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. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      @(#)ip_input.c  8.2 (Berkeley) 1/4/94
  63  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
  64  */
  65
  66 #define _IP_VHL
  67
  68 #include "opt_bootp.h"
  69 #include "opt_ipdn.h"
  70 #include "opt_ipdivert.h"
  71 #include "opt_ipstealth.h"
  72 #include "opt_ipsec.h"
  73 #include "opt_rss.h"
  74
  75 #include <sys/param.h>
  76 #include <sys/systm.h>
  77 #include <sys/mbuf.h>
  78 #include <sys/malloc.h>
  79 #include <sys/mpipe.h>
  80 #include <sys/domain.h>
  81 #include <sys/protosw.h>
  82 #include <sys/socket.h>
  83 #include <sys/time.h>
  84 #include <sys/globaldata.h>
  85 #include <sys/thread.h>
  86 #include <sys/kernel.h>
  87 #include <sys/syslog.h>
  88 #include <sys/sysctl.h>
  89 #include <sys/in_cksum.h>
  90 #include <sys/lock.h>
  91
  92 #include <sys/mplock2.h>
  93
  94 #include <machine/stdarg.h>
  95
  96 #include <net/if.h>
  97 #include <net/if_types.h>
  98 #include <net/if_var.h>
  99 #include <net/if_dl.h>
 100 #include <net/pfil.h>
 101 #include <net/route.h>
 102 #include <net/netisr2.h>
 103
 104 #include <netinet/in.h>
 105 #include <netinet/in_systm.h>
 106 #include <netinet/in_var.h>
 107 #include <netinet/ip.h>
 108 #include <netinet/in_pcb.h>
 109 #include <netinet/ip_var.h>
 110 #include <netinet/ip_icmp.h>
 111 #include <netinet/ip_divert.h>
 112 #include <netinet/ip_flow.h>
 113
 114 #include <sys/thread2.h>
 115 #include <sys/msgport2.h>
 116 #include <net/netmsg2.h>
 117
 118 #include <sys/socketvar.h>
 119
 120 #include <net/ipfw/ip_fw.h>
 121 #include <net/dummynet/ip_dummynet.h>
 122
 123 #ifdef IPSEC
 124 #include <netinet6/ipsec.h>
 125 #include <netproto/key/key.h>
 126 #endif
 127
 128 #ifdef FAST_IPSEC
 129 #include <netproto/ipsec/ipsec.h>
 130 #include <netproto/ipsec/key.h>
 131 #endif
 132
 133 int rsvp_on = 0;
 134 static int ip_rsvp_on;
 135 struct socket *ip_rsvpd;
 136
 137 int ipforwarding = 0;
 138 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
 139     &ipforwarding, 0, "Enable IP forwarding between interfaces");
 140
 141 static int ipsendredirects = 1; /* XXX */
 142 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
 143     &ipsendredirects, 0, "Enable sending IP redirects");
 144
 145 int ip_defttl = IPDEFTTL;
 146 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
 147     &ip_defttl, 0, "Maximum TTL on IP packets");
 148
 149 static int ip_dosourceroute = 0;
 150 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
 151     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
 152
 153 static int ip_acceptsourceroute = 0;
 154 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
 155     CTLFLAG_RW, &ip_acceptsourceroute, 0,
 156     "Enable accepting source routed IP packets");
 157
 158 static int ip_keepfaith = 0;
 159 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
 160     &ip_keepfaith, 0,
 161     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
 162
 163 static int maxnipq;
 164 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 165     &maxnipq, 0,
 166     "Maximum number of IPv4 fragment reassembly queue entries");
 167
 168 static int maxfragsperpacket;
 169 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 170     &maxfragsperpacket, 0,
 171     "Maximum number of IPv4 fragments allowed per packet");
 172
 173 static int ip_sendsourcequench = 0;
 174 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 175     &ip_sendsourcequench, 0,
 176     "Enable the transmission of source quench packets");
 177
 178 int ip_do_randomid = 1;
 179 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 180     &ip_do_randomid, 0,
 181     "Assign random ip_id values");
 182 /*
 183  * XXX - Setting ip_checkinterface mostly implements the receive side of
 184  * the Strong ES model described in RFC 1122, but since the routing table
 185  * and transmit implementation do not implement the Strong ES model,
 186  * setting this to 1 results in an odd hybrid.
 187  *
 188  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 189  * to translate the destination address to another local interface.
 190  *
 191  * XXX - ip_checkinterface must be disabled if you add IP aliases
 192  * to the loopback interface instead of the interface where the
 193  * packets for those addresses are received.
 194  */
 195 static int ip_checkinterface = 0;
 196 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 197     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 198
 199 static u_long ip_hash_count = 0;
 200 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, hash_count, CTLFLAG_RD,
 201     &ip_hash_count, 0, "Number of packets hashed by IP");
 202
 203 #ifdef RSS_DEBUG
 204 static u_long ip_rehash_count = 0;
 205 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, rehash_count, CTLFLAG_RD,
 206     &ip_rehash_count, 0, "Number of packets rehashed by IP");
 207
 208 static u_long ip_dispatch_fast = 0;
 209 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_fast_count, CTLFLAG_RD,
 210     &ip_dispatch_fast, 0, "Number of packets handled on current CPU");
 211
 212 static u_long ip_dispatch_slow = 0;
 213 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_slow_count, CTLFLAG_RD,
 214     &ip_dispatch_slow, 0, "Number of packets messaged to another CPU");
 215 #endif
 216
 217 #ifdef DIAGNOSTIC
 218 static int ipprintfs = 0;
 219 #endif
 220
 221 extern  struct domain inetdomain;
 222 extern  struct protosw inetsw[];
 223 u_char  ip_protox[IPPROTO_MAX];
 224 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 225 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 226                                                 /* inet addr hash table */
 227 u_long  in_ifaddrhmask;                         /* mask for hash table */
 228
 229 static struct mbuf *ipforward_mtemp[MAXCPU];
 230
 231 struct ip_stats ipstats_percpu[MAXCPU] __cachealign;
 232
 233 static int
 234 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 235 {
 236         int cpu, error = 0;
 237
 238         for (cpu = 0; cpu < ncpus; ++cpu) {
 239                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 240                                         sizeof(struct ip_stats))))
 241                         break;
 242                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 243                                        sizeof(struct ip_stats))))
 244                         break;
 245         }
 246
 247         return (error);
 248 }
 249 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 250     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 251
 252 /* Packet reassembly stuff */
 253 #define IPREASS_NHASH_LOG2      6
 254 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 255 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 256 #define IPREASS_HASH(x,y)                                               \
 257     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 258
 259 TAILQ_HEAD(ipqhead, ipq);
 260 struct ipfrag_queue {
 261         int                     nipq;
 262         struct netmsg_base      timeo_netmsg;
 263         struct netmsg_base      drain_netmsg;
 264         struct ipqhead          ipq[IPREASS_NHASH];
 265 } __cachealign;
 266
 267 static struct ipfrag_queue      ipfrag_queue_pcpu[MAXCPU];
 268
 269 #ifdef IPCTL_DEFMTU
 270 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 271     &ip_mtu, 0, "Default MTU");
 272 #endif
 273
 274 #ifdef IPSTEALTH
 275 static int ipstealth = 0;
 276 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 277 #else
 278 static const int ipstealth = 0;
 279 #endif
 280
 281 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
 282
 283 struct pfil_head inet_pfil_hook;
 284
 285 /*
 286  * struct ip_srcrt_opt is used to store packet state while it travels
 287  * through the stack.
 288  *
 289  * XXX Note that the code even makes assumptions on the size and
 290  * alignment of fields inside struct ip_srcrt so e.g. adding some
 291  * fields will break the code.  This needs to be fixed.
 292  *
 293  * We need to save the IP options in case a protocol wants to respond
 294  * to an incoming packet over the same route if the packet got here
 295  * using IP source routing.  This allows connection establishment and
 296  * maintenance when the remote end is on a network that is not known
 297  * to us.
 298  */
 299 struct ip_srcrt {
 300         struct  in_addr dst;                    /* final destination */
 301         char    nop;                            /* one NOP to align */
 302         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 303         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 304 };
 305
 306 struct ip_srcrt_opt {
 307         int             ip_nhops;
 308         struct ip_srcrt ip_srcrt;
 309 };
 310
 311 #define IPFRAG_MPIPE_MAX        4096
 312 #define MAXIPFRAG_MIN           ((IPFRAG_MPIPE_MAX * 2) / 256)
 313
 314 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 315 static struct malloc_pipe ipq_mpipe;
 316
 317 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
 318 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 319 static void             ip_freef(struct ipfrag_queue *, struct ipqhead *,
 320                             struct ipq *);
 321 static void             ip_input_handler(netmsg_t);
 322
 323 static void             ipfrag_timeo_dispatch(netmsg_t);
 324 static void             ipfrag_drain_dispatch(netmsg_t);
 325
 326 /*
 327  * IP initialization: fill in IP protocol switch table.
 328  * All protocols not implemented in kernel go to raw IP protocol handler.
 329  */
 330 void
 331 ip_init(void)
 332 {
 333         struct protosw *pr;
 334         int cpu, i;
 335
 336         /*
 337          * Make sure we can handle a reasonable number of fragments but
 338          * cap it at IPFRAG_MPIPE_MAX.
