tcp_syncache.c revision 168901
1/*- 2 * Copyright (c) 2001 McAfee, Inc. 3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Jonathan Lemon 7 * and McAfee Research, the Security Research Division of McAfee, Inc. under 8 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 9 * DARPA CHATS research program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: head/sys/netinet/tcp_syncache.c 168901 2007-04-20 13:36:48Z andre $ 33 */ 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_ipsec.h" 38#include "opt_mac.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/kernel.h> 43#include <sys/sysctl.h> 44#include <sys/lock.h> 45#include <sys/mutex.h> 46#include <sys/malloc.h> 47#include <sys/mbuf.h> 48#include <sys/md5.h> 49#include <sys/proc.h> /* for proc0 declaration */ 50#include <sys/random.h> 51#include <sys/socket.h> 52#include <sys/socketvar.h> 53 54#include <vm/uma.h> 55 56#include <net/if.h> 57#include <net/route.h> 58 59#include <netinet/in.h> 60#include <netinet/in_systm.h> 61#include <netinet/ip.h> 62#include <netinet/in_var.h> 63#include <netinet/in_pcb.h> 64#include <netinet/ip_var.h> 65#include <netinet/ip_options.h> 66#ifdef INET6 67#include <netinet/ip6.h> 68#include <netinet/icmp6.h> 69#include <netinet6/nd6.h> 70#include <netinet6/ip6_var.h> 71#include <netinet6/in6_pcb.h> 72#endif 73#include <netinet/tcp.h> 74#include <netinet/tcp_fsm.h> 75#include <netinet/tcp_seq.h> 76#include <netinet/tcp_timer.h> 77#include <netinet/tcp_var.h> 78#ifdef INET6 79#include <netinet6/tcp6_var.h> 80#endif 81 82#ifdef IPSEC 83#include <netinet6/ipsec.h> 84#ifdef INET6 85#include <netinet6/ipsec6.h> 86#endif 87#endif /*IPSEC*/ 88 89#ifdef FAST_IPSEC 90#include <netipsec/ipsec.h> 91#ifdef INET6 92#include <netipsec/ipsec6.h> 93#endif 94#include <netipsec/key.h> 95#endif /*FAST_IPSEC*/ 96 97#include <machine/in_cksum.h> 98 99#include <security/mac/mac_framework.h> 100 101static int tcp_syncookies = 1; 102SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 103 &tcp_syncookies, 0, 104 "Use TCP SYN cookies if the syncache overflows"); 105 106static int tcp_syncookiesonly = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW, 108 &tcp_syncookiesonly, 0, 109 "Use only TCP SYN cookies"); 110 111#define SYNCOOKIE_SECRET_SIZE 8 /* dwords */ 112#define SYNCOOKIE_LIFETIME 16 /* seconds */ 113 114struct syncache { 115 TAILQ_ENTRY(syncache) sc_hash; 116 struct in_conninfo sc_inc; /* addresses */ 117 u_long sc_rxttime; /* retransmit time */ 118 u_int16_t sc_rxmits; /* retransmit counter */ 119 120 u_int32_t sc_tsreflect; /* timestamp to reflect */ 121 u_int32_t sc_ts; /* our timestamp to send */ 122 u_int32_t sc_tsoff; /* ts offset w/ syncookies */ 123 u_int32_t sc_flowlabel; /* IPv6 flowlabel */ 124 tcp_seq sc_irs; /* seq from peer */ 125 tcp_seq sc_iss; /* our ISS */ 126 struct mbuf *sc_ipopts; /* source route */ 127 128 u_int16_t sc_peer_mss; /* peer's MSS */ 129 u_int16_t sc_wnd; /* advertised window */ 130 u_int8_t sc_ip_ttl; /* IPv4 TTL */ 131 u_int8_t sc_ip_tos; /* IPv4 TOS */ 132 u_int8_t sc_requested_s_scale:4, 133 sc_requested_r_scale:4; 134 u_int8_t sc_flags; 135#define SCF_NOOPT 0x01 /* no TCP options */ 136#define SCF_WINSCALE 0x02 /* negotiated window scaling */ 137#define SCF_TIMESTAMP 0x04 /* negotiated timestamps */ 138 /* MSS is implicit */ 139#define SCF_UNREACH 0x10 /* icmp unreachable received */ 140#define SCF_SIGNATURE 0x20 /* send MD5 digests */ 141#define SCF_SACK 0x80 /* send SACK option */ 142#ifdef MAC 143 struct label *sc_label; /* MAC label reference */ 144#endif 145}; 146 147struct syncache_head { 148 struct mtx sch_mtx; 149 TAILQ_HEAD(sch_head, syncache) sch_bucket; 150 struct callout sch_timer; 151 int sch_nextc; 152 u_int sch_length; 153 u_int sch_oddeven; 154 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE]; 155 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE]; 156 u_int sch_reseed; /* time_uptime, seconds */ 157}; 158 159static void syncache_drop(struct syncache *, struct syncache_head *); 160static void syncache_free(struct syncache *); 161static void syncache_insert(struct syncache *, struct syncache_head *); 162struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 163static int syncache_respond(struct syncache *); 164static struct socket *syncache_socket(struct syncache *, struct socket *, 165 struct mbuf *m); 166static void syncache_timer(void *); 167static void syncookie_generate(struct syncache_head *, struct syncache *, 168 u_int32_t *); 169static struct syncache 170 *syncookie_lookup(struct in_conninfo *, struct syncache_head *, 171 struct syncache *, struct tcpopt *, struct tcphdr *, 172 struct socket *); 173 174/* 175 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 176 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 177 * the odds are that the user has given up