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 nipq = 0;    /* total # of reass queues */
 164 static int maxnipq;
 165 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
 166     &maxnipq, 0,
 167     "Maximum number of IPv4 fragment reassembly queue entries");
 168
 169 static int maxfragsperpacket;
 170 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
 171     &maxfragsperpacket, 0,
 172     "Maximum number of IPv4 fragments allowed per packet");
 173
 174 static int ip_sendsourcequench = 0;
 175 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
 176     &ip_sendsourcequench, 0,
 177     "Enable the transmission of source quench packets");
 178
 179 int ip_do_randomid = 1;
 180 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
 181     &ip_do_randomid, 0,
 182     "Assign random ip_id values");
 183 /*
 184  * XXX - Setting ip_checkinterface mostly implements the receive side of
 185  * the Strong ES model described in RFC 1122, but since the routing table
 186  * and transmit implementation do not implement the Strong ES model,
 187  * setting this to 1 results in an odd hybrid.
 188  *
 189  * XXX - ip_checkinterface currently must be disabled if you use ipnat
 190  * to translate the destination address to another local interface.
 191  *
 192  * XXX - ip_checkinterface must be disabled if you add IP aliases
 193  * to the loopback interface instead of the interface where the
 194  * packets for those addresses are received.
 195  */
 196 static int ip_checkinterface = 0;
 197 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
 198     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
 199
 200 static u_long ip_hash_count = 0;
 201 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, hash_count, CTLFLAG_RD,
 202     &ip_hash_count, 0, "Number of packets hashed by IP");
 203
 204 #ifdef RSS_DEBUG
 205 static u_long ip_rehash_count = 0;
 206 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, rehash_count, CTLFLAG_RD,
 207     &ip_rehash_count, 0, "Number of packets rehashed by IP");
 208
 209 static u_long ip_dispatch_fast = 0;
 210 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_fast_count, CTLFLAG_RD,
 211     &ip_dispatch_fast, 0, "Number of packets handled on current CPU");
 212
 213 static u_long ip_dispatch_slow = 0;
 214 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_slow_count, CTLFLAG_RD,
 215     &ip_dispatch_slow, 0, "Number of packets messaged to another CPU");
 216 #endif
 217
 218 static struct lwkt_token ipq_token = LWKT_TOKEN_INITIALIZER(ipq_token);
 219
 220 #ifdef DIAGNOSTIC
 221 static int ipprintfs = 0;
 222 #endif
 223
 224 extern  struct domain inetdomain;
 225 extern  struct protosw inetsw[];
 226 u_char  ip_protox[IPPROTO_MAX];
 227 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
 228 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
 229                                                 /* inet addr hash table */
 230 u_long  in_ifaddrhmask;                         /* mask for hash table */
 231
 232 static struct mbuf *ipforward_mtemp[MAXCPU];
 233
 234 struct ip_stats ipstats_percpu[MAXCPU] __cachealign;
 235
 236 static int
 237 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
 238 {
 239         int cpu, error = 0;
 240
 241         for (cpu = 0; cpu < ncpus; ++cpu) {
 242                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
 243                                         sizeof(struct ip_stats))))
 244                         break;
 245                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
 246                                        sizeof(struct ip_stats))))
 247                         break;
 248         }
 249
 250         return (error);
 251 }
 252 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
 253     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
 254
 255 /* Packet reassembly stuff */
 256 #define IPREASS_NHASH_LOG2      6
 257 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
 258 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
 259 #define IPREASS_HASH(x,y)                                               \
 260     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
 261
 262 static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
 263
 264 #ifdef IPCTL_DEFMTU
 265 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
 266     &ip_mtu, 0, "Default MTU");
 267 #endif
 268
 269 #ifdef IPSTEALTH
 270 static int ipstealth = 0;
 271 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
 272 #else
 273 static const int ipstealth = 0;
 274 #endif
 275
 276 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
 277
 278 struct pfil_head inet_pfil_hook;
 279
 280 /*
 281  * struct ip_srcrt_opt is used to store packet state while it travels
 282  * through the stack.
 283  *
 284  * XXX Note that the code even makes assumptions on the size and
 285  * alignment of fields inside struct ip_srcrt so e.g. adding some
 286  * fields will break the code.  This needs to be fixed.
 287  *
 288  * We need to save the IP options in case a protocol wants to respond
 289  * to an incoming packet over the same route if the packet got here
 290  * using IP source routing.  This allows connection establishment and
 291  * maintenance when the remote end is on a network that is not known
 292  * to us.
 293  */
 294 struct ip_srcrt {
 295         struct  in_addr dst;                    /* final destination */
 296         char    nop;                            /* one NOP to align */
 297         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
 298         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
 299 };
 300
 301 struct ip_srcrt_opt {
 302         int             ip_nhops;
 303         struct ip_srcrt ip_srcrt;
 304 };
 305
 306 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
 307 static struct malloc_pipe ipq_mpipe;
 308
 309 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
 310 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
 311 static void             ip_freef(struct ipqhead *, struct ipq *);
 312 static void             ip_input_handler(netmsg_t);
 313
 314 /*
 315  * IP initialization: fill in IP protocol switch table.
 316  * All protocols not implemented in kernel go to raw IP protocol handler.
 317  */
 318 void
 319 ip_init(void)
 320 {
 321         struct protosw *pr;
 322         int i;
 323         int cpu;
 324
 325         /*
 326          * Make sure we can handle a reasonable number of fragments but
 327          * cap it at 4000 (XXX).
 328          */
 329         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
 330                     IFQ_MAXLEN, 4000, 0, NULL, NULL, NULL);
 331         for (i = 0; i < ncpus; ++i) {
 332                 TAILQ_INIT(&in_ifaddrheads[i]);
 333                 in_ifaddrhashtbls[i] =
 334                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
 335         }
 336         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
 337         if (pr == NULL)
 338                 panic("ip_init");
 339         for (i = 0; i < IPPROTO_MAX; i++)
 340                 ip_protox[i] = pr - inetsw;
 341         for (pr = inetdomain.dom_protosw;
 342              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
 343                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
 344                         if (pr->pr_protocol != IPPROTO_RAW)
 345                                 ip_protox[pr->pr_protocol] = pr - inetsw;
 346                 }
 347         }
 348
 349         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
 350         inet_pfil_hook.ph_af = AF_INET;
 351         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
 352                 kprintf("%s: WARNING: unable to register pfil hook, "
 353                         "error %d\n", __func__, i);
 354         }
 355
 356         for (i = 0; i < IPREASS_NHASH; i++)
 357                 TAILQ_INIT(&ipq[i]);
 358
 359         maxnipq = nmbclusters / 32;
 360         maxfragsperpacket = 16;
 361
 362         ip_id = time_second & 0xffff;   /* time_second survives reboots */
 363
 364         for (cpu = 0; cpu < ncpus; ++cpu) {
 365                 /*
 366                  * Initialize IP statistics counters for each CPU.
 367                  */
 368                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
 369
 370                 /*
 371                  * Preallocate mbuf template for forwarding
 372                  */
 373                 MGETHDR(ipforward_mtemp[cpu], MB_WAIT, MT_DATA);
 374         }
 375
 376         netisr_register(NETISR_IP, ip_input_handler, ip_hashfn_in);
 377         netisr_register_hashcheck(NETISR_IP, ip_hashcheck);
 378 }
 379
 380 /* Do transport protocol processing. */
 381 static void
 382 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
 383 {
 384         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
 385
 386         /*
 387          * Switch out to protocol's input routine.
 388          */
 389         PR_GET_MPLOCK(pr);
 390         pr->pr_input(&m, &hlen, ip->ip_p);
 391         PR_REL_MPLOCK(pr);
 392 }
 393
 394 static void
 395 transport_processing_handler(netmsg_t msg)
 396 {
 397         struct netmsg_packet *pmsg = &msg->packet;
 398         struct ip *ip;
 399         int hlen;
 400
 401         ip = mtod(pmsg->nm_packet, struct ip *);
 402         hlen = pmsg->base.lmsg.u.ms_result;
 403
 404         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
 405         /* msg was embedded in the mbuf, do not reply! */
 406 }
 407
 408 static void
 409 ip_input_handler(netmsg_t msg)
 410 {
 411         ip_input(msg->packet.nm_packet);
 412         /* msg was embedded in the mbuf, do not reply! */
 413 }
 414
 415 /*
 416  * IP input routine.  Checksum and byte swap header.  If fragmented
 417  * try to reassemble.  Process options.  Pass to next level.
