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 #include "opt_rss.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/mbuf.h> 45 #include <sys/malloc.h> 46 #include <sys/domain.h> 47 #include <sys/protosw.h> 48 #include <sys/socket.h> 49 #include <sys/time.h> 50 #include <sys/kernel.h> 51 #include <sys/lock.h> 52 #include <sys/rwlock.h> 53 #include <sys/sdt.h> 54 #include <sys/syslog.h> 55 #include <sys/sysctl.h> 56 57 #include <net/pfil.h> 58 #include <net/if.h> 59 #include <net/if_types.h> 60 #include <net/if_var.h> 61 #include <net/if_dl.h> 62 #include <net/route.h> 63 #include <net/netisr.h> 64 #include <net/rss_config.h> 65 #include <net/vnet.h> 66 67 #include <netinet/in.h> 68 #include <netinet/in_kdtrace.h> 69 #include <netinet/in_systm.h> 70 #include <netinet/in_var.h> 71 #include <netinet/ip.h> 72 #include <netinet/in_pcb.h> 73 #include <netinet/ip_var.h> 74 #include <netinet/ip_fw.h> 75 #include <netinet/ip_icmp.h> 76 #include <netinet/ip_options.h> 77 #include <machine/in_cksum.h> 78 #include <netinet/ip_carp.h> 79 #ifdef IPSEC 80 #include <netinet/ip_ipsec.h> 81 #endif /* IPSEC */ 82 #include <netinet/in_rss.h> 83 84 #include <sys/socketvar.h> 85 86 #include <security/mac/mac_framework.h> 87 88 #ifdef CTASSERT 89 CTASSERT(sizeof(struct ip) == 20); 90 #endif 91 92 struct rwlock in_ifaddr_lock; 93 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock"); 94 95 VNET_DEFINE(int, rsvp_on); 96 97 VNET_DEFINE(int, ipforwarding); 98 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_VNET | CTLFLAG_RW, 99 &VNET_NAME(ipforwarding), 0, 100 "Enable IP forwarding between interfaces"); 101 102 static VNET_DEFINE(int, ipsendredirects) = 1; /* XXX */ 103 #define V_ipsendredirects VNET(ipsendredirects) 104 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_VNET | CTLFLAG_RW, 105 &VNET_NAME(ipsendredirects), 0, 106 "Enable sending IP redirects"); 107 108 VNET_DEFINE(int, ip_do_randomid); 109 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_VNET | CTLFLAG_RW, 110 &VNET_NAME(ip_do_randomid), 0, 111 "Assign random ip_id values"); 112 113 /* 114 * XXX - Setting ip_checkinterface mostly implements the receive side of 115 * the Strong ES model described in RFC 1122, but since the routing table 116 * and transmit implementation do not implement the Strong ES model, 117 * setting this to 1 results in an odd hybrid. 118 * 119 * XXX - ip_checkinterface currently must be disabled if you use ipnat 120 * to translate the destination address to another local interface. 121 * 122 * XXX - ip_checkinterface must be disabled if you add IP aliases 123 * to the loopback interface instead of the interface where the 124 * packets for those addresses are received. 125 */ 126 static VNET_DEFINE(int, ip_checkinterface); 127 #define V_ip_checkinterface VNET(ip_checkinterface) 128 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_VNET | CTLFLAG_RW, 129 &VNET_NAME(ip_checkinterface), 0, 130 "Verify packet arrives on correct interface"); 131 132 VNET_DEFINE(struct pfil_head, inet_pfil_hook); /* Packet filter hooks */ 133 134 static struct netisr_handler ip_nh = { 135 .nh_name = "ip", 136 .nh_handler = ip_input, 137 .nh_proto = NETISR_IP, 138 #ifdef RSS 139 .nh_m2cpuid = rss_soft_m2cpuid, 140 .nh_policy = NETISR_POLICY_CPU, 141 .nh_dispatch = NETISR_DISPATCH_HYBRID, 142 #else 143 .nh_policy = NETISR_POLICY_FLOW, 144 #endif 145 }; 146 147 #ifdef RSS 148 /* 149 * Directly dispatched frames are currently assumed 150 * to have a flowid already calculated. 151 * 152 * It should likely have something that assert it 153 * actually has valid flow details. 154 */ 155 static struct netisr_handler ip_direct_nh = { 156 .nh_name = "ip_direct", 157 .nh_handler = ip_direct_input, 158 .nh_proto = NETISR_IP_DIRECT, 159 .nh_m2cpuid = rss_m2cpuid, 160 .nh_policy = NETISR_POLICY_CPU, 161 .nh_dispatch = NETISR_DISPATCH_HYBRID, 162 }; 163 #endif 164 165 extern struct domain inetdomain; 166 extern struct protosw inetsw[]; 167 u_char ip_protox[IPPROTO_MAX]; 168 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead); /* first inet address */ 169 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table */ 170 VNET_DEFINE(u_long, in_ifaddrhmask); /* mask for hash table */ 171 172 static VNET_DEFINE(uma_zone_t, ipq_zone); 173 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]); 174 static struct mtx ipqlock; 175 176 #define V_ipq_zone VNET(ipq_zone) 177 #define V_ipq VNET(ipq) 178 179 #define IPQ_LOCK() mtx_lock(&ipqlock) 180 #define IPQ_UNLOCK() mtx_unlock(&ipqlock) 181 #define IPQ_LOCK_INIT() mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF) 182 #define IPQ_LOCK_ASSERT() mtx_assert(&ipqlock, MA_OWNED) 183 184 static void maxnipq_update(void); 185 static void ipq_zone_change(void *); 186 static void ip_drain_locked(void); 187 188 static VNET_DEFINE(int, maxnipq); /* Administrative limit on # reass queues. */ 189 static VNET_DEFINE(int, nipq); /* Total # of reass queues */ 190 #define V_maxnipq VNET(maxnipq) 191 #define V_nipq VNET(nipq) 192 SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_VNET | CTLFLAG_RD, 193 &VNET_NAME(nipq), 0, 194 "Current number of IPv4 fragment reassembly queue entries"); 195 196 static VNET_DEFINE(int, maxfragsperpacket); 197 #define V_maxfragsperpacket VNET(maxfragsperpacket) 198 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(maxfragsperpacket), 0, 200 "Maximum number of IPv4 fragments allowed per packet"); 201 202 #ifdef IPCTL_DEFMTU 203 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW, 204 &ip_mtu, 0, "Default MTU"); 205 #endif 206 207 #ifdef IPSTEALTH 208 VNET_DEFINE(int, ipstealth); 209 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_VNET | CTLFLAG_RW, 210 &VNET_NAME(ipstealth), 0, 211 "IP stealth mode, no TTL decrementation on forwarding"); 212 #endif 213 214 static void ip_freef(struct ipqhead *, struct ipq *); 215 216 /* 217 * IP statistics are stored in the "array" of counter(9)s. 218 */ 219 VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat); 220 VNET_PCPUSTAT_SYSINIT(ipstat); 221 SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat, 222 "IP statistics (struct ipstat, netinet/ip_var.h)"); 223 224 #ifdef VIMAGE 225 VNET_PCPUSTAT_SYSUNINIT(ipstat); 226 #endif /* VIMAGE */ 227 228 /* 229 * Kernel module interface for updating ipstat. The argument is an index 230 * into ipstat treated as an array. 