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 1.95 2008/08/28 14:10:03 sephe Exp $
  69  */
  70
  71 #define _IP_VHL
  72
  73 #include "opt_bootp.h"
  74 #include "opt_ipfw.h"
  75 #include "opt_ipdn.h"
  76 #include "opt_ipdivert.h"
  77 #include "opt_ipfilter.h"
  78 #include "opt_ipstealth.h"
  79 #include "opt_ipsec.h"
  80
  81 #include <sys/param.h>
  82 #include <sys/systm.h>
  83 #include <sys/mbuf.h>
  84 #include <sys/malloc.h>
  85 #include <sys/mpipe.h>
  86 #include <sys/domain.h>
  87 #include <sys/protosw.h>
  88 #include <sys/socket.h>
  89 #include <sys/time.h>
  90 #include <sys/globaldata.h>
  91 #include <sys/thread.h>
  92 #include <sys/kernel.h>
  93 #include <sys/syslog.h>
  94 #include <sys/sysctl.h>
  95 #include <sys/in_cksum.h>
  96
  97 #include <machine/stdarg.h>
  98
  99 #include <net/if.h>
 100 #include <net/if_types.h>
 101 #include <net/if_var.h>
 102 #include <net/if_dl.h>
 103 #include <net/pfil.h>
 104 #include <net/route.h>
 105 #include <net/netisr.h>
 106
 107 #include <netinet/in.h>
 108 #include <netinet/in_systm.h>
 109 #include <netinet/in_var.h>
 110 #include <netinet/ip.h>
 111 #include <netinet/in_pcb.h>
 112 #include <netinet/ip_var.h>
 113 #include <netinet/ip_icmp.h>
 114 #ifdef IPDIVERT
 115 #include <netinet/ip_divert.h>
 116 #endif
 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 ipforwarding = 0;
 142 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 143     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 144
 145 static int ipsendredirects = 1; /* XXX */
 146 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 147     &ipsendredirects, 0, "Enable sending IP redirects");
 148
 149 int ip_defttl = IPDEFTTL;
 150 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 151     &ip_defttl, 0, "Maximum TTL on IP packets");
 152
 153 static int ip_dosourceroute = 0;
 154 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 155     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 156
 157 static int ip_acceptsourceroute = 0;
 158 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 159     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 160     "Enable accepting source routed IP packets");
 161
 162 static int ip_keepfaith = 0;
 163 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 164     &ip_keepfaith, 0,
 165     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 166
 167 static int nipq = 0;    /* total # of reass queues */
 168 static int maxnipq;
 169 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 170     &maxnipq, 0,
 171     "Maximum number of IPv4 fragment reassembly queue entries");
 172
 173 static int maxfragsperpacket;
 174 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 175     &maxfragsperpacket, 0,
 176     "Maximum number of IPv4 fragments allowed per packet");
 177
 178 static int ip_sendsourcequench = 0;
 179 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 180     &ip_sendsourcequench, 0,
 181     "Enable the transmission of source quench packets");
 182
 183 int ip_do_randomid = 1;
 184 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 185     &ip_do_randomid, 0,
 186     "Assign random ip_id values");
 187 /*
 188  * XXX - Setting ip_checkinterface mostly implements the receive side of
 189  * the Strong ES model described in RFC 1122, but since the routing table
 190  * and transmit implementation do not implement the Strong ES model,
 191  * setting this to 1 results in an odd hybrid.
 192  *
 193  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 194  * to translate the destination address to another local interface.
 195  *
 196  * XXX - ip_checkinterface must be disabled if you add IP aliases
 197  * to the loopback interface instead of the interface where the
 198  * packets for those addresses are received.
 199  */
 200 static int ip_checkinterface = 0;
 201 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 202     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 203
 204 #ifdef DIAGNOSTIC
 205 static int ipprintfs = 0;
 206 #endif
 207
 208 extern  struct domain inetdomain;
 209 extern  struct protosw inetsw[];
 210 u_char  ip_protox[IPPROTO_MAX];
 211 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 212 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 213                                                 /* inet addr hash table */
 214 u_long  in_ifaddrhmask;                         /* mask for hash table */
 215
 216 struct ip_stats ipstats_percpu[MAXCPU];
 217 #ifdef SMP
 218 static int
 219 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 220 {
 221         int cpu, error = 0;
 222
 223         for (cpu = 0; cpu < ncpus; ++cpu) {
 224                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 225                                         sizeof(struct ip_stats))))
 226                         break;
 227                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 228                                        sizeof(struct ip_stats))))
 229                         break;
 230         }
 231
 232         return (error);
 233 }
 234 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 235     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 236 #else
 237 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
 238     &ipstat, ip_stats, "IP statistics");
 239 #endif
 240
 241 /* Packet reassembly stuff */
 242 #define IPREASS_NHASH_LOG2      6
 243 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 244 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 245 #define IPREASS_HASH(x,y)                                               \
 246     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 247
 248 static struct ipq ipq[IPREASS_NHASH];
 249
 250 #ifdef IPCTL_DEFMTU
 251 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 252     &ip_mtu, 0, "Default MTU");
 253 #endif
 254
 255 #ifdef IPSTEALTH
 256 static int ipstealth = 0;
 257 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 258 #else
 259 static const int ipstealth = 0;
 260 #endif
 261
 262
 263 /* Firewall hooks */
 264 ip_fw_chk_t *ip_fw_chk_ptr;
 265 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
 266 int ip_fw_loaded;
 267 int fw_enable = 1;
 268 int fw_one_pass = 1;
 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 static struct mbuf      *ip_reass(struct mbuf *);
 306
 307 /*
 308  * IP initialization: fill in IP protocol switch table.
 309  * All protocols not implemented in kernel go to raw IP protocol handler.
 310  */
 311 void
 312 ip_init(void)
 313 {
 314         struct protosw *pr;
 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 &&
 339                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
 340                         ip_protox[pr->pr_protocol] = pr - inetsw;
 341
 342         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 343         inet_pfil_hook.ph_af = AF_INET;
 344         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 345                 kprintf("%s: WARNING: unable to register pfil hook, "
 346                         "error %d\n", __func__, i);
 347         }
 348
 349         for (i = 0; i < IPREASS_NHASH; i++)
 350                 ipq[i].next = ipq[i].prev = &ipq[i];
 351
 352         maxnipq = nmbclusters / 32;
 353         maxfragsperpacket = 16;
 354
 355         ip_id = time_second & 0xffff;
 356
 357         /*
 358          * Initialize IP statistics counters for each CPU.
 359          *
 360          */
 361 #ifdef SMP
 362         for (cpu = 0; cpu < ncpus; ++cpu) {
 363                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 364         }
 365 #else
 366         bzero(&ipstat, sizeof(struct ip_stats));
 367 #endif
 368
 369         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
 370 }
 371
 372 /*
 373  * XXX watch out this one. It is perhaps used as a cache for
 374  * the most recently used route ? it is cleared in in_addroute()
 375  * when a new route is successfully created.
 376  */
 377 struct route ipforward_rt[MAXCPU];
 378
 379 /* Do transport protocol processing. */
 380 static void
 381 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 382 {
 383         /*
 384          * Switch out to protocol's input routine.
 385          */
 386         (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
 387 }
 388
 389 static void
 390 transport_processing_handler(netmsg_t netmsg)
 391 {
 392         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
 393         struct ip *ip;
 394         int hlen;
 395
 396         ip = mtod(pmsg->nm_packet, struct ip *);
 397         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
 398
 399         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 400         /* netmsg was embedded in the mbuf, do not reply! */
 401 }
 402
 403 static void
 404 ip_input_handler(struct netmsg *msg0)
 405 {
 406         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
 407
 408         ip_input(m);
 409         /* msg0 was embedded in the mbuf, do not reply! */
 410 }
 411
 412 /*
 413  * IP input routine.  Checksum and byte swap header.  If fragmented
 414  * try to reassemble.  Process options.  Pass to next level.
