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                 if (ip_mrouter != NULL) {
 733                         /* XXX Multicast routing is not MPSAFE yet */
 734                         get_mplock();
 735
 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                         rel_mplock();
 753
 754                         /*
 755                          * The process-level routing daemon needs to receive
 756                          * all multicast IGMP packets, whether or not this
 757                          * host belongs to their destination groups.
 758                          */
 759                         if (ip->ip_p == IPPROTO_IGMP)
 760                                 goto ours;
 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                         ipstat.ips_notmember++;
 770                         m_freem(m);
 771                         return;
 772                 }
 773                 goto ours;
 774         }
 775         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
 776                 goto ours;
 777         if (ip->ip_dst.s_addr == INADDR_ANY)
 778                 goto ours;
 779
 780         /*
 781          * FAITH(Firewall Aided Internet Translator)
 782          */
 783         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 784                 if (ip_keepfaith) {
 785                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
 786                                 goto ours;
 787                 }
 788                 m_freem(m);
 789                 return;
 790         }
 791
 792         /*
 793          * Not for us; forward if possible and desirable.
 794          */
 795         if (!ipforwarding) {
 796                 ipstat.ips_cantforward++;
 797                 m_freem(m);
 798         } else {
 799 #ifdef IPSEC
 800                 /*
 801                  * Enforce inbound IPsec SPD.
 802                  */
 803                 if (ipsec4_in_reject(m, NULL)) {
 804                         ipsecstat.in_polvio++;
 805                         goto bad;
 806                 }
 807 #endif
 808 #ifdef FAST_IPSEC
 809                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 810                 crit_enter();
 811                 if (mtag != NULL) {
 812                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 813                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 814                 } else {
 815                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 816                                                    IP_FORWARDING, &error);
 817                 }
 818                 if (sp == NULL) {       /* NB: can happen if error */
 819                         crit_exit();
 820                         /*XXX error stat???*/
 821                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
 822                         goto bad;
 823                 }
 824
 825                 /*
 826                  * Check security policy against packet attributes.
 827                  */
 828                 error = ipsec_in_reject(sp, m);
 829                 KEY_FREESP(&sp);
 830                 crit_exit();
 831                 if (error) {
 832                         ipstat.ips_cantforward++;
 833                         goto bad;
 834                 }
 835 #endif
 836                 ip_forward(m, using_srcrt, next_hop);
 837         }
 838         return;
 839
 840 ours:
 841
 842         /*
 843          * IPSTEALTH: Process non-routing options only
 844          * if the packet is destined for us.
 845          */
 846         if (ipstealth &&
 847             hlen > sizeof(struct ip) &&
 848             ip_dooptions(m, 1, next_hop))
 849                 return;
 850
 851         /* Count the packet in the ip address stats */
 852         if (ia != NULL) {
 853                 IFA_STAT_INC(&ia->ia_ifa, ipackets, 1);
 854                 IFA_STAT_INC(&ia->ia_ifa, ibytes, m->m_pkthdr.len);
 855         }
 856
 857         /*
 858          * If offset or IP_MF are set, must reassemble.
 859          * Otherwise, nothing need be done.
 860          * (We could look in the reassembly queue to see
 861          * if the packet was previously fragmented,
 862          * but it's not worth the time; just let them time out.)
 863          */
 864         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
 865                 /*
 866                  * Attempt reassembly; if it succeeds, proceed.  ip_reass()
 867                  * will return a different mbuf.
 868                  *
 869                  * NOTE: ip_reass() returns m with M_HASH cleared to force
 870                  *       us to recharacterize the packet.
 871                  */
 872                 m = ip_reass(m);
 873                 if (m == NULL)
 874                         return;
 875                 ip = mtod(m, struct ip *);
 876
 877                 /* Get the header length of the reassembled packet */
 878                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 879         } else {
 880                 ip->ip_len -= hlen;
 881         }
 882
 883 #ifdef IPSEC
 884         /*
 885          * enforce IPsec policy checking if we are seeing last header.
 886          * note that we do not visit this with protocols with pcb layer
 887          * code - like udp/tcp/raw ip.
 888          */
 889         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
 890             ipsec4_in_reject(m, NULL)) {
 891                 ipsecstat.in_polvio++;
 892                 goto bad;
 893         }
 894 #endif
 895 #if FAST_IPSEC
 896         /*
 897          * enforce IPsec policy checking if we are seeing last header.
 898          * note that we do not visit this with protocols with pcb layer
 899          * code - like udp/tcp/raw ip.
 900          */
 901         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
 902                 /*
 903                  * Check if the packet has already had IPsec processing
 904                  * done.  If so, then just pass it along.  This tag gets
 905                  * set during AH, ESP, etc. input handling, before the
 906                  * packet is returned to the ip input queue for delivery.
 907                  */
 908                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
 909                 crit_enter();
 910                 if (mtag != NULL) {
 911                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 912                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
 913                 } else {
 914                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
 915                                                    IP_FORWARDING, &error);
 916                 }
 917                 if (sp != NULL) {
 918                         /*
 919                          * Check security policy against packet attributes.
 920                          */
 921                         error = ipsec_in_reject(sp, m);
 922                         KEY_FREESP(&sp);
 923                 } else {
 924                         /* XXX error stat??? */
 925                         error = EINVAL;
 926 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
 927                         goto bad;
 928                 }
 929                 crit_exit();
 930                 if (error)
 931                         goto bad;
 932         }
 933 #endif /* FAST_IPSEC */
 934
 935         /*
 936          * We must forward the packet to the correct protocol thread if
 937          * we are not already in it.
 938          *
 939          * NOTE: ip_len is now in host form.  ip_len is not adjusted
 940          *       further for protocol processing, instead we pass hlen
 941          *       to the protosw and let it deal with it.