attempting to connect by then. 178 */ 179#define SYNCACHE_MAXREXMTS 3 180 181/* Arbitrary values */ 182#define TCP_SYNCACHE_HASHSIZE 512 183#define TCP_SYNCACHE_BUCKETLIMIT 30 184 185struct tcp_syncache { 186 struct syncache_head *hashbase; 187 uma_zone_t zone; 188 u_int hashsize; 189 u_int hashmask; 190 u_int bucket_limit; 191 u_int cache_count; /* XXX: unprotected */ 192 u_int cache_limit; 193 u_int rexmt_limit; 194 u_int hash_secret; 195}; 196static struct tcp_syncache tcp_syncache; 197 198SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 199 200SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 201 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 202 203SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 204 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 205 206SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 207 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 208 209SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 210 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 211 212SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 213 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 214 215static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 216 217#define SYNCACHE_HASH(inc, mask) \ 218 ((tcp_syncache.hash_secret ^ \ 219 (inc)->inc_faddr.s_addr ^ \ 220 ((inc)->inc_faddr.s_addr >> 16) ^ \ 221 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 222 223#define SYNCACHE_HASH6(inc, mask) \ 224 ((tcp_syncache.hash_secret ^ \ 225 (inc)->inc6_faddr.s6_addr32[0] ^ \ 226 (inc)->inc6_faddr.s6_addr32[3] ^ \ 227 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 228 229#define ENDPTS_EQ(a, b) ( \ 230 (a)->ie_fport == (b)->ie_fport && \ 231 (a)->ie_lport == (b)->ie_lport && \ 232 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 233 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 234) 235 236#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 237 238#define SYNCACHE_TIMEOUT(sc, sch, co) do { \ 239 (sc)->sc_rxmits++; \ 240 (sc)->sc_rxttime = ticks + \ 241 TCPTV_RTOBASE * tcp_backoff[(sc)->sc_rxmits - 1]; \ 242 if ((sch)->sch_nextc > (sc)->sc_rxttime) \ 243 (sch)->sch_nextc = (sc)->sc_rxttime; \ 244 if (!TAILQ_EMPTY(&(sch)->sch_bucket) && !(co)) \ 245 callout_reset(&(sch)->sch_timer, \ 246 (sch)->sch_nextc - ticks, \ 247 syncache_timer, (void *)(sch)); \ 248} while (0) 249 250#define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 251#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 252#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 253 254/* 255 * Requires the syncache entry to be already removed from the bucket list. 256 */ 257static void 258syncache_free(struct syncache *sc) 259{ 260 if (sc->sc_ipopts) 261 (void) m_free(sc->sc_ipopts); 262#ifdef MAC 263 mac_destroy_syncache(&sc->sc_label); 264#endif 265 266 uma_zfree(tcp_syncache.zone, sc); 267} 268 269void 270syncache_init(void) 271{ 272 int i; 273 274 tcp_syncache.cache_count = 0; 275 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 276 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 277 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 278 tcp_syncache.hash_secret = arc4random(); 279 280 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 281 &tcp_syncache.hashsize); 282 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 283 &tcp_syncache.bucket_limit); 284 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 285 printf("WARNING: syncache hash size is not a power of 2.\n"); 286 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 287 } 288 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 289 290 /* Set limits. */ 291 tcp_syncache.cache_limit = 292 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 293 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 294 &tcp_syncache.cache_limit); 295 296 /* Allocate the hash table. */ 297 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 298 tcp_syncache.hashsize * sizeof(struct syncache_head), 299 M_SYNCACHE, M_WAITOK | M_ZERO); 300 301 /* Initialize the hash buckets. */ 302 for (i = 0; i < tcp_syncache.hashsize; i++) { 303 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 304 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 305 NULL, MTX_DEF); 306 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer, 307 &tcp_syncache.hashbase[i].sch_mtx, 0); 308 tcp_syncache.hashbase[i].sch_length = 0; 309 } 310 311 /* Create the syncache entry zone. */ 312 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 313 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 314 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 315} 316 317/* 318 * Inserts a syncache entry into the specified bucket row. 319 * Locks and unlocks the syncache_head autonomously. 