231 */ 232 void 233 kmod_ipstat_inc(int statnum) 234 { 235 236 counter_u64_add(VNET(ipstat)[statnum], 1); 237 } 238 239 void 240 kmod_ipstat_dec(int statnum) 241 { 242 243 counter_u64_add(VNET(ipstat)[statnum], -1); 244 } 245 246 static int 247 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS) 248 { 249 int error, qlimit; 250 251 netisr_getqlimit(&ip_nh, &qlimit); 252 error = sysctl_handle_int(oidp, &qlimit, 0, req); 253 if (error || !req->newptr) 254 return (error); 255 if (qlimit < 1) 256 return (EINVAL); 257 return (netisr_setqlimit(&ip_nh, qlimit)); 258 } 259 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, 260 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I", 261 "Maximum size of the IP input queue"); 262 263 static int 264 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS) 265 { 266 u_int64_t qdrops_long; 267 int error, qdrops; 268 269 netisr_getqdrops(&ip_nh, &qdrops_long); 270 qdrops = qdrops_long; 271 error = sysctl_handle_int(oidp, &qdrops, 0, req); 272 if (error || !req->newptr) 273 return (error); 274 if (qdrops != 0) 275 return (EINVAL); 276 netisr_clearqdrops(&ip_nh); 277 return (0); 278 } 279 280 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, 281 CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I", 282 "Number of packets dropped from the IP input queue"); 283 284 #ifdef RSS 285 static int 286 sysctl_netinet_intr_direct_queue_maxlen(SYSCTL_HANDLER_ARGS) 287 { 288 int error, qlimit; 289 290 netisr_getqlimit(&ip_direct_nh, &qlimit); 291 error = sysctl_handle_int(oidp, &qlimit, 0, req); 292 if (error || !req->newptr) 293 return (error); 294 if (qlimit < 1) 295 return (EINVAL); 296 return (netisr_setqlimit(&ip_direct_nh, qlimit)); 297 } 298 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_direct_queue_maxlen, 299 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_direct_queue_maxlen, "I", 300 "Maximum size of the IP direct input queue"); 301 302 static int 303 sysctl_netinet_intr_direct_queue_drops(SYSCTL_HANDLER_ARGS) 304 { 305 u_int64_t qdrops_long; 306 int error, qdrops; 307 308 netisr_getqdrops(&ip_direct_nh, &qdrops_long); 309 qdrops = qdrops_long; 310 error = sysctl_handle_int(oidp, &qdrops, 0, req); 311 if (error || !req->newptr) 312 return (error); 313 if (qdrops != 0) 314 return (EINVAL); 315 netisr_clearqdrops(&ip_direct_nh); 316 return (0); 317 } 318 319 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_direct_queue_drops, 320 CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_direct_queue_drops, "I", 321 "Number of packets dropped from the IP direct input queue"); 322 #endif /* RSS */ 323 324 /* 325 * IP initialization: fill in IP protocol switch table. 326 * All protocols not implemented in kernel go to raw IP protocol handler. 327 */ 328 void 329 ip_init(void) 330 { 331 struct protosw *pr; 332 int i; 333 334 V_ip_id = time_second & 0xffff; 335 336 TAILQ_INIT(&V_in_ifaddrhead); 337 V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask); 338 339 /* Initialize IP reassembly queue. */ 340 for (i = 0; i < IPREASS_NHASH; i++) 341 TAILQ_INIT(&V_ipq[i]); 342 V_maxnipq = nmbclusters / 32; 343 V_maxfragsperpacket = 16; 344 V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, 345 NULL, UMA_ALIGN_PTR, 0); 346 maxnipq_update(); 347 348 /* Initialize packet filter hooks. */ 349 V_inet_pfil_hook.ph_type = PFIL_TYPE_AF; 350 V_inet_pfil_hook.ph_af = AF_INET; 351 if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0) 352 printf("%s: WARNING: unable to register pfil hook, " 353 "error %d\n", __func__, i); 354 355 /* Skip initialization of globals for non-default instances. */ 356 if (!IS_DEFAULT_VNET(curvnet)) 357 return; 358 359 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 360 if (pr == NULL) 361 panic("ip_init: PF_INET not found"); 362 363 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */ 364 for (i = 0; i < IPPROTO_MAX; i++) 365 ip_protox[i] = pr - inetsw; 366 /* 367 * Cycle through IP protocols and put them into the appropriate place 368 * in ip_protox[]. 369 */ 370 for (pr = inetdomain.dom_protosw; 371 pr < inetdomain.dom_protoswNPROTOSW; pr++) 372 if (pr->pr_domain->dom_family == PF_INET && 373 pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) { 374 /* Be careful to only index valid IP protocols. */ 375 if (pr->pr_protocol < IPPROTO_MAX) 376 ip_protox[pr->pr_protocol] = pr - inetsw; 377 } 378 379 EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change, 380 NULL, EVENTHANDLER_PRI_ANY); 381 382 /* Initialize various other remaining things. */ 383 IPQ_LOCK_INIT(); 384 netisr_register(&ip_nh); 385 #ifdef RSS 386 netisr_register(&ip_direct_nh); 387 #endif 388 } 389 390 #ifdef VIMAGE 391 void 392 ip_destroy(void) 393 { 394 int i; 395 396 if ((i = pfil_head_unregister(&V_inet_pfil_hook)) != 0) 397 printf("%s: WARNING: unable to unregister pfil hook, " 398 "error %d\n", __func__, i); 399 400 /* Cleanup in_ifaddr hash table; should be empty. */ 401 hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask); 402 403 IPQ_LOCK(); 404 ip_drain_locked(); 405 IPQ_UNLOCK(); 406 407 uma_zdestroy(V_ipq_zone); 408 } 409 #endif 410 411 #ifdef RSS 412 /* 413 * IP direct input routine. 414 * 415 * This is called when reinjecting completed fragments where 416 * all of the previous checking and book-keeping has been done. 417 */ 418 void 419 ip_direct_input(struct mbuf *m) 420 { 421 struct ip *ip; 422 int hlen; 423 424 ip = mtod(m, struct ip *); 425 hlen = ip->ip_hl << 2; 426 427 IPSTAT_INC(ips_delivered); 428 (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); 429 return; 430 } 431 #endif 432 433 /* 434 * Ip input routine. Checksum and byte swap header. If fragmented 435 * try to reassemble. Process options. Pass to next level. 