 339          */
 340         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 341             IFQ_MAXLEN, IPFRAG_MPIPE_MAX, 0, NULL, NULL, NULL);
 342         for (cpu = 0; cpu < ncpus; ++cpu) {
 343                 TAILQ_INIT(&in_ifaddrheads[cpu]);
 344                 in_ifaddrhashtbls[cpu] =
 345                     hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 346         }
 347         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 348         if (pr == NULL)
 349                 panic("ip_init");
 350         for (i = 0; i < IPPROTO_MAX; i++)
 351                 ip_protox[i] = pr - inetsw;
 352         for (pr = inetdomain.dom_protosw;
 353              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
 354                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
 355                         if (pr->pr_protocol != IPPROTO_RAW)
 356                                 ip_protox[pr->pr_protocol] = pr - inetsw;
 357                 }
 358         }
 359
 360         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 361         inet_pfil_hook.ph_af = AF_INET;
 362         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 363                 kprintf("%s: WARNING: unable to register pfil hook, "
 364                         "error %d\n", __func__, i);
 365         }
 366
 367         maxnipq = (nmbclusters / 32) / ncpus2;
 368         if (maxnipq < MAXIPFRAG_MIN)
 369                 maxnipq = MAXIPFRAG_MIN;
 370         maxfragsperpacket = 16;
 371
 372         ip_id = time_second & 0xffff;   /* time_second survives reboots */
 373
 374         for (cpu = 0; cpu < ncpus; ++cpu) {
 375                 /*
 376                  * Initialize IP statistics counters for each CPU.
 377                  */
 378                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 379
 380                 /*
 381                  * Preallocate mbuf template for forwarding
 382                  */
 383                 MGETHDR(ipforward_mtemp[cpu], MB_WAIT, MT_DATA);
 384
 385                 /*
 386                  * Initialize per-cpu ip fragments queues
 387                  */
 388                 for (i = 0; i < IPREASS_NHASH; i++) {
 389                         struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[cpu];
 390
 391                         TAILQ_INIT(&fragq->ipq[i]);
 392                         netmsg_init(&fragq->timeo_netmsg, NULL,
 393                             &netisr_adone_rport, 0, ipfrag_timeo_dispatch);
 394                         netmsg_init(&fragq->drain_netmsg, NULL,
 395                             &netisr_adone_rport, 0, ipfrag_drain_dispatch);
 396                 }
 397         }
 398
 399         netisr_register(NETISR_IP, ip_input_handler, ip_hashfn_in);
 400         netisr_register_hashcheck(NETISR_IP, ip_hashcheck);
 401 }
 402
 403 /* Do transport protocol processing. */
 404 static void
 405 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 406 {
 407         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 408
 409         /*
 410          * Switch out to protocol's input routine.
 411          */
 412         PR_GET_MPLOCK(pr);
 413         pr->pr_input(&m, &hlen, ip->ip_p);
 414         PR_REL_MPLOCK(pr);
 415 }
 416
 417 static void
 418 transport_processing_handler(netmsg_t msg)
 419 {
 420         struct netmsg_packet *pmsg = &msg->packet;
 421         struct ip *ip;
 422         int hlen;
 423
 424         ip = mtod(pmsg->nm_packet, struct ip *);
 425         hlen = pmsg->base.lmsg.u.ms_result;
 426
 427         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 428         /* msg was embedded in the mbuf, do not reply! */
 429 }
 430
 431 static void
 432 ip_input_handler(netmsg_t msg)
 433 {
 434         ip_input(msg->packet.nm_packet);
 435         /* msg was embedded in the mbuf, do not reply! */
 436 }
 437
 438 /*
 439  * IP input routine.  Checksum and byte swap header.  If fragmented
 440  * try to reassemble.  Process options.  Pass to next level.
 441  */
 442 void
 443 ip_input(struct mbuf *m)
 444 {
 445         struct ip *ip;
 446         struct in_ifaddr *ia = NULL;
 447         struct in_ifaddr_container *iac;
 448         int hlen, checkif;
 449         u_short sum;
 450         struct in_addr pkt_dst;
 451         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 452         struct in_addr odst;                    /* original dst address(NAT) */
 453         struct m_tag *mtag;
 454         struct sockaddr_in *next_hop = NULL;
 455         lwkt_port_t port;
 456 #ifdef FAST_IPSEC
 457         struct tdb_ident *tdbi;
 458         struct secpolicy *sp;
 459         int error;
 460 #endif
 461
 462         M_ASSERTPKTHDR(m);
 463
 464         /*
 465          * This routine is called from numerous places which may not have
 466          * characterized the packet.
 467          */
 468         ip = mtod(m, struct ip *);
 469         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 470             ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK)) {
 471                 /*
 472                  * Force hash recalculation for fragments and multicast
 473                  * packets; hardware may not do it correctly.
 474                  * XXX add flag to indicate the hash is from hardware
 475                  */
 476                 m->m_flags &= ~M_HASH;
 477         }
 478         if ((m->m_flags & M_HASH) == 0) {
 479                 ip_hashfn(&m, 0, IP_MPORT_IN);
 480                 if (m == NULL)
 481                         return;
 482                 KKASSERT(m->m_flags & M_HASH);
 483
 484                 if (&curthread->td_msgport !=
 485                     netisr_hashport(m->m_pkthdr.hash)) {
 486                         netisr_queue(NETISR_IP, m);
 487                         /* Requeued to other netisr msgport; done */
 488                         return;
 489                 }
 490
 491                 /* mbuf could have been changed */
 492                 ip = mtod(m, struct ip *);
 493         }
 494
 495         /*
 496          * Pull out certain tags
 497          */
 498         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 499                 /* Next hop */
 500                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 501                 KKASSERT(mtag != NULL);
 502                 next_hop = m_tag_data(mtag);
 503         }
 504
 505         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 506                 /* dummynet already filtered us */
 507                 ip = mtod(m, struct ip *);
 508                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 509                 goto iphack;
 510         }
 511
 512         ipstat.ips_total++;
 513
 514         /* length checks already done in ip_hashfn() */
 515         KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
 516
 517         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 518                 ipstat.ips_badvers++;
 519                 goto bad;
 520         }
 521
 522         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 523         /* length checks already done in ip_hashfn() */
 524         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 525         KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
 526
 527         /* 127/8 must not appear on wire - RFC1122 */
 528         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 529             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 530                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 531                         ipstat.ips_badaddr++;
 532                         goto bad;
 533                 }
 534         }
 535
 536         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 537                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 538         } else {
 539                 if (hlen == sizeof(struct ip))
 540                         sum = in_cksum_hdr(ip);
 541                 else
 542                         sum = in_cksum(m, hlen);
 543         }
 544         if (sum != 0) {
 545                 ipstat.ips_badsum++;
 546                 goto bad;
 547         }
 548
 549 #ifdef ALTQ
 550         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 551                 /* packet is dropped by traffic conditioner */
 552                 return;
 553         }
 554 #endif
 555         /*
 556          * Convert fields to host representation.
 557          */
 558         ip->ip_len = ntohs(ip->ip_len);
 559         ip->ip_off = ntohs(ip->ip_off);
 560
 561         /* length checks already done in ip_hashfn() */
 562         KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
 563         KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
 564
 565         /*
 566          * Trim mbufs if longer than the IP header would have us expect.
 567          */
 568         if (m->m_pkthdr.len > ip->ip_len) {
 569                 if (m->m_len == m->m_pkthdr.len) {
 570                         m->m_len = ip->ip_len;
 571                         m->m_pkthdr.len = ip->ip_len;
 572                 } else {
 573                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 574                 }
 575         }
 576 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 577         /*
 578          * Bypass packet filtering for packets from a tunnel (gif).
 579          */
 580         if (ipsec_gethist(m, NULL))
 581                 goto pass;
 582 #endif
 583
 584         /*
 585          * IpHack's section.
 586          * Right now when no processing on packet has done
 587          * and it is still fresh out of network we do our black
 588          * deals with it.
 589          * - Firewall: deny/allow/divert
 590          * - Xlate: translate packet's addr/port (NAT).
 591          * - Pipe: pass pkt through dummynet.
 592          * - Wrap: fake packet's addr/port <unimpl.>
 593          * - Encapsulate: put it in another IP and send out. <unimp.>
 594          */
 595
 596 iphack:
 597         /*
 598          * If we've been forwarded from the output side, then
 599          * skip the firewall a second time
 600          */
 601         if (next_hop != NULL)
 602                 goto ours;
 603
 604         /* No pfil hooks */
 605         if (!pfil_has_hooks(&inet_pfil_hook)) {
 606                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 607                         /*
 608                          * Strip dummynet tags from stranded packets
 609                          */
 610                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 611                         KKASSERT(mtag != NULL);
 612                         m_tag_delete(m, mtag);
 613                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 614                 }
 615                 goto pass;
 616         }
 617
 618         /*
 619          * Run through list of hooks for input packets.
 620          *
 621          * NOTE!  If the packet is rewritten pf/ipfw/whoever must
 622          *        clear M_HASH.
 623          */
 624         odst = ip->ip_dst;
 625         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 626                 return;
 627         if (m == NULL)  /* consumed by filter */
 628                 return;
 629         ip = mtod(m, struct ip *);
 630         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 631         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 632
 633         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 634                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 635                 KKASSERT(mtag != NULL);
 636                 next_hop = m_tag_data(mtag);
 637         }
 638         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 639                 ip_dn_queue(m);
 640                 return;
 641         }
 642         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 643                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 644         }
 645 pass:
 646         /*
 647          * Process options and, if not destined for us,
 648          * ship it on.  ip_dooptions returns 1 when an
 649          * error was detected (causing an icmp message
 650          * to be sent and the original packet to be freed).
