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 1.97 2008/09/06 12:04:03 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         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 426         boolean_t needredispatch = FALSE;
 427         struct in_addr odst;                    /* original dst address(NAT) */
 428         struct m_tag *mtag;
 429         struct sockaddr_in *next_hop = NULL;
 430 #ifdef FAST_IPSEC
 431         struct tdb_ident *tdbi;
 432         struct secpolicy *sp;
 433         int error;
 434 #endif
 435
 436         M_ASSERTPKTHDR(m);
 437
 438         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 439                 /* Next hop */
 440                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 441                 KKASSERT(mtag != NULL);
 442                 next_hop = m_tag_data(mtag);
 443         }
 444
 445         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 446                 /* dummynet already filtered us */
 447                 ip = mtod(m, struct ip *);
 448                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 449                 goto iphack;
 450         }
 451
 452         ipstat.ips_total++;
 453
 454         /* length checks already done in ip_demux() */
 455         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
 456
 457         ip = mtod(m, struct ip *);
 458
 459         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 460                 ipstat.ips_badvers++;
 461                 goto bad;
 462         }
 463
 464         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 465         /* length checks already done in ip_demux() */
 466         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 467         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
 468
 469         /* 127/8 must not appear on wire - RFC1122 */
 470         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 471             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 472                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 473                         ipstat.ips_badaddr++;
 474                         goto bad;
 475                 }
 476         }
 477
 478         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 479                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 480         } else {
 481                 if (hlen == sizeof(struct ip)) {
 482                         sum = in_cksum_hdr(ip);
 483                 } else {
 484                         sum = in_cksum(m, hlen);
 485                 }
 486         }
 487         if (sum != 0) {
 488                 ipstat.ips_badsum++;
 489                 goto bad;
 490         }
 491
 492 #ifdef ALTQ
 493         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 494                 /* packet is dropped by traffic conditioner */
 495                 return;
 496         }
 497 #endif
 498         /*
 499          * Convert fields to host representation.
 500          */
 501         ip->ip_len = ntohs(ip->ip_len);
 502         if (ip->ip_len < hlen) {
 503                 ipstat.ips_badlen++;
 504                 goto bad;
 505         }
 506         ip->ip_off = ntohs(ip->ip_off);
 507
 508         /*
 509          * Check that the amount of data in the buffers
 510          * is as at least much as the IP header would have us expect.
 511          * Trim mbufs if longer than we expect.
 512          * Drop packet if shorter than we expect.
 513          */
 514         if (m->m_pkthdr.len < ip->ip_len) {
 515                 ipstat.ips_tooshort++;
 516                 goto bad;
 517         }
 518         if (m->m_pkthdr.len > ip->ip_len) {
 519                 if (m->m_len == m->m_pkthdr.len) {
 520                         m->m_len = ip->ip_len;
 521                         m->m_pkthdr.len = ip->ip_len;
 522                 } else
 523                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 524         }
 525 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 526         /*
 527          * Bypass packet filtering for packets from a tunnel (gif).
 528          */
 529         if (ipsec_gethist(m, NULL))
 530                 goto pass;
 531 #endif
 532
 533         /*
 534          * IpHack's section.
 535          * Right now when no processing on packet has done
 536          * and it is still fresh out of network we do our black
 537          * deals with it.
 538          * - Firewall: deny/allow/divert
 539          * - Xlate: translate packet's addr/port (NAT).
 540          * - Pipe: pass pkt through dummynet.
 541          * - Wrap: fake packet's addr/port <unimpl.>
 542          * - Encapsulate: put it in another IP and send out. <unimp.>
 543          */
 544
 545 iphack:
 546
 547         /*
 548          * Run through list of hooks for input packets.
 549          *
 550          * NB: Beware of the destination address changing (e.g.
 551          *     by NAT rewriting). When this happens, tell
 552          *     ip_forward to do the right thing.
 553          */
 554         if (pfil_has_hooks(&inet_pfil_hook)) {
 555                 odst = ip->ip_dst;
 556                 if (pfil_run_hooks(&inet_pfil_hook, &m,
 557                     m->m_pkthdr.rcvif, PFIL_IN)) {
 558                         return;
 559                 }
 560                 if (m == NULL)                  /* consumed by filter */
 561                         return;
 562                 ip = mtod(m, struct ip *);
 563                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 564         }
 565
 566         if (fw_enable && IPFW_LOADED) {
 567                 struct ip_fw_args args;
 568
 569                 /*
 570                  * If we've been forwarded from the output side, then
 571                  * skip the firewall a second time
 572                  */
 573                 if (next_hop != NULL)
 574                         goto ours;
 575
 576                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 577                         /* Extract info from dummynet tag */
 578                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 579                         KKASSERT(mtag != NULL);
 580                         args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
 581                         KKASSERT(args.rule != NULL);
 582
 583                         m_tag_delete(m, mtag);
 584                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 585                 } else {
 586                         args.rule = NULL;
 587                 }
 588
 589                 args.eh = NULL;
 590                 args.oif = NULL;
 591                 args.m = m;
 592                 i = ip_fw_chk_ptr(&args);
 593                 m = args.m;
 594
 595                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
 596                         if (m != NULL)
 597                                 m_freem(m);
 598                         return;
 599                 }
 600                 ip = mtod(m, struct ip *);      /* just in case m changed */
 601
 602                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 603                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 604                         KKASSERT(mtag != NULL);
 605                         next_hop = m_tag_data(mtag);
 606                 }
 607
 608                 if (i == 0 && next_hop == NULL) /* common case */
 609                         goto pass;
 610                 if (i & IP_FW_PORT_DYNT_FLAG) {
 611                         /* Send packet to the appropriate pipe */
 612                         ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
 613                         return;
 614                 }
 615 #ifdef IPDIVERT
 616                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
 617                         struct mbuf *clone = NULL;
 618                         int tee = 0;
 619
 620                         /* Divert or 'tee'? */
 621                         if (i & IP_FW_PORT_TEE_FLAG)
 622                                 tee = 1;
 623
 624                         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 625                                 const struct divert_info *divinfo;
 626                                 u_short frag_off;
 627
 628                                 /*
 629                                  * Only trust divert info in the fragment
 630                                  * at offset 0.