 942          */
 943         ipstat.ips_delivered++;
 944
 945         if ((m->m_flags & M_HASH) == 0) {
 946 #ifdef RSS_DEBUG
 947                 atomic_add_long(&ip_rehash_count, 1);
 948 #endif
 949                 ip->ip_len = htons(ip->ip_len + hlen);
 950                 ip->ip_off = htons(ip->ip_off);
 951
 952                 ip_hashfn(&m, 0, IP_MPORT_IN);
 953                 if (m == NULL)
 954                         return;
 955
 956                 ip = mtod(m, struct ip *);
 957                 ip->ip_len = ntohs(ip->ip_len) - hlen;
 958                 ip->ip_off = ntohs(ip->ip_off);
 959                 KKASSERT(m->m_flags & M_HASH);
 960         }
 961         port = netisr_hashport(m->m_pkthdr.hash);
 962
 963         if (port != &curthread->td_msgport) {
 964                 struct netmsg_packet *pmsg;
 965
 966 #ifdef RSS_DEBUG
 967                 atomic_add_long(&ip_dispatch_slow, 1);
 968 #endif
 969
 970                 pmsg = &m->m_hdr.mh_netmsg;
 971                 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
 972                             0, transport_processing_handler);
 973                 pmsg->nm_packet = m;
 974                 pmsg->base.lmsg.u.ms_result = hlen;
 975                 lwkt_sendmsg(port, &pmsg->base.lmsg);
 976         } else {
 977 #ifdef RSS_DEBUG
 978                 atomic_add_long(&ip_dispatch_fast, 1);
 979 #endif
 980                 transport_processing_oncpu(m, hlen, ip);
 981         }
 982         return;
 983
 984 bad:
 985         m_freem(m);
 986 }
 987
 988 /*
 989  * Take incoming datagram fragment and try to reassemble it into
 990  * whole datagram.  If a chain for reassembly of this datagram already
 991  * exists, then it is given as fp; otherwise have to make a chain.
 992  */
 993 struct mbuf *
 994 ip_reass(struct mbuf *m)
 995 {
 996         struct ip *ip = mtod(m, struct ip *);
 997         struct mbuf *p = NULL, *q, *nq;
 998         struct mbuf *n;
 999         struct ipq *fp = NULL;
1000         struct ipqhead *head;
1001         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1002         int i, next;
1003         u_short sum;
1004
1005         /* If maxnipq is 0, never accept fragments. */
1006         if (maxnipq == 0) {
1007                 ipstat.ips_fragments++;
1008                 ipstat.ips_fragdropped++;
1009                 m_freem(m);
1010                 return NULL;
1011         }
1012
1013         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1014         /*
1015          * Look for queue of fragments of this datagram.
1016          */
1017         lwkt_gettoken(&ipq_token);
1018         head = &ipq[sum];
1019         TAILQ_FOREACH(fp, head, ipq_list) {
1020                 if (ip->ip_id == fp->ipq_id &&
1021                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1022                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1023                     ip->ip_p == fp->ipq_p)
1024                         goto found;
1025         }
1026
1027         fp = NULL;
1028
1029         /*
1030          * Enforce upper bound on number of fragmented packets
1031          * for which we attempt reassembly;
1032          * If maxnipq is -1, accept all fragments without limitation.
1033          */
1034         if (nipq > maxnipq && maxnipq > 0) {
1035                 /*
1036                  * drop something from the tail of the current queue
1037                  * before proceeding further
1038                  */
1039                 struct ipq *q = TAILQ_LAST(head, ipqhead);
1040                 if (q == NULL) {
1041                         /*
1042                          * The current queue is empty,
1043                          * so drop from one of the others.
1044                          */
1045                         for (i = 0; i < IPREASS_NHASH; i++) {
1046                                 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
1047                                 if (r) {
1048                                         ipstat.ips_fragtimeout += r->ipq_nfrags;
1049                                         ip_freef(&ipq[i], r);
1050                                         break;
1051                                 }
1052                         }
1053                 } else {
1054                         ipstat.ips_fragtimeout += q->ipq_nfrags;
1055                         ip_freef(head, q);
1056                 }
1057         }
1058 found:
1059         /*
1060          * Adjust ip_len to not reflect header,
1061          * convert offset of this to bytes.
1062          */
1063         ip->ip_len -= hlen;
1064         if (ip->ip_off & IP_MF) {
1065                 /*
1066                  * Make sure that fragments have a data length
1067                  * that's a non-zero multiple of 8 bytes.
1068                  */
1069                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1070                         ipstat.ips_toosmall++; /* XXX */
1071                         m_freem(m);
1072                         goto done;
1073                 }
1074                 m->m_flags |= M_FRAG;
1075         } else {
1076                 m->m_flags &= ~M_FRAG;
1077         }
1078         ip->ip_off <<= 3;
1079
1080         ipstat.ips_fragments++;
1081         m->m_pkthdr.header = ip;
1082
1083         /*
1084          * If the hardware has not done csum over this fragment
1085          * then csum_data is not valid at all.
1086          */
1087         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1088             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1089                 m->m_pkthdr.csum_data = 0;
1090                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1091         }
1092
1093         /*
1094          * Presence of header sizes in mbufs
1095          * would confuse code below.
1096          */
1097         m->m_data += hlen;
1098         m->m_len -= hlen;
1099
1100         /*
1101          * If first fragment to arrive, create a reassembly queue.
1102          */
1103         if (fp == NULL) {
1104                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1105                         goto dropfrag;
1106                 TAILQ_INSERT_HEAD(head, fp, ipq_list);
1107                 nipq++;
1108                 fp->ipq_nfrags = 1;
1109                 fp->ipq_ttl = IPFRAGTTL;
1110                 fp->ipq_p = ip->ip_p;
1111                 fp->ipq_id = ip->ip_id;
1112                 fp->ipq_src = ip->ip_src;
1113                 fp->ipq_dst = ip->ip_dst;
1114                 fp->ipq_frags = m;
1115                 m->m_nextpkt = NULL;
1116                 goto inserted;
1117         } else {
1118                 fp->ipq_nfrags++;
1119         }
1120
1121 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1122
1123         /*
1124          * Find a segment which begins after this one does.
1125          */
1126         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1127                 if (GETIP(q)->ip_off > ip->ip_off)
1128                         break;
1129         }
1130
1131         /*
1132          * If there is a preceding segment, it may provide some of
1133          * our data already.  If so, drop the data from the incoming
1134          * segment.  If it provides all of our data, drop us, otherwise
1135          * stick new segment in the proper place.
1136          *
1137          * If some of the data is dropped from the the preceding
1138          * segment, then it's checksum is invalidated.
