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 1.108 2008/09/23 11:28:49 sephe Exp $
  69  */
  70
  71 #define _IP_VHL
  72
  73 #include "opt_bootp.h"
  74 #include "opt_ipfw.h"
  75 #include "opt_ipdn.h"
  76 #include "opt_ipdivert.h"
  77 #include "opt_ipfilter.h"
  78 #include "opt_ipstealth.h"
  79 #include "opt_ipsec.h"
  80
  81 #include <sys/param.h>
  82 #include <sys/systm.h>
  83 #include <sys/mbuf.h>
  84 #include <sys/malloc.h>
  85 #include <sys/mpipe.h>
  86 #include <sys/domain.h>
  87 #include <sys/protosw.h>
  88 #include <sys/socket.h>
  89 #include <sys/time.h>
  90 #include <sys/globaldata.h>
  91 #include <sys/thread.h>
  92 #include <sys/kernel.h>
  93 #include <sys/syslog.h>
  94 #include <sys/sysctl.h>
  95 #include <sys/in_cksum.h>
  96
  97 #include <machine/stdarg.h>
  98
  99 #include <net/if.h>
 100 #include <net/if_types.h>
 101 #include <net/if_var.h>
 102 #include <net/if_dl.h>
 103 #include <net/pfil.h>
 104 #include <net/route.h>
 105 #include <net/netisr.h>
 106
 107 #include <netinet/in.h>
 108 #include <netinet/in_systm.h>
 109 #include <netinet/in_var.h>
 110 #include <netinet/ip.h>
 111 #include <netinet/in_pcb.h>
 112 #include <netinet/ip_var.h>
 113 #include <netinet/ip_icmp.h>
 114 #include <netinet/ip_divert.h>
 115
 116 #include <sys/thread2.h>
 117 #include <sys/msgport2.h>
 118 #include <net/netmsg2.h>
 119
 120 #include <sys/socketvar.h>
 121
 122 #include <net/ipfw/ip_fw.h>
 123 #include <net/dummynet/ip_dummynet.h>
 124
 125 #ifdef IPSEC
 126 #include <netinet6/ipsec.h>
 127 #include <netproto/key/key.h>
 128 #endif
 129
 130 #ifdef FAST_IPSEC
 131 #include <netproto/ipsec/ipsec.h>
 132 #include <netproto/ipsec/key.h>
 133 #endif
 134
 135 int rsvp_on = 0;
 136 static int ip_rsvp_on;
 137 struct socket *ip_rsvpd;
 138
 139 int ipforwarding = 0;
 140 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 141     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 142
 143 static int ipsendredirects = 1; /* XXX */
 144 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 145     &ipsendredirects, 0, "Enable sending IP redirects");
 146
 147 int ip_defttl = IPDEFTTL;
 148 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 149     &ip_defttl, 0, "Maximum TTL on IP packets");
 150
 151 static int ip_dosourceroute = 0;
 152 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 153     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 154
 155 static int ip_acceptsourceroute = 0;
 156 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 157     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 158     "Enable accepting source routed IP packets");
 159
 160 static int ip_keepfaith = 0;
 161 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 162     &ip_keepfaith, 0,
 163     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 164
 165 static int nipq = 0;    /* total # of reass queues */
 166 static int maxnipq;
 167 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 168     &maxnipq, 0,
 169     "Maximum number of IPv4 fragment reassembly queue entries");
 170
 171 static int maxfragsperpacket;
 172 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 173     &maxfragsperpacket, 0,
 174     "Maximum number of IPv4 fragments allowed per packet");
 175
 176 static int ip_sendsourcequench = 0;
 177 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 178     &ip_sendsourcequench, 0,
 179     "Enable the transmission of source quench packets");
 180
 181 int ip_do_randomid = 1;
 182 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 183     &ip_do_randomid, 0,
 184     "Assign random ip_id values");
 185 /*
 186  * XXX - Setting ip_checkinterface mostly implements the receive side of
 187  * the Strong ES model described in RFC 1122, but since the routing table
 188  * and transmit implementation do not implement the Strong ES model,
 189  * setting this to 1 results in an odd hybrid.
 190  *
 191  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 192  * to translate the destination address to another local interface.
 193  *
 194  * XXX - ip_checkinterface must be disabled if you add IP aliases
 195  * to the loopback interface instead of the interface where the
 196  * packets for those addresses are received.
 197  */
 198 static int ip_checkinterface = 0;
 199 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 200     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 201
 202 #ifdef DIAGNOSTIC
 203 static int ipprintfs = 0;
 204 #endif
 205
 206 extern  int udp_mpsafe_proto;
 207 extern  int tcp_mpsafe_proto;
 208
 209 extern  struct domain inetdomain;
 210 extern  struct protosw inetsw[];
 211 u_char  ip_protox[IPPROTO_MAX];
 212 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 213 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 214                                                 /* inet addr hash table */
 215 u_long  in_ifaddrhmask;                         /* mask for hash table */
 216
 217 struct ip_stats ipstats_percpu[MAXCPU];
 218 #ifdef SMP
 219 static int
 220 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 221 {
 222         int cpu, error = 0;
 223
 224         for (cpu = 0; cpu < ncpus; ++cpu) {
 225                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 226                                         sizeof(struct ip_stats))))
 227                         break;
 228                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 229                                        sizeof(struct ip_stats))))
 230                         break;
 231         }
 232
 233         return (error);
 234 }
 235 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 236     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 237 #else
 238 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 239     &ipstat, ip_stats, "IP statistics");
 240 #endif
 241
 242 /* Packet reassembly stuff */
 243 #define IPREASS_NHASH_LOG2      6
 244 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 245 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 246 #define IPREASS_HASH(x,y)                                               \
 247     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 248
 249 static struct ipq ipq[IPREASS_NHASH];
 250
 251 #ifdef IPCTL_DEFMTU
 252 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 253     &ip_mtu, 0, "Default MTU");
 254 #endif
 255
 256 #ifdef IPSTEALTH
 257 static int ipstealth = 0;
 258 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 259 #else
 260 static const int ipstealth = 0;
 261 #endif
 262
 263 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
 264
 265 struct pfil_head inet_pfil_hook;
 266
 267 /*
 268  * struct ip_srcrt_opt is used to store packet state while it travels
 269  * through the stack.
 270  *
 271  * XXX Note that the code even makes assumptions on the size and
 272  * alignment of fields inside struct ip_srcrt so e.g. adding some
 273  * fields will break the code.  This needs to be fixed.
 274  *
 275  * We need to save the IP options in case a protocol wants to respond
 276  * to an incoming packet over the same route if the packet got here
 277  * using IP source routing.  This allows connection establishment and
 278  * maintenance when the remote end is on a network that is not known
 279  * to us.
 280  */
 281 struct ip_srcrt {
 282         struct  in_addr dst;                    /* final destination */
 283         char    nop;                            /* one NOP to align */
 284         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 285         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 286 };
 287
 288 struct ip_srcrt_opt {
 289         int             ip_nhops;
 290         struct ip_srcrt ip_srcrt;
 291 };
 292
 293 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 294 static struct malloc_pipe ipq_mpipe;
 295
 296 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
 297 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 298 static void             ip_freef(struct ipq *);
 299 static void             ip_input_handler(struct netmsg *);
 300
 301 /*
 302  * IP initialization: fill in IP protocol switch table.
 303  * All protocols not implemented in kernel go to raw IP protocol handler.
 304  */
 305 void
 306 ip_init(void)
 307 {
 308         struct protosw *pr;
 309         int i;
 310 #ifdef SMP
 311         int cpu;
 312 #endif
 313
 314         /*
 315          * Make sure we can handle a reasonable number of fragments but
 316          * cap it at 4000 (XXX).
