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 1.114 2008/10/28 03:07:28 sephe Exp $
  69  */
  70
  71 #define _IP_VHL
  72
  73 #include "opt_bootp.h"
  74 #include "opt_ipfw.h"
  75 #include "opt_ipdn.h"
  76 #include "opt_ipdivert.h"
  77 #include "opt_ipfilter.h"
  78 #include "opt_ipstealth.h"
  79 #include "opt_ipsec.h"
  80
  81 #include <sys/param.h>
  82 #include <sys/systm.h>
  83 #include <sys/mbuf.h>
  84 #include <sys/malloc.h>
  85 #include <sys/mpipe.h>
  86 #include <sys/domain.h>
  87 #include <sys/protosw.h>
  88 #include <sys/socket.h>
  89 #include <sys/time.h>
  90 #include <sys/globaldata.h>
  91 #include <sys/thread.h>
  92 #include <sys/kernel.h>
  93 #include <sys/syslog.h>
  94 #include <sys/sysctl.h>
  95 #include <sys/in_cksum.h>
  96 #include <sys/lock.h>
  97
  98 #include <machine/stdarg.h>
  99
 100 #include <net/if.h>
 101 #include <net/if_types.h>
 102 #include <net/if_var.h>
 103 #include <net/if_dl.h>
 104 #include <net/pfil.h>
 105 #include <net/route.h>
 106 #include <net/netisr.h>
 107
 108 #include <netinet/in.h>
 109 #include <netinet/in_systm.h>
 110 #include <netinet/in_var.h>
 111 #include <netinet/ip.h>
 112 #include <netinet/in_pcb.h>
 113 #include <netinet/ip_var.h>
 114 #include <netinet/ip_icmp.h>
 115 #include <netinet/ip_divert.h>
 116 #include <netinet/ip_flow.h>
 117
 118 #include <sys/thread2.h>
 119 #include <sys/msgport2.h>
 120 #include <net/netmsg2.h>
 121
 122 #include <sys/socketvar.h>
 123
 124 #include <net/ipfw/ip_fw.h>
 125 #include <net/dummynet/ip_dummynet.h>
 126
 127 #ifdef IPSEC
 128 #include <netinet6/ipsec.h>
 129 #include <netproto/key/key.h>
 130 #endif
 131
 132 #ifdef FAST_IPSEC
 133 #include <netproto/ipsec/ipsec.h>
 134 #include <netproto/ipsec/key.h>
 135 #endif
 136
 137 int rsvp_on = 0;
 138 static int ip_rsvp_on;
 139 struct socket *ip_rsvpd;
 140
 141 int ip_mpsafe = 0;
 142 TUNABLE_INT("net.inet.ip.mpsafe", &ip_mpsafe);
 143
 144 int ipforwarding = 0;
 145 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 146     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 147
 148 static int ipsendredirects = 1; /* XXX */
 149 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 150     &ipsendredirects, 0, "Enable sending IP redirects");
 151
 152 int ip_defttl = IPDEFTTL;
 153 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 154     &ip_defttl, 0, "Maximum TTL on IP packets");
 155
 156 static int ip_dosourceroute = 0;
 157 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 158     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 159
 160 static int ip_acceptsourceroute = 0;
 161 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 162     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 163     "Enable accepting source routed IP packets");
 164
 165 static int ip_keepfaith = 0;
 166 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 167     &ip_keepfaith, 0,
 168     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 169
 170 static int nipq = 0;    /* total # of reass queues */
 171 static int maxnipq;
 172 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 173     &maxnipq, 0,
 174     "Maximum number of IPv4 fragment reassembly queue entries");
 175
 176 static int maxfragsperpacket;
 177 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 178     &maxfragsperpacket, 0,
 179     "Maximum number of IPv4 fragments allowed per packet");
 180
 181 static int ip_sendsourcequench = 0;
 182 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 183     &ip_sendsourcequench, 0,
 184     "Enable the transmission of source quench packets");
 185
 186 int ip_do_randomid = 1;
 187 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 188     &ip_do_randomid, 0,
 189     "Assign random ip_id values");
 190 /*
 191  * XXX - Setting ip_checkinterface mostly implements the receive side of
 192  * the Strong ES model described in RFC 1122, but since the routing table
 193  * and transmit implementation do not implement the Strong ES model,
 194  * setting this to 1 results in an odd hybrid.
 195  *
 196  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 197  * to translate the destination address to another local interface.
 198  *
 199  * XXX - ip_checkinterface must be disabled if you add IP aliases
 200  * to the loopback interface instead of the interface where the
 201  * packets for those addresses are received.
 202  */
 203 static int ip_checkinterface = 0;
 204 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 205     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 206
 207 #ifdef DIAGNOSTIC
 208 static int ipprintfs = 0;
 209 #endif
 210
 211 extern  int udp_mpsafe_proto;
 212 extern  int tcp_mpsafe_proto;
 213
 214 extern  struct domain inetdomain;
 215 extern  struct protosw inetsw[];
 216 u_char  ip_protox[IPPROTO_MAX];
 217 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 218 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 219                                                 /* inet addr hash table */
 220 u_long  in_ifaddrhmask;                         /* mask for hash table */
 221
 222 struct ip_stats ipstats_percpu[MAXCPU];
 223 #ifdef SMP
 224 static int
 225 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 226 {
 227         int cpu, error = 0;
 228
 229         for (cpu = 0; cpu < ncpus; ++cpu) {
 230                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 231                                         sizeof(struct ip_stats))))
 232                         break;
 233                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 234                                        sizeof(struct ip_stats))))
 235                         break;
 236         }
 237
 238         return (error);
 239 }
 240 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 241     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 242 #else
 243 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 244     &ipstat, ip_stats, "IP statistics");
 245 #endif
 246
 247 /* Packet reassembly stuff */
 248 #define IPREASS_NHASH_LOG2      6
 249 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 250 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 251 #define IPREASS_HASH(x,y)                                               \
 252     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 253
 254 static struct ipq ipq[IPREASS_NHASH];
 255
 256 #ifdef IPCTL_DEFMTU
 257 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 258     &ip_mtu, 0, "Default MTU");
 259 #endif
 260
 261 #ifdef IPSTEALTH
 262 static int ipstealth = 0;
 263 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 264 #else
 265 static const int ipstealth = 0;
 266 #endif
 267
 268 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
 269
 270 struct pfil_head inet_pfil_hook;
 271
 272 /*
 273  * struct ip_srcrt_opt is used to store packet state while it travels
 274  * through the stack.
 275  *
 276  * XXX Note that the code even makes assumptions on the size and
 277  * alignment of fields inside struct ip_srcrt so e.g. adding some
 278  * fields will break the code.  This needs to be fixed.
 279  *
 280  * We need to save the IP options in case a protocol wants to respond
 281  * to an incoming packet over the same route if the packet got here
 282  * using IP source routing.  This allows connection establishment and
 283  * maintenance when the remote end is on a network that is not known
 284  * to us.
 285  */
 286 struct ip_srcrt {
 287         struct  in_addr dst;                    /* final destination */
 288         char    nop;                            /* one NOP to align */
 289         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 290         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 291 };
 292
 293 struct ip_srcrt_opt {
 294         int             ip_nhops;
 295         struct ip_srcrt ip_srcrt;
 296 };
 297
 298 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 299 static struct malloc_pipe ipq_mpipe;
 300
 301 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
 302 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 303 static void             ip_freef(struct ipq *);
 304 static void             ip_input_handler(struct netmsg *);
 305
 306 /*
 307  * IP initialization: fill in IP protocol switch table.
 308  * All protocols not implemented in kernel go to raw IP protocol handler.
 309  */
 310 void
 311 ip_init(void)
 312 {
 313         struct protosw *pr;
 314         uint32_t flags;
 315         int i;
 316 #ifdef SMP
 317         int cpu;
 318 #endif
 319
 320         /*
 321          * Make sure we can handle a reasonable number of fragments but
 322          * cap it at 4000 (XXX).