1139          */
1140         if (p) {
1141                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1142                 if (i > 0) {
1143                         if (i >= ip->ip_len)
1144                                 goto dropfrag;
1145                         m_adj(m, i);
1146                         m->m_pkthdr.csum_flags = 0;
1147                         ip->ip_off += i;
1148                         ip->ip_len -= i;
1149                 }
1150                 m->m_nextpkt = p->m_nextpkt;
1151                 p->m_nextpkt = m;
1152         } else {
1153                 m->m_nextpkt = fp->ipq_frags;
1154                 fp->ipq_frags = m;
1155         }
1156
1157         /*
1158          * While we overlap succeeding segments trim them or,
1159          * if they are completely covered, dequeue them.
1160          */
1161         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1162              q = nq) {
1163                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1164                 if (i < GETIP(q)->ip_len) {
1165                         GETIP(q)->ip_len -= i;
1166                         GETIP(q)->ip_off += i;
1167                         m_adj(q, i);
1168                         q->m_pkthdr.csum_flags = 0;
1169                         break;
1170                 }
1171                 nq = q->m_nextpkt;
1172                 m->m_nextpkt = nq;
1173                 ipstat.ips_fragdropped++;
1174                 fp->ipq_nfrags--;
1175                 q->m_nextpkt = NULL;
1176                 m_freem(q);
1177         }
1178
1179 inserted:
1180         /*
1181          * Check for complete reassembly and perform frag per packet
1182          * limiting.
1183          *
1184          * Frag limiting is performed here so that the nth frag has
1185          * a chance to complete the packet before we drop the packet.
1186          * As a result, n+1 frags are actually allowed per packet, but
1187          * only n will ever be stored. (n = maxfragsperpacket.)
1188          *
1189          */
1190         next = 0;
1191         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1192                 if (GETIP(q)->ip_off != next) {
1193                         if (fp->ipq_nfrags > maxfragsperpacket) {
1194                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1195                                 ip_freef(head, fp);
1196                         }
1197                         goto done;
1198                 }
1199                 next += GETIP(q)->ip_len;
1200         }
1201         /* Make sure the last packet didn't have the IP_MF flag */
1202         if (p->m_flags & M_FRAG) {
1203                 if (fp->ipq_nfrags > maxfragsperpacket) {
1204                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1205                         ip_freef(head, fp);
1206                 }
1207                 goto done;
1208         }
1209
1210         /*
1211          * Reassembly is complete.  Make sure the packet is a sane size.
1212          */
1213         q = fp->ipq_frags;
1214         ip = GETIP(q);
1215         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1216                 ipstat.ips_toolong++;
1217                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1218                 ip_freef(head, fp);
1219                 goto done;
1220         }
1221
1222         /*
1223          * Concatenate fragments.
1224          */
1225         m = q;
1226         n = m->m_next;
1227         m->m_next = NULL;
1228         m_cat(m, n);
1229         nq = q->m_nextpkt;
1230         q->m_nextpkt = NULL;
1231         for (q = nq; q != NULL; q = nq) {
1232                 nq = q->m_nextpkt;
1233                 q->m_nextpkt = NULL;
1234                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1235                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1236                 m_cat(m, q);
1237         }
1238
1239         /*
1240          * Clean up the 1's complement checksum.  Carry over 16 bits must
1241          * be added back.  This assumes no more then 65535 packet fragments
1242          * were reassembled.  A second carry can also occur (but not a third).
1243          */
1244         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1245                                 (m->m_pkthdr.csum_data >> 16);
1246         if (m->m_pkthdr.csum_data > 0xFFFF)
1247                 m->m_pkthdr.csum_data -= 0xFFFF;
1248
1249         /*
1250          * Create header for new ip packet by
1251          * modifying header of first packet;
1252          * dequeue and discard fragment reassembly header.
1253          * Make header visible.
1254          */
1255         ip->ip_len = next;
1256         ip->ip_src = fp->ipq_src;
1257         ip->ip_dst = fp->ipq_dst;
1258         TAILQ_REMOVE(head, fp, ipq_list);
1259         nipq--;
1260         mpipe_free(&ipq_mpipe, fp);
1261         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1262         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1263         /* some debugging cruft by sklower, below, will go away soon */
1264         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1265                 int plen = 0;
1266
1267                 for (n = m; n; n = n->m_next)
1268                         plen += n->m_len;
1269                 m->m_pkthdr.len = plen;
1270         }
1271
1272         /*
1273          * Reassembly complete, return the next protocol.
1274          *
1275          * Be sure to clear M_HASH to force the packet
1276          * to be re-characterized.
1277          *
1278          * Clear M_FRAG, we are no longer a fragment.
1279          */
1280         m->m_flags &= ~(M_HASH | M_FRAG);
1281
1282         ipstat.ips_reassembled++;
1283         lwkt_reltoken(&ipq_token);
1284         return (m);
1285
1286 dropfrag:
1287         ipstat.ips_fragdropped++;
1288         if (fp != NULL)
1289                 fp->ipq_nfrags--;
1290         m_freem(m);
1291 done:
1292         lwkt_reltoken(&ipq_token);
1293         return (NULL);
1294
1295 #undef GETIP
1296 }
1297
1298 /*
1299  * Free a fragment reassembly header and all
1300  * associated datagrams.
1301  *
1302  * Called with ipq_token held.
1303  */
1304 static void
1305 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1306 {
1307         struct mbuf *q;
1308
1309         /*
1310          * Remove first to protect against blocking
1311          */
1312         TAILQ_REMOVE(fhp, fp, ipq_list);
1313
1314         /*
1315          * Clean out at our leisure
1316          */
1317         while (fp->ipq_frags) {
1318                 q = fp->ipq_frags;
1319                 fp->ipq_frags = q->m_nextpkt;
1320                 q->m_nextpkt = NULL;
1321                 m_freem(q);
1322         }
1323         mpipe_free(&ipq_mpipe, fp);
1324         nipq--;
1325 }
1326
1327 /*
1328  * IP timer processing;
1329  * if a timer expires on a reassembly
1330  * queue, discard it.
