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 1.110 2008/09/24 14:26:39 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
 117 #include <sys/thread2.h>
 118 #include <sys/msgport2.h>
 119 #include <net/netmsg2.h>
 120
 121 #include <sys/socketvar.h>
 122
 123 #include <net/ipfw/ip_fw.h>
 124 #include <net/dummynet/ip_dummynet.h>
 125
 126 #ifdef IPSEC
 127 #include <netinet6/ipsec.h>
 128 #include <netproto/key/key.h>
 129 #endif
 130
 131 #ifdef FAST_IPSEC
 132 #include <netproto/ipsec/ipsec.h>
 133 #include <netproto/ipsec/key.h>
 134 #endif
 135
 136 int rsvp_on = 0;
 137 static int ip_rsvp_on;
 138 struct socket *ip_rsvpd;
 139
 140 int ipforwarding = 0;
 141 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 142     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 143
 144 static int ipsendredirects = 1; /* XXX */
 145 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 146     &ipsendredirects, 0, "Enable sending IP redirects");
 147
 148 int ip_defttl = IPDEFTTL;
 149 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 150     &ip_defttl, 0, "Maximum TTL on IP packets");
 151
 152 static int ip_dosourceroute = 0;
 153 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 154     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 155
 156 static int ip_acceptsourceroute = 0;
 157 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 158     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 159     "Enable accepting source routed IP packets");
 160
 161 static int ip_keepfaith = 0;
 162 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 163     &ip_keepfaith, 0,
 164     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 165
 166 static int nipq = 0;    /* total # of reass queues */
 167 static int maxnipq;
 168 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 169     &maxnipq, 0,
 170     "Maximum number of IPv4 fragment reassembly queue entries");
 171
 172 static int maxfragsperpacket;
 173 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 174     &maxfragsperpacket, 0,
 175     "Maximum number of IPv4 fragments allowed per packet");
 176
 177 static int ip_sendsourcequench = 0;
 178 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 179     &ip_sendsourcequench, 0,
 180     "Enable the transmission of source quench packets");
 181
 182 int ip_do_randomid = 1;
 183 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 184     &ip_do_randomid, 0,
 185     "Assign random ip_id values");
 186 /*
 187  * XXX - Setting ip_checkinterface mostly implements the receive side of
 188  * the Strong ES model described in RFC 1122, but since the routing table
 189  * and transmit implementation do not implement the Strong ES model,
 190  * setting this to 1 results in an odd hybrid.
 191  *
 192  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 193  * to translate the destination address to another local interface.
 194  *
 195  * XXX - ip_checkinterface must be disabled if you add IP aliases
 196  * to the loopback interface instead of the interface where the
 197  * packets for those addresses are received.
 198  */
 199 static int ip_checkinterface = 0;
 200 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 201     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 202
 203 #ifdef DIAGNOSTIC
 204 static int ipprintfs = 0;
 205 #endif
 206
 207 extern  int udp_mpsafe_proto;
 208 extern  int tcp_mpsafe_proto;
 209
 210 extern  struct domain inetdomain;
 211 extern  struct protosw inetsw[];
 212 u_char  ip_protox[IPPROTO_MAX];
 213 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 214 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 215                                                 /* inet addr hash table */
 216 u_long  in_ifaddrhmask;                         /* mask for hash table */
 217
 218 struct ip_stats ipstats_percpu[MAXCPU];
 219 #ifdef SMP
 220 static int
 221 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 222 {
 223         int cpu, error = 0;
 224
 225         for (cpu = 0; cpu < ncpus; ++cpu) {
 226                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 227                                         sizeof(struct ip_stats))))
 228                         break;
 229                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 230                                        sizeof(struct ip_stats))))
 231                         break;
 232         }
 233
 234         return (error);
 235 }
 236 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 237     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 238 #else
 239 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 240     &ipstat, ip_stats, "IP statistics");
 241 #endif
 242
 243 /* Packet reassembly stuff */
 244 #define IPREASS_NHASH_LOG2      6
 245 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 246 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 247 #define IPREASS_HASH(x,y)                                               \
 248     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 249
 250 static struct ipq ipq[IPREASS_NHASH];
 251
 252 #ifdef IPCTL_DEFMTU
 253 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 254     &ip_mtu, 0, "Default MTU");
 255 #endif
 256
 257 #ifdef IPSTEALTH
 258 static int ipstealth = 0;
 259 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 260 #else
 261 static const int ipstealth = 0;
 262 #endif
 263
 264 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
 265
 266 struct pfil_head inet_pfil_hook;
 267
 268 /*
 269  * struct ip_srcrt_opt is used to store packet state while it travels
 270  * through the stack.
 271  *
 272  * XXX Note that the code even makes assumptions on the size and
 273  * alignment of fields inside struct ip_srcrt so e.g. adding some
 274  * fields will break the code.  This needs to be fixed.
 275  *
 276  * We need to save the IP options in case a protocol wants to respond
 277  * to an incoming packet over the same route if the packet got here
 278  * using IP source routing.  This allows connection establishment and
 279  * maintenance when the remote end is on a network that is not known
 280  * to us.
 281  */
 282 struct ip_srcrt {
 283         struct  in_addr dst;                    /* final destination */
 284         char    nop;                            /* one NOP to align */
 285         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 286         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 287 };
 288
 289 struct ip_srcrt_opt {
 290         int             ip_nhops;
 291         struct ip_srcrt ip_srcrt;
 292 };
 293
 294 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 295 static struct malloc_pipe ipq_mpipe;
 296
 297 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
 298 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 299 static void             ip_freef(struct ipq *);
 300 static void             ip_input_handler(struct netmsg *);
 301
 302 /*
 303  * IP initialization: fill in IP protocol switch table.
 304  * All protocols not implemented in kernel go to raw IP protocol handler.
 305  */
 306 void
 307 ip_init(void)
 308 {
 309         struct protosw *pr;
 310         int i;
 311 #ifdef SMP
 312         int cpu;
 313 #endif
 314
 315         /*
 316          * Make sure we can handle a reasonable number of fragments but
 317          * cap it at 4000 (XXX).
 318          */
 319         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 320                     IFQ_MAXLEN, 4000, 0, NULL);
 321         for (i = 0; i < ncpus; ++i) {
 322                 TAILQ_INIT(&in_ifaddrheads[i]);
 323                 in_ifaddrhashtbls[i] =
 324                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 325         }
 326         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 327         if (pr == NULL)
 328                 panic("ip_init");
 329         for (i = 0; i < IPPROTO_MAX; i++)
 330                 ip_protox[i] = pr - inetsw;
 331         for (pr = inetdomain.dom_protosw;
 332              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
 333                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
 334                         if (pr->pr_protocol != IPPROTO_RAW)
 335                                 ip_protox[pr->pr_protocol] = pr - inetsw;
 336
 337                         /* XXX */
 338                         switch (pr->pr_protocol) {
 339                         case IPPROTO_TCP:
 340                                 if (tcp_mpsafe_proto)
 341                                         pr->pr_flags |= PR_MPSAFE;
 342                                 break;
 343
 344                         case IPPROTO_UDP:
 345                                 if (udp_mpsafe_proto)
 346                                         pr->pr_flags |= PR_MPSAFE;
 347                                 break;
 348                         }
 349                 }
 350         }
 351
 352         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 353         inet_pfil_hook.ph_af = AF_INET;
 354         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 355                 kprintf("%s: WARNING: unable to register pfil hook, "
 356                         "error %d\n", __func__, i);
 357         }
 358
 359         for (i = 0; i < IPREASS_NHASH; i++)
 360                 ipq[i].next = ipq[i].prev = &ipq[i];
 361
 362         maxnipq = nmbclusters / 32;
 363         maxfragsperpacket = 16;
 364
 365         ip_id = time_second & 0xffff;
 366
 367         /*
 368          * Initialize IP statistics counters for each CPU.