436 */ 437 void 438 ip_input(struct mbuf *m) 439 { 440 struct ip *ip = NULL; 441 struct in_ifaddr *ia = NULL; 442 struct ifaddr *ifa; 443 struct ifnet *ifp; 444 int checkif, hlen = 0; 445 uint16_t sum, ip_len; 446 int dchg = 0; /* dest changed after fw */ 447 struct in_addr odst; /* original dst address */ 448 449 M_ASSERTPKTHDR(m); 450 451 if (m->m_flags & M_FASTFWD_OURS) { 452 m->m_flags &= ~M_FASTFWD_OURS; 453 /* Set up some basics that will be used later. */ 454 ip = mtod(m, struct ip *); 455 hlen = ip->ip_hl << 2; 456 ip_len = ntohs(ip->ip_len); 457 goto ours; 458 } 459 460 IPSTAT_INC(ips_total); 461 462 if (m->m_pkthdr.len < sizeof(struct ip)) 463 goto tooshort; 464 465 if (m->m_len < sizeof (struct ip) && 466 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 467 IPSTAT_INC(ips_toosmall); 468 return; 469 } 470 ip = mtod(m, struct ip *); 471 472 if (ip->ip_v != IPVERSION) { 473 IPSTAT_INC(ips_badvers); 474 goto bad; 475 } 476 477 hlen = ip->ip_hl << 2; 478 if (hlen < sizeof(struct ip)) { /* minimum header length */ 479 IPSTAT_INC(ips_badhlen); 480 goto bad; 481 } 482 if (hlen > m->m_len) { 483 if ((m = m_pullup(m, hlen)) == NULL) { 484 IPSTAT_INC(ips_badhlen); 485 return; 486 } 487 ip = mtod(m, struct ip *); 488 } 489 490 IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL); 491 492 /* 127/8 must not appear on wire - RFC1122 */ 493 ifp = m->m_pkthdr.rcvif; 494 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 495 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 496 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 497 IPSTAT_INC(ips_badaddr); 498 goto bad; 499 } 500 } 501 502 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { 503 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); 504 } else { 505 if (hlen == sizeof(struct ip)) { 506 sum = in_cksum_hdr(ip); 507 } else { 508 sum = in_cksum(m, hlen); 509 } 510 } 511 if (sum) { 512 IPSTAT_INC(ips_badsum); 513 goto bad; 514 } 515 516 #ifdef ALTQ 517 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) 518 /* packet is dropped by traffic conditioner */ 519 return; 520 #endif 521 522 ip_len = ntohs(ip->ip_len); 523 if (ip_len < hlen) { 524 IPSTAT_INC(ips_badlen); 525 goto bad; 526 } 527 528 /* 529 * Check that the amount of data in the buffers 530 * is as at least much as the IP header would have us expect. 531 * Trim mbufs if longer than we expect. 532 * Drop packet if shorter than we expect. 533 */ 534 if (m->m_pkthdr.len < ip_len) { 535 tooshort: 536 IPSTAT_INC(ips_tooshort); 537 goto bad; 538 } 539 if (m->m_pkthdr.len > ip_len) { 540 if (m->m_len == m->m_pkthdr.len) { 541 m->m_len = ip_len; 542 m->m_pkthdr.len = ip_len; 543 } else 544 m_adj(m, ip_len - m->m_pkthdr.len); 545 } 546 547 #ifdef IPSEC 548 /* 549 * Bypass packet filtering for packets previously handled by IPsec. 550 */ 551 if (ip_ipsec_filtertunnel(m)) 552 goto passin; 553 #endif /* IPSEC */ 554 555 /* 556 * Run through list of hooks for input packets. 557 * 558 * NB: Beware of the destination address changing (e.g. 559 * by NAT rewriting). When this happens, tell 560 * ip_forward to do the right thing. 561 */ 562 563 /* Jump over all PFIL processing if hooks are not active. */ 564 if (!PFIL_HOOKED(&V_inet_pfil_hook)) 565 goto passin; 566 567 odst = ip->ip_dst; 568 if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0) 569 return; 570 if (m == NULL) /* consumed by filter */ 571 return; 572 573 ip = mtod(m, struct ip *); 574 dchg = (odst.s_addr != ip->ip_dst.s_addr); 575 ifp = m->m_pkthdr.rcvif; 576 577 if (m->m_flags & M_FASTFWD_OURS) { 578 m->m_flags &= ~M_FASTFWD_OURS; 579 goto ours; 580 } 581 if (m->m_flags & M_IP_NEXTHOP) { 582 dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL); 583 if (dchg != 0) { 584 /* 585 * Directly ship the packet on. This allows 586 * forwarding packets originally destined to us 587 * to some other directly connected host. 588 */ 589 ip_forward(m, 1); 590 return; 591 } 592 } 593 passin: 594 595 /* 596 * Process options and, if not destined for us, 597 * ship it on. ip_dooptions returns 1 when an 598 * error was detected (causing an icmp message 599 * to be sent and the original packet to be freed). 600 */ 601 if (hlen > sizeof (struct ip) && ip_dooptions(m, 0)) 602 return; 603 604 /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no 605 * matter if it is destined to another node, or whether it is 606 * a multicast one, RSVP wants it! and prevents it from being forwarded 607 * anywhere else. Also checks if the rsvp daemon is running before 608 * grabbing the packet. 609 */ 610 if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP) 611 goto ours; 612 613 /* 614 * Check our list of addresses, to see if the packet is for us. 615 * If we don't have any addresses, assume any unicast packet 616 * we receive might be for us (and let the upper layers deal 617 * with it). 618 */ 619 if (TAILQ_EMPTY(&V_in_ifaddrhead) && 620 (m->m_flags & (M_MCAST|M_BCAST)) == 0) 621 goto ours; 622 623 /* 624 * Enable a consistency check between the destination address 625 * and the arrival interface for a unicast packet (the RFC 1122 626 * strong ES model) if IP forwarding is disabled and the packet 627 * is not locally generated and the packet is not subject to 628 * 'ipfw fwd'. 629 * 630 * XXX - Checking also should be disabled if the destination 631 * address is ipnat'ed to a different interface. 632 * 633 * XXX - Checking is incompatible with IP aliases added 634 * to the loopback interface instead of the interface where 635 * the packets are received. 636 * 637 * XXX - This is the case for carp vhost IPs as well so we 638 * insert a workaround. If the packet got here, we already 639 * checked with carp_iamatch() and carp_forus(). 640 */ 641 checkif = V_ip_checkinterface && (V_ipforwarding == 0) && 642 ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) && 643 ifp->if_carp == NULL && (dchg == 0); 644 645 /* 646 * Check for exact addresses in the hash bucket. 647 */ 648 /* IN_IFADDR_RLOCK(); */ 649 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { 650 /* 651 * If the address matches, verify that the packet 652 * arrived via the correct interface if checking is 653 * enabled. 654 */ 655 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr && 656 (!checkif || ia->ia_ifp == ifp)) { 657 counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); 658 counter_u64_add(ia->ia_ifa.ifa_ibytes, 659 m->m_pkthdr.len); 660 /* IN_IFADDR_RUNLOCK(); */ 661 goto ours; 662 } 663 } 664 /* IN_IFADDR_RUNLOCK(); */ 665 666 /* 667 * Check for broadcast addresses. 668 * 669 * Only accept broadcast packets that arrive via the matching 670 * interface. Reception of forwarded directed broadcasts would 671 * be handled via ip_forward() and ether_output() with the loopback 672 * into the stack for SIMPLEX interfaces handled by ether_output(). 