ip_input.c revision 263307
1219820Sjeff/*- 2219820Sjeff * Copyright (c) 1982, 1986, 1988, 1993 3219820Sjeff * The Regents of the University of California. All rights reserved. 4219820Sjeff * 5219820Sjeff * Redistribution and use in source and binary forms, with or without 6219820Sjeff * modification, are permitted provided that the following conditions 7219820Sjeff * are met: 8219820Sjeff * 1. Redistributions of source code must retain the above copyright 9219820Sjeff * notice, this list of conditions and the following disclaimer. 10219820Sjeff * 2. Redistributions in binary form must reproduce the above copyright 11219820Sjeff * notice, this list of conditions and the following disclaimer in the 12219820Sjeff * documentation and/or other materials provided with the distribution. 13219820Sjeff * 4. Neither the name of the University nor the names of its contributors 14219820Sjeff * may be used to endorse or promote products derived from this software 15219820Sjeff * without specific prior written permission. 16219820Sjeff * 17219820Sjeff * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18219820Sjeff * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19219820Sjeff * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20219820Sjeff * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21219820Sjeff * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22219820Sjeff * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23219820Sjeff * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24219820Sjeff * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25219820Sjeff * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26219820Sjeff * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27219820Sjeff * SUCH DAMAGE. 28219820Sjeff * 29219820Sjeff * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 30219820Sjeff */ 31219820Sjeff 32219820Sjeff#include <sys/cdefs.h> 33219820Sjeff__FBSDID("$FreeBSD: stable/10/sys/netinet/ip_input.c 263307 2014-03-18 16:56:05Z glebius $"); 34219820Sjeff 35219820Sjeff#include "opt_bootp.h" 36219820Sjeff#include "opt_ipfw.h" 37219820Sjeff#include "opt_ipstealth.h" 38219820Sjeff#include "opt_ipsec.h" 39219820Sjeff#include "opt_kdtrace.h" 40219820Sjeff#include "opt_route.h" 41219820Sjeff 42219820Sjeff#include <sys/param.h> 43219820Sjeff#include <sys/systm.h> 44219820Sjeff#include <sys/mbuf.h> 45219820Sjeff#include <sys/malloc.h> 46219820Sjeff#include <sys/domain.h> 47219820Sjeff#include <sys/protosw.h> 48219820Sjeff#include <sys/socket.h> 49219820Sjeff#include <sys/time.h> 50219820Sjeff#include <sys/kernel.h> 51219820Sjeff#include <sys/lock.h> 52219820Sjeff#include <sys/rwlock.h> 53219820Sjeff#include <sys/sdt.h> 54219820Sjeff#include <sys/syslog.h> 55219820Sjeff#include <sys/sysctl.h> 56219820Sjeff 57219820Sjeff#include <net/pfil.h> 58219820Sjeff#include <net/if.h> 59219820Sjeff#include <net/if_types.h> 60219820Sjeff#include <net/if_var.h> 61219820Sjeff#include <net/if_dl.h> 62219820Sjeff#include <net/route.h> 63219820Sjeff#include <net/netisr.h> 64219820Sjeff#include <net/vnet.h> 65219820Sjeff 66219820Sjeff#include <netinet/in.h> 67219820Sjeff#include <netinet/in_kdtrace.h> 68219820Sjeff#include <netinet/in_systm.h> 69219820Sjeff#include <netinet/in_var.h> 70219820Sjeff#include <netinet/ip.h> 71219820Sjeff#include <netinet/in_pcb.h> 72219820Sjeff#include <netinet/ip_var.h> 73219820Sjeff#include <netinet/ip_fw.h> 74219820Sjeff#include <netinet/ip_icmp.h> 75219820Sjeff#include <netinet/ip_options.h> 76219820Sjeff#include <machine/in_cksum.h> 77219820Sjeff#include <netinet/ip_carp.h> 78219820Sjeff#ifdef IPSEC 79219820Sjeff#include <netinet/ip_ipsec.h> 80219820Sjeff#endif /* IPSEC */ 81219820Sjeff 82219820Sjeff#include <sys/socketvar.h> 83219820Sjeff 84219820Sjeff#include <security/mac/mac_framework.h> 85219820Sjeff 86219820Sjeff#ifdef CTASSERT 87219820SjeffCTASSERT(sizeof(struct ip) == 20); 88219820Sjeff#endif 89219820Sjeff 90219820Sjeffstruct rwlock in_ifaddr_lock; 91RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock"); 92 93VNET_DEFINE(int, rsvp_on); 94 95VNET_DEFINE(int, ipforwarding); 96SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW, 97 &VNET_NAME(ipforwarding), 0, 98 "Enable IP forwarding between interfaces"); 99 100static VNET_DEFINE(int, ipsendredirects) = 1; /* XXX */ 101#define V_ipsendredirects VNET(ipsendredirects) 102SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW, 103 &VNET_NAME(ipsendredirects), 0, 104 "Enable sending IP redirects"); 105 106static VNET_DEFINE(int, ip_keepfaith); 107#define V_ip_keepfaith VNET(ip_keepfaith) 108SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW, 109 &VNET_NAME(ip_keepfaith), 0, 110 "Enable packet capture for FAITH IPv4->IPv6 translater daemon"); 111 112static VNET_DEFINE(int, ip_sendsourcequench); 113#define V_ip_sendsourcequench VNET(ip_sendsourcequench) 114SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW, 115 &VNET_NAME(ip_sendsourcequench), 0, 116 "Enable the transmission of source quench packets"); 117 118VNET_DEFINE(int, ip_do_randomid); 119SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW, 120 &VNET_NAME(ip_do_randomid), 0, 121 "Assign random ip_id values"); 122 123/* 124 * XXX - Setting ip_checkinterface mostly implements the receive side of 125 * the Strong ES model described in RFC 1122, but since the routing table 126 * and transmit implementation do not implement the Strong ES model, 127 * setting this to 1 results in an odd hybrid. 128 * 129 * XXX - ip_checkinterface currently must be disabled if you use ipnat 130 * to translate the destination address to another local interface. 131 * 132 * XXX - ip_checkinterface must be disabled if you add IP aliases 133 * to the loopback interface instead of the interface where the 134 * packets for those addresses are received. 