 415  */
 416 void
 417 ip_input(struct mbuf *m)
 418 {
 419         struct ip *ip;
 420         struct in_ifaddr *ia = NULL;
 421         struct in_ifaddr_container *iac;
 422         int i, hlen, checkif;
 423         u_short sum;
 424         struct in_addr pkt_dst;
 425         struct ip_fw_args args;
 426         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 427         boolean_t needredispatch = FALSE;
 428         struct in_addr odst;                    /* original dst address(NAT) */
 429         struct m_tag *mtag;
 430         struct sockaddr_in *next_hop = NULL;
 431 #ifdef FAST_IPSEC
 432         struct tdb_ident *tdbi;
 433         struct secpolicy *sp;
 434         int error;
 435 #endif
 436
 437         args.eh = NULL;
 438         args.oif = NULL;
 439         args.rule = NULL;
 440
 441         M_ASSERTPKTHDR(m);
 442
 443         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 444                 /* Next hop */
 445                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 446                 KKASSERT(mtag != NULL);
 447                 next_hop = m_tag_data(mtag);
 448         }
 449
 450         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 451                 /* Extract info from dummynet tag */
 452                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 453                 KKASSERT(mtag != NULL);
 454                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
 455                 KKASSERT(args.rule != NULL);
 456
 457                 m_tag_delete(m, mtag);
 458                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 459
 460                 /* dummynet already filtered us */
 461                 ip = mtod(m, struct ip *);
 462                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 463                 goto iphack;
 464         }
 465
 466         ipstat.ips_total++;
 467
 468         /* length checks already done in ip_demux() */
 469         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 470
 471         ip = mtod(m, struct ip *);
 472
 473         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 474                 ipstat.ips_badvers++;
 475                 goto bad;
 476         }
 477
 478         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 479         /* length checks already done in ip_demux() */
 480         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 481         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 482
 483         /* 127/8 must not appear on wire - RFC1122 */
 484         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 485             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 486                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 487                         ipstat.ips_badaddr++;
 488                         goto bad;
 489                 }
 490         }
 491
 492         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 493                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 494         } else {
 495                 if (hlen == sizeof(struct ip)) {
 496                         sum = in_cksum_hdr(ip);
 497                 } else {
 498                         sum = in_cksum(m, hlen);
 499                 }
 500         }
 501         if (sum != 0) {
 502                 ipstat.ips_badsum++;
 503                 goto bad;
 504         }
 505
 506 #ifdef ALTQ
 507         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 508                 /* packet is dropped by traffic conditioner */
 509                 return;
 510         }
 511 #endif
 512         /*
 513          * Convert fields to host representation.
 514          */
 515         ip->ip_len = ntohs(ip->ip_len);
 516         if (ip->ip_len < hlen) {
 517                 ipstat.ips_badlen++;
 518                 goto bad;
 519         }
 520         ip->ip_off = ntohs(ip->ip_off);
 521
 522         /*
 523          * Check that the amount of data in the buffers
 524          * is as at least much as the IP header would have us expect.
 525          * Trim mbufs if longer than we expect.
 526          * Drop packet if shorter than we expect.
 527          */
 528         if (m->m_pkthdr.len < ip->ip_len) {
 529                 ipstat.ips_tooshort++;
 530                 goto bad;
 531         }
 532         if (m->m_pkthdr.len > ip->ip_len) {
 533                 if (m->m_len == m->m_pkthdr.len) {
 534                         m->m_len = ip->ip_len;
 535                         m->m_pkthdr.len = ip->ip_len;
 536                 } else
 537                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 538         }
 539 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 540         /*
 541          * Bypass packet filtering for packets from a tunnel (gif).
 542          */
 543         if (ipsec_gethist(m, NULL))
 544                 goto pass;
 545 #endif
 546
 547         /*
 548          * IpHack's section.
 549          * Right now when no processing on packet has done
 550          * and it is still fresh out of network we do our black
 551          * deals with it.
 552          * - Firewall: deny/allow/divert
 553          * - Xlate: translate packet's addr/port (NAT).
 554          * - Pipe: pass pkt through dummynet.
 555          * - Wrap: fake packet's addr/port <unimpl.>
 556          * - Encapsulate: put it in another IP and send out. <unimp.>
 557          */
 558
 559 iphack:
 560
 561         /*
 562          * Run through list of hooks for input packets.
 563          *
 564          * NB: Beware of the destination address changing (e.g.
 565          *     by NAT rewriting). When this happens, tell
 566          *     ip_forward to do the right thing.
 567          */
 568         if (pfil_has_hooks(&inet_pfil_hook)) {
 569                 odst = ip->ip_dst;
 570                 if (pfil_run_hooks(&inet_pfil_hook, &m,
 571                     m->m_pkthdr.rcvif, PFIL_IN)) {
 572                         return;
 573                 }
 574                 if (m == NULL)                  /* consumed by filter */
 575                         return;
 576                 ip = mtod(m, struct ip *);
 577                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 578         }
 579
 580         if (fw_enable && IPFW_LOADED) {
 581                 /*
 582                  * If we've been forwarded from the output side, then
 583                  * skip the firewall a second time
 584                  */
 585                 if (next_hop != NULL)
 586                         goto ours;
 587
 588                 args.m = m;
 589                 i = ip_fw_chk_ptr(&args);
 590                 m = args.m;
 591
 592                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
 593                         if (m != NULL)
 594                                 m_freem(m);
 595                         return;
 596                 }
 597                 ip = mtod(m, struct ip *);      /* just in case m changed */
 598
 599                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 600                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 601                         KKASSERT(mtag != NULL);
 602                         next_hop = m_tag_data(mtag);
 603                 }
 604
 605                 if (i == 0 && next_hop == NULL) /* common case */
 606                         goto pass;
 607                 if (i & IP_FW_PORT_DYNT_FLAG) {
 608                         /* Send packet to the appropriate pipe */
 609                         ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
 610                         return;
 611                 }
 612 #ifdef IPDIVERT
 613                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
 614                         struct mbuf *clone = NULL;
 615                         int tee = 0;
 616
 617                         /* Divert or 'tee'? */
 618                         if (i & IP_FW_PORT_TEE_FLAG)
 619                                 tee = 1;
 620
 621                         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 622                                 const struct divert_info *divinfo;
 623                                 u_short frag_off;
 624
 625                                 /*
 626                                  * Only trust divert info in the fragment
 627                                  * at offset 0.
 628                                  */
 629                                 frag_off = ip->ip_off << 3;
 630                                 if (frag_off != 0) {
 631                                         mtag = m_tag_find(m,
 632                                         PACKET_TAG_IPFW_DIVERT, NULL);
 633                                         m_tag_delete(m, mtag);
 634                                 }
 635
 636                                 /*
 637                                  * Attempt reassembly; if it succeeds, proceed.
 638                                  * ip_reass() will return a different mbuf.
 639                                  */
 640                                 m = ip_reass(m);
 641                                 if (m == NULL)
 642                                         return;
 643                                 ip = mtod(m, struct ip *);
 644
 645                                 needredispatch = TRUE;
 646
 647                                 /*
 648                                  * Get the header length of the reassembled
 649                                  * packet
 650                                  */
 651                                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 652
 653                                 /*
 654                                  * Restore original checksum before diverting
 655                                  * packet
 656                                  */
 657                                 ip->ip_len += hlen;
 658                                 ip->ip_len = htons(ip->ip_len);
 659                                 ip->ip_off = htons(ip->ip_off);
 660                                 ip->ip_sum = 0;
 661                                 if (hlen == sizeof(struct ip))
 662                                         ip->ip_sum = in_cksum_hdr(ip);
 663                                 else
 664                                         ip->ip_sum = in_cksum(m, hlen);
 665                                 ip->ip_off = ntohs(ip->ip_off);
 666                                 ip->ip_len = ntohs(ip->ip_len);
 667
 668                                 /*
 669                                  * Only use the saved divert info
 670                                  */
 671                                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT,
 672                                                   NULL);
 673                                 if (mtag == NULL) {
 674                                         /* Wrongly configured ipfw */
 675                                         kprintf("ip_input no divert info\n");
 676                                         m_freem(m);
 677                                         return;
 678                                 }
 679                                 divinfo = m_tag_data(mtag);
 680                                 tee = divinfo->tee;
 681                         }
 682
 683                         /*
 684                          * Divert or tee packet to the divert protocol if
 685                          * required.
 686                          */
 687
 688                         /* Clone packet if we're doing a 'tee' */
 689                         if (tee)
 690                                 clone = m_dup(m, MB_DONTWAIT);
 691
 692                         /*
 693                          * Restore packet header fields to original
 694                          * values
 695                          */
 696                         ip->ip_len = htons(ip->ip_len);
 697                         ip->ip_off = htons(ip->ip_off);
 698
 699                         /* Deliver packet to divert input routine */
 700                         divert_packet(m, 1);
 701                         ipstat.ips_delivered++;
 702
 703                         /* If 'tee', continue with original packet */
 704                         if (clone == NULL)
 705                                 return;
 706                         m = clone;
 707                         ip = mtod(m, struct ip *);
 708                         /*
 709                          * Jump backwards to complete processing of the
 710                          * packet. But first clear divert_info to avoid
 711                          * entering this block again.