1331  */
1332 void
1333 ip_slowtimo(void)
1334 {
1335         struct ipq *fp, *fp_temp;
1336         struct ipqhead *head;
1337         int i;
1338
1339         lwkt_gettoken(&ipq_token);
1340         for (i = 0; i < IPREASS_NHASH; i++) {
1341                 head = &ipq[i];
1342                 TAILQ_FOREACH_MUTABLE(fp, head, ipq_list, fp_temp) {
1343                         if (--fp->ipq_ttl == 0) {
1344                                 ipstat.ips_fragtimeout += fp->ipq_nfrags;
1345                                 ip_freef(head, fp);
1346                         }
1347                 }
1348         }
1349         /*
1350          * If we are over the maximum number of fragments
1351          * (due to the limit being lowered), drain off
1352          * enough to get down to the new limit.
1353          */
1354         if (maxnipq >= 0 && nipq > maxnipq) {
1355                 for (i = 0; i < IPREASS_NHASH; i++) {
1356                         head = &ipq[i];
1357                         while (nipq > maxnipq && !TAILQ_EMPTY(head)) {
1358                                 ipstat.ips_fragdropped +=
1359                                     TAILQ_FIRST(head)->ipq_nfrags;
1360                                 ip_freef(head, TAILQ_FIRST(head));
1361                         }
1362                 }
1363         }
1364         lwkt_reltoken(&ipq_token);
1365         ipflow_slowtimo();
1366 }
1367
1368 /*
1369  * Drain off all datagram fragments.
1370  */
1371 void
1372 ip_drain(void)
1373 {
1374         struct ipqhead *head;
1375         int i;
1376
1377         lwkt_gettoken(&ipq_token);
1378         for (i = 0; i < IPREASS_NHASH; i++) {
1379                 head = &ipq[i];
1380                 while (!TAILQ_EMPTY(head)) {
1381                         ipstat.ips_fragdropped += TAILQ_FIRST(head)->ipq_nfrags;
1382                         ip_freef(head, TAILQ_FIRST(head));
1383                 }
1384         }
1385         lwkt_reltoken(&ipq_token);
1386         in_rtqdrain();
1387 }
1388
1389 /*
1390  * Do option processing on a datagram,
1391  * possibly discarding it if bad options are encountered,
1392  * or forwarding it if source-routed.
1393  * The pass argument is used when operating in the IPSTEALTH
1394  * mode to tell what options to process:
1395  * [LS]SRR (pass 0) or the others (pass 1).
1396  * The reason for as many as two passes is that when doing IPSTEALTH,
1397  * non-routing options should be processed only if the packet is for us.
1398  * Returns 1 if packet has been forwarded/freed,
1399  * 0 if the packet should be processed further.
1400  */
1401 static int
1402 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1403 {
1404         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1405         struct ip *ip = mtod(m, struct ip *);
1406         u_char *cp;
1407         struct in_ifaddr *ia;
1408         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1409         boolean_t forward = FALSE;
1410         struct in_addr *sin, dst;
1411         n_time ntime;
1412
1413         dst = ip->ip_dst;
1414         cp = (u_char *)(ip + 1);
1415         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1416         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1417                 opt = cp[IPOPT_OPTVAL];
1418                 if (opt == IPOPT_EOL)
1419                         break;
1420                 if (opt == IPOPT_NOP)
1421                         optlen = 1;
1422                 else {
1423                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1424                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1425                                 goto bad;
1426                         }
1427                         optlen = cp[IPOPT_OLEN];
1428                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1429                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1430                                 goto bad;
1431                         }
1432                 }
1433                 switch (opt) {
1434
1435                 default:
1436                         break;
1437
1438                 /*
1439                  * Source routing with record.
1440                  * Find interface with current destination address.
1441                  * If none on this machine then drop if strictly routed,
1442                  * or do nothing if loosely routed.
1443                  * Record interface address and bring up next address
1444                  * component.  If strictly routed make sure next
1445                  * address is on directly accessible net.
1446                  */
1447                 case IPOPT_LSRR:
1448                 case IPOPT_SSRR:
1449                         if (ipstealth && pass > 0)
1450                                 break;
1451                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1452                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1453                                 goto bad;
1454                         }
1455                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1456                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1457                                 goto bad;
1458                         }
1459                         ipaddr.sin_addr = ip->ip_dst;
1460                         ia = (struct in_ifaddr *)
1461                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1462                         if (ia == NULL) {
1463                                 if (opt == IPOPT_SSRR) {
1464                                         type = ICMP_UNREACH;
1465                                         code = ICMP_UNREACH_SRCFAIL;
1466                                         goto bad;
1467                                 }
1468                                 if (!ip_dosourceroute)
1469                                         goto nosourcerouting;
1470                                 /*
1471                                  * Loose routing, and not at next destination
1472                                  * yet; nothing to do except forward.
1473                                  */
1474                                 break;
1475                         }
1476                         off--;                  /* 0 origin */
1477                         if (off > optlen - (int)sizeof(struct in_addr)) {
1478                                 /*
1479                                  * End of source route.  Should be for us.
1480                                  */
1481                                 if (!ip_acceptsourceroute)
1482                                         goto nosourcerouting;
1483                                 save_rte(m, cp, ip->ip_src);
1484                                 break;
1485                         }
1486                         if (ipstealth)
1487                                 goto dropit;
1488                         if (!ip_dosourceroute) {
1489                                 if (ipforwarding) {
1490                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1491
1492                                         /*
1493                                          * Acting as a router, so generate ICMP
1494                                          */
1495 nosourcerouting:
1496                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1497                                         log(LOG_WARNING,
1498                                             "attempted source route from %s to %s\n",
1499                                             inet_ntoa(ip->ip_src), buf);
1500                                         type = ICMP_UNREACH;
1501                                         code = ICMP_UNREACH_SRCFAIL;
1502                                         goto bad;
1503                                 } else {
1504                                         /*
1505                                          * Not acting as a router,
1506                                          * so silently drop.