135 */ 136static VNET_DEFINE(int, ip_checkinterface); 137#define V_ip_checkinterface VNET(ip_checkinterface) 138SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW, 139 &VNET_NAME(ip_checkinterface), 0, 140 "Verify packet arrives on correct interface"); 141 142VNET_DEFINE(struct pfil_head, inet_pfil_hook); /* Packet filter hooks */ 143 144static struct netisr_handler ip_nh = { 145 .nh_name = "ip", 146 .nh_handler = ip_input, 147 .nh_proto = NETISR_IP, 148 .nh_policy = NETISR_POLICY_FLOW, 149}; 150 151extern struct domain inetdomain; 152extern struct protosw inetsw[]; 153u_char ip_protox[IPPROTO_MAX]; 154VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead); /* first inet address */ 155VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table */ 156VNET_DEFINE(u_long, in_ifaddrhmask); /* mask for hash table */ 157 158static VNET_DEFINE(uma_zone_t, ipq_zone); 159static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]); 160static struct mtx ipqlock; 161 162#define V_ipq_zone VNET(ipq_zone) 163#define V_ipq VNET(ipq) 164 165#define IPQ_LOCK() mtx_lock(&ipqlock) 166#define IPQ_UNLOCK() mtx_unlock(&ipqlock) 167#define IPQ_LOCK_INIT() mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF) 168#define IPQ_LOCK_ASSERT() mtx_assert(&ipqlock, MA_OWNED) 169 170static void maxnipq_update(void); 171static void ipq_zone_change(void *); 172static void ip_drain_locked(void); 173 174static VNET_DEFINE(int, maxnipq); /* Administrative limit on # reass queues. */ 175static VNET_DEFINE(int, nipq); /* Total # of reass queues */ 176#define V_maxnipq VNET(maxnipq) 177#define V_nipq VNET(nipq) 178SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD, 179 &VNET_NAME(nipq), 0, 180 "Current number of IPv4 fragment reassembly queue entries"); 181 182static VNET_DEFINE(int, maxfragsperpacket); 183#define V_maxfragsperpacket VNET(maxfragsperpacket) 184SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW, 185 &VNET_NAME(maxfragsperpacket), 0, 186 "Maximum number of IPv4 fragments allowed per packet"); 187 188#ifdef IPCTL_DEFMTU 189SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW, 190 &ip_mtu, 0, "Default MTU"); 191#endif 192 193#ifdef IPSTEALTH 194VNET_DEFINE(int, ipstealth); 195SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, 196 &VNET_NAME(ipstealth), 0, 197 "IP stealth mode, no TTL decrementation on forwarding"); 198#endif 199 200static void ip_freef(struct ipqhead *, struct ipq *); 201 202/* 203 * IP statistics are stored in the "array" of counter(9)s. 204 */ 205VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat); 206VNET_PCPUSTAT_SYSINIT(ipstat); 207SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat, 208 "IP statistics (struct ipstat, netinet/ip_var.h)"); 209 210#ifdef VIMAGE 211VNET_PCPUSTAT_SYSUNINIT(ipstat); 212#endif /* VIMAGE */ 213 214/* 215 * Kernel module interface for updating ipstat. The argument is an index 216 * into ipstat treated as an array. 217 */ 218void 219kmod_ipstat_inc(int statnum) 220{ 221 222 counter_u64_add(VNET(ipstat)[statnum], 1); 223} 224 225void 226kmod_ipstat_dec(int statnum) 227{ 228 229 counter_u64_add(VNET(ipstat)[statnum], -1); 230} 231 232static int 233sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS) 234{ 235 int error, qlimit; 236 237 netisr_getqlimit(&ip_nh, &qlimit); 238 error = sysctl_handle_int(oidp, &qlimit, 0, req); 239 if (error || !req->newptr) 240 return (error); 241 if (qlimit < 1) 242 return (EINVAL); 243 return (netisr_setqlimit(&ip_nh, qlimit)); 244} 245SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, 246 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I", 247 "Maximum size of the IP input queue"); 248 249static int 250sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS) 251{ 252 u_int64_t qdrops_long; 253 int error, qdrops; 254 255 netisr_getqdrops(&ip_nh, &qdrops_long); 256 qdrops = qdrops_long; 257 error = sysctl_handle_int(oidp, &qdrops, 0, req); 258 if (error || !req->newptr) 259 return (error); 260 if (qdrops != 0) 261 return (EINVAL); 262 netisr_clearqdrops(&ip_nh); 263 return (0); 264} 265 266SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, 267 CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I", 268 "Number of packets dropped from the IP input queue"); 269 270/* 271 * IP initialization: fill in IP protocol switch table. 272 * All protocols not implemented in kernel go to raw IP protocol handler. 273 */ 274void 275ip_init(void) 276{ 277 struct protosw *pr; 278 int i; 279 280 V_ip_id = time_second & 0xffff; 281 282 TAILQ_INIT(&V_in_ifaddrhead); 283 V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask); 284 285 /* Initialize IP reassembly queue. */ 286 for (i = 0; i < IPREASS_NHASH; i++) 287 TAILQ_INIT(&V_ipq[i]); 288 V_maxnipq = nmbclusters / 32; 289 V_maxfragsperpacket = 16; 290 V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, 291 NULL, UMA_ALIGN_PTR, 0); 292 maxnipq_update(); 293 294 /* Initialize packet filter hooks. */ 295 V_inet_pfil_hook.ph_type = PFIL_TYPE_AF; 296 V_inet_pfil_hook.ph_af = AF_INET; 297 if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0) 298 printf("%s: WARNING: unable to register pfil hook, " 299 "error %d\n", __func__, i); 300 301 /* Skip initialization of globals for non-default instances. */ 302 if (!IS_DEFAULT_VNET(curvnet)) 303 return; 304 305 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 306 if (pr == NULL) 307 panic("ip_init: PF_INET not found"); 308 309 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */ 310 for (i = 0; i < IPPROTO_MAX; i++) 311 ip_protox[i] = pr - inetsw; 312 /* 313 * Cycle through IP protocols and put them into the appropriate place 314 * in ip_protox[]. 315 */ 316 for (pr = inetdomain.dom_protosw; 317 pr < inetdomain.dom_protoswNPROTOSW; pr++) 318 if (pr->pr_domain->dom_family == PF_INET && 319 pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) { 320 /* Be careful to only index valid IP protocols. */ 321 if (pr->pr_protocol < IPPROTO_MAX) 322 ip_protox[pr->pr_protocol] = pr - inetsw; 323 } 324 325 EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change, 326 NULL, EVENTHANDLER_PRI_ANY); 327 328 /* Initialize various other remaining things. */ 329 IPQ_LOCK_INIT(); 330 netisr_register(&ip_nh); 331} 332 333#ifdef VIMAGE 334void 335ip_destroy(void) 336{ 337 int i; 338 339 if ((i = pfil_head_unregister(&V_inet_pfil_hook)) != 0) 340 printf("%s: WARNING: unable to unregister pfil hook, " 341 "error %d\n", __func__, i); 342 343 /* Cleanup in_ifaddr hash table; should be empty. */ 344 hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask); 345 346 IPQ_LOCK(); 347 ip_drain_locked(); 348 IPQ_UNLOCK(); 349 350 uma_zdestroy(V_ipq_zone); 351} 352#endif 353 354/* 355 * Ip input routine. Checksum and byte swap header. If fragmented 356 * try to reassemble. Process options. Pass to next level. 