 712                          * We do not need to clear args.divert_rule as
 713                          * it will not be used.
 714                          *
 715                          * XXX
 716                          * Better safe than sorry, remove the DIVERT
 717                          * tag.
 718                          */
 719                         mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT,
 720                                           NULL);
 721                         KKASSERT(mtag != NULL);
 722                         m_tag_delete(m, mtag);
 723
 724                         goto pass;
 725                 }
 726 #endif
 727                 if (i == 0 && next_hop != NULL)
 728                         goto pass;
 729                 /*
 730                  * if we get here, the packet must be dropped
 731                  */
 732                 m_freem(m);
 733                 return;
 734         }
 735 pass:
 736
 737         /*
 738          * Process options and, if not destined for us,
 739          * ship it on.  ip_dooptions returns 1 when an
 740          * error was detected (causing an icmp message
 741          * to be sent and the original packet to be freed).
 742          */
 743         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 744                 return;
 745
 746         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 747          * matter if it is destined to another node, or whether it is
 748          * a multicast one, RSVP wants it! and prevents it from being forwarded
 749          * anywhere else. Also checks if the rsvp daemon is running before
 750          * grabbing the packet.
 751          */
 752         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 753                 goto ours;
 754
 755         /*
 756          * Check our list of addresses, to see if the packet is for us.
 757          * If we don't have any addresses, assume any unicast packet
 758          * we receive might be for us (and let the upper layers deal
 759          * with it).
 760          */
 761         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 762             !(m->m_flags & (M_MCAST | M_BCAST)))
 763                 goto ours;
 764
 765         /*
 766          * Cache the destination address of the packet; this may be
 767          * changed by use of 'ipfw fwd'.
 768          */
 769         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 770
 771         /*
 772          * Enable a consistency check between the destination address
 773          * and the arrival interface for a unicast packet (the RFC 1122
 774          * strong ES model) if IP forwarding is disabled and the packet
 775          * is not locally generated and the packet is not subject to
 776          * 'ipfw fwd'.
 777          *
 778          * XXX - Checking also should be disabled if the destination
 779          * address is ipnat'ed to a different interface.
 780          *
 781          * XXX - Checking is incompatible with IP aliases added
 782          * to the loopback interface instead of the interface where
 783          * the packets are received.
 784          */
 785         checkif = ip_checkinterface &&
 786                   !ipforwarding &&
 787                   m->m_pkthdr.rcvif != NULL &&
 788                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 789                   next_hop == NULL;
 790
 791         /*
 792          * Check for exact addresses in the hash bucket.
 793          */
 794         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 795                 ia = iac->ia;
 796
 797                 /*
 798                  * If the address matches, verify that the packet
 799                  * arrived via the correct interface if checking is
 800                  * enabled.
 801                  */
 802                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 803                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 804                         goto ours;
 805         }
 806         ia = NULL;
 807
 808         /*
 809          * Check for broadcast addresses.
 810          *
 811          * Only accept broadcast packets that arrive via the matching
 812          * interface.  Reception of forwarded directed broadcasts would
 813          * be handled via ip_forward() and ether_output() with the loopback
 814          * into the stack for SIMPLEX interfaces handled by ether_output().
 815          */
 816         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 817                 struct ifaddr_container *ifac;
 818
 819                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 820                               ifa_link) {
 821                         struct ifaddr *ifa = ifac->ifa;
 822
 823                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 824                                 continue;
 825                         if (ifa->ifa_addr->sa_family != AF_INET)
 826                                 continue;
 827                         ia = ifatoia(ifa);
 828                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 829                                                                 pkt_dst.s_addr)
 830                                 goto ours;
 831                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 832                                 goto ours;
 833 #ifdef BOOTP_COMPAT
 834                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 835                                 goto ours;
 836 #endif
 837                 }
 838         }
 839         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 840                 struct in_multi *inm;
 841
 842                 if (ip_mrouter != NULL) {
 843                         /*
 844                          * If we are acting as a multicast router, all
 845                          * incoming multicast packets are passed to the
 846                          * kernel-level multicast forwarding function.
 847                          * The packet is returned (relatively) intact; if
 848                          * ip_mforward() returns a non-zero value, the packet
 849                          * must be discarded, else it may be accepted below.
 850                          */
 851                         if (ip_mforward != NULL &&
 852                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 853                                 ipstat.ips_cantforward++;
 854                                 m_freem(m);
 855                                 return;
 856                         }
 857
 858                         /*
 859                          * The process-level routing daemon needs to receive
 860                          * all multicast IGMP packets, whether or not this
 861                          * host belongs to their destination groups.
 862                          */
 863                         if (ip->ip_p == IPPROTO_IGMP)
 864                                 goto ours;
 865                         ipstat.ips_forward++;
 866                 }
 867                 /*
 868                  * See if we belong to the destination multicast group on the
 869                  * arrival interface.
 870                  */
 871                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
 872                 if (inm == NULL) {
 873                         ipstat.ips_notmember++;
 874                         m_freem(m);
 875                         return;
 876                 }
 877                 goto ours;
 878         }
 879         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 880                 goto ours;
 881         if (ip->ip_dst.s_addr == INADDR_ANY)
 882                 goto ours;
 883
 884         /*
 885          * FAITH(Firewall Aided Internet Translator)
 886          */
 887         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 888                 if (ip_keepfaith) {
 889                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 890                                 goto ours;
 891                 }
 892                 m_freem(m);
 893                 return;
 894         }
 895
 896         /*
 897          * Not for us; forward if possible and desirable.
 898          */
 899         if (!ipforwarding) {
 900                 ipstat.ips_cantforward++;
 901                 m_freem(m);
 902         } else {
 903 #ifdef IPSEC
 904                 /*
 905                  * Enforce inbound IPsec SPD.
 906                  */
 907                 if (ipsec4_in_reject(m, NULL)) {
 908                         ipsecstat.in_polvio++;
 909                         goto bad;
 910                 }
 911 #endif
 912 #ifdef FAST_IPSEC
 913                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 914                 crit_enter();
 915                 if (mtag != NULL) {
 916                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 917                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 918                 } else {
 919                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 920                                                    IP_FORWARDING, &error);
 921                 }
 922                 if (sp == NULL) {       /* NB: can happen if error */
 923                         crit_exit();
 924                         /*XXX error stat???*/
 925                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 926                         goto bad;
 927                 }
 928
 929                 /*
 930                  * Check security policy against packet attributes.
 931                  */
 932                 error = ipsec_in_reject(sp, m);
 933                 KEY_FREESP(&sp);
 934                 crit_exit();
 935                 if (error) {
 936                         ipstat.ips_cantforward++;
 937                         goto bad;
 938                 }
 939 #endif
 940                 ip_forward(m, using_srcrt, next_hop);
 941         }
 942         return;
 943
 944 ours:
 945
 946         /*
 947          * IPSTEALTH: Process non-routing options only
 948          * if the packet is destined for us.
 949          */
 950         if (ipstealth &&
 951             hlen > sizeof(struct ip) &&
 952             ip_dooptions(m, 1, next_hop))
 953                 return;
 954
 955         /* Count the packet in the ip address stats */
 956         if (ia != NULL) {
 957                 ia->ia_ifa.if_ipackets++;
 958                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
 959         }
 960
 961         /*
 962          * If offset or IP_MF are set, must reassemble.
 963          * Otherwise, nothing need be done.
 964          * (We could look in the reassembly queue to see
 965          * if the packet was previously fragmented,
 966          * but it's not worth the time; just let them time out.)
 967          */
 968         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 969                 /*
 970                  * Attempt reassembly; if it succeeds, proceed.
 971                  * ip_reass() will return a different mbuf.
 972                  */
 973                 m = ip_reass(m);
 974                 if (m == NULL)
 975                         return;
 976                 ip = mtod(m, struct ip *);
 977
 978                 /* Get the header length of the reassembled packet */
 979                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 980
 981                 needredispatch = TRUE;
 982         } else {
 983                 ip->ip_len -= hlen;
 984         }
 985
 986 #ifdef IPSEC
 987         /*
 988          * enforce IPsec policy checking if we are seeing last header.
 989          * note that we do not visit this with protocols with pcb layer
 990          * code - like udp/tcp/raw ip.
 991          */
 992         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 993             ipsec4_in_reject(m, NULL)) {
 994                 ipsecstat.in_polvio++;
 995                 goto bad;
 996         }
 997 #endif
 998 #if FAST_IPSEC
 999         /*
1000          * enforce IPsec policy checking if we are seeing last header.