 369          *
 370          */
 371 #ifdef SMP
 372         for (cpu = 0; cpu < ncpus; ++cpu) {
 373                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 374         }
 375 #else
 376         bzero(&ipstat, sizeof(struct ip_stats));
 377 #endif
 378
 379         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler,
 380                         NETISR_FLAG_NOTMPSAFE);
 381 }
 382
 383 /*
 384  * XXX watch out this one. It is perhaps used as a cache for
 385  * the most recently used route ? it is cleared in in_addroute()
 386  * when a new route is successfully created.
 387  */
 388 struct route ipforward_rt[MAXCPU];
 389
 390 /* Do transport protocol processing. */
 391 static void
 392 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 393 {
 394         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 395
 396         /*
 397          * Switch out to protocol's input routine.
 398          */
 399         PR_GET_MPLOCK(pr);
 400         pr->pr_input(m, hlen, ip->ip_p);
 401         PR_REL_MPLOCK(pr);
 402 }
 403
 404 static void
 405 transport_processing_handler(netmsg_t netmsg)
 406 {
 407         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 408         struct ip *ip;
 409         int hlen;
 410
 411         ip = mtod(pmsg->nm_packet, struct ip *);
 412         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 413
 414         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 415         /* netmsg was embedded in the mbuf, do not reply! */
 416 }
 417
 418 static void
 419 ip_input_handler(struct netmsg *msg0)
 420 {
 421         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 422
 423         ip_input(m);
 424         /* msg0 was embedded in the mbuf, do not reply! */
 425 }
 426
 427 /*
 428  * IP input routine.  Checksum and byte swap header.  If fragmented
 429  * try to reassemble.  Process options.  Pass to next level.
 430  */
 431 void
 432 ip_input(struct mbuf *m)
 433 {
 434         struct ip *ip;
 435         struct in_ifaddr *ia = NULL;
 436         struct in_ifaddr_container *iac;
 437         int hlen, checkif;
 438         u_short sum;
 439         struct in_addr pkt_dst;
 440         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 441         boolean_t needredispatch = FALSE;
 442         struct in_addr odst;                    /* original dst address(NAT) */
 443         struct m_tag *mtag;
 444         struct sockaddr_in *next_hop = NULL;
 445 #ifdef FAST_IPSEC
 446         struct tdb_ident *tdbi;
 447         struct secpolicy *sp;
 448         int error;
 449 #endif
 450
 451         M_ASSERTPKTHDR(m);
 452
 453         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 454                 /* Next hop */
 455                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 456                 KKASSERT(mtag != NULL);
 457                 next_hop = m_tag_data(mtag);
 458         }
 459
 460         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 461                 /* dummynet already filtered us */
 462                 ip = mtod(m, struct ip *);
 463                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 464                 goto iphack;
 465         }
 466
 467         ipstat.ips_total++;
 468
 469         /* length checks already done in ip_demux() */
 470         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 471
 472         ip = mtod(m, struct ip *);
 473
 474         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 475                 ipstat.ips_badvers++;
 476                 goto bad;
 477         }
 478
 479         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 480         /* length checks already done in ip_demux() */
 481         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 482         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 483
 484         /* 127/8 must not appear on wire - RFC1122 */
 485         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 486             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 487                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 488                         ipstat.ips_badaddr++;
 489                         goto bad;
 490                 }
 491         }
 492
 493         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 494                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 495         } else {
 496                 if (hlen == sizeof(struct ip)) {
 497                         sum = in_cksum_hdr(ip);
 498                 } else {
 499                         sum = in_cksum(m, hlen);
 500                 }
 501         }
 502         if (sum != 0) {
 503                 ipstat.ips_badsum++;
 504                 goto bad;
 505         }
 506
 507 #ifdef ALTQ
 508         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 509                 /* packet is dropped by traffic conditioner */
 510                 return;
 511         }
 512 #endif
 513         /*
 514          * Convert fields to host representation.
 515          */
 516         ip->ip_len = ntohs(ip->ip_len);
 517         if (ip->ip_len < hlen) {
 518                 ipstat.ips_badlen++;
 519                 goto bad;
 520         }
 521         ip->ip_off = ntohs(ip->ip_off);
 522
 523         /*
 524          * Check that the amount of data in the buffers
 525          * is as at least much as the IP header would have us expect.
 526          * Trim mbufs if longer than we expect.
 527          * Drop packet if shorter than we expect.
 528          */
 529         if (m->m_pkthdr.len < ip->ip_len) {
 530                 ipstat.ips_tooshort++;
 531                 goto bad;
 532         }
 533         if (m->m_pkthdr.len > ip->ip_len) {
 534                 if (m->m_len == m->m_pkthdr.len) {
 535                         m->m_len = ip->ip_len;
 536                         m->m_pkthdr.len = ip->ip_len;
 537                 } else
 538                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 539         }
 540 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 541         /*
 542          * Bypass packet filtering for packets from a tunnel (gif).
 543          */
 544         if (ipsec_gethist(m, NULL))
 545                 goto pass;
 546 #endif
 547
 548         /*
 549          * IpHack's section.
 550          * Right now when no processing on packet has done
 551          * and it is still fresh out of network we do our black
 552          * deals with it.
 553          * - Firewall: deny/allow/divert
 554          * - Xlate: translate packet's addr/port (NAT).
 555          * - Pipe: pass pkt through dummynet.
 556          * - Wrap: fake packet's addr/port <unimpl.>
 557          * - Encapsulate: put it in another IP and send out. <unimp.>
 558          */
 559
 560 iphack:
 561         /*
 562          * If we've been forwarded from the output side, then
 563          * skip the firewall a second time
 564          */
 565         if (next_hop != NULL)
 566                 goto ours;
 567
 568         /* No pfil hooks */
 569         if (!pfil_has_hooks(&inet_pfil_hook)) {
 570                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 571                         /*
 572                          * Strip dummynet tags from stranded packets
 573                          */
 574                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 575                         KKASSERT(mtag != NULL);
 576                         m_tag_delete(m, mtag);
 577                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 578                 }
 579                 goto pass;
 580         }
 581
 582         /*
 583          * Run through list of hooks for input packets.
 584          *
 585          * NB: Beware of the destination address changing (e.g.
 586          *     by NAT rewriting). When this happens, tell
 587          *     ip_forward to do the right thing.
 588          */
 589         odst = ip->ip_dst;
 590         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 591                 return;
 592         if (m == NULL)  /* consumed by filter */
 593                 return;
 594         ip = mtod(m, struct ip *);
 595         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 596         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 597
 598         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 599                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 600                 KKASSERT(mtag != NULL);
 601                 next_hop = m_tag_data(mtag);
 602         }
 603         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 604                 ip_dn_queue(m);
 605                 return;
 606         }
 607         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 608                 needredispatch = TRUE;
 609                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 610         }
 611 pass:
 612         /*
 613          * Process options and, if not destined for us,
 614          * ship it on.  ip_dooptions returns 1 when an
 615          * error was detected (causing an icmp message
 616          * to be sent and the original packet to be freed).
 617          */
 618         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 619                 return;
 620
 621         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 622          * matter if it is destined to another node, or whether it is
 623          * a multicast one, RSVP wants it! and prevents it from being forwarded
 624          * anywhere else. Also checks if the rsvp daemon is running before
 625          * grabbing the packet.
 626          */
 627         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 628                 goto ours;
 629
 630         /*
 631          * Check our list of addresses, to see if the packet is for us.
 632          * If we don't have any addresses, assume any unicast packet
 633          * we receive might be for us (and let the upper layers deal
 634          * with it).
 635          */
 636         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 637             !(m->m_flags & (M_MCAST | M_BCAST)))
 638                 goto ours;
 639
 640         /*
 641          * Cache the destination address of the packet; this may be
 642          * changed by use of 'ipfw fwd'.
 643          */
 644         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 645
 646         /*
 647          * Enable a consistency check between the destination address
 648          * and the arrival interface for a unicast packet (the RFC 1122
 649          * strong ES model) if IP forwarding is disabled and the packet
 650          * is not locally generated and the packet is not subject to
 651          * 'ipfw fwd'.
 652          *
 653          * XXX - Checking also should be disabled if the destination
 654          * address is ipnat'ed to a different interface.