 317          */
 318         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 319                     IFQ_MAXLEN, 4000, 0, NULL);
 320         for (i = 0; i < ncpus; ++i) {
 321                 TAILQ_INIT(&in_ifaddrheads[i]);
 322                 in_ifaddrhashtbls[i] =
 323                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 324         }
 325         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 326         if (pr == NULL)
 327                 panic("ip_init");
 328         for (i = 0; i < IPPROTO_MAX; i++)
 329                 ip_protox[i] = pr - inetsw;
 330         for (pr = inetdomain.dom_protosw;
 331              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
 332                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
 333                         if (pr->pr_protocol != IPPROTO_RAW)
 334                                 ip_protox[pr->pr_protocol] = pr - inetsw;
 335
 336                         /* XXX */
 337                         switch (pr->pr_protocol) {
 338                         case IPPROTO_TCP:
 339                                 if (tcp_mpsafe_proto)
 340                                         pr->pr_flags |= PR_MPSAFE;
 341                                 break;
 342
 343                         case IPPROTO_UDP:
 344                                 if (udp_mpsafe_proto)
 345                                         pr->pr_flags |= PR_MPSAFE;
 346                                 break;
 347                         }
 348                 }
 349         }
 350
 351         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 352         inet_pfil_hook.ph_af = AF_INET;
 353         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 354                 kprintf("%s: WARNING: unable to register pfil hook, "
 355                         "error %d\n", __func__, i);
 356         }
 357
 358         for (i = 0; i < IPREASS_NHASH; i++)
 359                 ipq[i].next = ipq[i].prev = &ipq[i];
 360
 361         maxnipq = nmbclusters / 32;
 362         maxfragsperpacket = 16;
 363
 364         ip_id = time_second & 0xffff;
 365
 366         /*
 367          * Initialize IP statistics counters for each CPU.
 368          *
 369          */
 370 #ifdef SMP
 371         for (cpu = 0; cpu < ncpus; ++cpu) {
 372                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 373         }
 374 #else
 375         bzero(&ipstat, sizeof(struct ip_stats));
 376 #endif
 377
 378         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler, 0);
 379 }
 380
 381 /*
 382  * XXX watch out this one. It is perhaps used as a cache for
 383  * the most recently used route ? it is cleared in in_addroute()
 384  * when a new route is successfully created.
 385  */
 386 struct route ipforward_rt[MAXCPU];
 387
 388 /* Do transport protocol processing. */
 389 static void
 390 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 391 {
 392         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 393
 394         /*
 395          * Switch out to protocol's input routine.
 396          */
 397         PR_GET_MPLOCK(pr);
 398         pr->pr_input(m, hlen, ip->ip_p);
 399         PR_REL_MPLOCK(pr);
 400 }
 401
 402 static void
 403 transport_processing_handler(netmsg_t netmsg)
 404 {
 405         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 406         struct ip *ip;
 407         int hlen;
 408
 409         ip = mtod(pmsg->nm_packet, struct ip *);
 410         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 411
 412         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 413         /* netmsg was embedded in the mbuf, do not reply! */
 414 }
 415
 416 static void
 417 ip_input_handler(struct netmsg *msg0)
 418 {
 419         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 420
 421         ip_input(m);
 422         /* msg0 was embedded in the mbuf, do not reply! */
 423 }
 424
 425 /*
 426  * IP input routine.  Checksum and byte swap header.  If fragmented
 427  * try to reassemble.  Process options.  Pass to next level.
 428  */
 429 void
 430 ip_input(struct mbuf *m)
 431 {
 432         struct ip *ip;
 433         struct in_ifaddr *ia = NULL;
 434         struct in_ifaddr_container *iac;
 435         int hlen, checkif;
 436         u_short sum;
 437         struct in_addr pkt_dst;
 438         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 439         boolean_t needredispatch = FALSE;
 440         struct in_addr odst;                    /* original dst address(NAT) */
 441         struct m_tag *mtag;
 442         struct sockaddr_in *next_hop = NULL;
 443 #ifdef FAST_IPSEC
 444         struct tdb_ident *tdbi;
 445         struct secpolicy *sp;
 446         int error;
 447 #endif
 448
 449         M_ASSERTPKTHDR(m);
 450
 451         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 452                 /* Next hop */
 453                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 454                 KKASSERT(mtag != NULL);
 455                 next_hop = m_tag_data(mtag);
 456         }
 457
 458         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 459                 /* dummynet already filtered us */
 460                 ip = mtod(m, struct ip *);
 461                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 462                 goto iphack;
 463         }
 464
 465         ipstat.ips_total++;
 466
 467         /* length checks already done in ip_demux() */
 468         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 469
 470         ip = mtod(m, struct ip *);
 471
 472         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 473                 ipstat.ips_badvers++;
 474                 goto bad;
 475         }
 476
 477         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 478         /* length checks already done in ip_demux() */
 479         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 480         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 481
 482         /* 127/8 must not appear on wire - RFC1122 */
 483         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 484             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 485                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 486                         ipstat.ips_badaddr++;
 487                         goto bad;
 488                 }
 489         }
 490
 491         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 492                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 493         } else {
 494                 if (hlen == sizeof(struct ip)) {
 495                         sum = in_cksum_hdr(ip);
 496                 } else {
 497                         sum = in_cksum(m, hlen);
 498                 }
 499         }
 500         if (sum != 0) {
 501                 ipstat.ips_badsum++;
 502                 goto bad;
 503         }
 504
 505 #ifdef ALTQ
 506         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 507                 /* packet is dropped by traffic conditioner */
 508                 return;
 509         }
 510 #endif
 511         /*
 512          * Convert fields to host representation.
 513          */
 514         ip->ip_len = ntohs(ip->ip_len);
 515         if (ip->ip_len < hlen) {
 516                 ipstat.ips_badlen++;
 517                 goto bad;
 518         }
 519         ip->ip_off = ntohs(ip->ip_off);
 520
 521         /*
 522          * Check that the amount of data in the buffers
 523          * is as at least much as the IP header would have us expect.
 524          * Trim mbufs if longer than we expect.
 525          * Drop packet if shorter than we expect.
 526          */
 527         if (m->m_pkthdr.len < ip->ip_len) {
 528                 ipstat.ips_tooshort++;
 529                 goto bad;
 530         }
 531         if (m->m_pkthdr.len > ip->ip_len) {
 532                 if (m->m_len == m->m_pkthdr.len) {
 533                         m->m_len = ip->ip_len;
 534                         m->m_pkthdr.len = ip->ip_len;
 535                 } else
 536                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 537         }
 538 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 539         /*
 540          * Bypass packet filtering for packets from a tunnel (gif).
 541          */
 542         if (ipsec_gethist(m, NULL))
 543                 goto pass;
 544 #endif
 545
 546         /*
 547          * IpHack's section.
 548          * Right now when no processing on packet has done
 549          * and it is still fresh out of network we do our black
 550          * deals with it.
 551          * - Firewall: deny/allow/divert
 552          * - Xlate: translate packet's addr/port (NAT).
 553          * - Pipe: pass pkt through dummynet.
 554          * - Wrap: fake packet's addr/port <unimpl.>
 555          * - Encapsulate: put it in another IP and send out. <unimp.>
 556          */
 557
 558 iphack:
 559         /*
 560          * If we've been forwarded from the output side, then
 561          * skip the firewall a second time
 562          */
 563         if (next_hop != NULL)
 564                 goto ours;
 565
 566         /* No pfil hooks */
 567         if (!pfil_has_hooks(&inet_pfil_hook)) {
 568                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 569                         /*
 570                          * Strip dummynet tags from stranded packets
 571                          */
 572                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 573                         KKASSERT(mtag != NULL);
 574                         m_tag_delete(m, mtag);
 575                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 576                 }
 577                 goto pass;
 578         }
 579
 580         /*
 581          * Run through list of hooks for input packets.
 582          *
 583          * NB: Beware of the destination address changing (e.g.
 584          *     by NAT rewriting). When this happens, tell
 585          *     ip_forward to do the right thing.
 586          */
 587         odst = ip->ip_dst;
 588         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 589                 return;
 590         if (m == NULL)  /* consumed by filter */
 591                 return;
 592         ip = mtod(m, struct ip *);
 593         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 594         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 595
 596         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 597                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 598                 KKASSERT(mtag != NULL);
 599                 next_hop = m_tag_data(mtag);
 600         }
 601         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 602                 ip_dn_queue(m);
 603                 return;
 604         }
 605         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 606                 needredispatch = TRUE;
 607                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 608         }
 609 pass:
 610         /*
 611          * Process options and, if not destined for us,
 612          * ship it on.  ip_dooptions returns 1 when an
 613          * error was detected (causing an icmp message
 614          * to be sent and the original packet to be freed).