 323          */
 324         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 325                     IFQ_MAXLEN, 4000, 0, NULL);
 326         for (i = 0; i < ncpus; ++i) {
 327                 TAILQ_INIT(&in_ifaddrheads[i]);
 328                 in_ifaddrhashtbls[i] =
 329                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 330         }
 331         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 332         if (pr == NULL)
 333                 panic("ip_init");
 334         for (i = 0; i < IPPROTO_MAX; i++)
 335                 ip_protox[i] = pr - inetsw;
 336         for (pr = inetdomain.dom_protosw;
 337              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
 338                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
 339                         if (pr->pr_protocol != IPPROTO_RAW)
 340                                 ip_protox[pr->pr_protocol] = pr - inetsw;
 341
 342                         /* XXX */
 343                         switch (pr->pr_protocol) {
 344                         case IPPROTO_TCP:
 345                                 if (tcp_mpsafe_proto)
 346                                         pr->pr_flags |= PR_MPSAFE;
 347                                 break;
 348
 349                         case IPPROTO_UDP:
 350                                 if (udp_mpsafe_proto)
 351                                         pr->pr_flags |= PR_MPSAFE;
 352                                 break;
 353                         }
 354                 }
 355         }
 356
 357         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 358         inet_pfil_hook.ph_af = AF_INET;
 359         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 360                 kprintf("%s: WARNING: unable to register pfil hook, "
 361                         "error %d\n", __func__, i);
 362         }
 363
 364         for (i = 0; i < IPREASS_NHASH; i++)
 365                 ipq[i].next = ipq[i].prev = &ipq[i];
 366
 367         maxnipq = nmbclusters / 32;
 368         maxfragsperpacket = 16;
 369
 370         ip_id = time_second & 0xffff;
 371
 372         /*
 373          * Initialize IP statistics counters for each CPU.
 374          *
 375          */
 376 #ifdef SMP
 377         for (cpu = 0; cpu < ncpus; ++cpu) {
 378                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 379         }
 380 #else
 381         bzero(&ipstat, sizeof(struct ip_stats));
 382 #endif
 383
 384 #if defined(IPSEC) || defined(FAST_IPSEC)
 385         /* XXX IPSEC is not MPSAFE yet */
 386         flags = NETISR_FLAG_NOTMPSAFE;
 387 #else
 388         if (ip_mpsafe) {
 389                 kprintf("ip: MPSAFE\n");
 390                 flags = NETISR_FLAG_MPSAFE;
 391         } else {
 392                 flags = NETISR_FLAG_NOTMPSAFE;
 393         }
 394 #endif
 395         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler, flags);
 396 }
 397
 398 /*
 399  * XXX watch out this one. It is perhaps used as a cache for
 400  * the most recently used route ? it is cleared in in_addroute()
 401  * when a new route is successfully created.
 402  */
 403 struct route ipforward_rt[MAXCPU];
 404
 405 /* Do transport protocol processing. */
 406 static void
 407 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 408 {
 409         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 410
 411         /*
 412          * Switch out to protocol's input routine.
 413          */
 414         PR_GET_MPLOCK(pr);
 415         pr->pr_input(m, hlen, ip->ip_p);
 416         PR_REL_MPLOCK(pr);
 417 }
 418
 419 static void
 420 transport_processing_handler(netmsg_t netmsg)
 421 {
 422         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 423         struct ip *ip;
 424         int hlen;
 425
 426         ip = mtod(pmsg->nm_packet, struct ip *);
 427         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 428
 429         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 430         /* netmsg was embedded in the mbuf, do not reply! */
 431 }
 432
 433 static void
 434 ip_input_handler(struct netmsg *msg0)
 435 {
 436         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 437
 438         ip_input(m);
 439         /* msg0 was embedded in the mbuf, do not reply! */
 440 }
 441
 442 /*
 443  * IP input routine.  Checksum and byte swap header.  If fragmented
 444  * try to reassemble.  Process options.  Pass to next level.
 445  */
 446 void
 447 ip_input(struct mbuf *m)
 448 {
 449         struct ip *ip;
 450         struct in_ifaddr *ia = NULL;
 451         struct in_ifaddr_container *iac;
 452         int hlen, checkif;
 453         u_short sum;
 454         struct in_addr pkt_dst;
 455         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 456         boolean_t needredispatch = FALSE;
 457         struct in_addr odst;                    /* original dst address(NAT) */
 458         struct m_tag *mtag;
 459         struct sockaddr_in *next_hop = NULL;
 460 #ifdef FAST_IPSEC
 461         struct tdb_ident *tdbi;
 462         struct secpolicy *sp;
 463         int error;
 464 #endif
 465
 466         M_ASSERTPKTHDR(m);
 467
 468         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 469                 /* Next hop */
 470                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 471                 KKASSERT(mtag != NULL);
 472                 next_hop = m_tag_data(mtag);
 473         }
 474
 475         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 476                 /* dummynet already filtered us */
 477                 ip = mtod(m, struct ip *);
 478                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 479                 goto iphack;
 480         }
 481
 482         ipstat.ips_total++;
 483
 484         /* length checks already done in ip_mport() */
 485         KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
 486         ip = mtod(m, struct ip *);
 487
 488         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 489                 ipstat.ips_badvers++;
 490                 goto bad;
 491         }
 492
 493         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 494         /* length checks already done in ip_mport() */
 495         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 496         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 497
 498         /* 127/8 must not appear on wire - RFC1122 */
 499         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 500             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 501                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 502                         ipstat.ips_badaddr++;
 503                         goto bad;
 504                 }
 505         }
 506
 507         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 508                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 509         } else {
 510                 if (hlen == sizeof(struct ip))
 511                         sum = in_cksum_hdr(ip);
 512                 else
 513                         sum = in_cksum(m, hlen);
 514         }
 515         if (sum != 0) {
 516                 ipstat.ips_badsum++;
 517                 goto bad;
 518         }
 519
 520 #ifdef ALTQ
 521         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 522                 /* packet is dropped by traffic conditioner */
 523                 return;
 524         }
 525 #endif
 526         /*
 527          * Convert fields to host representation.
 528          */
 529         ip->ip_len = ntohs(ip->ip_len);
 530         if (ip->ip_len < hlen) {
 531                 ipstat.ips_badlen++;
 532                 goto bad;
 533         }
 534         ip->ip_off = ntohs(ip->ip_off);
 535
 536         /*
 537          * Check that the amount of data in the buffers
 538          * is as at least much as the IP header would have us expect.
 539          * Trim mbufs if longer than we expect.
 540          * Drop packet if shorter than we expect.
 541          */
 542         if (m->m_pkthdr.len < ip->ip_len) {
 543                 ipstat.ips_tooshort++;
 544                 goto bad;
 545         }
 546         if (m->m_pkthdr.len > ip->ip_len) {
 547                 if (m->m_len == m->m_pkthdr.len) {
 548                         m->m_len = ip->ip_len;
 549                         m->m_pkthdr.len = ip->ip_len;
 550                 } else {
 551                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 552                 }
 553         }
 554 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 555         /*
 556          * Bypass packet filtering for packets from a tunnel (gif).
 557          */
 558         if (ipsec_gethist(m, NULL))
 559                 goto pass;
 560 #endif
 561
 562         /*
 563          * IpHack's section.
 564          * Right now when no processing on packet has done
 565          * and it is still fresh out of network we do our black
 566          * deals with it.
 567          * - Firewall: deny/allow/divert
 568          * - Xlate: translate packet's addr/port (NAT).
 569          * - Pipe: pass pkt through dummynet.
 570          * - Wrap: fake packet's addr/port <unimpl.>
 571          * - Encapsulate: put it in another IP and send out. <unimp.>
 572          */
 573
 574 iphack:
 575         /*
 576          * If we've been forwarded from the output side, then
 577          * skip the firewall a second time
 578          */
 579         if (next_hop != NULL)
 580                 goto ours;
 581
 582         /* No pfil hooks */
 583         if (!pfil_has_hooks(&inet_pfil_hook)) {
 584                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 585                         /*
 586                          * Strip dummynet tags from stranded packets
 587                          */
 588                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 589                         KKASSERT(mtag != NULL);
 590                         m_tag_delete(m, mtag);
 591                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 592                 }
 593                 goto pass;
 594         }
 595
 596         /*
 597          * Run through list of hooks for input packets.
 598          *
 599          * NB: Beware of the destination address changing (e.g.
 600          *     by NAT rewriting). When this happens, tell
 601          *     ip_forward to do the right thing.