 651          */
 652         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 653                 return;
 654
 655         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 656          * matter if it is destined to another node, or whether it is
 657          * a multicast one, RSVP wants it! and prevents it from being forwarded
 658          * anywhere else. Also checks if the rsvp daemon is running before
 659          * grabbing the packet.
 660          */
 661         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 662                 goto ours;
 663
 664         /*
 665          * Check our list of addresses, to see if the packet is for us.
 666          * If we don't have any addresses, assume any unicast packet
 667          * we receive might be for us (and let the upper layers deal
 668          * with it).
 669          */
 670         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 671             !(m->m_flags & (M_MCAST | M_BCAST)))
 672                 goto ours;
 673
 674         /*
 675          * Cache the destination address of the packet; this may be
 676          * changed by use of 'ipfw fwd'.
 677          */
 678         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 679
 680         /*
 681          * Enable a consistency check between the destination address
 682          * and the arrival interface for a unicast packet (the RFC 1122
 683          * strong ES model) if IP forwarding is disabled and the packet
 684          * is not locally generated and the packet is not subject to
 685          * 'ipfw fwd'.
 686          *
 687          * XXX - Checking also should be disabled if the destination
 688          * address is ipnat'ed to a different interface.
 689          *
 690          * XXX - Checking is incompatible with IP aliases added
 691          * to the loopback interface instead of the interface where
 692          * the packets are received.
 693          */
 694         checkif = ip_checkinterface &&
 695                   !ipforwarding &&
 696                   m->m_pkthdr.rcvif != NULL &&
 697                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 698                   next_hop == NULL;
 699
 700         /*
 701          * Check for exact addresses in the hash bucket.
 702          */
 703         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 704                 ia = iac->ia;
 705
 706                 /*
 707                  * If the address matches, verify that the packet
 708                  * arrived via the correct interface if checking is
 709                  * enabled.
 710                  */
 711                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 712                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 713                         goto ours;
 714         }
 715         ia = NULL;
 716
 717         /*
 718          * Check for broadcast addresses.
 719          *
 720          * Only accept broadcast packets that arrive via the matching
 721          * interface.  Reception of forwarded directed broadcasts would
 722          * be handled via ip_forward() and ether_output() with the loopback
 723          * into the stack for SIMPLEX interfaces handled by ether_output().
 724          */
 725         if (m->m_pkthdr.rcvif != NULL &&
 726             m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 727                 struct ifaddr_container *ifac;
 728
 729                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 730                               ifa_link) {
 731                         struct ifaddr *ifa = ifac->ifa;
 732
 733                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 734                                 continue;
 735                         if (ifa->ifa_addr->sa_family != AF_INET)
 736                                 continue;
 737                         ia = ifatoia(ifa);
 738                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 739                                                                 pkt_dst.s_addr)
 740                                 goto ours;
 741                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 742                                 goto ours;
 743 #ifdef BOOTP_COMPAT
 744                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 745                                 goto ours;
 746 #endif
 747                 }
 748         }
 749         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 750                 struct in_multi *inm;
 751
 752                 if (ip_mrouter != NULL) {
 753                         /* XXX Multicast routing is not MPSAFE yet */
 754                         get_mplock();
 755
 756                         /*
 757                          * If we are acting as a multicast router, all
 758                          * incoming multicast packets are passed to the
 759                          * kernel-level multicast forwarding function.
 760                          * The packet is returned (relatively) intact; if
 761                          * ip_mforward() returns a non-zero value, the packet
 762                          * must be discarded, else it may be accepted below.
 763                          */
 764                         if (ip_mforward != NULL &&
 765                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 766                                 rel_mplock();
 767                                 ipstat.ips_cantforward++;
 768                                 m_freem(m);
 769                                 return;
 770                         }
 771
 772                         rel_mplock();
 773
 774                         /*
 775                          * The process-level routing daemon needs to receive
 776                          * all multicast IGMP packets, whether or not this
 777                          * host belongs to their destination groups.
 778                          */
 779                         if (ip->ip_p == IPPROTO_IGMP)
 780                                 goto ours;
 781                         ipstat.ips_forward++;
 782                 }
 783                 /*
 784                  * See if we belong to the destination multicast group on the
 785                  * arrival interface.
 786                  */
 787                 inm = IN_LOOKUP_MULTI(&ip->ip_dst, m->m_pkthdr.rcvif);
 788                 if (inm == NULL) {
 789                         ipstat.ips_notmember++;
 790                         m_freem(m);
 791                         return;
 792                 }
 793                 goto ours;
 794         }
 795         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 796                 goto ours;
 797         if (ip->ip_dst.s_addr == INADDR_ANY)
 798                 goto ours;
 799
 800         /*
 801          * FAITH(Firewall Aided Internet Translator)
 802          */
 803         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 804                 if (ip_keepfaith) {
 805                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 806                                 goto ours;
 807                 }
 808                 m_freem(m);
 809                 return;
 810         }
 811
 812         /*
 813          * Not for us; forward if possible and desirable.
 814          */
 815         if (!ipforwarding) {
 816                 ipstat.ips_cantforward++;
 817                 m_freem(m);
 818         } else {
 819 #ifdef IPSEC
 820                 /*
 821                  * Enforce inbound IPsec SPD.
 822                  */
 823                 if (ipsec4_in_reject(m, NULL)) {
 824                         ipsecstat.in_polvio++;
 825                         goto bad;
 826                 }
 827 #endif
 828 #ifdef FAST_IPSEC
 829                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 830                 crit_enter();
 831                 if (mtag != NULL) {
 832                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 833                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 834                 } else {
 835                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 836                                                    IP_FORWARDING, &error);
 837                 }
 838                 if (sp == NULL) {       /* NB: can happen if error */
 839                         crit_exit();
 840                         /*XXX error stat???*/
 841                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 842                         goto bad;
 843                 }
 844
 845                 /*
 846                  * Check security policy against packet attributes.
 847                  */
 848                 error = ipsec_in_reject(sp, m);
 849                 KEY_FREESP(&sp);
 850                 crit_exit();
 851                 if (error) {
 852                         ipstat.ips_cantforward++;
 853                         goto bad;
 854                 }
 855 #endif
 856                 ip_forward(m, using_srcrt, next_hop);
 857         }
 858         return;
 859
 860 ours:
 861
 862         /*
 863          * IPSTEALTH: Process non-routing options only
 864          * if the packet is destined for us.
 865          */
 866         if (ipstealth &&
 867             hlen > sizeof(struct ip) &&
 868             ip_dooptions(m, 1, next_hop))
 869                 return;
 870
 871         /* Count the packet in the ip address stats */
 872         if (ia != NULL) {
 873                 IFA_STAT_INC(&ia->ia_ifa, ipackets, 1);
 874                 IFA_STAT_INC(&ia->ia_ifa, ibytes, m->m_pkthdr.len);
 875         }
 876
 877         /*
 878          * If offset or IP_MF are set, must reassemble.
 879          * Otherwise, nothing need be done.
 880          * (We could look in the reassembly queue to see
 881          * if the packet was previously fragmented,
 882          * but it's not worth the time; just let them time out.)
 883          */
 884         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 885                 /*
 886                  * Attempt reassembly; if it succeeds, proceed.  ip_reass()
 887                  * will return a different mbuf.
 888                  *
 889                  * NOTE: ip_reass() returns m with M_HASH cleared to force
 890                  *       us to recharacterize the packet.
 891                  */
 892                 m = ip_reass(m);
 893                 if (m == NULL)
 894                         return;
 895                 ip = mtod(m, struct ip *);
 896
 897                 /* Get the header length of the reassembled packet */
 898                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 899         } else {
 900                 ip->ip_len -= hlen;
 901         }
 902
 903 #ifdef IPSEC
 904         /*
 905          * enforce IPsec policy checking if we are seeing last header.
 906          * note that we do not visit this with protocols with pcb layer
 907          * code - like udp/tcp/raw ip.
 908          */
 909         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 910             ipsec4_in_reject(m, NULL)) {
 911                 ipsecstat.in_polvio++;
 912                 goto bad;
 913         }
 914 #endif
 915 #if FAST_IPSEC
 916         /*
 917          * enforce IPsec policy checking if we are seeing last header.
 918          * note that we do not visit this with protocols with pcb layer
 919          * code - like udp/tcp/raw ip.
 920          */
 921         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 922                 /*
 923                  * Check if the packet has already had IPsec processing
 924                  * done.  If so, then just pass it along.  This tag gets
 925                  * set during AH, ESP, etc. input handling, before the
 926                  * packet is returned to the ip input queue for delivery.
 927                  */
 928                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 929                 crit_enter();
 930                 if (mtag != NULL) {
 931                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 932                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 933                 } else {
 934                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 935                                                    IP_FORWARDING, &error);
 936                 }
 937                 if (sp != NULL) {
 938                         /*
 939                          * Check security policy against packet attributes.