357 */ 358void 359ip_input(struct mbuf *m) 360{ 361 struct ip *ip = NULL; 362 struct in_ifaddr *ia = NULL; 363 struct ifaddr *ifa; 364 struct ifnet *ifp; 365 int checkif, hlen = 0; 366 uint16_t sum, ip_len; 367 int dchg = 0; /* dest changed after fw */ 368 struct in_addr odst; /* original dst address */ 369 370 M_ASSERTPKTHDR(m); 371 372 if (m->m_flags & M_FASTFWD_OURS) { 373 m->m_flags &= ~M_FASTFWD_OURS; 374 /* Set up some basics that will be used later. */ 375 ip = mtod(m, struct ip *); 376 hlen = ip->ip_hl << 2; 377 ip_len = ntohs(ip->ip_len); 378 goto ours; 379 } 380 381 IPSTAT_INC(ips_total); 382 383 if (m->m_pkthdr.len < sizeof(struct ip)) 384 goto tooshort; 385 386 if (m->m_len < sizeof (struct ip) && 387 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 388 IPSTAT_INC(ips_toosmall); 389 return; 390 } 391 ip = mtod(m, struct ip *); 392 393 if (ip->ip_v != IPVERSION) { 394 IPSTAT_INC(ips_badvers); 395 goto bad; 396 } 397 398 hlen = ip->ip_hl << 2; 399 if (hlen < sizeof(struct ip)) { /* minimum header length */ 400 IPSTAT_INC(ips_badhlen); 401 goto bad; 402 } 403 if (hlen > m->m_len) { 404 if ((m = m_pullup(m, hlen)) == NULL) { 405 IPSTAT_INC(ips_badhlen); 406 return; 407 } 408 ip = mtod(m, struct ip *); 409 } 410 411 IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL); 412 413 /* 127/8 must not appear on wire - RFC1122 */ 414 ifp = m->m_pkthdr.rcvif; 415 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 416 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 417 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 418 IPSTAT_INC(ips_badaddr); 419 goto bad; 420 } 421 } 422 423 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { 424 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); 425 } else { 426 if (hlen == sizeof(struct ip)) { 427 sum = in_cksum_hdr(ip); 428 } else { 429 sum = in_cksum(m, hlen); 430 } 431 } 432 if (sum) { 433 IPSTAT_INC(ips_badsum); 434 goto bad; 435 } 436 437#ifdef ALTQ 438 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) 439 /* packet is dropped by traffic conditioner */ 440 return; 441#endif 442 443 ip_len = ntohs(ip->ip_len); 444 if (ip_len < hlen) { 445 IPSTAT_INC(ips_badlen); 446 goto bad; 447 } 448 449 /* 450 * Check that the amount of data in the buffers 451 * is as at least much as the IP header would have us expect. 452 * Trim mbufs if longer than we expect. 453 * Drop packet if shorter than we expect. 454 */ 455 if (m->m_pkthdr.len < ip_len) { 456tooshort: 457 IPSTAT_INC(ips_tooshort); 458 goto bad; 459 } 460 if (m->m_pkthdr.len > ip_len) { 461 if (m->m_len == m->m_pkthdr.len) { 462 m->m_len = ip_len; 463 m->m_pkthdr.len = ip_len; 464 } else 465 m_adj(m, ip_len - m->m_pkthdr.len); 466 } 467#ifdef IPSEC 468 /* 469 * Bypass packet filtering for packets previously handled by IPsec. 470 */ 471 if (ip_ipsec_filtertunnel(m)) 472 goto passin; 473#endif /* IPSEC */ 474 475 /* 476 * Run through list of hooks for input packets. 477 * 478 * NB: Beware of the destination address changing (e.g. 479 * by NAT rewriting). When this happens, tell 480 * ip_forward to do the right thing. 481 */ 482 483 /* Jump over all PFIL processing if hooks are not active. */ 484 if (!PFIL_HOOKED(&V_inet_pfil_hook)) 485 goto passin; 486 487 odst = ip->ip_dst; 488 if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0) 489 return; 490 if (m == NULL) /* consumed by filter */ 491 return; 492 493 ip = mtod(m, struct ip *); 494 dchg = (odst.s_addr != ip->ip_dst.s_addr); 495 ifp = m->m_pkthdr.rcvif; 496 497 if (m->m_flags & M_FASTFWD_OURS) { 498 m->m_flags &= ~M_FASTFWD_OURS; 499 goto ours; 500 } 501 if (m->m_flags & M_IP_NEXTHOP) { 502 dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL); 503 if (dchg != 0) { 504 /* 505 * Directly ship the packet on. This allows 506 * forwarding packets originally destined to us 507 * to some other directly connected host. 508 */ 509 ip_forward(m, 1); 510 return; 511 } 512 } 513passin: 514 515 /* 516 * Process options and, if not destined for us, 517 * ship it on. ip_dooptions returns 1 when an 518 * error was detected (causing an icmp message 519 * to be sent and the original packet to be freed). 520 */ 521 if (hlen > sizeof (struct ip) && ip_dooptions(m, 0)) 522 return; 523 524 /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no 525 * matter if it is destined to another node, or whether it is 526 * a multicast one, RSVP wants it! and prevents it from being forwarded 527 * anywhere else. Also checks if the rsvp daemon is running before 528 * grabbing the packet. 529 */ 530 if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP) 531 goto ours; 532 533 /* 534 * Check our list of addresses, to see if the packet is for us. 535 * If we don't have any addresses, assume any unicast packet 536 * we receive might be for us (and let the upper layers deal 537 * with it). 538 */ 539 if (TAILQ_EMPTY(&V_in_ifaddrhead) && 540 (m->m_flags & (M_MCAST|M_BCAST)) == 0) 541 goto ours; 542 543 /* 544 * Enable a consistency check between the destination address 545 * and the arrival interface for a unicast packet (the RFC 1122 546 * strong ES model) if IP forwarding is disabled and the packet 547 * is not locally generated and the packet is not subject to 548 * 'ipfw fwd'. 549 * 550 * XXX - Checking also should be disabled if the destination 551 * address is ipnat'ed to a different interface. 552 * 553 * XXX - Checking is incompatible with IP aliases added 554 * to the loopback interface instead of the interface where 555 * the packets are received. 556 * 557 * XXX - This is the case for carp vhost IPs as well so we 558 * insert a workaround. If the packet got here, we already 559 * checked with carp_iamatch() and carp_forus(). 560 */ 561 checkif = V_ip_checkinterface && (V_ipforwarding == 0) && 562 ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) && 563 ifp->if_carp == NULL && (dchg == 0); 564 565 /* 566 * Check for exact addresses in the hash bucket. 567 */ 568 /* IN_IFADDR_RLOCK(); */ 569 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { 570 /* 571 * If the address matches, verify that the packet 572 * arrived via the correct interface if checking is 573 * enabled. 574 */ 575 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr && 576 (!checkif || ia->ia_ifp == ifp)) { 577 ifa_ref(&ia->ia_ifa); 578 /* IN_IFADDR_RUNLOCK(); */ 579 goto ours; 580 } 581 } 582 /* IN_IFADDR_RUNLOCK(); */ 583 584 /* 585 * Check for broadcast addresses. 