1001          * note that we do not visit this with protocols with pcb layer
1002          * code - like udp/tcp/raw ip.
1003          */
1004         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1005                 /*
1006                  * Check if the packet has already had IPsec processing
1007                  * done.  If so, then just pass it along.  This tag gets
1008                  * set during AH, ESP, etc. input handling, before the
1009                  * packet is returned to the ip input queue for delivery.
1010                  */
1011                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1012                 crit_enter();
1013                 if (mtag != NULL) {
1014                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
1015                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1016                 } else {
1017                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1018                                                    IP_FORWARDING, &error);
1019                 }
1020                 if (sp != NULL) {
1021                         /*
1022                          * Check security policy against packet attributes.
1023                          */
1024                         error = ipsec_in_reject(sp, m);
1025                         KEY_FREESP(&sp);
1026                 } else {
1027                         /* XXX error stat??? */
1028                         error = EINVAL;
1029 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1030                         goto bad;
1031                 }
1032                 crit_exit();
1033                 if (error)
1034                         goto bad;
1035         }
1036 #endif /* FAST_IPSEC */
1037
1038         ipstat.ips_delivered++;
1039         if (needredispatch) {
1040                 struct netmsg_packet *pmsg;
1041                 lwkt_port_t port;
1042
1043                 ip->ip_off = htons(ip->ip_off);
1044                 ip->ip_len = htons(ip->ip_len);
1045                 port = ip_mport_in(&m);
1046                 if (port == NULL)
1047                         return;
1048
1049                 pmsg = &m->m_hdr.mh_netmsg;
1050                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
1051                             transport_processing_handler);
1052                 pmsg->nm_packet = m;
1053                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
1054
1055                 ip = mtod(m, struct ip *);
1056                 ip->ip_len = ntohs(ip->ip_len);
1057                 ip->ip_off = ntohs(ip->ip_off);
1058                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
1059         } else {
1060                 transport_processing_oncpu(m, hlen, ip);
1061         }
1062         return;
1063
1064 bad:
1065         m_freem(m);
1066 }
1067
1068 /*
1069  * Take incoming datagram fragment and try to reassemble it into
1070  * whole datagram.  If a chain for reassembly of this datagram already
1071  * exists, then it is given as fp; otherwise have to make a chain.
1072  *
1073  * When IPDIVERT enabled, keep additional state with each packet that
1074  * tells us if we need to divert or tee the packet we're building.
1075  * In particular, *divinfo includes the port and TEE flag.
1076  */
1077
1078 static struct mbuf *
1079 ip_reass(struct mbuf *m)
1080 {
1081         struct ip *ip = mtod(m, struct ip *);
1082         struct mbuf *p = NULL, *q, *nq;
1083         struct mbuf *n;
1084         struct ipq *fp = NULL;
1085         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1086         int i, next;
1087         u_short sum;
1088
1089         /* If maxnipq is 0, never accept fragments. */
1090         if (maxnipq == 0) {
1091                 ipstat.ips_fragments++;
1092                 ipstat.ips_fragdropped++;
1093                 m_freem(m);
1094                 return NULL;
1095         }
1096
1097         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1098         /*
1099          * Look for queue of fragments of this datagram.
1100          */
1101         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
1102                 if (ip->ip_id == fp->ipq_id &&
1103                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1104                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1105                     ip->ip_p == fp->ipq_p)
1106                         goto found;
1107
1108         fp = NULL;
1109
1110         /*
1111          * Enforce upper bound on number of fragmented packets
1112          * for which we attempt reassembly;
1113          * If maxnipq is -1, accept all fragments without limitation.
1114          */
1115         if (nipq > maxnipq && maxnipq > 0) {
1116                 /*
1117                  * drop something from the tail of the current queue
1118                  * before proceeding further
1119                  */
1120                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1121                         for (i = 0; i < IPREASS_NHASH; i++) {
1122                                 if (ipq[i].prev != &ipq[i]) {
1123                                         ipstat.ips_fragtimeout +=
1124                                             ipq[i].prev->ipq_nfrags;
1125                                         ip_freef(ipq[i].prev);
1126                                         break;
1127                                 }
1128                         }
1129                 } else {
1130                         ipstat.ips_fragtimeout +=
1131                             ipq[sum].prev->ipq_nfrags;
1132                         ip_freef(ipq[sum].prev);
1133                 }
1134         }
1135 found:
1136         /*
1137          * Adjust ip_len to not reflect header,
1138          * convert offset of this to bytes.
1139          */
1140         ip->ip_len -= hlen;
1141         if (ip->ip_off & IP_MF) {
1142                 /*
1143                  * Make sure that fragments have a data length
1144                  * that's a non-zero multiple of 8 bytes.
1145                  */
1146                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1147                         ipstat.ips_toosmall++; /* XXX */
1148                         m_freem(m);
1149                         return NULL;
1150                 }
1151                 m->m_flags |= M_FRAG;
1152         } else
1153                 m->m_flags &= ~M_FRAG;
1154         ip->ip_off <<= 3;
1155
1156         ipstat.ips_fragments++;
1157         m->m_pkthdr.header = ip;
1158
1159         /*
1160          * If the hardware has not done csum over this fragment
1161          * then csum_data is not valid at all.
1162          */
1163         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1164             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1165                 m->m_pkthdr.csum_data = 0;
1166                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1167         }
1168
1169         /*
1170          * Presence of header sizes in mbufs
1171          * would confuse code below.
1172          */
1173         m->m_data += hlen;
1174         m->m_len -= hlen;
1175
1176         /*
1177          * If first fragment to arrive, create a reassembly queue.
1178          */
1179         if (fp == NULL) {
1180                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1181                         goto dropfrag;
1182                 insque(fp, &ipq[sum]);
1183                 nipq++;
1184                 fp->ipq_nfrags = 1;
1185                 fp->ipq_ttl = IPFRAGTTL;
1186                 fp->ipq_p = ip->ip_p;
1187                 fp->ipq_id = ip->ip_id;
1188                 fp->ipq_src = ip->ip_src;
1189                 fp->ipq_dst = ip->ip_dst;
1190                 fp->ipq_frags = m;
1191                 m->m_nextpkt = NULL;
1192                 goto inserted;
1193         } else {
1194                 fp->ipq_nfrags++;
1195         }
1196
1197 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1198
1199         /*
1200          * Find a segment which begins after this one does.
1201          */
1202         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1203                 if (GETIP(q)->ip_off > ip->ip_off)
1204                         break;
1205
1206         /*
1207          * If there is a preceding segment, it may provide some of
1208          * our data already.  If so, drop the data from the incoming
1209          * segment.  If it provides all of our data, drop us, otherwise
1210          * stick new segment in the proper place.
1211          *
1212          * If some of the data is dropped from the the preceding
1213          * segment, then it's checksum is invalidated.
1214          */
1215         if (p) {
1216                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1217                 if (i > 0) {
1218                         if (i >= ip->ip_len)
1219                                 goto dropfrag;
1220                         m_adj(m, i);
1221                         m->m_pkthdr.csum_flags = 0;
1222                         ip->ip_off += i;
1223                         ip->ip_len -= i;
1224                 }
1225                 m->m_nextpkt = p->m_nextpkt;
1226                 p->m_nextpkt = m;
1227         } else {
1228                 m->m_nextpkt = fp->ipq_frags;
1229                 fp->ipq_frags = m;
1230         }
1231
1232         /*
1233          * While we overlap succeeding segments trim them or,
1234          * if they are completely covered, dequeue them.
1235          */
1236         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1237              q = nq) {
1238                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1239                 if (i < GETIP(q)->ip_len) {
1240                         GETIP(q)->ip_len -= i;
1241                         GETIP(q)->ip_off += i;
1242                         m_adj(q, i);
1243                         q->m_pkthdr.csum_flags = 0;
1244                         break;
1245                 }
1246                 nq = q->m_nextpkt;
1247                 m->m_nextpkt = nq;
1248                 ipstat.ips_fragdropped++;
1249                 fp->ipq_nfrags--;
1250                 q->m_nextpkt = NULL;
1251                 m_freem(q);
1252         }
1253
1254 inserted:
1255         /*
1256          * Check for complete reassembly and perform frag per packet
1257          * limiting.
1258          *
1259          * Frag limiting is performed here so that the nth frag has
1260          * a chance to complete the packet before we drop the packet.