1507                                          */
1508 dropit:
1509                                         ipstat.ips_cantforward++;
1510                                         m_freem(m);
1511                                         return (1);
1512                                 }
1513                         }
1514
1515                         /*
1516                          * locate outgoing interface
1517                          */
1518                         memcpy(&ipaddr.sin_addr, cp + off,
1519                             sizeof ipaddr.sin_addr);
1520
1521                         if (opt == IPOPT_SSRR) {
1522 #define INA     struct in_ifaddr *
1523 #define SA      struct sockaddr *
1524                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1525                                                                         == NULL)
1526                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1527                         } else {
1528                                 ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1529                         }
1530                         if (ia == NULL) {
1531                                 type = ICMP_UNREACH;
1532                                 code = ICMP_UNREACH_SRCFAIL;
1533                                 goto bad;
1534                         }
1535                         ip->ip_dst = ipaddr.sin_addr;
1536                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1537                             sizeof(struct in_addr));
1538                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1539                         /*
1540                          * Let ip_intr's mcast routing check handle mcast pkts
1541                          */
1542                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1543                         break;
1544
1545                 case IPOPT_RR:
1546                         if (ipstealth && pass == 0)
1547                                 break;
1548                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1549                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1550                                 goto bad;
1551                         }
1552                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1553                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1554                                 goto bad;
1555                         }
1556                         /*
1557                          * If no space remains, ignore.
1558                          */
1559                         off--;                  /* 0 origin */
1560                         if (off > optlen - (int)sizeof(struct in_addr))
1561                                 break;
1562                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1563                             sizeof ipaddr.sin_addr);
1564                         /*
1565                          * locate outgoing interface; if we're the destination,
1566                          * use the incoming interface (should be same).
1567                          */
1568                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1569                             (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1570                                 type = ICMP_UNREACH;
1571                                 code = ICMP_UNREACH_HOST;
1572                                 goto bad;
1573                         }
1574                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1575                             sizeof(struct in_addr));
1576                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1577                         break;
1578
1579                 case IPOPT_TS:
1580                         if (ipstealth && pass == 0)
1581                                 break;
1582                         code = cp - (u_char *)ip;
1583                         if (optlen < 4 || optlen > 40) {
1584                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1585                                 goto bad;
1586                         }
1587                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1588                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1589                                 goto bad;
1590                         }
1591                         if (off > optlen - (int)sizeof(int32_t)) {
1592                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1593                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1594                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1595                                         goto bad;
1596                                 }
1597                                 break;
1598                         }
1599                         off--;                          /* 0 origin */
1600                         sin = (struct in_addr *)(cp + off);
1601                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1602
1603                         case IPOPT_TS_TSONLY:
1604                                 break;
1605
1606                         case IPOPT_TS_TSANDADDR:
1607                                 if (off + sizeof(n_time) +
1608                                     sizeof(struct in_addr) > optlen) {
1609                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1610                                         goto bad;
1611                                 }
1612                                 ipaddr.sin_addr = dst;
1613                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1614                                                             m->m_pkthdr.rcvif);
1615                                 if (ia == NULL)
1616                                         continue;
1617                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1618                                     sizeof(struct in_addr));
1619                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1620                                 off += sizeof(struct in_addr);
1621                                 break;
1622
1623                         case IPOPT_TS_PRESPEC:
1624                                 if (off + sizeof(n_time) +
1625                                     sizeof(struct in_addr) > optlen) {
1626                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1627                                         goto bad;
1628                                 }
1629                                 memcpy(&ipaddr.sin_addr, sin,
1630                                     sizeof(struct in_addr));
1631                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1632                                         continue;
1633                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1634                                 off += sizeof(struct in_addr);
1635                                 break;
1636
1637                         default:
1638                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1639                                 goto bad;
1640                         }
1641                         ntime = iptime();
1642                         memcpy(cp + off, &ntime, sizeof(n_time));
1643                         cp[IPOPT_OFFSET] += sizeof(n_time);
1644                 }
1645         }
1646         if (forward && ipforwarding) {
1647                 ip_forward(m, TRUE, next_hop);
1648                 return (1);
1649         }
1650         return (0);
1651 bad:
1652         icmp_error(m, type, code, 0, 0);
1653         ipstat.ips_badoptions++;
1654         return (1);
1655 }
1656
1657 /*
1658  * Given address of next destination (final or next hop),
1659  * return internet address info of interface to be used to get there.
1660  */
1661 struct in_ifaddr *
1662 ip_rtaddr(struct in_addr dst, struct route *ro0)
1663 {
1664         struct route sro, *ro;
1665         struct sockaddr_in *sin;
1666         struct in_ifaddr *ia;
1667
1668         if (ro0 != NULL) {
1669                 ro = ro0;
1670         } else {
1671                 bzero(&sro, sizeof(sro));
1672                 ro = &sro;
1673         }
1674
1675         sin = (struct sockaddr_in *)&ro->ro_dst;
1676
1677         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1678                 if (ro->ro_rt != NULL) {
1679                         RTFREE(ro->ro_rt);
1680                         ro->ro_rt = NULL;
1681                 }
1682                 sin->sin_family = AF_INET;
1683                 sin->sin_len = sizeof *sin;
1684                 sin->sin_addr = dst;
1685                 rtalloc_ign(ro, RTF_PRCLONING);
1686         }
1687
1688         if (ro->ro_rt == NULL)
1689                 return (NULL);
1690
1691         ia = ifatoia(ro->ro_rt->rt_ifa);
1692
1693         if (ro == &sro)
1694                 RTFREE(ro->ro_rt);
1695         return ia;
1696 }
1697
1698 /*
1699  * Save incoming source route for use in replies,
1700  * to be picked up later by ip_srcroute if the receiver is interested.
1701  */
1702 static void
1703 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1704 {
1705         struct m_tag *mtag;
1706         struct ip_srcrt_opt *opt;
1707         unsigned olen;
1708
1709         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1710         if (mtag == NULL)
1711                 return;
1712         opt = m_tag_data(mtag);
1713
1714         olen = option[IPOPT_OLEN];
1715 #ifdef DIAGNOSTIC
1716         if (ipprintfs)
1717                 kprintf("save_rte: olen %d\n", olen);
1718 #endif
1719         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1720                 m_tag_free(mtag);
1721                 return;
1722         }
1723         bcopy(option, opt->ip_srcrt.srcopt, olen);
1724         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1725         opt->ip_srcrt.dst = dst;
1726         m_tag_prepend(m, mtag);
1727 }
1728
1729 /*
1730  * Retrieve incoming source route for use in replies,
1731  * in the same form used by setsockopt.
1732  * The first hop is placed before the options, will be removed later.