 631                                  */
 632                                 frag_off = ip->ip_off << 3;
 633                                 if (frag_off != 0) {
 634                                         mtag = m_tag_find(m,
 635                                         PACKET_TAG_IPFW_DIVERT, NULL);
 636                                         m_tag_delete(m, mtag);
 637                                 }
 638
 639                                 /*
 640                                  * Attempt reassembly; if it succeeds, proceed.
 641                                  * ip_reass() will return a different mbuf.
 642                                  */
 643                                 m = ip_reass(m);
 644                                 if (m == NULL)
 645                                         return;
 646                                 ip = mtod(m, struct ip *);
 647
 648                                 needredispatch = TRUE;
 649
 650                                 /*
 651                                  * Get the header length of the reassembled
 652                                  * packet
 653                                  */
 654                                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 655
 656                                 /*
 657                                  * Restore original checksum before diverting
 658                                  * packet
 659                                  */
 660                                 ip->ip_len += hlen;
 661                                 ip->ip_len = htons(ip->ip_len);
 662                                 ip->ip_off = htons(ip->ip_off);
 663                                 ip->ip_sum = 0;
 664                                 if (hlen == sizeof(struct ip))
 665                                         ip->ip_sum = in_cksum_hdr(ip);
 666                                 else
 667                                         ip->ip_sum = in_cksum(m, hlen);
 668                                 ip->ip_off = ntohs(ip->ip_off);
 669                                 ip->ip_len = ntohs(ip->ip_len);
 670
 671                                 /*
 672                                  * Only use the saved divert info
 673                                  */
 674                                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT,
 675                                                   NULL);
 676                                 if (mtag == NULL) {
 677                                         /* Wrongly configured ipfw */
 678                                         kprintf("ip_input no divert info\n");
 679                                         m_freem(m);
 680                                         return;
 681                                 }
 682                                 divinfo = m_tag_data(mtag);
 683                                 tee = divinfo->tee;
 684                         }
 685
 686                         /*
 687                          * Divert or tee packet to the divert protocol if
 688                          * required.
 689                          */
 690
 691                         /* Clone packet if we're doing a 'tee' */
 692                         if (tee)
 693                                 clone = m_dup(m, MB_DONTWAIT);
 694
 695                         /*
 696                          * Restore packet header fields to original
 697                          * values
 698                          */
 699                         ip->ip_len = htons(ip->ip_len);
 700                         ip->ip_off = htons(ip->ip_off);
 701
 702                         /* Deliver packet to divert input routine */
 703                         divert_packet(m, 1);
 704                         ipstat.ips_delivered++;
 705
 706                         /* If 'tee', continue with original packet */
 707                         if (clone == NULL)
 708                                 return;
 709                         m = clone;
 710                         ip = mtod(m, struct ip *);
 711
 712                         /*
 713                          * Complete processing of the packet.
 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 static struct mbuf *
1074 ip_reass(struct mbuf *m)
1075 {
1076         struct ip *ip = mtod(m, struct ip *);
1077         struct mbuf *p = NULL, *q, *nq;
1078         struct mbuf *n;
1079         struct ipq *fp = NULL;
1080         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1081         int i, next;
1082         u_short sum;
1083
1084         /* If maxnipq is 0, never accept fragments. */
1085         if (maxnipq == 0) {
1086                 ipstat.ips_fragments++;
1087                 ipstat.ips_fragdropped++;
1088                 m_freem(m);
1089                 return NULL;
1090         }
1091
1092         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1093         /*
1094          * Look for queue of fragments of this datagram.
1095          */
1096         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
1097                 if (ip->ip_id == fp->ipq_id &&
1098                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1099                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1100                     ip->ip_p == fp->ipq_p)
1101                         goto found;
1102
1103         fp = NULL;
1104
1105         /*
1106          * Enforce upper bound on number of fragmented packets
1107          * for which we attempt reassembly;
1108          * If maxnipq is -1, accept all fragments without limitation.
1109          */
1110         if (nipq > maxnipq && maxnipq > 0) {
1111                 /*
1112                  * drop something from the tail of the current queue
1113                  * before proceeding further
1114                  */
1115                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1116                         for (i = 0; i < IPREASS_NHASH; i++) {
1117                                 if (ipq[i].prev != &ipq[i]) {
1118                                         ipstat.ips_fragtimeout +=
1119                                             ipq[i].prev->ipq_nfrags;
1120                                         ip_freef(ipq[i].prev);
1121                                         break;
1122                                 }
1123                         }
1124                 } else {
1125                         ipstat.ips_fragtimeout +=
1126                             ipq[sum].prev->ipq_nfrags;
1127                         ip_freef(ipq[sum].prev);
1128                 }
1129         }
1130 found:
1131         /*
1132          * Adjust ip_len to not reflect header,
1133          * convert offset of this to bytes.
1134          */
1135         ip->ip_len -= hlen;
1136         if (ip->ip_off & IP_MF) {
1137                 /*
1138                  * Make sure that fragments have a data length
1139                  * that's a non-zero multiple of 8 bytes.
1140                  */
1141                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1142                         ipstat.ips_toosmall++; /* XXX */
1143                         m_freem(m);
1144                         return NULL;
1145                 }
1146                 m->m_flags |= M_FRAG;
1147         } else
1148                 m->m_flags &= ~M_FRAG;
1149         ip->ip_off <<= 3;
1150
1151         ipstat.ips_fragments++;
1152         m->m_pkthdr.header = ip;
1153
1154         /*
1155          * If the hardware has not done csum over this fragment
1156          * then csum_data is not valid at all.
1157          */
1158         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1159             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1160                 m->m_pkthdr.csum_data = 0;
1161                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1162         }
1163
1164         /*
1165          * Presence of header sizes in mbufs
1166          * would confuse code below.
1167          */
1168         m->m_data += hlen;
1169         m->m_len -= hlen;
1170
1171         /*
1172          * If first fragment to arrive, create a reassembly queue.