586 * 587 * Only accept broadcast packets that arrive via the matching 588 * interface. Reception of forwarded directed broadcasts would 589 * be handled via ip_forward() and ether_output() with the loopback 590 * into the stack for SIMPLEX interfaces handled by ether_output(). 591 */ 592 if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { 593 IF_ADDR_RLOCK(ifp); 594 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 595 if (ifa->ifa_addr->sa_family != AF_INET) 596 continue; 597 ia = ifatoia(ifa); 598 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == 599 ip->ip_dst.s_addr) { 600 ifa_ref(ifa); 601 IF_ADDR_RUNLOCK(ifp); 602 goto ours; 603 } 604#ifdef BOOTP_COMPAT 605 if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) { 606 ifa_ref(ifa); 607 IF_ADDR_RUNLOCK(ifp); 608 goto ours; 609 } 610#endif 611 } 612 IF_ADDR_RUNLOCK(ifp); 613 ia = NULL; 614 } 615 /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */ 616 if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) { 617 IPSTAT_INC(ips_cantforward); 618 m_freem(m); 619 return; 620 } 621 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 622 if (V_ip_mrouter) { 623 /* 624 * If we are acting as a multicast router, all 625 * incoming multicast packets are passed to the 626 * kernel-level multicast forwarding function. 627 * The packet is returned (relatively) intact; if 628 * ip_mforward() returns a non-zero value, the packet 629 * must be discarded, else it may be accepted below. 630 */ 631 if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) { 632 IPSTAT_INC(ips_cantforward); 633 m_freem(m); 634 return; 635 } 636 637 /* 638 * The process-level routing daemon needs to receive 639 * all multicast IGMP packets, whether or not this 640 * host belongs to their destination groups. 641 */ 642 if (ip->ip_p == IPPROTO_IGMP) 643 goto ours; 644 IPSTAT_INC(ips_forward); 645 } 646 /* 647 * Assume the packet is for us, to avoid prematurely taking 648 * a lock on the in_multi hash. Protocols must perform 649 * their own filtering and update statistics accordingly. 650 */ 651 goto ours; 652 } 653 if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST) 654 goto ours; 655 if (ip->ip_dst.s_addr == INADDR_ANY) 656 goto ours; 657 658 /* 659 * FAITH(Firewall Aided Internet Translator) 660 */ 661 if (ifp && ifp->if_type == IFT_FAITH) { 662 if (V_ip_keepfaith) { 663 if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP) 664 goto ours; 665 } 666 m_freem(m); 667 return; 668 } 669 670 /* 671 * Not for us; forward if possible and desirable. 672 */ 673 if (V_ipforwarding == 0) { 674 IPSTAT_INC(ips_cantforward); 675 m_freem(m); 676 } else { 677#ifdef IPSEC 678 if (ip_ipsec_fwd(m)) 679 goto bad; 680#endif /* IPSEC */ 681 ip_forward(m, dchg); 682 } 683 return; 684 685ours: 686#ifdef IPSTEALTH 687 /* 688 * IPSTEALTH: Process non-routing options only 689 * if the packet is destined for us. 690 */ 691 if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) { 692 if (ia != NULL) 693 ifa_free(&ia->ia_ifa); 694 return; 695 } 696#endif /* IPSTEALTH */ 697 698 /* Count the packet in the ip address stats */ 699 if (ia != NULL) { 700 ia->ia_ifa.if_ipackets++; 701 ia->ia_ifa.if_ibytes += m->m_pkthdr.len; 702 ifa_free(&ia->ia_ifa); 703 } 704 705 /* 706 * Attempt reassembly; if it succeeds, proceed. 707 * ip_reass() will return a different mbuf. 708 */ 709 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) { 710 /* XXXGL: shouldn't we save & set m_flags? */ 711 m = ip_reass(m); 712 if (m == NULL) 713 return; 714 ip = mtod(m, struct ip *); 715 /* Get the header length of the reassembled packet */ 716 hlen = ip->ip_hl << 2; 717 } 718 719#ifdef IPSEC 720 /* 721 * enforce IPsec policy checking if we are seeing last header. 722 * note that we do not visit this with protocols with pcb layer 723 * code - like udp/tcp/raw ip. 724 */ 725 if (ip_ipsec_input(m)) 726 goto bad; 727#endif /* IPSEC */ 728 729 /* 730 * Switch out to protocol's input routine. 731 */ 732 IPSTAT_INC(ips_delivered); 733 734 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen); 735 return; 736bad: 737 m_freem(m); 738} 739 740/* 741 * After maxnipq has been updated, propagate the change to UMA. The UMA zone 742 * max has slightly different semantics than the sysctl, for historical 743 * reasons. 744 */ 745static void 746maxnipq_update(void) 747{ 748 749 /* 750 * -1 for unlimited allocation. 751 */ 752 if (V_maxnipq < 0) 753 uma_zone_set_max(V_ipq_zone, 0); 754 /* 755 * Positive number for specific bound. 756 */ 757 if (V_maxnipq > 0) 758 uma_zone_set_max(V_ipq_zone, V_maxnipq); 759 /* 760 * Zero specifies no further fragment queue allocation -- set the 761 * bound very low, but rely on implementation elsewhere to actually 762 * prevent allocation and reclaim current queues. 763 */ 764 if (V_maxnipq == 0) 765 uma_zone_set_max(V_ipq_zone, 1); 766} 767 768static void 769ipq_zone_change(void *tag) 770{ 771 772 if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) { 773 V_maxnipq = nmbclusters / 32; 774 maxnipq_update(); 775 } 776} 777 778static int 779sysctl_maxnipq(SYSCTL_HANDLER_ARGS) 780{ 781 int error, i; 782 783 i = V_maxnipq; 784 error = sysctl_handle_int(oidp, &i, 0, req); 785 if (error || !req->newptr) 786 return (error); 787 788 /* 789 * XXXRW: Might be a good idea to sanity check the argument and place 790 * an extreme upper bound. 791 */ 792 if (i < -1) 793 return (EINVAL); 794 V_maxnipq = i; 795 maxnipq_update(); 796 return (0); 797} 798 799SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW, 800 NULL, 0, sysctl_maxnipq, "I", 801 "Maximum number of IPv4 fragment reassembly queue entries"); 802 803#define M_IP_FRAG M_PROTO9 804 805/* 806 * Take incoming datagram fragment and try to reassemble it into 807 * whole datagram. If the argument is the first fragment or one 808 * in between the function will return NULL and store the mbuf 809 * in the fragment chain. If the argument is the last fragment 810 * the packet will be reassembled and the pointer to the new 811 * mbuf returned for further processing. Only m_tags attached 812 * to the first packet/fragment are preserved. 813 * The IP header is *NOT* adjusted out of iplen. 