320 */ 321static void 322syncache_insert(struct syncache *sc, struct syncache_head *sch) 323{ 324 struct syncache *sc2; 325 326 SCH_LOCK(sch); 327 328 /* 329 * Make sure that we don't overflow the per-bucket limit. 330 * If the bucket is full, toss the oldest element. 331 */ 332 if (sch->sch_length >= tcp_syncache.bucket_limit) { 333 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 334 ("sch->sch_length incorrect")); 335 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 336 syncache_drop(sc2, sch); 337 tcpstat.tcps_sc_bucketoverflow++; 338 } 339 340 /* Put it into the bucket. */ 341 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 342 sch->sch_length++; 343 344 /* Reinitialize the bucket row's timer. */ 345 SYNCACHE_TIMEOUT(sc, sch, 1); 346 347 SCH_UNLOCK(sch); 348 349 tcp_syncache.cache_count++; 350 tcpstat.tcps_sc_added++; 351} 352 353/* 354 * Remove and free entry from syncache bucket row. 355 * Expects locked syncache head. 356 */ 357static void 358syncache_drop(struct syncache *sc, struct syncache_head *sch) 359{ 360 361 SCH_LOCK_ASSERT(sch); 362 363 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 364 sch->sch_length--; 365 366 syncache_free(sc); 367 tcp_syncache.cache_count--; 368} 369 370/* 371 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 372 * If we have retransmitted an entry the maximum number of times, expire it. 373 * One separate timer for each bucket row. 374 */ 375static void 376syncache_timer(void *xsch) 377{ 378 struct syncache_head *sch = (struct syncache_head *)xsch; 379 struct syncache *sc, *nsc; 380 int tick = ticks; 381 382 /* NB: syncache_head has already been locked by the callout. */ 383 SCH_LOCK_ASSERT(sch); 384 385 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 386 /* 387 * We do not check if the listen socket still exists 388 * and accept the case where the listen socket may be 389 * gone by the time we resend the SYN/ACK. We do 390 * not expect this to happens often. If it does, 391 * then the RST will be sent by the time the remote 392 * host does the SYN/ACK->ACK. 393 */ 394 if (sc->sc_rxttime >= tick) { 395 if (sc->sc_rxttime < sch->sch_nextc) 396 sch->sch_nextc = sc->sc_rxttime; 397 continue; 398 } 399 400 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) { 401 syncache_drop(sc, sch); 402 tcpstat.tcps_sc_stale++; 403 continue; 404 } 405 406 (void) syncache_respond(sc); 407 tcpstat.tcps_sc_retransmitted++; 408 SYNCACHE_TIMEOUT(sc, sch, 0); 409 } 410 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 411 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 412 syncache_timer, (void *)(sch)); 413} 414 415/* 416 * Find an entry in the syncache. 417 * Returns always with locked syncache_head plus a matching entry or NULL. 418 */ 419struct syncache * 420syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 421{ 422 struct syncache *sc; 423 struct syncache_head *sch; 424 425#ifdef INET6 426 if (inc->inc_isipv6) { 427 sch = &tcp_syncache.hashbase[ 428 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 429 *schp = sch; 430 431 SCH_LOCK(sch); 432 433 /* Circle through bucket row to find matching entry. */ 434 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 435 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 436 return (sc); 437 } 438 } else 439#endif 440 { 441 sch = &tcp_syncache.hashbase[ 442 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 443 *schp = sch; 444 445 SCH_LOCK(sch); 446 447 /* Circle through bucket row to find matching entry. */ 448 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 449#ifdef INET6 450 if (sc->sc_inc.inc_isipv6) 451 continue; 452#endif 453 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 454 return (sc); 455 } 456 } 457 SCH_LOCK_ASSERT(*schp); 458 return (NULL); /* always returns with locked sch */ 459} 460 461/* 462 * This function is called when we get a RST for a 463 * non-existent connection, so that we can see if the 464 * connection is in the syn cache. If it is, zap it. 465 */ 466void 467syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 468{ 469 struct syncache *sc; 470 struct syncache_head *sch; 471 472 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 473 SCH_LOCK_ASSERT(sch); 474 if (sc == NULL) 475 goto done; 476 477 /* 478 * If the RST bit is set, check the sequence number to see 479 * if this is a valid reset segment. 480 * RFC 793 page 37: 481 * In all states except SYN-SENT, all reset (RST) segments 482 * are validated by checking their SEQ-fields. A reset is 483 * valid if its sequence number is in the window. 