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 /* 109 * XXX - Setting ip_checkinterface mostly implements the receive side of 110 * the Strong ES model described in RFC 1122, but since the routing table 111 * and transmit implementation do not implement the Strong ES model, 112 * setting this to 1 results in an odd hybrid. 113 * 114 * XXX - ip_checkinterface currently must be disabled if you use ipnat 115 * to translate the destination address to another local interface. 116 * 117 * XXX - ip_checkinterface must be disabled if you add IP aliases 118 * to the loopback interface instead of the interface where the 119 * packets for those addresses are received. 120 */ 121 static VNET_DEFINE(int, ip_checkinterface); 122 #define V_ip_checkinterface VNET(ip_checkinterface) 123 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_VNET | CTLFLAG_RW, 124 &VNET_NAME(ip_checkinterface), 0, 125 "Verify packet arrives on correct interface"); 126 127 VNET_DEFINE(struct pfil_head, inet_pfil_hook); /* Packet filter hooks */ 128 129 static struct netisr_handler ip_nh = { 130 .nh_name = "ip", 131 .nh_handler = ip_input, 132 .nh_proto = NETISR_IP, 133 #ifdef RSS 134 .nh_m2cpuid = rss_soft_m2cpuid, 135 .nh_policy = NETISR_POLICY_CPU, 136 .nh_dispatch = NETISR_DISPATCH_HYBRID, 137 #else 138 .nh_policy = NETISR_POLICY_FLOW, 139 #endif 140 }; 141 142 #ifdef RSS 143 /* 144 * Directly dispatched frames are currently assumed 145 * to have a flowid already calculated. 146 * 147 * It should likely have something that assert it 148 * actually has valid flow details. 149 */ 150 static struct netisr_handler ip_direct_nh = { 151 .nh_name = "ip_direct", 152 .nh_handler = ip_direct_input, 153 .nh_proto = NETISR_IP_DIRECT, 154 .nh_m2cpuid = rss_m2cpuid, 155 .nh_policy = NETISR_POLICY_CPU, 156 .nh_dispatch = NETISR_DISPATCH_HYBRID, 157 }; 158 #endif 159 160 extern struct domain inetdomain; 161 extern struct protosw inetsw[]; 162 u_char ip_protox[IPPROTO_MAX]; 163 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead); /* first inet address */ 164 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table */ 165 VNET_DEFINE(u_long, in_ifaddrhmask); /* mask for hash table */ 166 167 static VNET_DEFINE(uma_zone_t, ipq_zone); 168 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]); 169 static struct mtx_padalign ipqlock[IPREASS_NHASH]; 170 171 #define V_ipq_zone VNET(ipq_zone) 172 #define V_ipq VNET(ipq) 173 174 /* 175 * The ipqlock array is global, /not/ per-VNET. 176 */ 177 #define IPQ_LOCK(i) mtx_lock(&ipqlock[(i)]) 178 #define IPQ_UNLOCK(i) mtx_unlock(&ipqlock[(i)]) 179 #define IPQ_LOCK_INIT(i) mtx_init(&ipqlock[(i)], "ipqlock", NULL, MTX_DEF) 180 181 static void maxnipq_update(void); 182 static void ipq_zone_change(void *); 183 static void ip_drain_vnet(void); 184 static void ipq_free(struct ipqhead *, struct ipq *); 185 186 static inline void 187 ipq_timeout(struct ipqhead *head, struct ipq *fp) 188 { 189 190 IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); 191 ipq_free(head, fp); 192 } 193 194 static inline void 195 ipq_drop(struct ipqhead *head, struct ipq *fp) 196 { 197 198 IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 199 ipq_free(head, fp); 200 } 201 202 static VNET_DEFINE(int, maxnipq); /* Administrative limit on # reass queues. */ 203 static VNET_DEFINE(int, nipq); /* Total # of reass queues */ 204 #define V_maxnipq VNET(maxnipq) 205 #define V_nipq VNET(nipq) 206 SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_VNET | CTLFLAG_RD, 207 &VNET_NAME(nipq), 0, 208 "Current number of IPv4 fragment reassembly queue entries"); 209 210 static VNET_DEFINE(int, maxfragsperpacket); 211 #define V_maxfragsperpacket VNET(maxfragsperpacket) 212 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(maxfragsperpacket), 0, 214 "Maximum number of IPv4 fragments allowed per packet"); 215 216 #ifdef IPCTL_DEFMTU 217 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW, 218 &ip_mtu, 0, "Default MTU"); 219 #endif 220 221 #ifdef IPSTEALTH 222 VNET_DEFINE(int, ipstealth); 223 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(ipstealth), 0, 225 "IP stealth mode, no TTL decrementation on forwarding"); 226 #endif 227 228 /* 229 * IP statistics are stored in the "array" of counter(9)s. 230 */ 231 VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat); 232 VNET_PCPUSTAT_SYSINIT(ipstat); 233 SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat, 234 "IP statistics (struct ipstat, netinet/ip_var.h)"); 235 236 #ifdef VIMAGE 237 VNET_PCPUSTAT_SYSUNINIT(ipstat); 238 #endif /* VIMAGE */ 239 240 /* 241 * Kernel module interface for updating ipstat. The argument is an index 242 * into ipstat treated as an array. 243 */ 244 void 245 kmod_ipstat_inc(int statnum) 246 { 247 248 counter_u64_add(VNET(ipstat)[statnum], 1); 249 } 250 251 void 252 kmod_ipstat_dec(int statnum) 253 { 254 255 counter_u64_add(VNET(ipstat)[statnum], -1); 256 } 257 258 static int 259 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS) 260 { 261 int error, qlimit; 262 263 netisr_getqlimit(&ip_nh, &qlimit); 264 error = sysctl_handle_int(oidp, &qlimit, 0, req); 265 if (error || !req->newptr) 266 return (error); 267 if (qlimit < 1) 268 return (EINVAL); 269 return (netisr_setqlimit(&ip_nh, qlimit)); 270 } 271 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, 272 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I", 273 "Maximum size of the IP input queue"); 274 275 static int 276 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS) 277 { 278 u_int64_t qdrops_long; 279 int error, qdrops; 280 281 netisr_getqdrops(&ip_nh, &qdrops_long); 282 qdrops = qdrops_long; 283 error = sysctl_handle_int(oidp, &qdrops, 0, req); 284 if (error || !req->newptr) 285 return (error); 286 if (qdrops != 0) 287 return (EINVAL); 288 netisr_clearqdrops(&ip_nh); 289 return (0); 290 } 291 292 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, 293 CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I", 294 "Number of packets dropped from the IP input queue"); 295 296 #ifdef RSS 297 static int 298 sysctl_netinet_intr_direct_queue_maxlen(SYSCTL_HANDLER_ARGS) 299 { 300 int error, qlimit; 301 302 netisr_getqlimit(&ip_direct_nh, &qlimit); 303 error = sysctl_handle_int(oidp, &qlimit, 0, req); 304 if (error || !req->newptr) 305 return (error); 306 if (qlimit < 1) 307 return (EINVAL); 308 return (netisr_setqlimit(&ip_direct_nh, qlimit)); 309 } 310 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_direct_queue_maxlen, 311 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_direct_queue_maxlen, "I", 312 "Maximum size of the IP direct input queue"); 313 314 static int 315 sysctl_netinet_intr_direct_queue_drops(SYSCTL_HANDLER_ARGS) 316 { 317 u_int64_t qdrops_long; 318 int error, qdrops; 319 320 netisr_getqdrops(&ip_direct_nh, &qdrops_long); 321 qdrops = qdrops_long; 322 error = sysctl_handle_int(oidp, &qdrops, 0, req); 323 if (error || !req->newptr) 324 return (error); 325 if (qdrops != 0) 326 return (EINVAL); 327 netisr_clearqdrops(&ip_direct_nh); 328 return (0); 329 } 330 331 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_direct_queue_drops, 332 CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_direct_queue_drops, "I", 333 "Number of packets dropped from the IP direct input queue"); 334 #endif /* RSS */ 335 336 /* 337 * IP initialization: fill in IP protocol switch table. 338 * All protocols not implemented in kernel go to raw IP protocol handler. 339 */ 340 void 341 ip_init(void) 342 { 343 struct protosw *pr; 344 int i; 345 346 TAILQ_INIT(&V_in_ifaddrhead); 347 V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask); 348 349 /* Initialize IP reassembly queue. */ 350 for (i = 0; i < IPREASS_NHASH; i++) 351 TAILQ_INIT(&V_ipq[i]); 352 V_maxnipq = nmbclusters / 32; 353 V_maxfragsperpacket = 16; 354 V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, 355 NULL, UMA_ALIGN_PTR, 0); 356 maxnipq_update(); 357 358 /* Initialize packet filter hooks. */ 359 V_inet_pfil_hook.ph_type = PFIL_TYPE_AF; 360 V_inet_pfil_hook.ph_af = AF_INET; 361 if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0) 362 printf("%s: WARNING: unable to register pfil hook, " 363 "error %d\n", __func__, i); 364 365 /* Skip initialization of globals for non-default instances. */ 366 if (!IS_DEFAULT_VNET(curvnet)) 367 return; 368 369 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 370 if (pr == NULL) 371 panic("ip_init: PF_INET not found"); 372 373 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */ 374 for (i = 0; i < IPPROTO_MAX; i++) 375 ip_protox[i] = pr - inetsw; 376 /* 377 * Cycle through IP protocols and put them into the appropriate place 378 * in ip_protox[]. 379 */ 380 for (pr = inetdomain.dom_protosw; 381 pr < inetdomain.dom_protoswNPROTOSW; pr++) 382 if (pr->pr_domain->dom_family == PF_INET && 383 pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) { 384 /* Be careful to only index valid IP protocols. */ 385 if (pr->pr_protocol < IPPROTO_MAX) 386 ip_protox[pr->pr_protocol] = pr - inetsw; 387 } 388 389 EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change, 390 NULL, EVENTHANDLER_PRI_ANY); 391 392 /* Initialize various other remaining things. */ 393 for (i = 0; i < IPREASS_NHASH; i++) 394 IPQ_LOCK_INIT(i); 395 netisr_register(&ip_nh); 396 #ifdef RSS 397 netisr_register(&ip_direct_nh); 398 #endif 399 } 400 401 #ifdef VIMAGE 402 void 403 ip_destroy(void) 404 { 405 int i; 406 407 if ((i = pfil_head_unregister(&V_inet_pfil_hook)) != 0) 408 printf("%s: WARNING: unable to unregister pfil hook, " 409 "error %d\n", __func__, i); 410 411 /* Cleanup in_ifaddr hash table; should be empty. */ 412 hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask); 413 414 ip_drain_vnet(); 415 416 uma_zdestroy(V_ipq_zone); 417 } 418 #endif 419 420 #ifdef RSS 421 /* 422 * IP direct input routine. 423 * 424 * This is called when reinjecting completed fragments where 425 * all of the previous checking and book-keeping has been done. 426 */ 427 void 428 ip_direct_input(struct mbuf *m) 429 { 430 struct ip *ip; 431 int hlen; 432 433 ip = mtod(m, struct ip *); 434 hlen = ip->ip_hl << 2; 435 436 IPSTAT_INC(ips_delivered); 437 (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); 438 return; 439 } 440 #endif 441 442 /* 443 * Ip input routine. Checksum and byte swap header. If fragmented 444 * try to reassemble. Process options. Pass to next level. 445 */ 446 void 447 ip_input(struct mbuf *m) 448 { 449 struct ip *ip = NULL; 450 struct in_ifaddr *ia = NULL; 451 struct ifaddr *ifa; 452 struct ifnet *ifp; 453 int checkif, hlen = 0; 454 uint16_t sum, ip_len; 455 int dchg = 0; /* dest changed after fw */ 456 struct in_addr odst; /* original dst address */ 457 458 M_ASSERTPKTHDR(m); 459 460 if (m->m_flags & M_FASTFWD_OURS) { 461 m->m_flags &= ~M_FASTFWD_OURS; 462 /* Set up some basics that will be used later. */ 463 ip = mtod(m, struct ip *); 464 hlen = ip->ip_hl << 2; 465 ip_len = ntohs(ip->ip_len); 466 goto ours; 467 } 468 469 IPSTAT_INC(ips_total); 470 471 if (m->m_pkthdr.len < sizeof(struct ip)) 472 goto tooshort; 473 474 if (m->m_len < sizeof (struct ip) && 475 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 476 IPSTAT_INC(ips_toosmall); 477 return; 478 } 479 ip = mtod(m, struct ip *); 480 481 if (ip->ip_v != IPVERSION) { 482 IPSTAT_INC(ips_badvers); 483 goto bad; 484 } 485 486 hlen = ip->ip_hl << 2; 487 if (hlen < sizeof(struct ip)) { /* minimum header length */ 488 IPSTAT_INC(ips_badhlen); 489 goto bad; 490 } 491 if (hlen > m->m_len) { 492 if ((m = m_pullup(m, hlen)) == NULL) { 493 IPSTAT_INC(ips_badhlen); 494 return; 495 } 496 ip = mtod(m, struct ip *); 497 } 498 499 IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL); 500 501 /* 127/8 must not appear on wire - RFC1122 */ 502 ifp = m->m_pkthdr.rcvif; 503 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 504 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 505 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 506 IPSTAT_INC(ips_badaddr); 507 goto bad; 508 } 509 } 510 511 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { 512 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); 513 } else { 514 if (hlen == sizeof(struct ip)) { 515 sum = in_cksum_hdr(ip); 516 } else { 517 sum = in_cksum(m, hlen); 518 } 519 } 520 if (sum) { 521 IPSTAT_INC(ips_badsum); 522 goto bad; 523 } 524 525 #ifdef ALTQ 526 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) 527 /* packet is dropped by traffic conditioner */ 528 return; 529 #endif 530 531 ip_len = ntohs(ip->ip_len); 532 if (ip_len < hlen) { 533 IPSTAT_INC(ips_badlen); 534 goto bad; 535 } 536 537 /* 538 * Check that the amount of data in the buffers 539 * is as at least much as the IP header would have us expect. 