1261          * As a result, n+1 frags are actually allowed per packet, but
1262          * only n will ever be stored. (n = maxfragsperpacket.)
1263          *
1264          */
1265         next = 0;
1266         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1267                 if (GETIP(q)->ip_off != next) {
1268                         if (fp->ipq_nfrags > maxfragsperpacket) {
1269                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1270                                 ip_freef(fp);
1271                         }
1272                         return (NULL);
1273                 }
1274                 next += GETIP(q)->ip_len;
1275         }
1276         /* Make sure the last packet didn't have the IP_MF flag */
1277         if (p->m_flags & M_FRAG) {
1278                 if (fp->ipq_nfrags > maxfragsperpacket) {
1279                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1280                         ip_freef(fp);
1281                 }
1282                 return (NULL);
1283         }
1284
1285         /*
1286          * Reassembly is complete.  Make sure the packet is a sane size.
1287          */
1288         q = fp->ipq_frags;
1289         ip = GETIP(q);
1290         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1291                 ipstat.ips_toolong++;
1292                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1293                 ip_freef(fp);
1294                 return (NULL);
1295         }
1296
1297         /*
1298          * Concatenate fragments.
1299          */
1300         m = q;
1301         n = m->m_next;
1302         m->m_next = NULL;
1303         m_cat(m, n);
1304         nq = q->m_nextpkt;
1305         q->m_nextpkt = NULL;
1306         for (q = nq; q != NULL; q = nq) {
1307                 nq = q->m_nextpkt;
1308                 q->m_nextpkt = NULL;
1309                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1310                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1311                 m_cat(m, q);
1312         }
1313
1314         /*
1315          * Clean up the 1's complement checksum.  Carry over 16 bits must
1316          * be added back.  This assumes no more then 65535 packet fragments
1317          * were reassembled.  A second carry can also occur (but not a third).
1318          */
1319         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1320                                 (m->m_pkthdr.csum_data >> 16);
1321         if (m->m_pkthdr.csum_data > 0xFFFF)
1322                 m->m_pkthdr.csum_data -= 0xFFFF;
1323
1324         /*
1325          * Create header for new ip packet by
1326          * modifying header of first packet;
1327          * dequeue and discard fragment reassembly header.
1328          * Make header visible.
1329          */
1330         ip->ip_len = next;
1331         ip->ip_src = fp->ipq_src;
1332         ip->ip_dst = fp->ipq_dst;
1333         remque(fp);
1334         nipq--;
1335         mpipe_free(&ipq_mpipe, fp);
1336         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1337         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1338         /* some debugging cruft by sklower, below, will go away soon */
1339         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1340                 int plen = 0;
1341
1342                 for (n = m; n; n = n->m_next)
1343                         plen += n->m_len;
1344                 m->m_pkthdr.len = plen;
1345         }
1346
1347         ipstat.ips_reassembled++;
1348         return (m);
1349
1350 dropfrag:
1351         ipstat.ips_fragdropped++;
1352         if (fp != NULL)
1353                 fp->ipq_nfrags--;
1354         m_freem(m);
1355         return (NULL);
1356
1357 #undef GETIP
1358 }
1359
1360 /*
1361  * Free a fragment reassembly header and all
1362  * associated datagrams.
1363  */
1364 static void
1365 ip_freef(struct ipq *fp)
1366 {
1367         struct mbuf *q;
1368
1369         while (fp->ipq_frags) {
1370                 q = fp->ipq_frags;
1371                 fp->ipq_frags = q->m_nextpkt;
1372                 q->m_nextpkt = NULL;
1373                 m_freem(q);
1374         }
1375         remque(fp);
1376         mpipe_free(&ipq_mpipe, fp);
1377         nipq--;
1378 }
1379
1380 /*
1381  * IP timer processing;
1382  * if a timer expires on a reassembly
1383  * queue, discard it.
1384  */
1385 void
1386 ip_slowtimo(void)
1387 {
1388         struct ipq *fp;
1389         int i;
1390
1391         crit_enter();
1392         for (i = 0; i < IPREASS_NHASH; i++) {
1393                 fp = ipq[i].next;
1394                 if (fp == NULL)
1395                         continue;
1396                 while (fp != &ipq[i]) {
1397                         --fp->ipq_ttl;
1398                         fp = fp->next;
1399                         if (fp->prev->ipq_ttl == 0) {
1400                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1401                                 ip_freef(fp->prev);
1402                         }
1403                 }
1404         }
1405         /*
1406          * If we are over the maximum number of fragments
1407          * (due to the limit being lowered), drain off
1408          * enough to get down to the new limit.
1409          */
1410         if (maxnipq >= 0 && nipq > maxnipq) {
1411                 for (i = 0; i < IPREASS_NHASH; i++) {
1412                         while (nipq > maxnipq &&
1413                                 (ipq[i].next != &ipq[i])) {
1414                                 ipstat.ips_fragdropped +=
1415                                     ipq[i].next->ipq_nfrags;
1416                                 ip_freef(ipq[i].next);
1417                         }
1418                 }
1419         }
1420         ipflow_slowtimo();
1421         crit_exit();
1422 }
1423
1424 /*
1425  * Drain off all datagram fragments.
1426  */
1427 void
1428 ip_drain(void)
1429 {
1430         int i;
1431
1432         for (i = 0; i < IPREASS_NHASH; i++) {
1433                 while (ipq[i].next != &ipq[i]) {
1434                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1435                         ip_freef(ipq[i].next);
1436                 }
1437         }
1438         in_rtqdrain();
1439 }
1440
1441 /*
1442  * Do option processing on a datagram,
1443  * possibly discarding it if bad options are encountered,
1444  * or forwarding it if source-routed.
1445  * The pass argument is used when operating in the IPSTEALTH
1446  * mode to tell what options to process:
1447  * [LS]SRR (pass 0) or the others (pass 1).
1448  * The reason for as many as two passes is that when doing IPSTEALTH,
1449  * non-routing options should be processed only if the packet is for us.
1450  * Returns 1 if packet has been forwarded/freed,
1451  * 0 if the packet should be processed further.
1452  */
1453 static int
1454 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1455 {
1456         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1457         struct ip *ip = mtod(m, struct ip *);
1458         u_char *cp;
1459         struct in_ifaddr *ia;
1460         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1461         boolean_t forward = FALSE;
1462         struct in_addr *sin, dst;
1463         n_time ntime;
1464
1465         dst = ip->ip_dst;
1466         cp = (u_char *)(ip + 1);
1467         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1468         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1469                 opt = cp[IPOPT_OPTVAL];
1470                 if (opt == IPOPT_EOL)
1471                         break;
1472                 if (opt == IPOPT_NOP)
1473                         optlen = 1;
1474                 else {
1475                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1476                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1477                                 goto bad;
1478                         }
1479                         optlen = cp[IPOPT_OLEN];
1480                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1481                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1482                                 goto bad;
1483                         }
1484                 }
1485                 switch (opt) {
1486
1487                 default:
1488                         break;
1489
1490                 /*
1491                  * Source routing with record.
1492                  * Find interface with current destination address.
1493                  * If none on this machine then drop if strictly routed,
1494                  * or do nothing if loosely routed.
1495                  * Record interface address and bring up next address
1496                  * component.  If strictly routed make sure next
1497                  * address is on directly accessible net.
1498                  */
1499                 case IPOPT_LSRR:
1500                 case IPOPT_SSRR:
1501                         if (ipstealth && pass > 0)
1502                                 break;
1503                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1504                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1505                                 goto bad;
1506                         }
1507                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1508                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1509                                 goto bad;
1510                         }
1511                         ipaddr.sin_addr = ip->ip_dst;
1512                         ia = (struct in_ifaddr *)
1513                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1514                         if (ia == NULL) {
1515                                 if (opt == IPOPT_SSRR) {
1516                                         type = ICMP_UNREACH;
1517                                         code = ICMP_UNREACH_SRCFAIL;
1518                                         goto bad;
1519                                 }
1520                                 if (!ip_dosourceroute)
1521                                         goto nosourcerouting;
1522                                 /*
1523                                  * Loose routing, and not at next destination
1524                                  * yet; nothing to do except forward.
1525                                  */
1526                                 break;
1527                         }
1528                         off--;                  /* 0 origin */
1529                         if (off > optlen - (int)sizeof(struct in_addr)) {
1530                                 /*
1531                                  * End of source route.  Should be for us.