 655          *
 656          * XXX - Checking is incompatible with IP aliases added
 657          * to the loopback interface instead of the interface where
 658          * the packets are received.
 659          */
 660         checkif = ip_checkinterface &&
 661                   !ipforwarding &&
 662                   m->m_pkthdr.rcvif != NULL &&
 663                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 664                   next_hop == NULL;
 665
 666         /*
 667          * Check for exact addresses in the hash bucket.
 668          */
 669         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 670                 ia = iac->ia;
 671
 672                 /*
 673                  * If the address matches, verify that the packet
 674                  * arrived via the correct interface if checking is
 675                  * enabled.
 676                  */
 677                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 678                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 679                         goto ours;
 680         }
 681         ia = NULL;
 682
 683         /*
 684          * Check for broadcast addresses.
 685          *
 686          * Only accept broadcast packets that arrive via the matching
 687          * interface.  Reception of forwarded directed broadcasts would
 688          * be handled via ip_forward() and ether_output() with the loopback
 689          * into the stack for SIMPLEX interfaces handled by ether_output().
 690          */
 691         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 692                 struct ifaddr_container *ifac;
 693
 694                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 695                               ifa_link) {
 696                         struct ifaddr *ifa = ifac->ifa;
 697
 698                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 699                                 continue;
 700                         if (ifa->ifa_addr->sa_family != AF_INET)
 701                                 continue;
 702                         ia = ifatoia(ifa);
 703                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 704                                                                 pkt_dst.s_addr)
 705                                 goto ours;
 706                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 707                                 goto ours;
 708 #ifdef BOOTP_COMPAT
 709                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 710                                 goto ours;
 711 #endif
 712                 }
 713         }
 714         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 715                 struct in_multi *inm;
 716
 717                 if (ip_mrouter != NULL) {
 718                         /*
 719                          * If we are acting as a multicast router, all
 720                          * incoming multicast packets are passed to the
 721                          * kernel-level multicast forwarding function.
 722                          * The packet is returned (relatively) intact; if
 723                          * ip_mforward() returns a non-zero value, the packet
 724                          * must be discarded, else it may be accepted below.
 725                          */
 726                         if (ip_mforward != NULL &&
 727                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 728                                 ipstat.ips_cantforward++;
 729                                 m_freem(m);
 730                                 return;
 731                         }
 732
 733                         /*
 734                          * The process-level routing daemon needs to receive
 735                          * all multicast IGMP packets, whether or not this
 736                          * host belongs to their destination groups.
 737                          */
 738                         if (ip->ip_p == IPPROTO_IGMP)
 739                                 goto ours;
 740                         ipstat.ips_forward++;
 741                 }
 742                 /*
 743                  * See if we belong to the destination multicast group on the
 744                  * arrival interface.
 745                  */
 746                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 747                 if (inm == NULL) {
 748                         ipstat.ips_notmember++;
 749                         m_freem(m);
 750                         return;
 751                 }
 752                 goto ours;
 753         }
 754         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 755                 goto ours;
 756         if (ip->ip_dst.s_addr == INADDR_ANY)
 757                 goto ours;
 758
 759         /*
 760          * FAITH(Firewall Aided Internet Translator)
 761          */
 762         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 763                 if (ip_keepfaith) {
 764                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 765                                 goto ours;
 766                 }
 767                 m_freem(m);
 768                 return;
 769         }
 770
 771         /*
 772          * Not for us; forward if possible and desirable.
 773          */
 774         if (!ipforwarding) {
 775                 ipstat.ips_cantforward++;
 776                 m_freem(m);
 777         } else {
 778 #ifdef IPSEC
 779                 /*
 780                  * Enforce inbound IPsec SPD.
 781                  */
 782                 if (ipsec4_in_reject(m, NULL)) {
 783                         ipsecstat.in_polvio++;
 784                         goto bad;
 785                 }
 786 #endif
 787 #ifdef FAST_IPSEC
 788                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 789                 crit_enter();
 790                 if (mtag != NULL) {
 791                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 792                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 793                 } else {
 794                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 795                                                    IP_FORWARDING, &error);
 796                 }
 797                 if (sp == NULL) {       /* NB: can happen if error */
 798                         crit_exit();
 799                         /*XXX error stat???*/
 800                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 801                         goto bad;
 802                 }
 803
 804                 /*
 805                  * Check security policy against packet attributes.
 806                  */
 807                 error = ipsec_in_reject(sp, m);
 808                 KEY_FREESP(&sp);
 809                 crit_exit();
 810                 if (error) {
 811                         ipstat.ips_cantforward++;
 812                         goto bad;
 813                 }
 814 #endif
 815                 ip_forward(m, using_srcrt, next_hop);
 816         }
 817         return;
 818
 819 ours:
 820
 821         /*
 822          * IPSTEALTH: Process non-routing options only
 823          * if the packet is destined for us.
 824          */
 825         if (ipstealth &&
 826             hlen > sizeof(struct ip) &&
 827             ip_dooptions(m, 1, next_hop))
 828                 return;
 829
 830         /* Count the packet in the ip address stats */
 831         if (ia != NULL) {
 832                 ia->ia_ifa.if_ipackets++;
 833                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 834         }
 835
 836         /*
 837          * If offset or IP_MF are set, must reassemble.
 838          * Otherwise, nothing need be done.
 839          * (We could look in the reassembly queue to see
 840          * if the packet was previously fragmented,
 841          * but it's not worth the time; just let them time out.)
 842          */
 843         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 844                 /*
 845                  * Attempt reassembly; if it succeeds, proceed.
 846                  * ip_reass() will return a different mbuf.
 847                  */
 848                 m = ip_reass(m);
 849                 if (m == NULL)
 850                         return;
 851                 ip = mtod(m, struct ip *);
 852
 853                 /* Get the header length of the reassembled packet */
 854                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 855
 856                 needredispatch = TRUE;
 857         } else {
 858                 ip->ip_len -= hlen;
 859         }
 860
 861 #ifdef IPSEC
 862         /*
 863          * enforce IPsec policy checking if we are seeing last header.
 864          * note that we do not visit this with protocols with pcb layer
 865          * code - like udp/tcp/raw ip.
 866          */
 867         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 868             ipsec4_in_reject(m, NULL)) {
 869                 ipsecstat.in_polvio++;
 870                 goto bad;
 871         }
 872 #endif
 873 #if FAST_IPSEC
 874         /*
 875          * enforce IPsec policy checking if we are seeing last header.
 876          * note that we do not visit this with protocols with pcb layer
 877          * code - like udp/tcp/raw ip.
 878          */
 879         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 880                 /*
 881                  * Check if the packet has already had IPsec processing
 882                  * done.  If so, then just pass it along.  This tag gets
 883                  * set during AH, ESP, etc. input handling, before the
 884                  * packet is returned to the ip input queue for delivery.
 885                  */
 886                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 887                 crit_enter();
 888                 if (mtag != NULL) {
 889                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 890                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 891                 } else {
 892                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 893                                                    IP_FORWARDING, &error);
 894                 }
 895                 if (sp != NULL) {
 896                         /*
 897                          * Check security policy against packet attributes.
 898                          */
 899                         error = ipsec_in_reject(sp, m);
 900                         KEY_FREESP(&sp);
 901                 } else {
 902                         /* XXX error stat??? */
 903                         error = EINVAL;
 904 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 905                         goto bad;
 906                 }
 907                 crit_exit();
 908                 if (error)
 909                         goto bad;
 910         }
 911 #endif /* FAST_IPSEC */
 912
 913         ipstat.ips_delivered++;
 914         if (needredispatch) {
 915                 struct netmsg_packet *pmsg;
 916                 lwkt_port_t port;
 917
 918                 ip->ip_off = htons(ip->ip_off);
 919                 ip->ip_len = htons(ip->ip_len);
 920                 port = ip_mport_in(&m);
 921                 if (port == NULL)
 922                         return;
 923
 924                 pmsg = &m->m_hdr.mh_netmsg;
 925                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, MSGF_MPSAFE,
 926                             transport_processing_handler);
 927                 pmsg->nm_packet = m;
 928                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
 929
 930                 ip = mtod(m, struct ip *);
 931                 ip->ip_len = ntohs(ip->ip_len);
 932                 ip->ip_off = ntohs(ip->ip_off);
 933                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
 934         } else {
 935                 transport_processing_oncpu(m, hlen, ip);
 936         }
 937         return;
 938
 939 bad:
 940         m_freem(m);
 941 }
 942
 943 /*
 944  * Take incoming datagram fragment and try to reassemble it into
 945  * whole datagram.  If a chain for reassembly of this datagram already
 946  * exists, then it is given as fp; otherwise have to make a chain.