484 * 485 * The sequence number in the reset segment is normally an 486 * echo of our outgoing acknowlegement numbers, but some hosts 487 * send a reset with the sequence number at the rightmost edge 488 * of our receive window, and we have to handle this case. 489 */ 490 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 491 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 492 syncache_drop(sc, sch); 493 tcpstat.tcps_sc_reset++; 494 } 495done: 496 SCH_UNLOCK(sch); 497} 498 499void 500syncache_badack(struct in_conninfo *inc) 501{ 502 struct syncache *sc; 503 struct syncache_head *sch; 504 505 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 506 SCH_LOCK_ASSERT(sch); 507 if (sc != NULL) { 508 syncache_drop(sc, sch); 509 tcpstat.tcps_sc_badack++; 510 } 511 SCH_UNLOCK(sch); 512} 513 514void 515syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 516{ 517 struct syncache *sc; 518 struct syncache_head *sch; 519 520 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 521 SCH_LOCK_ASSERT(sch); 522 if (sc == NULL) 523 goto done; 524 525 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 526 if (ntohl(th->th_seq) != sc->sc_iss) 527 goto done; 528 529 /* 530 * If we've rertransmitted 3 times and this is our second error, 531 * we remove the entry. Otherwise, we allow it to continue on. 532 * This prevents us from incorrectly nuking an entry during a 533 * spurious network outage. 534 * 535 * See tcp_notify(). 536 */ 537 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 538 sc->sc_flags |= SCF_UNREACH; 539 goto done; 540 } 541 syncache_drop(sc, sch); 542 tcpstat.tcps_sc_unreach++; 543done: 544 SCH_UNLOCK(sch); 545} 546 547/* 548 * Build a new TCP socket structure from a syncache entry. 549 */ 550static struct socket * 551syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 552{ 553 struct inpcb *inp = NULL; 554 struct socket *so; 555 struct tcpcb *tp; 556 557 NET_ASSERT_GIANT(); 558 INP_INFO_WLOCK_ASSERT(&tcbinfo); 559 560 /* 561 * Ok, create the full blown connection, and set things up 562 * as they would have been set up if we had created the 563 * connection when the SYN arrived. If we can't create 564 * the connection, abort it. 565 */ 566 so = sonewconn(lso, SS_ISCONNECTED); 567 if (so == NULL) { 568 /* 569 * Drop the connection; we will send a RST if the peer 570 * retransmits the ACK, 571 */ 572 tcpstat.tcps_listendrop++; 573 goto abort2; 574 } 575#ifdef MAC 576 SOCK_LOCK(so); 577 mac_set_socket_peer_from_mbuf(m, so); 578 SOCK_UNLOCK(so); 579#endif 580 581 inp = sotoinpcb(so); 582 INP_LOCK(inp); 583 584 /* Insert new socket into PCB hash list. */ 585 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 586#ifdef INET6 587 if (sc->sc_inc.inc_isipv6) { 588 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 589 } else { 590 inp->inp_vflag &= ~INP_IPV6; 591 inp->inp_vflag |= INP_IPV4; 592#endif 593 inp->inp_laddr = sc->sc_inc.inc_laddr; 594#ifdef INET6 595 } 596#endif 597 inp->inp_lport = sc->sc_inc.inc_lport; 598 if (in_pcbinshash(inp) != 0) { 599 /* 600 * Undo the assignments above if we failed to 601 * put the PCB on the hash lists. 602 */ 603#ifdef INET6 604 if (sc->sc_inc.inc_isipv6) 605 inp->in6p_laddr = in6addr_any; 606 else 607#endif 608 inp->inp_laddr.s_addr = INADDR_ANY; 609 inp->inp_lport = 0; 610 goto abort; 611 } 612#ifdef IPSEC 613 /* Copy old policy into new socket's. */ 614 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 615 printf("syncache_socket: could not copy policy\n"); 616#endif 617#ifdef FAST_IPSEC 618 /* Copy old policy into new socket's. */ 619 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 620 printf("syncache_socket: could not copy policy\n"); 621#endif 622#ifdef INET6 623 if (sc->sc_inc.inc_isipv6) { 624 struct inpcb *oinp = sotoinpcb(lso); 625 struct in6_addr laddr6; 626 struct sockaddr_in6 sin6; 627 /* 628 * Inherit socket options from the listening socket. 629 * Note that in6p_inputopts are not (and should not be) 630 * copied, since it stores previously received options and is 631 * used to detect if each new option is different than the 632 * previous one and hence should be passed to a user. 633 * If we copied in6p_inputopts, a user would not be able to 634 * receive options just after calling the accept system call. 635 */ 636 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 637 if (oinp->in6p_outputopts) 638 inp->in6p_outputopts = 639 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 640 641 sin6.sin6_family = AF_INET6; 642 sin6.sin6_len = sizeof(sin6); 643 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 644 sin6.sin6_port = sc->sc_inc.inc_fport; 645 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 646 laddr6 = inp->in6p_laddr; 647 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 648 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 649 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 650 thread0.td_ucred)) { 651 inp->in6p_laddr = laddr6; 652 goto abort; 653 } 654 /* Override flowlabel from in6_pcbconnect. */ 655 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 656 inp->in6p_flowinfo |= sc->sc_flowlabel; 657 } else 658#endif 659 { 660 struct in_addr laddr; 661 