673 */ 674 if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { 675 IF_ADDR_RLOCK(ifp); 676 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 677 if (ifa->ifa_addr->sa_family != AF_INET) 678 continue; 679 ia = ifatoia(ifa); 680 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == 681 ip->ip_dst.s_addr) { 682 counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); 683 counter_u64_add(ia->ia_ifa.ifa_ibytes, 684 m->m_pkthdr.len); 685 IF_ADDR_RUNLOCK(ifp); 686 goto ours; 687 } 688 #ifdef BOOTP_COMPAT 689 if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) { 690 counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); 691 counter_u64_add(ia->ia_ifa.ifa_ibytes, 692 m->m_pkthdr.len); 693 IF_ADDR_RUNLOCK(ifp); 694 goto ours; 695 } 696 #endif 697 } 698 IF_ADDR_RUNLOCK(ifp); 699 ia = NULL; 700 } 701 /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */ 702 if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) { 703 IPSTAT_INC(ips_cantforward); 704 m_freem(m); 705 return; 706 } 707 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 708 if (V_ip_mrouter) { 709 /* 710 * If we are acting as a multicast router, all 711 * incoming multicast packets are passed to the 712 * kernel-level multicast forwarding function. 713 * The packet is returned (relatively) intact; if 714 * ip_mforward() returns a non-zero value, the packet 715 * must be discarded, else it may be accepted below. 716 */ 717 if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) { 718 IPSTAT_INC(ips_cantforward); 719 m_freem(m); 720 return; 721 } 722 723 /* 724 * The process-level routing daemon needs to receive 725 * all multicast IGMP packets, whether or not this 726 * host belongs to their destination groups. 727 */ 728 if (ip->ip_p == IPPROTO_IGMP) 729 goto ours; 730 IPSTAT_INC(ips_forward); 731 } 732 /* 733 * Assume the packet is for us, to avoid prematurely taking 734 * a lock on the in_multi hash. Protocols must perform 735 * their own filtering and update statistics accordingly. 736 */ 737 goto ours; 738 } 739 if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST) 740 goto ours; 741 if (ip->ip_dst.s_addr == INADDR_ANY) 742 goto ours; 743 744 /* 745 * Not for us; forward if possible and desirable. 746 */ 747 if (V_ipforwarding == 0) { 748 IPSTAT_INC(ips_cantforward); 749 m_freem(m); 750 } else { 751 ip_forward(m, dchg); 752 } 753 return; 754 755 ours: 756 #ifdef IPSTEALTH 757 /* 758 * IPSTEALTH: Process non-routing options only 759 * if the packet is destined for us. 760 */ 761 if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) 762 return; 763 #endif /* IPSTEALTH */ 764 765 /* 766 * Attempt reassembly; if it succeeds, proceed. 767 * ip_reass() will return a different mbuf. 768 */ 769 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) { 770 /* XXXGL: shouldn't we save & set m_flags? */ 771 m = ip_reass(m); 772 if (m == NULL) 773 return; 774 ip = mtod(m, struct ip *); 775 /* Get the header length of the reassembled packet */ 776 hlen = ip->ip_hl << 2; 777 } 778 779 #ifdef IPSEC 780 /* 781 * enforce IPsec policy checking if we are seeing last header. 782 * note that we do not visit this with protocols with pcb layer 783 * code - like udp/tcp/raw ip. 784 */ 785 if (ip_ipsec_input(m, ip->ip_p) != 0) 786 goto bad; 787 #endif /* IPSEC */ 788 789 /* 790 * Switch out to protocol's input routine. 791 */ 792 IPSTAT_INC(ips_delivered); 793 794 (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); 795 return; 796 bad: 797 m_freem(m); 798 } 799 800 /* 801 * After maxnipq has been updated, propagate the change to UMA. The UMA zone 802 * max has slightly different semantics than the sysctl, for historical 803 * reasons. 804 */ 805 static void 806 maxnipq_update(void) 807 { 808 809 /* 810 * -1 for unlimited allocation. 811 */ 812 if (V_maxnipq < 0) 813 uma_zone_set_max(V_ipq_zone, 0); 814 /* 815 * Positive number for specific bound. 816 */ 817 if (V_maxnipq > 0) 818 uma_zone_set_max(V_ipq_zone, V_maxnipq); 819 /* 820 * Zero specifies no further fragment queue allocation -- set the 821 * bound very low, but rely on implementation elsewhere to actually 822 * prevent allocation and reclaim current queues. 823 */ 824 if (V_maxnipq == 0) 825 uma_zone_set_max(V_ipq_zone, 1); 826 } 827 828 static void 829 ipq_zone_change(void *tag) 830 { 831 832 if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) { 833 V_maxnipq = nmbclusters / 32; 834 maxnipq_update(); 835 } 836 } 837 838 static int 839 sysctl_maxnipq(SYSCTL_HANDLER_ARGS) 840 { 841 int error, i; 842 843 i = V_maxnipq; 844 error = sysctl_handle_int(oidp, &i, 0, req); 845 if (error || !req->newptr) 846 return (error); 847 848 /* 849 * XXXRW: Might be a good idea to sanity check the argument and place 850 * an extreme upper bound. 851 */ 852 if (i < -1) 853 return (EINVAL); 854 V_maxnipq = i; 855 maxnipq_update(); 856 return (0); 857 } 858 859 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW, 860 NULL, 0, sysctl_maxnipq, "I", 861 "Maximum number of IPv4 fragment reassembly queue entries"); 862 863 #define M_IP_FRAG M_PROTO9 864 865 /* 866 * Take incoming datagram fragment and try to reassemble it into 867 * whole datagram. If the argument is the first fragment or one 868 * in between the function will return NULL and store the mbuf 869 * in the fragment chain. If the argument is the last fragment 870 * the packet will be reassembled and the pointer to the new 871 * mbuf returned for further processing. Only m_tags attached 872 * to the first packet/fragment are preserved. 873 * The IP header is *NOT* adjusted out of iplen. 874 */ 875 struct mbuf * 876 ip_reass(struct mbuf *m) 877 { 878 struct ip *ip; 879 struct mbuf *p, *q, *nq, *t; 880 struct ipq *fp = NULL; 881 struct ipqhead *head; 882 int i, hlen, next; 883 u_int8_t ecn, ecn0; 884 u_short hash; 885 #ifdef RSS 886 uint32_t rss_hash, rss_type; 887 #endif 888 889 /* If maxnipq or maxfragsperpacket are 0, never accept fragments. */ 890 if (V_maxnipq == 0 || V_maxfragsperpacket == 0) { 891 IPSTAT_INC(ips_fragments); 892 IPSTAT_INC(ips_fragdropped); 893 m_freem(m); 894 return (NULL); 895 } 896 897 ip = mtod(m, struct ip *); 898 hlen = ip->ip_hl << 2; 899 900 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id); 901 head = &V_ipq[hash]; 902 IPQ_LOCK(); 903 904 /* 905 * Look for queue of fragments 906 * of this datagram. 907 */ 908 TAILQ_FOREACH(fp, head, ipq_list) 909 if (ip->ip_id == fp->ipq_id && 910 ip->ip_src.s_addr == fp->ipq_src.s_addr && 911 ip->ip_dst.s_addr == fp->ipq_dst.s_addr && 912 #ifdef MAC 913 mac_ipq_match(m, fp) && 914 #endif 915 ip->ip_p == fp->ipq_p) 916 goto found; 917 918 fp = NULL; 919 920 /* 921 * Attempt to trim the number of allocated fragment queues if it 922 * exceeds the administrative limit. 923 */ 924 if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) { 925 /* 926 * drop something from the tail of the current queue 927 * before proceeding further 928 */ 929 struct ipq *q = TAILQ_LAST(head, ipqhead); 930 if (q == NULL) { /* gak */ 931 for (i = 0; i < IPREASS_NHASH; i++) { 932 struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead); 933 if (r) { 934 IPSTAT_ADD(ips_fragtimeout, 935 r->ipq_nfrags); 936 ip_freef(&V_ipq[i], r); 937 break; 938 } 939 } 940 } else { 941 IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags); 942 ip_freef(head, q); 943 } 944 } 945 946 found: 947 /* 948 * Adjust ip_len to not reflect header, 949 * convert offset of this to bytes. 