ip_reass.c revision 220878
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/netinet/ip_input.c 220878 2011-04-20 07:55:33Z bz $"); 34 35#include "opt_bootp.h" 36#include "opt_ipfw.h" 37#include "opt_ipstealth.h" 38#include "opt_ipsec.h" 39#include "opt_route.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/callout.h> 44#include <sys/mbuf.h> 45#include <sys/malloc.h> 46#include <sys/domain.h> 47#include <sys/protosw.h> 48#include <sys/socket.h> 49#include <sys/time.h> 50#include <sys/kernel.h> 51#include <sys/lock.h> 52#include <sys/rwlock.h> 53#include <sys/syslog.h> 54#include <sys/sysctl.h> 55 56#include <net/pfil.h> 57#include <net/if.h> 58#include <net/if_types.h> 59#include <net/if_var.h> 60#include <net/if_dl.h> 61#include <net/route.h> 62#include <net/netisr.h> 63#include <net/vnet.h> 64#include <net/flowtable.h> 65 66#include <netinet/in.h> 67#include <netinet/in_systm.h> 68#include <netinet/in_var.h> 69#include <netinet/ip.h> 70#include <netinet/in_pcb.h> 71#include <netinet/ip_var.h> 72#include <netinet/ip_fw.h> 73#include <netinet/ip_icmp.h> 74#include <netinet/ip_options.h> 75#include <machine/in_cksum.h> 76#include <netinet/ip_carp.h> 77#ifdef IPSEC 78#include <netinet/ip_ipsec.h> 79#endif /* IPSEC */ 80 81#include <sys/socketvar.h> 82 83#include <security/mac/mac_framework.h> 84 85#ifdef CTASSERT 86CTASSERT(sizeof(struct ip) == 20); 87#endif 88 89struct rwlock in_ifaddr_lock; 90RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock"); 91 92VNET_DEFINE(int, rsvp_on); 93 94VNET_DEFINE(int, ipforwarding); 95SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW, 96 &VNET_NAME(ipforwarding), 0, 97 "Enable IP forwarding between interfaces"); 98 99static VNET_DEFINE(int, ipsendredirects) = 1; /* XXX */ 100#define V_ipsendredirects VNET(ipsendredirects) 101SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW, 102 &VNET_NAME(ipsendredirects), 0, 103 "Enable sending IP redirects"); 104 105VNET_DEFINE(int, ip_defttl) = IPDEFTTL; 106SYSCTL_VNET_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW, 107 &VNET_NAME(ip_defttl), 0, 108 "Maximum TTL on IP packets"); 109 110static VNET_DEFINE(int, ip_keepfaith); 111#define V_ip_keepfaith VNET(ip_keepfaith) 112SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW, 113 &VNET_NAME(ip_keepfaith), 0, 114 "Enable packet capture for FAITH IPv4->IPv6 translater daemon"); 115 116static VNET_DEFINE(int, ip_sendsourcequench); 117#define V_ip_sendsourcequench VNET(ip_sendsourcequench) 118SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW, 119 &VNET_NAME(ip_sendsourcequench), 0, 120 "Enable the transmission of source quench packets"); 121 122VNET_DEFINE(int, ip_do_randomid); 123SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW, 124 &VNET_NAME(ip_do_randomid), 0, 125 "Assign random ip_id values"); 126 127/* 128 * XXX - Setting ip_checkinterface mostly implements the receive side of 129 * the Strong ES model described in RFC 1122, but since the routing table 130 * and transmit implementation do not implement the Strong ES model, 131 * setting this to 1 results in an odd hybrid. 132 * 133 * XXX - ip_checkinterface currently must be disabled if you use ipnat 134 * to translate the destination address to another local interface. 135 * 136 * XXX - ip_checkinterface must be disabled if you add IP aliases 137 * to the loopback interface instead of the interface where the 138 * packets for those addresses are received. 139 */ 140static VNET_DEFINE(int, ip_checkinterface); 141#define V_ip_checkinterface VNET(ip_checkinterface) 142SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW, 143 &VNET_NAME(ip_checkinterface), 0, 144 "Verify packet arrives on correct interface"); 145 146VNET_DEFINE(struct pfil_head, inet_pfil_hook); /* Packet filter hooks */ 147 148static struct netisr_handler ip_nh = { 149 .nh_name = "ip", 150 .nh_handler = ip_input, 151 .nh_proto = NETISR_IP, 152 .nh_policy = NETISR_POLICY_FLOW, 153}; 154 155extern struct domain inetdomain; 156extern struct protosw inetsw[]; 157u_char ip_protox[IPPROTO_MAX]; 158VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead); /* first inet address */ 159VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table */ 160VNET_DEFINE(u_long, in_ifaddrhmask); /* mask for hash table */ 161 162VNET_DEFINE(struct ipstat, ipstat); 163SYSCTL_VNET_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW, 164 &VNET_NAME(ipstat), ipstat, 165 "IP statistics (struct ipstat, netinet/ip_var.h)"); 166 167static VNET_DEFINE(uma_zone_t, ipq_zone); 168static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]); 169static struct mtx ipqlock; 170 171#define V_ipq_zone VNET(ipq_zone) 172#define V_ipq VNET(ipq) 173 174#define IPQ_LOCK() mtx_lock(&ipqlock) 175#define IPQ_UNLOCK() mtx_unlock(&ipqlock) 176#define IPQ_LOCK_INIT() mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF) 177#define IPQ_LOCK_ASSERT() mtx_assert(&ipqlock, MA_OWNED) 178 179static void maxnipq_update(void); 180static void ipq_zone_change(void *); 181static void ip_drain_locked(void); 182 183static VNET_DEFINE(int, maxnipq); /* Administrative limit on # reass queues. */ 184static VNET_DEFINE(int, nipq); /* Total # of reass queues */ 185#define V_maxnipq VNET(maxnipq) 186#define V_nipq VNET(nipq) 187SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD, 188 &VNET_NAME(nipq), 0, 189 "Current number of IPv4 fragment reassembly queue entries"); 190 191static VNET_DEFINE(int, maxfragsperpacket); 192#define V_maxfragsperpacket VNET(maxfragsperpacket) 193SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW, 194 &VNET_NAME(maxfragsperpacket), 0, 195 "Maximum number of IPv4 fragments allowed per packet"); 196 197struct callout ipport_tick_callout; 198 199#ifdef IPCTL_DEFMTU 200SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW, 201 &ip_mtu, 0, "Default MTU"); 202#endif 203 204#ifdef IPSTEALTH 205VNET_DEFINE(int, ipstealth); 206SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, 207 &VNET_NAME(ipstealth), 0, 208 "IP stealth mode, no TTL decrementation on forwarding"); 209#endif 210 211#ifdef FLOWTABLE 212static VNET_DEFINE(int, ip_output_flowtable_size) = 2048; 213VNET_DEFINE(struct flowtable *, ip_ft); 214#define V_ip_output_flowtable_size VNET(ip_output_flowtable_size) 215 216SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN, 217 &VNET_NAME(ip_output_flowtable_size), 2048, 218 "number of entries in the per-cpu output flow caches"); 219#endif 220 221static void ip_freef(struct ipqhead *, struct ipq *); 222 223/* 224 * Kernel module interface for updating ipstat. The argument is an index 225 * into ipstat treated as an array of u_long. While this encodes the general 226 * layout of ipstat into the caller, it doesn't encode its location, so that 227 * future changes to add, for example, per-CPU stats support won't cause 228 * binary compatibility problems for kernel modules. 