struct sockaddr_in sin; 662 663 inp->inp_options = ip_srcroute(m); 664 if (inp->inp_options == NULL) { 665 inp->inp_options = sc->sc_ipopts; 666 sc->sc_ipopts = NULL; 667 } 668 669 sin.sin_family = AF_INET; 670 sin.sin_len = sizeof(sin); 671 sin.sin_addr = sc->sc_inc.inc_faddr; 672 sin.sin_port = sc->sc_inc.inc_fport; 673 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 674 laddr = inp->inp_laddr; 675 if (inp->inp_laddr.s_addr == INADDR_ANY) 676 inp->inp_laddr = sc->sc_inc.inc_laddr; 677 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 678 thread0.td_ucred)) { 679 inp->inp_laddr = laddr; 680 goto abort; 681 } 682 } 683 tp = intotcpcb(inp); 684 tp->t_state = TCPS_SYN_RECEIVED; 685 tp->iss = sc->sc_iss; 686 tp->irs = sc->sc_irs; 687 tcp_rcvseqinit(tp); 688 tcp_sendseqinit(tp); 689 tp->snd_wl1 = sc->sc_irs; 690 tp->snd_max = tp->iss + 1; 691 tp->snd_nxt = tp->iss + 1; 692 tp->rcv_up = sc->sc_irs + 1; 693 tp->rcv_wnd = sc->sc_wnd; 694 tp->rcv_adv += tp->rcv_wnd; 695 tp->last_ack_sent = tp->rcv_nxt; 696 697 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 698 if (sc->sc_flags & SCF_NOOPT) 699 tp->t_flags |= TF_NOOPT; 700 else { 701 if (sc->sc_flags & SCF_WINSCALE) { 702 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 703 tp->snd_scale = sc->sc_requested_s_scale; 704 tp->request_r_scale = sc->sc_requested_r_scale; 705 } 706 if (sc->sc_flags & SCF_TIMESTAMP) { 707 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 708 tp->ts_recent = sc->sc_tsreflect; 709 tp->ts_recent_age = ticks; 710 tp->ts_offset = sc->sc_tsoff; 711 } 712#ifdef TCP_SIGNATURE 713 if (sc->sc_flags & SCF_SIGNATURE) 714 tp->t_flags |= TF_SIGNATURE; 715#endif 716 if (sc->sc_flags & SCF_SACK) { 717 tp->sack_enable = 1; 718 tp->t_flags |= TF_SACK_PERMIT; 719 } 720 } 721 722 /* 723 * Set up MSS and get cached values from tcp_hostcache. 724 * This might overwrite some of the defaults we just set. 725 */ 726 tcp_mss(tp, sc->sc_peer_mss); 727 728 /* 729 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 730 */ 731 if (sc->sc_rxmits > 1) 732 tp->snd_cwnd = tp->t_maxseg; 733 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 734 735 INP_UNLOCK(inp); 736 737 tcpstat.tcps_accepts++; 738 return (so); 739 740abort: 741 INP_UNLOCK(inp); 742abort2: 743 if (so != NULL) 744 soabort(so); 745 return (NULL); 746} 747 748/* 749 * This function gets called when we receive an ACK for a 750 * socket in the LISTEN state. We look up the connection 751 * in the syncache, and if its there, we pull it out of 752 * the cache and turn it into a full-blown connection in 753 * the SYN-RECEIVED state. 754 */ 755int 756syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 757 struct socket **lsop, struct mbuf *m) 758{ 759 struct syncache *sc; 760 struct syncache_head *sch; 761 struct socket *so; 762 struct syncache scs; 763 764 /* 765 * Global TCP locks are held because we manipulate the PCB lists 766 * and create a new socket. 767 */ 768 INP_INFO_WLOCK_ASSERT(&tcbinfo); 769 770 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 771 SCH_LOCK_ASSERT(sch); 772 if (sc == NULL) { 773 /* 774 * There is no syncache entry, so see if this ACK is 775 * a returning syncookie. To do this, first: 776 * A. See if this socket has had a syncache entry dropped in 777 * the past. We don't want to accept a bogus syncookie 778 * if we've never received a SYN. 779 * B. check that the syncookie is valid. If it is, then 780 * cobble up a fake syncache entry, and return. 781 */ 782 if (!tcp_syncookies) { 783 SCH_UNLOCK(sch); 784 goto failed; 785 } 786 bzero(&scs, sizeof(scs)); 787 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 788 SCH_UNLOCK(sch); 789 if (sc == NULL) 790 goto failed; 791 tcpstat.tcps_sc_recvcookie++; 792 } else { 793 /* Pull out the entry to unlock the bucket row. */ 794 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 795 sch->sch_length--; 796 tcp_syncache.cache_count--; 797 SCH_UNLOCK(sch); 798 } 799 800 /* 801 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 802 */ 803 if (th->th_ack != sc->sc_iss + 1) 804 goto failed; 805 806 so = syncache_socket(sc, *lsop, m); 807 808 if (so == NULL) { 809#if 0 810resetandabort: 811 /* XXXjlemon check this - is this correct? */ 812 (void) tcp_respond(NULL, m, m, th, 813 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 814#endif 815 m_freem(m); /* XXX: only needed for above */ 816 tcpstat.tcps_sc_aborted++; 817 if (sc != &scs) { 818 syncache_insert(sc, sch); /* try again later */ 819 sc = NULL; 820 } 821 goto failed; 822 } else 823 tcpstat.tcps_sc_completed++; 824 *lsop = so; 825 826 if (sc != &scs) 827 syncache_free(sc); 828 return (1); 829failed: 830 if (sc != NULL && sc != &scs) 831 syncache_free(sc); 832 return (0); 833} 834 835/* 836 * Given a LISTEN socket and an inbound SYN request, add 837 * this to the syn cache, and send back a segment: 838 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 839 * to the source. 