540 * Trim mbufs if longer than we expect. 541 * Drop packet if shorter than we expect. 542 */ 543 if (m->m_pkthdr.len < ip_len) { 544 tooshort: 545 IPSTAT_INC(ips_tooshort); 546 goto bad; 547 } 548 if (m->m_pkthdr.len > ip_len) { 549 if (m->m_len == m->m_pkthdr.len) { 550 m->m_len = ip_len; 551 m->m_pkthdr.len = ip_len; 552 } else 553 m_adj(m, ip_len - m->m_pkthdr.len); 554 } 555 556 #ifdef IPSEC 557 /* 558 * Bypass packet filtering for packets previously handled by IPsec. 559 */ 560 if (ip_ipsec_filtertunnel(m)) 561 goto passin; 562 #endif /* IPSEC */ 563 564 /* 565 * Run through list of hooks for input packets. 566 * 567 * NB: Beware of the destination address changing (e.g. 568 * by NAT rewriting). When this happens, tell 569 * ip_forward to do the right thing. 570 */ 571 572 /* Jump over all PFIL processing if hooks are not active. */ 573 if (!PFIL_HOOKED(&V_inet_pfil_hook)) 574 goto passin; 575 576 odst = ip->ip_dst; 577 if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0) 578 return; 579 if (m == NULL) /* consumed by filter */ 580 return; 581 582 ip = mtod(m, struct ip *); 583 dchg = (odst.s_addr != ip->ip_dst.s_addr); 584 ifp = m->m_pkthdr.rcvif; 585 586 if (m->m_flags & M_FASTFWD_OURS) { 587 m->m_flags &= ~M_FASTFWD_OURS; 588 goto ours; 589 } 590 if (m->m_flags & M_IP_NEXTHOP) { 591 dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL); 592 if (dchg != 0) { 593 /* 594 * Directly ship the packet on. This allows 595 * forwarding packets originally destined to us 596 * to some other directly connected host. 597 */ 598 ip_forward(m, 1); 599 return; 600 } 601 } 602 passin: 603 604 /* 605 * Process options and, if not destined for us, 606 * ship it on. ip_dooptions returns 1 when an 607 * error was detected (causing an icmp message 608 * to be sent and the original packet to be freed). 609 */ 610 if (hlen > sizeof (struct ip) && ip_dooptions(m, 0)) 611 return; 612 613 /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no 614 * matter if it is destined to another node, or whether it is 615 * a multicast one, RSVP wants it! and prevents it from being forwarded 616 * anywhere else. Also checks if the rsvp daemon is running before 617 * grabbing the packet. 618 */ 619 if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP) 620 goto ours; 621 622 /* 623 * Check our list of addresses, to see if the packet is for us. 624 * If we don't have any addresses, assume any unicast packet 625 * we receive might be for us (and let the upper layers deal 626 * with it). 627 */ 628 if (TAILQ_EMPTY(&V_in_ifaddrhead) && 629 (m->m_flags & (M_MCAST|M_BCAST)) == 0) 630 goto ours; 631 632 /* 633 * Enable a consistency check between the destination address 634 * and the arrival interface for a unicast packet (the RFC 1122 635 * strong ES model) if IP forwarding is disabled and the packet 636 * is not locally generated and the packet is not subject to 637 * 'ipfw fwd'. 638 * 639 * XXX - Checking also should be disabled if the destination 640 * address is ipnat'ed to a different interface. 641 * 642 * XXX - Checking is incompatible with IP aliases added 643 * to the loopback interface instead of the interface where 644 * the packets are received. 645 * 646 * XXX - This is the case for carp vhost IPs as well so we 647 * insert a workaround. If the packet got here, we already 648 * checked with carp_iamatch() and carp_forus(). 649 */ 650 checkif = V_ip_checkinterface && (V_ipforwarding == 0) && 651 ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) && 652 ifp->if_carp == NULL && (dchg == 0); 653 654 /* 655 * Check for exact addresses in the hash bucket. 656 */ 657 /* IN_IFADDR_RLOCK(); */ 658 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { 659 /* 660 * If the address matches, verify that the packet 661 * arrived via the correct interface if checking is 662 * enabled. 663 */ 664 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr && 665 (!checkif || ia->ia_ifp == ifp)) { 666 counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); 667 counter_u64_add(ia->ia_ifa.ifa_ibytes, 668 m->m_pkthdr.len); 669 /* IN_IFADDR_RUNLOCK(); */ 670 goto ours; 671 } 672 } 673 /* IN_IFADDR_RUNLOCK(); */ 674 675 /* 676 * Check for broadcast addresses. 677 * 678 * Only accept broadcast packets that arrive via the matching 679 * interface. Reception of forwarded directed broadcasts would 680 * be handled via ip_forward() and ether_output() with the loopback 681 * into the stack for SIMPLEX interfaces handled by ether_output(). 682 */ 683 if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { 684 IF_ADDR_RLOCK(ifp); 685 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 686 if (ifa->ifa_addr->sa_family != AF_INET) 687 continue; 688 ia = ifatoia(ifa); 689 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == 690 ip->ip_dst.s_addr) { 691 counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); 692 counter_u64_add(ia->ia_ifa.ifa_ibytes, 693 m->m_pkthdr.len); 694 IF_ADDR_RUNLOCK(ifp); 695 goto ours; 696 } 697 #ifdef BOOTP_COMPAT 698 if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) { 699 counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); 700 counter_u64_add(ia->ia_ifa.ifa_ibytes, 701 m->m_pkthdr.len); 702 IF_ADDR_RUNLOCK(ifp); 703 goto ours; 704 } 705 #endif 706 } 707 IF_ADDR_RUNLOCK(ifp); 708 ia = NULL; 709 } 710 /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */ 711 if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) { 712 IPSTAT_INC(ips_cantforward); 713 m_freem(m); 714 return; 715 } 716 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 717 if (V_ip_mrouter) { 718 /* 719 * If we are acting as a multicast router, all 720 * incoming multicast packets are passed to the 721 * kernel-level multicast forwarding function. 722 * The packet is returned (relatively) intact; if 723 * ip_mforward() returns a non-zero value, the packet 724 * must be discarded, else it may be accepted below. 725 */ 726 if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) { 727 IPSTAT_INC(ips_cantforward); 728 m_freem(m); 729 return; 730 } 731 732 /* 733 * The process-level routing daemon needs to receive 734 * all multicast IGMP packets, whether or not this 735 * host belongs to their destination groups. 