1532                                  */
1533                                 if (!ip_acceptsourceroute)
1534                                         goto nosourcerouting;
1535                                 save_rte(m, cp, ip->ip_src);
1536                                 break;
1537                         }
1538                         if (ipstealth)
1539                                 goto dropit;
1540                         if (!ip_dosourceroute) {
1541                                 if (ipforwarding) {
1542                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1543
1544                                         /*
1545                                          * Acting as a router, so generate ICMP
1546                                          */
1547 nosourcerouting:
1548                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1549                                         log(LOG_WARNING,
1550                                             "attempted source route from %s to %s\n",
1551                                             inet_ntoa(ip->ip_src), buf);
1552                                         type = ICMP_UNREACH;
1553                                         code = ICMP_UNREACH_SRCFAIL;
1554                                         goto bad;
1555                                 } else {
1556                                         /*
1557                                          * Not acting as a router,
1558                                          * so silently drop.
1559                                          */
1560 dropit:
1561                                         ipstat.ips_cantforward++;
1562                                         m_freem(m);
1563                                         return (1);
1564                                 }
1565                         }
1566
1567                         /*
1568                          * locate outgoing interface
1569                          */
1570                         memcpy(&ipaddr.sin_addr, cp + off,
1571                             sizeof ipaddr.sin_addr);
1572
1573                         if (opt == IPOPT_SSRR) {
1574 #define INA     struct in_ifaddr *
1575 #define SA      struct sockaddr *
1576                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1577                                                                         == NULL)
1578                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1579                         } else
1580                                 ia = ip_rtaddr(ipaddr.sin_addr,
1581                                                &ipforward_rt[mycpuid]);
1582                         if (ia == NULL) {
1583                                 type = ICMP_UNREACH;
1584                                 code = ICMP_UNREACH_SRCFAIL;
1585                                 goto bad;
1586                         }
1587                         ip->ip_dst = ipaddr.sin_addr;
1588                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1589                             sizeof(struct in_addr));
1590                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1591                         /*
1592                          * Let ip_intr's mcast routing check handle mcast pkts
1593                          */
1594                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1595                         break;
1596
1597                 case IPOPT_RR:
1598                         if (ipstealth && pass == 0)
1599                                 break;
1600                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1601                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1602                                 goto bad;
1603                         }
1604                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1605                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1606                                 goto bad;
1607                         }
1608                         /*
1609                          * If no space remains, ignore.
1610                          */
1611                         off--;                  /* 0 origin */
1612                         if (off > optlen - (int)sizeof(struct in_addr))
1613                                 break;
1614                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1615                             sizeof ipaddr.sin_addr);
1616                         /*
1617                          * locate outgoing interface; if we're the destination,
1618                          * use the incoming interface (should be same).
1619                          */
1620                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1621                             (ia = ip_rtaddr(ipaddr.sin_addr,
1622                                             &ipforward_rt[mycpuid]))
1623                                                                      == NULL) {
1624                                 type = ICMP_UNREACH;
1625                                 code = ICMP_UNREACH_HOST;
1626                                 goto bad;
1627                         }
1628                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1629                             sizeof(struct in_addr));
1630                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1631                         break;
1632
1633                 case IPOPT_TS:
1634                         if (ipstealth && pass == 0)
1635                                 break;
1636                         code = cp - (u_char *)ip;
1637                         if (optlen < 4 || optlen > 40) {
1638                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1639                                 goto bad;
1640                         }
1641                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1642                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1643                                 goto bad;
1644                         }
1645                         if (off > optlen - (int)sizeof(int32_t)) {
1646                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1647                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1648                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1649                                         goto bad;
1650                                 }
1651                                 break;
1652                         }
1653                         off--;                          /* 0 origin */
1654                         sin = (struct in_addr *)(cp + off);
1655                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1656
1657                         case IPOPT_TS_TSONLY:
1658                                 break;
1659
1660                         case IPOPT_TS_TSANDADDR:
1661                                 if (off + sizeof(n_time) +
1662                                     sizeof(struct in_addr) > optlen) {
1663                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1664                                         goto bad;
1665                                 }
1666                                 ipaddr.sin_addr = dst;
1667                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1668                                                             m->m_pkthdr.rcvif);
1669                                 if (ia == NULL)
1670                                         continue;
1671                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1672                                     sizeof(struct in_addr));
1673                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1674                                 off += sizeof(struct in_addr);
1675                                 break;
1676
1677                         case IPOPT_TS_PRESPEC:
1678                                 if (off + sizeof(n_time) +
1679                                     sizeof(struct in_addr) > optlen) {
1680                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1681                                         goto bad;
1682                                 }
1683                                 memcpy(&ipaddr.sin_addr, sin,
1684                                     sizeof(struct in_addr));
1685                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1686                                         continue;
1687                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1688                                 off += sizeof(struct in_addr);
1689                                 break;
1690
1691                         default:
1692                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1693                                 goto bad;
1694                         }
1695                         ntime = iptime();
1696                         memcpy(cp + off, &ntime, sizeof(n_time));
1697                         cp[IPOPT_OFFSET] += sizeof(n_time);
1698                 }
1699         }
1700         if (forward && ipforwarding) {
1701                 ip_forward(m, TRUE, next_hop);
1702                 return (1);
1703         }
1704         return (0);
1705 bad:
1706         icmp_error(m, type, code, 0, 0);
1707         ipstat.ips_badoptions++;
1708         return (1);
1709 }
1710
1711 /*
1712  * Given address of next destination (final or next hop),
1713  * return internet address info of interface to be used to get there.
1714  */
1715 struct in_ifaddr *
1716 ip_rtaddr(struct in_addr dst, struct route *ro)
1717 {
1718         struct sockaddr_in *sin;
1719
1720         sin = (struct sockaddr_in *)&ro->ro_dst;
1721
1722         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1723                 if (ro->ro_rt != NULL) {
1724                         RTFREE(ro->ro_rt);
1725                         ro->ro_rt = NULL;
1726                 }
1727                 sin->sin_family = AF_INET;
1728                 sin->sin_len = sizeof *sin;
1729                 sin->sin_addr = dst;
1730                 rtalloc_ign(ro, RTF_PRCLONING);
1731         }
1732
1733         if (ro->ro_rt == NULL)
1734                 return (NULL);
1735
1736         return (ifatoia(ro->ro_rt->rt_ifa));
1737 }
1738
1739 /*
1740  * Save incoming source route for use in replies,
1741  * to be picked up later by ip_srcroute if the receiver is interested.
1742  */
1743 static void
1744 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1745 {
1746         struct m_tag *mtag;
1747         struct ip_srcrt_opt *opt;
1748         unsigned olen;
1749
1750         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1751         if (mtag == NULL)
1752                 return;
1753         opt = m_tag_data(mtag);
1754
1755         olen = option[IPOPT_OLEN];
1756 #ifdef DIAGNOSTIC
1757         if (ipprintfs)
1758                 kprintf("save_rte: olen %d\n", olen);
1759 #endif
1760         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1761                 m_tag_free(mtag);
1762                 return;
1763         }
1764         bcopy(option, opt->ip_srcrt.srcopt, olen);
1765         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1766         opt->ip_srcrt.dst = dst;
1767         m_tag_prepend(m, mtag);
1768 }
1769
1770 /*
1771  * Retrieve incoming source route for use in replies,
1772  * in the same form used by setsockopt.
1773  * The first hop is placed before the options, will be removed later.
1774  */
1775 struct mbuf *
1776 ip_srcroute(struct mbuf *m0)
1777 {
1778         struct in_addr *p, *q;
1779         struct mbuf *m;
1780         struct m_tag *mtag;
1781         struct ip_srcrt_opt *opt;
1782
1783         if (m0 == NULL)
1784                 return NULL;
1785
1786         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1787         if (mtag == NULL)
1788                 return NULL;
1789         opt = m_tag_data(mtag);
1790
1791         if (opt->ip_nhops == 0)
1792                 return (NULL);
1793         m = m_get(MB_DONTWAIT, MT_HEADER);
1794         if (m == NULL)
1795                 return (NULL);
1796
1797 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1798
1799         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1800         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1801                    sizeof(struct in_addr) + OPTSIZ;
1802 #ifdef DIAGNOSTIC
1803         if (ipprintfs) {
1804                 kprintf("ip_srcroute: nhops %d mlen %d",
1805                         opt->ip_nhops, m->m_len);
1806         }
1807 #endif
1808
1809         /*
1810          * First save first hop for return route
1811          */
1812         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1813         *(mtod(m, struct in_addr *)) = *p--;
1814 #ifdef DIAGNOSTIC
1815         if (ipprintfs)
1816                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1817 #endif
1818
1819         /*
1820          * Copy option fields and padding (nop) to mbuf.