 602          */
 603         odst = ip->ip_dst;
 604         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 605                 return;
 606         if (m == NULL)  /* consumed by filter */
 607                 return;
 608         ip = mtod(m, struct ip *);
 609         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 610         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 611
 612         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 613                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 614                 KKASSERT(mtag != NULL);
 615                 next_hop = m_tag_data(mtag);
 616         }
 617         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 618                 ip_dn_queue(m);
 619                 return;
 620         }
 621         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 622                 needredispatch = TRUE;
 623                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 624         }
 625 pass:
 626         /*
 627          * Process options and, if not destined for us,
 628          * ship it on.  ip_dooptions returns 1 when an
 629          * error was detected (causing an icmp message
 630          * to be sent and the original packet to be freed).
 631          */
 632         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 633                 return;
 634
 635         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 636          * matter if it is destined to another node, or whether it is
 637          * a multicast one, RSVP wants it! and prevents it from being forwarded
 638          * anywhere else. Also checks if the rsvp daemon is running before
 639          * grabbing the packet.
 640          */
 641         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 642                 goto ours;
 643
 644         /*
 645          * Check our list of addresses, to see if the packet is for us.
 646          * If we don't have any addresses, assume any unicast packet
 647          * we receive might be for us (and let the upper layers deal
 648          * with it).
 649          */
 650         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 651             !(m->m_flags & (M_MCAST | M_BCAST)))
 652                 goto ours;
 653
 654         /*
 655          * Cache the destination address of the packet; this may be
 656          * changed by use of 'ipfw fwd'.
 657          */
 658         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 659
 660         /*
 661          * Enable a consistency check between the destination address
 662          * and the arrival interface for a unicast packet (the RFC 1122
 663          * strong ES model) if IP forwarding is disabled and the packet
 664          * is not locally generated and the packet is not subject to
 665          * 'ipfw fwd'.
 666          *
 667          * XXX - Checking also should be disabled if the destination
 668          * address is ipnat'ed to a different interface.
 669          *
 670          * XXX - Checking is incompatible with IP aliases added
 671          * to the loopback interface instead of the interface where
 672          * the packets are received.
 673          */
 674         checkif = ip_checkinterface &&
 675                   !ipforwarding &&
 676                   m->m_pkthdr.rcvif != NULL &&
 677                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 678                   next_hop == NULL;
 679
 680         /*
 681          * Check for exact addresses in the hash bucket.
 682          */
 683         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 684                 ia = iac->ia;
 685
 686                 /*
 687                  * If the address matches, verify that the packet
 688                  * arrived via the correct interface if checking is
 689                  * enabled.
 690                  */
 691                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 692                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 693                         goto ours;
 694         }
 695         ia = NULL;
 696
 697         /*
 698          * Check for broadcast addresses.
 699          *
 700          * Only accept broadcast packets that arrive via the matching
 701          * interface.  Reception of forwarded directed broadcasts would
 702          * be handled via ip_forward() and ether_output() with the loopback
 703          * into the stack for SIMPLEX interfaces handled by ether_output().
 704          */
 705         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 706                 struct ifaddr_container *ifac;
 707
 708                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 709                               ifa_link) {
 710                         struct ifaddr *ifa = ifac->ifa;
 711
 712                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 713                                 continue;
 714                         if (ifa->ifa_addr->sa_family != AF_INET)
 715                                 continue;
 716                         ia = ifatoia(ifa);
 717                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 718                                                                 pkt_dst.s_addr)
 719                                 goto ours;
 720                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 721                                 goto ours;
 722 #ifdef BOOTP_COMPAT
 723                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 724                                 goto ours;
 725 #endif
 726                 }
 727         }
 728         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 729                 struct in_multi *inm;
 730
 731                 /* XXX Multicast is not MPSAFE yet */
 732                 get_mplock();
 733
 734                 if (ip_mrouter != NULL) {
 735                         /*
 736                          * If we are acting as a multicast router, all
 737                          * incoming multicast packets are passed to the
 738                          * kernel-level multicast forwarding function.
 739                          * The packet is returned (relatively) intact; if
 740                          * ip_mforward() returns a non-zero value, the packet
 741                          * must be discarded, else it may be accepted below.
 742                          */
 743                         if (ip_mforward != NULL &&
 744                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 745                                 rel_mplock();
 746                                 ipstat.ips_cantforward++;
 747                                 m_freem(m);
 748                                 return;
 749                         }
 750
 751                         /*
 752                          * The process-level routing daemon needs to receive
 753                          * all multicast IGMP packets, whether or not this
 754                          * host belongs to their destination groups.
 755                          */
 756                         if (ip->ip_p == IPPROTO_IGMP) {
 757                                 rel_mplock();
 758                                 goto ours;
 759                         }
 760                         ipstat.ips_forward++;
 761                 }
 762                 /*
 763                  * See if we belong to the destination multicast group on the
 764                  * arrival interface.
 765                  */
 766                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 767                 if (inm == NULL) {
 768                         rel_mplock();
 769                         ipstat.ips_notmember++;
 770                         m_freem(m);
 771                         return;
 772                 }
 773
 774                 rel_mplock();
 775                 goto ours;
 776         }
 777         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 778                 goto ours;
 779         if (ip->ip_dst.s_addr == INADDR_ANY)
 780                 goto ours;
 781
 782         /*
 783          * FAITH(Firewall Aided Internet Translator)
 784          */
 785         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 786                 if (ip_keepfaith) {
 787                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 788                                 goto ours;
 789                 }
 790                 m_freem(m);
 791                 return;
 792         }
 793
 794         /*
 795          * Not for us; forward if possible and desirable.
 796          */
 797         if (!ipforwarding) {
 798                 ipstat.ips_cantforward++;
 799                 m_freem(m);
 800         } else {
 801 #ifdef IPSEC
 802                 /*
 803                  * Enforce inbound IPsec SPD.
 804                  */
 805                 if (ipsec4_in_reject(m, NULL)) {
 806                         ipsecstat.in_polvio++;
 807                         goto bad;
 808                 }
 809 #endif
 810 #ifdef FAST_IPSEC
 811                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 812                 crit_enter();
 813                 if (mtag != NULL) {
 814                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 815                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 816                 } else {
 817                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 818                                                    IP_FORWARDING, &error);
 819                 }
 820                 if (sp == NULL) {       /* NB: can happen if error */
 821                         crit_exit();
 822                         /*XXX error stat???*/
 823                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 824                         goto bad;
 825                 }
 826
 827                 /*
 828                  * Check security policy against packet attributes.
 829                  */
 830                 error = ipsec_in_reject(sp, m);
 831                 KEY_FREESP(&sp);
 832                 crit_exit();
 833                 if (error) {
 834                         ipstat.ips_cantforward++;
 835                         goto bad;
 836                 }
 837 #endif
 838                 ip_forward(m, using_srcrt, next_hop);
 839         }
 840         return;
 841
 842 ours:
 843
 844         /*
 845          * IPSTEALTH: Process non-routing options only
 846          * if the packet is destined for us.
 847          */
 848         if (ipstealth &&
 849             hlen > sizeof(struct ip) &&
 850             ip_dooptions(m, 1, next_hop))
 851                 return;
 852
 853         /* Count the packet in the ip address stats */
 854         if (ia != NULL) {
 855                 ia->ia_ifa.if_ipackets++;
 856                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 857         }
 858
 859         /*
 860          * If offset or IP_MF are set, must reassemble.
 861          * Otherwise, nothing need be done.
 862          * (We could look in the reassembly queue to see
 863          * if the packet was previously fragmented,
 864          * but it's not worth the time; just let them time out.)
 865          */
 866         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 867                 /*
 868                  * Attempt reassembly; if it succeeds, proceed.
 869                  * ip_reass() will return a different mbuf.
 870                  */
 871                 m = ip_reass(m);
 872                 if (m == NULL)
 873                         return;
 874                 ip = mtod(m, struct ip *);
 875
 876                 /* Get the header length of the reassembled packet */
 877                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 878
 879                 needredispatch = TRUE;
 880         } else {
 881                 ip->ip_len -= hlen;
 882         }
 883
 884 #ifdef IPSEC
 885         /*
 886          * enforce IPsec policy checking if we are seeing last header.
 887          * note that we do not visit this with protocols with pcb layer
 888          * code - like udp/tcp/raw ip.
 889          */
 890         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 891             ipsec4_in_reject(m, NULL)) {
 892                 ipsecstat.in_polvio++;
 893                 goto bad;
 894         }
 895 #endif
 896 #if FAST_IPSEC
 897         /*
 898          * enforce IPsec policy checking if we are seeing last header.
 899          * note that we do not visit this with protocols with pcb layer
 900          * code - like udp/tcp/raw ip.