 418  */
 419 void
 420 ip_input(struct mbuf *m)
 421 {
 422         struct ip *ip;
 423         struct in_ifaddr *ia = NULL;
 424         struct in_ifaddr_container *iac;
 425         int hlen, checkif;
 426         u_short sum;
 427         struct in_addr pkt_dst;
 428         boolean_t check_msgport = FALSE;
 429         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
 430         struct in_addr odst;                    /* original dst address(NAT) */
 431         struct m_tag *mtag;
 432         struct sockaddr_in *next_hop = NULL;
 433         lwkt_port_t port;
 434 #ifdef FAST_IPSEC
 435         struct tdb_ident *tdbi;
 436         struct secpolicy *sp;
 437         int error;
 438 #endif
 439
 440         M_ASSERTPKTHDR(m);
 441
 442         /*
 443          * This routine is called from numerous places which may not have
 444          * characterized the packet.
 445          */
 446         if ((m->m_flags & M_HASH) == 0) {
 447                 atomic_add_long(&ip_hash_count, 1);
 448                 ip_hashfn(&m, 0, IP_MPORT_IN);
 449                 if (m == NULL)
 450                         return;
 451                 KKASSERT(m->m_flags & M_HASH);
 452                 check_msgport = TRUE;
 453         }
 454         ip = mtod(m, struct ip *);
 455
 456         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 457                 /*
 458                  * XXX handle multicast on CPU0 for now.
 459                  *
 460                  * This could happen for packets hashed by hardware
 461                  * using RSS, which does not differentiate multicast
 462                  * packets from unicast packets.
 463                  */
 464                 m->m_pkthdr.hash = 0;
 465                 check_msgport = TRUE;
 466         }
 467
 468         if (check_msgport &&
 469             &curthread->td_msgport != netisr_hashport(m->m_pkthdr.hash)) {
 470                 netisr_queue(NETISR_IP, m);
 471                 /* Requeued to other netisr msgport; done */
 472                 return;
 473         }
 474
 475         /*
 476          * Pull out certain tags
 477          */
 478         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 479                 /* Next hop */
 480                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 481                 KKASSERT(mtag != NULL);
 482                 next_hop = m_tag_data(mtag);
 483         }
 484
 485         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 486                 /* dummynet already filtered us */
 487                 ip = mtod(m, struct ip *);
 488                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 489                 goto iphack;
 490         }
 491
 492         ipstat.ips_total++;
 493
 494         /* length checks already done in ip_hashfn() */
 495         KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
 496
 497         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
 498                 ipstat.ips_badvers++;
 499                 goto bad;
 500         }
 501
 502         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 503         /* length checks already done in ip_hashfn() */
 504         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
 505         KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
 506
 507         /* 127/8 must not appear on wire - RFC1122 */
 508         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 509             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 510                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
 511                         ipstat.ips_badaddr++;
 512                         goto bad;
 513                 }
 514         }
 515
 516         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
 517                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
 518         } else {
 519                 if (hlen == sizeof(struct ip))
 520                         sum = in_cksum_hdr(ip);
 521                 else
 522                         sum = in_cksum(m, hlen);
 523         }
 524         if (sum != 0) {
 525                 ipstat.ips_badsum++;
 526                 goto bad;
 527         }
 528
 529 #ifdef ALTQ
 530         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
 531                 /* packet is dropped by traffic conditioner */
 532                 return;
 533         }
 534 #endif
 535         /*
 536          * Convert fields to host representation.
 537          */
 538         ip->ip_len = ntohs(ip->ip_len);
 539         ip->ip_off = ntohs(ip->ip_off);
 540
 541         /* length checks already done in ip_hashfn() */
 542         KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
 543         KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
 544
 545         /*
 546          * Trim mbufs if longer than the IP header would have us expect.
 547          */
 548         if (m->m_pkthdr.len > ip->ip_len) {
 549                 if (m->m_len == m->m_pkthdr.len) {
 550                         m->m_len = ip->ip_len;
 551                         m->m_pkthdr.len = ip->ip_len;
 552                 } else {
 553                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
 554                 }
 555         }
 556 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
 557         /*
 558          * Bypass packet filtering for packets from a tunnel (gif).
 559          */
 560         if (ipsec_gethist(m, NULL))
 561                 goto pass;
 562 #endif
 563
 564         /*
 565          * IpHack's section.
 566          * Right now when no processing on packet has done
 567          * and it is still fresh out of network we do our black
 568          * deals with it.
 569          * - Firewall: deny/allow/divert
 570          * - Xlate: translate packet's addr/port (NAT).
 571          * - Pipe: pass pkt through dummynet.
 572          * - Wrap: fake packet's addr/port <unimpl.>
 573          * - Encapsulate: put it in another IP and send out. <unimp.>
 574          */
 575
 576 iphack:
 577         /*
 578          * If we've been forwarded from the output side, then
 579          * skip the firewall a second time
 580          */
 581         if (next_hop != NULL)
 582                 goto ours;
 583
 584         /* No pfil hooks */
 585         if (!pfil_has_hooks(&inet_pfil_hook)) {
 586                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 587                         /*
 588                          * Strip dummynet tags from stranded packets
 589                          */
 590                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 591                         KKASSERT(mtag != NULL);
 592                         m_tag_delete(m, mtag);
 593                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
 594                 }
 595                 goto pass;
 596         }
 597
 598         /*
 599          * Run through list of hooks for input packets.
 600          *
 601          * NOTE!  If the packet is rewritten pf/ipfw/whoever must
 602          *        clear M_HASH.
 603          */
 604         odst = ip->ip_dst;
 605         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
 606                 return;
 607         if (m == NULL)  /* consumed by filter */
 608                 return;
 609         ip = mtod(m, struct ip *);
 610         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 611         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
 612
 613         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
 614                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 615                 KKASSERT(mtag != NULL);
 616                 next_hop = m_tag_data(mtag);
 617         }
 618         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
 619                 ip_dn_queue(m);
 620                 return;
 621         }
 622         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
 623                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
 624         }
 625 pass:
 626         /*
 627          * Process options and, if not destined for us,
 628          * ship it on.  ip_dooptions returns 1 when an
 629          * error was detected (causing an icmp message
 630          * to be sent and the original packet to be freed).
 631          */
 632         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
 633                 return;
 634
 635         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
 636          * matter if it is destined to another node, or whether it is
 637          * a multicast one, RSVP wants it! and prevents it from being forwarded
 638          * anywhere else. Also checks if the rsvp daemon is running before
 639          * grabbing the packet.
 640          */
 641         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
 642                 goto ours;
 643
 644         /*
 645          * Check our list of addresses, to see if the packet is for us.
 646          * If we don't have any addresses, assume any unicast packet
 647          * we receive might be for us (and let the upper layers deal
 648          * with it).
 649          */
 650         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
 651             !(m->m_flags & (M_MCAST | M_BCAST)))
 652                 goto ours;
 653
 654         /*
 655          * Cache the destination address of the packet; this may be
 656          * changed by use of 'ipfw fwd'.
 657          */
 658         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
 659
 660         /*
 661          * Enable a consistency check between the destination address
 662          * and the arrival interface for a unicast packet (the RFC 1122
 663          * strong ES model) if IP forwarding is disabled and the packet
 664          * is not locally generated and the packet is not subject to
 665          * 'ipfw fwd'.
 666          *
 667          * XXX - Checking also should be disabled if the destination
 668          * address is ipnat'ed to a different interface.
 669          *
 670          * XXX - Checking is incompatible with IP aliases added
 671          * to the loopback interface instead of the interface where
 672          * the packets are received.
 673          */
 674         checkif = ip_checkinterface &&
 675                   !ipforwarding &&
 676                   m->m_pkthdr.rcvif != NULL &&
 677                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
 678                   next_hop == NULL;
 679
 680         /*
 681          * Check for exact addresses in the hash bucket.
 682          */
 683         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
 684                 ia = iac->ia;
 685
 686                 /*
 687                  * If the address matches, verify that the packet
 688                  * arrived via the correct interface if checking is
 689                  * enabled.
 690                  */
 691                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
 692                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
 693                         goto ours;
 694         }
 695         ia = NULL;
 696
 697         /*
 698          * Check for broadcast addresses.
 699          *
 700          * Only accept broadcast packets that arrive via the matching
 701          * interface.  Reception of forwarded directed broadcasts would
 702          * be handled via ip_forward() and ether_output() with the loopback
 703          * into the stack for SIMPLEX interfaces handled by ether_output().