814 */ 815struct mbuf * 816ip_reass(struct mbuf *m) 817{ 818 struct ip *ip; 819 struct mbuf *p, *q, *nq, *t; 820 struct ipq *fp = NULL; 821 struct ipqhead *head; 822 int i, hlen, next; 823 u_int8_t ecn, ecn0; 824 u_short hash; 825 826 /* If maxnipq or maxfragsperpacket are 0, never accept fragments. */ 827 if (V_maxnipq == 0 || V_maxfragsperpacket == 0) { 828 IPSTAT_INC(ips_fragments); 829 IPSTAT_INC(ips_fragdropped); 830 m_freem(m); 831 return (NULL); 832 } 833 834 ip = mtod(m, struct ip *); 835 hlen = ip->ip_hl << 2; 836 837 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id); 838 head = &V_ipq[hash]; 839 IPQ_LOCK(); 840 841 /* 842 * Look for queue of fragments 843 * of this datagram. 844 */ 845 TAILQ_FOREACH(fp, head, ipq_list) 846 if (ip->ip_id == fp->ipq_id && 847 ip->ip_src.s_addr == fp->ipq_src.s_addr && 848 ip->ip_dst.s_addr == fp->ipq_dst.s_addr && 849#ifdef MAC 850 mac_ipq_match(m, fp) && 851#endif 852 ip->ip_p == fp->ipq_p) 853 goto found; 854 855 fp = NULL; 856 857 /* 858 * Attempt to trim the number of allocated fragment queues if it 859 * exceeds the administrative limit. 860 */ 861 if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) { 862 /* 863 * drop something from the tail of the current queue 864 * before proceeding further 865 */ 866 struct ipq *q = TAILQ_LAST(head, ipqhead); 867 if (q == NULL) { /* gak */ 868 for (i = 0; i < IPREASS_NHASH; i++) { 869 struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead); 870 if (r) { 871 IPSTAT_ADD(ips_fragtimeout, 872 r->ipq_nfrags); 873 ip_freef(&V_ipq[i], r); 874 break; 875 } 876 } 877 } else { 878 IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags); 879 ip_freef(head, q); 880 } 881 } 882 883found: 884 /* 885 * Adjust ip_len to not reflect header, 886 * convert offset of this to bytes. 887 */ 888 ip->ip_len = htons(ntohs(ip->ip_len) - hlen); 889 if (ip->ip_off & htons(IP_MF)) { 890 /* 891 * Make sure that fragments have a data length 892 * that's a non-zero multiple of 8 bytes. 893 */ 894 if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) { 895 IPSTAT_INC(ips_toosmall); /* XXX */ 896 goto dropfrag; 897 } 898 m->m_flags |= M_IP_FRAG; 899 } else 900 m->m_flags &= ~M_IP_FRAG; 901 ip->ip_off = htons(ntohs(ip->ip_off) << 3); 902 903 /* 904 * Attempt reassembly; if it succeeds, proceed. 905 * ip_reass() will return a different mbuf. 906 */ 907 IPSTAT_INC(ips_fragments); 908 m->m_pkthdr.PH_loc.ptr = ip; 909 910 /* Previous ip_reass() started here. */ 911 /* 912 * Presence of header sizes in mbufs 913 * would confuse code below. 914 */ 915 m->m_data += hlen; 916 m->m_len -= hlen; 917 918 /* 919 * If first fragment to arrive, create a reassembly queue. 920 */ 921 if (fp == NULL) { 922 fp = uma_zalloc(V_ipq_zone, M_NOWAIT); 923 if (fp == NULL) 924 goto dropfrag; 925#ifdef MAC 926 if (mac_ipq_init(fp, M_NOWAIT) != 0) { 927 uma_zfree(V_ipq_zone, fp); 928 fp = NULL; 929 goto dropfrag; 930 } 931 mac_ipq_create(m, fp); 932#endif 933 TAILQ_INSERT_HEAD(head, fp, ipq_list); 934 V_nipq++; 935 fp->ipq_nfrags = 1; 936 fp->ipq_ttl = IPFRAGTTL; 937 fp->ipq_p = ip->ip_p; 938 fp->ipq_id = ip->ip_id; 939 fp->ipq_src = ip->ip_src; 940 fp->ipq_dst = ip->ip_dst; 941 fp->ipq_frags = m; 942 m->m_nextpkt = NULL; 943 goto done; 944 } else { 945 fp->ipq_nfrags++; 946#ifdef MAC 947 mac_ipq_update(m, fp); 948#endif 949 } 950 951#define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) 952 953 /* 954 * Handle ECN by comparing this segment with the first one; 955 * if CE is set, do not lose CE. 956 * drop if CE and not-ECT are mixed for the same packet. 957 */ 958 ecn = ip->ip_tos & IPTOS_ECN_MASK; 959 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; 960 if (ecn == IPTOS_ECN_CE) { 961 if (ecn0 == IPTOS_ECN_NOTECT) 962 goto dropfrag; 963 if (ecn0 != IPTOS_ECN_CE) 964 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; 965 } 966 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) 967 goto dropfrag; 968 969 /* 970 * Find a segment which begins after this one does. 971 */ 972 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) 973 if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) 974 break; 975 976 /* 977 * If there is a preceding segment, it may provide some of 978 * our data already. If so, drop the data from the incoming 979 * segment. If it provides all of our data, drop us, otherwise 980 * stick new segment in the proper place. 981 * 982 * If some of the data is dropped from the preceding 983 * segment, then it's checksum is invalidated. 984 */ 985 if (p) { 986 i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - 987 ntohs(ip->ip_off); 988 if (i > 0) { 989 if (i >= ntohs(ip->ip_len)) 990 goto dropfrag; 991 m_adj(m, i); 992 m->m_pkthdr.csum_flags = 0; 993 ip->ip_off = htons(ntohs(ip->ip_off) + i); 994 ip->ip_len = htons(ntohs(ip->ip_len) - i); 995 } 996 m->m_nextpkt = p->m_nextpkt; 997 p->m_nextpkt = m; 998 } else { 999 m->m_nextpkt = fp->ipq_frags; 1000 fp->ipq_frags = m; 1001 } 1002 1003 /* 1004 * While we overlap succeeding segments trim them or, 1005 * if they are completely covered, dequeue them. 1006 */ 1007 for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > 1008 ntohs(GETIP(q)->ip_off); q = nq) { 1009 i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - 1010 ntohs(GETIP(q)->ip_off); 1011 if (i < ntohs(GETIP(q)->ip_len)) { 1012 GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); 1013 GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); 1014 m_adj(q, i); 1015 q->m_pkthdr.csum_flags = 0; 1016 break; 1017 } 1018 nq = q->m_nextpkt; 1019 m->m_nextpkt = nq; 1020 IPSTAT_INC(ips_fragdropped); 1021 fp->ipq_nfrags--; 1022 m_freem(q); 1023 } 1024 1025 /* 1026 * Check for complete reassembly and perform frag per packet 1027 * limiting. 1028 * 1029 * Frag limiting is performed here so that the nth frag has 1030 * a chance to complete the packet before we drop the packet. 1031 * As a result, n+1 frags are actually allowed per packet, but 1032 * only n will ever be stored. (n = maxfragsperpacket.) 1033 * 1034 */ 1035 next = 0; 1036 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { 1037 if (ntohs(GETIP(q)->ip_off) != next) { 1038 if (fp->ipq_nfrags > V_maxfragsperpacket) { 1039 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1040 ip_freef(head, fp); 1041 } 1042 goto done; 1043 } 1044 next += ntohs(GETIP(q)->ip_len); 1045 } 1046 /* Make sure the last packet didn't have the IP_MF flag */ 1047 if (p->m_flags & M_IP_FRAG) { 1048 if (fp->ipq_nfrags > V_maxfragsperpacket) { 1049 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1050 ip_freef(head, fp); 1051 } 1052 goto done; 1053 } 1054 1055 /* 1056 * Reassembly is complete. Make sure the packet is a sane size. 1057 */ 1058 q = fp->ipq_frags; 1059 ip = GETIP(q); 1060 if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { 1061 IPSTAT_INC(ips_toolong); 1062 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1063 ip_freef(head, fp); 1064 goto done; 1065 } 1066 1067 /* 1068 * Concatenate fragments. 