1173          */
1174         if (fp == NULL) {
1175                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1176                         goto dropfrag;
1177                 insque(fp, &ipq[sum]);
1178                 nipq++;
1179                 fp->ipq_nfrags = 1;
1180                 fp->ipq_ttl = IPFRAGTTL;
1181                 fp->ipq_p = ip->ip_p;
1182                 fp->ipq_id = ip->ip_id;
1183                 fp->ipq_src = ip->ip_src;
1184                 fp->ipq_dst = ip->ip_dst;
1185                 fp->ipq_frags = m;
1186                 m->m_nextpkt = NULL;
1187                 goto inserted;
1188         } else {
1189                 fp->ipq_nfrags++;
1190         }
1191
1192 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1193
1194         /*
1195          * Find a segment which begins after this one does.
1196          */
1197         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1198                 if (GETIP(q)->ip_off > ip->ip_off)
1199                         break;
1200
1201         /*
1202          * If there is a preceding segment, it may provide some of
1203          * our data already.  If so, drop the data from the incoming
1204          * segment.  If it provides all of our data, drop us, otherwise
1205          * stick new segment in the proper place.
1206          *
1207          * If some of the data is dropped from the the preceding
1208          * segment, then it's checksum is invalidated.
1209          */
1210         if (p) {
1211                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1212                 if (i > 0) {
1213                         if (i >= ip->ip_len)
1214                                 goto dropfrag;
1215                         m_adj(m, i);
1216                         m->m_pkthdr.csum_flags = 0;
1217                         ip->ip_off += i;
1218                         ip->ip_len -= i;
1219                 }
1220                 m->m_nextpkt = p->m_nextpkt;
1221                 p->m_nextpkt = m;
1222         } else {
1223                 m->m_nextpkt = fp->ipq_frags;
1224                 fp->ipq_frags = m;
1225         }
1226
1227         /*
1228          * While we overlap succeeding segments trim them or,
1229          * if they are completely covered, dequeue them.
1230          */
1231         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1232              q = nq) {
1233                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1234                 if (i < GETIP(q)->ip_len) {
1235                         GETIP(q)->ip_len -= i;
1236                         GETIP(q)->ip_off += i;
1237                         m_adj(q, i);
1238                         q->m_pkthdr.csum_flags = 0;
1239                         break;
1240                 }
1241                 nq = q->m_nextpkt;
1242                 m->m_nextpkt = nq;
1243                 ipstat.ips_fragdropped++;
1244                 fp->ipq_nfrags--;
1245                 q->m_nextpkt = NULL;
1246                 m_freem(q);
1247         }
1248
1249 inserted:
1250         /*
1251          * Check for complete reassembly and perform frag per packet
1252          * limiting.
1253          *
1254          * Frag limiting is performed here so that the nth frag has
1255          * a chance to complete the packet before we drop the packet.
1256          * As a result, n+1 frags are actually allowed per packet, but
1257          * only n will ever be stored. (n = maxfragsperpacket.)
1258          *
1259          */
1260         next = 0;
1261         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1262                 if (GETIP(q)->ip_off != next) {
1263                         if (fp->ipq_nfrags > maxfragsperpacket) {
1264                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1265                                 ip_freef(fp);
1266                         }
1267                         return (NULL);
1268                 }
1269                 next += GETIP(q)->ip_len;
1270         }
1271         /* Make sure the last packet didn't have the IP_MF flag */
1272         if (p->m_flags & M_FRAG) {
1273                 if (fp->ipq_nfrags > maxfragsperpacket) {
1274                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1275                         ip_freef(fp);
1276                 }
1277                 return (NULL);
1278         }
1279
1280         /*
1281          * Reassembly is complete.  Make sure the packet is a sane size.
1282          */
1283         q = fp->ipq_frags;
1284         ip = GETIP(q);
1285         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1286                 ipstat.ips_toolong++;
1287                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1288                 ip_freef(fp);
1289                 return (NULL);
1290         }
1291
1292         /*
1293          * Concatenate fragments.
1294          */
1295         m = q;
1296         n = m->m_next;
1297         m->m_next = NULL;
1298         m_cat(m, n);
1299         nq = q->m_nextpkt;
1300         q->m_nextpkt = NULL;
1301         for (q = nq; q != NULL; q = nq) {
1302                 nq = q->m_nextpkt;
1303                 q->m_nextpkt = NULL;
1304                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1305                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1306                 m_cat(m, q);
1307         }
1308
1309         /*
1310          * Clean up the 1's complement checksum.  Carry over 16 bits must
1311          * be added back.  This assumes no more then 65535 packet fragments
1312          * were reassembled.  A second carry can also occur (but not a third).
1313          */
1314         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1315                                 (m->m_pkthdr.csum_data >> 16);
1316         if (m->m_pkthdr.csum_data > 0xFFFF)
1317                 m->m_pkthdr.csum_data -= 0xFFFF;
1318
1319         /*
1320          * Create header for new ip packet by
1321          * modifying header of first packet;
1322          * dequeue and discard fragment reassembly header.
1323          * Make header visible.
1324          */
1325         ip->ip_len = next;
1326         ip->ip_src = fp->ipq_src;
1327         ip->ip_dst = fp->ipq_dst;
1328         remque(fp);
1329         nipq--;
1330         mpipe_free(&ipq_mpipe, fp);
1331         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1332         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1333         /* some debugging cruft by sklower, below, will go away soon */
1334         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1335                 int plen = 0;
1336
1337                 for (n = m; n; n = n->m_next)
1338                         plen += n->m_len;
1339                 m->m_pkthdr.len = plen;
1340         }
1341
1342         ipstat.ips_reassembled++;
1343         return (m);
1344
1345 dropfrag:
1346         ipstat.ips_fragdropped++;
1347         if (fp != NULL)
1348                 fp->ipq_nfrags--;
1349         m_freem(m);
1350         return (NULL);
1351
1352 #undef GETIP
1353 }
1354
1355 /*
1356  * Free a fragment reassembly header and all
1357  * associated datagrams.
1358  */
1359 static void
1360 ip_freef(struct ipq *fp)
1361 {
1362         struct mbuf *q;
1363
1364         while (fp->ipq_frags) {
1365                 q = fp->ipq_frags;
1366                 fp->ipq_frags = q->m_nextpkt;
1367                 q->m_nextpkt = NULL;
1368                 m_freem(q);
1369         }
1370         remque(fp);
1371         mpipe_free(&ipq_mpipe, fp);
1372         nipq--;
1373 }
1374
1375 /*
1376  * IP timer processing;
1377  * if a timer expires on a reassembly
1378  * queue, discard it.