 704          */
 705         if (m->m_pkthdr.rcvif != NULL &&
 706             m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
 707                 struct ifaddr_container *ifac;
 708
 709                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
 710                               ifa_link) {
 711                         struct ifaddr *ifa = ifac->ifa;
 712
 713                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
 714                                 continue;
 715                         if (ifa->ifa_addr->sa_family != AF_INET)
 716                                 continue;
 717                         ia = ifatoia(ifa);
 718                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 719                                                                 pkt_dst.s_addr)
 720                                 goto ours;
 721                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
 722                                 goto ours;
 723 #ifdef BOOTP_COMPAT
 724                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
 725                                 goto ours;
 726 #endif
 727                 }
 728         }
 729         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 730                 struct in_multi *inm;
 731
 732                 /* XXX Multicast is not MPSAFE yet */
 733                 get_mplock();
 734
 735                 if (ip_mrouter != NULL) {
 736                         /*
 737                          * If we are acting as a multicast router, all
 738                          * incoming multicast packets are passed to the
 739                          * kernel-level multicast forwarding function.
 740                          * The packet is returned (relatively) intact; if
 741                          * ip_mforward() returns a non-zero value, the packet
 742                          * must be discarded, else it may be accepted below.
 743                          */
 744                         if (ip_mforward != NULL &&
 745                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
 746                                 rel_mplock();
 747                                 ipstat.ips_cantforward++;
 748                                 m_freem(m);
 749                                 return;
 750                         }
 751
 752                         /*
 753                          * The process-level routing daemon needs to receive
 754                          * all multicast IGMP packets, whether or not this
 755                          * host belongs to their destination groups.
 756                          */
 757                         if (ip->ip_p == IPPROTO_IGMP) {
 758                                 rel_mplock();
 759                                 goto ours;
 760                         }
 761                         ipstat.ips_forward++;
 762                 }
 763                 /*
 764                  * See if we belong to the destination multicast group on the
 765                  * arrival interface.
 766                  */
 767                 inm = IN_LOOKUP_MULTI(&ip->ip_dst, m->m_pkthdr.rcvif);
 768                 if (inm == NULL) {
 769                         rel_mplock();
 770                         ipstat.ips_notmember++;
 771                         m_freem(m);
 772                         return;
 773                 }
 774
 775                 rel_mplock();
 776                 goto ours;
 777         }
 778         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 779                 goto ours;
 780         if (ip->ip_dst.s_addr == INADDR_ANY)
 781                 goto ours;
 782
 783         /*
 784          * FAITH(Firewall Aided Internet Translator)
 785          */
 786         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 787                 if (ip_keepfaith) {
 788                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 789                                 goto ours;
 790                 }
 791                 m_freem(m);
 792                 return;
 793         }
 794
 795         /*
 796          * Not for us; forward if possible and desirable.
 797          */
 798         if (!ipforwarding) {
 799                 ipstat.ips_cantforward++;
 800                 m_freem(m);
 801         } else {
 802 #ifdef IPSEC
 803                 /*
 804                  * Enforce inbound IPsec SPD.
 805                  */
 806                 if (ipsec4_in_reject(m, NULL)) {
 807                         ipsecstat.in_polvio++;
 808                         goto bad;
 809                 }
 810 #endif
 811 #ifdef FAST_IPSEC
 812                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 813                 crit_enter();
 814                 if (mtag != NULL) {
 815                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 816                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 817                 } else {
 818                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 819                                                    IP_FORWARDING, &error);
 820                 }
 821                 if (sp == NULL) {       /* NB: can happen if error */
 822                         crit_exit();
 823                         /*XXX error stat???*/
 824                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 825                         goto bad;
 826                 }
 827
 828                 /*
 829                  * Check security policy against packet attributes.
 830                  */
 831                 error = ipsec_in_reject(sp, m);
 832                 KEY_FREESP(&sp);
 833                 crit_exit();
 834                 if (error) {
 835                         ipstat.ips_cantforward++;
 836                         goto bad;
 837                 }
 838 #endif
 839                 ip_forward(m, using_srcrt, next_hop);
 840         }
 841         return;
 842
 843 ours:
 844
 845         /*
 846          * IPSTEALTH: Process non-routing options only
 847          * if the packet is destined for us.
 848          */
 849         if (ipstealth &&
 850             hlen > sizeof(struct ip) &&
 851             ip_dooptions(m, 1, next_hop))
 852                 return;
 853
 854         /* Count the packet in the ip address stats */
 855         if (ia != NULL) {
 856                 IFA_STAT_INC(&ia->ia_ifa, ipackets, 1);
 857                 IFA_STAT_INC(&ia->ia_ifa, ibytes, m->m_pkthdr.len);
 858         }
 859
 860         /*
 861          * If offset or IP_MF are set, must reassemble.
 862          * Otherwise, nothing need be done.
 863          * (We could look in the reassembly queue to see
 864          * if the packet was previously fragmented,
 865          * but it's not worth the time; just let them time out.)
 866          */
 867         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 868                 /*
 869                  * Attempt reassembly; if it succeeds, proceed.  ip_reass()
 870                  * will return a different mbuf.
 871                  *
 872                  * NOTE: ip_reass() returns m with M_HASH cleared to force
 873                  *       us to recharacterize the packet.
 874                  */
 875                 m = ip_reass(m);
 876                 if (m == NULL)
 877                         return;
 878                 ip = mtod(m, struct ip *);
 879
 880                 /* Get the header length of the reassembled packet */
 881                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 882         } else {
 883                 ip->ip_len -= hlen;
 884         }
 885
 886 #ifdef IPSEC
 887         /*
 888          * enforce IPsec policy checking if we are seeing last header.
 889          * note that we do not visit this with protocols with pcb layer
 890          * code - like udp/tcp/raw ip.
 891          */
 892         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 893             ipsec4_in_reject(m, NULL)) {
 894                 ipsecstat.in_polvio++;
 895                 goto bad;
 896         }
 897 #endif
 898 #if FAST_IPSEC
 899         /*
 900          * enforce IPsec policy checking if we are seeing last header.
 901          * note that we do not visit this with protocols with pcb layer
 902          * code - like udp/tcp/raw ip.
 903          */
 904         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 905                 /*
 906                  * Check if the packet has already had IPsec processing
 907                  * done.  If so, then just pass it along.  This tag gets
 908                  * set during AH, ESP, etc. input handling, before the
 909                  * packet is returned to the ip input queue for delivery.
 910                  */
 911                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 912                 crit_enter();
 913                 if (mtag != NULL) {
 914                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 915                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 916                 } else {
 917                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 918                                                    IP_FORWARDING, &error);
 919                 }
 920                 if (sp != NULL) {
 921                         /*
 922                          * Check security policy against packet attributes.
 923                          */
 924                         error = ipsec_in_reject(sp, m);
 925                         KEY_FREESP(&sp);
 926                 } else {
 927                         /* XXX error stat??? */
 928                         error = EINVAL;
 929 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 930                         goto bad;
 931                 }
 932                 crit_exit();
 933                 if (error)
 934                         goto bad;
 935         }
 936 #endif /* FAST_IPSEC */
 937
 938         /*
 939          * We must forward the packet to the correct protocol thread if
 940          * we are not already in it.
 941          *
 942          * NOTE: ip_len is now in host form.  ip_len is not adjusted
 943          *       further for protocol processing, instead we pass hlen
 944          *       to the protosw and let it deal with it.
 945          */
 946         ipstat.ips_delivered++;
 947
 948         if ((m->m_flags & M_HASH) == 0) {
 949 #ifdef RSS_DEBUG
 950                 atomic_add_long(&ip_rehash_count, 1);
 951 #endif
 952                 ip->ip_len = htons(ip->ip_len + hlen);
 953                 ip->ip_off = htons(ip->ip_off);
 954
 955                 ip_hashfn(&m, 0, IP_MPORT_IN);
 956                 if (m == NULL)
 957                         return;
 958
 959                 ip = mtod(m, struct ip *);
 960                 ip->ip_len = ntohs(ip->ip_len) - hlen;
 961                 ip->ip_off = ntohs(ip->ip_off);
 962                 KKASSERT(m->m_flags & M_HASH);
 963         }
 964         port = netisr_hashport(m->m_pkthdr.hash);
 965
 966         if (port != &curthread->td_msgport) {
 967                 struct netmsg_packet *pmsg;
 968
 969 #ifdef RSS_DEBUG
 970                 atomic_add_long(&ip_dispatch_slow, 1);
 971 #endif
 972
 973                 pmsg = &m->m_hdr.mh_netmsg;
 974                 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
 975                             0, transport_processing_handler);
 976                 pmsg->nm_packet = m;
 977                 pmsg->base.lmsg.u.ms_result = hlen;
 978                 lwkt_sendmsg(port, &pmsg->base.lmsg);
 979         } else {
 980 #ifdef RSS_DEBUG
 981                 atomic_add_long(&ip_dispatch_fast, 1);
 982 #endif
 983                 transport_processing_oncpu(m, hlen, ip);
 984         }
 985         return;
 986
 987 bad:
 988         m_freem(m);
 989 }
 990
 991 /*
 992  * Take incoming datagram fragment and try to reassemble it into
 993  * whole datagram.  If a chain for reassembly of this datagram already
 994  * exists, then it is given as fp; otherwise have to make a chain.