 947  */
 948 struct mbuf *
 949 ip_reass(struct mbuf *m)
 950 {
 951         struct ip *ip = mtod(m, struct ip *);
 952         struct mbuf *p = NULL, *q, *nq;
 953         struct mbuf *n;
 954         struct ipq *fp = NULL;
 955         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 956         int i, next;
 957         u_short sum;
 958
 959         /* If maxnipq is 0, never accept fragments. */
 960         if (maxnipq == 0) {
 961                 ipstat.ips_fragments++;
 962                 ipstat.ips_fragdropped++;
 963                 m_freem(m);
 964                 return NULL;
 965         }
 966
 967         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 968         /*
 969          * Look for queue of fragments of this datagram.
 970          */
 971         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 972                 if (ip->ip_id == fp->ipq_id &&
 973                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 974                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 975                     ip->ip_p == fp->ipq_p)
 976                         goto found;
 977
 978         fp = NULL;
 979
 980         /*
 981          * Enforce upper bound on number of fragmented packets
 982          * for which we attempt reassembly;
 983          * If maxnipq is -1, accept all fragments without limitation.
 984          */
 985         if (nipq > maxnipq && maxnipq > 0) {
 986                 /*
 987                  * drop something from the tail of the current queue
 988                  * before proceeding further
 989                  */
 990                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
 991                         for (i = 0; i < IPREASS_NHASH; i++) {
 992                                 if (ipq[i].prev != &ipq[i]) {
 993                                         ipstat.ips_fragtimeout +=
 994                                             ipq[i].prev->ipq_nfrags;
 995                                         ip_freef(ipq[i].prev);
 996                                         break;
 997                                 }
 998                         }
 999                 } else {
1000                         ipstat.ips_fragtimeout +=
1001                             ipq[sum].prev->ipq_nfrags;
1002                         ip_freef(ipq[sum].prev);
1003                 }
1004         }
1005 found:
1006         /*
1007          * Adjust ip_len to not reflect header,
1008          * convert offset of this to bytes.
1009          */
1010         ip->ip_len -= hlen;
1011         if (ip->ip_off & IP_MF) {
1012                 /*
1013                  * Make sure that fragments have a data length
1014                  * that's a non-zero multiple of 8 bytes.
1015                  */
1016                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1017                         ipstat.ips_toosmall++; /* XXX */
1018                         m_freem(m);
1019                         return NULL;
1020                 }
1021                 m->m_flags |= M_FRAG;
1022         } else
1023                 m->m_flags &= ~M_FRAG;
1024         ip->ip_off <<= 3;
1025
1026         ipstat.ips_fragments++;
1027         m->m_pkthdr.header = ip;
1028
1029         /*
1030          * If the hardware has not done csum over this fragment
1031          * then csum_data is not valid at all.
1032          */
1033         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1034             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1035                 m->m_pkthdr.csum_data = 0;
1036                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1037         }
1038
1039         /*
1040          * Presence of header sizes in mbufs
1041          * would confuse code below.
1042          */
1043         m->m_data += hlen;
1044         m->m_len -= hlen;
1045
1046         /*
1047          * If first fragment to arrive, create a reassembly queue.
1048          */
1049         if (fp == NULL) {
1050                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1051                         goto dropfrag;
1052                 insque(fp, &ipq[sum]);
1053                 nipq++;
1054                 fp->ipq_nfrags = 1;
1055                 fp->ipq_ttl = IPFRAGTTL;
1056                 fp->ipq_p = ip->ip_p;
1057                 fp->ipq_id = ip->ip_id;
1058                 fp->ipq_src = ip->ip_src;
1059                 fp->ipq_dst = ip->ip_dst;
1060                 fp->ipq_frags = m;
1061                 m->m_nextpkt = NULL;
1062                 goto inserted;
1063         } else {
1064                 fp->ipq_nfrags++;
1065         }
1066
1067 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1068
1069         /*
1070          * Find a segment which begins after this one does.
1071          */
1072         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1073                 if (GETIP(q)->ip_off > ip->ip_off)
1074                         break;
1075
1076         /*
1077          * If there is a preceding segment, it may provide some of
1078          * our data already.  If so, drop the data from the incoming
1079          * segment.  If it provides all of our data, drop us, otherwise
1080          * stick new segment in the proper place.
1081          *
1082          * If some of the data is dropped from the the preceding
1083          * segment, then it's checksum is invalidated.
1084          */
1085         if (p) {
1086                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1087                 if (i > 0) {
1088                         if (i >= ip->ip_len)
1089                                 goto dropfrag;
1090                         m_adj(m, i);
1091                         m->m_pkthdr.csum_flags = 0;
1092                         ip->ip_off += i;
1093                         ip->ip_len -= i;
1094                 }
1095                 m->m_nextpkt = p->m_nextpkt;
1096                 p->m_nextpkt = m;
1097         } else {
1098                 m->m_nextpkt = fp->ipq_frags;
1099                 fp->ipq_frags = m;
1100         }
1101
1102         /*
1103          * While we overlap succeeding segments trim them or,
1104          * if they are completely covered, dequeue them.
1105          */
1106         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1107              q = nq) {
1108                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1109                 if (i < GETIP(q)->ip_len) {
1110                         GETIP(q)->ip_len -= i;
1111                         GETIP(q)->ip_off += i;
1112                         m_adj(q, i);
1113                         q->m_pkthdr.csum_flags = 0;
1114                         break;
1115                 }
1116                 nq = q->m_nextpkt;
1117                 m->m_nextpkt = nq;
1118                 ipstat.ips_fragdropped++;
1119                 fp->ipq_nfrags--;
1120                 q->m_nextpkt = NULL;
1121                 m_freem(q);
1122         }
1123
1124 inserted:
1125         /*
1126          * Check for complete reassembly and perform frag per packet
1127          * limiting.
1128          *
1129          * Frag limiting is performed here so that the nth frag has
1130          * a chance to complete the packet before we drop the packet.
1131          * As a result, n+1 frags are actually allowed per packet, but
1132          * only n will ever be stored. (n = maxfragsperpacket.)
1133          *
1134          */
1135         next = 0;
1136         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1137                 if (GETIP(q)->ip_off != next) {
1138                         if (fp->ipq_nfrags > maxfragsperpacket) {
1139                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1140                                 ip_freef(fp);
1141                         }
1142                         return (NULL);
1143                 }
1144                 next += GETIP(q)->ip_len;
1145         }
1146         /* Make sure the last packet didn't have the IP_MF flag */
1147         if (p->m_flags & M_FRAG) {
1148                 if (fp->ipq_nfrags > maxfragsperpacket) {
1149                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1150                         ip_freef(fp);
1151                 }
1152                 return (NULL);
1153         }
1154
1155         /*
1156          * Reassembly is complete.  Make sure the packet is a sane size.
1157          */
1158         q = fp->ipq_frags;
1159         ip = GETIP(q);
1160         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1161                 ipstat.ips_toolong++;
1162                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1163                 ip_freef(fp);
1164                 return (NULL);
1165         }
1166
1167         /*
1168          * Concatenate fragments.
1169          */
1170         m = q;
1171         n = m->m_next;
1172         m->m_next = NULL;
1173         m_cat(m, n);
1174         nq = q->m_nextpkt;
1175         q->m_nextpkt = NULL;
1176         for (q = nq; q != NULL; q = nq) {
1177                 nq = q->m_nextpkt;
1178                 q->m_nextpkt = NULL;
1179                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1180                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1181                 m_cat(m, q);
1182         }
1183
1184         /*
1185          * Clean up the 1's complement checksum.  Carry over 16 bits must
1186          * be added back.  This assumes no more then 65535 packet fragments
1187          * were reassembled.  A second carry can also occur (but not a third).