 615          */
 616         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 617                 return;
 618
 619         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 620          * matter if it is destined to another node, or whether it is
 621          * a multicast one, RSVP wants it! and prevents it from being forwarded
 622          * anywhere else. Also checks if the rsvp daemon is running before
 623          * grabbing the packet.
 624          */
 625         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 626                 goto ours;
 627
 628         /*
 629          * Check our list of addresses, to see if the packet is for us.
 630          * If we don't have any addresses, assume any unicast packet
 631          * we receive might be for us (and let the upper layers deal
 632          * with it).
 633          */
 634         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 635             !(m->m_flags & (M_MCAST | M_BCAST)))
 636                 goto ours;
 637
 638         /*
 639          * Cache the destination address of the packet; this may be
 640          * changed by use of 'ipfw fwd'.
 641          */
 642         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 643
 644         /*
 645          * Enable a consistency check between the destination address
 646          * and the arrival interface for a unicast packet (the RFC 1122
 647          * strong ES model) if IP forwarding is disabled and the packet
 648          * is not locally generated and the packet is not subject to
 649          * 'ipfw fwd'.
 650          *
 651          * XXX - Checking also should be disabled if the destination
 652          * address is ipnat'ed to a different interface.
 653          *
 654          * XXX - Checking is incompatible with IP aliases added
 655          * to the loopback interface instead of the interface where
 656          * the packets are received.
 657          */
 658         checkif = ip_checkinterface &&
 659                   !ipforwarding &&
 660                   m->m_pkthdr.rcvif != NULL &&
 661                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 662                   next_hop == NULL;
 663
 664         /*
 665          * Check for exact addresses in the hash bucket.
 666          */
 667         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 668                 ia = iac->ia;
 669
 670                 /*
 671                  * If the address matches, verify that the packet
 672                  * arrived via the correct interface if checking is
 673                  * enabled.
 674                  */
 675                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 676                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 677                         goto ours;
 678         }
 679         ia = NULL;
 680
 681         /*
 682          * Check for broadcast addresses.
 683          *
 684          * Only accept broadcast packets that arrive via the matching
 685          * interface.  Reception of forwarded directed broadcasts would
 686          * be handled via ip_forward() and ether_output() with the loopback
 687          * into the stack for SIMPLEX interfaces handled by ether_output().
 688          */
 689         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 690                 struct ifaddr_container *ifac;
 691
 692                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 693                               ifa_link) {
 694                         struct ifaddr *ifa = ifac->ifa;
 695
 696                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 697                                 continue;
 698                         if (ifa->ifa_addr->sa_family != AF_INET)
 699                                 continue;
 700                         ia = ifatoia(ifa);
 701                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 702                                                                 pkt_dst.s_addr)
 703                                 goto ours;
 704                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 705                                 goto ours;
 706 #ifdef BOOTP_COMPAT
 707                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 708                                 goto ours;
 709 #endif
 710                 }
 711         }
 712         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 713                 struct in_multi *inm;
 714
 715                 if (ip_mrouter != NULL) {
 716                         /*
 717                          * If we are acting as a multicast router, all
 718                          * incoming multicast packets are passed to the
 719                          * kernel-level multicast forwarding function.
 720                          * The packet is returned (relatively) intact; if
 721                          * ip_mforward() returns a non-zero value, the packet
 722                          * must be discarded, else it may be accepted below.
 723                          */
 724                         if (ip_mforward != NULL &&
 725                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 726                                 ipstat.ips_cantforward++;
 727                                 m_freem(m);
 728                                 return;
 729                         }
 730
 731                         /*
 732                          * The process-level routing daemon needs to receive
 733                          * all multicast IGMP packets, whether or not this
 734                          * host belongs to their destination groups.
 735                          */
 736                         if (ip->ip_p == IPPROTO_IGMP)
 737                                 goto ours;
 738                         ipstat.ips_forward++;
 739                 }
 740                 /*
 741                  * See if we belong to the destination multicast group on the
 742                  * arrival interface.
 743                  */
 744                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 745                 if (inm == NULL) {
 746                         ipstat.ips_notmember++;
 747                         m_freem(m);
 748                         return;
 749                 }
 750                 goto ours;
 751         }
 752         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 753                 goto ours;
 754         if (ip->ip_dst.s_addr == INADDR_ANY)
 755                 goto ours;
 756
 757         /*
 758          * FAITH(Firewall Aided Internet Translator)
 759          */
 760         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 761                 if (ip_keepfaith) {
 762                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 763                                 goto ours;
 764                 }
 765                 m_freem(m);
 766                 return;
 767         }
 768
 769         /*
 770          * Not for us; forward if possible and desirable.
 771          */
 772         if (!ipforwarding) {
 773                 ipstat.ips_cantforward++;
 774                 m_freem(m);
 775         } else {
 776 #ifdef IPSEC
 777                 /*
 778                  * Enforce inbound IPsec SPD.
 779                  */
 780                 if (ipsec4_in_reject(m, NULL)) {
 781                         ipsecstat.in_polvio++;
 782                         goto bad;
 783                 }
 784 #endif
 785 #ifdef FAST_IPSEC
 786                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 787                 crit_enter();
 788                 if (mtag != NULL) {
 789                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 790                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 791                 } else {
 792                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 793                                                    IP_FORWARDING, &error);
 794                 }
 795                 if (sp == NULL) {       /* NB: can happen if error */
 796                         crit_exit();
 797                         /*XXX error stat???*/
 798                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 799                         goto bad;
 800                 }
 801
 802                 /*
 803                  * Check security policy against packet attributes.
 804                  */
 805                 error = ipsec_in_reject(sp, m);
 806                 KEY_FREESP(&sp);
 807                 crit_exit();
 808                 if (error) {
 809                         ipstat.ips_cantforward++;
 810                         goto bad;
 811                 }
 812 #endif
 813                 ip_forward(m, using_srcrt, next_hop);
 814         }
 815         return;
 816
 817 ours:
 818
 819         /*
 820          * IPSTEALTH: Process non-routing options only
 821          * if the packet is destined for us.
 822          */
 823         if (ipstealth &&
 824             hlen > sizeof(struct ip) &&
 825             ip_dooptions(m, 1, next_hop))
 826                 return;
 827
 828         /* Count the packet in the ip address stats */
 829         if (ia != NULL) {
 830                 ia->ia_ifa.if_ipackets++;
 831                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 832         }
 833
 834         /*
 835          * If offset or IP_MF are set, must reassemble.
 836          * Otherwise, nothing need be done.
 837          * (We could look in the reassembly queue to see
 838          * if the packet was previously fragmented,
 839          * but it's not worth the time; just let them time out.)
 840          */
 841         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 842                 /*
 843                  * Attempt reassembly; if it succeeds, proceed.
 844                  * ip_reass() will return a different mbuf.
 845                  */
 846                 m = ip_reass(m);
 847                 if (m == NULL)
 848                         return;
 849                 ip = mtod(m, struct ip *);
 850
 851                 /* Get the header length of the reassembled packet */
 852                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 853
 854                 needredispatch = TRUE;
 855         } else {
 856                 ip->ip_len -= hlen;
 857         }
 858
 859 #ifdef IPSEC
 860         /*
 861          * enforce IPsec policy checking if we are seeing last header.
 862          * note that we do not visit this with protocols with pcb layer
 863          * code - like udp/tcp/raw ip.
 864          */
 865         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 866             ipsec4_in_reject(m, NULL)) {
 867                 ipsecstat.in_polvio++;
 868                 goto bad;
 869         }
 870 #endif
 871 #if FAST_IPSEC
 872         /*
 873          * enforce IPsec policy checking if we are seeing last header.
 874          * note that we do not visit this with protocols with pcb layer
 875          * code - like udp/tcp/raw ip.
 876          */
 877         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 878                 /*
 879                  * Check if the packet has already had IPsec processing
 880                  * done.  If so, then just pass it along.  This tag gets
 881                  * set during AH, ESP, etc. input handling, before the
 882                  * packet is returned to the ip input queue for delivery.
 883                  */
 884                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 885                 crit_enter();
 886                 if (mtag != NULL) {
 887                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 888                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 889                 } else {
 890                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 891                                                    IP_FORWARDING, &error);
 892                 }
 893                 if (sp != NULL) {
 894                         /*
 895                          * Check security policy against packet attributes.