840 * 841 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 842 * Doing so would require that we hold onto the data and deliver it 843 * to the application. However, if we are the target of a SYN-flood 844 * DoS attack, an attacker could send data which would eventually 845 * consume all available buffer space if it were ACKed. By not ACKing 846 * the data, we avoid this DoS scenario. 847 */ 848int 849syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 850 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 851{ 852 struct tcpcb *tp; 853 struct socket *so; 854 struct syncache *sc = NULL; 855 struct syncache_head *sch; 856 struct mbuf *ipopts = NULL; 857 u_int32_t flowtmp; 858 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 859#ifdef INET6 860 int autoflowlabel = 0; 861#endif 862#ifdef MAC 863 struct label *maclabel; 864#endif 865 struct syncache scs; 866 867 INP_INFO_WLOCK_ASSERT(&tcbinfo); 868 INP_LOCK_ASSERT(inp); /* listen socket */ 869 870 /* 871 * Combine all so/tp operations very early to drop the INP lock as 872 * soon as possible. 873 */ 874 so = *lsop; 875 tp = sototcpcb(so); 876 877#ifdef INET6 878 if (inc->inc_isipv6 && 879 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 880 autoflowlabel = 1; 881#endif 882 ip_ttl = inp->inp_ip_ttl; 883 ip_tos = inp->inp_ip_tos; 884 win = sbspace(&so->so_rcv); 885 sb_hiwat = so->so_rcv.sb_hiwat; 886 noopt = (tp->t_flags & TF_NOOPT); 887 888 so = NULL; 889 tp = NULL; 890 891#ifdef MAC 892 if (mac_init_syncache(&maclabel) != 0) { 893 INP_UNLOCK(inp); 894 INP_INFO_WUNLOCK(&tcbinfo); 895 goto done; 896 } else 897 mac_init_syncache_from_inpcb(maclabel, inp); 898#endif 899 INP_UNLOCK(inp); 900 INP_INFO_WUNLOCK(&tcbinfo); 901 902 /* 903 * Remember the IP options, if any. 904 */ 905#ifdef INET6 906 if (!inc->inc_isipv6) 907#endif 908 ipopts = ip_srcroute(m); 909 910 /* 911 * See if we already have an entry for this connection. 912 * If we do, resend the SYN,ACK, and reset the retransmit timer. 913 * 914 * XXX: should the syncache be re-initialized with the contents 915 * of the new SYN here (which may have different options?) 916 */ 917 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 918 SCH_LOCK_ASSERT(sch); 919 if (sc != NULL) { 920 tcpstat.tcps_sc_dupsyn++; 921 if (ipopts) { 922 /* 923 * If we were remembering a previous source route, 924 * forget it and use the new one we've been given. 925 */ 926 if (sc->sc_ipopts) 927 (void) m_free(sc->sc_ipopts); 928 sc->sc_ipopts = ipopts; 929 } 930 /* 931 * Update timestamp if present. 932 */ 933 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 934 sc->sc_tsreflect = to->to_tsval; 935 else 936 sc->sc_flags &= ~SCF_TIMESTAMP; 937#ifdef MAC 938 /* 939 * Since we have already unconditionally allocated label 940 * storage, free it up. The syncache entry will already 941 * have an initialized label we can use. 942 */ 943 mac_destroy_syncache(&maclabel); 944 KASSERT(sc->sc_label != NULL, 945 ("%s: label not initialized", __func__)); 946#endif 947 if (syncache_respond(sc) == 0) { 948 SYNCACHE_TIMEOUT(sc, sch, 1); 949 tcpstat.tcps_sndacks++; 950 tcpstat.tcps_sndtotal++; 951 } 952 SCH_UNLOCK(sch); 953 goto done; 954 } 955 956 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 957 if (sc == NULL) { 958 /* 959 * The zone allocator couldn't provide more entries. 960 * Treat this as if the cache was full; drop the oldest 961 * entry and insert the new one. 962 */ 963 tcpstat.tcps_sc_zonefail++; 964 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 965 syncache_drop(sc, sch); 966 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 967 if (sc == NULL) { 968 if (tcp_syncookies) { 969 bzero(&scs, sizeof(scs)); 970 sc = &scs; 971 } else { 972 SCH_UNLOCK(sch); 973 if (ipopts) 974 (void) m_free(ipopts); 975 goto done; 976 } 977 } 978 } 979 980 /* 981 * Fill in the syncache values. 982 */ 983#ifdef MAC 984 sc->sc_label = maclabel; 985#endif 986 sc->sc_ipopts = ipopts; 987 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 988#ifdef INET6 989 if (!inc->inc_isipv6) 990#endif 991 { 992 sc->sc_ip_tos = ip_tos; 993 sc->sc_ip_ttl = ip_ttl; 994 } 995 996 sc->sc_irs = th->th_seq; 997 sc->sc_iss = arc4random(); 998 sc->sc_flags = 0; 999 sc->sc_flowlabel = 0; 1000 1001 /* 1002 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1003 * win was derived from socket earlier in the function. 1004 */ 1005 win = imax(win, 0); 1006 win = imin(win, TCP_MAXWIN); 1007 sc->sc_wnd = win; 1008 1009 if (tcp_do_rfc1323) { 1010 /* 1011 * A timestamp received in a SYN makes 1012 * it ok to send timestamp requests and replies. 