229 */ 230void 231kmod_ipstat_inc(int statnum) 232{ 233 234 (*((u_long *)&V_ipstat + statnum))++; 235} 236 237void 238kmod_ipstat_dec(int statnum) 239{ 240 241 (*((u_long *)&V_ipstat + statnum))--; 242} 243 244static int 245sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS) 246{ 247 int error, qlimit; 248 249 netisr_getqlimit(&ip_nh, &qlimit); 250 error = sysctl_handle_int(oidp, &qlimit, 0, req); 251 if (error || !req->newptr) 252 return (error); 253 if (qlimit < 1) 254 return (EINVAL); 255 return (netisr_setqlimit(&ip_nh, qlimit)); 256} 257SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, 258 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I", 259 "Maximum size of the IP input queue"); 260 261static int 262sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS) 263{ 264 u_int64_t qdrops_long; 265 int error, qdrops; 266 267 netisr_getqdrops(&ip_nh, &qdrops_long); 268 qdrops = qdrops_long; 269 error = sysctl_handle_int(oidp, &qdrops, 0, req); 270 if (error || !req->newptr) 271 return (error); 272 if (qdrops != 0) 273 return (EINVAL); 274 netisr_clearqdrops(&ip_nh); 275 return (0); 276} 277 278SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, 279 CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I", 280 "Number of packets dropped from the IP input queue"); 281 282/* 283 * IP initialization: fill in IP protocol switch table. 284 * All protocols not implemented in kernel go to raw IP protocol handler. 285 */ 286void 287ip_init(void) 288{ 289 struct protosw *pr; 290 int i; 291 292 V_ip_id = time_second & 0xffff; 293 294 TAILQ_INIT(&V_in_ifaddrhead); 295 V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask); 296 297 /* Initialize IP reassembly queue. */ 298 for (i = 0; i < IPREASS_NHASH; i++) 299 TAILQ_INIT(&V_ipq[i]); 300 V_maxnipq = nmbclusters / 32; 301 V_maxfragsperpacket = 16; 302 V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, 303 NULL, UMA_ALIGN_PTR, 0); 304 maxnipq_update(); 305 306 /* Initialize packet filter hooks. */ 307 V_inet_pfil_hook.ph_type = PFIL_TYPE_AF; 308 V_inet_pfil_hook.ph_af = AF_INET; 309 if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0) 310 printf("%s: WARNING: unable to register pfil hook, " 311 "error %d\n", __func__, i); 312 313#ifdef FLOWTABLE 314 if (TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size", 315 &V_ip_output_flowtable_size)) { 316 if (V_ip_output_flowtable_size < 256) 317 V_ip_output_flowtable_size = 256; 318 if (!powerof2(V_ip_output_flowtable_size)) { 319 printf("flowtable must be power of 2 size\n"); 320 V_ip_output_flowtable_size = 2048; 321 } 322 } else { 323 /* 324 * round up to the next power of 2 325 */ 326 V_ip_output_flowtable_size = 1 << fls((1024 + maxusers * 64)-1); 327 } 328 V_ip_ft = flowtable_alloc("ipv4", V_ip_output_flowtable_size, FL_PCPU); 329#endif 330 331 /* Skip initialization of globals for non-default instances. */ 332 if (!IS_DEFAULT_VNET(curvnet)) 333 return; 334 335 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 336 if (pr == NULL) 337 panic("ip_init: PF_INET not found"); 338 339 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */ 340 for (i = 0; i < IPPROTO_MAX; i++) 341 ip_protox[i] = pr - inetsw; 342 /* 343 * Cycle through IP protocols and put them into the appropriate place 344 * in ip_protox[]. 345 */ 346 for (pr = inetdomain.dom_protosw; 347 pr < inetdomain.dom_protoswNPROTOSW; pr++) 348 if (pr->pr_domain->dom_family == PF_INET && 349 pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) { 350 /* Be careful to only index valid IP protocols. */ 351 if (pr->pr_protocol < IPPROTO_MAX) 352 ip_protox[pr->pr_protocol] = pr - inetsw; 353 } 354 355 /* Start ipport_tick. */ 356 callout_init(&ipport_tick_callout, CALLOUT_MPSAFE); 357 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 358 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 359 SHUTDOWN_PRI_DEFAULT); 360 EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change, 361 NULL, EVENTHANDLER_PRI_ANY); 362 363 /* Initialize various other remaining things. */ 364 IPQ_LOCK_INIT(); 365 netisr_register(&ip_nh); 366} 367 368#ifdef VIMAGE 369void 370ip_destroy(void) 371{ 372 373 /* Cleanup in_ifaddr hash table; should be empty. */ 374 hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask); 375 376 IPQ_LOCK(); 377 ip_drain_locked(); 378 IPQ_UNLOCK(); 379 380 uma_zdestroy(V_ipq_zone); 381} 382#endif 383 384void 385ip_fini(void *xtp) 386{ 387 388 callout_stop(&ipport_tick_callout); 389} 390 391/* 392 * Ip input routine. Checksum and byte swap header. If fragmented 393 * try to reassemble. Process options. Pass to next level. 394 */ 395void 396ip_input(struct mbuf *m) 397{ 398 struct ip *ip = NULL; 399 struct in_ifaddr *ia = NULL; 400 struct ifaddr *ifa; 401 struct ifnet *ifp; 402 int checkif, hlen = 0; 403 u_short sum; 404 int dchg = 0; /* dest changed after fw */ 405 struct in_addr odst; /* original dst address */ 406 407 M_ASSERTPKTHDR(m); 408 409 if (m->m_flags & M_FASTFWD_OURS) { 410 /* 411 * Firewall or NAT changed destination to local. 412 * We expect ip_len and ip_off to be in host byte order. 