 896                          */
 897                         error = ipsec_in_reject(sp, m);
 898                         KEY_FREESP(&sp);
 899                 } else {
 900                         /* XXX error stat??? */
 901                         error = EINVAL;
 902 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 903                         goto bad;
 904                 }
 905                 crit_exit();
 906                 if (error)
 907                         goto bad;
 908         }
 909 #endif /* FAST_IPSEC */
 910
 911         ipstat.ips_delivered++;
 912         if (needredispatch) {
 913                 struct netmsg_packet *pmsg;
 914                 lwkt_port_t port;
 915
 916                 ip->ip_off = htons(ip->ip_off);
 917                 ip->ip_len = htons(ip->ip_len);
 918                 port = ip_mport_in(&m);
 919                 if (port == NULL)
 920                         return;
 921
 922                 pmsg = &m->m_hdr.mh_netmsg;
 923                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, MSGF_MPSAFE,
 924                             transport_processing_handler);
 925                 pmsg->nm_packet = m;
 926                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
 927
 928                 ip = mtod(m, struct ip *);
 929                 ip->ip_len = ntohs(ip->ip_len);
 930                 ip->ip_off = ntohs(ip->ip_off);
 931                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
 932         } else {
 933                 transport_processing_oncpu(m, hlen, ip);
 934         }
 935         return;
 936
 937 bad:
 938         m_freem(m);
 939 }
 940
 941 /*
 942  * Take incoming datagram fragment and try to reassemble it into
 943  * whole datagram.  If a chain for reassembly of this datagram already
 944  * exists, then it is given as fp; otherwise have to make a chain.
 945  */
 946 struct mbuf *
 947 ip_reass(struct mbuf *m)
 948 {
 949         struct ip *ip = mtod(m, struct ip *);
 950         struct mbuf *p = NULL, *q, *nq;
 951         struct mbuf *n;
 952         struct ipq *fp = NULL;
 953         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 954         int i, next;
 955         u_short sum;
 956
 957         /* If maxnipq is 0, never accept fragments. */
 958         if (maxnipq == 0) {
 959                 ipstat.ips_fragments++;
 960                 ipstat.ips_fragdropped++;
 961                 m_freem(m);
 962                 return NULL;
 963         }
 964
 965         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 966         /*
 967          * Look for queue of fragments of this datagram.
 968          */
 969         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 970                 if (ip->ip_id == fp->ipq_id &&
 971                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 972                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 973                     ip->ip_p == fp->ipq_p)
 974                         goto found;
 975
 976         fp = NULL;
 977
 978         /*
 979          * Enforce upper bound on number of fragmented packets
 980          * for which we attempt reassembly;
 981          * If maxnipq is -1, accept all fragments without limitation.
 982          */
 983         if (nipq > maxnipq && maxnipq > 0) {
 984                 /*
 985                  * drop something from the tail of the current queue
 986                  * before proceeding further
 987                  */
 988                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
 989                         for (i = 0; i < IPREASS_NHASH; i++) {
 990                                 if (ipq[i].prev != &ipq[i]) {
 991                                         ipstat.ips_fragtimeout +=
 992                                             ipq[i].prev->ipq_nfrags;
 993                                         ip_freef(ipq[i].prev);
 994                                         break;
 995                                 }
 996                         }
 997                 } else {
 998                         ipstat.ips_fragtimeout +=
 999                             ipq[sum].prev->ipq_nfrags;
1000                         ip_freef(ipq[sum].prev);
1001                 }
1002         }
1003 found:
1004         /*
1005          * Adjust ip_len to not reflect header,
1006          * convert offset of this to bytes.
1007          */
1008         ip->ip_len -= hlen;
1009         if (ip->ip_off & IP_MF) {
1010                 /*
1011                  * Make sure that fragments have a data length
1012                  * that's a non-zero multiple of 8 bytes.
1013                  */
1014                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1015                         ipstat.ips_toosmall++; /* XXX */
1016                         m_freem(m);
1017                         return NULL;
1018                 }
1019                 m->m_flags |= M_FRAG;
1020         } else
1021                 m->m_flags &= ~M_FRAG;
1022         ip->ip_off <<= 3;
1023
1024         ipstat.ips_fragments++;
1025         m->m_pkthdr.header = ip;
1026
1027         /*
1028          * If the hardware has not done csum over this fragment
1029          * then csum_data is not valid at all.
1030          */
1031         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1032             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1033                 m->m_pkthdr.csum_data = 0;
1034                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1035         }
1036
1037         /*
1038          * Presence of header sizes in mbufs
1039          * would confuse code below.
1040          */
1041         m->m_data += hlen;
1042         m->m_len -= hlen;
1043
1044         /*
1045          * If first fragment to arrive, create a reassembly queue.
1046          */
1047         if (fp == NULL) {
1048                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1049                         goto dropfrag;
1050                 insque(fp, &ipq[sum]);
1051                 nipq++;
1052                 fp->ipq_nfrags = 1;
1053                 fp->ipq_ttl = IPFRAGTTL;
1054                 fp->ipq_p = ip->ip_p;
1055                 fp->ipq_id = ip->ip_id;
1056                 fp->ipq_src = ip->ip_src;
1057                 fp->ipq_dst = ip->ip_dst;
1058                 fp->ipq_frags = m;
1059                 m->m_nextpkt = NULL;
1060                 goto inserted;
1061         } else {
1062                 fp->ipq_nfrags++;
1063         }
1064
1065 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1066
1067         /*
1068          * Find a segment which begins after this one does.
1069          */
1070         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1071                 if (GETIP(q)->ip_off > ip->ip_off)
1072                         break;
1073
1074         /*
1075          * If there is a preceding segment, it may provide some of
1076          * our data already.  If so, drop the data from the incoming
1077          * segment.  If it provides all of our data, drop us, otherwise
1078          * stick new segment in the proper place.
1079          *
1080          * If some of the data is dropped from the the preceding
1081          * segment, then it's checksum is invalidated.
1082          */
1083         if (p) {
1084                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1085                 if (i > 0) {
1086                         if (i >= ip->ip_len)
1087                                 goto dropfrag;
1088                         m_adj(m, i);
1089                         m->m_pkthdr.csum_flags = 0;
1090                         ip->ip_off += i;
1091                         ip->ip_len -= i;
1092                 }
1093                 m->m_nextpkt = p->m_nextpkt;
1094                 p->m_nextpkt = m;
1095         } else {
1096                 m->m_nextpkt = fp->ipq_frags;
1097                 fp->ipq_frags = m;
1098         }
1099
1100         /*
1101          * While we overlap succeeding segments trim them or,
1102          * if they are completely covered, dequeue them.
1103          */
1104         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1105              q = nq) {
1106                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1107                 if (i < GETIP(q)->ip_len) {
1108                         GETIP(q)->ip_len -= i;
1109                         GETIP(q)->ip_off += i;
1110                         m_adj(q, i);
1111                         q->m_pkthdr.csum_flags = 0;
1112                         break;
1113                 }
1114                 nq = q->m_nextpkt;
1115                 m->m_nextpkt = nq;
1116                 ipstat.ips_fragdropped++;
1117                 fp->ipq_nfrags--;
1118                 q->m_nextpkt = NULL;
1119                 m_freem(q);
1120         }
1121
1122 inserted:
1123         /*
1124          * Check for complete reassembly and perform frag per packet
1125          * limiting.
1126          *
1127          * Frag limiting is performed here so that the nth frag has
1128          * a chance to complete the packet before we drop the packet.
1129          * As a result, n+1 frags are actually allowed per packet, but
1130          * only n will ever be stored. (n = maxfragsperpacket.)
1131          *
1132          */
1133         next = 0;
1134         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1135                 if (GETIP(q)->ip_off != next) {
1136                         if (fp->ipq_nfrags > maxfragsperpacket) {
1137                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1138                                 ip_freef(fp);
1139                         }
1140                         return (NULL);
1141                 }
1142                 next += GETIP(q)->ip_len;
1143         }
1144         /* Make sure the last packet didn't have the IP_MF flag */
1145         if (p->m_flags & M_FRAG) {
1146                 if (fp->ipq_nfrags > maxfragsperpacket) {
1147                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1148                         ip_freef(fp);
1149                 }
1150                 return (NULL);
1151         }
1152
1153         /*
1154          * Reassembly is complete.  Make sure the packet is a sane size.