 940                          */
 941                         error = ipsec_in_reject(sp, m);
 942                         KEY_FREESP(&sp);
 943                 } else {
 944                         /* XXX error stat??? */
 945                         error = EINVAL;
 946 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 947                         crit_exit();
 948                         goto bad;
 949                 }
 950                 crit_exit();
 951                 if (error)
 952                         goto bad;
 953         }
 954 #endif /* FAST_IPSEC */
 955
 956         /*
 957          * We must forward the packet to the correct protocol thread if
 958          * we are not already in it.
 959          *
 960          * NOTE: ip_len is now in host form.  ip_len is not adjusted
 961          *       further for protocol processing, instead we pass hlen
 962          *       to the protosw and let it deal with it.
 963          */
 964         ipstat.ips_delivered++;
 965
 966         if ((m->m_flags & M_HASH) == 0) {
 967 #ifdef RSS_DEBUG
 968                 atomic_add_long(&ip_rehash_count, 1);
 969 #endif
 970                 ip->ip_len = htons(ip->ip_len + hlen);
 971                 ip->ip_off = htons(ip->ip_off);
 972
 973                 ip_hashfn(&m, 0, IP_MPORT_IN);
 974                 if (m == NULL)
 975                         return;
 976
 977                 ip = mtod(m, struct ip *);
 978                 ip->ip_len = ntohs(ip->ip_len) - hlen;
 979                 ip->ip_off = ntohs(ip->ip_off);
 980                 KKASSERT(m->m_flags & M_HASH);
 981         }
 982         port = netisr_hashport(m->m_pkthdr.hash);
 983
 984         if (port != &curthread->td_msgport) {
 985                 struct netmsg_packet *pmsg;
 986
 987 #ifdef RSS_DEBUG
 988                 atomic_add_long(&ip_dispatch_slow, 1);
 989 #endif
 990
 991                 pmsg = &m->m_hdr.mh_netmsg;
 992                 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
 993                             0, transport_processing_handler);
 994                 pmsg->nm_packet = m;
 995                 pmsg->base.lmsg.u.ms_result = hlen;
 996                 lwkt_sendmsg(port, &pmsg->base.lmsg);
 997         } else {
 998 #ifdef RSS_DEBUG
 999                 atomic_add_long(&ip_dispatch_fast, 1);
1000 #endif
1001                 transport_processing_oncpu(m, hlen, ip);
1002         }
1003         return;
1004
1005 bad:
1006         m_freem(m);
1007 }
1008
1009 /*
1010  * Take incoming datagram fragment and try to reassemble it into
1011  * whole datagram.  If a chain for reassembly of this datagram already
1012  * exists, then it is given as fp; otherwise have to make a chain.
1013  */
1014 struct mbuf *
1015 ip_reass(struct mbuf *m)
1016 {
1017         struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[mycpuid];
1018         struct ip *ip = mtod(m, struct ip *);
1019         struct mbuf *p = NULL, *q, *nq;
1020         struct mbuf *n;
1021         struct ipq *fp = NULL;
1022         struct ipqhead *head;
1023         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1024         int i, next;
1025         u_short sum;
1026
1027         /* If maxnipq is 0, never accept fragments. */
1028         if (maxnipq == 0) {
1029                 ipstat.ips_fragments++;
1030                 ipstat.ips_fragdropped++;
1031                 m_freem(m);
1032                 return NULL;
1033         }
1034
1035         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1036         /*
1037          * Look for queue of fragments of this datagram.
1038          */
1039         head = &fragq->ipq[sum];
1040         TAILQ_FOREACH(fp, head, ipq_list) {
1041                 if (ip->ip_id == fp->ipq_id &&
1042                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1043                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1044                     ip->ip_p == fp->ipq_p)
1045                         goto found;
1046         }
1047
1048         fp = NULL;
1049
1050         /*
1051          * Enforce upper bound on number of fragmented packets
1052          * for which we attempt reassembly;
1053          * If maxnipq is -1, accept all fragments without limitation.
1054          */
1055         if (fragq->nipq > maxnipq && maxnipq > 0) {
1056                 /*
1057                  * drop something from the tail of the current queue
1058                  * before proceeding further
1059                  */
1060                 struct ipq *q = TAILQ_LAST(head, ipqhead);
1061                 if (q == NULL) {
1062                         /*
1063                          * The current queue is empty,
1064                          * so drop from one of the others.
1065                          */
1066                         for (i = 0; i < IPREASS_NHASH; i++) {
1067                                 struct ipq *r = TAILQ_LAST(&fragq->ipq[i],
1068                                     ipqhead);
1069                                 if (r) {
1070                                         ipstat.ips_fragtimeout += r->ipq_nfrags;
1071                                         ip_freef(fragq, &fragq->ipq[i], r);
1072                                         break;
1073                                 }
1074                         }
1075                 } else {
1076                         ipstat.ips_fragtimeout += q->ipq_nfrags;
1077                         ip_freef(fragq, head, q);
1078                 }
1079         }
1080 found:
1081         /*
1082          * Adjust ip_len to not reflect header,
1083          * convert offset of this to bytes.
1084          */
1085         ip->ip_len -= hlen;
1086         if (ip->ip_off & IP_MF) {
1087                 /*
1088                  * Make sure that fragments have a data length
1089                  * that's a non-zero multiple of 8 bytes.
1090                  */
1091                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1092                         ipstat.ips_toosmall++; /* XXX */
1093                         m_freem(m);
1094                         goto done;
1095                 }
1096                 m->m_flags |= M_FRAG;
1097         } else {
1098                 m->m_flags &= ~M_FRAG;
1099         }
1100         ip->ip_off <<= 3;
1101
1102         ipstat.ips_fragments++;
1103         m->m_pkthdr.header = ip;
1104
1105         /*
1106          * If the hardware has not done csum over this fragment
1107          * then csum_data is not valid at all.
1108          */
1109         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1110             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1111                 m->m_pkthdr.csum_data = 0;
1112                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1113         }
1114
1115         /*
1116          * Presence of header sizes in mbufs
1117          * would confuse code below.
1118          */
1119         m->m_data += hlen;
1120         m->m_len -= hlen;
1121
1122         /*
1123          * If first fragment to arrive, create a reassembly queue.
1124          */
1125         if (fp == NULL) {
1126                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1127                         goto dropfrag;
1128                 TAILQ_INSERT_HEAD(head, fp, ipq_list);
1129                 fragq->nipq++;
1130                 fp->ipq_nfrags = 1;
1131                 fp->ipq_ttl = IPFRAGTTL;
1132                 fp->ipq_p = ip->ip_p;
1133                 fp->ipq_id = ip->ip_id;
1134                 fp->ipq_src = ip->ip_src;
1135                 fp->ipq_dst = ip->ip_dst;
1136                 fp->ipq_frags = m;
1137                 m->m_nextpkt = NULL;
1138                 goto inserted;
1139         } else {
1140                 fp->ipq_nfrags++;
1141         }
1142
1143 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1144
1145         /*
1146          * Find a segment which begins after this one does.
1147          */
1148         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1149                 if (GETIP(q)->ip_off > ip->ip_off)
1150                         break;
1151         }
1152
1153         /*
1154          * If there is a preceding segment, it may provide some of
1155          * our data already.  If so, drop the data from the incoming
1156          * segment.  If it provides all of our data, drop us, otherwise
1157          * stick new segment in the proper place.
1158          *
1159          * If some of the data is dropped from the the preceding
1160          * segment, then it's checksum is invalidated.
1161          */
1162         if (p) {
1163                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1164                 if (i > 0) {
1165                         if (i >= ip->ip_len)
1166                                 goto dropfrag;
1167                         m_adj(m, i);
1168                         m->m_pkthdr.csum_flags = 0;
1169                         ip->ip_off += i;
1170                         ip->ip_len -= i;
1171                 }
1172                 m->m_nextpkt = p->m_nextpkt;
1173                 p->m_nextpkt = m;
1174         } else {
1175                 m->m_nextpkt = fp->ipq_frags;
1176                 fp->ipq_frags = m;
1177         }
1178
1179         /*
1180          * While we overlap succeeding segments trim them or,
1181          * if they are completely covered, dequeue them.
1182          */
1183         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1184              q = nq) {
1185                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1186                 if (i < GETIP(q)->ip_len) {
1187                         GETIP(q)->ip_len -= i;
1188                         GETIP(q)->ip_off += i;
1189                         m_adj(q, i);
1190                         q->m_pkthdr.csum_flags = 0;
1191                         break;
1192                 }
1193                 nq = q->m_nextpkt;
1194                 m->m_nextpkt = nq;
1195                 ipstat.ips_fragdropped++;
1196                 fp->ipq_nfrags--;
1197                 q->m_nextpkt = NULL;
1198                 m_freem(q);
1199         }
1200
1201 inserted:
1202         /*
1203          * Check for complete reassembly and perform frag per packet
1204          * limiting.
1205          *
1206          * Frag limiting is performed here so that the nth frag has
1207          * a chance to complete the packet before we drop the packet.
1208          * As a result, n+1 frags are actually allowed per packet, but
1209          * only n will ever be stored. (n = maxfragsperpacket.)