 995  */
 996 struct mbuf *
 997 ip_reass(struct mbuf *m)
 998 {
 999         struct ip *ip = mtod(m, struct ip *);
1000         struct mbuf *p = NULL, *q, *nq;
1001         struct mbuf *n;
1002         struct ipq *fp = NULL;
1003         struct ipqhead *head;
1004         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1005         int i, next;
1006         u_short sum;
1007
1008         /* If maxnipq is 0, never accept fragments. */
1009         if (maxnipq == 0) {
1010                 ipstat.ips_fragments++;
1011                 ipstat.ips_fragdropped++;
1012                 m_freem(m);
1013                 return NULL;
1014         }
1015
1016         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1017         /*
1018          * Look for queue of fragments of this datagram.
1019          */
1020         lwkt_gettoken(&ipq_token);
1021         head = &ipq[sum];
1022         TAILQ_FOREACH(fp, head, ipq_list) {
1023                 if (ip->ip_id == fp->ipq_id &&
1024                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1025                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1026                     ip->ip_p == fp->ipq_p)
1027                         goto found;
1028         }
1029
1030         fp = NULL;
1031
1032         /*
1033          * Enforce upper bound on number of fragmented packets
1034          * for which we attempt reassembly;
1035          * If maxnipq is -1, accept all fragments without limitation.
1036          */
1037         if (nipq > maxnipq && maxnipq > 0) {
1038                 /*
1039                  * drop something from the tail of the current queue
1040                  * before proceeding further
1041                  */
1042                 struct ipq *q = TAILQ_LAST(head, ipqhead);
1043                 if (q == NULL) {
1044                         /*
1045                          * The current queue is empty,
1046                          * so drop from one of the others.
1047                          */
1048                         for (i = 0; i < IPREASS_NHASH; i++) {
1049                                 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
1050                                 if (r) {
1051                                         ipstat.ips_fragtimeout += r->ipq_nfrags;
1052                                         ip_freef(&ipq[i], r);
1053                                         break;
1054                                 }
1055                         }
1056                 } else {
1057                         ipstat.ips_fragtimeout += q->ipq_nfrags;
1058                         ip_freef(head, q);
1059                 }
1060         }
1061 found:
1062         /*
1063          * Adjust ip_len to not reflect header,
1064          * convert offset of this to bytes.
1065          */
1066         ip->ip_len -= hlen;
1067         if (ip->ip_off & IP_MF) {
1068                 /*
1069                  * Make sure that fragments have a data length
1070                  * that's a non-zero multiple of 8 bytes.
1071                  */
1072                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1073                         ipstat.ips_toosmall++; /* XXX */
1074                         m_freem(m);
1075                         goto done;
1076                 }
1077                 m->m_flags |= M_FRAG;
1078         } else {
1079                 m->m_flags &= ~M_FRAG;
1080         }
1081         ip->ip_off <<= 3;
1082
1083         ipstat.ips_fragments++;
1084         m->m_pkthdr.header = ip;
1085
1086         /*
1087          * If the hardware has not done csum over this fragment
1088          * then csum_data is not valid at all.
1089          */
1090         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1091             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1092                 m->m_pkthdr.csum_data = 0;
1093                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1094         }
1095
1096         /*
1097          * Presence of header sizes in mbufs
1098          * would confuse code below.
1099          */
1100         m->m_data += hlen;
1101         m->m_len -= hlen;
1102
1103         /*
1104          * If first fragment to arrive, create a reassembly queue.
1105          */
1106         if (fp == NULL) {
1107                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1108                         goto dropfrag;
1109                 TAILQ_INSERT_HEAD(head, fp, ipq_list);
1110                 nipq++;
1111                 fp->ipq_nfrags = 1;
1112                 fp->ipq_ttl = IPFRAGTTL;
1113                 fp->ipq_p = ip->ip_p;
1114                 fp->ipq_id = ip->ip_id;
1115                 fp->ipq_src = ip->ip_src;
1116                 fp->ipq_dst = ip->ip_dst;
1117                 fp->ipq_frags = m;
1118                 m->m_nextpkt = NULL;
1119                 goto inserted;
1120         } else {
1121                 fp->ipq_nfrags++;
1122         }
1123
1124 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1125
1126         /*
1127          * Find a segment which begins after this one does.
1128          */
1129         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1130                 if (GETIP(q)->ip_off > ip->ip_off)
1131                         break;
1132         }
1133
1134         /*
1135          * If there is a preceding segment, it may provide some of
1136          * our data already.  If so, drop the data from the incoming
1137          * segment.  If it provides all of our data, drop us, otherwise
1138          * stick new segment in the proper place.
1139          *
1140          * If some of the data is dropped from the the preceding
1141          * segment, then it's checksum is invalidated.
1142          */
1143         if (p) {
1144                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1145                 if (i > 0) {
1146                         if (i >= ip->ip_len)
1147                                 goto dropfrag;
1148                         m_adj(m, i);
1149                         m->m_pkthdr.csum_flags = 0;
1150                         ip->ip_off += i;
1151                         ip->ip_len -= i;
1152                 }
1153                 m->m_nextpkt = p->m_nextpkt;
1154                 p->m_nextpkt = m;
1155         } else {
1156                 m->m_nextpkt = fp->ipq_frags;
1157                 fp->ipq_frags = m;
1158         }
1159
1160         /*
1161          * While we overlap succeeding segments trim them or,
1162          * if they are completely covered, dequeue them.
1163          */
1164         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1165              q = nq) {
1166                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1167                 if (i < GETIP(q)->ip_len) {
1168                         GETIP(q)->ip_len -= i;
1169                         GETIP(q)->ip_off += i;
1170                         m_adj(q, i);
1171                         q->m_pkthdr.csum_flags = 0;
1172                         break;
1173                 }
1174                 nq = q->m_nextpkt;
1175                 m->m_nextpkt = nq;
1176                 ipstat.ips_fragdropped++;
1177                 fp->ipq_nfrags--;
1178                 q->m_nextpkt = NULL;
1179                 m_freem(q);
1180         }
1181
1182 inserted:
1183         /*
1184          * Check for complete reassembly and perform frag per packet
1185          * limiting.
1186          *
1187          * Frag limiting is performed here so that the nth frag has
1188          * a chance to complete the packet before we drop the packet.
1189          * As a result, n+1 frags are actually allowed per packet, but
1190          * only n will ever be stored. (n = maxfragsperpacket.)
1191          *
1192          */
1193         next = 0;
1194         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1195                 if (GETIP(q)->ip_off != next) {
1196                         if (fp->ipq_nfrags > maxfragsperpacket) {
1197                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1198                                 ip_freef(head, fp);
1199                         }
1200                         goto done;
1201                 }
1202                 next += GETIP(q)->ip_len;
1203         }
1204         /* Make sure the last packet didn't have the IP_MF flag */
1205         if (p->m_flags & M_FRAG) {
1206                 if (fp->ipq_nfrags > maxfragsperpacket) {
1207                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1208                         ip_freef(head, fp);
1209                 }
1210                 goto done;
1211         }
1212
1213         /*
1214          * Reassembly is complete.  Make sure the packet is a sane size.
1215          */
1216         q = fp->ipq_frags;
1217         ip = GETIP(q);
1218         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1219                 ipstat.ips_toolong++;
1220                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1221                 ip_freef(head, fp);
1222                 goto done;
1223         }
1224
1225         /*
1226          * Concatenate fragments.
1227          */
1228         m = q;
1229         n = m->m_next;
1230         m->m_next = NULL;
1231         m_cat(m, n);
1232         nq = q->m_nextpkt;
1233         q->m_nextpkt = NULL;
1234         for (q = nq; q != NULL; q = nq) {
1235                 nq = q->m_nextpkt;
1236                 q->m_nextpkt = NULL;
1237                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1238                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1239                 m_cat(m, q);
1240         }
1241
1242         /*
1243          * Clean up the 1's complement checksum.  Carry over 16 bits must
1244          * be added back.  This assumes no more then 65535 packet fragments
1245          * were reassembled.  A second carry can also occur (but not a third).