736 */ 737 if (ip->ip_p == IPPROTO_IGMP) 738 goto ours; 739 IPSTAT_INC(ips_forward); 740 } 741 /* 742 * Assume the packet is for us, to avoid prematurely taking 743 * a lock on the in_multi hash. Protocols must perform 744 * their own filtering and update statistics accordingly. 745 */ 746 goto ours; 747 } 748 if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST) 749 goto ours; 750 if (ip->ip_dst.s_addr == INADDR_ANY) 751 goto ours; 752 753 /* 754 * Not for us; forward if possible and desirable. 755 */ 756 if (V_ipforwarding == 0) { 757 IPSTAT_INC(ips_cantforward); 758 m_freem(m); 759 } else { 760 ip_forward(m, dchg); 761 } 762 return; 763 764 ours: 765 #ifdef IPSTEALTH 766 /* 767 * IPSTEALTH: Process non-routing options only 768 * if the packet is destined for us. 769 */ 770 if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) 771 return; 772 #endif /* IPSTEALTH */ 773 774 /* 775 * Attempt reassembly; if it succeeds, proceed. 776 * ip_reass() will return a different mbuf. 777 */ 778 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) { 779 /* XXXGL: shouldn't we save & set m_flags? */ 780 m = ip_reass(m); 781 if (m == NULL) 782 return; 783 ip = mtod(m, struct ip *); 784 /* Get the header length of the reassembled packet */ 785 hlen = ip->ip_hl << 2; 786 } 787 788 #ifdef IPSEC 789 /* 790 * enforce IPsec policy checking if we are seeing last header. 791 * note that we do not visit this with protocols with pcb layer 792 * code - like udp/tcp/raw ip. 793 */ 794 if (ip_ipsec_input(m, ip->ip_p) != 0) 795 goto bad; 796 #endif /* IPSEC */ 797 798 /* 799 * Switch out to protocol's input routine. 800 */ 801 IPSTAT_INC(ips_delivered); 802 803 (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); 804 return; 805 bad: 806 m_freem(m); 807 } 808 809 /* 810 * After maxnipq has been updated, propagate the change to UMA. The UMA zone 811 * max has slightly different semantics than the sysctl, for historical 812 * reasons. 813 */ 814 static void 815 maxnipq_update(void) 816 { 817 818 /* 819 * -1 for unlimited allocation. 820 */ 821 if (V_maxnipq < 0) 822 uma_zone_set_max(V_ipq_zone, 0); 823 /* 824 * Positive number for specific bound. 825 */ 826 if (V_maxnipq > 0) 827 uma_zone_set_max(V_ipq_zone, V_maxnipq); 828 /* 829 * Zero specifies no further fragment queue allocation. 830 */ 831 if (V_maxnipq == 0) { 832 uma_zone_set_max(V_ipq_zone, 1); 833 ip_drain_vnet(); 834 } 835 } 836 837 static void 838 ipq_zone_change(void *tag) 839 { 840 841 if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) { 842 V_maxnipq = nmbclusters / 32; 843 maxnipq_update(); 844 } 845 } 846 847 static int 848 sysctl_maxnipq(SYSCTL_HANDLER_ARGS) 849 { 850 int error, i; 851 852 i = V_maxnipq; 853 error = sysctl_handle_int(oidp, &i, 0, req); 854 if (error || !req->newptr) 855 return (error); 856 857 /* 858 * XXXRW: Might be a good idea to sanity check the argument and place 859 * an extreme upper bound. 860 */ 861 if (i < -1) 862 return (EINVAL); 863 V_maxnipq = i; 864 maxnipq_update(); 865 return (0); 866 } 867 868 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW, 869 NULL, 0, sysctl_maxnipq, "I", 870 "Maximum number of IPv4 fragment reassembly queue entries"); 871 872 #define M_IP_FRAG M_PROTO9 873 874 /* 875 * Attempt to purge something from the reassembly queue to make 876 * room. 877 * 878 * Must be called without any IPQ locks held, as it will attempt 879 * to lock each in turn. 880 * 881 * 'skip_bucket' is the bucket with which to skip over, or -1 to 882 * not skip over anything. 883 * 884 * Returns the bucket being freed, or -1 for no action. 885 */ 886 static int 887 ip_reass_purge_element(int skip_bucket) 888 { 889 int i; 890 struct ipq *r; 891 892 for (i = 0; i < IPREASS_NHASH; i++) { 893 if (skip_bucket > -1 && i == skip_bucket) 894 continue; 895 IPQ_LOCK(i); 896 r = TAILQ_LAST(&V_ipq[i], ipqhead); 897 if (r) { 898 ipq_timeout(&V_ipq[i], r); 899 IPQ_UNLOCK(i); 900 return (i); 901 } 902 IPQ_UNLOCK(i); 903 } 904 return (-1); 905 } 906 907 /* 908 * Take incoming datagram fragment and try to reassemble it into 909 * whole datagram. If the argument is the first fragment or one 910 * in between the function will return NULL and store the mbuf 911 * in the fragment chain. If the argument is the last fragment 912 * the packet will be reassembled and the pointer to the new 913 * mbuf returned for further processing. Only m_tags attached 914 * to the first packet/fragment are preserved. 915 * The IP header is *NOT* adjusted out of iplen. 916 */ 917 struct mbuf * 918 ip_reass(struct mbuf *m) 919 { 920 struct ip *ip; 921 struct mbuf *p, *q, *nq, *t; 922 struct ipq *fp = NULL; 923 struct ipqhead *head; 924 int i, hlen, next; 925 u_int8_t ecn, ecn0; 926 u_short hash; 927 #ifdef RSS 928 uint32_t rss_hash, rss_type; 929 #endif 930 int do_purge = 0; 931 932 /* If maxnipq or maxfragsperpacket are 0, never accept fragments. */ 933 if (V_maxnipq == 0 || V_maxfragsperpacket == 0) { 934 IPSTAT_INC(ips_fragments); 935 IPSTAT_INC(ips_fragdropped); 936 m_freem(m); 937 return (NULL); 938 } 939 940 ip = mtod(m, struct ip *); 941 hlen = ip->ip_hl << 2; 942 943 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id); 944 head = &V_ipq[hash]; 945 IPQ_LOCK(hash); 946 947 /* 948 * Look for queue of fragments 949 * of this datagram. 950 */ 951 TAILQ_FOREACH(fp, head, ipq_list) 952 if (ip->ip_id == fp->ipq_id && 953 ip->ip_src.s_addr == fp->ipq_src.s_addr && 954 ip->ip_dst.s_addr == fp->ipq_dst.s_addr && 955 #ifdef MAC 956 mac_ipq_match(m, fp) && 957 #endif 958 ip->ip_p == fp->ipq_p) 959 goto found; 960 961 fp = NULL; 962 963 /* 964 * Attempt to trim the number of allocated fragment queues if it 965 * exceeds the administrative limit. 966 */ 967 if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) { 968 /* 969 * drop something from the tail of the current queue 970 * before proceeding further 971 */ 972 struct ipq *q = TAILQ_LAST(head, ipqhead); 973 if (q == NULL) { /* gak */ 974 /* 975 * Defer doing this until later; when the 976 * lock is no longer held. 977 */ 978 do_purge = 1; 979 } else 980 ipq_timeout(head, q); 981 } 982 983 found: 984 /* 985 * Adjust ip_len to not reflect header, 986 * convert offset of this to bytes. 987 */ 988 ip->ip_len = htons(ntohs(ip->ip_len) - hlen); 989 if (ip->ip_off & htons(IP_MF)) { 990 /* 991 * Make sure that fragments have a data length 992 * that's a non-zero multiple of 8 bytes. 