 901          */
 902         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 903                 /*
 904                  * Check if the packet has already had IPsec processing
 905                  * done.  If so, then just pass it along.  This tag gets
 906                  * set during AH, ESP, etc. input handling, before the
 907                  * packet is returned to the ip input queue for delivery.
 908                  */
 909                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 910                 crit_enter();
 911                 if (mtag != NULL) {
 912                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 913                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 914                 } else {
 915                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 916                                                    IP_FORWARDING, &error);
 917                 }
 918                 if (sp != NULL) {
 919                         /*
 920                          * Check security policy against packet attributes.
 921                          */
 922                         error = ipsec_in_reject(sp, m);
 923                         KEY_FREESP(&sp);
 924                 } else {
 925                         /* XXX error stat??? */
 926                         error = EINVAL;
 927 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 928                         goto bad;
 929                 }
 930                 crit_exit();
 931                 if (error)
 932                         goto bad;
 933         }
 934 #endif /* FAST_IPSEC */
 935
 936         ipstat.ips_delivered++;
 937         if (needredispatch) {
 938                 struct netmsg_packet *pmsg;
 939                 lwkt_port_t port;
 940
 941                 ip->ip_off = htons(ip->ip_off);
 942                 ip->ip_len = htons(ip->ip_len);
 943                 port = ip_mport_in(&m);
 944                 if (port == NULL)
 945                         return;
 946
 947                 pmsg = &m->m_hdr.mh_netmsg;
 948                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, MSGF_MPSAFE,
 949                             transport_processing_handler);
 950                 pmsg->nm_packet = m;
 951                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
 952
 953                 ip = mtod(m, struct ip *);
 954                 ip->ip_len = ntohs(ip->ip_len);
 955                 ip->ip_off = ntohs(ip->ip_off);
 956                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
 957         } else {
 958                 transport_processing_oncpu(m, hlen, ip);
 959         }
 960         return;
 961
 962 bad:
 963         m_freem(m);
 964 }
 965
 966 /*
 967  * Take incoming datagram fragment and try to reassemble it into
 968  * whole datagram.  If a chain for reassembly of this datagram already
 969  * exists, then it is given as fp; otherwise have to make a chain.
 970  */
 971 struct mbuf *
 972 ip_reass(struct mbuf *m)
 973 {
 974         struct ip *ip = mtod(m, struct ip *);
 975         struct mbuf *p = NULL, *q, *nq;
 976         struct mbuf *n;
 977         struct ipq *fp = NULL;
 978         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 979         int i, next;
 980         u_short sum;
 981
 982         /* If maxnipq is 0, never accept fragments. */
 983         if (maxnipq == 0) {
 984                 ipstat.ips_fragments++;
 985                 ipstat.ips_fragdropped++;
 986                 m_freem(m);
 987                 return NULL;
 988         }
 989
 990         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
 991         /*
 992          * Look for queue of fragments of this datagram.
 993          */
 994         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
 995                 if (ip->ip_id == fp->ipq_id &&
 996                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
 997                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
 998                     ip->ip_p == fp->ipq_p)
 999                         goto found;
1000
1001         fp = NULL;
1002
1003         /*
1004          * Enforce upper bound on number of fragmented packets
1005          * for which we attempt reassembly;
1006          * If maxnipq is -1, accept all fragments without limitation.
1007          */
1008         if (nipq > maxnipq && maxnipq > 0) {
1009                 /*
1010                  * drop something from the tail of the current queue
1011                  * before proceeding further
1012                  */
1013                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1014                         for (i = 0; i < IPREASS_NHASH; i++) {
1015                                 if (ipq[i].prev != &ipq[i]) {
1016                                         ipstat.ips_fragtimeout +=
1017                                             ipq[i].prev->ipq_nfrags;
1018                                         ip_freef(ipq[i].prev);
1019                                         break;
1020                                 }
1021                         }
1022                 } else {
1023                         ipstat.ips_fragtimeout +=
1024                             ipq[sum].prev->ipq_nfrags;
1025                         ip_freef(ipq[sum].prev);
1026                 }
1027         }
1028 found:
1029         /*
1030          * Adjust ip_len to not reflect header,
1031          * convert offset of this to bytes.
1032          */
1033         ip->ip_len -= hlen;
1034         if (ip->ip_off & IP_MF) {
1035                 /*
1036                  * Make sure that fragments have a data length
1037                  * that's a non-zero multiple of 8 bytes.
1038                  */
1039                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1040                         ipstat.ips_toosmall++; /* XXX */
1041                         m_freem(m);
1042                         return NULL;
1043                 }
1044                 m->m_flags |= M_FRAG;
1045         } else
1046                 m->m_flags &= ~M_FRAG;
1047         ip->ip_off <<= 3;
1048
1049         ipstat.ips_fragments++;
1050         m->m_pkthdr.header = ip;
1051
1052         /*
1053          * If the hardware has not done csum over this fragment
1054          * then csum_data is not valid at all.
1055          */
1056         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1057             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1058                 m->m_pkthdr.csum_data = 0;
1059                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1060         }
1061
1062         /*
1063          * Presence of header sizes in mbufs
1064          * would confuse code below.
1065          */
1066         m->m_data += hlen;
1067         m->m_len -= hlen;
1068
1069         /*
1070          * If first fragment to arrive, create a reassembly queue.
1071          */
1072         if (fp == NULL) {
1073                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1074                         goto dropfrag;
1075                 insque(fp, &ipq[sum]);
1076                 nipq++;
1077                 fp->ipq_nfrags = 1;
1078                 fp->ipq_ttl = IPFRAGTTL;
1079                 fp->ipq_p = ip->ip_p;
1080                 fp->ipq_id = ip->ip_id;
1081                 fp->ipq_src = ip->ip_src;
1082                 fp->ipq_dst = ip->ip_dst;
1083                 fp->ipq_frags = m;
1084                 m->m_nextpkt = NULL;
1085                 goto inserted;
1086         } else {
1087                 fp->ipq_nfrags++;
1088         }
1089
1090 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1091
1092         /*
1093          * Find a segment which begins after this one does.
1094          */
1095         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1096                 if (GETIP(q)->ip_off > ip->ip_off)
1097                         break;
1098
1099         /*
1100          * If there is a preceding segment, it may provide some of
1101          * our data already.  If so, drop the data from the incoming
1102          * segment.  If it provides all of our data, drop us, otherwise
1103          * stick new segment in the proper place.
1104          *
1105          * If some of the data is dropped from the the preceding
1106          * segment, then it's checksum is invalidated.
1107          */
1108         if (p) {
1109                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1110                 if (i > 0) {
1111                         if (i >= ip->ip_len)
1112                                 goto dropfrag;
1113                         m_adj(m, i);
1114                         m->m_pkthdr.csum_flags = 0;
1115                         ip->ip_off += i;
1116                         ip->ip_len -= i;
1117                 }
1118                 m->m_nextpkt = p->m_nextpkt;
1119                 p->m_nextpkt = m;
1120         } else {
1121                 m->m_nextpkt = fp->ipq_frags;
1122                 fp->ipq_frags = m;
1123         }
1124
1125         /*
1126          * While we overlap succeeding segments trim them or,
1127          * if they are completely covered, dequeue them.
1128          */
1129         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1130              q = nq) {
1131                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1132                 if (i < GETIP(q)->ip_len) {
1133                         GETIP(q)->ip_len -= i;
1134                         GETIP(q)->ip_off += i;
1135                         m_adj(q, i);
1136                         q->m_pkthdr.csum_flags = 0;
1137                         break;
1138                 }
1139                 nq = q->m_nextpkt;
1140                 m->m_nextpkt = nq;
1141                 ipstat.ips_fragdropped++;
1142                 fp->ipq_nfrags--;
1143                 q->m_nextpkt = NULL;
1144                 m_freem(q);
1145         }
1146
1147 inserted:
1148         /*
1149          * Check for complete reassembly and perform frag per packet
1150          * limiting.
1151          *
1152          * Frag limiting is performed here so that the nth frag has
1153          * a chance to complete the packet before we drop the packet.
1154          * As a result, n+1 frags are actually allowed per packet, but
1155          * only n will ever be stored. (n = maxfragsperpacket.)