1246          */
1247         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1248                                 (m->m_pkthdr.csum_data >> 16);
1249         if (m->m_pkthdr.csum_data > 0xFFFF)
1250                 m->m_pkthdr.csum_data -= 0xFFFF;
1251
1252         /*
1253          * Create header for new ip packet by
1254          * modifying header of first packet;
1255          * dequeue and discard fragment reassembly header.
1256          * Make header visible.
1257          */
1258         ip->ip_len = next;
1259         ip->ip_src = fp->ipq_src;
1260         ip->ip_dst = fp->ipq_dst;
1261         TAILQ_REMOVE(head, fp, ipq_list);
1262         nipq--;
1263         mpipe_free(&ipq_mpipe, fp);
1264         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1265         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1266         /* some debugging cruft by sklower, below, will go away soon */
1267         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1268                 int plen = 0;
1269
1270                 for (n = m; n; n = n->m_next)
1271                         plen += n->m_len;
1272                 m->m_pkthdr.len = plen;
1273         }
1274
1275         /*
1276          * Reassembly complete, return the next protocol.
1277          *
1278          * Be sure to clear M_HASH to force the packet
1279          * to be re-characterized.
1280          *
1281          * Clear M_FRAG, we are no longer a fragment.
1282          */
1283         m->m_flags &= ~(M_HASH | M_FRAG);
1284
1285         ipstat.ips_reassembled++;
1286         lwkt_reltoken(&ipq_token);
1287         return (m);
1288
1289 dropfrag:
1290         ipstat.ips_fragdropped++;
1291         if (fp != NULL)
1292                 fp->ipq_nfrags--;
1293         m_freem(m);
1294 done:
1295         lwkt_reltoken(&ipq_token);
1296         return (NULL);
1297
1298 #undef GETIP
1299 }
1300
1301 /*
1302  * Free a fragment reassembly header and all
1303  * associated datagrams.
1304  *
1305  * Called with ipq_token held.
1306  */
1307 static void
1308 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1309 {
1310         struct mbuf *q;
1311
1312         /*
1313          * Remove first to protect against blocking
1314          */
1315         TAILQ_REMOVE(fhp, fp, ipq_list);
1316
1317         /*
1318          * Clean out at our leisure
1319          */
1320         while (fp->ipq_frags) {
1321                 q = fp->ipq_frags;
1322                 fp->ipq_frags = q->m_nextpkt;
1323                 q->m_nextpkt = NULL;
1324                 m_freem(q);
1325         }
1326         mpipe_free(&ipq_mpipe, fp);
1327         nipq--;
1328 }
1329
1330 /*
1331  * IP timer processing;
1332  * if a timer expires on a reassembly
1333  * queue, discard it.
1334  */
1335 void
1336 ip_slowtimo(void)
1337 {
1338         struct ipq *fp, *fp_temp;
1339         struct ipqhead *head;
1340         int i;
1341
1342         lwkt_gettoken(&ipq_token);
1343         for (i = 0; i < IPREASS_NHASH; i++) {
1344                 head = &ipq[i];
1345                 TAILQ_FOREACH_MUTABLE(fp, head, ipq_list, fp_temp) {
1346                         if (--fp->ipq_ttl == 0) {
1347                                 ipstat.ips_fragtimeout += fp->ipq_nfrags;
1348                                 ip_freef(head, fp);
1349                         }
1350                 }
1351         }
1352         /*
1353          * If we are over the maximum number of fragments
1354          * (due to the limit being lowered), drain off
1355          * enough to get down to the new limit.
1356          */
1357         if (maxnipq >= 0 && nipq > maxnipq) {
1358                 for (i = 0; i < IPREASS_NHASH; i++) {
1359                         head = &ipq[i];
1360                         while (nipq > maxnipq && !TAILQ_EMPTY(head)) {
1361                                 ipstat.ips_fragdropped +=
1362                                     TAILQ_FIRST(head)->ipq_nfrags;
1363                                 ip_freef(head, TAILQ_FIRST(head));
1364                         }
1365                 }
1366         }
1367         lwkt_reltoken(&ipq_token);
1368         ipflow_slowtimo();
1369 }
1370
1371 /*
1372  * Drain off all datagram fragments.
1373  */
1374 void
1375 ip_drain(void)
1376 {
1377         struct ipqhead *head;
1378         int i;
1379
1380         lwkt_gettoken(&ipq_token);
1381         for (i = 0; i < IPREASS_NHASH; i++) {
1382                 head = &ipq[i];
1383                 while (!TAILQ_EMPTY(head)) {
1384                         ipstat.ips_fragdropped += TAILQ_FIRST(head)->ipq_nfrags;
1385                         ip_freef(head, TAILQ_FIRST(head));
1386                 }
1387         }
1388         lwkt_reltoken(&ipq_token);
1389         in_rtqdrain();
1390 }
1391
1392 /*
1393  * Do option processing on a datagram,
1394  * possibly discarding it if bad options are encountered,
1395  * or forwarding it if source-routed.
1396  * The pass argument is used when operating in the IPSTEALTH
1397  * mode to tell what options to process:
1398  * [LS]SRR (pass 0) or the others (pass 1).
1399  * The reason for as many as two passes is that when doing IPSTEALTH,
1400  * non-routing options should be processed only if the packet is for us.
1401  * Returns 1 if packet has been forwarded/freed,
1402  * 0 if the packet should be processed further.
1403  */
1404 static int
1405 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1406 {
1407         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1408         struct ip *ip = mtod(m, struct ip *);
1409         u_char *cp;
1410         struct in_ifaddr *ia;
1411         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1412         boolean_t forward = FALSE;
1413         struct in_addr *sin, dst;
1414         n_time ntime;
1415
1416         dst = ip->ip_dst;
1417         cp = (u_char *)(ip + 1);
1418         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1419         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1420                 opt = cp[IPOPT_OPTVAL];
1421                 if (opt == IPOPT_EOL)
1422                         break;
1423                 if (opt == IPOPT_NOP)
1424                         optlen = 1;
1425                 else {
1426                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1427                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1428                                 goto bad;
1429                         }
1430                         optlen = cp[IPOPT_OLEN];
1431                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1432                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1433                                 goto bad;
1434                         }
1435                 }
1436                 switch (opt) {
1437
1438                 default:
1439                         break;
1440
1441                 /*
1442                  * Source routing with record.
1443                  * Find interface with current destination address.
1444                  * If none on this machine then drop if strictly routed,
1445                  * or do nothing if loosely routed.
1446                  * Record interface address and bring up next address
1447                  * component.  If strictly routed make sure next
1448                  * address is on directly accessible net.
1449                  */
1450                 case IPOPT_LSRR:
1451                 case IPOPT_SSRR:
1452                         if (ipstealth && pass > 0)
1453                                 break;
1454                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1455                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1456                                 goto bad;
1457                         }
1458                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1459                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1460                                 goto bad;
1461                         }
1462                         ipaddr.sin_addr = ip->ip_dst;
1463                         ia = (struct in_ifaddr *)
1464                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1465                         if (ia == NULL) {
1466                                 if (opt == IPOPT_SSRR) {
1467                                         type = ICMP_UNREACH;
1468                                         code = ICMP_UNREACH_SRCFAIL;
1469                                         goto bad;
1470                                 }
1471                                 if (!ip_dosourceroute)
1472                                         goto nosourcerouting;
1473                                 /*
1474                                  * Loose routing, and not at next destination
1475                                  * yet; nothing to do except forward.
1476                                  */
1477                                 break;
1478                         }
1479                         off--;                  /* 0 origin */
1480                         if (off > optlen - (int)sizeof(struct in_addr)) {
1481                                 /*
1482                                  * End of source route.  Should be for us.
1483                                  */
1484                                 if (!ip_acceptsourceroute)
1485                                         goto nosourcerouting;
1486                                 save_rte(m, cp, ip->ip_src);
1487                                 break;
1488                         }
1489                         if (ipstealth)
1490                                 goto dropit;
1491                         if (!ip_dosourceroute) {
1492                                 if (ipforwarding) {
1493                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1494
1495                                         /*
1496                                          * Acting as a router, so generate ICMP
1497                                          */
1498 nosourcerouting:
1499                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1500                                         log(LOG_WARNING,
1501                                             "attempted source route from %s to %s\n",
1502                                             inet_ntoa(ip->ip_src), buf);
1503                                         type = ICMP_UNREACH;
1504                                         code = ICMP_UNREACH_SRCFAIL;
1505                                         goto bad;
1506                                 } else {
1507                                         /*
1508                                          * Not acting as a router,
1509                                          * so silently drop.
