
2075         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2076                 struct timeval tv;
2077
2078                 microtime(&tv);
2079                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2080                     SCM_TIMESTAMP, SOL_SOCKET);
2081                 if (*mp)
2082                         mp = &(*mp)->m_next;
2083         }
2084         if (inp->inp_flags & INP_RECVDSTADDR) {
2085                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2086                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2087                 if (*mp)
2088                         mp = &(*mp)->m_next;
2089         }
2090         if (inp->inp_flags & INP_RECVTTL) {
2091                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2092                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2093                 if (*mp)
2094                         mp = &(*mp)->m_next;
2095         }
2096 #ifdef notyet
2097         /* XXX
2098          * Moving these out of udp_input() made them even more broken
2099          * than they already were.
2100          */
2101         /* options were tossed already */
2102         if (inp->inp_flags & INP_RECVOPTS) {
2103                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2104                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2105                 if (*mp)
2106                         mp = &(*mp)->m_next;
2107         }
2108         /* ip_srcroute doesn't do what we want here, need to fix */
2109         if (inp->inp_flags & INP_RECVRETOPTS) {
2110                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2111                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2112                 if (*mp)
2113                         mp = &(*mp)->m_next;
2114         }
2115 #endif
2116         if (inp->inp_flags & INP_RECVIF) {
2117                 struct ifnet *ifp;
2118                 struct sdlbuf {
2119                         struct sockaddr_dl sdl;
2120                         u_char  pad[32];
2121                 } sdlbuf;
2122                 struct sockaddr_dl *sdp;
2123                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2124
2125                 if (((ifp = m->m_pkthdr.rcvif)) &&
2126                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2127                         sdp = IF_LLSOCKADDR(ifp);
2128                         /*
2129                          * Change our mind and don't try copy.
2130                          */
2131                         if ((sdp->sdl_family != AF_LINK) ||
2132                             (sdp->sdl_len > sizeof(sdlbuf))) {
2133                                 goto makedummy;
2134                         }
2135                         bcopy(sdp, sdl2, sdp->sdl_len);
2136                 } else {
2137 makedummy:
2138                         sdl2->sdl_len =
2139                             offsetof(struct sockaddr_dl, sdl_data[0]);
2140                         sdl2->sdl_family = AF_LINK;
2141                         sdl2->sdl_index = 0;
2142                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2143                 }
2144                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2145                         IP_RECVIF, IPPROTO_IP);
2146                 if (*mp)
2147                         mp = &(*mp)->m_next;
2148         }
2149 }
2150
2151 /*
2152  * XXX these routines are called from the upper part of the kernel.
2153  *
2154  * They could also be moved to ip_mroute.c, since all the RSVP
2155  *  handling is done there already.
2156  */
2157 int
2158 ip_rsvp_init(struct socket *so)
2159 {
2160         if (so->so_type != SOCK_RAW ||
2161             so->so_proto->pr_protocol != IPPROTO_RSVP)
2162                 return EOPNOTSUPP;
2163
2164         if (ip_rsvpd != NULL)
2165                 return EADDRINUSE;
2166
2167         ip_rsvpd = so;
2168         /*
2169          * This may seem silly, but we need to be sure we don't over-increment
2170          * the RSVP counter, in case something slips up.
2171          */
2172         if (!ip_rsvp_on) {
2173                 ip_rsvp_on = 1;
2174                 rsvp_on++;
2175         }
2176
2177         return 0;
2178 }
2179
2180 int
2181 ip_rsvp_done(void)
2182 {
2183         ip_rsvpd = NULL;
2184         /*
2185          * This may seem silly, but we need to be sure we don't over-decrement
2186          * the RSVP counter, in case something slips up.
2187          */
2188         if (ip_rsvp_on) {
2189                 ip_rsvp_on = 0;
2190                 rsvp_on--;
2191         }
2192         return 0;
2193 }
2194
2195 void
2196 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2197 {
2198         int off, proto;
2199         __va_list ap;
2200
2201         __va_start(ap, m);
2202         off = __va_arg(ap, int);
2203         proto = __va_arg(ap, int);
2204         __va_end(ap);
2205
2206         if (rsvp_input_p) { /* call the real one if loaded */
2207                 rsvp_input_p(m, off, proto);
2208                 return;
2209         }
2210
2211         /* Can still get packets with rsvp_on = 0 if there is a local member
2212          * of the group to which the RSVP packet is addressed.  But in this
2213          * case we want to throw the packet away.
2214          */
2215
2216         if (!rsvp_on) {
2217                 m_freem(m);
2218                 return;
2219         }
2220
2221         if (ip_rsvpd != NULL) {
2222                 rip_input(m, off, proto);
2223                 return;
2224         }
2225         /* Drop the packet */
2226         m_freem(m);
2227 }