1156          *
1157          */
1158         next = 0;
1159         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1160                 if (GETIP(q)->ip_off != next) {
1161                         if (fp->ipq_nfrags > maxfragsperpacket) {
1162                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1163                                 ip_freef(fp);
1164                         }
1165                         return (NULL);
1166                 }
1167                 next += GETIP(q)->ip_len;
1168         }
1169         /* Make sure the last packet didn't have the IP_MF flag */
1170         if (p->m_flags & M_FRAG) {
1171                 if (fp->ipq_nfrags > maxfragsperpacket) {
1172                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1173                         ip_freef(fp);
1174                 }
1175                 return (NULL);
1176         }
1177
1178         /*
1179          * Reassembly is complete.  Make sure the packet is a sane size.
1180          */
1181         q = fp->ipq_frags;
1182         ip = GETIP(q);
1183         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1184                 ipstat.ips_toolong++;
1185                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1186                 ip_freef(fp);
1187                 return (NULL);
1188         }
1189
1190         /*
1191          * Concatenate fragments.
1192          */
1193         m = q;
1194         n = m->m_next;
1195         m->m_next = NULL;
1196         m_cat(m, n);
1197         nq = q->m_nextpkt;
1198         q->m_nextpkt = NULL;
1199         for (q = nq; q != NULL; q = nq) {
1200                 nq = q->m_nextpkt;
1201                 q->m_nextpkt = NULL;
1202                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1203                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1204                 m_cat(m, q);
1205         }
1206
1207         /*
1208          * Clean up the 1's complement checksum.  Carry over 16 bits must
1209          * be added back.  This assumes no more then 65535 packet fragments
1210          * were reassembled.  A second carry can also occur (but not a third).
1211          */
1212         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1213                                 (m->m_pkthdr.csum_data >> 16);
1214         if (m->m_pkthdr.csum_data > 0xFFFF)
1215                 m->m_pkthdr.csum_data -= 0xFFFF;
1216
1217         /*
1218          * Create header for new ip packet by
1219          * modifying header of first packet;
1220          * dequeue and discard fragment reassembly header.
1221          * Make header visible.
1222          */
1223         ip->ip_len = next;
1224         ip->ip_src = fp->ipq_src;
1225         ip->ip_dst = fp->ipq_dst;
1226         remque(fp);
1227         nipq--;
1228         mpipe_free(&ipq_mpipe, fp);
1229         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1230         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1231         /* some debugging cruft by sklower, below, will go away soon */
1232         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1233                 int plen = 0;
1234
1235                 for (n = m; n; n = n->m_next)
1236                         plen += n->m_len;
1237                 m->m_pkthdr.len = plen;
1238         }
1239
1240         ipstat.ips_reassembled++;
1241         return (m);
1242
1243 dropfrag:
1244         ipstat.ips_fragdropped++;
1245         if (fp != NULL)
1246                 fp->ipq_nfrags--;
1247         m_freem(m);
1248         return (NULL);
1249
1250 #undef GETIP
1251 }
1252
1253 /*
1254  * Free a fragment reassembly header and all
1255  * associated datagrams.
1256  */
1257 static void
1258 ip_freef(struct ipq *fp)
1259 {
1260         struct mbuf *q;
1261
1262         while (fp->ipq_frags) {
1263                 q = fp->ipq_frags;
1264                 fp->ipq_frags = q->m_nextpkt;
1265                 q->m_nextpkt = NULL;
1266                 m_freem(q);
1267         }
1268         remque(fp);
1269         mpipe_free(&ipq_mpipe, fp);
1270         nipq--;
1271 }
1272
1273 /*
1274  * IP timer processing;
1275  * if a timer expires on a reassembly
1276  * queue, discard it.
1277  */
1278 void
1279 ip_slowtimo(void)
1280 {
1281         struct ipq *fp;
1282         int i;
1283
1284         crit_enter();
1285         for (i = 0; i < IPREASS_NHASH; i++) {
1286                 fp = ipq[i].next;
1287                 if (fp == NULL)
1288                         continue;
1289                 while (fp != &ipq[i]) {
1290                         --fp->ipq_ttl;
1291                         fp = fp->next;
1292                         if (fp->prev->ipq_ttl == 0) {
1293                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1294                                 ip_freef(fp->prev);
1295                         }
1296                 }
1297         }
1298         /*
1299          * If we are over the maximum number of fragments
1300          * (due to the limit being lowered), drain off
1301          * enough to get down to the new limit.
1302          */
1303         if (maxnipq >= 0 && nipq > maxnipq) {
1304                 for (i = 0; i < IPREASS_NHASH; i++) {
1305                         while (nipq > maxnipq &&
1306                                 (ipq[i].next != &ipq[i])) {
1307                                 ipstat.ips_fragdropped +=
1308                                     ipq[i].next->ipq_nfrags;
1309                                 ip_freef(ipq[i].next);
1310                         }
1311                 }
1312         }
1313         ipflow_slowtimo();
1314         crit_exit();
1315 }
1316
1317 /*
1318  * Drain off all datagram fragments.
1319  */
1320 void
1321 ip_drain(void)
1322 {
1323         int i;
1324
1325         for (i = 0; i < IPREASS_NHASH; i++) {
1326                 while (ipq[i].next != &ipq[i]) {
1327                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1328                         ip_freef(ipq[i].next);
1329                 }
1330         }
1331         in_rtqdrain();
1332 }
1333
1334 /*
1335  * Do option processing on a datagram,
1336  * possibly discarding it if bad options are encountered,
1337  * or forwarding it if source-routed.
1338  * The pass argument is used when operating in the IPSTEALTH
1339  * mode to tell what options to process:
1340  * [LS]SRR (pass 0) or the others (pass 1).
1341  * The reason for as many as two passes is that when doing IPSTEALTH,
1342  * non-routing options should be processed only if the packet is for us.
1343  * Returns 1 if packet has been forwarded/freed,
1344  * 0 if the packet should be processed further.
1345  */
1346 static int
1347 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1348 {
1349         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1350         struct ip *ip = mtod(m, struct ip *);
1351         u_char *cp;
1352         struct in_ifaddr *ia;
1353         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1354         boolean_t forward = FALSE;
1355         struct in_addr *sin, dst;
1356         n_time ntime;
1357
1358         dst = ip->ip_dst;
1359         cp = (u_char *)(ip + 1);
1360         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1361         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1362                 opt = cp[IPOPT_OPTVAL];
1363                 if (opt == IPOPT_EOL)
1364                         break;
1365                 if (opt == IPOPT_NOP)
1366                         optlen = 1;
1367                 else {
1368                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1369                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1370                                 goto bad;
1371                         }
1372                         optlen = cp[IPOPT_OLEN];
1373                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1374                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1375                                 goto bad;
1376                         }
1377                 }
1378                 switch (opt) {
1379
1380                 default:
1381                         break;
1382
1383                 /*
1384                  * Source routing with record.
1385                  * Find interface with current destination address.
1386                  * If none on this machine then drop if strictly routed,
1387                  * or do nothing if loosely routed.
1388                  * Record interface address and bring up next address
1389                  * component.  If strictly routed make sure next
1390                  * address is on directly accessible net.
1391                  */
1392                 case IPOPT_LSRR:
1393                 case IPOPT_SSRR:
1394                         if (ipstealth && pass > 0)
1395                                 break;
1396                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1397                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1398                                 goto bad;
1399                         }
1400                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1401                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1402                                 goto bad;
1403                         }
1404                         ipaddr.sin_addr = ip->ip_dst;
1405                         ia = (struct in_ifaddr *)
1406                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1407                         if (ia == NULL) {
1408                                 if (opt == IPOPT_SSRR) {
1409                                         type = ICMP_UNREACH;
1410                                         code = ICMP_UNREACH_SRCFAIL;
1411                                         goto bad;
1412                                 }
1413                                 if (!ip_dosourceroute)
1414                                         goto nosourcerouting;
1415                                 /*
1416                                  * Loose routing, and not at next destination
1417                                  * yet; nothing to do except forward.
1418                                  */
1419                                 break;
1420                         }
1421                         off--;                  /* 0 origin */
1422                         if (off > optlen - (int)sizeof(struct in_addr)) {
1423                                 /*
1424                                  * End of source route.  Should be for us.