1510                                          */
1511 dropit:
1512                                         ipstat.ips_cantforward++;
1513                                         m_freem(m);
1514                                         return (1);
1515                                 }
1516                         }
1517
1518                         /*
1519                          * locate outgoing interface
1520                          */
1521                         memcpy(&ipaddr.sin_addr, cp + off,
1522                             sizeof ipaddr.sin_addr);
1523
1524                         if (opt == IPOPT_SSRR) {
1525 #define INA     struct in_ifaddr *
1526 #define SA      struct sockaddr *
1527                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1528                                                                         == NULL)
1529                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1530                         } else {
1531                                 ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1532                         }
1533                         if (ia == NULL) {
1534                                 type = ICMP_UNREACH;
1535                                 code = ICMP_UNREACH_SRCFAIL;
1536                                 goto bad;
1537                         }
1538                         ip->ip_dst = ipaddr.sin_addr;
1539                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1540                             sizeof(struct in_addr));
1541                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1542                         /*
1543                          * Let ip_intr's mcast routing check handle mcast pkts
1544                          */
1545                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1546                         break;
1547
1548                 case IPOPT_RR:
1549                         if (ipstealth && pass == 0)
1550                                 break;
1551                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1552                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1553                                 goto bad;
1554                         }
1555                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1556                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1557                                 goto bad;
1558                         }
1559                         /*
1560                          * If no space remains, ignore.
1561                          */
1562                         off--;                  /* 0 origin */
1563                         if (off > optlen - (int)sizeof(struct in_addr))
1564                                 break;
1565                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1566                             sizeof ipaddr.sin_addr);
1567                         /*
1568                          * locate outgoing interface; if we're the destination,
1569                          * use the incoming interface (should be same).
1570                          */
1571                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1572                             (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1573                                 type = ICMP_UNREACH;
1574                                 code = ICMP_UNREACH_HOST;
1575                                 goto bad;
1576                         }
1577                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1578                             sizeof(struct in_addr));
1579                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1580                         break;
1581
1582                 case IPOPT_TS:
1583                         if (ipstealth && pass == 0)
1584                                 break;
1585                         code = cp - (u_char *)ip;
1586                         if (optlen < 4 || optlen > 40) {
1587                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1588                                 goto bad;
1589                         }
1590                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1591                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1592                                 goto bad;
1593                         }
1594                         if (off > optlen - (int)sizeof(int32_t)) {
1595                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1596                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1597                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1598                                         goto bad;
1599                                 }
1600                                 break;
1601                         }
1602                         off--;                          /* 0 origin */
1603                         sin = (struct in_addr *)(cp + off);
1604                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1605
1606                         case IPOPT_TS_TSONLY:
1607                                 break;
1608
1609                         case IPOPT_TS_TSANDADDR:
1610                                 if (off + sizeof(n_time) +
1611                                     sizeof(struct in_addr) > optlen) {
1612                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1613                                         goto bad;
1614                                 }
1615                                 ipaddr.sin_addr = dst;
1616                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1617                                                             m->m_pkthdr.rcvif);
1618                                 if (ia == NULL)
1619                                         continue;
1620                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1621                                     sizeof(struct in_addr));
1622                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1623                                 off += sizeof(struct in_addr);
1624                                 break;
1625
1626                         case IPOPT_TS_PRESPEC:
1627                                 if (off + sizeof(n_time) +
1628                                     sizeof(struct in_addr) > optlen) {
1629                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1630                                         goto bad;
1631                                 }
1632                                 memcpy(&ipaddr.sin_addr, sin,
1633                                     sizeof(struct in_addr));
1634                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1635                                         continue;
1636                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1637                                 off += sizeof(struct in_addr);
1638                                 break;
1639
1640                         default:
1641                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1642                                 goto bad;
1643                         }
1644                         ntime = iptime();
1645                         memcpy(cp + off, &ntime, sizeof(n_time));
1646                         cp[IPOPT_OFFSET] += sizeof(n_time);
1647                 }
1648         }
1649         if (forward && ipforwarding) {
1650                 ip_forward(m, TRUE, next_hop);
1651                 return (1);
1652         }
1653         return (0);
1654 bad:
1655         icmp_error(m, type, code, 0, 0);
1656         ipstat.ips_badoptions++;
1657         return (1);
1658 }
1659
1660 /*
1661  * Given address of next destination (final or next hop),
1662  * return internet address info of interface to be used to get there.
1663  */
1664 struct in_ifaddr *
1665 ip_rtaddr(struct in_addr dst, struct route *ro0)
1666 {
1667         struct route sro, *ro;
1668         struct sockaddr_in *sin;
1669         struct in_ifaddr *ia;
1670
1671         if (ro0 != NULL) {
1672                 ro = ro0;
1673         } else {
1674                 bzero(&sro, sizeof(sro));
1675                 ro = &sro;
1676         }
1677
1678         sin = (struct sockaddr_in *)&ro->ro_dst;
1679
1680         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1681                 if (ro->ro_rt != NULL) {
1682                         RTFREE(ro->ro_rt);
1683                         ro->ro_rt = NULL;
1684                 }
1685                 sin->sin_family = AF_INET;
1686                 sin->sin_len = sizeof *sin;
1687                 sin->sin_addr = dst;
1688                 rtalloc_ign(ro, RTF_PRCLONING);
1689         }
1690
1691         if (ro->ro_rt == NULL)
1692                 return (NULL);
1693
1694         ia = ifatoia(ro->ro_rt->rt_ifa);
1695
1696         if (ro == &sro)
1697                 RTFREE(ro->ro_rt);
1698         return ia;
1699 }
1700
1701 /*
1702  * Save incoming source route for use in replies,
1703  * to be picked up later by ip_srcroute if the receiver is interested.
1704  */
1705 static void
1706 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1707 {
1708         struct m_tag *mtag;
1709         struct ip_srcrt_opt *opt;
1710         unsigned olen;
1711
1712         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1713         if (mtag == NULL)
1714                 return;
1715         opt = m_tag_data(mtag);
1716
1717         olen = option[IPOPT_OLEN];
1718 #ifdef DIAGNOSTIC
1719         if (ipprintfs)
1720                 kprintf("save_rte: olen %d\n", olen);
1721 #endif
1722         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1723                 m_tag_free(mtag);
1724                 return;
1725         }
1726         bcopy(option, opt->ip_srcrt.srcopt, olen);
1727         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1728         opt->ip_srcrt.dst = dst;
1729         m_tag_prepend(m, mtag);
1730 }
1731
1732 /*
1733  * Retrieve incoming source route for use in replies,
1734  * in the same form used by setsockopt.
1735  * The first hop is placed before the options, will be removed later.
1736  */
1737 struct mbuf *
1738 ip_srcroute(struct mbuf *m0)
1739 {
1740         struct in_addr *p, *q;
1741         struct mbuf *m;
1742         struct m_tag *mtag;
1743         struct ip_srcrt_opt *opt;
1744
1745         if (m0 == NULL)
1746                 return NULL;
1747
1748         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1749         if (mtag == NULL)
1750                 return NULL;
1751         opt = m_tag_data(mtag);
1752
1753         if (opt->ip_nhops == 0)
1754                 return (NULL);
1755         m = m_get(MB_DONTWAIT, MT_HEADER);
1756         if (m == NULL)
1757                 return (NULL);
1758
1759 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1760
1761         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1762         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1763                    sizeof(struct in_addr) + OPTSIZ;
1764 #ifdef DIAGNOSTIC
1765         if (ipprintfs) {
1766                 kprintf("ip_srcroute: nhops %d mlen %d",
1767                         opt->ip_nhops, m->m_len);
1768         }
1769 #endif
1770
1771         /*
1772          * First save first hop for return route
1773          */
1774         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1775         *(mtod(m, struct in_addr *)) = *p--;
1776 #ifdef DIAGNOSTIC
1777         if (ipprintfs)
1778                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1779 #endif
1780
1781         /*
1782          * Copy option fields and padding (nop) to mbuf.
1783          */
1784         opt->ip_srcrt.nop = IPOPT_NOP;
1785         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1786         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1787             OPTSIZ);
1788         q = (struct in_addr *)(mtod(m, caddr_t) +
1789             sizeof(struct in_addr) + OPTSIZ);
1790 #undef OPTSIZ
1791         /*
1792          * Record return path as an IP source route,
1793          * reversing the path (pointers are now aligned).