1210          *
1211          */
1212         next = 0;
1213         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1214                 if (GETIP(q)->ip_off != next) {
1215                         if (fp->ipq_nfrags > maxfragsperpacket) {
1216                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1217                                 ip_freef(fragq, head, fp);
1218                         }
1219                         goto done;
1220                 }
1221                 next += GETIP(q)->ip_len;
1222         }
1223         /* Make sure the last packet didn't have the IP_MF flag */
1224         if (p->m_flags & M_FRAG) {
1225                 if (fp->ipq_nfrags > maxfragsperpacket) {
1226                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1227                         ip_freef(fragq, head, fp);
1228                 }
1229                 goto done;
1230         }
1231
1232         /*
1233          * Reassembly is complete.  Make sure the packet is a sane size.
1234          */
1235         q = fp->ipq_frags;
1236         ip = GETIP(q);
1237         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1238                 ipstat.ips_toolong++;
1239                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1240                 ip_freef(fragq, head, fp);
1241                 goto done;
1242         }
1243
1244         /*
1245          * Concatenate fragments.
1246          */
1247         m = q;
1248         n = m->m_next;
1249         m->m_next = NULL;
1250         m_cat(m, n);
1251         nq = q->m_nextpkt;
1252         q->m_nextpkt = NULL;
1253         for (q = nq; q != NULL; q = nq) {
1254                 nq = q->m_nextpkt;
1255                 q->m_nextpkt = NULL;
1256                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1257                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1258                 m_cat(m, q);
1259         }
1260
1261         /*
1262          * Clean up the 1's complement checksum.  Carry over 16 bits must
1263          * be added back.  This assumes no more then 65535 packet fragments
1264          * were reassembled.  A second carry can also occur (but not a third).
1265          */
1266         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1267                                 (m->m_pkthdr.csum_data >> 16);
1268         if (m->m_pkthdr.csum_data > 0xFFFF)
1269                 m->m_pkthdr.csum_data -= 0xFFFF;
1270
1271         /*
1272          * Create header for new ip packet by
1273          * modifying header of first packet;
1274          * dequeue and discard fragment reassembly header.
1275          * Make header visible.
1276          */
1277         ip->ip_len = next;
1278         ip->ip_src = fp->ipq_src;
1279         ip->ip_dst = fp->ipq_dst;
1280         TAILQ_REMOVE(head, fp, ipq_list);
1281         fragq->nipq--;
1282         mpipe_free(&ipq_mpipe, fp);
1283         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1284         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1285         /* some debugging cruft by sklower, below, will go away soon */
1286         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1287                 int plen = 0;
1288
1289                 for (n = m; n; n = n->m_next)
1290                         plen += n->m_len;
1291                 m->m_pkthdr.len = plen;
1292         }
1293
1294         /*
1295          * Reassembly complete, return the next protocol.
1296          *
1297          * Be sure to clear M_HASH to force the packet
1298          * to be re-characterized.
1299          *
1300          * Clear M_FRAG, we are no longer a fragment.
1301          */
1302         m->m_flags &= ~(M_HASH | M_FRAG);
1303
1304         ipstat.ips_reassembled++;
1305         return (m);
1306
1307 dropfrag:
1308         ipstat.ips_fragdropped++;
1309         if (fp != NULL)
1310                 fp->ipq_nfrags--;
1311         m_freem(m);
1312 done:
1313         return (NULL);
1314
1315 #undef GETIP
1316 }
1317
1318 /*
1319  * Free a fragment reassembly header and all
1320  * associated datagrams.
1321  */
1322 static void
1323 ip_freef(struct ipfrag_queue *fragq, struct ipqhead *fhp, struct ipq *fp)
1324 {
1325         struct mbuf *q;
1326
1327         /*
1328          * Remove first to protect against blocking
1329          */
1330         TAILQ_REMOVE(fhp, fp, ipq_list);
1331
1332         /*
1333          * Clean out at our leisure
1334          */
1335         while (fp->ipq_frags) {
1336                 q = fp->ipq_frags;
1337                 fp->ipq_frags = q->m_nextpkt;
1338                 q->m_nextpkt = NULL;
1339                 m_freem(q);
1340         }
1341         mpipe_free(&ipq_mpipe, fp);
1342         fragq->nipq--;
1343 }
1344
1345 /*
1346  * If a timer expires on a reassembly queue, discard it.
1347  */
1348 static void
1349 ipfrag_timeo_dispatch(netmsg_t nmsg)
1350 {
1351         struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[mycpuid];
1352         struct ipq *fp, *fp_temp;
1353         struct ipqhead *head;
1354         int i;
1355
1356         crit_enter();
1357         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1358         crit_exit();
1359
1360         for (i = 0; i < IPREASS_NHASH; i++) {
1361                 head = &fragq->ipq[i];
1362                 TAILQ_FOREACH_MUTABLE(fp, head, ipq_list, fp_temp) {
1363                         if (--fp->ipq_ttl == 0) {
1364                                 ipstat.ips_fragtimeout += fp->ipq_nfrags;
1365                                 ip_freef(fragq, head, fp);
1366                         }
1367                 }
1368         }
1369         /*
1370          * If we are over the maximum number of fragments
1371          * (due to the limit being lowered), drain off
1372          * enough to get down to the new limit.
1373          */
1374         if (maxnipq >= 0 && fragq->nipq > maxnipq) {
1375                 for (i = 0; i < IPREASS_NHASH; i++) {
1376                         head = &fragq->ipq[i];
1377                         while (fragq->nipq > maxnipq && !TAILQ_EMPTY(head)) {
1378                                 ipstat.ips_fragdropped +=
1379                                     TAILQ_FIRST(head)->ipq_nfrags;
1380                                 ip_freef(fragq, head, TAILQ_FIRST(head));
1381                         }
1382                 }
1383         }
1384 }
1385
1386 static void
1387 ipfrag_timeo_ipi(void *arg __unused)
1388 {
1389         int cpu = mycpuid;
1390         struct lwkt_msg *msg = &ipfrag_queue_pcpu[cpu].timeo_netmsg.lmsg;
1391
1392         crit_enter();
1393         if (msg->ms_flags & MSGF_DONE)
1394                 lwkt_sendmsg_oncpu(netisr_cpuport(cpu), msg);
1395         crit_exit();
1396 }
1397
1398 static void
1399 ipfrag_slowtimo(void)
1400 {
1401         cpumask_t mask = 0;
1402         int i;
1403
1404         for (i = 0; i < ncpus; ++i)
1405                 mask |= CPUMASK(i);
1406         mask &= smp_active_mask;
1407         if (mask != 0)
1408                 lwkt_send_ipiq_mask(mask, ipfrag_timeo_ipi, NULL);
1409 }
1410
1411 /*
1412  * IP timer processing
1413  */
1414 void
1415 ip_slowtimo(void)
1416 {
1417         ipfrag_slowtimo();
1418         ipflow_slowtimo();
1419 }
1420
1421 /*
1422  * Drain off all datagram fragments.
1423  */
1424 static void
1425 ipfrag_drain_dispatch(netmsg_t nmsg)
1426 {
1427         struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[mycpuid];
1428         struct ipqhead *head;
1429         int i;
1430
1431         crit_enter();
1432         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1433         crit_exit();
1434
1435         for (i = 0; i < IPREASS_NHASH; i++) {
1436                 head = &fragq->ipq[i];
1437                 while (!TAILQ_EMPTY(head)) {
1438                         ipstat.ips_fragdropped += TAILQ_FIRST(head)->ipq_nfrags;
1439                         ip_freef(fragq, head, TAILQ_FIRST(head));
1440                 }
1441         }
1442 }
1443
1444 static void
1445 ipfrag_drain_ipi(void *arg __unused)
1446 {
1447         int cpu = mycpuid;
1448         struct lwkt_msg *msg = &ipfrag_queue_pcpu[cpu].drain_netmsg.lmsg;
1449
1450         crit_enter();
1451         if (msg->ms_flags & MSGF_DONE)
1452                 lwkt_sendmsg_oncpu(netisr_cpuport(cpu), msg);
1453         crit_exit();
1454 }
1455
1456 static void
1457 ipfrag_drain(void)
1458 {
1459         cpumask_t mask = 0;
1460         int i;
1461
1462         for (i = 0; i < ncpus; ++i)
1463                 mask |= CPUMASK(i);
1464         mask &= smp_active_mask;
1465         if (mask != 0)
1466                 lwkt_send_ipiq_mask(mask, ipfrag_drain_ipi, NULL);
1467 }
1468
1469 void
1470 ip_drain(void)
1471 {
1472         ipfrag_drain();
1473         in_rtqdrain();
1474 }
1475
1476 /*
1477  * Do option processing on a datagram,
1478  * possibly discarding it if bad options are encountered,
1479  * or forwarding it if source-routed.
1480  * The pass argument is used when operating in the IPSTEALTH
1481  * mode to tell what options to process:
1482  * [LS]SRR (pass 0) or the others (pass 1).
1483  * The reason for as many as two passes is that when doing IPSTEALTH,
1484  * non-routing options should be processed only if the packet is for us.
1485  * Returns 1 if packet has been forwarded/freed,
1486  * 0 if the packet should be processed further.
1487  */
1488 static int
1489 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1490 {
1491         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1492         struct ip *ip = mtod(m, struct ip *);
1493         u_char *cp;
1494         struct in_ifaddr *ia;
1495         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1496         boolean_t forward = FALSE;
1497         struct in_addr *sin, dst;
1498         n_time ntime;
1499
1500         dst = ip->ip_dst;
1501         cp = (u_char *)(ip + 1);
1502         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1503         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1504                 opt = cp[IPOPT_OPTVAL];
1505                 if (opt == IPOPT_EOL)
1506                         break;
1507                 if (opt == IPOPT_NOP)
1508                         optlen = 1;
1509                 else {
1510                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1511                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1512                                 goto bad;
1513                         }
1514                         optlen = cp[IPOPT_OLEN];
1515                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1516                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1517                                 goto bad;
1518                         }
1519                 }
1520                 switch (opt) {
1521
1522                 default:
1523                         break;
1524
1525                 /*
1526                  * Source routing with record.
