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