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