1155          */
1156         q = fp->ipq_frags;
1157         ip = GETIP(q);
1158         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1159                 ipstat.ips_toolong++;
1160                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1161                 ip_freef(fp);
1162                 return (NULL);
1163         }
1164
1165         /*
1166          * Concatenate fragments.
1167          */
1168         m = q;
1169         n = m->m_next;
1170         m->m_next = NULL;
1171         m_cat(m, n);
1172         nq = q->m_nextpkt;
1173         q->m_nextpkt = NULL;
1174         for (q = nq; q != NULL; q = nq) {
1175                 nq = q->m_nextpkt;
1176                 q->m_nextpkt = NULL;
1177                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1178                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1179                 m_cat(m, q);
1180         }
1181
1182         /*
1183          * Clean up the 1's complement checksum.  Carry over 16 bits must
1184          * be added back.  This assumes no more then 65535 packet fragments
1185          * were reassembled.  A second carry can also occur (but not a third).
1186          */
1187         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1188                                 (m->m_pkthdr.csum_data >> 16);
1189         if (m->m_pkthdr.csum_data > 0xFFFF)
1190                 m->m_pkthdr.csum_data -= 0xFFFF;
1191
1192         /*
1193          * Create header for new ip packet by
1194          * modifying header of first packet;
1195          * dequeue and discard fragment reassembly header.
1196          * Make header visible.
1197          */
1198         ip->ip_len = next;
1199         ip->ip_src = fp->ipq_src;
1200         ip->ip_dst = fp->ipq_dst;
1201         remque(fp);
1202         nipq--;
1203         mpipe_free(&ipq_mpipe, fp);
1204         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1205         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1206         /* some debugging cruft by sklower, below, will go away soon */
1207         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1208                 int plen = 0;
1209
1210                 for (n = m; n; n = n->m_next)
1211                         plen += n->m_len;
1212                 m->m_pkthdr.len = plen;
1213         }
1214
1215         ipstat.ips_reassembled++;
1216         return (m);
1217
1218 dropfrag:
1219         ipstat.ips_fragdropped++;
1220         if (fp != NULL)
1221                 fp->ipq_nfrags--;
1222         m_freem(m);
1223         return (NULL);
1224
1225 #undef GETIP
1226 }
1227
1228 /*
1229  * Free a fragment reassembly header and all
1230  * associated datagrams.
1231  */
1232 static void
1233 ip_freef(struct ipq *fp)
1234 {
1235         struct mbuf *q;
1236
1237         while (fp->ipq_frags) {
1238                 q = fp->ipq_frags;
1239                 fp->ipq_frags = q->m_nextpkt;
1240                 q->m_nextpkt = NULL;
1241                 m_freem(q);
1242         }
1243         remque(fp);
1244         mpipe_free(&ipq_mpipe, fp);
1245         nipq--;
1246 }
1247
1248 /*
1249  * IP timer processing;
1250  * if a timer expires on a reassembly
1251  * queue, discard it.
1252  */
1253 void
1254 ip_slowtimo(void)
1255 {
1256         struct ipq *fp;
1257         int i;
1258
1259         crit_enter();
1260         for (i = 0; i < IPREASS_NHASH; i++) {
1261                 fp = ipq[i].next;
1262                 if (fp == NULL)
1263                         continue;
1264                 while (fp != &ipq[i]) {
1265                         --fp->ipq_ttl;
1266                         fp = fp->next;
1267                         if (fp->prev->ipq_ttl == 0) {
1268                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1269                                 ip_freef(fp->prev);
1270                         }
1271                 }
1272         }
1273         /*
1274          * If we are over the maximum number of fragments
1275          * (due to the limit being lowered), drain off
1276          * enough to get down to the new limit.
1277          */
1278         if (maxnipq >= 0 && nipq > maxnipq) {
1279                 for (i = 0; i < IPREASS_NHASH; i++) {
1280                         while (nipq > maxnipq &&
1281                                 (ipq[i].next != &ipq[i])) {
1282                                 ipstat.ips_fragdropped +=
1283                                     ipq[i].next->ipq_nfrags;
1284                                 ip_freef(ipq[i].next);
1285                         }
1286                 }
1287         }
1288         ipflow_slowtimo();
1289         crit_exit();
1290 }
1291
1292 /*
1293  * Drain off all datagram fragments.
1294  */
1295 void
1296 ip_drain(void)
1297 {
1298         int i;
1299
1300         for (i = 0; i < IPREASS_NHASH; i++) {
1301                 while (ipq[i].next != &ipq[i]) {
1302                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1303                         ip_freef(ipq[i].next);
1304                 }
1305         }
1306         in_rtqdrain();
1307 }
1308
1309 /*
1310  * Do option processing on a datagram,
1311  * possibly discarding it if bad options are encountered,
1312  * or forwarding it if source-routed.
1313  * The pass argument is used when operating in the IPSTEALTH
1314  * mode to tell what options to process:
1315  * [LS]SRR (pass 0) or the others (pass 1).
1316  * The reason for as many as two passes is that when doing IPSTEALTH,
1317  * non-routing options should be processed only if the packet is for us.
1318  * Returns 1 if packet has been forwarded/freed,
1319  * 0 if the packet should be processed further.
1320  */
1321 static int
1322 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1323 {
1324         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1325         struct ip *ip = mtod(m, struct ip *);
1326         u_char *cp;
1327         struct in_ifaddr *ia;
1328         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1329         boolean_t forward = FALSE;
1330         struct in_addr *sin, dst;
1331         n_time ntime;
1332
1333         dst = ip->ip_dst;
1334         cp = (u_char *)(ip + 1);
1335         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1336         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1337                 opt = cp[IPOPT_OPTVAL];
1338                 if (opt == IPOPT_EOL)
1339                         break;
1340                 if (opt == IPOPT_NOP)
1341                         optlen = 1;
1342                 else {
1343                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1344                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1345                                 goto bad;
1346                         }
1347                         optlen = cp[IPOPT_OLEN];
1348                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1349                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1350                                 goto bad;
1351                         }
1352                 }
1353                 switch (opt) {
1354
1355                 default:
1356                         break;
1357
1358                 /*
1359                  * Source routing with record.
1360                  * Find interface with current destination address.
1361                  * If none on this machine then drop if strictly routed,
1362                  * or do nothing if loosely routed.
1363                  * Record interface address and bring up next address
1364                  * component.  If strictly routed make sure next
1365                  * address is on directly accessible net.
1366                  */
1367                 case IPOPT_LSRR:
1368                 case IPOPT_SSRR:
1369                         if (ipstealth && pass > 0)
1370                                 break;
1371                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1372                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1373                                 goto bad;
1374                         }
1375                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1376                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1377                                 goto bad;
1378                         }
1379                         ipaddr.sin_addr = ip->ip_dst;
1380                         ia = (struct in_ifaddr *)
1381                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1382                         if (ia == NULL) {
1383                                 if (opt == IPOPT_SSRR) {
1384                                         type = ICMP_UNREACH;
1385                                         code = ICMP_UNREACH_SRCFAIL;
1386                                         goto bad;
1387                                 }
1388                                 if (!ip_dosourceroute)
1389                                         goto nosourcerouting;
1390                                 /*
1391                                  * Loose routing, and not at next destination
1392                                  * yet; nothing to do except forward.
1393                                  */
1394                                 break;
1395                         }
1396                         off--;                  /* 0 origin */
1397                         if (off > optlen - (int)sizeof(struct in_addr)) {
1398                                 /*
1399                                  * End of source route.  Should be for us.
1400                                  */
1401                                 if (!ip_acceptsourceroute)
1402                                         goto nosourcerouting;
1403                                 save_rte(m, cp, ip->ip_src);
1404                                 break;
1405                         }
1406                         if (ipstealth)
1407                                 goto dropit;
1408                         if (!ip_dosourceroute) {
1409                                 if (ipforwarding) {
1410                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1411
1412                                         /*
1413                                          * Acting as a router, so generate ICMP
1414                                          */
1415 nosourcerouting:
1416                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1417                                         log(LOG_WARNING,
1418                                             "attempted source route from %s to %s\n",
1419                                             inet_ntoa(ip->ip_src), buf);
1420                                         type = ICMP_UNREACH;
1421                                         code = ICMP_UNREACH_SRCFAIL;
1422                                         goto bad;
1423                                 } else {
1424                                         /*
1425                                          * Not acting as a router,
1426                                          * so silently drop.