1425                                  */
1426                                 if (!ip_acceptsourceroute)
1427                                         goto nosourcerouting;
1428                                 save_rte(m, cp, ip->ip_src);
1429                                 break;
1430                         }
1431                         if (ipstealth)
1432                                 goto dropit;
1433                         if (!ip_dosourceroute) {
1434                                 if (ipforwarding) {
1435                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1436
1437                                         /*
1438                                          * Acting as a router, so generate ICMP
1439                                          */
1440 nosourcerouting:
1441                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1442                                         log(LOG_WARNING,
1443                                             "attempted source route from %s to %s\n",
1444                                             inet_ntoa(ip->ip_src), buf);
1445                                         type = ICMP_UNREACH;
1446                                         code = ICMP_UNREACH_SRCFAIL;
1447                                         goto bad;
1448                                 } else {
1449                                         /*
1450                                          * Not acting as a router,
1451                                          * so silently drop.
1452                                          */
1453 dropit:
1454                                         ipstat.ips_cantforward++;
1455                                         m_freem(m);
1456                                         return (1);
1457                                 }
1458                         }
1459
1460                         /*
1461                          * locate outgoing interface
1462                          */
1463                         memcpy(&ipaddr.sin_addr, cp + off,
1464                             sizeof ipaddr.sin_addr);
1465
1466                         if (opt == IPOPT_SSRR) {
1467 #define INA     struct in_ifaddr *
1468 #define SA      struct sockaddr *
1469                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1470                                                                         == NULL)
1471                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1472                         } else
1473                                 ia = ip_rtaddr(ipaddr.sin_addr,
1474                                                &ipforward_rt[mycpuid]);
1475                         if (ia == NULL) {
1476                                 type = ICMP_UNREACH;
1477                                 code = ICMP_UNREACH_SRCFAIL;
1478                                 goto bad;
1479                         }
1480                         ip->ip_dst = ipaddr.sin_addr;
1481                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1482                             sizeof(struct in_addr));
1483                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1484                         /*
1485                          * Let ip_intr's mcast routing check handle mcast pkts
1486                          */
1487                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1488                         break;
1489
1490                 case IPOPT_RR:
1491                         if (ipstealth && pass == 0)
1492                                 break;
1493                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1494                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1495                                 goto bad;
1496                         }
1497                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1498                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1499                                 goto bad;
1500                         }
1501                         /*
1502                          * If no space remains, ignore.
1503                          */
1504                         off--;                  /* 0 origin */
1505                         if (off > optlen - (int)sizeof(struct in_addr))
1506                                 break;
1507                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1508                             sizeof ipaddr.sin_addr);
1509                         /*
1510                          * locate outgoing interface; if we're the destination,
1511                          * use the incoming interface (should be same).
1512                          */
1513                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1514                             (ia = ip_rtaddr(ipaddr.sin_addr,
1515                                             &ipforward_rt[mycpuid]))
1516                                                                      == NULL) {
1517                                 type = ICMP_UNREACH;
1518                                 code = ICMP_UNREACH_HOST;
1519                                 goto bad;
1520                         }
1521                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1522                             sizeof(struct in_addr));
1523                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1524                         break;
1525
1526                 case IPOPT_TS:
1527                         if (ipstealth && pass == 0)
1528                                 break;
1529                         code = cp - (u_char *)ip;
1530                         if (optlen < 4 || optlen > 40) {
1531                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1532                                 goto bad;
1533                         }
1534                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1535                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1536                                 goto bad;
1537                         }
1538                         if (off > optlen - (int)sizeof(int32_t)) {
1539                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1540                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1541                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1542                                         goto bad;
1543                                 }
1544                                 break;
1545                         }
1546                         off--;                          /* 0 origin */
1547                         sin = (struct in_addr *)(cp + off);
1548                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1549
1550                         case IPOPT_TS_TSONLY:
1551                                 break;
1552
1553                         case IPOPT_TS_TSANDADDR:
1554                                 if (off + sizeof(n_time) +
1555                                     sizeof(struct in_addr) > optlen) {
1556                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1557                                         goto bad;
1558                                 }
1559                                 ipaddr.sin_addr = dst;
1560                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1561                                                             m->m_pkthdr.rcvif);
1562                                 if (ia == NULL)
1563                                         continue;
1564                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1565                                     sizeof(struct in_addr));
1566                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1567                                 off += sizeof(struct in_addr);
1568                                 break;
1569
1570                         case IPOPT_TS_PRESPEC:
1571                                 if (off + sizeof(n_time) +
1572                                     sizeof(struct in_addr) > optlen) {
1573                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1574                                         goto bad;
1575                                 }
1576                                 memcpy(&ipaddr.sin_addr, sin,
1577                                     sizeof(struct in_addr));
1578                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1579                                         continue;
1580                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1581                                 off += sizeof(struct in_addr);
1582                                 break;
1583
1584                         default:
1585                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1586                                 goto bad;
1587                         }
1588                         ntime = iptime();
1589                         memcpy(cp + off, &ntime, sizeof(n_time));
1590                         cp[IPOPT_OFFSET] += sizeof(n_time);
1591                 }
1592         }
1593         if (forward && ipforwarding) {
1594                 ip_forward(m, TRUE, next_hop);
1595                 return (1);
1596         }
1597         return (0);
1598 bad:
1599         icmp_error(m, type, code, 0, 0);
1600         ipstat.ips_badoptions++;
1601         return (1);
1602 }
1603
1604 /*
1605  * Given address of next destination (final or next hop),
1606  * return internet address info of interface to be used to get there.
1607  */
1608 struct in_ifaddr *
1609 ip_rtaddr(struct in_addr dst, struct route *ro)
1610 {
1611         struct sockaddr_in *sin;
1612
1613         sin = (struct sockaddr_in *)&ro->ro_dst;
1614
1615         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1616                 if (ro->ro_rt != NULL) {
1617                         RTFREE(ro->ro_rt);
1618                         ro->ro_rt = NULL;
1619                 }
1620                 sin->sin_family = AF_INET;
1621                 sin->sin_len = sizeof *sin;
1622                 sin->sin_addr = dst;
1623                 rtalloc_ign(ro, RTF_PRCLONING);
1624         }
1625
1626         if (ro->ro_rt == NULL)
1627                 return (NULL);
1628
1629         return (ifatoia(ro->ro_rt->rt_ifa));
1630 }
1631
1632 /*
1633  * Save incoming source route for use in replies,
1634  * to be picked up later by ip_srcroute if the receiver is interested.
1635  */
1636 static void
1637 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1638 {
1639         struct m_tag *mtag;
1640         struct ip_srcrt_opt *opt;
1641         unsigned olen;
1642
1643         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1644         if (mtag == NULL)
1645                 return;
1646         opt = m_tag_data(mtag);
1647
1648         olen = option[IPOPT_OLEN];
1649 #ifdef DIAGNOSTIC
1650         if (ipprintfs)
1651                 kprintf("save_rte: olen %d\n", olen);
1652 #endif
1653         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1654                 m_tag_free(mtag);
1655                 return;
1656         }
1657         bcopy(option, opt->ip_srcrt.srcopt, olen);
1658         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1659         opt->ip_srcrt.dst = dst;
1660         m_tag_prepend(m, mtag);
1661 }
1662
1663 /*
1664  * Retrieve incoming source route for use in replies,
1665  * in the same form used by setsockopt.
1666  * The first hop is placed before the options, will be removed later.
1667  */
1668 struct mbuf *
1669 ip_srcroute(struct mbuf *m0)
1670 {
1671         struct in_addr *p, *q;
1672         struct mbuf *m;
1673         struct m_tag *mtag;
1674         struct ip_srcrt_opt *opt;
1675
1676         if (m0 == NULL)
1677                 return NULL;
1678
1679         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1680         if (mtag == NULL)
1681                 return NULL;
1682         opt = m_tag_data(mtag);
1683
1684         if (opt->ip_nhops == 0)
1685                 return (NULL);
1686         m = m_get(MB_DONTWAIT, MT_HEADER);
1687         if (m == NULL)
1688                 return (NULL);
1689
1690 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1691
1692         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1693         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1694                    sizeof(struct in_addr) + OPTSIZ;
1695 #ifdef DIAGNOSTIC
1696         if (ipprintfs) {
1697                 kprintf("ip_srcroute: nhops %d mlen %d",
1698                         opt->ip_nhops, m->m_len);
1699         }
1700 #endif
1701
1702         /*
1703          * First save first hop for return route
1704          */
1705         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1706         *(mtod(m, struct in_addr *)) = *p--;
1707 #ifdef DIAGNOSTIC
1708         if (ipprintfs)
1709                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1710 #endif
1711
1712         /*
1713          * Copy option fields and padding (nop) to mbuf.