1069 */ 1070 m = q; 1071 t = m->m_next; 1072 m->m_next = NULL; 1073 m_cat(m, t); 1074 nq = q->m_nextpkt; 1075 q->m_nextpkt = NULL; 1076 for (q = nq; q != NULL; q = nq) { 1077 nq = q->m_nextpkt; 1078 q->m_nextpkt = NULL; 1079 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; 1080 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; 1081 m_cat(m, q); 1082 } 1083 /* 1084 * In order to do checksumming faster we do 'end-around carry' here 1085 * (and not in for{} loop), though it implies we are not going to 1086 * reassemble more than 64k fragments. 1087 */ 1088 m->m_pkthdr.csum_data = 1089 (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16); 1090#ifdef MAC 1091 mac_ipq_reassemble(fp, m); 1092 mac_ipq_destroy(fp); 1093#endif 1094 1095 /* 1096 * Create header for new ip packet by modifying header of first 1097 * packet; dequeue and discard fragment reassembly header. 1098 * Make header visible. 1099 */ 1100 ip->ip_len = htons((ip->ip_hl << 2) + next); 1101 ip->ip_src = fp->ipq_src; 1102 ip->ip_dst = fp->ipq_dst; 1103 TAILQ_REMOVE(head, fp, ipq_list); 1104 V_nipq--; 1105 uma_zfree(V_ipq_zone, fp); 1106 m->m_len += (ip->ip_hl << 2); 1107 m->m_data -= (ip->ip_hl << 2); 1108 /* some debugging cruft by sklower, below, will go away soon */ 1109 if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ 1110 m_fixhdr(m); 1111 IPSTAT_INC(ips_reassembled); 1112 IPQ_UNLOCK(); 1113 return (m); 1114 1115dropfrag: 1116 IPSTAT_INC(ips_fragdropped); 1117 if (fp != NULL) 1118 fp->ipq_nfrags--; 1119 m_freem(m); 1120done: 1121 IPQ_UNLOCK(); 1122 return (NULL); 1123 1124#undef GETIP 1125} 1126 1127/* 1128 * Free a fragment reassembly header and all 1129 * associated datagrams. 1130 */ 1131static void 1132ip_freef(struct ipqhead *fhp, struct ipq *fp) 1133{ 1134 struct mbuf *q; 1135 1136 IPQ_LOCK_ASSERT(); 1137 1138 while (fp->ipq_frags) { 1139 q = fp->ipq_frags; 1140 fp->ipq_frags = q->m_nextpkt; 1141 m_freem(q); 1142 } 1143 TAILQ_REMOVE(fhp, fp, ipq_list); 1144 uma_zfree(V_ipq_zone, fp); 1145 V_nipq--; 1146} 1147 1148/* 1149 * IP timer processing; 1150 * if a timer expires on a reassembly 1151 * queue, discard it. 1152 */ 1153void 1154ip_slowtimo(void) 1155{ 1156 VNET_ITERATOR_DECL(vnet_iter); 1157 struct ipq *fp; 1158 int i; 1159 1160 VNET_LIST_RLOCK_NOSLEEP(); 1161 IPQ_LOCK(); 1162 VNET_FOREACH(vnet_iter) { 1163 CURVNET_SET(vnet_iter); 1164 for (i = 0; i < IPREASS_NHASH; i++) { 1165 for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) { 1166 struct ipq *fpp; 1167 1168 fpp = fp; 1169 fp = TAILQ_NEXT(fp, ipq_list); 1170 if(--fpp->ipq_ttl == 0) { 1171 IPSTAT_ADD(ips_fragtimeout, 1172 fpp->ipq_nfrags); 1173 ip_freef(&V_ipq[i], fpp); 1174 } 1175 } 1176 } 1177 /* 1178 * If we are over the maximum number of fragments 1179 * (due to the limit being lowered), drain off 1180 * enough to get down to the new limit. 1181 */ 1182 if (V_maxnipq >= 0 && V_nipq > V_maxnipq) { 1183 for (i = 0; i < IPREASS_NHASH; i++) { 1184 while (V_nipq > V_maxnipq && 1185 !TAILQ_EMPTY(&V_ipq[i])) { 1186 IPSTAT_ADD(ips_fragdropped, 1187 TAILQ_FIRST(&V_ipq[i])->ipq_nfrags); 1188 ip_freef(&V_ipq[i], 1189 TAILQ_FIRST(&V_ipq[i])); 1190 } 1191 } 1192 } 1193 CURVNET_RESTORE(); 1194 } 1195 IPQ_UNLOCK(); 1196 VNET_LIST_RUNLOCK_NOSLEEP(); 1197} 1198 1199/* 1200 * Drain off all datagram fragments. 1201 */ 1202static void 1203ip_drain_locked(void) 1204{ 1205 int i; 1206 1207 IPQ_LOCK_ASSERT(); 1208 1209 for (i = 0; i < IPREASS_NHASH; i++) { 1210 while(!TAILQ_EMPTY(&V_ipq[i])) { 1211 IPSTAT_ADD(ips_fragdropped, 1212 TAILQ_FIRST(&V_ipq[i])->ipq_nfrags); 1213 ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i])); 1214 } 1215 } 1216} 1217 1218void 1219ip_drain(void) 1220{ 1221 VNET_ITERATOR_DECL(vnet_iter); 1222 1223 VNET_LIST_RLOCK_NOSLEEP(); 1224 IPQ_LOCK(); 1225 VNET_FOREACH(vnet_iter) { 1226 CURVNET_SET(vnet_iter); 1227 ip_drain_locked(); 1228 CURVNET_RESTORE(); 1229 } 1230 IPQ_UNLOCK(); 1231 VNET_LIST_RUNLOCK_NOSLEEP(); 1232 in_rtqdrain(); 1233} 1234 1235/* 1236 * The protocol to be inserted into ip_protox[] must be already registered 1237 * in inetsw[], either statically or through pf_proto_register(). 1238 */ 1239int 1240ipproto_register(short ipproto) 1241{ 1242 struct protosw *pr; 1243 1244 /* Sanity checks. */ 1245 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1246 return (EPROTONOSUPPORT); 1247 1248 /* 1249 * The protocol slot must not be occupied by another protocol 1250 * already. An index pointing to IPPROTO_RAW is unused. 1251 */ 1252 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1253 if (pr == NULL) 1254 return (EPFNOSUPPORT); 1255 if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */ 1256 return (EEXIST); 1257 1258 /* Find the protocol position in inetsw[] and set the index. */ 1259 for (pr = inetdomain.dom_protosw; 1260 pr < inetdomain.dom_protoswNPROTOSW; pr++) { 1261 if (pr->pr_domain->dom_family == PF_INET && 1262 pr->pr_protocol && pr->pr_protocol == ipproto) { 1263 ip_protox[pr->pr_protocol] = pr - inetsw; 1264 return (0); 1265 } 1266 } 1267 return (EPROTONOSUPPORT); 1268} 1269 1270int 1271ipproto_unregister(short ipproto) 1272{ 1273 struct protosw *pr; 1274 1275 /* Sanity checks. */ 1276 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1277 return (EPROTONOSUPPORT); 1278 1279 /* Check if the protocol was indeed registered. */ 1280 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1281 if (pr == NULL) 1282 return (EPFNOSUPPORT); 1283 if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */ 1284 return (ENOENT); 1285 1286 /* Reset the protocol slot to IPPROTO_RAW. */ 1287 ip_protox[ipproto] = pr - inetsw; 1288 return (0); 1289} 1290 1291/* 1292 * Given address of next destination (final or next hop), return (referenced) 1293 * internet address info of interface to be used to get there. 