1188          */
1189         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1190                                 (m->m_pkthdr.csum_data >> 16);
1191         if (m->m_pkthdr.csum_data > 0xFFFF)
1192                 m->m_pkthdr.csum_data -= 0xFFFF;
1193
1194         /*
1195          * Create header for new ip packet by
1196          * modifying header of first packet;
1197          * dequeue and discard fragment reassembly header.
1198          * Make header visible.
1199          */
1200         ip->ip_len = next;
1201         ip->ip_src = fp->ipq_src;
1202         ip->ip_dst = fp->ipq_dst;
1203         remque(fp);
1204         nipq--;
1205         mpipe_free(&ipq_mpipe, fp);
1206         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1207         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1208         /* some debugging cruft by sklower, below, will go away soon */
1209         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1210                 int plen = 0;
1211
1212                 for (n = m; n; n = n->m_next)
1213                         plen += n->m_len;
1214                 m->m_pkthdr.len = plen;
1215         }
1216
1217         ipstat.ips_reassembled++;
1218         return (m);
1219
1220 dropfrag:
1221         ipstat.ips_fragdropped++;
1222         if (fp != NULL)
1223                 fp->ipq_nfrags--;
1224         m_freem(m);
1225         return (NULL);
1226
1227 #undef GETIP
1228 }
1229
1230 /*
1231  * Free a fragment reassembly header and all
1232  * associated datagrams.
1233  */
1234 static void
1235 ip_freef(struct ipq *fp)
1236 {
1237         struct mbuf *q;
1238
1239         while (fp->ipq_frags) {
1240                 q = fp->ipq_frags;
1241                 fp->ipq_frags = q->m_nextpkt;
1242                 q->m_nextpkt = NULL;
1243                 m_freem(q);
1244         }
1245         remque(fp);
1246         mpipe_free(&ipq_mpipe, fp);
1247         nipq--;
1248 }
1249
1250 /*
1251  * IP timer processing;
1252  * if a timer expires on a reassembly
1253  * queue, discard it.
1254  */
1255 void
1256 ip_slowtimo(void)
1257 {
1258         struct ipq *fp;
1259         int i;
1260
1261         crit_enter();
1262         for (i = 0; i < IPREASS_NHASH; i++) {
1263                 fp = ipq[i].next;
1264                 if (fp == NULL)
1265                         continue;
1266                 while (fp != &ipq[i]) {
1267                         --fp->ipq_ttl;
1268                         fp = fp->next;
1269                         if (fp->prev->ipq_ttl == 0) {
1270                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1271                                 ip_freef(fp->prev);
1272                         }
1273                 }
1274         }
1275         /*
1276          * If we are over the maximum number of fragments
1277          * (due to the limit being lowered), drain off
1278          * enough to get down to the new limit.
1279          */
1280         if (maxnipq >= 0 && nipq > maxnipq) {
1281                 for (i = 0; i < IPREASS_NHASH; i++) {
1282                         while (nipq > maxnipq &&
1283                                 (ipq[i].next != &ipq[i])) {
1284                                 ipstat.ips_fragdropped +=
1285                                     ipq[i].next->ipq_nfrags;
1286                                 ip_freef(ipq[i].next);
1287                         }
1288                 }
1289         }
1290         ipflow_slowtimo();
1291         crit_exit();
1292 }
1293
1294 /*
1295  * Drain off all datagram fragments.
1296  */
1297 void
1298 ip_drain(void)
1299 {
1300         int i;
1301
1302         for (i = 0; i < IPREASS_NHASH; i++) {
1303                 while (ipq[i].next != &ipq[i]) {
1304                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1305                         ip_freef(ipq[i].next);
1306                 }
1307         }
1308         in_rtqdrain();
1309 }
1310
1311 /*
1312  * Do option processing on a datagram,
1313  * possibly discarding it if bad options are encountered,
1314  * or forwarding it if source-routed.
1315  * The pass argument is used when operating in the IPSTEALTH
1316  * mode to tell what options to process:
1317  * [LS]SRR (pass 0) or the others (pass 1).
1318  * The reason for as many as two passes is that when doing IPSTEALTH,
1319  * non-routing options should be processed only if the packet is for us.
1320  * Returns 1 if packet has been forwarded/freed,
1321  * 0 if the packet should be processed further.
1322  */
1323 static int
1324 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1325 {
1326         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1327         struct ip *ip = mtod(m, struct ip *);
1328         u_char *cp;
1329         struct in_ifaddr *ia;
1330         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1331         boolean_t forward = FALSE;
1332         struct in_addr *sin, dst;
1333         n_time ntime;
1334
1335         dst = ip->ip_dst;
1336         cp = (u_char *)(ip + 1);
1337         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1338         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1339                 opt = cp[IPOPT_OPTVAL];
1340                 if (opt == IPOPT_EOL)
1341                         break;
1342                 if (opt == IPOPT_NOP)
1343                         optlen = 1;
1344                 else {
1345                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1346                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1347                                 goto bad;
1348                         }
1349                         optlen = cp[IPOPT_OLEN];
1350                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1351                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1352                                 goto bad;
1353                         }
1354                 }
1355                 switch (opt) {
1356
1357                 default:
1358                         break;
1359
1360                 /*
1361                  * Source routing with record.
1362                  * Find interface with current destination address.
1363                  * If none on this machine then drop if strictly routed,
1364                  * or do nothing if loosely routed.
1365                  * Record interface address and bring up next address
1366                  * component.  If strictly routed make sure next
1367                  * address is on directly accessible net.
1368                  */
1369                 case IPOPT_LSRR:
1370                 case IPOPT_SSRR:
1371                         if (ipstealth && pass > 0)
1372                                 break;
1373                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1374                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1375                                 goto bad;
1376                         }
1377                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1378                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1379                                 goto bad;
1380                         }
1381                         ipaddr.sin_addr = ip->ip_dst;
1382                         ia = (struct in_ifaddr *)
1383                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1384                         if (ia == NULL) {
1385                                 if (opt == IPOPT_SSRR) {
1386                                         type = ICMP_UNREACH;
1387                                         code = ICMP_UNREACH_SRCFAIL;
1388                                         goto bad;
1389                                 }
1390                                 if (!ip_dosourceroute)
1391                                         goto nosourcerouting;
1392                                 /*
1393                                  * Loose routing, and not at next destination
1394                                  * yet; nothing to do except forward.
1395                                  */
1396                                 break;
1397                         }
1398                         off--;                  /* 0 origin */
1399                         if (off > optlen - (int)sizeof(struct in_addr)) {
1400                                 /*
1401                                  * End of source route.  Should be for us.
1402                                  */
1403                                 if (!ip_acceptsourceroute)
1404                                         goto nosourcerouting;
1405                                 save_rte(m, cp, ip->ip_src);
1406                                 break;
1407                         }
1408                         if (ipstealth)
1409                                 goto dropit;
1410                         if (!ip_dosourceroute) {
1411                                 if (ipforwarding) {
1412                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1413
1414                                         /*
1415                                          * Acting as a router, so generate ICMP
1416                                          */
1417 nosourcerouting:
1418                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1419                                         log(LOG_WARNING,
1420                                             "attempted source route from %s to %s\n",
1421                                             inet_ntoa(ip->ip_src), buf);
1422                                         type = ICMP_UNREACH;
1423                                         code = ICMP_UNREACH_SRCFAIL;
1424                                         goto bad;
1425                                 } else {
1426                                         /*
1427                                          * Not acting as a router,
1428                                          * so silently drop.