1527                  * Find interface with current destination address.
1528                  * If none on this machine then drop if strictly routed,
1529                  * or do nothing if loosely routed.
1530                  * Record interface address and bring up next address
1531                  * component.  If strictly routed make sure next
1532                  * address is on directly accessible net.
1533                  */
1534                 case IPOPT_LSRR:
1535                 case IPOPT_SSRR:
1536                         if (ipstealth && pass > 0)
1537                                 break;
1538                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1539                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1540                                 goto bad;
1541                         }
1542                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1543                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1544                                 goto bad;
1545                         }
1546                         ipaddr.sin_addr = ip->ip_dst;
1547                         ia = (struct in_ifaddr *)
1548                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1549                         if (ia == NULL) {
1550                                 if (opt == IPOPT_SSRR) {
1551                                         type = ICMP_UNREACH;
1552                                         code = ICMP_UNREACH_SRCFAIL;
1553                                         goto bad;
1554                                 }
1555                                 if (!ip_dosourceroute)
1556                                         goto nosourcerouting;
1557                                 /*
1558                                  * Loose routing, and not at next destination
1559                                  * yet; nothing to do except forward.
1560                                  */
1561                                 break;
1562                         }
1563                         off--;                  /* 0 origin */
1564                         if (off > optlen - (int)sizeof(struct in_addr)) {
1565                                 /*
1566                                  * End of source route.  Should be for us.
1567                                  */
1568                                 if (!ip_acceptsourceroute)
1569                                         goto nosourcerouting;
1570                                 save_rte(m, cp, ip->ip_src);
1571                                 break;
1572                         }
1573                         if (ipstealth)
1574                                 goto dropit;
1575                         if (!ip_dosourceroute) {
1576                                 if (ipforwarding) {
1577                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1578
1579                                         /*
1580                                          * Acting as a router, so generate ICMP
1581                                          */
1582 nosourcerouting:
1583                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1584                                         log(LOG_WARNING,
1585                                             "attempted source route from %s to %s\n",
1586                                             inet_ntoa(ip->ip_src), buf);
1587                                         type = ICMP_UNREACH;
1588                                         code = ICMP_UNREACH_SRCFAIL;
1589                                         goto bad;
1590                                 } else {
1591                                         /*
1592                                          * Not acting as a router,
1593                                          * so silently drop.
1594                                          */
1595 dropit:
1596                                         ipstat.ips_cantforward++;
1597                                         m_freem(m);
1598                                         return (1);
1599                                 }
1600                         }
1601
1602                         /*
1603                          * locate outgoing interface
1604                          */
1605                         memcpy(&ipaddr.sin_addr, cp + off,
1606                             sizeof ipaddr.sin_addr);
1607
1608                         if (opt == IPOPT_SSRR) {
1609 #define INA     struct in_ifaddr *
1610 #define SA      struct sockaddr *
1611                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1612                                                                         == NULL)
1613                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1614                         } else {
1615                                 ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1616                         }
1617                         if (ia == NULL) {
1618                                 type = ICMP_UNREACH;
1619                                 code = ICMP_UNREACH_SRCFAIL;
1620                                 goto bad;
1621                         }
1622                         ip->ip_dst = ipaddr.sin_addr;
1623                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1624                             sizeof(struct in_addr));
1625                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1626                         /*
1627                          * Let ip_intr's mcast routing check handle mcast pkts
1628                          */
1629                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1630                         break;
1631
1632                 case IPOPT_RR:
1633                         if (ipstealth && pass == 0)
1634                                 break;
1635                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1636                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1637                                 goto bad;
1638                         }
1639                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1640                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1641                                 goto bad;
1642                         }
1643                         /*
1644                          * If no space remains, ignore.
1645                          */
1646                         off--;                  /* 0 origin */
1647                         if (off > optlen - (int)sizeof(struct in_addr))
1648                                 break;
1649                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1650                             sizeof ipaddr.sin_addr);
1651                         /*
1652                          * locate outgoing interface; if we're the destination,
1653                          * use the incoming interface (should be same).
1654                          */
1655                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1656                             (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1657                                 type = ICMP_UNREACH;
1658                                 code = ICMP_UNREACH_HOST;
1659                                 goto bad;
1660                         }
1661                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1662                             sizeof(struct in_addr));
1663                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1664                         break;
1665
1666                 case IPOPT_TS:
1667                         if (ipstealth && pass == 0)
1668                                 break;
1669                         code = cp - (u_char *)ip;
1670                         if (optlen < 4 || optlen > 40) {
1671                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1672                                 goto bad;
1673                         }
1674                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1675                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1676                                 goto bad;
1677                         }
1678                         if (off > optlen - (int)sizeof(int32_t)) {
1679                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1680                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1681                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1682                                         goto bad;
1683                                 }
1684                                 break;
1685                         }
1686                         off--;                          /* 0 origin */
1687                         sin = (struct in_addr *)(cp + off);
1688                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1689
1690                         case IPOPT_TS_TSONLY:
1691                                 break;
1692
1693                         case IPOPT_TS_TSANDADDR:
1694                                 if (off + sizeof(n_time) +
1695                                     sizeof(struct in_addr) > optlen) {
1696                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1697                                         goto bad;
1698                                 }
1699                                 ipaddr.sin_addr = dst;
1700                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1701                                                             m->m_pkthdr.rcvif);
1702                                 if (ia == NULL)
1703                                         continue;
1704                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1705                                     sizeof(struct in_addr));
1706                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1707                                 off += sizeof(struct in_addr);
1708                                 break;
1709
1710                         case IPOPT_TS_PRESPEC:
1711                                 if (off + sizeof(n_time) +
1712                                     sizeof(struct in_addr) > optlen) {
1713                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1714                                         goto bad;
1715                                 }
1716                                 memcpy(&ipaddr.sin_addr, sin,
1717                                     sizeof(struct in_addr));
1718                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1719                                         continue;
1720                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1721                                 off += sizeof(struct in_addr);
1722                                 break;
1723
1724                         default:
1725                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1726                                 goto bad;
1727                         }
1728                         ntime = iptime();
1729                         memcpy(cp + off, &ntime, sizeof(n_time));
1730                         cp[IPOPT_OFFSET] += sizeof(n_time);
1731                 }
1732         }
1733         if (forward && ipforwarding) {
1734                 ip_forward(m, TRUE, next_hop);
1735                 return (1);
1736         }
1737         return (0);
1738 bad:
1739         icmp_error(m, type, code, 0, 0);
1740         ipstat.ips_badoptions++;
1741         return (1);
1742 }
1743
1744 /*
1745  * Given address of next destination (final or next hop),
1746  * return internet address info of interface to be used to get there.
1747  */
1748 struct in_ifaddr *
1749 ip_rtaddr(struct in_addr dst, struct route *ro0)
1750 {
1751         struct route sro, *ro;
1752         struct sockaddr_in *sin;
1753         struct in_ifaddr *ia;
1754
1755         if (ro0 != NULL) {
1756                 ro = ro0;
1757         } else {
1758                 bzero(&sro, sizeof(sro));
1759                 ro = &sro;
1760         }
1761
1762         sin = (struct sockaddr_in *)&ro->ro_dst;
1763
1764         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1765                 if (ro->ro_rt != NULL) {
1766                         RTFREE(ro->ro_rt);
1767                         ro->ro_rt = NULL;
1768                 }
1769                 sin->sin_family = AF_INET;
1770                 sin->sin_len = sizeof *sin;
1771                 sin->sin_addr = dst;
1772                 rtalloc_ign(ro, RTF_PRCLONING);
1773         }
1774
1775         if (ro->ro_rt == NULL)
1776                 return (NULL);
1777
1778         ia = ifatoia(ro->ro_rt->rt_ifa);
1779
1780         if (ro == &sro)
1781                 RTFREE(ro->ro_rt);
1782         return ia;
1783 }
1784
1785 /*
1786  * Save incoming source route for use in replies,
1787  * to be picked up later by ip_srcroute if the receiver is interested.
1788  */
1789 static void
1790 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1791 {
1792         struct m_tag *mtag;
1793         struct ip_srcrt_opt *opt;
1794         unsigned olen;
1795
1796         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1797         if (mtag == NULL)
1798                 return;
1799         opt = m_tag_data(mtag);
1800
1801         olen = option[IPOPT_OLEN];
1802 #ifdef DIAGNOSTIC
1803         if (ipprintfs)
1804                 kprintf("save_rte: olen %d\n", olen);
1805 #endif
1806         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1807                 m_tag_free(mtag);
1808                 return;
1809         }
1810         bcopy(option, opt->ip_srcrt.srcopt, olen);
1811         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1812         opt->ip_srcrt.dst = dst;
1813         m_tag_prepend(m, mtag);
1814 }
1815
1816 /*
1817  * Retrieve incoming source route for use in replies,
1818  * in the same form used by setsockopt.
1819  * The first hop is placed before the options, will be removed later.