993 */ 994 if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) { 995 IPSTAT_INC(ips_toosmall); /* XXX */ 996 goto dropfrag; 997 } 998 m->m_flags |= M_IP_FRAG; 999 } else 1000 m->m_flags &= ~M_IP_FRAG; 1001 ip->ip_off = htons(ntohs(ip->ip_off) << 3); 1002 1003 /* 1004 * Attempt reassembly; if it succeeds, proceed. 1005 * ip_reass() will return a different mbuf. 1006 */ 1007 IPSTAT_INC(ips_fragments); 1008 m->m_pkthdr.PH_loc.ptr = ip; 1009 1010 /* Previous ip_reass() started here. */ 1011 /* 1012 * Presence of header sizes in mbufs 1013 * would confuse code below. 1014 */ 1015 m->m_data += hlen; 1016 m->m_len -= hlen; 1017 1018 /* 1019 * If first fragment to arrive, create a reassembly queue. 1020 */ 1021 if (fp == NULL) { 1022 fp = uma_zalloc(V_ipq_zone, M_NOWAIT); 1023 if (fp == NULL) 1024 goto dropfrag; 1025 #ifdef MAC 1026 if (mac_ipq_init(fp, M_NOWAIT) != 0) { 1027 uma_zfree(V_ipq_zone, fp); 1028 fp = NULL; 1029 goto dropfrag; 1030 } 1031 mac_ipq_create(m, fp); 1032 #endif 1033 TAILQ_INSERT_HEAD(head, fp, ipq_list); 1034 V_nipq++; 1035 fp->ipq_nfrags = 1; 1036 fp->ipq_ttl = IPFRAGTTL; 1037 fp->ipq_p = ip->ip_p; 1038 fp->ipq_id = ip->ip_id; 1039 fp->ipq_src = ip->ip_src; 1040 fp->ipq_dst = ip->ip_dst; 1041 fp->ipq_frags = m; 1042 m->m_nextpkt = NULL; 1043 goto done; 1044 } else { 1045 fp->ipq_nfrags++; 1046 #ifdef MAC 1047 mac_ipq_update(m, fp); 1048 #endif 1049 } 1050 1051 #define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) 1052 1053 /* 1054 * Handle ECN by comparing this segment with the first one; 1055 * if CE is set, do not lose CE. 1056 * drop if CE and not-ECT are mixed for the same packet. 1057 */ 1058 ecn = ip->ip_tos & IPTOS_ECN_MASK; 1059 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; 1060 if (ecn == IPTOS_ECN_CE) { 1061 if (ecn0 == IPTOS_ECN_NOTECT) 1062 goto dropfrag; 1063 if (ecn0 != IPTOS_ECN_CE) 1064 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; 1065 } 1066 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) 1067 goto dropfrag; 1068 1069 /* 1070 * Find a segment which begins after this one does. 1071 */ 1072 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) 1073 if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) 1074 break; 1075 1076 /* 1077 * If there is a preceding segment, it may provide some of 1078 * our data already. If so, drop the data from the incoming 1079 * segment. If it provides all of our data, drop us, otherwise 1080 * stick new segment in the proper place. 1081 * 1082 * If some of the data is dropped from the preceding 1083 * segment, then it's checksum is invalidated. 1084 */ 1085 if (p) { 1086 i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - 1087 ntohs(ip->ip_off); 1088 if (i > 0) { 1089 if (i >= ntohs(ip->ip_len)) 1090 goto dropfrag; 1091 m_adj(m, i); 1092 m->m_pkthdr.csum_flags = 0; 1093 ip->ip_off = htons(ntohs(ip->ip_off) + i); 1094 ip->ip_len = htons(ntohs(ip->ip_len) - i); 1095 } 1096 m->m_nextpkt = p->m_nextpkt; 1097 p->m_nextpkt = m; 1098 } else { 1099 m->m_nextpkt = fp->ipq_frags; 1100 fp->ipq_frags = m; 1101 } 1102 1103 /* 1104 * While we overlap succeeding segments trim them or, 1105 * if they are completely covered, dequeue them. 1106 */ 1107 for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > 1108 ntohs(GETIP(q)->ip_off); q = nq) { 1109 i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - 1110 ntohs(GETIP(q)->ip_off); 1111 if (i < ntohs(GETIP(q)->ip_len)) { 1112 GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); 1113 GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); 1114 m_adj(q, i); 1115 q->m_pkthdr.csum_flags = 0; 1116 break; 1117 } 1118 nq = q->m_nextpkt; 1119 m->m_nextpkt = nq; 1120 IPSTAT_INC(ips_fragdropped); 1121 fp->ipq_nfrags--; 1122 m_freem(q); 1123 } 1124 1125 /* 1126 * Check for complete reassembly and perform frag per packet 1127 * limiting. 1128 * 1129 * Frag limiting is performed here so that the nth frag has 1130 * a chance to complete the packet before we drop the packet. 1131 * As a result, n+1 frags are actually allowed per packet, but 1132 * only n will ever be stored. (n = maxfragsperpacket.) 1133 * 1134 */ 1135 next = 0; 1136 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { 1137 if (ntohs(GETIP(q)->ip_off) != next) { 1138 if (fp->ipq_nfrags > V_maxfragsperpacket) 1139 ipq_drop(head, fp); 1140 goto done; 1141 } 1142 next += ntohs(GETIP(q)->ip_len); 1143 } 1144 /* Make sure the last packet didn't have the IP_MF flag */ 1145 if (p->m_flags & M_IP_FRAG) { 1146 if (fp->ipq_nfrags > V_maxfragsperpacket) 1147 ipq_drop(head, fp); 1148 goto done; 1149 } 1150 1151 /* 1152 * Reassembly is complete. Make sure the packet is a sane size. 1153 */ 1154 q = fp->ipq_frags; 1155 ip = GETIP(q); 1156 if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { 1157 IPSTAT_INC(ips_toolong); 1158 ipq_drop(head, fp); 1159 goto done; 1160 } 1161 1162 /* 1163 * Concatenate fragments. 1164 */ 1165 m = q; 1166 t = m->m_next; 1167 m->m_next = NULL; 1168 m_cat(m, t); 1169 nq = q->m_nextpkt; 1170 q->m_nextpkt = NULL; 1171 for (q = nq; q != NULL; q = nq) { 1172 nq = q->m_nextpkt; 1173 q->m_nextpkt = NULL; 1174 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; 1175 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; 1176 m_cat(m, q); 1177 } 1178 /* 1179 * In order to do checksumming faster we do 'end-around carry' here 1180 * (and not in for{} loop), though it implies we are not going to 1181 * reassemble more than 64k fragments. 1182 */ 1183 while (m->m_pkthdr.csum_data & 0xffff0000) 1184 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 1185 (m->m_pkthdr.csum_data >> 16); 1186 #ifdef MAC 1187 mac_ipq_reassemble(fp, m); 1188 mac_ipq_destroy(fp); 1189 #endif 1190 1191 /* 1192 * Create header for new ip packet by modifying header of first 1193 * packet; dequeue and discard fragment reassembly header. 1194 * Make header visible. 1195 */ 1196 ip->ip_len = htons((ip->ip_hl << 2) + next); 1197 ip->ip_src = fp->ipq_src; 1198 ip->ip_dst = fp->ipq_dst; 1199 TAILQ_REMOVE(head, fp, ipq_list); 1200 V_nipq--; 1201 uma_zfree(V_ipq_zone, fp); 1202 m->m_len += (ip->ip_hl << 2); 1203 m->m_data -= (ip->ip_hl << 2); 1204 /* some debugging cruft by sklower, below, will go away soon */ 1205 if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ 1206 m_fixhdr(m); 1207 IPSTAT_INC(ips_reassembled); 1208 IPQ_UNLOCK(hash); 1209 1210 /* 1211 * Do the delayed purge to keep fragment counts under 1212 * the configured maximum. 