1427                                          */
1428 dropit:
1429                                         ipstat.ips_cantforward++;
1430                                         m_freem(m);
1431                                         return (1);
1432                                 }
1433                         }
1434
1435                         /*
1436                          * locate outgoing interface
1437                          */
1438                         memcpy(&ipaddr.sin_addr, cp + off,
1439                             sizeof ipaddr.sin_addr);
1440
1441                         if (opt == IPOPT_SSRR) {
1442 #define INA     struct in_ifaddr *
1443 #define SA      struct sockaddr *
1444                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1445                                                                         == NULL)
1446                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1447                         } else
1448                                 ia = ip_rtaddr(ipaddr.sin_addr,
1449                                                &ipforward_rt[mycpuid]);
1450                         if (ia == NULL) {
1451                                 type = ICMP_UNREACH;
1452                                 code = ICMP_UNREACH_SRCFAIL;
1453                                 goto bad;
1454                         }
1455                         ip->ip_dst = ipaddr.sin_addr;
1456                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1457                             sizeof(struct in_addr));
1458                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1459                         /*
1460                          * Let ip_intr's mcast routing check handle mcast pkts
1461                          */
1462                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1463                         break;
1464
1465                 case IPOPT_RR:
1466                         if (ipstealth && pass == 0)
1467                                 break;
1468                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1469                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1470                                 goto bad;
1471                         }
1472                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1473                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1474                                 goto bad;
1475                         }
1476                         /*
1477                          * If no space remains, ignore.
1478                          */
1479                         off--;                  /* 0 origin */
1480                         if (off > optlen - (int)sizeof(struct in_addr))
1481                                 break;
1482                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1483                             sizeof ipaddr.sin_addr);
1484                         /*
1485                          * locate outgoing interface; if we're the destination,
1486                          * use the incoming interface (should be same).
1487                          */
1488                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1489                             (ia = ip_rtaddr(ipaddr.sin_addr,
1490                                             &ipforward_rt[mycpuid]))
1491                                                                      == NULL) {
1492                                 type = ICMP_UNREACH;
1493                                 code = ICMP_UNREACH_HOST;
1494                                 goto bad;
1495                         }
1496                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1497                             sizeof(struct in_addr));
1498                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1499                         break;
1500
1501                 case IPOPT_TS:
1502                         if (ipstealth && pass == 0)
1503                                 break;
1504                         code = cp - (u_char *)ip;
1505                         if (optlen < 4 || optlen > 40) {
1506                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1507                                 goto bad;
1508                         }
1509                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1510                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1511                                 goto bad;
1512                         }
1513                         if (off > optlen - (int)sizeof(int32_t)) {
1514                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1515                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1516                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1517                                         goto bad;
1518                                 }
1519                                 break;
1520                         }
1521                         off--;                          /* 0 origin */
1522                         sin = (struct in_addr *)(cp + off);
1523                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1524
1525                         case IPOPT_TS_TSONLY:
1526                                 break;
1527
1528                         case IPOPT_TS_TSANDADDR:
1529                                 if (off + sizeof(n_time) +
1530                                     sizeof(struct in_addr) > optlen) {
1531                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1532                                         goto bad;
1533                                 }
1534                                 ipaddr.sin_addr = dst;
1535                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1536                                                             m->m_pkthdr.rcvif);
1537                                 if (ia == NULL)
1538                                         continue;
1539                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1540                                     sizeof(struct in_addr));
1541                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1542                                 off += sizeof(struct in_addr);
1543                                 break;
1544
1545                         case IPOPT_TS_PRESPEC:
1546                                 if (off + sizeof(n_time) +
1547                                     sizeof(struct in_addr) > optlen) {
1548                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1549                                         goto bad;
1550                                 }
1551                                 memcpy(&ipaddr.sin_addr, sin,
1552                                     sizeof(struct in_addr));
1553                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1554                                         continue;
1555                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1556                                 off += sizeof(struct in_addr);
1557                                 break;
1558
1559                         default:
1560                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1561                                 goto bad;
1562                         }
1563                         ntime = iptime();
1564                         memcpy(cp + off, &ntime, sizeof(n_time));
1565                         cp[IPOPT_OFFSET] += sizeof(n_time);
1566                 }
1567         }
1568         if (forward && ipforwarding) {
1569                 ip_forward(m, TRUE, next_hop);
1570                 return (1);
1571         }
1572         return (0);
1573 bad:
1574         icmp_error(m, type, code, 0, 0);
1575         ipstat.ips_badoptions++;
1576         return (1);
1577 }
1578
1579 /*
1580  * Given address of next destination (final or next hop),
1581  * return internet address info of interface to be used to get there.
1582  */
1583 struct in_ifaddr *
1584 ip_rtaddr(struct in_addr dst, struct route *ro)
1585 {
1586         struct sockaddr_in *sin;
1587
1588         sin = (struct sockaddr_in *)&ro->ro_dst;
1589
1590         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1591                 if (ro->ro_rt != NULL) {
1592                         RTFREE(ro->ro_rt);
1593                         ro->ro_rt = NULL;
1594                 }
1595                 sin->sin_family = AF_INET;
1596                 sin->sin_len = sizeof *sin;
1597                 sin->sin_addr = dst;
1598                 rtalloc_ign(ro, RTF_PRCLONING);
1599         }
1600
1601         if (ro->ro_rt == NULL)
1602                 return (NULL);
1603
1604         return (ifatoia(ro->ro_rt->rt_ifa));
1605 }
1606
1607 /*
1608  * Save incoming source route for use in replies,
1609  * to be picked up later by ip_srcroute if the receiver is interested.
1610  */
1611 static void
1612 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1613 {
1614         struct m_tag *mtag;
1615         struct ip_srcrt_opt *opt;
1616         unsigned olen;
1617
1618         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1619         if (mtag == NULL)
1620                 return;
1621         opt = m_tag_data(mtag);
1622
1623         olen = option[IPOPT_OLEN];
1624 #ifdef DIAGNOSTIC
1625         if (ipprintfs)
1626                 kprintf("save_rte: olen %d\n", olen);
1627 #endif
1628         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1629                 m_tag_free(mtag);
1630                 return;
1631         }
1632         bcopy(option, opt->ip_srcrt.srcopt, olen);
1633         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1634         opt->ip_srcrt.dst = dst;
1635         m_tag_prepend(m, mtag);
1636 }
1637
1638 /*
1639  * Retrieve incoming source route for use in replies,
1640  * in the same form used by setsockopt.
1641  * The first hop is placed before the options, will be removed later.
1642  */
1643 struct mbuf *
1644 ip_srcroute(struct mbuf *m0)
1645 {
1646         struct in_addr *p, *q;
1647         struct mbuf *m;
1648         struct m_tag *mtag;
1649         struct ip_srcrt_opt *opt;
1650
1651         if (m0 == NULL)
1652                 return NULL;
1653
1654         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1655         if (mtag == NULL)
1656                 return NULL;
1657         opt = m_tag_data(mtag);
1658
1659         if (opt->ip_nhops == 0)
1660                 return (NULL);
1661         m = m_get(MB_DONTWAIT, MT_HEADER);
1662         if (m == NULL)
1663                 return (NULL);
1664
1665 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1666
1667         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1668         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1669                    sizeof(struct in_addr) + OPTSIZ;
1670 #ifdef DIAGNOSTIC
1671         if (ipprintfs) {
1672                 kprintf("ip_srcroute: nhops %d mlen %d",
1673                         opt->ip_nhops, m->m_len);
1674         }
1675 #endif
1676
1677         /*
1678          * First save first hop for return route
1679          */
1680         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1681         *(mtod(m, struct in_addr *)) = *p--;
1682 #ifdef DIAGNOSTIC
1683         if (ipprintfs)
1684                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1685 #endif
1686
1687         /*
1688          * Copy option fields and padding (nop) to mbuf.
1689          */
1690         opt->ip_srcrt.nop = IPOPT_NOP;
1691         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1692         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1693             OPTSIZ);
1694         q = (struct in_addr *)(mtod(m, caddr_t) +
1695             sizeof(struct in_addr) + OPTSIZ);
1696 #undef OPTSIZ
1697         /*
1698          * Record return path as an IP source route,
1699          * reversing the path (pointers are now aligned).