1794          */
1795         while (p >= opt->ip_srcrt.route) {
1796 #ifdef DIAGNOSTIC
1797                 if (ipprintfs)
1798                         kprintf(" %x", ntohl(q->s_addr));
1799 #endif
1800                 *q++ = *p--;
1801         }
1802         /*
1803          * Last hop goes to final destination.
1804          */
1805         *q = opt->ip_srcrt.dst;
1806         m_tag_delete(m0, mtag);
1807 #ifdef DIAGNOSTIC
1808         if (ipprintfs)
1809                 kprintf(" %x\n", ntohl(q->s_addr));
1810 #endif
1811         return (m);
1812 }
1813
1814 /*
1815  * Strip out IP options.
1816  */
1817 void
1818 ip_stripoptions(struct mbuf *m)
1819 {
1820         int datalen;
1821         struct ip *ip = mtod(m, struct ip *);
1822         caddr_t opts;
1823         int optlen;
1824
1825         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1826         opts = (caddr_t)(ip + 1);
1827         datalen = m->m_len - (sizeof(struct ip) + optlen);
1828         bcopy(opts + optlen, opts, datalen);
1829         m->m_len -= optlen;
1830         if (m->m_flags & M_PKTHDR)
1831                 m->m_pkthdr.len -= optlen;
1832         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1833 }
1834
1835 u_char inetctlerrmap[PRC_NCMDS] = {
1836         0,              0,              0,              0,
1837         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1838         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1839         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1840         0,              0,              0,              0,
1841         ENOPROTOOPT,    ECONNREFUSED
1842 };
1843
1844 /*
1845  * Forward a packet.  If some error occurs return the sender
1846  * an icmp packet.  Note we can't always generate a meaningful
1847  * icmp message because icmp doesn't have a large enough repertoire
1848  * of codes and types.
1849  *
1850  * If not forwarding, just drop the packet.  This could be confusing
1851  * if ipforwarding was zero but some routing protocol was advancing
1852  * us as a gateway to somewhere.  However, we must let the routing
1853  * protocol deal with that.
1854  *
1855  * The using_srcrt parameter indicates whether the packet is being forwarded
1856  * via a source route.
1857  */
1858 void
1859 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1860 {
1861         struct ip *ip = mtod(m, struct ip *);
1862         struct rtentry *rt;
1863         struct route fwd_ro;
1864         int error, type = 0, code = 0, destmtu = 0;
1865         struct mbuf *mcopy, *mtemp = NULL;
1866         n_long dest;
1867         struct in_addr pkt_dst;
1868
1869         dest = INADDR_ANY;
1870         /*
1871          * Cache the destination address of the packet; this may be
1872          * changed by use of 'ipfw fwd'.
1873          */
1874         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1875
1876 #ifdef DIAGNOSTIC
1877         if (ipprintfs)
1878                 kprintf("forward: src %x dst %x ttl %x\n",
1879                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1880 #endif
1881
1882         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1883                 ipstat.ips_cantforward++;
1884                 m_freem(m);
1885                 return;
1886         }
1887         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1888                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1889                 return;
1890         }
1891
1892         bzero(&fwd_ro, sizeof(fwd_ro));
1893         ip_rtaddr(pkt_dst, &fwd_ro);
1894         if (fwd_ro.ro_rt == NULL) {
1895                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1896                 return;
1897         }
1898         rt = fwd_ro.ro_rt;
1899
1900         if (curthread->td_type == TD_TYPE_NETISR) {
1901                 /*
1902                  * Save the IP header and at most 8 bytes of the payload,
1903                  * in case we need to generate an ICMP message to the src.
1904                  */
1905                 mtemp = ipforward_mtemp[mycpuid];
1906                 KASSERT((mtemp->m_flags & M_EXT) == 0 &&
1907                     mtemp->m_data == mtemp->m_pktdat &&
1908                     m_tag_first(mtemp) == NULL,
1909                     ("ip_forward invalid mtemp1"));
1910
1911                 if (!m_dup_pkthdr(mtemp, m, MB_DONTWAIT)) {
1912                         /*
1913                          * It's probably ok if the pkthdr dup fails (because
1914                          * the deep copy of the tag chain failed), but for now
1915                          * be conservative and just discard the copy since
1916                          * code below may some day want the tags.
1917                          */
1918                         mtemp = NULL;
1919                 } else {
1920                         mtemp->m_type = m->m_type;
1921                         mtemp->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1922                             (int)ip->ip_len);
1923                         mtemp->m_pkthdr.len = mtemp->m_len;
1924                         m_copydata(m, 0, mtemp->m_len, mtod(mtemp, caddr_t));
1925                 }
1926         }
1927
1928         if (!ipstealth)
1929                 ip->ip_ttl -= IPTTLDEC;
1930
1931         /*
1932          * If forwarding packet using same interface that it came in on,
1933          * perhaps should send a redirect to sender to shortcut a hop.
1934          * Only send redirect if source is sending directly to us,
1935          * and if packet was not source routed (or has any options).
1936          * Also, don't send redirect if forwarding using a default route
1937          * or a route modified by a redirect.
1938          */
1939         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1940             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1941             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1942             ipsendredirects && !using_srcrt && next_hop == NULL) {
1943                 u_long src = ntohl(ip->ip_src.s_addr);
1944                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1945
1946                 if (rt_ifa != NULL &&
1947                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1948                         if (rt->rt_flags & RTF_GATEWAY)
1949                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1950                         else
1951                                 dest = pkt_dst.s_addr;
1952                         /*
1953                          * Router requirements says to only send
1954                          * host redirects.
1955                          */
1956                         type = ICMP_REDIRECT;
1957                         code = ICMP_REDIRECT_HOST;
1958 #ifdef DIAGNOSTIC
1959                         if (ipprintfs)
1960                                 kprintf("redirect (%d) to %x\n", code, dest);
1961 #endif
1962                 }
1963         }
1964
1965         error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
1966         if (error == 0) {
1967                 ipstat.ips_forward++;
1968                 if (type == 0) {
1969                         if (mtemp)
1970                                 ipflow_create(&fwd_ro, mtemp);
1971                         goto done;
1972                 } else {
1973                         ipstat.ips_redirectsent++;
1974                 }
1975         } else {
1976                 ipstat.ips_cantforward++;
1977         }
1978
1979         if (mtemp == NULL)
1980                 goto done;
1981
1982         /*
1983          * Errors that do not require generating ICMP message
1984          */
1985         switch (error) {
1986         case ENOBUFS:
1987                 /*
1988                  * A router should not generate ICMP_SOURCEQUENCH as
1989                  * required in RFC1812 Requirements for IP Version 4 Routers.
1990                  * Source quench could be a big problem under DoS attacks,
1991                  * or if the underlying interface is rate-limited.
1992                  * Those who need source quench packets may re-enable them
1993                  * via the net.inet.ip.sendsourcequench sysctl.
1994                  */
1995                 if (!ip_sendsourcequench)
1996                         goto done;
1997                 break;
1998
1999         case EACCES:                    /* ipfw denied packet */
2000                 goto done;
2001         }
2002
2003         KASSERT((mtemp->m_flags & M_EXT) == 0 &&
2004             mtemp->m_data == mtemp->m_pktdat,
2005             ("ip_forward invalid mtemp2"));
2006         mcopy = m_copym(mtemp, 0, mtemp->m_len, MB_DONTWAIT);
2007         if (mcopy == NULL)
2008                 goto done;
2009
2010         /*
2011          * Send ICMP message.
2012          */
2013         switch (error) {
2014         case 0:                         /* forwarded, but need redirect */
2015                 /* type, code set above */
2016                 break;
2017
2018         case ENETUNREACH:               /* shouldn't happen, checked above */
2019         case EHOSTUNREACH:
2020         case ENETDOWN:
2021         case EHOSTDOWN:
2022         default:
2023                 type = ICMP_UNREACH;
2024                 code = ICMP_UNREACH_HOST;
2025                 break;
2026
2027         case EMSGSIZE:
2028                 type = ICMP_UNREACH;
2029                 code = ICMP_UNREACH_NEEDFRAG;
2030 #ifdef IPSEC
2031                 /*
2032                  * If the packet is routed over IPsec tunnel, tell the
2033                  * originator the tunnel MTU.
2034                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2035                  * XXX quickhack!!!