1213 * 1214 * This is delayed so that it's not done with another IPQ bucket 1215 * lock held. 1216 * 1217 * Note that we pass in the bucket to /skip/ over, not 1218 * the bucket to /purge/. 1219 */ 1220 if (do_purge) 1221 ip_reass_purge_element(hash); 1222 1223 #ifdef RSS 1224 /* 1225 * Query the RSS layer for the flowid / flowtype for the 1226 * mbuf payload. 1227 * 1228 * For now, just assume we have to calculate a new one. 1229 * Later on we should check to see if the assigned flowid matches 1230 * what RSS wants for the given IP protocol and if so, just keep it. 1231 * 1232 * We then queue into the relevant netisr so it can be dispatched 1233 * to the correct CPU. 1234 * 1235 * Note - this may return 1, which means the flowid in the mbuf 1236 * is correct for the configured RSS hash types and can be used. 1237 */ 1238 if (rss_mbuf_software_hash_v4(m, 0, &rss_hash, &rss_type) == 0) { 1239 m->m_pkthdr.flowid = rss_hash; 1240 M_HASHTYPE_SET(m, rss_type); 1241 } 1242 1243 /* 1244 * Queue/dispatch for reprocessing. 1245 * 1246 * Note: this is much slower than just handling the frame in the 1247 * current receive context. It's likely worth investigating 1248 * why this is. 1249 */ 1250 netisr_dispatch(NETISR_IP_DIRECT, m); 1251 return (NULL); 1252 #endif 1253 1254 /* Handle in-line */ 1255 return (m); 1256 1257 dropfrag: 1258 IPSTAT_INC(ips_fragdropped); 1259 if (fp != NULL) 1260 fp->ipq_nfrags--; 1261 m_freem(m); 1262 done: 1263 IPQ_UNLOCK(hash); 1264 return (NULL); 1265 1266 #undef GETIP 1267 } 1268 1269 /* 1270 * Free a fragment reassembly header and all 1271 * associated datagrams. 1272 */ 1273 static void 1274 ipq_free(struct ipqhead *fhp, struct ipq *fp) 1275 { 1276 struct mbuf *q; 1277 1278 while (fp->ipq_frags) { 1279 q = fp->ipq_frags; 1280 fp->ipq_frags = q->m_nextpkt; 1281 m_freem(q); 1282 } 1283 TAILQ_REMOVE(fhp, fp, ipq_list); 1284 uma_zfree(V_ipq_zone, fp); 1285 V_nipq--; 1286 } 1287 1288 /* 1289 * IP timer processing; 1290 * if a timer expires on a reassembly 1291 * queue, discard it. 1292 */ 1293 void 1294 ip_slowtimo(void) 1295 { 1296 VNET_ITERATOR_DECL(vnet_iter); 1297 struct ipq *fp, *tmp; 1298 int i; 1299 1300 VNET_LIST_RLOCK_NOSLEEP(); 1301 VNET_FOREACH(vnet_iter) { 1302 CURVNET_SET(vnet_iter); 1303 for (i = 0; i < IPREASS_NHASH; i++) { 1304 IPQ_LOCK(i); 1305 TAILQ_FOREACH_SAFE(fp, &V_ipq[i], ipq_list, tmp) 1306 if (--fp->ipq_ttl == 0) 1307 ipq_timeout(&V_ipq[i], fp); 1308 IPQ_UNLOCK(i); 1309 } 1310 /* 1311 * If we are over the maximum number of fragments 1312 * (due to the limit being lowered), drain off 1313 * enough to get down to the new limit. 1314 */ 1315 if (V_maxnipq >= 0 && V_nipq > V_maxnipq) { 1316 for (i = 0; i < IPREASS_NHASH; i++) { 1317 IPQ_LOCK(i); 1318 while (V_nipq > V_maxnipq && 1319 !TAILQ_EMPTY(&V_ipq[i])) 1320 ipq_drop(&V_ipq[i], 1321 TAILQ_FIRST(&V_ipq[i])); 1322 IPQ_UNLOCK(i); 1323 } 1324 } 1325 CURVNET_RESTORE(); 1326 } 1327 VNET_LIST_RUNLOCK_NOSLEEP(); 1328 } 1329 1330 /* 1331 * Drain off all datagram fragments. 1332 */ 1333 static void 1334 ip_drain_vnet(void) 1335 { 1336 int i; 1337 1338 for (i = 0; i < IPREASS_NHASH; i++) { 1339 IPQ_LOCK(i); 1340 while(!TAILQ_EMPTY(&V_ipq[i])) 1341 ipq_drop(&V_ipq[i], TAILQ_FIRST(&V_ipq[i])); 1342 IPQ_UNLOCK(i); 1343 } 1344 } 1345 1346 void 1347 ip_drain(void) 1348 { 1349 VNET_ITERATOR_DECL(vnet_iter); 1350 1351 VNET_LIST_RLOCK_NOSLEEP(); 1352 VNET_FOREACH(vnet_iter) { 1353 CURVNET_SET(vnet_iter); 1354 ip_drain_vnet(); 1355 CURVNET_RESTORE(); 1356 } 1357 VNET_LIST_RUNLOCK_NOSLEEP(); 1358 } 1359 1360 /* 1361 * The protocol to be inserted into ip_protox[] must be already registered 1362 * in inetsw[], either statically or through pf_proto_register(). 1363 */ 1364 int 1365 ipproto_register(short ipproto) 1366 { 1367 struct protosw *pr; 1368 1369 /* Sanity checks. */ 1370 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1371 return (EPROTONOSUPPORT); 1372 1373 /* 1374 * The protocol slot must not be occupied by another protocol 1375 * already. An index pointing to IPPROTO_RAW is unused. 1376 */ 1377 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1378 if (pr == NULL) 1379 return (EPFNOSUPPORT); 1380 if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */ 1381 return (EEXIST); 1382 1383 /* Find the protocol position in inetsw[] and set the index. */ 1384 for (pr = inetdomain.dom_protosw; 1385 pr < inetdomain.dom_protoswNPROTOSW; pr++) { 1386 if (pr->pr_domain->dom_family == PF_INET && 1387 pr->pr_protocol && pr->pr_protocol == ipproto) { 1388 ip_protox[pr->pr_protocol] = pr - inetsw; 1389 return (0); 1390 } 1391 } 1392 return (EPROTONOSUPPORT); 1393 } 1394 1395 int 1396 ipproto_unregister(short ipproto) 1397 { 1398 struct protosw *pr; 1399 1400 /* Sanity checks. */ 1401 if (ipproto <= 0 || ipproto >= IPPROTO_MAX) 1402 return (EPROTONOSUPPORT); 1403 1404 /* Check if the protocol was indeed registered. */ 1405 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); 1406 if (pr == NULL) 1407 return (EPFNOSUPPORT); 1408 if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */ 1409 return (ENOENT); 1410 1411 /* Reset the protocol slot to IPPROTO_RAW. */ 1412 ip_protox[ipproto] = pr - inetsw; 1413 return (0); 1414 } 1415 1416 /* 1417 * Given address of next destination (final or next hop), return (referenced) 1418 * internet address info of interface to be used to get there. 1419 */ 1420 struct in_ifaddr * 1421 ip_rtaddr(struct in_addr dst, u_int fibnum) 1422 { 1423 struct route sro; 1424 struct sockaddr_in *sin; 1425 struct in_ifaddr *ia; 1426 1427 bzero(&sro, sizeof(sro)); 1428 sin = (struct sockaddr_in *)&sro.ro_dst; 1429 sin->sin_family = AF_INET; 1430 sin->sin_len = sizeof(*sin); 1431 sin->sin_addr = dst; 1432 in_rtalloc_ign(&sro, 0, fibnum); 1433 1434 if (sro.ro_rt == NULL) 1435 return (NULL); 1436 1437 ia = ifatoia(sro.ro_rt->rt_ifa); 1438 ifa_ref(&ia->ia_ifa); 1439 RTFREE(sro.ro_rt); 1440 return (ia); 1441 } 1442 1443 u_char inetctlerrmap[PRC_NCMDS] = { 1444 0, 0, 0, 0, 1445 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH, 1446 EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED, 1447 EMSGSIZE, EHOSTUNREACH, 0, 0, 1448 0, 0, EHOSTUNREACH, 0, 1449 ENOPROTOOPT, ECONNREFUSED 1450 }; 1451 1452 /* 1453 * Forward a packet. If some error occurs return the sender 1454 * an icmp packet. Note we can't always generate a meaningful 1455 * icmp message because icmp doesn't have a large enough repertoire 1456 * of codes and types. 