413 */ 414 m->m_flags &= ~M_FASTFWD_OURS; 415 /* Set up some basics that will be used later. */ 416 ip = mtod(m, struct ip *); 417 hlen = ip->ip_hl << 2; 418 goto ours; 419 } 420 421 IPSTAT_INC(ips_total); 422 423 if (m->m_pkthdr.len < sizeof(struct ip)) 424 goto tooshort; 425 426 if (m->m_len < sizeof (struct ip) && 427 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 428 IPSTAT_INC(ips_toosmall); 429 return; 430 } 431 ip = mtod(m, struct ip *); 432 433 if (ip->ip_v != IPVERSION) { 434 IPSTAT_INC(ips_badvers); 435 goto bad; 436 } 437 438 hlen = ip->ip_hl << 2; 439 if (hlen < sizeof(struct ip)) { /* minimum header length */ 440 IPSTAT_INC(ips_badhlen); 441 goto bad; 442 } 443 if (hlen > m->m_len) { 444 if ((m = m_pullup(m, hlen)) == NULL) { 445 IPSTAT_INC(ips_badhlen); 446 return; 447 } 448 ip = mtod(m, struct ip *); 449 } 450 451 /* 127/8 must not appear on wire - RFC1122 */ 452 ifp = m->m_pkthdr.rcvif; 453 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 454 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 455 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 456 IPSTAT_INC(ips_badaddr); 457 goto bad; 458 } 459 } 460 461 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { 462 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); 463 } else { 464 if (hlen == sizeof(struct ip)) { 465 sum = in_cksum_hdr(ip); 466 } else { 467 sum = in_cksum(m, hlen); 468 } 469 } 470 if (sum) { 471 IPSTAT_INC(ips_badsum); 472 goto bad; 473 } 474 475#ifdef ALTQ 476 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) 477 /* packet is dropped by traffic conditioner */ 478 return; 479#endif 480 481 /* 482 * Convert fields to host representation. 483 */ 484 ip->ip_len = ntohs(ip->ip_len); 485 if (ip->ip_len < hlen) { 486 IPSTAT_INC(ips_badlen); 487 goto bad; 488 } 489 ip->ip_off = ntohs(ip->ip_off); 490 491 /* 492 * Check that the amount of data in the buffers 493 * is as at least much as the IP header would have us expect. 494 * Trim mbufs if longer than we expect. 495 * Drop packet if shorter than we expect. 496 */ 497 if (m->m_pkthdr.len < ip->ip_len) { 498tooshort: 499 IPSTAT_INC(ips_tooshort); 500 goto bad; 501 } 502 if (m->m_pkthdr.len > ip->ip_len) { 503 if (m->m_len == m->m_pkthdr.len) { 504 m->m_len = ip->ip_len; 505 m->m_pkthdr.len = ip->ip_len; 506 } else 507 m_adj(m, ip->ip_len - m->m_pkthdr.len); 508 } 509#ifdef IPSEC 510 /* 511 * Bypass packet filtering for packets from a tunnel (gif). 512 */ 513 if (ip_ipsec_filtertunnel(m)) 514 goto passin; 515#endif /* IPSEC */ 516 517 /* 518 * Run through list of hooks for input packets. 519 * 520 * NB: Beware of the destination address changing (e.g. 521 * by NAT rewriting). When this happens, tell 522 * ip_forward to do the right thing. 523 */ 524 525 /* Jump over all PFIL processing if hooks are not active. */ 526 if (!PFIL_HOOKED(&V_inet_pfil_hook)) 527 goto passin; 528 529 odst = ip->ip_dst; 530 if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0) 531 return; 532 if (m == NULL) /* consumed by filter */ 533 return; 534 535 ip = mtod(m, struct ip *); 536 dchg = (odst.s_addr != ip->ip_dst.s_addr); 537 ifp = m->m_pkthdr.rcvif; 538 539#ifdef IPFIREWALL_FORWARD 540 if (m->m_flags & M_FASTFWD_OURS) { 541 m->m_flags &= ~M_FASTFWD_OURS; 542 goto ours; 543 } 544 if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) { 545 /* 546 * Directly ship the packet on. This allows forwarding 547 * packets originally destined to us to some other directly 548 * connected host. 549 */ 550 ip_forward(m, dchg); 551 return; 552 } 553#endif /* IPFIREWALL_FORWARD */ 554 555passin: 556 /* 557 * Process options and, if not destined for us, 558 * ship it on. ip_dooptions returns 1 when an 559 * error was detected (causing an icmp message 560 * to be sent and the original packet to be freed). 561 */ 562 if (hlen > sizeof (struct ip) && ip_dooptions(m, 0)) 563 return; 564 565 /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no 566 * matter if it is destined to another node, or whether it is 567 * a multicast one, RSVP wants it! and prevents it from being forwarded 568 * anywhere else. Also checks if the rsvp daemon is running before 569 * grabbing the packet. 570 */ 571 if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP) 572 goto ours; 573 574 /* 575 * Check our list of addresses, to see if the packet is for us. 576 * If we don't have any addresses, assume any unicast packet 577 * we receive might be for us (and let the upper layers deal 578 * with it). 579 */ 580 if (TAILQ_EMPTY(&V_in_ifaddrhead) && 581 (m->m_flags & (M_MCAST|M_BCAST)) == 0) 582 goto ours; 583 584 /* 585 * Enable a consistency check between the destination address 586 * and the arrival interface for a unicast packet (the RFC 1122 587 * strong ES model) if IP forwarding is disabled and the packet 588 * is not locally generated and the packet is not subject to 589 * 'ipfw fwd'. 590 * 591 * XXX - Checking also should be disabled if the destination 592 * address is ipnat'ed to a different interface. 593 * 594 * XXX - Checking is incompatible with IP aliases added 595 * to the loopback interface instead of the interface where 596 * the packets are received. 597 * 598 * XXX - This is the case for carp vhost IPs as well so we 599 * insert a workaround. If the packet got here, we already 600 * checked with carp_iamatch() and carp_forus(). 601 */ 602 checkif = V_ip_checkinterface && (V_ipforwarding == 0) && 603 ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) && 604 ifp->if_carp == NULL && (dchg == 0); 605 606 /* 607 * Check for exact addresses in the hash bucket. 608 */ 609 /* IN_IFADDR_RLOCK(); */ 610 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { 611 /* 612 * If the address matches, verify that the packet 613 * arrived via the correct interface if checking is 614 * enabled. 615 */ 616 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr && 617 (!checkif || ia->ia_ifp == ifp)) { 618 ifa_ref(&ia->ia_ifa); 619 /* IN_IFADDR_RUNLOCK(); */ 620 goto ours; 621 } 622 } 623 /* IN_IFADDR_RUNLOCK(); */ 624 625 /* 626 * Check for broadcast addresses. 627 * 628 * Only accept broadcast packets that arrive via the matching 629 * interface. Reception of forwarded directed broadcasts would 630 * be handled via ip_forward() and ether_output() with the loopback 631 * into the stack for SIMPLEX interfaces handled by ether_output(). 