950 */ 951 ip->ip_len = htons(ntohs(ip->ip_len) - hlen); 952 if (ip->ip_off & htons(IP_MF)) { 953 /* 954 * Make sure that fragments have a data length 955 * that's a non-zero multiple of 8 bytes. 956 */ 957 if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) { 958 IPSTAT_INC(ips_toosmall); /* XXX */ 959 goto dropfrag; 960 } 961 m->m_flags |= M_IP_FRAG; 962 } else 963 m->m_flags &= ~M_IP_FRAG; 964 ip->ip_off = htons(ntohs(ip->ip_off) << 3); 965 966 /* 967 * Attempt reassembly; if it succeeds, proceed. 968 * ip_reass() will return a different mbuf. 969 */ 970 IPSTAT_INC(ips_fragments); 971 m->m_pkthdr.PH_loc.ptr = ip; 972 973 /* Previous ip_reass() started here. */ 974 /* 975 * Presence of header sizes in mbufs 976 * would confuse code below. 977 */ 978 m->m_data += hlen; 979 m->m_len -= hlen; 980 981 /* 982 * If first fragment to arrive, create a reassembly queue. 983 */ 984 if (fp == NULL) { 985 fp = uma_zalloc(V_ipq_zone, M_NOWAIT); 986 if (fp == NULL) 987 goto dropfrag; 988 #ifdef MAC 989 if (mac_ipq_init(fp, M_NOWAIT) != 0) { 990 uma_zfree(V_ipq_zone, fp); 991 fp = NULL; 992 goto dropfrag; 993 } 994 mac_ipq_create(m, fp); 995 #endif 996 TAILQ_INSERT_HEAD(head, fp, ipq_list); 997 V_nipq++; 998 fp->ipq_nfrags = 1; 999 fp->ipq_ttl = IPFRAGTTL; 1000 fp->ipq_p = ip->ip_p; 1001 fp->ipq_id = ip->ip_id; 1002 fp->ipq_src = ip->ip_src; 1003 fp->ipq_dst = ip->ip_dst; 1004 fp->ipq_frags = m; 1005 m->m_nextpkt = NULL; 1006 goto done; 1007 } else { 1008 fp->ipq_nfrags++; 1009 #ifdef MAC 1010 mac_ipq_update(m, fp); 1011 #endif 1012 } 1013 1014 #define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) 1015 1016 /* 1017 * Handle ECN by comparing this segment with the first one; 1018 * if CE is set, do not lose CE. 1019 * drop if CE and not-ECT are mixed for the same packet. 1020 */ 1021 ecn = ip->ip_tos & IPTOS_ECN_MASK; 1022 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; 1023 if (ecn == IPTOS_ECN_CE) { 1024 if (ecn0 == IPTOS_ECN_NOTECT) 1025 goto dropfrag; 1026 if (ecn0 != IPTOS_ECN_CE) 1027 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; 1028 } 1029 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) 1030 goto dropfrag; 1031 1032 /* 1033 * Find a segment which begins after this one does. 1034 */ 1035 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) 1036 if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) 1037 break; 1038 1039 /* 1040 * If there is a preceding segment, it may provide some of 1041 * our data already. If so, drop the data from the incoming 1042 * segment. If it provides all of our data, drop us, otherwise 1043 * stick new segment in the proper place. 1044 * 1045 * If some of the data is dropped from the preceding 1046 * segment, then it's checksum is invalidated. 1047 */ 1048 if (p) { 1049 i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - 1050 ntohs(ip->ip_off); 1051 if (i > 0) { 1052 if (i >= ntohs(ip->ip_len)) 1053 goto dropfrag; 1054 m_adj(m, i); 1055 m->m_pkthdr.csum_flags = 0; 1056 ip->ip_off = htons(ntohs(ip->ip_off) + i); 1057 ip->ip_len = htons(ntohs(ip->ip_len) - i); 1058 } 1059 m->m_nextpkt = p->m_nextpkt; 1060 p->m_nextpkt = m; 1061 } else { 1062 m->m_nextpkt = fp->ipq_frags; 1063 fp->ipq_frags = m; 1064 } 1065 1066 /* 1067 * While we overlap succeeding segments trim them or, 1068 * if they are completely covered, dequeue them. 1069 */ 1070 for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > 1071 ntohs(GETIP(q)->ip_off); q = nq) { 1072 i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - 1073 ntohs(GETIP(q)->ip_off); 1074 if (i < ntohs(GETIP(q)->ip_len)) { 1075 GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); 1076 GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); 1077 m_adj(q, i); 1078 q->m_pkthdr.csum_flags = 0; 1079 break; 1080 } 1081 nq = q->m_nextpkt; 1082 m->m_nextpkt = nq; 1083 IPSTAT_INC(ips_fragdropped); 1084 fp->ipq_nfrags--; 1085 m_freem(q); 1086 } 1087 1088 /* 1089 * Check for complete reassembly and perform frag per packet 1090 * limiting. 1091 * 1092 * Frag limiting is performed here so that the nth frag has 1093 * a chance to complete the packet before we drop the packet. 1094 * As a result, n+1 frags are actually allowed per packet, but 1095 * only n will ever be stored. (n = maxfragsperpacket.) 1096 * 1097 */ 1098 next = 0; 1099 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { 1100 if (ntohs(GETIP(q)->ip_off) != next) { 1101 if (fp->ipq_nfrags > V_maxfragsperpacket) { 1102 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1103 ip_freef(head, fp); 1104 } 1105 goto done; 1106 } 1107 next += ntohs(GETIP(q)->ip_len); 1108 } 1109 /* Make sure the last packet didn't have the IP_MF flag */ 1110 if (p->m_flags & M_IP_FRAG) { 1111 if (fp->ipq_nfrags > V_maxfragsperpacket) { 1112 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1113 ip_freef(head, fp); 1114 } 1115 goto done; 1116 } 1117 1118 /* 1119 * Reassembly is complete. Make sure the packet is a sane size. 1120 */ 1121 q = fp->ipq_frags; 1122 ip = GETIP(q); 1123 if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { 1124 IPSTAT_INC(ips_toolong); 1125 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 1126 ip_freef(head, fp); 1127 goto done; 1128 } 1129 1130 /* 1131 * Concatenate fragments. 1132 */ 1133 m = q; 1134 t = m->m_next; 1135 m->m_next = NULL; 1136 m_cat(m, t); 1137 nq = q->m_nextpkt; 1138 q->m_nextpkt = NULL; 1139 for (q = nq; q != NULL; q = nq) { 1140 nq = q->m_nextpkt; 1141 q->m_nextpkt = NULL; 1142 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; 1143 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; 1144 m_cat(m, q); 1145 } 1146 /* 1147 * In order to do checksumming faster we do 'end-around carry' here 1148 * (and not in for{} loop), though it implies we are not going to 1149 * reassemble more than 64k fragments. 1150 */ 1151 while (m->m_pkthdr.csum_data & 0xffff0000) 1152 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 1153 (m->m_pkthdr.csum_data >> 16); 1154 #ifdef MAC 1155 mac_ipq_reassemble(fp, m); 1156 mac_ipq_destroy(fp); 1157 #endif 1158 1159 /* 1160 * Create header for new ip packet by modifying header of first 1161 * packet; dequeue and discard fragment reassembly header. 