1379  */
1380 void
1381 ip_slowtimo(void)
1382 {
1383         struct ipq *fp;
1384         int i;
1385
1386         crit_enter();
1387         for (i = 0; i < IPREASS_NHASH; i++) {
1388                 fp = ipq[i].next;
1389                 if (fp == NULL)
1390                         continue;
1391                 while (fp != &ipq[i]) {
1392                         --fp->ipq_ttl;
1393                         fp = fp->next;
1394                         if (fp->prev->ipq_ttl == 0) {
1395                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1396                                 ip_freef(fp->prev);
1397                         }
1398                 }
1399         }
1400         /*
1401          * If we are over the maximum number of fragments
1402          * (due to the limit being lowered), drain off
1403          * enough to get down to the new limit.
1404          */
1405         if (maxnipq >= 0 && nipq > maxnipq) {
1406                 for (i = 0; i < IPREASS_NHASH; i++) {
1407                         while (nipq > maxnipq &&
1408                                 (ipq[i].next != &ipq[i])) {
1409                                 ipstat.ips_fragdropped +=
1410                                     ipq[i].next->ipq_nfrags;
1411                                 ip_freef(ipq[i].next);
1412                         }
1413                 }
1414         }
1415         ipflow_slowtimo();
1416         crit_exit();
1417 }
1418
1419 /*
1420  * Drain off all datagram fragments.
1421  */
1422 void
1423 ip_drain(void)
1424 {
1425         int i;
1426
1427         for (i = 0; i < IPREASS_NHASH; i++) {
1428                 while (ipq[i].next != &ipq[i]) {
1429                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1430                         ip_freef(ipq[i].next);
1431                 }
1432         }
1433         in_rtqdrain();
1434 }
1435
1436 /*
1437  * Do option processing on a datagram,
1438  * possibly discarding it if bad options are encountered,
1439  * or forwarding it if source-routed.
1440  * The pass argument is used when operating in the IPSTEALTH
1441  * mode to tell what options to process:
1442  * [LS]SRR (pass 0) or the others (pass 1).
1443  * The reason for as many as two passes is that when doing IPSTEALTH,
1444  * non-routing options should be processed only if the packet is for us.
1445  * Returns 1 if packet has been forwarded/freed,
1446  * 0 if the packet should be processed further.
1447  */
1448 static int
1449 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1450 {
1451         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1452         struct ip *ip = mtod(m, struct ip *);
1453         u_char *cp;
1454         struct in_ifaddr *ia;
1455         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1456         boolean_t forward = FALSE;
1457         struct in_addr *sin, dst;
1458         n_time ntime;
1459
1460         dst = ip->ip_dst;
1461         cp = (u_char *)(ip + 1);
1462         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1463         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1464                 opt = cp[IPOPT_OPTVAL];
1465                 if (opt == IPOPT_EOL)
1466                         break;
1467                 if (opt == IPOPT_NOP)
1468                         optlen = 1;
1469                 else {
1470                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1471                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1472                                 goto bad;
1473                         }
1474                         optlen = cp[IPOPT_OLEN];
1475                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1476                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1477                                 goto bad;
1478                         }
1479                 }
1480                 switch (opt) {
1481
1482                 default:
1483                         break;
1484
1485                 /*
1486                  * Source routing with record.
1487                  * Find interface with current destination address.
1488                  * If none on this machine then drop if strictly routed,
1489                  * or do nothing if loosely routed.
1490                  * Record interface address and bring up next address
1491                  * component.  If strictly routed make sure next
1492                  * address is on directly accessible net.
1493                  */
1494                 case IPOPT_LSRR:
1495                 case IPOPT_SSRR:
1496                         if (ipstealth && pass > 0)
1497                                 break;
1498                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1499                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1500                                 goto bad;
1501                         }
1502                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1503                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1504                                 goto bad;
1505                         }
1506                         ipaddr.sin_addr = ip->ip_dst;
1507                         ia = (struct in_ifaddr *)
1508                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1509                         if (ia == NULL) {
1510                                 if (opt == IPOPT_SSRR) {
1511                                         type = ICMP_UNREACH;
1512                                         code = ICMP_UNREACH_SRCFAIL;
1513                                         goto bad;
1514                                 }
1515                                 if (!ip_dosourceroute)
1516                                         goto nosourcerouting;
1517                                 /*
1518                                  * Loose routing, and not at next destination
1519                                  * yet; nothing to do except forward.
1520                                  */
1521                                 break;
1522                         }
1523                         off--;                  /* 0 origin */
1524                         if (off > optlen - (int)sizeof(struct in_addr)) {
1525                                 /*
1526                                  * End of source route.  Should be for us.
1527                                  */
1528                                 if (!ip_acceptsourceroute)
1529                                         goto nosourcerouting;
1530                                 save_rte(m, cp, ip->ip_src);
1531                                 break;
1532                         }
1533                         if (ipstealth)
1534                                 goto dropit;
1535                         if (!ip_dosourceroute) {
1536                                 if (ipforwarding) {
1537                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1538
1539                                         /*
1540                                          * Acting as a router, so generate ICMP
1541                                          */
1542 nosourcerouting:
1543                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1544                                         log(LOG_WARNING,
1545                                             "attempted source route from %s to %s\n",
1546                                             inet_ntoa(ip->ip_src), buf);
1547                                         type = ICMP_UNREACH;
1548                                         code = ICMP_UNREACH_SRCFAIL;
1549                                         goto bad;
1550                                 } else {
1551                                         /*
1552                                          * Not acting as a router,
1553                                          * so silently drop.