632 */ 633 if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { 634 IF_ADDR_LOCK(ifp); 635 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 636 if (ifa->ifa_addr->sa_family != AF_INET) 637 continue; 638 ia = ifatoia(ifa); 639 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == 640 ip->ip_dst.s_addr) { 641 ifa_ref(ifa); 642 IF_ADDR_UNLOCK(ifp); 643 goto ours; 644 } 645 if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) { 646 ifa_ref(ifa); 647 IF_ADDR_UNLOCK(ifp); 648 goto ours; 649 } 650#ifdef BOOTP_COMPAT 651 if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) { 652 ifa_ref(ifa); 653 IF_ADDR_UNLOCK(ifp); 654 goto ours; 655 } 656#endif 657 } 658 IF_ADDR_UNLOCK(ifp); 659 ia = NULL; 660 } 661 /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */ 662 if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) { 663 IPSTAT_INC(ips_cantforward); 664 m_freem(m); 665 return; 666 } 667 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 668 if (V_ip_mrouter) { 669 /* 670 * If we are acting as a multicast router, all 671 * incoming multicast packets are passed to the 672 * kernel-level multicast forwarding function. 673 * The packet is returned (relatively) intact; if 674 * ip_mforward() returns a non-zero value, the packet 675 * must be discarded, else it may be accepted below. 676 */ 677 if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) { 678 IPSTAT_INC(ips_cantforward); 679 m_freem(m); 680 return; 681 } 682 683 /* 684 * The process-level routing daemon needs to receive 685 * all multicast IGMP packets, whether or not this 686 * host belongs to their destination groups. 687 */ 688 if (ip->ip_p == IPPROTO_IGMP) 689 goto ours; 690 IPSTAT_INC(ips_forward); 691 } 692 /* 693 * Assume the packet is for us, to avoid prematurely taking 694 * a lock on the in_multi hash. Protocols must perform 695 * their own filtering and update statistics accordingly. 696 */ 697 goto ours; 698 } 699 if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST) 700 goto ours; 701 if (ip->ip_dst.s_addr == INADDR_ANY) 702 goto ours; 703 704 /* 705 * FAITH(Firewall Aided Internet Translator) 706 */ 707 if (ifp && ifp->if_type == IFT_FAITH) { 708 if (V_ip_keepfaith) { 709 if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP) 710 goto ours; 711 } 712 m_freem(m); 713 return; 714 } 715 716 /* 717 * Not for us; forward if possible and desirable. 718 */ 719 if (V_ipforwarding == 0) { 720 IPSTAT_INC(ips_cantforward); 721 m_freem(m); 722 } else { 723#ifdef IPSEC 724 if (ip_ipsec_fwd(m)) 725 goto bad; 726#endif /* IPSEC */ 727 ip_forward(m, dchg); 728 } 729 return; 730 731ours: 732#ifdef IPSTEALTH 733 /* 734 * IPSTEALTH: Process non-routing options only 735 * if the packet is destined for us. 736 */ 737 if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) { 738 if (ia != NULL) 739 ifa_free(&ia->ia_ifa); 740 return; 741 } 742#endif /* IPSTEALTH */ 743 744 /* Count the packet in the ip address stats */ 745 if (ia != NULL) { 746 ia->ia_ifa.if_ipackets++; 747 ia->ia_ifa.if_ibytes += m->m_pkthdr.len; 748 ifa_free(&ia->ia_ifa); 749 } 750 751 /* 752 * Attempt reassembly; if it succeeds, proceed. 753 * ip_reass() will return a different mbuf. 754 */ 755 if (ip->ip_off & (IP_MF | IP_OFFMASK)) { 756 m = ip_reass(m); 757 if (m == NULL) 758 return; 759 ip = mtod(m, struct ip *); 760 /* Get the header length of the reassembled packet */ 761 hlen = ip->ip_hl << 2; 762 } 763 764 /* 765 * Further protocols expect the packet length to be w/o the 766 * IP header. 767 */ 768 ip->ip_len -= hlen; 769 770#ifdef IPSEC 771 /* 772 * enforce IPsec policy checking if we are seeing last header. 773 * note that we do not visit this with protocols with pcb layer 774 * code - like udp/tcp/raw ip. 775 */ 776 if (ip_ipsec_input(m)) 777 goto bad; 778#endif /* IPSEC */ 779 780 /* 781 * Switch out to protocol's input routine. 782 */ 783 IPSTAT_INC(ips_delivered); 784 785 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen); 786 return; 787bad: 788 m_freem(m); 789} 790 791/* 792 * After maxnipq has been updated, propagate the change to UMA. The UMA zone 793 * max has slightly different semantics than the sysctl, for historical 794 * reasons. 795 */ 796static void 797maxnipq_update(void) 798{ 799 800 /* 801 * -1 for unlimited allocation. 802 */ 803 if (V_maxnipq < 0) 804 uma_zone_set_max(V_ipq_zone, 0); 805 /* 806 * Positive number for specific bound. 807 */ 808 if (V_maxnipq > 0) 809 uma_zone_set_max(V_ipq_zone, V_maxnipq); 810 /* 811 * Zero specifies no further fragment queue allocation -- set the 812 * bound very low, but rely on implementation elsewhere to actually 813 * prevent allocation and reclaim current queues. 814 */ 815 if (V_maxnipq == 0) 816 uma_zone_set_max(V_ipq_zone, 1); 817} 818 819static void 820ipq_zone_change(void *tag) 821{ 822 823 if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) { 824 V_maxnipq = nmbclusters / 32; 825 maxnipq_update(); 826 } 827} 828 829static int 830sysctl_maxnipq(SYSCTL_HANDLER_ARGS) 831{ 832 int error, i; 833 834 i = V_maxnipq; 835 error = sysctl_handle_int(oidp, &i, 0, req); 836 if (error || !req->newptr) 837 return (error); 838 839 /* 840 * XXXRW: Might be a good idea to sanity check the argument and place 841 * an extreme upper bound. 842 */ 843 if (i < -1) 844 return (EINVAL); 845 V_maxnipq = i; 846 maxnipq_update(); 847 return (0); 848} 849 850SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW, 851 NULL, 0, sysctl_maxnipq, "I", 852 "Maximum number of IPv4 fragment reassembly queue entries"); 853 854/* 855 * Take incoming datagram fragment and try to reassemble it into 856 * whole datagram. If the argument is the first fragment or one 857 * in between the function will return NULL and store the mbuf 858 * in the fragment chain. If the argument is the last fragment 859 * the packet will be reassembled and the pointer to the new 860 * mbuf returned for further processing. Only m_tags attached 861 * to the first packet/fragment are preserved. 862 * The IP header is *NOT* adjusted out of iplen. 863 */ 864struct mbuf * 865ip_reass(struct mbuf *m) 866{ 867 struct ip *ip; 868 struct mbuf *p, *q, *nq, *t; 869 struct ipq *fp = NULL; 870 struct ipqhead *head; 871 int i, hlen, next; 872 u_int8_t ecn, ecn0; 873 u_short hash; 874 875 /* If maxnipq or maxfragsperpacket are 0, never accept fragments. */ 876 if (V_maxnipq == 0 || V_maxfragsperpacket == 0) { 877 IPSTAT_INC(ips_fragments); 878 IPSTAT_INC(ips_fragdropped); 879 m_freem(m); 880 return (NULL); 881 } 882 883 ip = mtod(m, struct ip *); 884 hlen = ip->ip_hl << 2; 885 886 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id); 887 head = &V_ipq[hash]; 888 IPQ_LOCK(); 889 890 /* 891 * Look for queue of fragments 892 * of this datagram. 