1821          */
1822         opt->ip_srcrt.nop = IPOPT_NOP;
1823         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1824         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1825             OPTSIZ);
1826         q = (struct in_addr *)(mtod(m, caddr_t) +
1827             sizeof(struct in_addr) + OPTSIZ);
1828 #undef OPTSIZ
1829         /*
1830          * Record return path as an IP source route,
1831          * reversing the path (pointers are now aligned).
1832          */
1833         while (p >= opt->ip_srcrt.route) {
1834 #ifdef DIAGNOSTIC
1835                 if (ipprintfs)
1836                         kprintf(" %x", ntohl(q->s_addr));
1837 #endif
1838                 *q++ = *p--;
1839         }
1840         /*
1841          * Last hop goes to final destination.
1842          */
1843         *q = opt->ip_srcrt.dst;
1844         m_tag_delete(m0, mtag);
1845 #ifdef DIAGNOSTIC
1846         if (ipprintfs)
1847                 kprintf(" %x\n", ntohl(q->s_addr));
1848 #endif
1849         return (m);
1850 }
1851
1852 /*
1853  * Strip out IP options.
1854  */
1855 void
1856 ip_stripoptions(struct mbuf *m)
1857 {
1858         int datalen;
1859         struct ip *ip = mtod(m, struct ip *);
1860         caddr_t opts;
1861         int optlen;
1862
1863         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1864         opts = (caddr_t)(ip + 1);
1865         datalen = m->m_len - (sizeof(struct ip) + optlen);
1866         bcopy(opts + optlen, opts, datalen);
1867         m->m_len -= optlen;
1868         if (m->m_flags & M_PKTHDR)
1869                 m->m_pkthdr.len -= optlen;
1870         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1871 }
1872
1873 u_char inetctlerrmap[PRC_NCMDS] = {
1874         0,              0,              0,              0,
1875         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1876         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1877         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1878         0,              0,              0,              0,
1879         ENOPROTOOPT,    ECONNREFUSED
1880 };
1881
1882 /*
1883  * Forward a packet.  If some error occurs return the sender
1884  * an icmp packet.  Note we can't always generate a meaningful
1885  * icmp message because icmp doesn't have a large enough repertoire
1886  * of codes and types.
1887  *
1888  * If not forwarding, just drop the packet.  This could be confusing
1889  * if ipforwarding was zero but some routing protocol was advancing
1890  * us as a gateway to somewhere.  However, we must let the routing
1891  * protocol deal with that.
1892  *
1893  * The using_srcrt parameter indicates whether the packet is being forwarded
1894  * via a source route.
1895  */
1896 void
1897 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1898 {
1899         struct ip *ip = mtod(m, struct ip *);
1900         struct sockaddr_in *ipforward_rtaddr;
1901         struct rtentry *rt;
1902         int error, type = 0, code = 0, destmtu = 0;
1903         struct mbuf *mcopy;
1904         n_long dest;
1905         struct in_addr pkt_dst;
1906         struct route *cache_rt = &ipforward_rt[mycpuid];
1907
1908         dest = INADDR_ANY;
1909         /*
1910          * Cache the destination address of the packet; this may be
1911          * changed by use of 'ipfw fwd'.
1912          */
1913         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1914
1915 #ifdef DIAGNOSTIC
1916         if (ipprintfs)
1917                 kprintf("forward: src %x dst %x ttl %x\n",
1918                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1919 #endif
1920
1921         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1922                 ipstat.ips_cantforward++;
1923                 m_freem(m);
1924                 return;
1925         }
1926         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1927                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1928                 return;
1929         }
1930
1931         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1932         if (cache_rt->ro_rt == NULL ||
1933             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1934                 if (cache_rt->ro_rt != NULL) {
1935                         RTFREE(cache_rt->ro_rt);
1936                         cache_rt->ro_rt = NULL;
1937                 }
1938                 ipforward_rtaddr->sin_family = AF_INET;
1939                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1940                 ipforward_rtaddr->sin_addr = pkt_dst;
1941                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1942                 if (cache_rt->ro_rt == NULL) {
1943                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1944                         return;
1945                 }
1946         }
1947         rt = cache_rt->ro_rt;
1948
1949         /*
1950          * Save the IP header and at most 8 bytes of the payload,
1951          * in case we need to generate an ICMP message to the src.
1952          *
1953          * XXX this can be optimized a lot by saving the data in a local
1954          * buffer on the stack (72 bytes at most), and only allocating the
1955          * mbuf if really necessary. The vast majority of the packets
1956          * are forwarded without having to send an ICMP back (either
1957          * because unnecessary, or because rate limited), so we are
1958          * really we are wasting a lot of work here.
1959          *
1960          * We don't use m_copy() because it might return a reference
1961          * to a shared cluster. Both this function and ip_output()
1962          * assume exclusive access to the IP header in `m', so any
1963          * data in a cluster may change before we reach icmp_error().
1964          */
1965         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1966         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1967                 /*
1968                  * It's probably ok if the pkthdr dup fails (because
1969                  * the deep copy of the tag chain failed), but for now
1970                  * be conservative and just discard the copy since
1971                  * code below may some day want the tags.
1972                  */
1973                 m_free(mcopy);
1974                 mcopy = NULL;
1975         }
1976         if (mcopy != NULL) {
1977                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1978                     (int)ip->ip_len);
1979                 mcopy->m_pkthdr.len = mcopy->m_len;
1980                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1981         }
1982
1983         if (!ipstealth)
1984                 ip->ip_ttl -= IPTTLDEC;
1985
1986         /*
1987          * If forwarding packet using same interface that it came in on,
1988          * perhaps should send a redirect to sender to shortcut a hop.
1989          * Only send redirect if source is sending directly to us,
1990          * and if packet was not source routed (or has any options).
1991          * Also, don't send redirect if forwarding using a default route
1992          * or a route modified by a redirect.
1993          */
1994         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1995             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1996             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1997             ipsendredirects && !using_srcrt && next_hop == NULL) {
1998                 u_long src = ntohl(ip->ip_src.s_addr);
1999                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
2000
2001                 if (rt_ifa != NULL &&
2002                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
2003                         if (rt->rt_flags & RTF_GATEWAY)
2004                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
2005                         else
2006                                 dest = pkt_dst.s_addr;
2007                         /*
2008                          * Router requirements says to only send
2009                          * host redirects.
2010                          */
2011                         type = ICMP_REDIRECT;
2012                         code = ICMP_REDIRECT_HOST;
2013 #ifdef DIAGNOSTIC
2014                         if (ipprintfs)
2015                                 kprintf("redirect (%d) to %x\n", code, dest);
2016 #endif
2017                 }
2018         }
2019
2020         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
2021         if (error == 0) {
2022                 ipstat.ips_forward++;
2023                 if (type == 0) {
2024                         if (mcopy) {
2025                                 ipflow_create(cache_rt, mcopy);
2026                                 m_freem(mcopy);
2027                         }
2028                         return;         /* most common case */
2029                 } else {
2030                         ipstat.ips_redirectsent++;
2031                 }
2032         } else {
2033                 ipstat.ips_cantforward++;
2034         }
2035
2036         if (mcopy == NULL)
2037                 return;
2038
2039         /*
2040          * Send ICMP message.
2041          */
2042
2043         switch (error) {
2044
2045         case 0:                         /* forwarded, but need redirect */
2046                 /* type, code set above */
2047                 break;
2048
2049         case ENETUNREACH:               /* shouldn't happen, checked above */
2050         case EHOSTUNREACH:
2051         case ENETDOWN:
2052         case EHOSTDOWN:
2053         default:
2054                 type = ICMP_UNREACH;
2055                 code = ICMP_UNREACH_HOST;
2056                 break;
2057
2058         case EMSGSIZE:
2059                 type = ICMP_UNREACH;
2060                 code = ICMP_UNREACH_NEEDFRAG;
2061 #ifdef IPSEC
2062                 /*
2063                  * If the packet is routed over IPsec tunnel, tell the
2064                  * originator the tunnel MTU.
2065                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2066                  * XXX quickhack!!!
2067                  */
2068                 if (cache_rt->ro_rt != NULL) {
2069                         struct secpolicy *sp = NULL;
2070                         int ipsecerror;
2071                         int ipsechdr;
2072                         struct route *ro;
2073
2074                         sp = ipsec4_getpolicybyaddr(mcopy,
2075                                                     IPSEC_DIR_OUTBOUND,
2076                                                     IP_FORWARDING,
2077                                                     &ipsecerror);
2078
2079                         if (sp == NULL)
2080                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2081                         else {
2082                                 /* count IPsec header size */
2083                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2084                                                          IPSEC_DIR_OUTBOUND,
2085                                                          NULL);
2086
2087                                 /*
2088                                  * find the correct route for outer IPv4
2089                                  * header, compute tunnel MTU.