2036                  */
2037                 if (fwd_ro.ro_rt != NULL) {
2038                         struct secpolicy *sp = NULL;
2039                         int ipsecerror;
2040                         int ipsechdr;
2041                         struct route *ro;
2042
2043                         sp = ipsec4_getpolicybyaddr(mcopy,
2044                                                     IPSEC_DIR_OUTBOUND,
2045                                                     IP_FORWARDING,
2046                                                     &ipsecerror);
2047
2048                         if (sp == NULL)
2049                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2050                         else {
2051                                 /* count IPsec header size */
2052                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2053                                                          IPSEC_DIR_OUTBOUND,
2054                                                          NULL);
2055
2056                                 /*
2057                                  * find the correct route for outer IPv4
2058                                  * header, compute tunnel MTU.
2059                                  *
2060                                  */
2061                                 if (sp->req != NULL && sp->req->sav != NULL &&
2062                                     sp->req->sav->sah != NULL) {
2063                                         ro = &sp->req->sav->sah->sa_route;
2064                                         if (ro->ro_rt != NULL &&
2065                                             ro->ro_rt->rt_ifp != NULL) {
2066                                                 destmtu =
2067                                                     ro->ro_rt->rt_ifp->if_mtu;
2068                                                 destmtu -= ipsechdr;
2069                                         }
2070                                 }
2071
2072                                 key_freesp(sp);
2073                         }
2074                 }
2075 #elif FAST_IPSEC
2076                 /*
2077                  * If the packet is routed over IPsec tunnel, tell the
2078                  * originator the tunnel MTU.
2079                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2080                  * XXX quickhack!!!
2081                  */
2082                 if (fwd_ro.ro_rt != NULL) {
2083                         struct secpolicy *sp = NULL;
2084                         int ipsecerror;
2085                         int ipsechdr;
2086                         struct route *ro;
2087
2088                         sp = ipsec_getpolicybyaddr(mcopy,
2089                                                    IPSEC_DIR_OUTBOUND,
2090                                                    IP_FORWARDING,
2091                                                    &ipsecerror);
2092
2093                         if (sp == NULL)
2094                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2095                         else {
2096                                 /* count IPsec header size */
2097                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2098                                                          IPSEC_DIR_OUTBOUND,
2099                                                          NULL);
2100
2101                                 /*
2102                                  * find the correct route for outer IPv4
2103                                  * header, compute tunnel MTU.
2104                                  */
2105
2106                                 if (sp->req != NULL &&
2107                                     sp->req->sav != NULL &&
2108                                     sp->req->sav->sah != NULL) {
2109                                         ro = &sp->req->sav->sah->sa_route;
2110                                         if (ro->ro_rt != NULL &&
2111                                             ro->ro_rt->rt_ifp != NULL) {
2112                                                 destmtu =
2113                                                     ro->ro_rt->rt_ifp->if_mtu;
2114                                                 destmtu -= ipsechdr;
2115                                         }
2116                                 }
2117
2118                                 KEY_FREESP(&sp);
2119                         }
2120                 }
2121 #else /* !IPSEC && !FAST_IPSEC */
2122                 if (fwd_ro.ro_rt != NULL)
2123                         destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2124 #endif /*IPSEC*/
2125                 ipstat.ips_cantfrag++;
2126                 break;
2127
2128         case ENOBUFS:
2129                 type = ICMP_SOURCEQUENCH;
2130                 code = 0;
2131                 break;
2132
2133         case EACCES:                    /* ipfw denied packet */
2134                 panic("ip_forward EACCES should not reach");
2135         }
2136         icmp_error(mcopy, type, code, dest, destmtu);
2137 done:
2138         if (mtemp != NULL)
2139                 m_tag_delete_chain(mtemp);
2140         if (fwd_ro.ro_rt != NULL)
2141                 RTFREE(fwd_ro.ro_rt);
2142 }
2143
2144 void
2145 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2146                struct mbuf *m)
2147 {
2148         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2149                 struct timeval tv;
2150
2151                 microtime(&tv);
2152                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2153                     SCM_TIMESTAMP, SOL_SOCKET);
2154                 if (*mp)
2155                         mp = &(*mp)->m_next;
2156         }
2157         if (inp->inp_flags & INP_RECVDSTADDR) {
2158                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2159                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2160                 if (*mp)
2161                         mp = &(*mp)->m_next;
2162         }
2163         if (inp->inp_flags & INP_RECVTTL) {
2164                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2165                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2166                 if (*mp)
2167                         mp = &(*mp)->m_next;
2168         }
2169 #ifdef notyet
2170         /* XXX
2171          * Moving these out of udp_input() made them even more broken
2172          * than they already were.
2173          */
2174         /* options were tossed already */
2175         if (inp->inp_flags & INP_RECVOPTS) {
2176                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2177                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2178                 if (*mp)
2179                         mp = &(*mp)->m_next;
2180         }
2181         /* ip_srcroute doesn't do what we want here, need to fix */
2182         if (inp->inp_flags & INP_RECVRETOPTS) {
2183                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2184                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2185                 if (*mp)
2186                         mp = &(*mp)->m_next;
2187         }
2188 #endif
2189         if (inp->inp_flags & INP_RECVIF) {
2190                 struct ifnet *ifp;
2191                 struct sdlbuf {
2192                         struct sockaddr_dl sdl;
2193                         u_char  pad[32];
2194                 } sdlbuf;
2195                 struct sockaddr_dl *sdp;
2196                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2197
2198                 if (((ifp = m->m_pkthdr.rcvif)) &&
2199                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2200                         sdp = IF_LLSOCKADDR(ifp);
2201                         /*
2202                          * Change our mind and don't try copy.
2203                          */
2204                         if ((sdp->sdl_family != AF_LINK) ||
2205                             (sdp->sdl_len > sizeof(sdlbuf))) {
2206                                 goto makedummy;
2207                         }
2208                         bcopy(sdp, sdl2, sdp->sdl_len);
2209                 } else {
2210 makedummy:
2211                         sdl2->sdl_len =
2212                             offsetof(struct sockaddr_dl, sdl_data[0]);
2213                         sdl2->sdl_family = AF_LINK;
2214                         sdl2->sdl_index = 0;
2215                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2216                 }
2217                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2218                         IP_RECVIF, IPPROTO_IP);
2219                 if (*mp)
2220                         mp = &(*mp)->m_next;
2221         }
2222 }
2223
2224 /*
2225  * XXX these routines are called from the upper part of the kernel.
2226  *
2227  * They could also be moved to ip_mroute.c, since all the RSVP
2228  *  handling is done there already.
2229  */
2230 int
2231 ip_rsvp_init(struct socket *so)
2232 {
2233         if (so->so_type != SOCK_RAW ||
2234             so->so_proto->pr_protocol != IPPROTO_RSVP)
2235                 return EOPNOTSUPP;
2236
2237         if (ip_rsvpd != NULL)
2238                 return EADDRINUSE;
2239
2240         ip_rsvpd = so;
2241         /*
2242          * This may seem silly, but we need to be sure we don't over-increment
2243          * the RSVP counter, in case something slips up.
2244          */
2245         if (!ip_rsvp_on) {
2246                 ip_rsvp_on = 1;
2247                 rsvp_on++;
2248         }
2249
2250         return 0;
2251 }
2252
2253 int
2254 ip_rsvp_done(void)
2255 {
2256         ip_rsvpd = NULL;
2257         /*
2258          * This may seem silly, but we need to be sure we don't over-decrement
2259          * the RSVP counter, in case something slips up.
2260          */
2261         if (ip_rsvp_on) {
2262                 ip_rsvp_on = 0;
2263                 rsvp_on--;
2264         }
2265         return 0;
2266 }
2267
2268 int
2269 rsvp_input(struct mbuf **mp, int *offp, int proto)
2270 {
2271         struct mbuf *m = *mp;
2272
2273         *mp = NULL;
2274
2275         if (rsvp_input_p) { /* call the real one if loaded */
2276                 *mp = m;
2277                 rsvp_input_p(mp, offp, proto);
2278                 return(IPPROTO_DONE);
2279         }
2280
2281         /* Can still get packets with rsvp_on = 0 if there is a local member
2282          * of the group to which the RSVP packet is addressed.  But in this
2283          * case we want to throw the packet away.
2284          */
2285
2286         if (!rsvp_on) {
2287                 m_freem(m);
2288                 return(IPPROTO_DONE);
2289         }
2290
2291         if (ip_rsvpd != NULL) {
2292                 *mp = m;
2293                 rip_input(mp, offp, proto);
2294                 return(IPPROTO_DONE);
2295         }
2296         /* Drop the packet */
2297         m_freem(m);
2298         return(IPPROTO_DONE);
2299 }