1733  */
1734 struct mbuf *
1735 ip_srcroute(struct mbuf *m0)
1736 {
1737         struct in_addr *p, *q;
1738         struct mbuf *m;
1739         struct m_tag *mtag;
1740         struct ip_srcrt_opt *opt;
1741
1742         if (m0 == NULL)
1743                 return NULL;
1744
1745         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1746         if (mtag == NULL)
1747                 return NULL;
1748         opt = m_tag_data(mtag);
1749
1750         if (opt->ip_nhops == 0)
1751                 return (NULL);
1752         m = m_get(MB_DONTWAIT, MT_HEADER);
1753         if (m == NULL)
1754                 return (NULL);
1755
1756 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1757
1758         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1759         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1760                    sizeof(struct in_addr) + OPTSIZ;
1761 #ifdef DIAGNOSTIC
1762         if (ipprintfs) {
1763                 kprintf("ip_srcroute: nhops %d mlen %d",
1764                         opt->ip_nhops, m->m_len);
1765         }
1766 #endif
1767
1768         /*
1769          * First save first hop for return route
1770          */
1771         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1772         *(mtod(m, struct in_addr *)) = *p--;
1773 #ifdef DIAGNOSTIC
1774         if (ipprintfs)
1775                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1776 #endif
1777
1778         /*
1779          * Copy option fields and padding (nop) to mbuf.
1780          */
1781         opt->ip_srcrt.nop = IPOPT_NOP;
1782         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1783         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1784             OPTSIZ);
1785         q = (struct in_addr *)(mtod(m, caddr_t) +
1786             sizeof(struct in_addr) + OPTSIZ);
1787 #undef OPTSIZ
1788         /*
1789          * Record return path as an IP source route,
1790          * reversing the path (pointers are now aligned).
1791          */
1792         while (p >= opt->ip_srcrt.route) {
1793 #ifdef DIAGNOSTIC
1794                 if (ipprintfs)
1795                         kprintf(" %x", ntohl(q->s_addr));
1796 #endif
1797                 *q++ = *p--;
1798         }
1799         /*
1800          * Last hop goes to final destination.
1801          */
1802         *q = opt->ip_srcrt.dst;
1803         m_tag_delete(m0, mtag);
1804 #ifdef DIAGNOSTIC
1805         if (ipprintfs)
1806                 kprintf(" %x\n", ntohl(q->s_addr));
1807 #endif
1808         return (m);
1809 }
1810
1811 /*
1812  * Strip out IP options.
1813  */
1814 void
1815 ip_stripoptions(struct mbuf *m)
1816 {
1817         int datalen;
1818         struct ip *ip = mtod(m, struct ip *);
1819         caddr_t opts;
1820         int optlen;
1821
1822         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1823         opts = (caddr_t)(ip + 1);
1824         datalen = m->m_len - (sizeof(struct ip) + optlen);
1825         bcopy(opts + optlen, opts, datalen);
1826         m->m_len -= optlen;
1827         if (m->m_flags & M_PKTHDR)
1828                 m->m_pkthdr.len -= optlen;
1829         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1830 }
1831
1832 u_char inetctlerrmap[PRC_NCMDS] = {
1833         0,              0,              0,              0,
1834         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1835         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1836         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1837         0,              0,              0,              0,
1838         ENOPROTOOPT,    ECONNREFUSED
1839 };
1840
1841 /*
1842  * Forward a packet.  If some error occurs return the sender
1843  * an icmp packet.  Note we can't always generate a meaningful
1844  * icmp message because icmp doesn't have a large enough repertoire
1845  * of codes and types.
1846  *
1847  * If not forwarding, just drop the packet.  This could be confusing
1848  * if ipforwarding was zero but some routing protocol was advancing
1849  * us as a gateway to somewhere.  However, we must let the routing
1850  * protocol deal with that.
1851  *
1852  * The using_srcrt parameter indicates whether the packet is being forwarded
1853  * via a source route.
1854  */
1855 void
1856 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1857 {
1858         struct ip *ip = mtod(m, struct ip *);
1859         struct rtentry *rt;
1860         struct route fwd_ro;
1861         int error, type = 0, code = 0, destmtu = 0;
1862         struct mbuf *mcopy, *mtemp = NULL;
1863         n_long dest;
1864         struct in_addr pkt_dst;
1865
1866         dest = INADDR_ANY;
1867         /*
1868          * Cache the destination address of the packet; this may be
1869          * changed by use of 'ipfw fwd'.
1870          */
1871         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1872
1873 #ifdef DIAGNOSTIC
1874         if (ipprintfs)
1875                 kprintf("forward: src %x dst %x ttl %x\n",
1876                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1877 #endif
1878
1879         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1880                 ipstat.ips_cantforward++;
1881                 m_freem(m);
1882                 return;
1883         }
1884         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1885                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1886                 return;
1887         }
1888
1889         bzero(&fwd_ro, sizeof(fwd_ro));
1890         ip_rtaddr(pkt_dst, &fwd_ro);
1891         if (fwd_ro.ro_rt == NULL) {
1892                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1893                 return;
1894         }
1895         rt = fwd_ro.ro_rt;
1896
1897         if (curthread->td_type == TD_TYPE_NETISR) {
1898                 /*
1899                  * Save the IP header and at most 8 bytes of the payload,
1900                  * in case we need to generate an ICMP message to the src.
1901                  */
1902                 mtemp = ipforward_mtemp[mycpuid];
1903                 KASSERT((mtemp->m_flags & M_EXT) == 0 &&
1904                     mtemp->m_data == mtemp->m_pktdat &&
1905                     m_tag_first(mtemp) == NULL,
1906                     ("ip_forward invalid mtemp1"));
1907
1908                 if (!m_dup_pkthdr(mtemp, m, MB_DONTWAIT)) {
1909                         /*
1910                          * It's probably ok if the pkthdr dup fails (because
1911                          * the deep copy of the tag chain failed), but for now
1912                          * be conservative and just discard the copy since
1913                          * code below may some day want the tags.
1914                          */
1915                         mtemp = NULL;
1916                 } else {
1917                         mtemp->m_type = m->m_type;
1918                         mtemp->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1919                             (int)ip->ip_len);
1920                         mtemp->m_pkthdr.len = mtemp->m_len;
1921                         m_copydata(m, 0, mtemp->m_len, mtod(mtemp, caddr_t));
1922                 }
1923         }
1924
1925         if (!ipstealth)
1926                 ip->ip_ttl -= IPTTLDEC;
1927
1928         /*
1929          * If forwarding packet using same interface that it came in on,
1930          * perhaps should send a redirect to sender to shortcut a hop.