893 */ 894 TAILQ_FOREACH(fp, head, ipq_list) 895 if (ip->ip_id == fp->ipq_id && 896 ip->ip_src.s_addr == fp->ipq_src.s_addr && 897 ip->ip_dst.s_addr == fp->ipq_dst.s_addr && 898#ifdef MAC 899 mac_ipq_match(m, fp) && 900#endif 901 ip->ip_p == fp->ipq_p) 902 goto found; 903 904 fp = NULL; 905 906 /* 907 * Attempt to trim the number of allocated fragment queues if it 908 * exceeds the administrative limit. 909 */ 910 if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) { 911 /* 912 * drop something from the tail of the current queue 913 * before proceeding further 914 */ 915 struct ipq *q = TAILQ_LAST(head, ipqhead); 916 if (q == NULL) { /* gak */ 917 for (i = 0; i < IPREASS_NHASH; i++) { 918 struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead); 919 if (r) { 920 IPSTAT_ADD(ips_fragtimeout, 921 r->ipq_nfrags); 922 ip_freef(&V_ipq[i], r); 923 break; 924 } 925 } 926 } else { 927 IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags); 928 ip_freef(head, q); 929 } 930 } 931 932found: 933 /* 934 * Adjust ip_len to not reflect header, 935 * convert offset of this to bytes. 936 */ 937 ip->ip_len -= hlen; 938 if (ip->ip_off & IP_MF) { 939 /* 940 * Make sure that fragments have a data length 941 * that's a non-zero multiple of 8 bytes. 942 */ 943 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) { 944 IPSTAT_INC(ips_toosmall); /* XXX */ 945 goto dropfrag; 946 } 947 m->m_flags |= M_FRAG; 948 } else 949 m->m_flags &= ~M_FRAG; 950 ip->ip_off <<= 3; 951 952 953 /* 954 * Attempt reassembly; if it succeeds, proceed. 955 * ip_reass() will return a different mbuf. 956 */ 957 IPSTAT_INC(ips_fragments); 958 m->m_pkthdr.header = ip; 959 960 /* Previous ip_reass() started here. */ 961 /* 962 * Presence of header sizes in mbufs 963 * would confuse code below. 964 */ 965 m->m_data += hlen; 966 m->m_len -= hlen; 967 968 /* 969 * If first fragment to arrive, create a reassembly queue. 970 */ 971 if (fp == NULL) { 972 fp = uma_zalloc(V_ipq_zone, M_NOWAIT); 973 if (fp == NULL) 974 goto dropfrag; 975#ifdef MAC 976 if (mac_ipq_init(fp, M_NOWAIT) != 0) { 977 uma_zfree(V_ipq_zone, fp); 978 fp = NULL; 979 goto dropfrag; 980 } 981 mac_ipq_create(m, fp); 982#endif 983 TAILQ_INSERT_HEAD(head, fp, ipq_list); 984 V_nipq++; 985 fp->ipq_nfrags = 1; 986 fp->ipq_ttl = IPFRAGTTL; 987 fp->ipq_p = ip->ip_p; 988 fp->ipq_id = ip->ip_id; 989 fp->ipq_src = ip->ip_src; 990 fp->ipq_dst = ip->ip_dst; 991 fp->ipq_frags = m; 992 m->m_nextpkt = NULL; 993 goto done; 994 } else { 995 fp->ipq_nfrags++; 996#ifdef MAC 997 mac_ipq_update(m, fp); 998#endif 999 } 1000 1001#define GETIP(m) ((struct ip*)((m)->m_pkthdr.header)) 1002 1003 /* 1004 * Handle ECN by comparing this segment with the first one; 1005 * if CE is set, do not lose CE. 1006 * drop if CE and not-ECT are mixed for the same packet. 1007 */ 1008 ecn = ip->ip_tos & IPTOS_ECN_MASK; 1009 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; 1010 if (ecn == IPTOS_ECN_CE) { 1011 if (ecn0 == IPTOS_ECN_NOTECT) 1012 goto dropfrag; 1013 if (ecn0 != IPTOS_ECN_CE) 1014 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; 1015 } 1016 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) 1017 goto dropfrag; 1018 1019 /* 1020 * Find a segment which begins after this one does. 1021 */ 1022 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) 1023 if (GETIP(q)->ip_off > ip->ip_off) 1024 break; 1025 1026 /* 1027 * If there is a preceding segment, it may provide some of 1028 * our data already. If so, drop the data from the incoming 1029 * segment. If it provides all of our data, drop us, otherwise 1030 * stick new segment in the proper place. 1031 * 1032 * If some of the data is dropped from the preceding 1033 * segment, then it's checksum is invalidated. 1034 */ 1035 if (p) { 1036 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off; 1037 if (i > 0) { 1038 if (i >= ip->ip_len) 1039 goto dropfrag; 1040 m_adj(m, i); 1041 m->m_pkthdr.csum_flags = 0; 1042 ip->ip_off += i; 1043 ip->ip_len -= i; 1044 } 1045 m->m_nextpkt = p->m_nextpkt; 1046 p->m_nextpkt = m; 1047 } else { 1048 m->m_nextpkt = fp->ipq_frags; 1049 fp->ipq_frags = m; 1050 } 1051 1052 /* 1053 * While we overlap succeeding segments trim them or, 1054 * if they are completely covered, dequeue them. 1055 */ 1056 for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off; 1057 q = nq) { 1058 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off; 1059 if (i < GETIP(q)->ip_len) { 1060 GETIP(q)->ip_len -= i; 1061 GETIP(q)->ip_off += i; 1062 m_adj(q, i); 1063 q->m_pkthdr.csum_flags = 0; 1064 break; 1065 } 1066 nq = q->m_nextpkt; 1067 m->m_nextpkt = nq; 1068 IPSTAT_INC(ips_fragdropped); 1069 fp->ipq_nfrags--; 1070 m_freem(q); 1071 } 1072 1073 /* 1074 * Check for complete reassembly and perform frag per packet 1075 * limiting. 1076 * 1077 * Frag limiting is performed here so that the nth frag has 1078 * a chance to complete the packet before we drop the packet. 1079 * As a result, n+1 frags are actually allowed per packet, but 1080 * only n will ever be stored. (n = maxfragsperpacket.) 1081 * 1082 */ 1083 next = 0; 1084 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { 1085 if (GETIP(q)->ip_off != next) { 1086 if (fp->ipq_nfrags > V_maxfragsperpacket) { 1087 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1088 ip_freef(head, fp); 1089 } 1090 goto done; 1091 } 1092 next += GETIP(q)->ip_len; 1093 } 1094 /* Make sure the last packet didn't have the IP_MF flag */ 1095 if (p->m_flags & M_FRAG) { 1096 if (fp->ipq_nfrags > V_maxfragsperpacket) { 1097 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1098 ip_freef(head, fp); 1099 } 1100 goto done; 1101 } 1102 1103 /* 1104 * Reassembly is complete. Make sure the packet is a sane size. 1105 */ 1106 q = fp->ipq_frags; 1107 ip = GETIP(q); 1108 if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { 1109 IPSTAT_INC(ips_toolong); 1110 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1111 ip_freef(head, fp); 1112 goto done; 1113 } 1114 1115 /* 1116 * Concatenate fragments. 