1162 * Make header visible. 1163 */ 1164 ip->ip_len = htons((ip->ip_hl << 2) + next); 1165 ip->ip_src = fp->ipq_src; 1166 ip->ip_dst = fp->ipq_dst; 1167 TAILQ_REMOVE(head, fp, ipq_list); 1168 V_nipq--; 1169 uma_zfree(V_ipq_zone, fp); 1170 m->m_len += (ip->ip_hl << 2); 1171 m->m_data -= (ip->ip_hl << 2); 1172 /* some debugging cruft by sklower, below, will go away soon */ 1173 if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ 1174 m_fixhdr(m); 1175 IPSTAT_INC(ips_reassembled); 1176 IPQ_UNLOCK(); 1177 1178 #ifdef RSS 1179 /* 1180 * Query the RSS layer for the flowid / flowtype for the 1181 * mbuf payload. 1182 * 1183 * For now, just assume we have to calculate a new one. 1184 * Later on we should check to see if the assigned flowid matches 1185 * what RSS wants for the given IP protocol and if so, just keep it. 1186 * 1187 * We then queue into the relevant netisr so it can be dispatched 1188 * to the correct CPU. 1189 * 1190 * Note - this may return 1, which means the flowid in the mbuf 1191 * is correct for the configured RSS hash types and can be used. 1192 */ 1193 if (rss_mbuf_software_hash_v4(m, 0, &rss_hash, &rss_type) == 0) { 1194 m->m_pkthdr.flowid = rss_hash; 1195 M_HASHTYPE_SET(m, rss_type); 1196 } 1197 1198 /* 1199 * Queue/dispatch for reprocessing. 1200 * 1201 * Note: this is much slower than just handling the frame in the 1202 * current receive context. It's likely worth investigating 1203 * why this is. 1204 */ 1205 netisr_dispatch(NETISR_IP_DIRECT, m); 1206 return (NULL); 1207 #endif 1208 1209 /* Handle in-line */ 1210 return (m); 1211 1212 dropfrag: 1213 IPSTAT_INC(ips_fragdropped); 1214 if (fp != NULL) 1215 fp->ipq_nfrags--; 1216 m_freem(m); 1217 done: 1218 IPQ_UNLOCK(); 1219 return (NULL); 1220 1221 #undef GETIP 1222 } 1223 1224 /* 1225 * Free a fragment reassembly header and all 1226 * associated datagrams. 1227 */ 1228 static void 1229 ip_freef(struct ipqhead *fhp, struct ipq *fp) 1230 { 1231 struct mbuf *q; 1232 1233 IPQ_LOCK_ASSERT(); 1234 1235 while (fp->ipq_frags) { 1236 q = fp->ipq_frags; 1237 fp->ipq_frags = q->m_nextpkt; 1238 m_freem(q); 1239 } 1240 TAILQ_REMOVE(fhp, fp, ipq_list); 1241 uma_zfree(V_ipq_zone, fp); 1242 V_nipq--; 1243 } 1244 1245 /* 1246 * IP timer processing; 1247 * if a timer expires on a reassembly 1248 * queue, discard it. 1249 */ 1250 void 1251 ip_slowtimo(void) 1252 { 1253 VNET_ITERATOR_DECL(vnet_iter); 1254 struct ipq *fp; 1255 int i; 1256 1257 VNET_LIST_RLOCK_NOSLEEP(); 1258 IPQ_LOCK(); 1259 VNET_FOREACH(vnet_iter) { 1260 CURVNET_SET(vnet_iter); 1261 for (i = 0; i < IPREASS_NHASH; i++) { 1262 for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) { 1263 struct ipq *fpp; 1264 1265 fpp = fp; 1266 fp = TAILQ_NEXT(fp, ipq_list); 1267 if(--fpp->ipq_ttl == 0) { 1268 IPSTAT_ADD(ips_fragtimeout, 1269 fpp->ipq_nfrags); 1270 ip_freef(&V_ipq[i], fpp); 1271 } 1272 } 1273 } 1274 /* 1275 * If we are over the maximum number of fragments 1276 * (due to the limit being lowered), drain off 1277 * enough to get down to the new limit. 1278 */ 1279 if (V_maxnipq >= 0 && V_nipq > V_maxnipq) { 1280 for (i = 0; i < IPREASS_NHASH; i++) { 1281 while (V_nipq > V_maxnipq && 1282 !TAILQ_EMPTY(&V_ipq[i])) { 1283 IPSTAT_ADD(ips_fragdropped, 1284 TAILQ_FIRST(&V_ipq[i])->ipq_nfrags); 1285 ip_freef(&V_ipq[i], 1286 TAILQ_FIRST(&V_ipq[i])); 1287 } 1288 } 1289 } 1290 CURVNET_RESTORE(); 1291 } 1292 IPQ_UNLOCK(); 1293 VNET_LIST_RUNLOCK_NOSLEEP(); 1294 } 1295 1296 /* 1297 * Drain off all datagram fragments. 1298 */ 1299 static void 1300 ip_drain_locked(void) 1301 { 1302 int i; 1303 1304 IPQ_LOCK_ASSERT(); 1305 1306 for (i = 0; i < IPREASS_NHASH; i++) { 1307 while(!TAILQ_EMPTY(&V_ipq[i])) { 1308 IPSTAT_ADD(ips_fragdropped, 1309 TAILQ_FIRST(&V_ipq[i])->ipq_nfrags); 1310 ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i])); 1311 } 1312 } 1313 } 1314 1315 void 1316 ip_drain(void) 1317 { 1318 VNET_ITERATOR_DECL(vnet_iter); 1319 1320 VNET_LIST_RLOCK_NOSLEEP(); 1321 IPQ_LOCK(); 1322 VNET_FOREACH(vnet_iter) { 1323 CURVNET_SET(vnet_iter); 1324 ip_drain_locked(); 1325 CURVNET_RESTORE(); 1326 } 1327 IPQ_UNLOCK(); 1328 VNET_LIST_RUNLOCK_NOSLEEP(); 1329 } 1330 1331 /* 1332 * The protocol to be inserted into ip_protox[] must be already registered 1333 * in inetsw[], either statically or through pf_proto_register(). 1334 */ 1335 int 1336 ipproto_register(short ipproto) 1337 { 1338 struct protosw *pr; 1339 1340 /* Sanity checks. */ 1341 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1342 return (EPROTONOSUPPORT); 1343 1344 /* 1345 * The protocol slot must not be occupied by another protocol 1346 * already. An index pointing to IPPROTO_RAW is unused. 1347 */ 1348 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1349 if (pr == NULL) 1350 return (EPFNOSUPPORT); 1351 if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */ 1352 return (EEXIST); 1353 1354 /* Find the protocol position in inetsw[] and set the index. */ 1355 for (pr = inetdomain.dom_protosw; 1356 pr < inetdomain.dom_protoswNPROTOSW; pr++) { 1357 if (pr->pr_domain->dom_family == PF_INET && 1358 pr->pr_protocol && pr->pr_protocol == ipproto) { 1359 ip_protox[pr->pr_protocol] = pr - inetsw; 1360 return (0); 1361 } 1362 } 1363 return (EPROTONOSUPPORT); 1364 } 1365 1366 int 1367 ipproto_unregister(short ipproto) 1368 { 1369 struct protosw *pr; 1370 1371 /* Sanity checks. */ 1372 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1373 return (EPROTONOSUPPORT); 1374 1375 /* Check if the protocol was indeed registered. */ 1376 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1377 if (pr == NULL) 1378 return (EPFNOSUPPORT); 1379 if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */ 1380 return (ENOENT); 1381 1382 /* Reset the protocol slot to IPPROTO_RAW. */ 1383 ip_protox[ipproto] = pr - inetsw; 1384 return (0); 1385 } 1386 1387 /* 1388 * Given address of next destination (final or next hop), return (referenced) 1389 * internet address info of interface to be used to get there. 