1700          */
1701         while (p >= opt->ip_srcrt.route) {
1702 #ifdef DIAGNOSTIC
1703                 if (ipprintfs)
1704                         kprintf(" %x", ntohl(q->s_addr));
1705 #endif
1706                 *q++ = *p--;
1707         }
1708         /*
1709          * Last hop goes to final destination.
1710          */
1711         *q = opt->ip_srcrt.dst;
1712         m_tag_delete(m0, mtag);
1713 #ifdef DIAGNOSTIC
1714         if (ipprintfs)
1715                 kprintf(" %x\n", ntohl(q->s_addr));
1716 #endif
1717         return (m);
1718 }
1719
1720 /*
1721  * Strip out IP options.
1722  */
1723 void
1724 ip_stripoptions(struct mbuf *m)
1725 {
1726         int datalen;
1727         struct ip *ip = mtod(m, struct ip *);
1728         caddr_t opts;
1729         int optlen;
1730
1731         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1732         opts = (caddr_t)(ip + 1);
1733         datalen = m->m_len - (sizeof(struct ip) + optlen);
1734         bcopy(opts + optlen, opts, datalen);
1735         m->m_len -= optlen;
1736         if (m->m_flags & M_PKTHDR)
1737                 m->m_pkthdr.len -= optlen;
1738         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1739 }
1740
1741 u_char inetctlerrmap[PRC_NCMDS] = {
1742         0,              0,              0,              0,
1743         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1744         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1745         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1746         0,              0,              0,              0,
1747         ENOPROTOOPT,    ECONNREFUSED
1748 };
1749
1750 /*
1751  * Forward a packet.  If some error occurs return the sender
1752  * an icmp packet.  Note we can't always generate a meaningful
1753  * icmp message because icmp doesn't have a large enough repertoire
1754  * of codes and types.
1755  *
1756  * If not forwarding, just drop the packet.  This could be confusing
1757  * if ipforwarding was zero but some routing protocol was advancing
1758  * us as a gateway to somewhere.  However, we must let the routing
1759  * protocol deal with that.
1760  *
1761  * The using_srcrt parameter indicates whether the packet is being forwarded
1762  * via a source route.
1763  */
1764 void
1765 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1766 {
1767         struct ip *ip = mtod(m, struct ip *);
1768         struct sockaddr_in *ipforward_rtaddr;
1769         struct rtentry *rt;
1770         int error, type = 0, code = 0, destmtu = 0;
1771         struct mbuf *mcopy;
1772         n_long dest;
1773         struct in_addr pkt_dst;
1774         struct route *cache_rt = &ipforward_rt[mycpuid];
1775
1776         dest = INADDR_ANY;
1777         /*
1778          * Cache the destination address of the packet; this may be
1779          * changed by use of 'ipfw fwd'.
1780          */
1781         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1782
1783 #ifdef DIAGNOSTIC
1784         if (ipprintfs)
1785                 kprintf("forward: src %x dst %x ttl %x\n",
1786                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1787 #endif
1788
1789         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1790                 ipstat.ips_cantforward++;
1791                 m_freem(m);
1792                 return;
1793         }
1794         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1795                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1796                 return;
1797         }
1798
1799         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1800         if (cache_rt->ro_rt == NULL ||
1801             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1802                 if (cache_rt->ro_rt != NULL) {
1803                         RTFREE(cache_rt->ro_rt);
1804                         cache_rt->ro_rt = NULL;
1805                 }
1806                 ipforward_rtaddr->sin_family = AF_INET;
1807                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1808                 ipforward_rtaddr->sin_addr = pkt_dst;
1809                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1810                 if (cache_rt->ro_rt == NULL) {
1811                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1812                         return;
1813                 }
1814         }
1815         rt = cache_rt->ro_rt;
1816
1817         /*
1818          * Save the IP header and at most 8 bytes of the payload,
1819          * in case we need to generate an ICMP message to the src.
1820          *
1821          * XXX this can be optimized a lot by saving the data in a local
1822          * buffer on the stack (72 bytes at most), and only allocating the
1823          * mbuf if really necessary. The vast majority of the packets
1824          * are forwarded without having to send an ICMP back (either
1825          * because unnecessary, or because rate limited), so we are
1826          * really we are wasting a lot of work here.
1827          *
1828          * We don't use m_copy() because it might return a reference
1829          * to a shared cluster. Both this function and ip_output()
1830          * assume exclusive access to the IP header in `m', so any
1831          * data in a cluster may change before we reach icmp_error().
1832          */
1833         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1834         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1835                 /*
1836                  * It's probably ok if the pkthdr dup fails (because
1837                  * the deep copy of the tag chain failed), but for now
1838                  * be conservative and just discard the copy since
1839                  * code below may some day want the tags.
1840                  */
1841                 m_free(mcopy);
1842                 mcopy = NULL;
1843         }
1844         if (mcopy != NULL) {
1845                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1846                     (int)ip->ip_len);
1847                 mcopy->m_pkthdr.len = mcopy->m_len;
1848                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1849         }
1850
1851         if (!ipstealth)
1852                 ip->ip_ttl -= IPTTLDEC;
1853
1854         /*
1855          * If forwarding packet using same interface that it came in on,
1856          * perhaps should send a redirect to sender to shortcut a hop.
1857          * Only send redirect if source is sending directly to us,
1858          * and if packet was not source routed (or has any options).
1859          * Also, don't send redirect if forwarding using a default route
1860          * or a route modified by a redirect.
1861          */
1862         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1863             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1864             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1865             ipsendredirects && !using_srcrt && next_hop == NULL) {
1866                 u_long src = ntohl(ip->ip_src.s_addr);
1867                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1868
1869                 if (rt_ifa != NULL &&
1870                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1871                         if (rt->rt_flags & RTF_GATEWAY)
1872                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1873                         else
1874                                 dest = pkt_dst.s_addr;
1875                         /*
1876                          * Router requirements says to only send
1877                          * host redirects.
1878                          */
1879                         type = ICMP_REDIRECT;
1880                         code = ICMP_REDIRECT_HOST;
1881 #ifdef DIAGNOSTIC
1882                         if (ipprintfs)
1883                                 kprintf("redirect (%d) to %x\n", code, dest);
1884 #endif
1885                 }
1886         }
1887
1888         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1889         if (error == 0) {
1890                 ipstat.ips_forward++;
1891                 if (type == 0) {
1892                         if (mcopy) {
1893                                 ipflow_create(cache_rt, mcopy);
1894                                 m_freem(mcopy);
1895                         }
1896                         return;         /* most common case */
1897                 } else {
1898                         ipstat.ips_redirectsent++;
1899                 }
1900         } else {
1901                 ipstat.ips_cantforward++;
1902         }
1903
1904         if (mcopy == NULL)
1905                 return;
1906
1907         /*
1908          * Send ICMP message.
1909          */
1910
1911         switch (error) {
1912
1913         case 0:                         /* forwarded, but need redirect */
1914                 /* type, code set above */
1915                 break;
1916
1917         case ENETUNREACH:               /* shouldn't happen, checked above */
1918         case EHOSTUNREACH:
1919         case ENETDOWN:
1920         case EHOSTDOWN:
1921         default:
1922                 type = ICMP_UNREACH;
1923                 code = ICMP_UNREACH_HOST;
1924                 break;
1925
1926         case EMSGSIZE:
1927                 type = ICMP_UNREACH;
1928                 code = ICMP_UNREACH_NEEDFRAG;
1929 #ifdef IPSEC
1930                 /*
1931                  * If the packet is routed over IPsec tunnel, tell the
1932                  * originator the tunnel MTU.
1933                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1934                  * XXX quickhack!!!
1935                  */
1936                 if (cache_rt->ro_rt != NULL) {
1937                         struct secpolicy *sp = NULL;
1938                         int ipsecerror;
1939                         int ipsechdr;
1940                         struct route *ro;
1941
1942                         sp = ipsec4_getpolicybyaddr(mcopy,
1943                                                     IPSEC_DIR_OUTBOUND,
1944                                                     IP_FORWARDING,
1945                                                     &ipsecerror);
1946
1947                         if (sp == NULL)
1948                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1949                         else {
1950                                 /* count IPsec header size */
1951                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1952                                                          IPSEC_DIR_OUTBOUND,
1953                                                          NULL);
1954
1955                                 /*
1956                                  * find the correct route for outer IPv4
1957                                  * header, compute tunnel MTU.