1117 */ 1118 m = q; 1119 t = m->m_next; 1120 m->m_next = NULL; 1121 m_cat(m, t); 1122 nq = q->m_nextpkt; 1123 q->m_nextpkt = NULL; 1124 for (q = nq; q != NULL; q = nq) { 1125 nq = q->m_nextpkt; 1126 q->m_nextpkt = NULL; 1127 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; 1128 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; 1129 m_cat(m, q); 1130 } 1131 /* 1132 * In order to do checksumming faster we do 'end-around carry' here 1133 * (and not in for{} loop), though it implies we are not going to 1134 * reassemble more than 64k fragments. 1135 */ 1136 m->m_pkthdr.csum_data = 1137 (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16); 1138#ifdef MAC 1139 mac_ipq_reassemble(fp, m); 1140 mac_ipq_destroy(fp); 1141#endif 1142 1143 /* 1144 * Create header for new ip packet by modifying header of first 1145 * packet; dequeue and discard fragment reassembly header. 1146 * Make header visible. 1147 */ 1148 ip->ip_len = (ip->ip_hl << 2) + next; 1149 ip->ip_src = fp->ipq_src; 1150 ip->ip_dst = fp->ipq_dst; 1151 TAILQ_REMOVE(head, fp, ipq_list); 1152 V_nipq--; 1153 uma_zfree(V_ipq_zone, fp); 1154 m->m_len += (ip->ip_hl << 2); 1155 m->m_data -= (ip->ip_hl << 2); 1156 /* some debugging cruft by sklower, below, will go away soon */ 1157 if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ 1158 m_fixhdr(m); 1159 IPSTAT_INC(ips_reassembled); 1160 IPQ_UNLOCK(); 1161 return (m); 1162 1163dropfrag: 1164 IPSTAT_INC(ips_fragdropped); 1165 if (fp != NULL) 1166 fp->ipq_nfrags--; 1167 m_freem(m); 1168done: 1169 IPQ_UNLOCK(); 1170 return (NULL); 1171 1172#undef GETIP 1173} 1174 1175/* 1176 * Free a fragment reassembly header and all 1177 * associated datagrams. 1178 */ 1179static void 1180ip_freef(struct ipqhead *fhp, struct ipq *fp) 1181{ 1182 struct mbuf *q; 1183 1184 IPQ_LOCK_ASSERT(); 1185 1186 while (fp->ipq_frags) { 1187 q = fp->ipq_frags; 1188 fp->ipq_frags = q->m_nextpkt; 1189 m_freem(q); 1190 } 1191 TAILQ_REMOVE(fhp, fp, ipq_list); 1192 uma_zfree(V_ipq_zone, fp); 1193 V_nipq--; 1194} 1195 1196/* 1197 * IP timer processing; 1198 * if a timer expires on a reassembly 1199 * queue, discard it. 1200 */ 1201void 1202ip_slowtimo(void) 1203{ 1204 VNET_ITERATOR_DECL(vnet_iter); 1205 struct ipq *fp; 1206 int i; 1207 1208 VNET_LIST_RLOCK_NOSLEEP(); 1209 IPQ_LOCK(); 1210 VNET_FOREACH(vnet_iter) { 1211 CURVNET_SET(vnet_iter); 1212 for (i = 0; i < IPREASS_NHASH; i++) { 1213 for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) { 1214 struct ipq *fpp; 1215 1216 fpp = fp; 1217 fp = TAILQ_NEXT(fp, ipq_list); 1218 if(--fpp->ipq_ttl == 0) { 1219 IPSTAT_ADD(ips_fragtimeout, 1220 fpp->ipq_nfrags); 1221 ip_freef(&V_ipq[i], fpp); 1222 } 1223 } 1224 } 1225 /* 1226 * If we are over the maximum number of fragments 1227 * (due to the limit being lowered), drain off 1228 * enough to get down to the new limit. 1229 */ 1230 if (V_maxnipq >= 0 && V_nipq > V_maxnipq) { 1231 for (i = 0; i < IPREASS_NHASH; i++) { 1232 while (V_nipq > V_maxnipq && 1233 !TAILQ_EMPTY(&V_ipq[i])) { 1234 IPSTAT_ADD(ips_fragdropped, 1235 TAILQ_FIRST(&V_ipq[i])->ipq_nfrags); 1236 ip_freef(&V_ipq[i], 1237 TAILQ_FIRST(&V_ipq[i])); 1238 } 1239 } 1240 } 1241 CURVNET_RESTORE(); 1242 } 1243 IPQ_UNLOCK(); 1244 VNET_LIST_RUNLOCK_NOSLEEP(); 1245} 1246 1247/* 1248 * Drain off all datagram fragments. 1249 */ 1250static void 1251ip_drain_locked(void) 1252{ 1253 int i; 1254 1255 IPQ_LOCK_ASSERT(); 1256 1257 for (i = 0; i < IPREASS_NHASH; i++) { 1258 while(!TAILQ_EMPTY(&V_ipq[i])) { 1259 IPSTAT_ADD(ips_fragdropped, 1260 TAILQ_FIRST(&V_ipq[i])->ipq_nfrags); 1261 ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i])); 1262 } 1263 } 1264} 1265 1266void 1267ip_drain(void) 1268{ 1269 VNET_ITERATOR_DECL(vnet_iter); 1270 1271 VNET_LIST_RLOCK_NOSLEEP(); 1272 IPQ_LOCK(); 1273 VNET_FOREACH(vnet_iter) { 1274 CURVNET_SET(vnet_iter); 1275 ip_drain_locked(); 1276 CURVNET_RESTORE(); 1277 } 1278 IPQ_UNLOCK(); 1279 VNET_LIST_RUNLOCK_NOSLEEP(); 1280 in_rtqdrain(); 1281} 1282 1283/* 1284 * The protocol to be inserted into ip_protox[] must be already registered 1285 * in inetsw[], either statically or through pf_proto_register(). 1286 */ 1287int 1288ipproto_register(short ipproto) 1289{ 1290 struct protosw *pr; 1291 1292 /* Sanity checks. */ 1293 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1294 return (EPROTONOSUPPORT); 1295 1296 /* 1297 * The protocol slot must not be occupied by another protocol 1298 * already. An index pointing to IPPROTO_RAW is unused. 1299 */ 1300 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1301 if (pr == NULL) 1302 return (EPFNOSUPPORT); 1303 if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */ 1304 return (EEXIST); 1305 1306 /* Find the protocol position in inetsw[] and set the index. */ 1307 for (pr = inetdomain.dom_protosw; 1308 pr < inetdomain.dom_protoswNPROTOSW; pr++) { 1309 if (pr->pr_domain->dom_family == PF_INET && 1310 pr->pr_protocol && pr->pr_protocol == ipproto) { 1311 ip_protox[pr->pr_protocol] = pr - inetsw; 1312 return (0); 1313 } 1314 } 1315 return (EPROTONOSUPPORT); 1316} 1317 1318int 1319ipproto_unregister(short ipproto) 1320{ 1321 struct protosw *pr; 1322 1323 /* Sanity checks. */ 1324 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1325 return (EPROTONOSUPPORT); 1326 1327 /* Check if the protocol was indeed registered. */ 1328 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1329 if (pr == NULL) 1330 return (EPFNOSUPPORT); 1331 if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */ 1332 return (ENOENT); 1333 1334 /* Reset the protocol slot to IPPROTO_RAW. */ 1335 ip_protox[ipproto] = pr - inetsw; 1336 return (0); 1337} 1338 1339/* 1340 * Given address of next destination (final or next hop), return (referenced) 1341 * internet address info of interface to be used to get there. 