1457 * 1458 * If not forwarding, just drop the packet. This could be confusing 1459 * if ipforwarding was zero but some routing protocol was advancing 1460 * us as a gateway to somewhere. However, we must let the routing 1461 * protocol deal with that. 1462 * 1463 * The srcrt parameter indicates whether the packet is being forwarded 1464 * via a source route. 1465 */ 1466 void 1467 ip_forward(struct mbuf *m, int srcrt) 1468 { 1469 struct ip *ip = mtod(m, struct ip *); 1470 struct in_ifaddr *ia; 1471 struct mbuf *mcopy; 1472 struct in_addr dest; 1473 struct route ro; 1474 int error, type = 0, code = 0, mtu = 0; 1475 1476 if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) { 1477 IPSTAT_INC(ips_cantforward); 1478 m_freem(m); 1479 return; 1480 } 1481 #ifdef IPSEC 1482 if (ip_ipsec_fwd(m) != 0) { 1483 IPSTAT_INC(ips_cantforward); 1484 m_freem(m); 1485 return; 1486 } 1487 #endif /* IPSEC */ 1488 #ifdef IPSTEALTH 1489 if (!V_ipstealth) { 1490 #endif 1491 if (ip->ip_ttl <= IPTTLDEC) { 1492 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 1493 0, 0); 1494 return; 1495 } 1496 #ifdef IPSTEALTH 1497 } 1498 #endif 1499 1500 ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m)); 1501 #ifndef IPSEC 1502 /* 1503 * 'ia' may be NULL if there is no route for this destination. 1504 * In case of IPsec, Don't discard it just yet, but pass it to 1505 * ip_output in case of outgoing IPsec policy. 1506 */ 1507 if (!srcrt && ia == NULL) { 1508 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 1509 return; 1510 } 1511 #endif 1512 1513 /* 1514 * Save the IP header and at most 8 bytes of the payload, 1515 * in case we need to generate an ICMP message to the src. 1516 * 1517 * XXX this can be optimized a lot by saving the data in a local 1518 * buffer on the stack (72 bytes at most), and only allocating the 1519 * mbuf if really necessary. The vast majority of the packets 1520 * are forwarded without having to send an ICMP back (either 1521 * because unnecessary, or because rate limited), so we are 1522 * really we are wasting a lot of work here. 1523 * 1524 * We don't use m_copy() because it might return a reference 1525 * to a shared cluster. Both this function and ip_output() 1526 * assume exclusive access to the IP header in `m', so any 1527 * data in a cluster may change before we reach icmp_error(). 1528 */ 1529 mcopy = m_gethdr(M_NOWAIT, m->m_type); 1530 if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) { 1531 /* 1532 * It's probably ok if the pkthdr dup fails (because 1533 * the deep copy of the tag chain failed), but for now 1534 * be conservative and just discard the copy since 1535 * code below may some day want the tags. 1536 */ 1537 m_free(mcopy); 1538 mcopy = NULL; 1539 } 1540 if (mcopy != NULL) { 1541 mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy)); 1542 mcopy->m_pkthdr.len = mcopy->m_len; 1543 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t)); 1544 } 1545 1546 #ifdef IPSTEALTH 1547 if (!V_ipstealth) { 1548 #endif 1549 ip->ip_ttl -= IPTTLDEC; 1550 #ifdef IPSTEALTH 1551 } 1552 #endif 1553 1554 /* 1555 * If forwarding packet using same interface that it came in on, 1556 * perhaps should send a redirect to sender to shortcut a hop. 1557 * Only send redirect if source is sending directly to us, 1558 * and if packet was not source routed (or has any options). 1559 * Also, don't send redirect if forwarding using a default route 1560 * or a route modified by a redirect. 1561 */ 1562 dest.s_addr = 0; 1563 if (!srcrt && V_ipsendredirects && 1564 ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) { 1565 struct sockaddr_in *sin; 1566 struct rtentry *rt; 1567 1568 bzero(&ro, sizeof(ro)); 1569 sin = (struct sockaddr_in *)&ro.ro_dst; 1570 sin->sin_family = AF_INET; 1571 sin->sin_len = sizeof(*sin); 1572 sin->sin_addr = ip->ip_dst; 1573 in_rtalloc_ign(&ro, 0, M_GETFIB(m)); 1574 1575 rt = ro.ro_rt; 1576 1577 if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 && 1578 satosin(rt_key(rt))->sin_addr.s_addr != 0) { 1579 #define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa)) 1580 u_long src = ntohl(ip->ip_src.s_addr); 1581 1582 if (RTA(rt) && 1583 (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) { 1584 if (rt->rt_flags & RTF_GATEWAY) 1585 dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr; 1586 else 1587 dest.s_addr = ip->ip_dst.s_addr; 1588 /* Router requirements says to only send host redirects */ 1589 type = ICMP_REDIRECT; 1590 code = ICMP_REDIRECT_HOST; 1591 } 1592 } 1593 if (rt) 1594 RTFREE(rt); 1595 } 1596 1597 /* 1598 * Try to cache the route MTU from ip_output so we can consider it for 1599 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191. 1600 */ 1601 bzero(&ro, sizeof(ro)); 1602 1603 error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL); 1604 1605 if (error == EMSGSIZE && ro.ro_rt) 1606 mtu = ro.ro_rt->rt_mtu; 1607 RO_RTFREE(&ro); 1608 1609 if (error) 1610 IPSTAT_INC(ips_cantforward); 1611 else { 1612 IPSTAT_INC(ips_forward); 1613 if (type) 1614 IPSTAT_INC(ips_redirectsent); 1615 else { 1616 if (mcopy) 1617 m_freem(mcopy); 1618 if (ia != NULL) 1619 ifa_free(&ia->ia_ifa); 1620 return; 1621 } 1622 } 1623 if (mcopy == NULL) { 1624 if (ia != NULL) 1625 ifa_free(&ia->ia_ifa); 1626 return; 1627 } 1628 1629 switch (error) { 1630 1631 case 0: /* forwarded, but need redirect */ 1632 /* type, code set above */ 1633 break; 1634 1635 case ENETUNREACH: 1636 case EHOSTUNREACH: 1637 case ENETDOWN: 1638 case EHOSTDOWN: 1639 default: 1640 type = ICMP_UNREACH; 1641 code = ICMP_UNREACH_HOST; 1642 break; 1643 1644 case EMSGSIZE: 1645 type = ICMP_UNREACH; 1646 code = ICMP_UNREACH_NEEDFRAG; 1647 1648 #ifdef IPSEC 1649 /* 1650 * If IPsec is configured for this path, 1651 * override any possibly mtu value set by ip_output. 