1429                                          */
1430 dropit:
1431                                         ipstat.ips_cantforward++;
1432                                         m_freem(m);
1433                                         return (1);
1434                                 }
1435                         }
1436
1437                         /*
1438                          * locate outgoing interface
1439                          */
1440                         memcpy(&ipaddr.sin_addr, cp + off,
1441                             sizeof ipaddr.sin_addr);
1442
1443                         if (opt == IPOPT_SSRR) {
1444 #define INA     struct in_ifaddr *
1445 #define SA      struct sockaddr *
1446                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1447                                                                         == NULL)
1448                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1449                         } else
1450                                 ia = ip_rtaddr(ipaddr.sin_addr,
1451                                                &ipforward_rt[mycpuid]);
1452                         if (ia == NULL) {
1453                                 type = ICMP_UNREACH;
1454                                 code = ICMP_UNREACH_SRCFAIL;
1455                                 goto bad;
1456                         }
1457                         ip->ip_dst = ipaddr.sin_addr;
1458                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1459                             sizeof(struct in_addr));
1460                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1461                         /*
1462                          * Let ip_intr's mcast routing check handle mcast pkts
1463                          */
1464                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1465                         break;
1466
1467                 case IPOPT_RR:
1468                         if (ipstealth && pass == 0)
1469                                 break;
1470                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1471                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1472                                 goto bad;
1473                         }
1474                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1475                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1476                                 goto bad;
1477                         }
1478                         /*
1479                          * If no space remains, ignore.
1480                          */
1481                         off--;                  /* 0 origin */
1482                         if (off > optlen - (int)sizeof(struct in_addr))
1483                                 break;
1484                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1485                             sizeof ipaddr.sin_addr);
1486                         /*
1487                          * locate outgoing interface; if we're the destination,
1488                          * use the incoming interface (should be same).
1489                          */
1490                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1491                             (ia = ip_rtaddr(ipaddr.sin_addr,
1492                                             &ipforward_rt[mycpuid]))
1493                                                                      == NULL) {
1494                                 type = ICMP_UNREACH;
1495                                 code = ICMP_UNREACH_HOST;
1496                                 goto bad;
1497                         }
1498                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1499                             sizeof(struct in_addr));
1500                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1501                         break;
1502
1503                 case IPOPT_TS:
1504                         if (ipstealth && pass == 0)
1505                                 break;
1506                         code = cp - (u_char *)ip;
1507                         if (optlen < 4 || optlen > 40) {
1508                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1509                                 goto bad;
1510                         }
1511                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1512                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1513                                 goto bad;
1514                         }
1515                         if (off > optlen - (int)sizeof(int32_t)) {
1516                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1517                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1518                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1519                                         goto bad;
1520                                 }
1521                                 break;
1522                         }
1523                         off--;                          /* 0 origin */
1524                         sin = (struct in_addr *)(cp + off);
1525                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1526
1527                         case IPOPT_TS_TSONLY:
1528                                 break;
1529
1530                         case IPOPT_TS_TSANDADDR:
1531                                 if (off + sizeof(n_time) +
1532                                     sizeof(struct in_addr) > optlen) {
1533                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1534                                         goto bad;
1535                                 }
1536                                 ipaddr.sin_addr = dst;
1537                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1538                                                             m->m_pkthdr.rcvif);
1539                                 if (ia == NULL)
1540                                         continue;
1541                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1542                                     sizeof(struct in_addr));
1543                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1544                                 off += sizeof(struct in_addr);
1545                                 break;
1546
1547                         case IPOPT_TS_PRESPEC:
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                                 memcpy(&ipaddr.sin_addr, sin,
1554                                     sizeof(struct in_addr));
1555                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1556                                         continue;
1557                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1558                                 off += sizeof(struct in_addr);
1559                                 break;
1560
1561                         default:
1562                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1563                                 goto bad;
1564                         }
1565                         ntime = iptime();
1566                         memcpy(cp + off, &ntime, sizeof(n_time));
1567                         cp[IPOPT_OFFSET] += sizeof(n_time);
1568                 }
1569         }
1570         if (forward && ipforwarding) {
1571                 ip_forward(m, TRUE, next_hop);
1572                 return (1);
1573         }
1574         return (0);
1575 bad:
1576         icmp_error(m, type, code, 0, 0);
1577         ipstat.ips_badoptions++;
1578         return (1);
1579 }
1580
1581 /*
1582  * Given address of next destination (final or next hop),
1583  * return internet address info of interface to be used to get there.
1584  */
1585 struct in_ifaddr *
1586 ip_rtaddr(struct in_addr dst, struct route *ro)
1587 {
1588         struct sockaddr_in *sin;
1589
1590         sin = (struct sockaddr_in *)&ro->ro_dst;
1591
1592         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1593                 if (ro->ro_rt != NULL) {
1594                         RTFREE(ro->ro_rt);
1595                         ro->ro_rt = NULL;
1596                 }
1597                 sin->sin_family = AF_INET;
1598                 sin->sin_len = sizeof *sin;
1599                 sin->sin_addr = dst;
1600                 rtalloc_ign(ro, RTF_PRCLONING);
1601         }
1602
1603         if (ro->ro_rt == NULL)
1604                 return (NULL);
1605
1606         return (ifatoia(ro->ro_rt->rt_ifa));
1607 }
1608
1609 /*
1610  * Save incoming source route for use in replies,
1611  * to be picked up later by ip_srcroute if the receiver is interested.
1612  */
1613 static void
1614 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1615 {
1616         struct m_tag *mtag;
1617         struct ip_srcrt_opt *opt;
1618         unsigned olen;
1619
1620         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1621         if (mtag == NULL)
1622                 return;
1623         opt = m_tag_data(mtag);
1624
1625         olen = option[IPOPT_OLEN];
1626 #ifdef DIAGNOSTIC
1627         if (ipprintfs)
1628                 kprintf("save_rte: olen %d\n", olen);
1629 #endif
1630         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1631                 m_tag_free(mtag);
1632                 return;
1633         }
1634         bcopy(option, opt->ip_srcrt.srcopt, olen);
1635         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1636         opt->ip_srcrt.dst = dst;
1637         m_tag_prepend(m, mtag);
1638 }
1639
1640 /*
1641  * Retrieve incoming source route for use in replies,
1642  * in the same form used by setsockopt.
1643  * The first hop is placed before the options, will be removed later.
1644  */
1645 struct mbuf *
1646 ip_srcroute(struct mbuf *m0)
1647 {
1648         struct in_addr *p, *q;
1649         struct mbuf *m;
1650         struct m_tag *mtag;
1651         struct ip_srcrt_opt *opt;
1652
1653         if (m0 == NULL)
1654                 return NULL;
1655
1656         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1657         if (mtag == NULL)
1658                 return NULL;
1659         opt = m_tag_data(mtag);
1660
1661         if (opt->ip_nhops == 0)
1662                 return (NULL);
1663         m = m_get(MB_DONTWAIT, MT_HEADER);
1664         if (m == NULL)
1665                 return (NULL);
1666
1667 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1668
1669         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1670         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1671                    sizeof(struct in_addr) + OPTSIZ;
1672 #ifdef DIAGNOSTIC
1673         if (ipprintfs) {
1674                 kprintf("ip_srcroute: nhops %d mlen %d",
1675                         opt->ip_nhops, m->m_len);
1676         }
1677 #endif
1678
1679         /*
1680          * First save first hop for return route
1681          */
1682         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1683         *(mtod(m, struct in_addr *)) = *p--;
1684 #ifdef DIAGNOSTIC
1685         if (ipprintfs)
1686                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1687 #endif
1688
1689         /*
1690          * Copy option fields and padding (nop) to mbuf.
1691          */
1692         opt->ip_srcrt.nop = IPOPT_NOP;
1693         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1694         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1695             OPTSIZ);
1696         q = (struct in_addr *)(mtod(m, caddr_t) +
1697             sizeof(struct in_addr) + OPTSIZ);
1698 #undef OPTSIZ
1699         /*
1700          * Record return path as an IP source route,
1701          * reversing the path (pointers are now aligned).
1702          */
1703         while (p >= opt->ip_srcrt.route) {
1704 #ifdef DIAGNOSTIC
1705                 if (ipprintfs)
1706                         kprintf(" %x", ntohl(q->s_addr));
1707 #endif
1708                 *q++ = *p--;
1709         }
1710         /*
1711          * Last hop goes to final destination.