1714          */
1715         opt->ip_srcrt.nop = IPOPT_NOP;
1716         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1717         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1718             OPTSIZ);
1719         q = (struct in_addr *)(mtod(m, caddr_t) +
1720             sizeof(struct in_addr) + OPTSIZ);
1721 #undef OPTSIZ
1722         /*
1723          * Record return path as an IP source route,
1724          * reversing the path (pointers are now aligned).
1725          */
1726         while (p >= opt->ip_srcrt.route) {
1727 #ifdef DIAGNOSTIC
1728                 if (ipprintfs)
1729                         kprintf(" %x", ntohl(q->s_addr));
1730 #endif
1731                 *q++ = *p--;
1732         }
1733         /*
1734          * Last hop goes to final destination.
1735          */
1736         *q = opt->ip_srcrt.dst;
1737         m_tag_delete(m0, mtag);
1738 #ifdef DIAGNOSTIC
1739         if (ipprintfs)
1740                 kprintf(" %x\n", ntohl(q->s_addr));
1741 #endif
1742         return (m);
1743 }
1744
1745 /*
1746  * Strip out IP options.
1747  */
1748 void
1749 ip_stripoptions(struct mbuf *m)
1750 {
1751         int datalen;
1752         struct ip *ip = mtod(m, struct ip *);
1753         caddr_t opts;
1754         int optlen;
1755
1756         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1757         opts = (caddr_t)(ip + 1);
1758         datalen = m->m_len - (sizeof(struct ip) + optlen);
1759         bcopy(opts + optlen, opts, datalen);
1760         m->m_len -= optlen;
1761         if (m->m_flags & M_PKTHDR)
1762                 m->m_pkthdr.len -= optlen;
1763         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1764 }
1765
1766 u_char inetctlerrmap[PRC_NCMDS] = {
1767         0,              0,              0,              0,
1768         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1769         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1770         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1771         0,              0,              0,              0,
1772         ENOPROTOOPT,    ECONNREFUSED
1773 };
1774
1775 /*
1776  * Forward a packet.  If some error occurs return the sender
1777  * an icmp packet.  Note we can't always generate a meaningful
1778  * icmp message because icmp doesn't have a large enough repertoire
1779  * of codes and types.
1780  *
1781  * If not forwarding, just drop the packet.  This could be confusing
1782  * if ipforwarding was zero but some routing protocol was advancing
1783  * us as a gateway to somewhere.  However, we must let the routing
1784  * protocol deal with that.
1785  *
1786  * The using_srcrt parameter indicates whether the packet is being forwarded
1787  * via a source route.
1788  */
1789 void
1790 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1791 {
1792         struct ip *ip = mtod(m, struct ip *);
1793         struct sockaddr_in *ipforward_rtaddr;
1794         struct rtentry *rt;
1795         int error, type = 0, code = 0, destmtu = 0;
1796         struct mbuf *mcopy;
1797         n_long dest;
1798         struct in_addr pkt_dst;
1799         struct route *cache_rt = &ipforward_rt[mycpuid];
1800
1801         dest = INADDR_ANY;
1802         /*
1803          * Cache the destination address of the packet; this may be
1804          * changed by use of 'ipfw fwd'.
1805          */
1806         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1807
1808 #ifdef DIAGNOSTIC
1809         if (ipprintfs)
1810                 kprintf("forward: src %x dst %x ttl %x\n",
1811                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1812 #endif
1813
1814         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1815                 ipstat.ips_cantforward++;
1816                 m_freem(m);
1817                 return;
1818         }
1819         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1820                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1821                 return;
1822         }
1823
1824         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1825         if (cache_rt->ro_rt == NULL ||
1826             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1827                 if (cache_rt->ro_rt != NULL) {
1828                         RTFREE(cache_rt->ro_rt);
1829                         cache_rt->ro_rt = NULL;
1830                 }
1831                 ipforward_rtaddr->sin_family = AF_INET;
1832                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1833                 ipforward_rtaddr->sin_addr = pkt_dst;
1834                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1835                 if (cache_rt->ro_rt == NULL) {
1836                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1837                         return;
1838                 }
1839         }
1840         rt = cache_rt->ro_rt;
1841
1842         /*
1843          * Save the IP header and at most 8 bytes of the payload,
1844          * in case we need to generate an ICMP message to the src.
1845          *
1846          * XXX this can be optimized a lot by saving the data in a local
1847          * buffer on the stack (72 bytes at most), and only allocating the
1848          * mbuf if really necessary. The vast majority of the packets
1849          * are forwarded without having to send an ICMP back (either
1850          * because unnecessary, or because rate limited), so we are
1851          * really we are wasting a lot of work here.
1852          *
1853          * We don't use m_copy() because it might return a reference
1854          * to a shared cluster. Both this function and ip_output()
1855          * assume exclusive access to the IP header in `m', so any
1856          * data in a cluster may change before we reach icmp_error().
1857          */
1858         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1859         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1860                 /*
1861                  * It's probably ok if the pkthdr dup fails (because
1862                  * the deep copy of the tag chain failed), but for now
1863                  * be conservative and just discard the copy since
1864                  * code below may some day want the tags.
1865                  */
1866                 m_free(mcopy);
1867                 mcopy = NULL;
1868         }
1869         if (mcopy != NULL) {
1870                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1871                     (int)ip->ip_len);
1872                 mcopy->m_pkthdr.len = mcopy->m_len;
1873                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1874         }
1875
1876         if (!ipstealth)
1877                 ip->ip_ttl -= IPTTLDEC;
1878
1879         /*
1880          * If forwarding packet using same interface that it came in on,
1881          * perhaps should send a redirect to sender to shortcut a hop.
1882          * Only send redirect if source is sending directly to us,
1883          * and if packet was not source routed (or has any options).
1884          * Also, don't send redirect if forwarding using a default route
1885          * or a route modified by a redirect.
1886          */
1887         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1888             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1889             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1890             ipsendredirects && !using_srcrt && next_hop == NULL) {
1891                 u_long src = ntohl(ip->ip_src.s_addr);
1892                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1893
1894                 if (rt_ifa != NULL &&
1895                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1896                         if (rt->rt_flags & RTF_GATEWAY)
1897                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1898                         else
1899                                 dest = pkt_dst.s_addr;
1900                         /*
1901                          * Router requirements says to only send
1902                          * host redirects.
1903                          */
1904                         type = ICMP_REDIRECT;
1905                         code = ICMP_REDIRECT_HOST;
1906 #ifdef DIAGNOSTIC
1907                         if (ipprintfs)
1908                                 kprintf("redirect (%d) to %x\n", code, dest);
1909 #endif
1910                 }
1911         }
1912
1913         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1914         if (error == 0) {
1915                 ipstat.ips_forward++;
1916                 if (type == 0) {
1917                         if (mcopy) {
1918                                 ipflow_create(cache_rt, mcopy);
1919                                 m_freem(mcopy);
1920                         }
1921                         return;         /* most common case */
1922                 } else {
1923                         ipstat.ips_redirectsent++;
1924                 }
1925         } else {
1926                 ipstat.ips_cantforward++;
1927         }
1928
1929         if (mcopy == NULL)
1930                 return;
1931
1932         /*
1933          * Send ICMP message.
1934          */
1935
1936         switch (error) {
1937
1938         case 0:                         /* forwarded, but need redirect */
1939                 /* type, code set above */
1940                 break;
1941
1942         case ENETUNREACH:               /* shouldn't happen, checked above */
1943         case EHOSTUNREACH:
1944         case ENETDOWN:
1945         case EHOSTDOWN:
1946         default:
1947                 type = ICMP_UNREACH;
1948                 code = ICMP_UNREACH_HOST;
1949                 break;
1950
1951         case EMSGSIZE:
1952                 type = ICMP_UNREACH;
1953                 code = ICMP_UNREACH_NEEDFRAG;
1954 #ifdef IPSEC
1955                 /*
1956                  * If the packet is routed over IPsec tunnel, tell the
1957                  * originator the tunnel MTU.
1958                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1959                  * XXX quickhack!!!
1960                  */
1961                 if (cache_rt->ro_rt != NULL) {
1962                         struct secpolicy *sp = NULL;
1963                         int ipsecerror;
1964                         int ipsechdr;
1965                         struct route *ro;
1966
1967                         sp = ipsec4_getpolicybyaddr(mcopy,
1968                                                     IPSEC_DIR_OUTBOUND,
1969                                                     IP_FORWARDING,
1970                                                     &ipsecerror);
1971
1972                         if (sp == NULL)
1973                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1974                         else {
1975                                 /* count IPsec header size */
1976                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1977                                                          IPSEC_DIR_OUTBOUND,
1978                                                          NULL);
1979
1980                                 /*
1981                                  * find the correct route for outer IPv4
1982                                  * header, compute tunnel MTU.