1342 */ 1343struct in_ifaddr * 1344ip_rtaddr(struct in_addr dst, u_int fibnum) 1345{ 1346 struct route sro; 1347 struct sockaddr_in *sin; 1348 struct in_ifaddr *ia; 1349 1350 bzero(&sro, sizeof(sro)); 1351 sin = (struct sockaddr_in *)&sro.ro_dst; 1352 sin->sin_family = AF_INET; 1353 sin->sin_len = sizeof(*sin); 1354 sin->sin_addr = dst; 1355 in_rtalloc_ign(&sro, 0, fibnum); 1356 1357 if (sro.ro_rt == NULL) 1358 return (NULL); 1359 1360 ia = ifatoia(sro.ro_rt->rt_ifa); 1361 ifa_ref(&ia->ia_ifa); 1362 RTFREE(sro.ro_rt); 1363 return (ia); 1364} 1365 1366u_char inetctlerrmap[PRC_NCMDS] = { 1367 0, 0, 0, 0, 1368 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH, 1369 EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED, 1370 EMSGSIZE, EHOSTUNREACH, 0, 0, 1371 0, 0, EHOSTUNREACH, 0, 1372 ENOPROTOOPT, ECONNREFUSED 1373}; 1374 1375/* 1376 * Forward a packet. If some error occurs return the sender 1377 * an icmp packet. Note we can't always generate a meaningful 1378 * icmp message because icmp doesn't have a large enough repertoire 1379 * of codes and types. 1380 * 1381 * If not forwarding, just drop the packet. This could be confusing 1382 * if ipforwarding was zero but some routing protocol was advancing 1383 * us as a gateway to somewhere. However, we must let the routing 1384 * protocol deal with that. 1385 * 1386 * The srcrt parameter indicates whether the packet is being forwarded 1387 * via a source route. 1388 */ 1389void 1390ip_forward(struct mbuf *m, int srcrt) 1391{ 1392 struct ip *ip = mtod(m, struct ip *); 1393 struct in_ifaddr *ia; 1394 struct mbuf *mcopy; 1395 struct in_addr dest; 1396 struct route ro; 1397 int error, type = 0, code = 0, mtu = 0; 1398 1399 if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) { 1400 IPSTAT_INC(ips_cantforward); 1401 m_freem(m); 1402 return; 1403 } 1404#ifdef IPSTEALTH 1405 if (!V_ipstealth) { 1406#endif 1407 if (ip->ip_ttl <= IPTTLDEC) { 1408 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 1409 0, 0); 1410 return; 1411 } 1412#ifdef IPSTEALTH 1413 } 1414#endif 1415 1416 ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m)); 1417#ifndef IPSEC 1418 /* 1419 * 'ia' may be NULL if there is no route for this destination. 1420 * In case of IPsec, Don't discard it just yet, but pass it to 1421 * ip_output in case of outgoing IPsec policy. 1422 */ 1423 if (!srcrt && ia == NULL) { 1424 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 1425 return; 1426 } 1427#endif 1428 1429 /* 1430 * Save the IP header and at most 8 bytes of the payload, 1431 * in case we need to generate an ICMP message to the src. 1432 * 1433 * XXX this can be optimized a lot by saving the data in a local 1434 * buffer on the stack (72 bytes at most), and only allocating the 1435 * mbuf if really necessary. The vast majority of the packets 1436 * are forwarded without having to send an ICMP back (either 1437 * because unnecessary, or because rate limited), so we are 1438 * really we are wasting a lot of work here. 1439 * 1440 * We don't use m_copy() because it might return a reference 1441 * to a shared cluster. Both this function and ip_output() 1442 * assume exclusive access to the IP header in `m', so any 1443 * data in a cluster may change before we reach icmp_error(). 1444 */ 1445 MGETHDR(mcopy, M_DONTWAIT, m->m_type); 1446 if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) { 1447 /* 1448 * It's probably ok if the pkthdr dup fails (because 1449 * the deep copy of the tag chain failed), but for now 1450 * be conservative and just discard the copy since 1451 * code below may some day want the tags. 1452 */ 1453 m_free(mcopy); 1454 mcopy = NULL; 1455 } 1456 if (mcopy != NULL) { 1457 mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy)); 1458 mcopy->m_pkthdr.len = mcopy->m_len; 1459 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t)); 1460 } 1461 1462#ifdef IPSTEALTH 1463 if (!V_ipstealth) { 1464#endif 1465 ip->ip_ttl -= IPTTLDEC; 1466#ifdef IPSTEALTH 1467 } 1468#endif 1469 1470 /* 1471 * If forwarding packet using same interface that it came in on, 1472 * perhaps should send a redirect to sender to shortcut a hop. 1473 * Only send redirect if source is sending directly to us, 1474 * and if packet was not source routed (or has any options). 1475 * Also, don't send redirect if forwarding using a default route 1476 * or a route modified by a redirect. 1477 */ 1478 dest.s_addr = 0; 1479 if (!srcrt && V_ipsendredirects && 1480 ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) { 1481 struct sockaddr_in *sin; 1482 struct rtentry *rt; 1483 1484 bzero(&ro, sizeof(ro)); 1485 sin = (struct sockaddr_in *)&ro.ro_dst; 1486 sin->sin_family = AF_INET; 1487 sin->sin_len = sizeof(*sin); 1488 sin->sin_addr = ip->ip_dst; 1489 in_rtalloc_ign(&ro, 0, M_GETFIB(m)); 1490 1491 rt = ro.ro_rt; 1492 1493 if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 && 1494 satosin(rt_key(rt))->sin_addr.s_addr != 0) { 1495#define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa)) 1496 u_long src = ntohl(ip->ip_src.s_addr); 1497 1498 if (RTA(rt) && 1499 (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) { 1500 if (rt->rt_flags & RTF_GATEWAY) 1501 dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr; 1502 else 1503 dest.s_addr = ip->ip_dst.s_addr; 1504 /* Router requirements says to only send host redirects */ 1505 type = ICMP_REDIRECT; 1506 code = ICMP_REDIRECT_HOST; 1507 } 1508 } 1509 if (rt) 1510 RTFREE(rt); 1511 } 1512 1513 /* 1514 * Try to cache the route MTU from ip_output so we can consider it for 1515 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191. 1516 */ 1517 bzero(&ro, sizeof(ro)); 1518 1519 error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL); 1520 1521 if (error == EMSGSIZE && ro.ro_rt) 1522 mtu = ro.ro_rt->rt_rmx.rmx_mtu; 1523 if (ro.ro_rt) 1524 RTFREE(ro.ro_rt); 1525 1526 if (error) 1527 IPSTAT_INC(ips_cantforward); 1528 else { 1529 IPSTAT_INC(ips_forward); 1530 if (type) 1531 IPSTAT_INC(ips_redirectsent); 1532 else { 1533 if (mcopy) 1534 m_freem(mcopy); 1535 if (ia != NULL) 1536 ifa_free(&ia->ia_ifa); 1537 return; 1538 } 1539 } 1540 if (mcopy == NULL) { 1541 if (ia != NULL) 1542 ifa_free(&ia->ia_ifa); 1543 return; 1544 } 1545 1546 switch (error) { 1547 1548 case 0: /* forwarded, but need redirect */ 1549 /* type, code set above */ 1550 break; 1551 1552 case ENETUNREACH: 1553 case EHOSTUNREACH: 1554 case ENETDOWN: 1555 case EHOSTDOWN: 1556 default: 1557 type = ICMP_UNREACH; 1558 code = ICMP_UNREACH_HOST; 1559 break; 1560 1561 case EMSGSIZE: 1562 type = ICMP_UNREACH; 1563 code = ICMP_UNREACH_NEEDFRAG; 1564 1565#ifdef IPSEC 1566 /* 1567 * If IPsec is configured for this path, 1568 * override any possibly mtu value set by ip_output. 