1712          */
1713         *q = opt->ip_srcrt.dst;
1714         m_tag_delete(m0, mtag);
1715 #ifdef DIAGNOSTIC
1716         if (ipprintfs)
1717                 kprintf(" %x\n", ntohl(q->s_addr));
1718 #endif
1719         return (m);
1720 }
1721
1722 /*
1723  * Strip out IP options.
1724  */
1725 void
1726 ip_stripoptions(struct mbuf *m)
1727 {
1728         int datalen;
1729         struct ip *ip = mtod(m, struct ip *);
1730         caddr_t opts;
1731         int optlen;
1732
1733         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1734         opts = (caddr_t)(ip + 1);
1735         datalen = m->m_len - (sizeof(struct ip) + optlen);
1736         bcopy(opts + optlen, opts, datalen);
1737         m->m_len -= optlen;
1738         if (m->m_flags & M_PKTHDR)
1739                 m->m_pkthdr.len -= optlen;
1740         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1741 }
1742
1743 u_char inetctlerrmap[PRC_NCMDS] = {
1744         0,              0,              0,              0,
1745         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1746         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1747         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1748         0,              0,              0,              0,
1749         ENOPROTOOPT,    ECONNREFUSED
1750 };
1751
1752 /*
1753  * Forward a packet.  If some error occurs return the sender
1754  * an icmp packet.  Note we can't always generate a meaningful
1755  * icmp message because icmp doesn't have a large enough repertoire
1756  * of codes and types.
1757  *
1758  * If not forwarding, just drop the packet.  This could be confusing
1759  * if ipforwarding was zero but some routing protocol was advancing
1760  * us as a gateway to somewhere.  However, we must let the routing
1761  * protocol deal with that.
1762  *
1763  * The using_srcrt parameter indicates whether the packet is being forwarded
1764  * via a source route.
1765  */
1766 void
1767 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1768 {
1769         struct ip *ip = mtod(m, struct ip *);
1770         struct sockaddr_in *ipforward_rtaddr;
1771         struct rtentry *rt;
1772         int error, type = 0, code = 0, destmtu = 0;
1773         struct mbuf *mcopy;
1774         n_long dest;
1775         struct in_addr pkt_dst;
1776         struct route *cache_rt = &ipforward_rt[mycpuid];
1777
1778         dest = INADDR_ANY;
1779         /*
1780          * Cache the destination address of the packet; this may be
1781          * changed by use of 'ipfw fwd'.
1782          */
1783         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1784
1785 #ifdef DIAGNOSTIC
1786         if (ipprintfs)
1787                 kprintf("forward: src %x dst %x ttl %x\n",
1788                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1789 #endif
1790
1791         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1792                 ipstat.ips_cantforward++;
1793                 m_freem(m);
1794                 return;
1795         }
1796         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1797                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1798                 return;
1799         }
1800
1801         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1802         if (cache_rt->ro_rt == NULL ||
1803             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1804                 if (cache_rt->ro_rt != NULL) {
1805                         RTFREE(cache_rt->ro_rt);
1806                         cache_rt->ro_rt = NULL;
1807                 }
1808                 ipforward_rtaddr->sin_family = AF_INET;
1809                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1810                 ipforward_rtaddr->sin_addr = pkt_dst;
1811                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1812                 if (cache_rt->ro_rt == NULL) {
1813                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1814                         return;
1815                 }
1816         }
1817         rt = cache_rt->ro_rt;
1818
1819         /*
1820          * Save the IP header and at most 8 bytes of the payload,
1821          * in case we need to generate an ICMP message to the src.
1822          *
1823          * XXX this can be optimized a lot by saving the data in a local
1824          * buffer on the stack (72 bytes at most), and only allocating the
1825          * mbuf if really necessary. The vast majority of the packets
1826          * are forwarded without having to send an ICMP back (either
1827          * because unnecessary, or because rate limited), so we are
1828          * really we are wasting a lot of work here.
1829          *
1830          * We don't use m_copy() because it might return a reference
1831          * to a shared cluster. Both this function and ip_output()
1832          * assume exclusive access to the IP header in `m', so any
1833          * data in a cluster may change before we reach icmp_error().
1834          */
1835         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1836         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1837                 /*
1838                  * It's probably ok if the pkthdr dup fails (because
1839                  * the deep copy of the tag chain failed), but for now
1840                  * be conservative and just discard the copy since
1841                  * code below may some day want the tags.
1842                  */
1843                 m_free(mcopy);
1844                 mcopy = NULL;
1845         }
1846         if (mcopy != NULL) {
1847                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1848                     (int)ip->ip_len);
1849                 mcopy->m_pkthdr.len = mcopy->m_len;
1850                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1851         }
1852
1853         if (!ipstealth)
1854                 ip->ip_ttl -= IPTTLDEC;
1855
1856         /*
1857          * If forwarding packet using same interface that it came in on,
1858          * perhaps should send a redirect to sender to shortcut a hop.
1859          * Only send redirect if source is sending directly to us,
1860          * and if packet was not source routed (or has any options).
1861          * Also, don't send redirect if forwarding using a default route
1862          * or a route modified by a redirect.
1863          */
1864         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1865             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1866             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1867             ipsendredirects && !using_srcrt && next_hop == NULL) {
1868                 u_long src = ntohl(ip->ip_src.s_addr);
1869                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1870
1871                 if (rt_ifa != NULL &&
1872                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1873                         if (rt->rt_flags & RTF_GATEWAY)
1874                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1875                         else
1876                                 dest = pkt_dst.s_addr;
1877                         /*
1878                          * Router requirements says to only send
1879                          * host redirects.
1880                          */
1881                         type = ICMP_REDIRECT;
1882                         code = ICMP_REDIRECT_HOST;
1883 #ifdef DIAGNOSTIC
1884                         if (ipprintfs)
1885                                 kprintf("redirect (%d) to %x\n", code, dest);
1886 #endif
1887                 }
1888         }
1889
1890         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1891         if (error == 0) {
1892                 ipstat.ips_forward++;
1893                 if (type == 0) {
1894                         if (mcopy) {
1895                                 ipflow_create(cache_rt, mcopy);
1896                                 m_freem(mcopy);
1897                         }
1898                         return;         /* most common case */
1899                 } else {
1900                         ipstat.ips_redirectsent++;
1901                 }
1902         } else {
1903                 ipstat.ips_cantforward++;
1904         }
1905
1906         if (mcopy == NULL)
1907                 return;
1908
1909         /*
1910          * Send ICMP message.
1911          */
1912
1913         switch (error) {
1914
1915         case 0:                         /* forwarded, but need redirect */
1916                 /* type, code set above */
1917                 break;
1918
1919         case ENETUNREACH:               /* shouldn't happen, checked above */
1920         case EHOSTUNREACH:
1921         case ENETDOWN:
1922         case EHOSTDOWN:
1923         default:
1924                 type = ICMP_UNREACH;
1925                 code = ICMP_UNREACH_HOST;
1926                 break;
1927
1928         case EMSGSIZE:
1929                 type = ICMP_UNREACH;
1930                 code = ICMP_UNREACH_NEEDFRAG;
1931 #ifdef IPSEC
1932                 /*
1933                  * If the packet is routed over IPsec tunnel, tell the
1934                  * originator the tunnel MTU.
1935                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1936                  * XXX quickhack!!!
1937                  */
1938                 if (cache_rt->ro_rt != NULL) {
1939                         struct secpolicy *sp = NULL;
1940                         int ipsecerror;
1941                         int ipsechdr;
1942                         struct route *ro;
1943
1944                         sp = ipsec4_getpolicybyaddr(mcopy,
1945                                                     IPSEC_DIR_OUTBOUND,
1946                                                     IP_FORWARDING,
1947                                                     &ipsecerror);
1948
1949                         if (sp == NULL)
1950                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1951                         else {
1952                                 /* count IPsec header size */
1953                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1954                                                          IPSEC_DIR_OUTBOUND,
1955                                                          NULL);
1956
1957                                 /*
1958                                  * find the correct route for outer IPv4
1959                                  * header, compute tunnel MTU.