1294 */ 1295struct in_ifaddr * 1296ip_rtaddr(struct in_addr dst, u_int fibnum) 1297{ 1298 struct route sro; 1299 struct sockaddr_in *sin; 1300 struct in_ifaddr *ia; 1301 1302 bzero(&sro, sizeof(sro)); 1303 sin = (struct sockaddr_in *)&sro.ro_dst; 1304 sin->sin_family = AF_INET; 1305 sin->sin_len = sizeof(*sin); 1306 sin->sin_addr = dst; 1307 in_rtalloc_ign(&sro, 0, fibnum); 1308 1309 if (sro.ro_rt == NULL) 1310 return (NULL); 1311 1312 ia = ifatoia(sro.ro_rt->rt_ifa); 1313 ifa_ref(&ia->ia_ifa); 1314 RTFREE(sro.ro_rt); 1315 return (ia); 1316} 1317 1318u_char inetctlerrmap[PRC_NCMDS] = { 1319 0, 0, 0, 0, 1320 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH, 1321 EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED, 1322 EMSGSIZE, EHOSTUNREACH, 0, 0, 1323 0, 0, EHOSTUNREACH, 0, 1324 ENOPROTOOPT, ECONNREFUSED 1325}; 1326 1327/* 1328 * Forward a packet. If some error occurs return the sender 1329 * an icmp packet. Note we can't always generate a meaningful 1330 * icmp message because icmp doesn't have a large enough repertoire 1331 * of codes and types. 1332 * 1333 * If not forwarding, just drop the packet. This could be confusing 1334 * if ipforwarding was zero but some routing protocol was advancing 1335 * us as a gateway to somewhere. However, we must let the routing 1336 * protocol deal with that. 1337 * 1338 * The srcrt parameter indicates whether the packet is being forwarded 1339 * via a source route. 1340 */ 1341void 1342ip_forward(struct mbuf *m, int srcrt) 1343{ 1344 struct ip *ip = mtod(m, struct ip *); 1345 struct in_ifaddr *ia; 1346 struct mbuf *mcopy; 1347 struct in_addr dest; 1348 struct route ro; 1349 int error, type = 0, code = 0, mtu = 0; 1350 1351 if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) { 1352 IPSTAT_INC(ips_cantforward); 1353 m_freem(m); 1354 return; 1355 } 1356#ifdef IPSTEALTH 1357 if (!V_ipstealth) { 1358#endif 1359 if (ip->ip_ttl <= IPTTLDEC) { 1360 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 1361 0, 0); 1362 return; 1363 } 1364#ifdef IPSTEALTH 1365 } 1366#endif 1367 1368 ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m)); 1369#ifndef IPSEC 1370 /* 1371 * 'ia' may be NULL if there is no route for this destination. 1372 * In case of IPsec, Don't discard it just yet, but pass it to 1373 * ip_output in case of outgoing IPsec policy. 1374 */ 1375 if (!srcrt && ia == NULL) { 1376 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 1377 return; 1378 } 1379#endif 1380 1381 /* 1382 * Save the IP header and at most 8 bytes of the payload, 1383 * in case we need to generate an ICMP message to the src. 1384 * 1385 * XXX this can be optimized a lot by saving the data in a local 1386 * buffer on the stack (72 bytes at most), and only allocating the 1387 * mbuf if really necessary. The vast majority of the packets 1388 * are forwarded without having to send an ICMP back (either 1389 * because unnecessary, or because rate limited), so we are 1390 * really we are wasting a lot of work here. 1391 * 1392 * We don't use m_copy() because it might return a reference 1393 * to a shared cluster. Both this function and ip_output() 1394 * assume exclusive access to the IP header in `m', so any 1395 * data in a cluster may change before we reach icmp_error(). 1396 */ 1397 mcopy = m_gethdr(M_NOWAIT, m->m_type); 1398 if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) { 1399 /* 1400 * It's probably ok if the pkthdr dup fails (because 1401 * the deep copy of the tag chain failed), but for now 1402 * be conservative and just discard the copy since 1403 * code below may some day want the tags. 1404 */ 1405 m_free(mcopy); 1406 mcopy = NULL; 1407 } 1408 if (mcopy != NULL) { 1409 mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy)); 1410 mcopy->m_pkthdr.len = mcopy->m_len; 1411 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t)); 1412 } 1413 1414#ifdef IPSTEALTH 1415 if (!V_ipstealth) { 1416#endif 1417 ip->ip_ttl -= IPTTLDEC; 1418#ifdef IPSTEALTH 1419 } 1420#endif 1421 1422 /* 1423 * If forwarding packet using same interface that it came in on, 1424 * perhaps should send a redirect to sender to shortcut a hop. 1425 * Only send redirect if source is sending directly to us, 1426 * and if packet was not source routed (or has any options). 1427 * Also, don't send redirect if forwarding using a default route 1428 * or a route modified by a redirect. 1429 */ 1430 dest.s_addr = 0; 1431 if (!srcrt && V_ipsendredirects && 1432 ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) { 1433 struct sockaddr_in *sin; 1434 struct rtentry *rt; 1435 1436 bzero(&ro, sizeof(ro)); 1437 sin = (struct sockaddr_in *)&ro.ro_dst; 1438 sin->sin_family = AF_INET; 1439 sin->sin_len = sizeof(*sin); 1440 sin->sin_addr = ip->ip_dst; 1441 in_rtalloc_ign(&ro, 0, M_GETFIB(m)); 1442 1443 rt = ro.ro_rt; 1444 1445 if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 && 1446 satosin(rt_key(rt))->sin_addr.s_addr != 0) { 1447#define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa)) 1448 u_long src = ntohl(ip->ip_src.s_addr); 1449 1450 if (RTA(rt) && 1451 (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) { 1452 if (rt->rt_flags & RTF_GATEWAY) 1453 dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr; 1454 else 1455 dest.s_addr = ip->ip_dst.s_addr; 1456 /* Router requirements says to only send host redirects */ 1457 type = ICMP_REDIRECT; 1458 code = ICMP_REDIRECT_HOST; 1459 } 1460 } 1461 if (rt) 1462 RTFREE(rt); 1463 } 1464 1465 /* 1466 * Try to cache the route MTU from ip_output so we can consider it for 1467 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191. 1468 */ 1469 bzero(&ro, sizeof(ro)); 1470 1471 error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL); 1472 1473 if (error == EMSGSIZE && ro.ro_rt) 1474 mtu = ro.ro_rt->rt_rmx.rmx_mtu; 1475 RO_RTFREE(&ro); 1476 1477 if (error) 1478 IPSTAT_INC(ips_cantforward); 1479 else { 1480 IPSTAT_INC(ips_forward); 1481 if (type) 1482 IPSTAT_INC(ips_redirectsent); 1483 else { 1484 if (mcopy) 1485 m_freem(mcopy); 1486 if (ia != NULL) 1487 ifa_free(&ia->ia_ifa); 1488 return; 1489 } 1490 } 1491 if (mcopy == NULL) { 1492 if (ia != NULL) 1493 ifa_free(&ia->ia_ifa); 1494 return; 1495 } 1496 1497 switch (error) { 1498 1499 case 0: /* forwarded, but need redirect */ 1500 /* type, code set above */ 1501 break; 1502 1503 case ENETUNREACH: 1504 case EHOSTUNREACH: 1505 case ENETDOWN: 1506 case EHOSTDOWN: 1507 default: 1508 type = ICMP_UNREACH; 1509 code = ICMP_UNREACH_HOST; 1510 break; 1511 1512 case EMSGSIZE: 1513 type = ICMP_UNREACH; 1514 code = ICMP_UNREACH_NEEDFRAG; 1515 1516#ifdef IPSEC 1517 /* 1518 * If IPsec is configured for this path, 1519 * override any possibly mtu value set by ip_output. 