1931          * Only send redirect if source is sending directly to us,
1932          * and if packet was not source routed (or has any options).
1933          * Also, don't send redirect if forwarding using a default route
1934          * or a route modified by a redirect.
1935          */
1936         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1937             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1938             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1939             ipsendredirects && !using_srcrt && next_hop == NULL) {
1940                 u_long src = ntohl(ip->ip_src.s_addr);
1941                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1942
1943                 if (rt_ifa != NULL &&
1944                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1945                         if (rt->rt_flags & RTF_GATEWAY)
1946                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1947                         else
1948                                 dest = pkt_dst.s_addr;
1949                         /*
1950                          * Router requirements says to only send
1951                          * host redirects.
1952                          */
1953                         type = ICMP_REDIRECT;
1954                         code = ICMP_REDIRECT_HOST;
1955 #ifdef DIAGNOSTIC
1956                         if (ipprintfs)
1957                                 kprintf("redirect (%d) to %x\n", code, dest);
1958 #endif
1959                 }
1960         }
1961
1962         error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
1963         if (error == 0) {
1964                 ipstat.ips_forward++;
1965                 if (type == 0) {
1966                         if (mtemp)
1967                                 ipflow_create(&fwd_ro, mtemp);
1968                         goto done;
1969                 } else {
1970                         ipstat.ips_redirectsent++;
1971                 }
1972         } else {
1973                 ipstat.ips_cantforward++;
1974         }
1975
1976         if (mtemp == NULL)
1977                 goto done;
1978
1979         /*
1980          * Errors that do not require generating ICMP message
1981          */
1982         switch (error) {
1983         case ENOBUFS:
1984                 /*
1985                  * A router should not generate ICMP_SOURCEQUENCH as
1986                  * required in RFC1812 Requirements for IP Version 4 Routers.
1987                  * Source quench could be a big problem under DoS attacks,
1988                  * or if the underlying interface is rate-limited.
1989                  * Those who need source quench packets may re-enable them
1990                  * via the net.inet.ip.sendsourcequench sysctl.
1991                  */
1992                 if (!ip_sendsourcequench)
1993                         goto done;
1994                 break;
1995
1996         case EACCES:                    /* ipfw denied packet */
1997                 goto done;
1998         }
1999
2000         KASSERT((mtemp->m_flags & M_EXT) == 0 &&
2001             mtemp->m_data == mtemp->m_pktdat,
2002             ("ip_forward invalid mtemp2"));
2003         mcopy = m_copym(mtemp, 0, mtemp->m_len, MB_DONTWAIT);
2004         if (mcopy == NULL)
2005                 goto done;
2006
2007         /*
2008          * Send ICMP message.
2009          */
2010         switch (error) {
2011         case 0:                         /* forwarded, but need redirect */
2012                 /* type, code set above */
2013                 break;
2014
2015         case ENETUNREACH:               /* shouldn't happen, checked above */
2016         case EHOSTUNREACH:
2017         case ENETDOWN:
2018         case EHOSTDOWN:
2019         default:
2020                 type = ICMP_UNREACH;
2021                 code = ICMP_UNREACH_HOST;
2022                 break;
2023
2024         case EMSGSIZE:
2025                 type = ICMP_UNREACH;
2026                 code = ICMP_UNREACH_NEEDFRAG;
2027 #ifdef IPSEC
2028                 /*
2029                  * If the packet is routed over IPsec tunnel, tell the
2030                  * originator the tunnel MTU.
2031                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2032                  * XXX quickhack!!!
2033                  */
2034                 if (fwd_ro.ro_rt != NULL) {
2035                         struct secpolicy *sp = NULL;
2036                         int ipsecerror;
2037                         int ipsechdr;
2038                         struct route *ro;
2039
2040                         sp = ipsec4_getpolicybyaddr(mcopy,
2041                                                     IPSEC_DIR_OUTBOUND,
2042                                                     IP_FORWARDING,
2043                                                     &ipsecerror);
2044
2045                         if (sp == NULL)
2046                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2047                         else {
2048                                 /* count IPsec header size */
2049                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2050                                                          IPSEC_DIR_OUTBOUND,
2051                                                          NULL);
2052
2053                                 /*
2054                                  * find the correct route for outer IPv4
2055                                  * header, compute tunnel MTU.
2056                                  *
2057                                  */
2058                                 if (sp->req != NULL && sp->req->sav != NULL &&
2059                                     sp->req->sav->sah != NULL) {
2060                                         ro = &sp->req->sav->sah->sa_route;
2061                                         if (ro->ro_rt != NULL &&
2062                                             ro->ro_rt->rt_ifp != NULL) {
2063                                                 destmtu =
2064                                                     ro->ro_rt->rt_ifp->if_mtu;
2065                                                 destmtu -= ipsechdr;
2066                                         }
2067                                 }
2068
2069                                 key_freesp(sp);
2070                         }
2071                 }
2072 #elif FAST_IPSEC
2073                 /*
2074                  * If the packet is routed over IPsec tunnel, tell the
2075                  * originator the tunnel MTU.
2076                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2077                  * XXX quickhack!!!
2078                  */
2079                 if (fwd_ro.ro_rt != NULL) {
2080                         struct secpolicy *sp = NULL;
2081                         int ipsecerror;
2082                         int ipsechdr;
2083                         struct route *ro;
2084
2085                         sp = ipsec_getpolicybyaddr(mcopy,
2086                                                    IPSEC_DIR_OUTBOUND,
2087                                                    IP_FORWARDING,
2088                                                    &ipsecerror);
2089
2090                         if (sp == NULL)
2091                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2092                         else {
2093                                 /* count IPsec header size */
2094                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2095                                                          IPSEC_DIR_OUTBOUND,
2096                                                          NULL);
2097
2098                                 /*
2099                                  * find the correct route for outer IPv4
2100                                  * header, compute tunnel MTU.