1013 */ 1014 if (to->to_flags & TOF_TS) { 1015 sc->sc_tsreflect = to->to_tsval; 1016 sc->sc_flags |= SCF_TIMESTAMP; 1017 } 1018 if (to->to_flags & TOF_SCALE) { 1019 int wscale = 0; 1020 1021 /* 1022 * Compute proper scaling value from buffer space. 1023 * Leave enough room for the socket buffer to grow 1024 * with auto sizing. This allows us to scale the 1025 * receive buffer over a wide range while not losing 1026 * any efficiency or fine granularity. 1027 * 1028 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1029 * or <SYN,ACK>) segment itself is never scaled. 1030 */ 1031 while (wscale < TCP_MAX_WINSHIFT && 1032 (0x1 << wscale) < tcp_minmss) 1033 wscale++; 1034 sc->sc_requested_r_scale = wscale; 1035 sc->sc_requested_s_scale = to->to_wscale; 1036 sc->sc_flags |= SCF_WINSCALE; 1037 } 1038 } 1039#ifdef TCP_SIGNATURE 1040 /* 1041 * If listening socket requested TCP digests, and received SYN 1042 * contains the option, flag this in the syncache so that 1043 * syncache_respond() will do the right thing with the SYN+ACK. 1044 * XXX: Currently we always record the option by default and will 1045 * attempt to use it in syncache_respond(). 1046 */ 1047 if (to->to_flags & TOF_SIGNATURE) 1048 sc->sc_flags |= SCF_SIGNATURE; 1049#endif 1050 if (to->to_flags & TOF_SACK) 1051 sc->sc_flags |= SCF_SACK; 1052 if (to->to_flags & TOF_MSS) 1053 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1054 if (noopt) 1055 sc->sc_flags |= SCF_NOOPT; 1056 1057 if (tcp_syncookies) { 1058 syncookie_generate(sch, sc, &flowtmp); 1059#ifdef INET6 1060 if (autoflowlabel) 1061 sc->sc_flowlabel = flowtmp; 1062#endif 1063 } else { 1064#ifdef INET6 1065 if (autoflowlabel) 1066 sc->sc_flowlabel = 1067 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1068#endif 1069 } 1070 SCH_UNLOCK(sch); 1071 1072 /* 1073 * Do a standard 3-way handshake. 1074 */ 1075 if (syncache_respond(sc) == 0) { 1076 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs) 1077 syncache_free(sc); 1078 else if (sc != &scs) 1079 syncache_insert(sc, sch); /* locks and unlocks sch */ 1080 tcpstat.tcps_sndacks++; 1081 tcpstat.tcps_sndtotal++; 1082 } else { 1083 if (sc != &scs) 1084 syncache_free(sc); 1085 tcpstat.tcps_sc_dropped++; 1086 } 1087 1088done: 1089#ifdef MAC 1090 if (sc == &scs) 1091 mac_destroy_syncache(&maclabel); 1092#endif 1093 *lsop = NULL; 1094 m_freem(m); 1095 return (1); 1096} 1097 1098static int 1099syncache_respond(struct syncache *sc) 1100{ 1101 struct ip *ip = NULL; 1102 struct mbuf *m; 1103 struct tcphdr *th; 1104 int optlen, error; 1105 u_int16_t hlen, tlen, mssopt; 1106 struct tcpopt to; 1107#ifdef INET6 1108 struct ip6_hdr *ip6 = NULL; 1109#endif 1110 1111 hlen = 1112#ifdef INET6 1113 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1114#endif 1115 sizeof(struct ip); 1116 tlen = hlen + sizeof(struct tcphdr); 1117 1118 /* Determine MSS we advertize to other end of connection. */ 1119 mssopt = tcp_mssopt(&sc->sc_inc); 1120 if (sc->sc_peer_mss) 1121 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss); 1122 1123 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1124 KASSERT(max_linkhdr + tlen + MAX_TCPOPTLEN <= MHLEN, 1125 ("syncache: mbuf too small")); 1126 1127 /* Create the IP+TCP header from scratch. */ 1128 m = m_gethdr(M_DONTWAIT, MT_DATA); 1129 if (m == NULL) 1130 return (ENOBUFS); 1131#ifdef MAC 1132 mac_create_mbuf_from_syncache(sc->sc_label, m); 1133#endif 1134 m->m_data += max_linkhdr; 1135 m->m_len = tlen; 1136 m->m_pkthdr.len = tlen; 1137 m->m_pkthdr.rcvif = NULL; 1138 1139#ifdef INET6 1140 if (sc->sc_inc.inc_isipv6) { 1141 ip6 = mtod(m, struct ip6_hdr *); 1142 ip6->ip6_vfc = IPV6_VERSION; 1143 ip6->ip6_nxt = IPPROTO_TCP; 1144 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1145 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1146 ip6->ip6_plen = htons(tlen - hlen); 1147 /* ip6_hlim is set after checksum */ 1148 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1149 ip6->ip6_flow |= sc->sc_flowlabel; 1150 1151 th = (struct tcphdr *)(ip6 + 1); 1152 } else 1153#endif 1154 { 1155 ip = mtod(m, struct ip *); 1156 ip->ip_v = IPVERSION; 1157 ip->ip_hl = sizeof(struct ip) >> 2; 1158 ip->ip_len = tlen; 1159 ip->ip_id = 0; 1160 ip->ip_off = 0; 1161 ip->ip_sum = 0; 1162 ip->ip_p = IPPROTO_TCP; 1163 ip->ip_src = sc->sc_inc.inc_laddr; 1164 ip->ip_dst = sc->sc_inc.inc_faddr; 1165 ip->ip_ttl = sc->sc_ip_ttl; 1166 ip->ip_tos = sc->sc_ip_tos; 1167 1168 /* 1169 * See if we should do MTU discovery. Route lookups are 1170 * expensive, so we will only unset the DF bit if: 1171 * 1172 * 1) path_mtu_discovery is disabled 1173 * 2) the SCF_UNREACH flag has been set 1174 */ 1175 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1176 ip->ip_off |= IP_DF; 1177 1178 th = (struct tcphdr *)(ip + 1); 1179 } 1180 th->th_sport = sc->sc_inc.inc_lport; 1181 th->th_dport = sc->sc_inc.inc_fport; 1182 1183 th->th_seq = htonl(sc->sc_iss); 1184 th->th_ack = htonl(sc->sc_irs + 1); 1185 