1520 */ 1521 mtu = ip_ipsec_mtu(mcopy, mtu); 1522#endif /* IPSEC */ 1523 /* 1524 * If the MTU was set before make sure we are below the 1525 * interface MTU. 1526 * If the MTU wasn't set before use the interface mtu or 1527 * fall back to the next smaller mtu step compared to the 1528 * current packet size. 1529 */ 1530 if (mtu != 0) { 1531 if (ia != NULL) 1532 mtu = min(mtu, ia->ia_ifp->if_mtu); 1533 } else { 1534 if (ia != NULL) 1535 mtu = ia->ia_ifp->if_mtu; 1536 else 1537 mtu = ip_next_mtu(ntohs(ip->ip_len), 0); 1538 } 1539 IPSTAT_INC(ips_cantfrag); 1540 break; 1541 1542 case ENOBUFS: 1543 /* 1544 * A router should not generate ICMP_SOURCEQUENCH as 1545 * required in RFC1812 Requirements for IP Version 4 Routers. 1546 * Source quench could be a big problem under DoS attacks, 1547 * or if the underlying interface is rate-limited. 1548 * Those who need source quench packets may re-enable them 1549 * via the net.inet.ip.sendsourcequench sysctl. 1550 */ 1551 if (V_ip_sendsourcequench == 0) { 1552 m_freem(mcopy); 1553 if (ia != NULL) 1554 ifa_free(&ia->ia_ifa); 1555 return; 1556 } else { 1557 type = ICMP_SOURCEQUENCH; 1558 code = 0; 1559 } 1560 break; 1561 1562 case EACCES: /* ipfw denied packet */ 1563 m_freem(mcopy); 1564 if (ia != NULL) 1565 ifa_free(&ia->ia_ifa); 1566 return; 1567 } 1568 if (ia != NULL) 1569 ifa_free(&ia->ia_ifa); 1570 icmp_error(mcopy, type, code, dest.s_addr, mtu); 1571} 1572 1573void 1574ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip, 1575 struct mbuf *m) 1576{ 1577 1578 if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) { 1579 struct bintime bt; 1580 1581 bintime(&bt); 1582 if (inp->inp_socket->so_options & SO_BINTIME) { 1583 *mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt), 1584 SCM_BINTIME, SOL_SOCKET); 1585 if (*mp) 1586 mp = &(*mp)->m_next; 1587 } 1588 if (inp->inp_socket->so_options & SO_TIMESTAMP) { 1589 struct timeval tv; 1590 1591 bintime2timeval(&bt, &tv); 1592 *mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv), 1593 SCM_TIMESTAMP, SOL_SOCKET); 1594 if (*mp) 1595 mp = &(*mp)->m_next; 1596 } 1597 } 1598 if (inp->inp_flags & INP_RECVDSTADDR) { 1599 *mp = sbcreatecontrol((caddr_t)&ip->ip_dst, 1600 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP); 1601 if (*mp) 1602 mp = &(*mp)->m_next; 1603 } 1604 if (inp->inp_flags & INP_RECVTTL) { 1605 *mp = sbcreatecontrol((caddr_t)&ip->ip_ttl, 1606 sizeof(u_char), IP_RECVTTL, IPPROTO_IP); 1607 if (*mp) 1608 mp = &(*mp)->m_next; 1609 } 1610#ifdef notyet 1611 /* XXX 1612 * Moving these out of udp_input() made them even more broken 1613 * than they already were. 1614 */ 1615 /* options were tossed already */ 1616 if (inp->inp_flags & INP_RECVOPTS) { 1617 *mp = sbcreatecontrol((caddr_t)opts_deleted_above, 1618 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP); 1619 if (*mp) 1620 mp = &(*mp)->m_next; 1621 } 1622 /* ip_srcroute doesn't do what we want here, need to fix */ 1623 if (inp->inp_flags & INP_RECVRETOPTS) { 1624 *mp = sbcreatecontrol((caddr_t)ip_srcroute(m), 1625 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP); 1626 if (*mp) 1627 mp = &(*mp)->m_next; 1628 } 1629#endif 1630 if (inp->inp_flags & INP_RECVIF) { 1631 struct ifnet *ifp; 1632 struct sdlbuf { 1633 struct sockaddr_dl sdl; 1634 u_char pad[32]; 1635 } sdlbuf; 1636 struct sockaddr_dl *sdp; 1637 struct sockaddr_dl *sdl2 = &sdlbuf.sdl; 1638 1639 if ((ifp = m->m_pkthdr.rcvif) && 1640 ifp->if_index && ifp->if_index <= V_if_index) { 1641 sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr; 1642 /* 1643 * Change our mind and don't try copy. 1644 */ 1645 if (sdp->sdl_family != AF_LINK || 1646 sdp->sdl_len > sizeof(sdlbuf)) { 1647 goto makedummy; 1648 } 1649 bcopy(sdp, sdl2, sdp->sdl_len); 1650 } else { 1651makedummy: 1652 sdl2->sdl_len = 1653 offsetof(struct sockaddr_dl, sdl_data[0]); 1654 sdl2->sdl_family = AF_LINK; 1655 sdl2->sdl_index = 0; 1656 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0; 1657 } 1658 *mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len, 1659 IP_RECVIF, IPPROTO_IP); 1660 if (*mp) 1661 mp = &(*mp)->m_next; 1662 } 1663 if (inp->inp_flags & INP_RECVTOS) { 1664 *mp = sbcreatecontrol((caddr_t)&ip->ip_tos, 1665 sizeof(u_char), IP_RECVTOS, IPPROTO_IP); 1666 if (*mp) 1667 mp = &(*mp)->m_next; 1668 } 1669} 1670 1671/* 1672 * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the 1673 * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on 1674 * locking. This code remains in ip_input.c as ip_mroute.c is optionally 1675 * compiled. 1676 */ 1677static VNET_DEFINE(int, ip_rsvp_on); 1678VNET_DEFINE(struct socket *, ip_rsvpd); 1679 1680#define V_ip_rsvp_on VNET(ip_rsvp_on) 1681 1682int 1683ip_rsvp_init(struct socket *so) 1684{ 1685 1686 if (so->so_type != SOCK_RAW || 1687 so->so_proto->pr_protocol != IPPROTO_RSVP) 1688 return EOPNOTSUPP; 1689 1690 if (V_ip_rsvpd != NULL) 1691 return EADDRINUSE; 1692 1693 V_ip_rsvpd = so; 1694 /* 1695 * This may seem silly, but we need to be sure we don't over-increment 1696 * the RSVP counter, in case something slips up. 1697 */ 1698 if (!V_ip_rsvp_on) { 1699 V_ip_rsvp_on = 1; 1700 V_rsvp_on++; 1701 } 1702 1703 return 0; 1704} 1705 1706int 1707ip_rsvp_done(void) 1708{ 1709 1710 V_ip_rsvpd = NULL; 1711 /* 1712 * This may seem silly, but we need to be sure we don't over-decrement 1713 * the RSVP counter, in case something slips up. 1714 */ 1715 if (V_ip_rsvp_on) { 1716 V_ip_rsvp_on = 0; 1717 V_rsvp_on--; 1718 } 1719 return 0; 1720} 1721 1722void 1723rsvp_input(struct mbuf *m, int off) /* XXX must fixup manually */ 1724{ 1725 1726 if (rsvp_input_p) { /* call the real one if loaded */ 1727 rsvp_input_p(m, off); 1728 return; 1729 } 1730 1731 /* Can still get packets with rsvp_on = 0 if there is a local member 1732 * of the group to which the RSVP packet is addressed. But in this 1733 * case we want to throw the packet away. 1734 */ 1735 1736 if (!V_rsvp_on) { 1737 m_freem(m); 1738 return; 1739 } 1740 1741 if (V_ip_rsvpd != NULL) { 1742 rip_input(m, off); 1743 return; 1744 } 1745 /* Drop the packet */ 1746 m_freem(m); 1747} 1748