1554                                          */
1555 dropit:
1556                                         ipstat.ips_cantforward++;
1557                                         m_freem(m);
1558                                         return (1);
1559                                 }
1560                         }
1561
1562                         /*
1563                          * locate outgoing interface
1564                          */
1565                         memcpy(&ipaddr.sin_addr, cp + off,
1566                             sizeof ipaddr.sin_addr);
1567
1568                         if (opt == IPOPT_SSRR) {
1569 #define INA     struct in_ifaddr *
1570 #define SA      struct sockaddr *
1571                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1572                                                                         == NULL)
1573                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1574                         } else
1575                                 ia = ip_rtaddr(ipaddr.sin_addr,
1576                                                &ipforward_rt[mycpuid]);
1577                         if (ia == NULL) {
1578                                 type = ICMP_UNREACH;
1579                                 code = ICMP_UNREACH_SRCFAIL;
1580                                 goto bad;
1581                         }
1582                         ip->ip_dst = ipaddr.sin_addr;
1583                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1584                             sizeof(struct in_addr));
1585                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1586                         /*
1587                          * Let ip_intr's mcast routing check handle mcast pkts
1588                          */
1589                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1590                         break;
1591
1592                 case IPOPT_RR:
1593                         if (ipstealth && pass == 0)
1594                                 break;
1595                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1596                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1597                                 goto bad;
1598                         }
1599                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1600                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1601                                 goto bad;
1602                         }
1603                         /*
1604                          * If no space remains, ignore.
1605                          */
1606                         off--;                  /* 0 origin */
1607                         if (off > optlen - (int)sizeof(struct in_addr))
1608                                 break;
1609                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1610                             sizeof ipaddr.sin_addr);
1611                         /*
1612                          * locate outgoing interface; if we're the destination,
1613                          * use the incoming interface (should be same).
1614                          */
1615                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1616                             (ia = ip_rtaddr(ipaddr.sin_addr,
1617                                             &ipforward_rt[mycpuid]))
1618                                                                      == NULL) {
1619                                 type = ICMP_UNREACH;
1620                                 code = ICMP_UNREACH_HOST;
1621                                 goto bad;
1622                         }
1623                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1624                             sizeof(struct in_addr));
1625                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1626                         break;
1627
1628                 case IPOPT_TS:
1629                         if (ipstealth && pass == 0)
1630                                 break;
1631                         code = cp - (u_char *)ip;
1632                         if (optlen < 4 || optlen > 40) {
1633                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1634                                 goto bad;
1635                         }
1636                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1637                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1638                                 goto bad;
1639                         }
1640                         if (off > optlen - (int)sizeof(int32_t)) {
1641                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1642                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1643                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1644                                         goto bad;
1645                                 }
1646                                 break;
1647                         }
1648                         off--;                          /* 0 origin */
1649                         sin = (struct in_addr *)(cp + off);
1650                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1651
1652                         case IPOPT_TS_TSONLY:
1653                                 break;
1654
1655                         case IPOPT_TS_TSANDADDR:
1656                                 if (off + sizeof(n_time) +
1657                                     sizeof(struct in_addr) > optlen) {
1658                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1659                                         goto bad;
1660                                 }
1661                                 ipaddr.sin_addr = dst;
1662                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1663                                                             m->m_pkthdr.rcvif);
1664                                 if (ia == NULL)
1665                                         continue;
1666                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1667                                     sizeof(struct in_addr));
1668                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1669                                 off += sizeof(struct in_addr);
1670                                 break;
1671
1672                         case IPOPT_TS_PRESPEC:
1673                                 if (off + sizeof(n_time) +
1674                                     sizeof(struct in_addr) > optlen) {
1675                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1676                                         goto bad;
1677                                 }
1678                                 memcpy(&ipaddr.sin_addr, sin,
1679                                     sizeof(struct in_addr));
1680                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1681                                         continue;
1682                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1683                                 off += sizeof(struct in_addr);
1684                                 break;
1685
1686                         default:
1687                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1688                                 goto bad;
1689                         }
1690                         ntime = iptime();
1691                         memcpy(cp + off, &ntime, sizeof(n_time));
1692                         cp[IPOPT_OFFSET] += sizeof(n_time);
1693                 }
1694         }
1695         if (forward && ipforwarding) {
1696                 ip_forward(m, TRUE, next_hop);
1697                 return (1);
1698         }
1699         return (0);
1700 bad:
1701         icmp_error(m, type, code, 0, 0);
1702         ipstat.ips_badoptions++;
1703         return (1);
1704 }
1705
1706 /*
1707  * Given address of next destination (final or next hop),
1708  * return internet address info of interface to be used to get there.
1709  */
1710 struct in_ifaddr *
1711 ip_rtaddr(struct in_addr dst, struct route *ro)
1712 {
1713         struct sockaddr_in *sin;
1714
1715         sin = (struct sockaddr_in *)&ro->ro_dst;
1716
1717         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1718                 if (ro->ro_rt != NULL) {
1719                         RTFREE(ro->ro_rt);
1720                         ro->ro_rt = NULL;
1721                 }
1722                 sin->sin_family = AF_INET;
1723                 sin->sin_len = sizeof *sin;
1724                 sin->sin_addr = dst;
1725                 rtalloc_ign(ro, RTF_PRCLONING);
1726         }
1727
1728         if (ro->ro_rt == NULL)
1729                 return (NULL);
1730
1731         return (ifatoia(ro->ro_rt->rt_ifa));
1732 }
1733
1734 /*
1735  * Save incoming source route for use in replies,
1736  * to be picked up later by ip_srcroute if the receiver is interested.
1737  */
1738 static void
1739 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1740 {
1741         struct m_tag *mtag;
1742         struct ip_srcrt_opt *opt;
1743         unsigned olen;
1744
1745         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1746         if (mtag == NULL)
1747                 return;
1748         opt = m_tag_data(mtag);
1749
1750         olen = option[IPOPT_OLEN];
1751 #ifdef DIAGNOSTIC
1752         if (ipprintfs)
1753                 kprintf("save_rte: olen %d\n", olen);
1754 #endif
1755         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1756                 m_tag_free(mtag);
1757                 return;
1758         }
1759         bcopy(option, opt->ip_srcrt.srcopt, olen);
1760         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1761         opt->ip_srcrt.dst = dst;
1762         m_tag_prepend(m, mtag);
1763 }
1764
1765 /*
1766  * Retrieve incoming source route for use in replies,
1767  * in the same form used by setsockopt.