1652 */ 1653 mtu = ip_ipsec_mtu(mcopy, mtu); 1654 #endif /* IPSEC */ 1655 /* 1656 * If the MTU was set before make sure we are below the 1657 * interface MTU. 1658 * If the MTU wasn't set before use the interface mtu or 1659 * fall back to the next smaller mtu step compared to the 1660 * current packet size. 1661 */ 1662 if (mtu != 0) { 1663 if (ia != NULL) 1664 mtu = min(mtu, ia->ia_ifp->if_mtu); 1665 } else { 1666 if (ia != NULL) 1667 mtu = ia->ia_ifp->if_mtu; 1668 else 1669 mtu = ip_next_mtu(ntohs(ip->ip_len), 0); 1670 } 1671 IPSTAT_INC(ips_cantfrag); 1672 break; 1673 1674 case ENOBUFS: 1675 case EACCES: /* ipfw denied packet */ 1676 m_freem(mcopy); 1677 if (ia != NULL) 1678 ifa_free(&ia->ia_ifa); 1679 return; 1680 } 1681 if (ia != NULL) 1682 ifa_free(&ia->ia_ifa); 1683 icmp_error(mcopy, type, code, dest.s_addr, mtu); 1684 } 1685 1686 void 1687 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip, 1688 struct mbuf *m) 1689 { 1690 1691 if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) { 1692 struct bintime bt; 1693 1694 bintime(&bt); 1695 if (inp->inp_socket->so_options & SO_BINTIME) { 1696 *mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt), 1697 SCM_BINTIME, SOL_SOCKET); 1698 if (*mp) 1699 mp = &(*mp)->m_next; 1700 } 1701 if (inp->inp_socket->so_options & SO_TIMESTAMP) { 1702 struct timeval tv; 1703 1704 bintime2timeval(&bt, &tv); 1705 *mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv), 1706 SCM_TIMESTAMP, SOL_SOCKET); 1707 if (*mp) 1708 mp = &(*mp)->m_next; 1709 } 1710 } 1711 if (inp->inp_flags & INP_RECVDSTADDR) { 1712 *mp = sbcreatecontrol((caddr_t)&ip->ip_dst, 1713 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP); 1714 if (*mp) 1715 mp = &(*mp)->m_next; 1716 } 1717 if (inp->inp_flags & INP_RECVTTL) { 1718 *mp = sbcreatecontrol((caddr_t)&ip->ip_ttl, 1719 sizeof(u_char), IP_RECVTTL, IPPROTO_IP); 1720 if (*mp) 1721 mp = &(*mp)->m_next; 1722 } 1723 #ifdef notyet 1724 /* XXX 1725 * Moving these out of udp_input() made them even more broken 1726 * than they already were. 1727 */ 1728 /* options were tossed already */ 1729 if (inp->inp_flags & INP_RECVOPTS) { 1730 *mp = sbcreatecontrol((caddr_t)opts_deleted_above, 1731 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP); 1732 if (*mp) 1733 mp = &(*mp)->m_next; 1734 } 1735 /* ip_srcroute doesn't do what we want here, need to fix */ 1736 if (inp->inp_flags & INP_RECVRETOPTS) { 1737 *mp = sbcreatecontrol((caddr_t)ip_srcroute(m), 1738 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP); 1739 if (*mp) 1740 mp = &(*mp)->m_next; 1741 } 1742 #endif 1743 if (inp->inp_flags & INP_RECVIF) { 1744 struct ifnet *ifp; 1745 struct sdlbuf { 1746 struct sockaddr_dl sdl; 1747 u_char pad[32]; 1748 } sdlbuf; 1749 struct sockaddr_dl *sdp; 1750 struct sockaddr_dl *sdl2 = &sdlbuf.sdl; 1751 1752 if ((ifp = m->m_pkthdr.rcvif) && 1753 ifp->if_index && ifp->if_index <= V_if_index) { 1754 sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr; 1755 /* 1756 * Change our mind and don't try copy. 1757 */ 1758 if (sdp->sdl_family != AF_LINK || 1759 sdp->sdl_len > sizeof(sdlbuf)) { 1760 goto makedummy; 1761 } 1762 bcopy(sdp, sdl2, sdp->sdl_len); 1763 } else { 1764 makedummy: 1765 sdl2->sdl_len = 1766 offsetof(struct sockaddr_dl, sdl_data[0]); 1767 sdl2->sdl_family = AF_LINK; 1768 sdl2->sdl_index = 0; 1769 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0; 1770 } 1771 *mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len, 1772 IP_RECVIF, IPPROTO_IP); 1773 if (*mp) 1774 mp = &(*mp)->m_next; 1775 } 1776 if (inp->inp_flags & INP_RECVTOS) { 1777 *mp = sbcreatecontrol((caddr_t)&ip->ip_tos, 1778 sizeof(u_char), IP_RECVTOS, IPPROTO_IP); 1779 if (*mp) 1780 mp = &(*mp)->m_next; 1781 } 1782 1783 if (inp->inp_flags2 & INP_RECVFLOWID) { 1784 uint32_t flowid, flow_type; 1785 1786 flowid = m->m_pkthdr.flowid; 1787 flow_type = M_HASHTYPE_GET(m); 1788 1789 /* 1790 * XXX should handle the failure of one or the 1791 * other - don't populate both? 1792 */ 1793 *mp = sbcreatecontrol((caddr_t) &flowid, 1794 sizeof(uint32_t), IP_FLOWID, IPPROTO_IP); 1795 if (*mp) 1796 mp = &(*mp)->m_next; 1797 *mp = sbcreatecontrol((caddr_t) &flow_type, 1798 sizeof(uint32_t), IP_FLOWTYPE, IPPROTO_IP); 1799 if (*mp) 1800 mp = &(*mp)->m_next; 1801 } 1802 1803 #ifdef RSS 1804 if (inp->inp_flags2 & INP_RECVRSSBUCKETID) { 1805 uint32_t flowid, flow_type; 1806 uint32_t rss_bucketid; 1807 1808 flowid = m->m_pkthdr.flowid; 1809 flow_type = M_HASHTYPE_GET(m); 1810 1811 if (rss_hash2bucket(flowid, flow_type, &rss_bucketid) == 0) { 1812 *mp = sbcreatecontrol((caddr_t) &rss_bucketid, 1813 sizeof(uint32_t), IP_RSSBUCKETID, IPPROTO_IP); 1814 if (*mp) 1815 mp = &(*mp)->m_next; 1816 } 1817 } 1818 #endif 1819 } 1820 1821 /* 1822 * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the 1823 * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on 1824 * locking. This code remains in ip_input.c as ip_mroute.c is optionally 1825 * compiled. 1826 */ 1827 static VNET_DEFINE(int, ip_rsvp_on); 1828 VNET_DEFINE(struct socket *, ip_rsvpd); 1829 1830 #define V_ip_rsvp_on VNET(ip_rsvp_on) 1831 1832 int 1833 ip_rsvp_init(struct socket *so) 1834 { 1835 1836 if (so->so_type != SOCK_RAW || 1837 so->so_proto->pr_protocol != IPPROTO_RSVP) 1838 return EOPNOTSUPP; 1839 1840 if (V_ip_rsvpd != NULL) 1841 return EADDRINUSE; 1842 1843 V_ip_rsvpd = so; 1844 /* 1845 * This may seem silly, but we need to be sure we don't over-increment 1846 * the RSVP counter, in case something slips up. 1847 */ 1848 if (!V_ip_rsvp_on) { 1849 V_ip_rsvp_on = 1; 1850 V_rsvp_on++; 1851 } 1852 1853 return 0; 1854 } 1855 1856 int 1857 ip_rsvp_done(void) 1858 { 1859 1860 V_ip_rsvpd = NULL; 1861 /* 1862 * This may seem silly, but we need to be sure we don't over-decrement 1863 * the RSVP counter, in case something slips up. 1864 */ 1865 if (V_ip_rsvp_on) { 1866 V_ip_rsvp_on = 0; 1867 V_rsvp_on--; 1868 } 1869 return 0; 1870 } 1871 1872 int 1873 rsvp_input(struct mbuf **mp, int *offp, int proto) 1874 { 1875 struct mbuf *m; 1876 1877 m = *mp; 1878 *mp = NULL; 1879 1880 if (rsvp_input_p) { /* call the real one if loaded */ 1881 *mp = m; 1882 rsvp_input_p(mp, offp, proto); 1883 return (IPPROTO_DONE); 1884 } 1885 1886 /* Can still get packets with rsvp_on = 0 if there is a local member 1887 * of the group to which the RSVP packet is addressed. But in this 1888 * case we want to throw the packet away. 1889 */ 1890 1891 if (!V_rsvp_on) { 1892 m_freem(m); 1893 return (IPPROTO_DONE); 1894 } 1895 1896 if (V_ip_rsvpd != NULL) { 1897 *mp = m; 1898 rip_input(mp, offp, proto); 1899 return (IPPROTO_DONE); 1900 } 1901 /* Drop the packet */ 1902 m_freem(m); 1903 return (IPPROTO_DONE); 1904 } 1905