1820  */
1821 struct mbuf *
1822 ip_srcroute(struct mbuf *m0)
1823 {
1824         struct in_addr *p, *q;
1825         struct mbuf *m;
1826         struct m_tag *mtag;
1827         struct ip_srcrt_opt *opt;
1828
1829         if (m0 == NULL)
1830                 return NULL;
1831
1832         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1833         if (mtag == NULL)
1834                 return NULL;
1835         opt = m_tag_data(mtag);
1836
1837         if (opt->ip_nhops == 0)
1838                 return (NULL);
1839         m = m_get(MB_DONTWAIT, MT_HEADER);
1840         if (m == NULL)
1841                 return (NULL);
1842
1843 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1844
1845         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1846         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1847                    sizeof(struct in_addr) + OPTSIZ;
1848 #ifdef DIAGNOSTIC
1849         if (ipprintfs) {
1850                 kprintf("ip_srcroute: nhops %d mlen %d",
1851                         opt->ip_nhops, m->m_len);
1852         }
1853 #endif
1854
1855         /*
1856          * First save first hop for return route
1857          */
1858         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1859         *(mtod(m, struct in_addr *)) = *p--;
1860 #ifdef DIAGNOSTIC
1861         if (ipprintfs)
1862                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1863 #endif
1864
1865         /*
1866          * Copy option fields and padding (nop) to mbuf.
1867          */
1868         opt->ip_srcrt.nop = IPOPT_NOP;
1869         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1870         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1871             OPTSIZ);
1872         q = (struct in_addr *)(mtod(m, caddr_t) +
1873             sizeof(struct in_addr) + OPTSIZ);
1874 #undef OPTSIZ
1875         /*
1876          * Record return path as an IP source route,
1877          * reversing the path (pointers are now aligned).
1878          */
1879         while (p >= opt->ip_srcrt.route) {
1880 #ifdef DIAGNOSTIC
1881                 if (ipprintfs)
1882                         kprintf(" %x", ntohl(q->s_addr));
1883 #endif
1884                 *q++ = *p--;
1885         }
1886         /*
1887          * Last hop goes to final destination.
1888          */
1889         *q = opt->ip_srcrt.dst;
1890         m_tag_delete(m0, mtag);
1891 #ifdef DIAGNOSTIC
1892         if (ipprintfs)
1893                 kprintf(" %x\n", ntohl(q->s_addr));
1894 #endif
1895         return (m);
1896 }
1897
1898 /*
1899  * Strip out IP options.
1900  */
1901 void
1902 ip_stripoptions(struct mbuf *m)
1903 {
1904         int datalen;
1905         struct ip *ip = mtod(m, struct ip *);
1906         caddr_t opts;
1907         int optlen;
1908
1909         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1910         opts = (caddr_t)(ip + 1);
1911         datalen = m->m_len - (sizeof(struct ip) + optlen);
1912         bcopy(opts + optlen, opts, datalen);
1913         m->m_len -= optlen;
1914         if (m->m_flags & M_PKTHDR)
1915                 m->m_pkthdr.len -= optlen;
1916         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1917 }
1918
1919 u_char inetctlerrmap[PRC_NCMDS] = {
1920         0,              0,              0,              0,
1921         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1922         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1923         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1924         0,              0,              0,              0,
1925         ENOPROTOOPT,    ECONNREFUSED
1926 };
1927
1928 /*
1929  * Forward a packet.  If some error occurs return the sender
1930  * an icmp packet.  Note we can't always generate a meaningful
1931  * icmp message because icmp doesn't have a large enough repertoire
1932  * of codes and types.
1933  *
1934  * If not forwarding, just drop the packet.  This could be confusing
1935  * if ipforwarding was zero but some routing protocol was advancing
1936  * us as a gateway to somewhere.  However, we must let the routing
1937  * protocol deal with that.
1938  *
1939  * The using_srcrt parameter indicates whether the packet is being forwarded
1940  * via a source route.
1941  */
1942 void
1943 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1944 {
1945         struct ip *ip = mtod(m, struct ip *);
1946         struct rtentry *rt;
1947         struct route fwd_ro;
1948         int error, type = 0, code = 0, destmtu = 0;
1949         struct mbuf *mcopy, *mtemp = NULL;
1950         n_long dest;
1951         struct in_addr pkt_dst;
1952
1953         dest = INADDR_ANY;
1954         /*
1955          * Cache the destination address of the packet; this may be
1956          * changed by use of 'ipfw fwd'.
1957          */
1958         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1959
1960 #ifdef DIAGNOSTIC
1961         if (ipprintfs)
1962                 kprintf("forward: src %x dst %x ttl %x\n",
1963                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1964 #endif
1965
1966         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1967                 ipstat.ips_cantforward++;
1968                 m_freem(m);
1969                 return;
1970         }
1971         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1972                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1973                 return;
1974         }
1975
1976         bzero(&fwd_ro, sizeof(fwd_ro));
1977         ip_rtaddr(pkt_dst, &fwd_ro);
1978         if (fwd_ro.ro_rt == NULL) {
1979                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1980                 return;
1981         }
1982         rt = fwd_ro.ro_rt;
1983
1984         if (curthread->td_type == TD_TYPE_NETISR) {
1985                 /*
1986                  * Save the IP header and at most 8 bytes of the payload,
1987                  * in case we need to generate an ICMP message to the src.
1988                  */
1989                 mtemp = ipforward_mtemp[mycpuid];
1990                 KASSERT((mtemp->m_flags & M_EXT) == 0 &&
1991                     mtemp->m_data == mtemp->m_pktdat &&
1992                     m_tag_first(mtemp) == NULL,
1993                     ("ip_forward invalid mtemp1"));
1994
1995                 if (!m_dup_pkthdr(mtemp, m, MB_DONTWAIT)) {
1996                         /*
1997                          * It's probably ok if the pkthdr dup fails (because
1998                          * the deep copy of the tag chain failed), but for now
1999                          * be conservative and just discard the copy since
2000                          * code below may some day want the tags.
2001                          */
2002                         mtemp = NULL;
2003                 } else {
2004                         mtemp->m_type = m->m_type;
2005                         mtemp->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
2006                             (int)ip->ip_len);
2007                         mtemp->m_pkthdr.len = mtemp->m_len;
2008                         m_copydata(m, 0, mtemp->m_len, mtod(mtemp, caddr_t));
2009                 }
2010         }
2011
2012         if (!ipstealth)
2013                 ip->ip_ttl -= IPTTLDEC;
2014
2015         /*
2016          * If forwarding packet using same interface that it came in on,
2017          * perhaps should send a redirect to sender to shortcut a hop.
2018          * Only send redirect if source is sending directly to us,
2019          * and if packet was not source routed (or has any options).
2020          * Also, don't send redirect if forwarding using a default route
2021          * or a route modified by a redirect.
2022          */
2023         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
2024             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
2025             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
2026             ipsendredirects && !using_srcrt && next_hop == NULL) {
2027                 u_long src = ntohl(ip->ip_src.s_addr);
2028                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
2029
2030                 if (rt_ifa != NULL &&
2031                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
2032                         if (rt->rt_flags & RTF_GATEWAY)
2033                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
2034                         else
2035                                 dest = pkt_dst.s_addr;
2036                         /*
2037                          * Router requirements says to only send
2038                          * host redirects.
2039                          */
2040                         type = ICMP_REDIRECT;
2041                         code = ICMP_REDIRECT_HOST;
2042 #ifdef DIAGNOSTIC
2043                         if (ipprintfs)
2044                                 kprintf("redirect (%d) to %x\n", code, dest);
2045 #endif
2046                 }
2047         }
2048
2049         error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
2050         if (error == 0) {
2051                 ipstat.ips_forward++;
2052                 if (type == 0) {
2053                         if (mtemp)
2054                                 ipflow_create(&fwd_ro, mtemp);
2055                         goto done;
2056                 } else {
2057                         ipstat.ips_redirectsent++;
2058                 }
2059         } else {
2060                 ipstat.ips_cantforward++;
2061         }
2062
2063         if (mtemp == NULL)
2064                 goto done;
2065
2066         /*
2067          * Errors that do not require generating ICMP message
2068          */
2069         switch (error) {
2070         case ENOBUFS:
2071                 /*
2072                  * A router should not generate ICMP_SOURCEQUENCH as
2073                  * required in RFC1812 Requirements for IP Version 4 Routers.
2074                  * Source quench could be a big problem under DoS attacks,
2075                  * or if the underlying interface is rate-limited.
2076                  * Those who need source quench packets may re-enable them
2077                  * via the net.inet.ip.sendsourcequench sysctl.
2078                  */
2079                 if (!ip_sendsourcequench)
2080                         goto done;
2081                 break;
2082
2083         case EACCES:                    /* ipfw denied packet */
2084                 goto done;
2085         }
2086
2087         KASSERT((mtemp->m_flags & M_EXT) == 0 &&
2088             mtemp->m_data == mtemp->m_pktdat,
2089             ("ip_forward invalid mtemp2"));
2090         mcopy = m_copym(mtemp, 0, mtemp->m_len, MB_DONTWAIT);
2091         if (mcopy == NULL)
2092                 goto done;
2093
2094         /*
2095          * Send ICMP message.