1768  * The first hop is placed before the options, will be removed later.
1769  */
1770 struct mbuf *
1771 ip_srcroute(struct mbuf *m0)
1772 {
1773         struct in_addr *p, *q;
1774         struct mbuf *m;
1775         struct m_tag *mtag;
1776         struct ip_srcrt_opt *opt;
1777
1778         if (m0 == NULL)
1779                 return NULL;
1780
1781         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1782         if (mtag == NULL)
1783                 return NULL;
1784         opt = m_tag_data(mtag);
1785
1786         if (opt->ip_nhops == 0)
1787                 return (NULL);
1788         m = m_get(MB_DONTWAIT, MT_HEADER);
1789         if (m == NULL)
1790                 return (NULL);
1791
1792 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1793
1794         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1795         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1796                    sizeof(struct in_addr) + OPTSIZ;
1797 #ifdef DIAGNOSTIC
1798         if (ipprintfs) {
1799                 kprintf("ip_srcroute: nhops %d mlen %d",
1800                         opt->ip_nhops, m->m_len);
1801         }
1802 #endif
1803
1804         /*
1805          * First save first hop for return route
1806          */
1807         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1808         *(mtod(m, struct in_addr *)) = *p--;
1809 #ifdef DIAGNOSTIC
1810         if (ipprintfs)
1811                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1812 #endif
1813
1814         /*
1815          * Copy option fields and padding (nop) to mbuf.
1816          */
1817         opt->ip_srcrt.nop = IPOPT_NOP;
1818         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1819         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1820             OPTSIZ);
1821         q = (struct in_addr *)(mtod(m, caddr_t) +
1822             sizeof(struct in_addr) + OPTSIZ);
1823 #undef OPTSIZ
1824         /*
1825          * Record return path as an IP source route,
1826          * reversing the path (pointers are now aligned).
1827          */
1828         while (p >= opt->ip_srcrt.route) {
1829 #ifdef DIAGNOSTIC
1830                 if (ipprintfs)
1831                         kprintf(" %x", ntohl(q->s_addr));
1832 #endif
1833                 *q++ = *p--;
1834         }
1835         /*
1836          * Last hop goes to final destination.
1837          */
1838         *q = opt->ip_srcrt.dst;
1839         m_tag_delete(m0, mtag);
1840 #ifdef DIAGNOSTIC
1841         if (ipprintfs)
1842                 kprintf(" %x\n", ntohl(q->s_addr));
1843 #endif
1844         return (m);
1845 }
1846
1847 /*
1848  * Strip out IP options.
1849  */
1850 void
1851 ip_stripoptions(struct mbuf *m)
1852 {
1853         int datalen;
1854         struct ip *ip = mtod(m, struct ip *);
1855         caddr_t opts;
1856         int optlen;
1857
1858         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1859         opts = (caddr_t)(ip + 1);
1860         datalen = m->m_len - (sizeof(struct ip) + optlen);
1861         bcopy(opts + optlen, opts, datalen);
1862         m->m_len -= optlen;
1863         if (m->m_flags & M_PKTHDR)
1864                 m->m_pkthdr.len -= optlen;
1865         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1866 }
1867
1868 u_char inetctlerrmap[PRC_NCMDS] = {
1869         0,              0,              0,              0,
1870         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1871         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1872         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1873         0,              0,              0,              0,
1874         ENOPROTOOPT,    ECONNREFUSED
1875 };
1876
1877 /*
1878  * Forward a packet.  If some error occurs return the sender
1879  * an icmp packet.  Note we can't always generate a meaningful
1880  * icmp message because icmp doesn't have a large enough repertoire
1881  * of codes and types.
1882  *
1883  * If not forwarding, just drop the packet.  This could be confusing
1884  * if ipforwarding was zero but some routing protocol was advancing
1885  * us as a gateway to somewhere.  However, we must let the routing
1886  * protocol deal with that.
1887  *
1888  * The using_srcrt parameter indicates whether the packet is being forwarded
1889  * via a source route.
1890  */
1891 void
1892 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1893 {
1894         struct ip *ip = mtod(m, struct ip *);
1895         struct sockaddr_in *ipforward_rtaddr;
1896         struct rtentry *rt;
1897         int error, type = 0, code = 0, destmtu = 0;
1898         struct mbuf *mcopy;
1899         n_long dest;
1900         struct in_addr pkt_dst;
1901         struct route *cache_rt = &ipforward_rt[mycpuid];
1902
1903         dest = INADDR_ANY;
1904         /*
1905          * Cache the destination address of the packet; this may be
1906          * changed by use of 'ipfw fwd'.
1907          */
1908         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1909
1910 #ifdef DIAGNOSTIC
1911         if (ipprintfs)
1912                 kprintf("forward: src %x dst %x ttl %x\n",
1913                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1914 #endif
1915
1916         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1917                 ipstat.ips_cantforward++;
1918                 m_freem(m);
1919                 return;
1920         }
1921         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1922                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1923                 return;
1924         }
1925
1926         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1927         if (cache_rt->ro_rt == NULL ||
1928             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1929                 if (cache_rt->ro_rt != NULL) {
1930                         RTFREE(cache_rt->ro_rt);
1931                         cache_rt->ro_rt = NULL;
1932                 }
1933                 ipforward_rtaddr->sin_family = AF_INET;
1934                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1935                 ipforward_rtaddr->sin_addr = pkt_dst;
1936                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1937                 if (cache_rt->ro_rt == NULL) {
1938                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1939                         return;
1940                 }
1941         }
1942         rt = cache_rt->ro_rt;
1943
1944         /*
1945          * Save the IP header and at most 8 bytes of the payload,
1946          * in case we need to generate an ICMP message to the src.
1947          *
1948          * XXX this can be optimized a lot by saving the data in a local
1949          * buffer on the stack (72 bytes at most), and only allocating the
1950          * mbuf if really necessary. The vast majority of the packets
1951          * are forwarded without having to send an ICMP back (either
1952          * because unnecessary, or because rate limited), so we are
1953          * really we are wasting a lot of work here.