1569 */ 1570 mtu = ip_ipsec_mtu(mcopy, mtu); 1571#endif /* IPSEC */ 1572 /* 1573 * If the MTU was set before make sure we are below the 1574 * interface MTU. 1575 * If the MTU wasn't set before use the interface mtu or 1576 * fall back to the next smaller mtu step compared to the 1577 * current packet size. 1578 */ 1579 if (mtu != 0) { 1580 if (ia != NULL) 1581 mtu = min(mtu, ia->ia_ifp->if_mtu); 1582 } else { 1583 if (ia != NULL) 1584 mtu = ia->ia_ifp->if_mtu; 1585 else 1586 mtu = ip_next_mtu(ip->ip_len, 0); 1587 } 1588 IPSTAT_INC(ips_cantfrag); 1589 break; 1590 1591 case ENOBUFS: 1592 /* 1593 * A router should not generate ICMP_SOURCEQUENCH as 1594 * required in RFC1812 Requirements for IP Version 4 Routers. 1595 * Source quench could be a big problem under DoS attacks, 1596 * or if the underlying interface is rate-limited. 1597 * Those who need source quench packets may re-enable them 1598 * via the net.inet.ip.sendsourcequench sysctl. 1599 */ 1600 if (V_ip_sendsourcequench == 0) { 1601 m_freem(mcopy); 1602 if (ia != NULL) 1603 ifa_free(&ia->ia_ifa); 1604 return; 1605 } else { 1606 type = ICMP_SOURCEQUENCH; 1607 code = 0; 1608 } 1609 break; 1610 1611 case EACCES: /* ipfw denied packet */ 1612 m_freem(mcopy); 1613 if (ia != NULL) 1614 ifa_free(&ia->ia_ifa); 1615 return; 1616 } 1617 if (ia != NULL) 1618 ifa_free(&ia->ia_ifa); 1619 icmp_error(mcopy, type, code, dest.s_addr, mtu); 1620} 1621 1622void 1623ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip, 1624 struct mbuf *m) 1625{ 1626 1627 if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) { 1628 struct bintime bt; 1629 1630 bintime(&bt); 1631 if (inp->inp_socket->so_options & SO_BINTIME) { 1632 *mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt), 1633 SCM_BINTIME, SOL_SOCKET); 1634 if (*mp) 1635 mp = &(*mp)->m_next; 1636 } 1637 if (inp->inp_socket->so_options & SO_TIMESTAMP) { 1638 struct timeval tv; 1639 1640 bintime2timeval(&bt, &tv); 1641 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv), 1642 SCM_TIMESTAMP, SOL_SOCKET); 1643 if (*mp) 1644 mp = &(*mp)->m_next; 1645 } 1646 } 1647 if (inp->inp_flags & INP_RECVDSTADDR) { 1648 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst, 1649 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP); 1650 if (*mp) 1651 mp = &(*mp)->m_next; 1652 } 1653 if (inp->inp_flags & INP_RECVTTL) { 1654 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl, 1655 sizeof(u_char), IP_RECVTTL, IPPROTO_IP); 1656 if (*mp) 1657 mp = &(*mp)->m_next; 1658 } 1659#ifdef notyet 1660 /* XXX 1661 * Moving these out of udp_input() made them even more broken 1662 * than they already were. 1663 */ 1664 /* options were tossed already */ 1665 if (inp->inp_flags & INP_RECVOPTS) { 1666 *mp = sbcreatecontrol((caddr_t) opts_deleted_above, 1667 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP); 1668 if (*mp) 1669 mp = &(*mp)->m_next; 1670 } 1671 /* ip_srcroute doesn't do what we want here, need to fix */ 1672 if (inp->inp_flags & INP_RECVRETOPTS) { 1673 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m), 1674 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP); 1675 if (*mp) 1676 mp = &(*mp)->m_next; 1677 } 1678#endif 1679 if (inp->inp_flags & INP_RECVIF) { 1680 struct ifnet *ifp; 1681 struct sdlbuf { 1682 struct sockaddr_dl sdl; 1683 u_char pad[32]; 1684 } sdlbuf; 1685 struct sockaddr_dl *sdp; 1686 struct sockaddr_dl *sdl2 = &sdlbuf.sdl; 1687 1688 if (((ifp = m->m_pkthdr.rcvif)) 1689 && ( ifp->if_index && (ifp->if_index <= V_if_index))) { 1690 sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr; 1691 /* 1692 * Change our mind and don't try copy. 1693 */ 1694 if ((sdp->sdl_family != AF_LINK) 1695 || (sdp->sdl_len > sizeof(sdlbuf))) { 1696 goto makedummy; 1697 } 1698 bcopy(sdp, sdl2, sdp->sdl_len); 1699 } else { 1700makedummy: 1701 sdl2->sdl_len 1702 = offsetof(struct sockaddr_dl, sdl_data[0]); 1703 sdl2->sdl_family = AF_LINK; 1704 sdl2->sdl_index = 0; 1705 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0; 1706 } 1707 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len, 1708 IP_RECVIF, IPPROTO_IP); 1709 if (*mp) 1710 mp = &(*mp)->m_next; 1711 } 1712} 1713 1714/* 1715 * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the 1716 * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on 1717 * locking. This code remains in ip_input.c as ip_mroute.c is optionally 1718 * compiled. 1719 */ 1720static VNET_DEFINE(int, ip_rsvp_on); 1721VNET_DEFINE(struct socket *, ip_rsvpd); 1722 1723#define V_ip_rsvp_on VNET(ip_rsvp_on) 1724 1725int 1726ip_rsvp_init(struct socket *so) 1727{ 1728 1729 if (so->so_type != SOCK_RAW || 1730 so->so_proto->pr_protocol != IPPROTO_RSVP) 1731 return EOPNOTSUPP; 1732 1733 if (V_ip_rsvpd != NULL) 1734 return EADDRINUSE; 1735 1736 V_ip_rsvpd = so; 1737 /* 1738 * This may seem silly, but we need to be sure we don't over-increment 1739 * the RSVP counter, in case something slips up. 1740 */ 1741 if (!V_ip_rsvp_on) { 1742 V_ip_rsvp_on = 1; 1743 V_rsvp_on++; 1744 } 1745 1746 return 0; 1747} 1748 1749int 1750ip_rsvp_done(void) 1751{ 1752 1753 V_ip_rsvpd = NULL; 1754 /* 1755 * This may seem silly, but we need to be sure we don't over-decrement 1756 * the RSVP counter, in case something slips up. 1757 */ 1758 if (V_ip_rsvp_on) { 1759 V_ip_rsvp_on = 0; 1760 V_rsvp_on--; 1761 } 1762 return 0; 1763} 1764 1765void 1766rsvp_input(struct mbuf *m, int off) /* XXX must fixup manually */ 1767{ 1768 1769 if (rsvp_input_p) { /* call the real one if loaded */ 1770 rsvp_input_p(m, off); 1771 return; 1772 } 1773 1774 /* Can still get packets with rsvp_on = 0 if there is a local member 1775 * of the group to which the RSVP packet is addressed. But in this 1776 * case we want to throw the packet away. 1777 */ 1778 1779 if (!V_rsvp_on) { 1780 m_freem(m); 1781 return; 1782 } 1783 1784 if (V_ip_rsvpd != NULL) { 1785 rip_input(m, off); 1786 return; 1787 } 1788 /* Drop the packet */ 1789 m_freem(m); 1790} 1791