2096          */
2097         switch (error) {
2098         case 0:                         /* forwarded, but need redirect */
2099                 /* type, code set above */
2100                 break;
2101
2102         case ENETUNREACH:               /* shouldn't happen, checked above */
2103         case EHOSTUNREACH:
2104         case ENETDOWN:
2105         case EHOSTDOWN:
2106         default:
2107                 type = ICMP_UNREACH;
2108                 code = ICMP_UNREACH_HOST;
2109                 break;
2110
2111         case EMSGSIZE:
2112                 type = ICMP_UNREACH;
2113                 code = ICMP_UNREACH_NEEDFRAG;
2114 #ifdef IPSEC
2115                 /*
2116                  * If the packet is routed over IPsec tunnel, tell the
2117                  * originator the tunnel MTU.
2118                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2119                  * XXX quickhack!!!
2120                  */
2121                 if (fwd_ro.ro_rt != NULL) {
2122                         struct secpolicy *sp = NULL;
2123                         int ipsecerror;
2124                         int ipsechdr;
2125                         struct route *ro;
2126
2127                         sp = ipsec4_getpolicybyaddr(mcopy,
2128                                                     IPSEC_DIR_OUTBOUND,
2129                                                     IP_FORWARDING,
2130                                                     &ipsecerror);
2131
2132                         if (sp == NULL)
2133                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2134                         else {
2135                                 /* count IPsec header size */
2136                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2137                                                          IPSEC_DIR_OUTBOUND,
2138                                                          NULL);
2139
2140                                 /*
2141                                  * find the correct route for outer IPv4
2142                                  * header, compute tunnel MTU.
2143                                  *
2144                                  */
2145                                 if (sp->req != NULL && sp->req->sav != NULL &&
2146                                     sp->req->sav->sah != NULL) {
2147                                         ro = &sp->req->sav->sah->sa_route;
2148                                         if (ro->ro_rt != NULL &&
2149                                             ro->ro_rt->rt_ifp != NULL) {
2150                                                 destmtu =
2151                                                     ro->ro_rt->rt_ifp->if_mtu;
2152                                                 destmtu -= ipsechdr;
2153                                         }
2154                                 }
2155
2156                                 key_freesp(sp);
2157                         }
2158                 }
2159 #elif FAST_IPSEC
2160                 /*
2161                  * If the packet is routed over IPsec tunnel, tell the
2162                  * originator the tunnel MTU.
2163                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2164                  * XXX quickhack!!!
2165                  */
2166                 if (fwd_ro.ro_rt != NULL) {
2167                         struct secpolicy *sp = NULL;
2168                         int ipsecerror;
2169                         int ipsechdr;
2170                         struct route *ro;
2171
2172                         sp = ipsec_getpolicybyaddr(mcopy,
2173                                                    IPSEC_DIR_OUTBOUND,
2174                                                    IP_FORWARDING,
2175                                                    &ipsecerror);
2176
2177                         if (sp == NULL)
2178                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2179                         else {
2180                                 /* count IPsec header size */
2181                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2182                                                          IPSEC_DIR_OUTBOUND,
2183                                                          NULL);
2184
2185                                 /*
2186                                  * find the correct route for outer IPv4
2187                                  * header, compute tunnel MTU.
2188                                  */
2189
2190                                 if (sp->req != NULL &&
2191                                     sp->req->sav != NULL &&
2192                                     sp->req->sav->sah != NULL) {
2193                                         ro = &sp->req->sav->sah->sa_route;
2194                                         if (ro->ro_rt != NULL &&
2195                                             ro->ro_rt->rt_ifp != NULL) {
2196                                                 destmtu =
2197                                                     ro->ro_rt->rt_ifp->if_mtu;
2198                                                 destmtu -= ipsechdr;
2199                                         }
2200                                 }
2201
2202                                 KEY_FREESP(&sp);
2203                         }
2204                 }
2205 #else /* !IPSEC && !FAST_IPSEC */
2206                 if (fwd_ro.ro_rt != NULL)
2207                         destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2208 #endif /*IPSEC*/
2209                 ipstat.ips_cantfrag++;
2210                 break;
2211
2212         case ENOBUFS:
2213                 type = ICMP_SOURCEQUENCH;
2214                 code = 0;
2215                 break;
2216
2217         case EACCES:                    /* ipfw denied packet */
2218                 panic("ip_forward EACCES should not reach");
2219         }
2220         icmp_error(mcopy, type, code, dest, destmtu);
2221 done:
2222         if (mtemp != NULL)
2223                 m_tag_delete_chain(mtemp);
2224         if (fwd_ro.ro_rt != NULL)
2225                 RTFREE(fwd_ro.ro_rt);
2226 }
2227
2228 void
2229 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2230                struct mbuf *m)
2231 {
2232         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2233                 struct timeval tv;
2234
2235                 microtime(&tv);
2236                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2237                     SCM_TIMESTAMP, SOL_SOCKET);
2238                 if (*mp)
2239                         mp = &(*mp)->m_next;
2240         }
2241         if (inp->inp_flags & INP_RECVDSTADDR) {
2242                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2243                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2244                 if (*mp)
2245                         mp = &(*mp)->m_next;
2246         }
2247         if (inp->inp_flags & INP_RECVTTL) {
2248                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2249                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2250                 if (*mp)
2251                         mp = &(*mp)->m_next;
2252         }
2253 #ifdef notyet
2254         /* XXX
2255          * Moving these out of udp_input() made them even more broken
2256          * than they already were.
2257          */
2258         /* options were tossed already */
2259         if (inp->inp_flags & INP_RECVOPTS) {
2260                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2261                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2262                 if (*mp)
2263                         mp = &(*mp)->m_next;
2264         }
2265         /* ip_srcroute doesn't do what we want here, need to fix */
2266         if (inp->inp_flags & INP_RECVRETOPTS) {
2267                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2268                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2269                 if (*mp)
2270                         mp = &(*mp)->m_next;
2271         }
2272 #endif
2273         if (inp->inp_flags & INP_RECVIF) {
2274                 struct ifnet *ifp;
2275                 struct sdlbuf {
2276                         struct sockaddr_dl sdl;
2277                         u_char  pad[32];
2278                 } sdlbuf;
2279                 struct sockaddr_dl *sdp;
2280                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2281
2282                 if (((ifp = m->m_pkthdr.rcvif)) &&
2283                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2284                         sdp = IF_LLSOCKADDR(ifp);
2285                         /*
2286                          * Change our mind and don't try copy.
2287                          */
2288                         if ((sdp->sdl_family != AF_LINK) ||
2289                             (sdp->sdl_len > sizeof(sdlbuf))) {
2290                                 goto makedummy;
2291                         }
2292                         bcopy(sdp, sdl2, sdp->sdl_len);
2293                 } else {
2294 makedummy:
2295                         sdl2->sdl_len =
2296                             offsetof(struct sockaddr_dl, sdl_data[0]);
2297                         sdl2->sdl_family = AF_LINK;
2298                         sdl2->sdl_index = 0;
2299                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2300                 }
2301                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2302                         IP_RECVIF, IPPROTO_IP);
2303                 if (*mp)
2304                         mp = &(*mp)->m_next;
2305         }
2306 }
2307
2308 /*
2309  * XXX these routines are called from the upper part of the kernel.
2310  *
2311  * They could also be moved to ip_mroute.c, since all the RSVP
2312  *  handling is done there already.
2313  */
2314 int
2315 ip_rsvp_init(struct socket *so)
2316 {
2317         if (so->so_type != SOCK_RAW ||
2318             so->so_proto->pr_protocol != IPPROTO_RSVP)
2319                 return EOPNOTSUPP;
2320
2321         if (ip_rsvpd != NULL)
2322                 return EADDRINUSE;
2323
2324         ip_rsvpd = so;
2325         /*
2326          * This may seem silly, but we need to be sure we don't over-increment
2327          * the RSVP counter, in case something slips up.
2328          */
2329         if (!ip_rsvp_on) {
2330                 ip_rsvp_on = 1;
2331                 rsvp_on++;
2332         }
2333
2334         return 0;
2335 }
2336
2337 int
2338 ip_rsvp_done(void)
2339 {
2340         ip_rsvpd = NULL;
2341         /*
2342          * This may seem silly, but we need to be sure we don't over-decrement
2343          * the RSVP counter, in case something slips up.
2344          */
2345         if (ip_rsvp_on) {
2346                 ip_rsvp_on = 0;
2347                 rsvp_on--;
2348         }
2349         return 0;
2350 }
2351
2352 int
2353 rsvp_input(struct mbuf **mp, int *offp, int proto)
2354 {
2355         struct mbuf *m = *mp;
2356
2357         *mp = NULL;
2358
2359         if (rsvp_input_p) { /* call the real one if loaded */
2360                 *mp = m;
2361                 rsvp_input_p(mp, offp, proto);
2362                 return(IPPROTO_DONE);
2363         }
2364
2365         /* Can still get packets with rsvp_on = 0 if there is a local member
2366          * of the group to which the RSVP packet is addressed.  But in this
2367          * case we want to throw the packet away.
2368          */
2369
2370         if (!rsvp_on) {
2371                 m_freem(m);
2372                 return(IPPROTO_DONE);
2373         }
2374
2375         if (ip_rsvpd != NULL) {
2376                 *mp = m;
2377                 rip_input(mp, offp, proto);
2378                 return(IPPROTO_DONE);
2379         }
2380         /* Drop the packet */
2381         m_freem(m);
2382         return(IPPROTO_DONE);
2383 }