1954          *
1955          * We don't use m_copy() because it might return a reference
1956          * to a shared cluster. Both this function and ip_output()
1957          * assume exclusive access to the IP header in `m', so any
1958          * data in a cluster may change before we reach icmp_error().
1959          */
1960         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1961         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1962                 /*
1963                  * It's probably ok if the pkthdr dup fails (because
1964                  * the deep copy of the tag chain failed), but for now
1965                  * be conservative and just discard the copy since
1966                  * code below may some day want the tags.
1967                  */
1968                 m_free(mcopy);
1969                 mcopy = NULL;
1970         }
1971         if (mcopy != NULL) {
1972                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1973                     (int)ip->ip_len);
1974                 mcopy->m_pkthdr.len = mcopy->m_len;
1975                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1976         }
1977
1978         if (!ipstealth)
1979                 ip->ip_ttl -= IPTTLDEC;
1980
1981         /*
1982          * If forwarding packet using same interface that it came in on,
1983          * perhaps should send a redirect to sender to shortcut a hop.
1984          * Only send redirect if source is sending directly to us,
1985          * and if packet was not source routed (or has any options).
1986          * Also, don't send redirect if forwarding using a default route
1987          * or a route modified by a redirect.
1988          */
1989         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1990             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1991             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1992             ipsendredirects && !using_srcrt && next_hop == NULL) {
1993                 u_long src = ntohl(ip->ip_src.s_addr);
1994                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1995
1996                 if (rt_ifa != NULL &&
1997                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1998                         if (rt->rt_flags & RTF_GATEWAY)
1999                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
2000                         else
2001                                 dest = pkt_dst.s_addr;
2002                         /*
2003                          * Router requirements says to only send
2004                          * host redirects.
2005                          */
2006                         type = ICMP_REDIRECT;
2007                         code = ICMP_REDIRECT_HOST;
2008 #ifdef DIAGNOSTIC
2009                         if (ipprintfs)
2010                                 kprintf("redirect (%d) to %x\n", code, dest);
2011 #endif
2012                 }
2013         }
2014
2015         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
2016         if (error == 0) {
2017                 ipstat.ips_forward++;
2018                 if (type == 0) {
2019                         if (mcopy) {
2020                                 ipflow_create(cache_rt, mcopy);
2021                                 m_freem(mcopy);
2022                         }
2023                         return;         /* most common case */
2024                 } else {
2025                         ipstat.ips_redirectsent++;
2026                 }
2027         } else {
2028                 ipstat.ips_cantforward++;
2029         }
2030
2031         if (mcopy == NULL)
2032                 return;
2033
2034         /*
2035          * Send ICMP message.
2036          */
2037
2038         switch (error) {
2039
2040         case 0:                         /* forwarded, but need redirect */
2041                 /* type, code set above */
2042                 break;
2043
2044         case ENETUNREACH:               /* shouldn't happen, checked above */
2045         case EHOSTUNREACH:
2046         case ENETDOWN:
2047         case EHOSTDOWN:
2048         default:
2049                 type = ICMP_UNREACH;
2050                 code = ICMP_UNREACH_HOST;
2051                 break;
2052
2053         case EMSGSIZE:
2054                 type = ICMP_UNREACH;
2055                 code = ICMP_UNREACH_NEEDFRAG;
2056 #ifdef IPSEC
2057                 /*
2058                  * If the packet is routed over IPsec tunnel, tell the
2059                  * originator the tunnel MTU.
2060                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2061                  * XXX quickhack!!!
2062                  */
2063                 if (cache_rt->ro_rt != NULL) {
2064                         struct secpolicy *sp = NULL;
2065                         int ipsecerror;
2066                         int ipsechdr;
2067                         struct route *ro;
2068
2069                         sp = ipsec4_getpolicybyaddr(mcopy,
2070                                                     IPSEC_DIR_OUTBOUND,
2071                                                     IP_FORWARDING,
2072                                                     &ipsecerror);
2073
2074                         if (sp == NULL)
2075                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2076                         else {
2077                                 /* count IPsec header size */
2078                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2079                                                          IPSEC_DIR_OUTBOUND,
2080                                                          NULL);
2081
2082                                 /*
2083                                  * find the correct route for outer IPv4
2084                                  * header, compute tunnel MTU.
2085                                  *
2086                                  */
2087                                 if (sp->req != NULL && sp->req->sav != NULL &&
2088                                     sp->req->sav->sah != NULL) {
2089                                         ro = &sp->req->sav->sah->sa_route;
2090                                         if (ro->ro_rt != NULL &&
2091                                             ro->ro_rt->rt_ifp != NULL) {
2092                                                 destmtu =
2093                                                     ro->ro_rt->rt_ifp->if_mtu;
2094                                                 destmtu -= ipsechdr;
2095                                         }
2096                                 }
2097
2098                                 key_freesp(sp);
2099                         }
2100                 }
2101 #elif FAST_IPSEC
2102                 /*
2103                  * If the packet is routed over IPsec tunnel, tell the
2104                  * originator the tunnel MTU.
2105                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2106                  * XXX quickhack!!!
2107                  */
2108                 if (cache_rt->ro_rt != NULL) {
2109                         struct secpolicy *sp = NULL;
2110                         int ipsecerror;
2111                         int ipsechdr;
2112                         struct route *ro;
2113
2114                         sp = ipsec_getpolicybyaddr(mcopy,
2115                                                    IPSEC_DIR_OUTBOUND,
2116                                                    IP_FORWARDING,
2117                                                    &ipsecerror);
2118
2119                         if (sp == NULL)
2120                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2121                         else {
2122                                 /* count IPsec header size */
2123                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2124                                                          IPSEC_DIR_OUTBOUND,
2125                                                          NULL);
2126
2127                                 /*
2128                                  * find the correct route for outer IPv4
2129                                  * header, compute tunnel MTU.