1390 */ 1391 struct in_ifaddr * 1392 ip_rtaddr(struct in_addr dst, u_int fibnum) 1393 { 1394 struct route sro; 1395 struct sockaddr_in *sin; 1396 struct in_ifaddr *ia; 1397 1398 bzero(&sro, sizeof(sro)); 1399 sin = (struct sockaddr_in *)&sro.ro_dst; 1400 sin->sin_family = AF_INET; 1401 sin->sin_len = sizeof(*sin); 1402 sin->sin_addr = dst; 1403 in_rtalloc_ign(&sro, 0, fibnum); 1404 1405 if (sro.ro_rt == NULL) 1406 return (NULL); 1407 1408 ia = ifatoia(sro.ro_rt->rt_ifa); 1409 ifa_ref(&ia->ia_ifa); 1410 RTFREE(sro.ro_rt); 1411 return (ia); 1412 } 1413 1414 u_char inetctlerrmap[PRC_NCMDS] = { 1415 0, 0, 0, 0, 1416 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH, 1417 EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED, 1418 EMSGSIZE, EHOSTUNREACH, 0, 0, 1419 0, 0, EHOSTUNREACH, 0, 1420 ENOPROTOOPT, ECONNREFUSED 1421 }; 1422 1423 /* 1424 * Forward a packet. If some error occurs return the sender 1425 * an icmp packet. Note we can't always generate a meaningful 1426 * icmp message because icmp doesn't have a large enough repertoire 1427 * of codes and types. 1428 * 1429 * If not forwarding, just drop the packet. This could be confusing 1430 * if ipforwarding was zero but some routing protocol was advancing 1431 * us as a gateway to somewhere. However, we must let the routing 1432 * protocol deal with that. 1433 * 1434 * The srcrt parameter indicates whether the packet is being forwarded 1435 * via a source route. 1436 */ 1437 void 1438 ip_forward(struct mbuf *m, int srcrt) 1439 { 1440 struct ip *ip = mtod(m, struct ip *); 1441 struct in_ifaddr *ia; 1442 struct mbuf *mcopy; 1443 struct in_addr dest; 1444 struct route ro; 1445 int error, type = 0, code = 0, mtu = 0; 1446 1447 if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) { 1448 IPSTAT_INC(ips_cantforward); 1449 m_freem(m); 1450 return; 1451 } 1452 #ifdef IPSEC 1453 if (ip_ipsec_fwd(m) != 0) { 1454 IPSTAT_INC(ips_cantforward); 1455 m_freem(m); 1456 return; 1457 } 1458 #endif /* IPSEC */ 1459 #ifdef IPSTEALTH 1460 if (!V_ipstealth) { 1461 #endif 1462 if (ip->ip_ttl <= IPTTLDEC) { 1463 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 1464 0, 0); 1465 return; 1466 } 1467 #ifdef IPSTEALTH 1468 } 1469 #endif 1470 1471 ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m)); 1472 #ifndef IPSEC 1473 /* 1474 * 'ia' may be NULL if there is no route for this destination. 1475 * In case of IPsec, Don't discard it just yet, but pass it to 1476 * ip_output in case of outgoing IPsec policy. 1477 */ 1478 if (!srcrt && ia == NULL) { 1479 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 1480 return; 1481 } 1482 #endif 1483 1484 /* 1485 * Save the IP header and at most 8 bytes of the payload, 1486 * in case we need to generate an ICMP message to the src. 1487 * 1488 * XXX this can be optimized a lot by saving the data in a local 1489 * buffer on the stack (72 bytes at most), and only allocating the 1490 * mbuf if really necessary. The vast majority of the packets 1491 * are forwarded without having to send an ICMP back (either 1492 * because unnecessary, or because rate limited), so we are 1493 * really we are wasting a lot of work here. 1494 * 1495 * We don't use m_copy() because it might return a reference 1496 * to a shared cluster. Both this function and ip_output() 1497 * assume exclusive access to the IP header in `m', so any 1498 * data in a cluster may change before we reach icmp_error(). 1499 */ 1500 mcopy = m_gethdr(M_NOWAIT, m->m_type); 1501 if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) { 1502 /* 1503 * It's probably ok if the pkthdr dup fails (because 1504 * the deep copy of the tag chain failed), but for now 1505 * be conservative and just discard the copy since 1506 * code below may some day want the tags. 1507 */ 1508 m_free(mcopy); 1509 mcopy = NULL; 1510 } 1511 if (mcopy != NULL) { 1512 mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy)); 1513 mcopy->m_pkthdr.len = mcopy->m_len; 1514 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t)); 1515 } 1516 1517 #ifdef IPSTEALTH 1518 if (!V_ipstealth) { 1519 #endif 1520 ip->ip_ttl -= IPTTLDEC; 1521 #ifdef IPSTEALTH 1522 } 1523 #endif 1524 1525 /* 1526 * If forwarding packet using same interface that it came in on, 1527 * perhaps should send a redirect to sender to shortcut a hop. 1528 * Only send redirect if source is sending directly to us, 1529 * and if packet was not source routed (or has any options). 1530 * Also, don't send redirect if forwarding using a default route 1531 * or a route modified by a redirect. 1532 */ 1533 dest.s_addr = 0; 1534 if (!srcrt && V_ipsendredirects && 1535 ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) { 1536 struct sockaddr_in *sin; 1537 struct rtentry *rt; 1538 1539 bzero(&ro, sizeof(ro)); 1540 sin = (struct sockaddr_in *)&ro.ro_dst; 1541 sin->sin_family = AF_INET; 1542 sin->sin_len = sizeof(*sin); 1543 sin->sin_addr = ip->ip_dst; 1544 in_rtalloc_ign(&ro, 0, M_GETFIB(m)); 1545 1546 rt = ro.ro_rt; 1547 1548 if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 && 1549 satosin(rt_key(rt))->sin_addr.s_addr != 0) { 1550 #define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa)) 1551 u_long src = ntohl(ip->ip_src.s_addr); 1552 1553 if (RTA(rt) && 1554 (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) { 1555 if (rt->rt_flags & RTF_GATEWAY) 1556 dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr; 1557 else 1558 dest.s_addr = ip->ip_dst.s_addr; 1559 /* Router requirements says to only send host redirects */ 1560 type = ICMP_REDIRECT; 1561 code = ICMP_REDIRECT_HOST; 1562 } 1563 } 1564 if (rt) 1565 RTFREE(rt); 1566 } 1567 1568 /* 1569 * Try to cache the route MTU from ip_output so we can consider it for 1570 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191. 1571 */ 1572 bzero(&ro, sizeof(ro)); 1573 1574 error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL); 1575 1576 if (error == EMSGSIZE && ro.ro_rt) 1577 mtu = ro.ro_rt->rt_mtu; 1578 RO_RTFREE(&ro); 1579 1580 if (error) 1581 IPSTAT_INC(ips_cantforward); 1582 else { 1583 IPSTAT_INC(ips_forward); 1584 if (type) 1585 IPSTAT_INC(ips_redirectsent); 1586 else { 1587 if (mcopy) 1588 m_freem(mcopy); 1589 if (ia != NULL) 1590 ifa_free(&ia->ia_ifa); 1591 return; 1592 } 1593 } 1594 if (mcopy == NULL) { 1595 if (ia != NULL) 1596 ifa_free(&ia->ia_ifa); 1597 return; 1598 } 1599 1600 switch (error) { 1601 1602 case 0: /* forwarded, but need redirect */ 1603 /* type, code set above */ 1604 break; 1605 1606 case ENETUNREACH: 1607 case EHOSTUNREACH: 1608 case ENETDOWN: 1609 case EHOSTDOWN: 1610 default: 1611 type = ICMP_UNREACH; 1612 code = ICMP_UNREACH_HOST; 1613 break; 1614 1615 case EMSGSIZE: 1616 type = ICMP_UNREACH; 1617 code = ICMP_UNREACH_NEEDFRAG; 1618 1619 #ifdef IPSEC 1620 /* 1621 * If IPsec is configured for this path, 1622 * override any possibly mtu value set by ip_output. 