1960                                  *
1961                                  */
1962                                 if (sp->req != NULL && sp->req->sav != NULL &&
1963                                     sp->req->sav->sah != NULL) {
1964                                         ro = &sp->req->sav->sah->sa_route;
1965                                         if (ro->ro_rt != NULL &&
1966                                             ro->ro_rt->rt_ifp != NULL) {
1967                                                 destmtu =
1968                                                     ro->ro_rt->rt_ifp->if_mtu;
1969                                                 destmtu -= ipsechdr;
1970                                         }
1971                                 }
1972
1973                                 key_freesp(sp);
1974                         }
1975                 }
1976 #elif FAST_IPSEC
1977                 /*
1978                  * If the packet is routed over IPsec tunnel, tell the
1979                  * originator the tunnel MTU.
1980                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1981                  * XXX quickhack!!!
1982                  */
1983                 if (cache_rt->ro_rt != NULL) {
1984                         struct secpolicy *sp = NULL;
1985                         int ipsecerror;
1986                         int ipsechdr;
1987                         struct route *ro;
1988
1989                         sp = ipsec_getpolicybyaddr(mcopy,
1990                                                    IPSEC_DIR_OUTBOUND,
1991                                                    IP_FORWARDING,
1992                                                    &ipsecerror);
1993
1994                         if (sp == NULL)
1995                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1996                         else {
1997                                 /* count IPsec header size */
1998                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1999                                                          IPSEC_DIR_OUTBOUND,
2000                                                          NULL);
2001
2002                                 /*
2003                                  * find the correct route for outer IPv4
2004                                  * header, compute tunnel MTU.
2005                                  */
2006
2007                                 if (sp->req != NULL &&
2008                                     sp->req->sav != NULL &&
2009                                     sp->req->sav->sah != NULL) {
2010                                         ro = &sp->req->sav->sah->sa_route;
2011                                         if (ro->ro_rt != NULL &&
2012                                             ro->ro_rt->rt_ifp != NULL) {
2013                                                 destmtu =
2014                                                     ro->ro_rt->rt_ifp->if_mtu;
2015                                                 destmtu -= ipsechdr;
2016                                         }
2017                                 }
2018
2019                                 KEY_FREESP(&sp);
2020                         }
2021                 }
2022 #else /* !IPSEC && !FAST_IPSEC */
2023                 if (cache_rt->ro_rt != NULL)
2024                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2025 #endif /*IPSEC*/
2026                 ipstat.ips_cantfrag++;
2027                 break;
2028
2029         case ENOBUFS:
2030                 /*
2031                  * A router should not generate ICMP_SOURCEQUENCH as
2032                  * required in RFC1812 Requirements for IP Version 4 Routers.
2033                  * Source quench could be a big problem under DoS attacks,
2034                  * or if the underlying interface is rate-limited.
2035                  * Those who need source quench packets may re-enable them
2036                  * via the net.inet.ip.sendsourcequench sysctl.
2037                  */
2038                 if (!ip_sendsourcequench) {
2039                         m_freem(mcopy);
2040                         return;
2041                 } else {
2042                         type = ICMP_SOURCEQUENCH;
2043                         code = 0;
2044                 }
2045                 break;
2046
2047         case EACCES:                    /* ipfw denied packet */
2048                 m_freem(mcopy);
2049                 return;
2050         }
2051         icmp_error(mcopy, type, code, dest, destmtu);
2052 }
2053
2054 void
2055 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2056                struct mbuf *m)
2057 {
2058         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2059                 struct timeval tv;
2060
2061                 microtime(&tv);
2062                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2063                     SCM_TIMESTAMP, SOL_SOCKET);
2064                 if (*mp)
2065                         mp = &(*mp)->m_next;
2066         }
2067         if (inp->inp_flags & INP_RECVDSTADDR) {
2068                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2069                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2070                 if (*mp)
2071                         mp = &(*mp)->m_next;
2072         }
2073         if (inp->inp_flags & INP_RECVTTL) {
2074                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2075                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2076                 if (*mp)
2077                         mp = &(*mp)->m_next;
2078         }
2079 #ifdef notyet
2080         /* XXX
2081          * Moving these out of udp_input() made them even more broken
2082          * than they already were.
2083          */
2084         /* options were tossed already */
2085         if (inp->inp_flags & INP_RECVOPTS) {
2086                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2087                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2088                 if (*mp)
2089                         mp = &(*mp)->m_next;
2090         }
2091         /* ip_srcroute doesn't do what we want here, need to fix */
2092         if (inp->inp_flags & INP_RECVRETOPTS) {
2093                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2094                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2095                 if (*mp)
2096                         mp = &(*mp)->m_next;
2097         }
2098 #endif
2099         if (inp->inp_flags & INP_RECVIF) {
2100                 struct ifnet *ifp;
2101                 struct sdlbuf {
2102                         struct sockaddr_dl sdl;
2103                         u_char  pad[32];
2104                 } sdlbuf;
2105                 struct sockaddr_dl *sdp;
2106                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2107
2108                 if (((ifp = m->m_pkthdr.rcvif)) &&
2109                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2110                         sdp = IF_LLSOCKADDR(ifp);
2111                         /*
2112                          * Change our mind and don't try copy.
2113                          */
2114                         if ((sdp->sdl_family != AF_LINK) ||
2115                             (sdp->sdl_len > sizeof(sdlbuf))) {
2116                                 goto makedummy;
2117                         }
2118                         bcopy(sdp, sdl2, sdp->sdl_len);
2119                 } else {
2120 makedummy:
2121                         sdl2->sdl_len =
2122                             offsetof(struct sockaddr_dl, sdl_data[0]);
2123                         sdl2->sdl_family = AF_LINK;
2124                         sdl2->sdl_index = 0;
2125                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2126                 }
2127                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2128                         IP_RECVIF, IPPROTO_IP);
2129                 if (*mp)
2130                         mp = &(*mp)->m_next;
2131         }
2132 }
2133
2134 /*
2135  * XXX these routines are called from the upper part of the kernel.
2136  *
2137  * They could also be moved to ip_mroute.c, since all the RSVP
2138  *  handling is done there already.
2139  */
2140 int
2141 ip_rsvp_init(struct socket *so)
2142 {
2143         if (so->so_type != SOCK_RAW ||
2144             so->so_proto->pr_protocol != IPPROTO_RSVP)
2145                 return EOPNOTSUPP;
2146
2147         if (ip_rsvpd != NULL)
2148                 return EADDRINUSE;
2149
2150         ip_rsvpd = so;
2151         /*
2152          * This may seem silly, but we need to be sure we don't over-increment
2153          * the RSVP counter, in case something slips up.
2154          */
2155         if (!ip_rsvp_on) {
2156                 ip_rsvp_on = 1;
2157                 rsvp_on++;
2158         }
2159
2160         return 0;
2161 }
2162
2163 int
2164 ip_rsvp_done(void)
2165 {
2166         ip_rsvpd = NULL;
2167         /*
2168          * This may seem silly, but we need to be sure we don't over-decrement
2169          * the RSVP counter, in case something slips up.
2170          */
2171         if (ip_rsvp_on) {
2172                 ip_rsvp_on = 0;
2173                 rsvp_on--;
2174         }
2175         return 0;
2176 }
2177
2178 void
2179 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2180 {
2181         int off, proto;
2182         __va_list ap;
2183
2184         __va_start(ap, m);
2185         off = __va_arg(ap, int);
2186         proto = __va_arg(ap, int);
2187         __va_end(ap);
2188
2189         if (rsvp_input_p) { /* call the real one if loaded */
2190                 rsvp_input_p(m, off, proto);
2191                 return;
2192         }
2193
2194         /* Can still get packets with rsvp_on = 0 if there is a local member
2195          * of the group to which the RSVP packet is addressed.  But in this
2196          * case we want to throw the packet away.
2197          */
2198
2199         if (!rsvp_on) {
2200                 m_freem(m);
2201                 return;
2202         }
2203
2204         if (ip_rsvpd != NULL) {
2205                 rip_input(m, off, proto);
2206                 return;
2207         }
2208         /* Drop the packet */
2209         m_freem(m);
2210 }