2130                                  */
2131
2132                                 if (sp->req != NULL &&
2133                                     sp->req->sav != NULL &&
2134                                     sp->req->sav->sah != NULL) {
2135                                         ro = &sp->req->sav->sah->sa_route;
2136                                         if (ro->ro_rt != NULL &&
2137                                             ro->ro_rt->rt_ifp != NULL) {
2138                                                 destmtu =
2139                                                     ro->ro_rt->rt_ifp->if_mtu;
2140                                                 destmtu -= ipsechdr;
2141                                         }
2142                                 }
2143
2144                                 KEY_FREESP(&sp);
2145                         }
2146                 }
2147 #else /* !IPSEC && !FAST_IPSEC */
2148                 if (cache_rt->ro_rt != NULL)
2149                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2150 #endif /*IPSEC*/
2151                 ipstat.ips_cantfrag++;
2152                 break;
2153
2154         case ENOBUFS:
2155                 /*
2156                  * A router should not generate ICMP_SOURCEQUENCH as
2157                  * required in RFC1812 Requirements for IP Version 4 Routers.
2158                  * Source quench could be a big problem under DoS attacks,
2159                  * or if the underlying interface is rate-limited.
2160                  * Those who need source quench packets may re-enable them
2161                  * via the net.inet.ip.sendsourcequench sysctl.
2162                  */
2163                 if (!ip_sendsourcequench) {
2164                         m_freem(mcopy);
2165                         return;
2166                 } else {
2167                         type = ICMP_SOURCEQUENCH;
2168                         code = 0;
2169                 }
2170                 break;
2171
2172         case EACCES:                    /* ipfw denied packet */
2173                 m_freem(mcopy);
2174                 return;
2175         }
2176         icmp_error(mcopy, type, code, dest, destmtu);
2177 }
2178
2179 void
2180 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2181                struct mbuf *m)
2182 {
2183         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2184                 struct timeval tv;
2185
2186                 microtime(&tv);
2187                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2188                     SCM_TIMESTAMP, SOL_SOCKET);
2189                 if (*mp)
2190                         mp = &(*mp)->m_next;
2191         }
2192         if (inp->inp_flags & INP_RECVDSTADDR) {
2193                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2194                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2195                 if (*mp)
2196                         mp = &(*mp)->m_next;
2197         }
2198         if (inp->inp_flags & INP_RECVTTL) {
2199                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2200                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2201                 if (*mp)
2202                         mp = &(*mp)->m_next;
2203         }
2204 #ifdef notyet
2205         /* XXX
2206          * Moving these out of udp_input() made them even more broken
2207          * than they already were.
2208          */
2209         /* options were tossed already */
2210         if (inp->inp_flags & INP_RECVOPTS) {
2211                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2212                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2213                 if (*mp)
2214                         mp = &(*mp)->m_next;
2215         }
2216         /* ip_srcroute doesn't do what we want here, need to fix */
2217         if (inp->inp_flags & INP_RECVRETOPTS) {
2218                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2219                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2220                 if (*mp)
2221                         mp = &(*mp)->m_next;
2222         }
2223 #endif
2224         if (inp->inp_flags & INP_RECVIF) {
2225                 struct ifnet *ifp;
2226                 struct sdlbuf {
2227                         struct sockaddr_dl sdl;
2228                         u_char  pad[32];
2229                 } sdlbuf;
2230                 struct sockaddr_dl *sdp;
2231                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2232
2233                 if (((ifp = m->m_pkthdr.rcvif)) &&
2234                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2235                         sdp = IF_LLSOCKADDR(ifp);
2236                         /*
2237                          * Change our mind and don't try copy.
2238                          */
2239                         if ((sdp->sdl_family != AF_LINK) ||
2240                             (sdp->sdl_len > sizeof(sdlbuf))) {
2241                                 goto makedummy;
2242                         }
2243                         bcopy(sdp, sdl2, sdp->sdl_len);
2244                 } else {
2245 makedummy:
2246                         sdl2->sdl_len =
2247                             offsetof(struct sockaddr_dl, sdl_data[0]);
2248                         sdl2->sdl_family = AF_LINK;
2249                         sdl2->sdl_index = 0;
2250                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2251                 }
2252                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2253                         IP_RECVIF, IPPROTO_IP);
2254                 if (*mp)
2255                         mp = &(*mp)->m_next;
2256         }
2257 }
2258
2259 /*
2260  * XXX these routines are called from the upper part of the kernel.
2261  *
2262  * They could also be moved to ip_mroute.c, since all the RSVP
2263  *  handling is done there already.
2264  */
2265 int
2266 ip_rsvp_init(struct socket *so)
2267 {
2268         if (so->so_type != SOCK_RAW ||
2269             so->so_proto->pr_protocol != IPPROTO_RSVP)
2270                 return EOPNOTSUPP;
2271
2272         if (ip_rsvpd != NULL)
2273                 return EADDRINUSE;
2274
2275         ip_rsvpd = so;
2276         /*
2277          * This may seem silly, but we need to be sure we don't over-increment
2278          * the RSVP counter, in case something slips up.
2279          */
2280         if (!ip_rsvp_on) {
2281                 ip_rsvp_on = 1;
2282                 rsvp_on++;
2283         }
2284
2285         return 0;
2286 }
2287
2288 int
2289 ip_rsvp_done(void)
2290 {
2291         ip_rsvpd = NULL;
2292         /*
2293          * This may seem silly, but we need to be sure we don't over-decrement
2294          * the RSVP counter, in case something slips up.
2295          */
2296         if (ip_rsvp_on) {
2297                 ip_rsvp_on = 0;
2298                 rsvp_on--;
2299         }
2300         return 0;
2301 }
2302
2303 void
2304 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2305 {
2306         int off, proto;
2307         __va_list ap;
2308
2309         __va_start(ap, m);
2310         off = __va_arg(ap, int);
2311         proto = __va_arg(ap, int);
2312         __va_end(ap);
2313
2314         if (rsvp_input_p) { /* call the real one if loaded */
2315                 rsvp_input_p(m, off, proto);
2316                 return;
2317         }
2318
2319         /* Can still get packets with rsvp_on = 0 if there is a local member
2320          * of the group to which the RSVP packet is addressed.  But in this
2321          * case we want to throw the packet away.
2322          */
2323
2324         if (!rsvp_on) {
2325                 m_freem(m);
2326                 return;
2327         }
2328
2329         if (ip_rsvpd != NULL) {
2330                 rip_input(m, off, proto);
2331                 return;
2332         }
2333         /* Drop the packet */
2334         m_freem(m);
2335 }