2090                                  *
2091                                  */
2092                                 if (sp->req != NULL && sp->req->sav != NULL &&
2093                                     sp->req->sav->sah != NULL) {
2094                                         ro = &sp->req->sav->sah->sa_route;
2095                                         if (ro->ro_rt != NULL &&
2096                                             ro->ro_rt->rt_ifp != NULL) {
2097                                                 destmtu =
2098                                                     ro->ro_rt->rt_ifp->if_mtu;
2099                                                 destmtu -= ipsechdr;
2100                                         }
2101                                 }
2102
2103                                 key_freesp(sp);
2104                         }
2105                 }
2106 #elif FAST_IPSEC
2107                 /*
2108                  * If the packet is routed over IPsec tunnel, tell the
2109                  * originator the tunnel MTU.
2110                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2111                  * XXX quickhack!!!
2112                  */
2113                 if (cache_rt->ro_rt != NULL) {
2114                         struct secpolicy *sp = NULL;
2115                         int ipsecerror;
2116                         int ipsechdr;
2117                         struct route *ro;
2118
2119                         sp = ipsec_getpolicybyaddr(mcopy,
2120                                                    IPSEC_DIR_OUTBOUND,
2121                                                    IP_FORWARDING,
2122                                                    &ipsecerror);
2123
2124                         if (sp == NULL)
2125                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2126                         else {
2127                                 /* count IPsec header size */
2128                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2129                                                          IPSEC_DIR_OUTBOUND,
2130                                                          NULL);
2131
2132                                 /*
2133                                  * find the correct route for outer IPv4
2134                                  * header, compute tunnel MTU.
2135                                  */
2136
2137                                 if (sp->req != NULL &&
2138                                     sp->req->sav != NULL &&
2139                                     sp->req->sav->sah != NULL) {
2140                                         ro = &sp->req->sav->sah->sa_route;
2141                                         if (ro->ro_rt != NULL &&
2142                                             ro->ro_rt->rt_ifp != NULL) {
2143                                                 destmtu =
2144                                                     ro->ro_rt->rt_ifp->if_mtu;
2145                                                 destmtu -= ipsechdr;
2146                                         }
2147                                 }
2148
2149                                 KEY_FREESP(&sp);
2150                         }
2151                 }
2152 #else /* !IPSEC && !FAST_IPSEC */
2153                 if (cache_rt->ro_rt != NULL)
2154                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2155 #endif /*IPSEC*/
2156                 ipstat.ips_cantfrag++;
2157                 break;
2158
2159         case ENOBUFS:
2160                 /*
2161                  * A router should not generate ICMP_SOURCEQUENCH as
2162                  * required in RFC1812 Requirements for IP Version 4 Routers.
2163                  * Source quench could be a big problem under DoS attacks,
2164                  * or if the underlying interface is rate-limited.
2165                  * Those who need source quench packets may re-enable them
2166                  * via the net.inet.ip.sendsourcequench sysctl.
2167                  */
2168                 if (!ip_sendsourcequench) {
2169                         m_freem(mcopy);
2170                         return;
2171                 } else {
2172                         type = ICMP_SOURCEQUENCH;
2173                         code = 0;
2174                 }
2175                 break;
2176
2177         case EACCES:                    /* ipfw denied packet */
2178                 m_freem(mcopy);
2179                 return;
2180         }
2181         icmp_error(mcopy, type, code, dest, destmtu);
2182 }
2183
2184 void
2185 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2186                struct mbuf *m)
2187 {
2188         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2189                 struct timeval tv;
2190
2191                 microtime(&tv);
2192                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2193                     SCM_TIMESTAMP, SOL_SOCKET);
2194                 if (*mp)
2195                         mp = &(*mp)->m_next;
2196         }
2197         if (inp->inp_flags & INP_RECVDSTADDR) {
2198                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2199                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2200                 if (*mp)
2201                         mp = &(*mp)->m_next;
2202         }
2203         if (inp->inp_flags & INP_RECVTTL) {
2204                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2205                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2206                 if (*mp)
2207                         mp = &(*mp)->m_next;
2208         }
2209 #ifdef notyet
2210         /* XXX
2211          * Moving these out of udp_input() made them even more broken
2212          * than they already were.
2213          */
2214         /* options were tossed already */
2215         if (inp->inp_flags & INP_RECVOPTS) {
2216                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2217                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2218                 if (*mp)
2219                         mp = &(*mp)->m_next;
2220         }
2221         /* ip_srcroute doesn't do what we want here, need to fix */
2222         if (inp->inp_flags & INP_RECVRETOPTS) {
2223                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2224                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2225                 if (*mp)
2226                         mp = &(*mp)->m_next;
2227         }
2228 #endif
2229         if (inp->inp_flags & INP_RECVIF) {
2230                 struct ifnet *ifp;
2231                 struct sdlbuf {
2232                         struct sockaddr_dl sdl;
2233                         u_char  pad[32];
2234                 } sdlbuf;
2235                 struct sockaddr_dl *sdp;
2236                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2237
2238                 if (((ifp = m->m_pkthdr.rcvif)) &&
2239                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2240                         sdp = IF_LLSOCKADDR(ifp);
2241                         /*
2242                          * Change our mind and don't try copy.
2243                          */
2244                         if ((sdp->sdl_family != AF_LINK) ||
2245                             (sdp->sdl_len > sizeof(sdlbuf))) {
2246                                 goto makedummy;
2247                         }
2248                         bcopy(sdp, sdl2, sdp->sdl_len);
2249                 } else {
2250 makedummy:
2251                         sdl2->sdl_len =
2252                             offsetof(struct sockaddr_dl, sdl_data[0]);
2253                         sdl2->sdl_family = AF_LINK;
2254                         sdl2->sdl_index = 0;
2255                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2256                 }
2257                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2258                         IP_RECVIF, IPPROTO_IP);
2259                 if (*mp)
2260                         mp = &(*mp)->m_next;
2261         }
2262 }
2263
2264 /*
2265  * XXX these routines are called from the upper part of the kernel.
2266  *
2267  * They could also be moved to ip_mroute.c, since all the RSVP
2268  *  handling is done there already.
2269  */
2270 int
2271 ip_rsvp_init(struct socket *so)
2272 {
2273         if (so->so_type != SOCK_RAW ||
2274             so->so_proto->pr_protocol != IPPROTO_RSVP)
2275                 return EOPNOTSUPP;
2276
2277         if (ip_rsvpd != NULL)
2278                 return EADDRINUSE;
2279
2280         ip_rsvpd = so;
2281         /*
2282          * This may seem silly, but we need to be sure we don't over-increment
2283          * the RSVP counter, in case something slips up.
2284          */
2285         if (!ip_rsvp_on) {
2286                 ip_rsvp_on = 1;
2287                 rsvp_on++;
2288         }
2289
2290         return 0;
2291 }
2292
2293 int
2294 ip_rsvp_done(void)
2295 {
2296         ip_rsvpd = NULL;
2297         /*
2298          * This may seem silly, but we need to be sure we don't over-decrement
2299          * the RSVP counter, in case something slips up.
2300          */
2301         if (ip_rsvp_on) {
2302                 ip_rsvp_on = 0;
2303                 rsvp_on--;
2304         }
2305         return 0;
2306 }
2307
2308 void
2309 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2310 {
2311         int off, proto;
2312         __va_list ap;
2313
2314         __va_start(ap, m);
2315         off = __va_arg(ap, int);
2316         proto = __va_arg(ap, int);
2317         __va_end(ap);
2318
2319         if (rsvp_input_p) { /* call the real one if loaded */
2320                 rsvp_input_p(m, off, proto);
2321                 return;
2322         }
2323
2324         /* Can still get packets with rsvp_on = 0 if there is a local member
2325          * of the group to which the RSVP packet is addressed.  But in this
2326          * case we want to throw the packet away.
2327          */
2328
2329         if (!rsvp_on) {
2330                 m_freem(m);
2331                 return;
2332         }
2333
2334         if (ip_rsvpd != NULL) {
2335                 rip_input(m, off, proto);
2336                 return;
2337         }
2338         /* Drop the packet */
2339         m_freem(m);
2340 }