1623 */ 1624 mtu = ip_ipsec_mtu(mcopy, mtu); 1625 #endif /* IPSEC */ 1626 /* 1627 * If the MTU was set before make sure we are below the 1628 * interface MTU. 1629 * If the MTU wasn't set before use the interface mtu or 1630 * fall back to the next smaller mtu step compared to the 1631 * current packet size. 1632 */ 1633 if (mtu != 0) { 1634 if (ia != NULL) 1635 mtu = min(mtu, ia->ia_ifp->if_mtu); 1636 } else { 1637 if (ia != NULL) 1638 mtu = ia->ia_ifp->if_mtu; 1639 else 1640 mtu = ip_next_mtu(ntohs(ip->ip_len), 0); 1641 } 1642 IPSTAT_INC(ips_cantfrag); 1643 break; 1644 1645 case ENOBUFS: 1646 case EACCES: /* ipfw denied packet */ 1647 m_freem(mcopy); 1648 if (ia != NULL) 1649 ifa_free(&ia->ia_ifa); 1650 return; 1651 } 1652 if (ia != NULL) 1653 ifa_free(&ia->ia_ifa); 1654 icmp_error(mcopy, type, code, dest.s_addr, mtu); 1655 } 1656 1657 void 1658 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip, 1659 struct mbuf *m) 1660 { 1661 1662 if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) { 1663 struct bintime bt; 1664 1665 bintime(&bt); 1666 if (inp->inp_socket->so_options & SO_BINTIME) { 1667 *mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt), 1668 SCM_BINTIME, SOL_SOCKET); 1669 if (*mp) 1670 mp = &(*mp)->m_next; 1671 } 1672 if (inp->inp_socket->so_options & SO_TIMESTAMP) { 1673 struct timeval tv; 1674 1675 bintime2timeval(&bt, &tv); 1676 *mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv), 1677 SCM_TIMESTAMP, SOL_SOCKET); 1678 if (*mp) 1679 mp = &(*mp)->m_next; 1680 } 1681 } 1682 if (inp->inp_flags & INP_RECVDSTADDR) { 1683 *mp = sbcreatecontrol((caddr_t)&ip->ip_dst, 1684 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP); 1685 if (*mp) 1686 mp = &(*mp)->m_next; 1687 } 1688 if (inp->inp_flags & INP_RECVTTL) { 1689 *mp = sbcreatecontrol((caddr_t)&ip->ip_ttl, 1690 sizeof(u_char), IP_RECVTTL, IPPROTO_IP); 1691 if (*mp) 1692 mp = &(*mp)->m_next; 1693 } 1694 #ifdef notyet 1695 /* XXX 1696 * Moving these out of udp_input() made them even more broken 1697 * than they already were. 1698 */ 1699 /* options were tossed already */ 1700 if (inp->inp_flags & INP_RECVOPTS) { 1701 *mp = sbcreatecontrol((caddr_t)opts_deleted_above, 1702 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP); 1703 if (*mp) 1704 mp = &(*mp)->m_next; 1705 } 1706 /* ip_srcroute doesn't do what we want here, need to fix */ 1707 if (inp->inp_flags & INP_RECVRETOPTS) { 1708 *mp = sbcreatecontrol((caddr_t)ip_srcroute(m), 1709 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP); 1710 if (*mp) 1711 mp = &(*mp)->m_next; 1712 } 1713 #endif 1714 if (inp->inp_flags & INP_RECVIF) { 1715 struct ifnet *ifp; 1716 struct sdlbuf { 1717 struct sockaddr_dl sdl; 1718 u_char pad[32]; 1719 } sdlbuf; 1720 struct sockaddr_dl *sdp; 1721 struct sockaddr_dl *sdl2 = &sdlbuf.sdl; 1722 1723 if ((ifp = m->m_pkthdr.rcvif) && 1724 ifp->if_index && ifp->if_index <= V_if_index) { 1725 sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr; 1726 /* 1727 * Change our mind and don't try copy. 1728 */ 1729 if (sdp->sdl_family != AF_LINK || 1730 sdp->sdl_len > sizeof(sdlbuf)) { 1731 goto makedummy; 1732 } 1733 bcopy(sdp, sdl2, sdp->sdl_len); 1734 } else { 1735 makedummy: 1736 sdl2->sdl_len = 1737 offsetof(struct sockaddr_dl, sdl_data[0]); 1738 sdl2->sdl_family = AF_LINK; 1739 sdl2->sdl_index = 0; 1740 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0; 1741 } 1742 *mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len, 1743 IP_RECVIF, IPPROTO_IP); 1744 if (*mp) 1745 mp = &(*mp)->m_next; 1746 } 1747 if (inp->inp_flags & INP_RECVTOS) { 1748 *mp = sbcreatecontrol((caddr_t)&ip->ip_tos, 1749 sizeof(u_char), IP_RECVTOS, IPPROTO_IP); 1750 if (*mp) 1751 mp = &(*mp)->m_next; 1752 } 1753 1754 if (inp->inp_flags2 & INP_RECVFLOWID) { 1755 uint32_t flowid, flow_type; 1756 1757 flowid = m->m_pkthdr.flowid; 1758 flow_type = M_HASHTYPE_GET(m); 1759 1760 /* 1761 * XXX should handle the failure of one or the 1762 * other - don't populate both? 1763 */ 1764 *mp = sbcreatecontrol((caddr_t) &flowid, 1765 sizeof(uint32_t), IP_FLOWID, IPPROTO_IP); 1766 if (*mp) 1767 mp = &(*mp)->m_next; 1768 *mp = sbcreatecontrol((caddr_t) &flow_type, 1769 sizeof(uint32_t), IP_FLOWTYPE, IPPROTO_IP); 1770 if (*mp) 1771 mp = &(*mp)->m_next; 1772 } 1773 1774 #ifdef RSS 1775 if (inp->inp_flags2 & INP_RECVRSSBUCKETID) { 1776 uint32_t flowid, flow_type; 1777 uint32_t rss_bucketid; 1778 1779 flowid = m->m_pkthdr.flowid; 1780 flow_type = M_HASHTYPE_GET(m); 1781 1782 if (rss_hash2bucket(flowid, flow_type, &rss_bucketid) == 0) { 1783 *mp = sbcreatecontrol((caddr_t) &rss_bucketid, 1784 sizeof(uint32_t), IP_RSSBUCKETID, IPPROTO_IP); 1785 if (*mp) 1786 mp = &(*mp)->m_next; 1787 } 1788 } 1789 #endif 1790 } 1791 1792 /* 1793 * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the 1794 * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on 1795 * locking. This code remains in ip_input.c as ip_mroute.c is optionally 1796 * compiled. 1797 */ 1798 static VNET_DEFINE(int, ip_rsvp_on); 1799 VNET_DEFINE(struct socket *, ip_rsvpd); 1800 1801 #define V_ip_rsvp_on VNET(ip_rsvp_on) 1802 1803 int 1804 ip_rsvp_init(struct socket *so) 1805 { 1806 1807 if (so->so_type != SOCK_RAW || 1808 so->so_proto->pr_protocol != IPPROTO_RSVP) 1809 return EOPNOTSUPP; 1810 1811 if (V_ip_rsvpd != NULL) 1812 return EADDRINUSE; 1813 1814 V_ip_rsvpd = so; 1815 /* 1816 * This may seem silly, but we need to be sure we don't over-increment 1817 * the RSVP counter, in case something slips up. 1818 */ 1819 if (!V_ip_rsvp_on) { 1820 V_ip_rsvp_on = 1; 1821 V_rsvp_on++; 1822 } 1823 1824 return 0; 1825 } 1826 1827 int 1828 ip_rsvp_done(void) 1829 { 1830 1831 V_ip_rsvpd = NULL; 1832 /* 1833 * This may seem silly, but we need to be sure we don't over-decrement 1834 * the RSVP counter, in case something slips up. 1835 */ 1836 if (V_ip_rsvp_on) { 1837 V_ip_rsvp_on = 0; 1838 V_rsvp_on--; 1839 } 1840 return 0; 1841 } 1842 1843 int 1844 rsvp_input(struct mbuf **mp, int *offp, int proto) 1845 { 1846 struct mbuf *m; 1847 1848 m = *mp; 1849 *mp = NULL; 1850 1851 if (rsvp_input_p) { /* call the real one if loaded */ 1852 *mp = m; 1853 rsvp_input_p(mp, offp, proto); 1854 return (IPPROTO_DONE); 1855 } 1856 1857 /* Can still get packets with rsvp_on = 0 if there is a local member 1858 * of the group to which the RSVP packet is addressed. But in this 1859 * case we want to throw the packet away. 1860 */ 1861 1862 if (!V_rsvp_on) { 1863 m_freem(m); 1864 return (IPPROTO_DONE); 1865 } 1866 1867 if (V_ip_rsvpd != NULL) { 1868 *mp = m; 1869 rip_input(mp, offp, proto); 1870 return (IPPROTO_DONE); 1871 } 1872 /* Drop the packet */ 1873 m_freem(m); 1874 return (IPPROTO_DONE); 1875 } 1876