1983                                  *
1984                                  */
1985                                 if (sp->req != NULL && sp->req->sav != NULL &&
1986                                     sp->req->sav->sah != NULL) {
1987                                         ro = &sp->req->sav->sah->sa_route;
1988                                         if (ro->ro_rt != NULL &&
1989                                             ro->ro_rt->rt_ifp != NULL) {
1990                                                 destmtu =
1991                                                     ro->ro_rt->rt_ifp->if_mtu;
1992                                                 destmtu -= ipsechdr;
1993                                         }
1994                                 }
1995
1996                                 key_freesp(sp);
1997                         }
1998                 }
1999 #elif FAST_IPSEC
2000                 /*
2001                  * If the packet is routed over IPsec tunnel, tell the
2002                  * originator the tunnel MTU.
2003                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2004                  * XXX quickhack!!!
2005                  */
2006                 if (cache_rt->ro_rt != NULL) {
2007                         struct secpolicy *sp = NULL;
2008                         int ipsecerror;
2009                         int ipsechdr;
2010                         struct route *ro;
2011
2012                         sp = ipsec_getpolicybyaddr(mcopy,
2013                                                    IPSEC_DIR_OUTBOUND,
2014                                                    IP_FORWARDING,
2015                                                    &ipsecerror);
2016
2017                         if (sp == NULL)
2018                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2019                         else {
2020                                 /* count IPsec header size */
2021                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2022                                                          IPSEC_DIR_OUTBOUND,
2023                                                          NULL);
2024
2025                                 /*
2026                                  * find the correct route for outer IPv4
2027                                  * header, compute tunnel MTU.
2028                                  */
2029
2030                                 if (sp->req != NULL &&
2031                                     sp->req->sav != NULL &&
2032                                     sp->req->sav->sah != NULL) {
2033                                         ro = &sp->req->sav->sah->sa_route;
2034                                         if (ro->ro_rt != NULL &&
2035                                             ro->ro_rt->rt_ifp != NULL) {
2036                                                 destmtu =
2037                                                     ro->ro_rt->rt_ifp->if_mtu;
2038                                                 destmtu -= ipsechdr;
2039                                         }
2040                                 }
2041
2042                                 KEY_FREESP(&sp);
2043                         }
2044                 }
2045 #else /* !IPSEC && !FAST_IPSEC */
2046                 if (cache_rt->ro_rt != NULL)
2047                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2048 #endif /*IPSEC*/
2049                 ipstat.ips_cantfrag++;
2050                 break;
2051
2052         case ENOBUFS:
2053                 /*
2054                  * A router should not generate ICMP_SOURCEQUENCH as
2055                  * required in RFC1812 Requirements for IP Version 4 Routers.
2056                  * Source quench could be a big problem under DoS attacks,
2057                  * or if the underlying interface is rate-limited.
2058                  * Those who need source quench packets may re-enable them
2059                  * via the net.inet.ip.sendsourcequench sysctl.
2060                  */
2061                 if (!ip_sendsourcequench) {
2062                         m_freem(mcopy);
2063                         return;
2064                 } else {
2065                         type = ICMP_SOURCEQUENCH;
2066                         code = 0;
2067                 }
2068                 break;
2069
2070         case EACCES:                    /* ipfw denied packet */
2071                 m_freem(mcopy);
2072                 return;
2073         }
2074         icmp_error(mcopy, type, code, dest, destmtu);
2075 }
2076
2077 void
2078 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2079                struct mbuf *m)
2080 {
2081         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2082                 struct timeval tv;
2083
2084                 microtime(&tv);
2085                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2086                     SCM_TIMESTAMP, SOL_SOCKET);
2087                 if (*mp)
2088                         mp = &(*mp)->m_next;
2089         }
2090         if (inp->inp_flags & INP_RECVDSTADDR) {
2091                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2092                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2093                 if (*mp)
2094                         mp = &(*mp)->m_next;
2095         }
2096         if (inp->inp_flags & INP_RECVTTL) {
2097                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2098                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2099                 if (*mp)
2100                         mp = &(*mp)->m_next;
2101         }
2102 #ifdef notyet
2103         /* XXX
2104          * Moving these out of udp_input() made them even more broken
2105          * than they already were.
2106          */
2107         /* options were tossed already */
2108         if (inp->inp_flags & INP_RECVOPTS) {
2109                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2110                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2111                 if (*mp)
2112                         mp = &(*mp)->m_next;
2113         }
2114         /* ip_srcroute doesn't do what we want here, need to fix */
2115         if (inp->inp_flags & INP_RECVRETOPTS) {
2116                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2117                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2118                 if (*mp)
2119                         mp = &(*mp)->m_next;
2120         }
2121 #endif
2122         if (inp->inp_flags & INP_RECVIF) {
2123                 struct ifnet *ifp;
2124                 struct sdlbuf {
2125                         struct sockaddr_dl sdl;
2126                         u_char  pad[32];
2127                 } sdlbuf;
2128                 struct sockaddr_dl *sdp;
2129                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2130
2131                 if (((ifp = m->m_pkthdr.rcvif)) &&
2132                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2133                         sdp = IF_LLSOCKADDR(ifp);
2134                         /*
2135                          * Change our mind and don't try copy.
2136                          */
2137                         if ((sdp->sdl_family != AF_LINK) ||
2138                             (sdp->sdl_len > sizeof(sdlbuf))) {
2139                                 goto makedummy;
2140                         }
2141                         bcopy(sdp, sdl2, sdp->sdl_len);
2142                 } else {
2143 makedummy:
2144                         sdl2->sdl_len =
2145                             offsetof(struct sockaddr_dl, sdl_data[0]);
2146                         sdl2->sdl_family = AF_LINK;
2147                         sdl2->sdl_index = 0;
2148                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2149                 }
2150                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2151                         IP_RECVIF, IPPROTO_IP);
2152                 if (*mp)
2153                         mp = &(*mp)->m_next;
2154         }
2155 }
2156
2157 /*
2158  * XXX these routines are called from the upper part of the kernel.
2159  *
2160  * They could also be moved to ip_mroute.c, since all the RSVP
2161  *  handling is done there already.
2162  */
2163 int
2164 ip_rsvp_init(struct socket *so)
2165 {
2166         if (so->so_type != SOCK_RAW ||
2167             so->so_proto->pr_protocol != IPPROTO_RSVP)
2168                 return EOPNOTSUPP;
2169
2170         if (ip_rsvpd != NULL)
2171                 return EADDRINUSE;
2172
2173         ip_rsvpd = so;
2174         /*
2175          * This may seem silly, but we need to be sure we don't over-increment
2176          * the RSVP counter, in case something slips up.
2177          */
2178         if (!ip_rsvp_on) {
2179                 ip_rsvp_on = 1;
2180                 rsvp_on++;
2181         }
2182
2183         return 0;
2184 }
2185
2186 int
2187 ip_rsvp_done(void)
2188 {
2189         ip_rsvpd = NULL;
2190         /*
2191          * This may seem silly, but we need to be sure we don't over-decrement
2192          * the RSVP counter, in case something slips up.
2193          */
2194         if (ip_rsvp_on) {
2195                 ip_rsvp_on = 0;
2196                 rsvp_on--;
2197         }
2198         return 0;
2199 }
2200
2201 void
2202 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2203 {
2204         int off, proto;
2205         __va_list ap;
2206
2207         __va_start(ap, m);
2208         off = __va_arg(ap, int);
2209         proto = __va_arg(ap, int);
2210         __va_end(ap);
2211
2212         if (rsvp_input_p) { /* call the real one if loaded */
2213                 rsvp_input_p(m, off, proto);
2214                 return;
2215         }
2216
2217         /* Can still get packets with rsvp_on = 0 if there is a local member
2218          * of the group to which the RSVP packet is addressed.  But in this
2219          * case we want to throw the packet away.
2220          */
2221
2222         if (!rsvp_on) {
2223                 m_freem(m);
2224                 return;
2225         }
2226
2227         if (ip_rsvpd != NULL) {
2228                 rip_input(m, off, proto);
2229                 return;
2230         }
2231         /* Drop the packet */
2232         m_freem(m);
2233 }