2101                                  */
2102
2103                                 if (sp->req != NULL &&
2104                                     sp->req->sav != NULL &&
2105                                     sp->req->sav->sah != NULL) {
2106                                         ro = &sp->req->sav->sah->sa_route;
2107                                         if (ro->ro_rt != NULL &&
2108                                             ro->ro_rt->rt_ifp != NULL) {
2109                                                 destmtu =
2110                                                     ro->ro_rt->rt_ifp->if_mtu;
2111                                                 destmtu -= ipsechdr;
2112                                         }
2113                                 }
2114
2115                                 KEY_FREESP(&sp);
2116                         }
2117                 }
2118 #else /* !IPSEC && !FAST_IPSEC */
2119                 if (fwd_ro.ro_rt != NULL)
2120                         destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2121 #endif /*IPSEC*/
2122                 ipstat.ips_cantfrag++;
2123                 break;
2124
2125         case ENOBUFS:
2126                 type = ICMP_SOURCEQUENCH;
2127                 code = 0;
2128                 break;
2129
2130         case EACCES:                    /* ipfw denied packet */
2131                 panic("ip_forward EACCES should not reach");
2132         }
2133         icmp_error(mcopy, type, code, dest, destmtu);
2134 done:
2135         if (mtemp != NULL)
2136                 m_tag_delete_chain(mtemp);
2137         if (fwd_ro.ro_rt != NULL)
2138                 RTFREE(fwd_ro.ro_rt);
2139 }
2140
2141 void
2142 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2143                struct mbuf *m)
2144 {
2145         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2146                 struct timeval tv;
2147
2148                 microtime(&tv);
2149                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2150                     SCM_TIMESTAMP, SOL_SOCKET);
2151                 if (*mp)
2152                         mp = &(*mp)->m_next;
2153         }
2154         if (inp->inp_flags & INP_RECVDSTADDR) {
2155                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2156                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2157                 if (*mp)
2158                         mp = &(*mp)->m_next;
2159         }
2160         if (inp->inp_flags & INP_RECVTTL) {
2161                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2162                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2163                 if (*mp)
2164                         mp = &(*mp)->m_next;
2165         }
2166 #ifdef notyet
2167         /* XXX
2168          * Moving these out of udp_input() made them even more broken
2169          * than they already were.
2170          */
2171         /* options were tossed already */
2172         if (inp->inp_flags & INP_RECVOPTS) {
2173                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2174                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2175                 if (*mp)
2176                         mp = &(*mp)->m_next;
2177         }
2178         /* ip_srcroute doesn't do what we want here, need to fix */
2179         if (inp->inp_flags & INP_RECVRETOPTS) {
2180                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2181                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2182                 if (*mp)
2183                         mp = &(*mp)->m_next;
2184         }
2185 #endif
2186         if (inp->inp_flags & INP_RECVIF) {
2187                 struct ifnet *ifp;
2188                 struct sdlbuf {
2189                         struct sockaddr_dl sdl;
2190                         u_char  pad[32];
2191                 } sdlbuf;
2192                 struct sockaddr_dl *sdp;
2193                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2194
2195                 if (((ifp = m->m_pkthdr.rcvif)) &&
2196                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2197                         sdp = IF_LLSOCKADDR(ifp);
2198                         /*
2199                          * Change our mind and don't try copy.
2200                          */
2201                         if ((sdp->sdl_family != AF_LINK) ||
2202                             (sdp->sdl_len > sizeof(sdlbuf))) {
2203                                 goto makedummy;
2204                         }
2205                         bcopy(sdp, sdl2, sdp->sdl_len);
2206                 } else {
2207 makedummy:
2208                         sdl2->sdl_len =
2209                             offsetof(struct sockaddr_dl, sdl_data[0]);
2210                         sdl2->sdl_family = AF_LINK;
2211                         sdl2->sdl_index = 0;
2212                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2213                 }
2214                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2215                         IP_RECVIF, IPPROTO_IP);
2216                 if (*mp)
2217                         mp = &(*mp)->m_next;
2218         }
2219 }
2220
2221 /*
2222  * XXX these routines are called from the upper part of the kernel.
2223  *
2224  * They could also be moved to ip_mroute.c, since all the RSVP
2225  *  handling is done there already.
2226  */
2227 int
2228 ip_rsvp_init(struct socket *so)
2229 {
2230         if (so->so_type != SOCK_RAW ||
2231             so->so_proto->pr_protocol != IPPROTO_RSVP)
2232                 return EOPNOTSUPP;
2233
2234         if (ip_rsvpd != NULL)
2235                 return EADDRINUSE;
2236
2237         ip_rsvpd = so;
2238         /*
2239          * This may seem silly, but we need to be sure we don't over-increment
2240          * the RSVP counter, in case something slips up.
2241          */
2242         if (!ip_rsvp_on) {
2243                 ip_rsvp_on = 1;
2244                 rsvp_on++;
2245         }
2246
2247         return 0;
2248 }
2249
2250 int
2251 ip_rsvp_done(void)
2252 {
2253         ip_rsvpd = NULL;
2254         /*
2255          * This may seem silly, but we need to be sure we don't over-decrement
2256          * the RSVP counter, in case something slips up.
2257          */
2258         if (ip_rsvp_on) {
2259                 ip_rsvp_on = 0;
2260                 rsvp_on--;
2261         }
2262         return 0;
2263 }
2264
2265 int
2266 rsvp_input(struct mbuf **mp, int *offp, int proto)
2267 {
2268         struct mbuf *m = *mp;
2269
2270         *mp = NULL;
2271
2272         if (rsvp_input_p) { /* call the real one if loaded */
2273                 *mp = m;
2274                 rsvp_input_p(mp, offp, proto);
2275                 return(IPPROTO_DONE);
2276         }
2277
2278         /* Can still get packets with rsvp_on = 0 if there is a local member
2279          * of the group to which the RSVP packet is addressed.  But in this
2280          * case we want to throw the packet away.
2281          */
2282
2283         if (!rsvp_on) {
2284                 m_freem(m);
2285                 return(IPPROTO_DONE);
2286         }
2287
2288         if (ip_rsvpd != NULL) {
2289                 *mp = m;
2290                 rip_input(mp, offp, proto);
2291                 return(IPPROTO_DONE);
2292         }
2293         /* Drop the packet */
2294         m_freem(m);
2295         return(IPPROTO_DONE);
2296 }