1958                                  *
1959                                  */
1960                                 if (sp->req != NULL && sp->req->sav != NULL &&
1961                                     sp->req->sav->sah != NULL) {
1962                                         ro = &sp->req->sav->sah->sa_route;
1963                                         if (ro->ro_rt != NULL &&
1964                                             ro->ro_rt->rt_ifp != NULL) {
1965                                                 destmtu =
1966                                                     ro->ro_rt->rt_ifp->if_mtu;
1967                                                 destmtu -= ipsechdr;
1968                                         }
1969                                 }
1970
1971                                 key_freesp(sp);
1972                         }
1973                 }
1974 #elif FAST_IPSEC
1975                 /*
1976                  * If the packet is routed over IPsec tunnel, tell the
1977                  * originator the tunnel MTU.
1978                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1979                  * XXX quickhack!!!
1980                  */
1981                 if (cache_rt->ro_rt != NULL) {
1982                         struct secpolicy *sp = NULL;
1983                         int ipsecerror;
1984                         int ipsechdr;
1985                         struct route *ro;
1986
1987                         sp = ipsec_getpolicybyaddr(mcopy,
1988                                                    IPSEC_DIR_OUTBOUND,
1989                                                    IP_FORWARDING,
1990                                                    &ipsecerror);
1991
1992                         if (sp == NULL)
1993                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1994                         else {
1995                                 /* count IPsec header size */
1996                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1997                                                          IPSEC_DIR_OUTBOUND,
1998                                                          NULL);
1999
2000                                 /*
2001                                  * find the correct route for outer IPv4
2002                                  * header, compute tunnel MTU.
2003                                  */
2004
2005                                 if (sp->req != NULL &&
2006                                     sp->req->sav != NULL &&
2007                                     sp->req->sav->sah != NULL) {
2008                                         ro = &sp->req->sav->sah->sa_route;
2009                                         if (ro->ro_rt != NULL &&
2010                                             ro->ro_rt->rt_ifp != NULL) {
2011                                                 destmtu =
2012                                                     ro->ro_rt->rt_ifp->if_mtu;
2013                                                 destmtu -= ipsechdr;
2014                                         }
2015                                 }
2016
2017                                 KEY_FREESP(&sp);
2018                         }
2019                 }
2020 #else /* !IPSEC && !FAST_IPSEC */
2021                 if (cache_rt->ro_rt != NULL)
2022                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2023 #endif /*IPSEC*/
2024                 ipstat.ips_cantfrag++;
2025                 break;
2026
2027         case ENOBUFS:
2028                 /*
2029                  * A router should not generate ICMP_SOURCEQUENCH as
2030                  * required in RFC1812 Requirements for IP Version 4 Routers.
2031                  * Source quench could be a big problem under DoS attacks,
2032                  * or if the underlying interface is rate-limited.
2033                  * Those who need source quench packets may re-enable them
2034                  * via the net.inet.ip.sendsourcequench sysctl.
2035                  */
2036                 if (!ip_sendsourcequench) {
2037                         m_freem(mcopy);
2038                         return;
2039                 } else {
2040                         type = ICMP_SOURCEQUENCH;
2041                         code = 0;
2042                 }
2043                 break;
2044
2045         case EACCES:                    /* ipfw denied packet */
2046                 m_freem(mcopy);
2047                 return;
2048         }
2049         icmp_error(mcopy, type, code, dest, destmtu);
2050 }
2051
2052 void
2053 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2054                struct mbuf *m)
2055 {
2056         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2057                 struct timeval tv;
2058
2059                 microtime(&tv);
2060                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2061                     SCM_TIMESTAMP, SOL_SOCKET);
2062                 if (*mp)
2063                         mp = &(*mp)->m_next;
2064         }
2065         if (inp->inp_flags & INP_RECVDSTADDR) {
2066                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2067                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2068                 if (*mp)
2069                         mp = &(*mp)->m_next;
2070         }
2071         if (inp->inp_flags & INP_RECVTTL) {
2072                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2073                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2074                 if (*mp)
2075                         mp = &(*mp)->m_next;
2076         }
2077 #ifdef notyet
2078         /* XXX
2079          * Moving these out of udp_input() made them even more broken
2080          * than they already were.
2081          */
2082         /* options were tossed already */
2083         if (inp->inp_flags & INP_RECVOPTS) {
2084                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2085                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2086                 if (*mp)
2087                         mp = &(*mp)->m_next;
2088         }
2089         /* ip_srcroute doesn't do what we want here, need to fix */
2090         if (inp->inp_flags & INP_RECVRETOPTS) {
2091                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2092                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2093                 if (*mp)
2094                         mp = &(*mp)->m_next;
2095         }
2096 #endif
2097         if (inp->inp_flags & INP_RECVIF) {
2098                 struct ifnet *ifp;
2099                 struct sdlbuf {
2100                         struct sockaddr_dl sdl;
2101                         u_char  pad[32];
2102                 } sdlbuf;
2103                 struct sockaddr_dl *sdp;
2104                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2105
2106                 if (((ifp = m->m_pkthdr.rcvif)) &&
2107                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2108                         sdp = IF_LLSOCKADDR(ifp);
2109                         /*
2110                          * Change our mind and don't try copy.
2111                          */
2112                         if ((sdp->sdl_family != AF_LINK) ||
2113                             (sdp->sdl_len > sizeof(sdlbuf))) {
2114                                 goto makedummy;
2115                         }
2116                         bcopy(sdp, sdl2, sdp->sdl_len);
2117                 } else {
2118 makedummy:
2119                         sdl2->sdl_len =
2120                             offsetof(struct sockaddr_dl, sdl_data[0]);
2121                         sdl2->sdl_family = AF_LINK;
2122                         sdl2->sdl_index = 0;
2123                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2124                 }
2125                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2126                         IP_RECVIF, IPPROTO_IP);
2127                 if (*mp)
2128                         mp = &(*mp)->m_next;
2129         }
2130 }
2131
2132 /*
2133  * XXX these routines are called from the upper part of the kernel.
2134  *
2135  * They could also be moved to ip_mroute.c, since all the RSVP
2136  *  handling is done there already.
2137  */
2138 int
2139 ip_rsvp_init(struct socket *so)
2140 {
2141         if (so->so_type != SOCK_RAW ||
2142             so->so_proto->pr_protocol != IPPROTO_RSVP)
2143                 return EOPNOTSUPP;
2144
2145         if (ip_rsvpd != NULL)
2146                 return EADDRINUSE;
2147
2148         ip_rsvpd = so;
2149         /*
2150          * This may seem silly, but we need to be sure we don't over-increment
2151          * the RSVP counter, in case something slips up.
2152          */
2153         if (!ip_rsvp_on) {
2154                 ip_rsvp_on = 1;
2155                 rsvp_on++;
2156         }
2157
2158         return 0;
2159 }
2160
2161 int
2162 ip_rsvp_done(void)
2163 {
2164         ip_rsvpd = NULL;
2165         /*
2166          * This may seem silly, but we need to be sure we don't over-decrement
2167          * the RSVP counter, in case something slips up.
2168          */
2169         if (ip_rsvp_on) {
2170                 ip_rsvp_on = 0;
2171                 rsvp_on--;
2172         }
2173         return 0;
2174 }
2175
2176 void
2177 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2178 {
2179         int off, proto;
2180         __va_list ap;
2181
2182         __va_start(ap, m);
2183         off = __va_arg(ap, int);
2184         proto = __va_arg(ap, int);
2185         __va_end(ap);
2186
2187         if (rsvp_input_p) { /* call the real one if loaded */
2188                 rsvp_input_p(m, off, proto);
2189                 return;
2190         }
2191
2192         /* Can still get packets with rsvp_on = 0 if there is a local member
2193          * of the group to which the RSVP packet is addressed.  But in this
2194          * case we want to throw the packet away.
2195          */
2196
2197         if (!rsvp_on) {
2198                 m_freem(m);
2199                 return;
2200         }
2201
2202         if (ip_rsvpd != NULL) {
2203                 rip_input(m, off, proto);
2204                 return;
2205         }
2206         /* Drop the packet */
2207         m_freem(m);
2208 }