th->th_off = sizeof(struct tcphdr) >> 2; 1186 th->th_x2 = 0; 1187 th->th_flags = TH_SYN|TH_ACK; 1188 th->th_win = htons(sc->sc_wnd); 1189 th->th_urp = 0; 1190 1191 /* Tack on the TCP options. */ 1192 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1193 to.to_flags = 0; 1194 1195 to.to_mss = mssopt; 1196 to.to_flags = TOF_MSS; 1197 if (sc->sc_flags & SCF_WINSCALE) { 1198 to.to_wscale = sc->sc_requested_r_scale; 1199 to.to_flags |= TOF_SCALE; 1200 } 1201 if (sc->sc_flags & SCF_TIMESTAMP) { 1202 /* Virgin timestamp or TCP cookie enhanced one. */ 1203 to.to_tsval = sc->sc_ts ? sc->sc_ts : ticks; 1204 to.to_tsecr = sc->sc_tsreflect; 1205 to.to_flags |= TOF_TS; 1206 } 1207 if (sc->sc_flags & SCF_SACK) 1208 to.to_flags |= TOF_SACKPERM; 1209#ifdef TCP_SIGNATURE 1210 if (sc->sc_flags & SCF_SIGNATURE) 1211 to.to_flags |= TOF_SIGNATURE; 1212#endif 1213 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1214 1215#ifdef TCP_SIGNATURE 1216 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1217 to.to_signature, IPSEC_DIR_OUTBOUND); 1218#endif 1219 1220 /* Adjust headers by option size. */ 1221 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1222 m->m_len += optlen; 1223 m->m_pkthdr.len += optlen; 1224#ifdef INET6 1225 if (sc->sc_inc.inc_isipv6) 1226 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1227 else 1228#endif 1229 ip->ip_len += optlen; 1230 } else 1231 optlen = 0; 1232 1233#ifdef INET6 1234 if (sc->sc_inc.inc_isipv6) { 1235 th->th_sum = 0; 1236 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1237 tlen + optlen - hlen); 1238 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1239 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1240 } else 1241#endif 1242 { 1243 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1244 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1245 m->m_pkthdr.csum_flags = CSUM_TCP; 1246 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1247 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1248 } 1249 return (error); 1250} 1251 1252/* 1253 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1254 * receive and to be able to handle SYN floods from bogus source addresses 1255 * (where we will never receive any reply). SYN floods try to exhaust all 1256 * our memory and available slots in the SYN cache table to cause a denial 1257 * of service to legitimate users of the local host. 1258 * 1259 * The idea of SYN cookies is to encode and include all necessary information 1260 * about the connection setup state within the SYN-ACK we send back and thus 1261 * to get along without keeping any local state until the ACK to the SYN-ACK 1262 * arrives (if ever). Everything we need to know should be available from 1263 * the information we encoded in the SYN-ACK. 1264 * 1265 * More information about the theory behind SYN cookies and its first 1266 * discussion and specification can be found at: 1267 * http://cr.yp.to/syncookies.html (overview) 1268 * http://cr.yp.to/syncookies/archive (gory details) 1269 * 1270 * This implementation extends the orginal idea and first implementation 1271 * of FreeBSD by using not only the initial sequence number field to store 1272 * information but also the timestamp field if present. This way we can 1273 * keep track of the entire state we need to know to recreate the session in 1274 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1275 * these days. For those that do not we still have to live with the known 1276 * shortcomings of the ISN only SYN cookies. 1277 * 1278 * Cookie layers: 1279 * 1280 * Initial sequence number we send: 1281 * 31|................................|0 1282 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1283 * D = MD5 Digest (first dword) 1284 * M = MSS index 1285 * R = Rotation of secret 1286 * P = Odd or Even secret 1287 * 1288 * The MD5 Digest is computed with over following parameters: 1289 * a) randomly rotated secret 1290 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1291 * c) the received initial sequence number from remote host 1292 * d) the rotation offset and odd/even bit 1293 * 1294 * Timestamp we send: 1295 * 31|................................|0 1296 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1297 * D = MD5 Digest (third dword) (only as filler) 1298 * S = Requested send window scale 1299 * R = Requested receive window scale 1300 * A = SACK allowed 1301 * 5 = TCP-MD5 enabled (not implemented yet) 1302 * XORed with MD5 Digest (forth dword) 1303 * 1304 * The timestamp isn't cryptographically secure and doesn't need to be. 1305 * The double use of the MD5 digest dwords ties it to a specific remote/ 1306 * local host/port, remote initial sequence number and our local time 1307 * limited secret. A received timestamp is reverted (XORed) and then 1308 * the contained MD5 dword is compared to the computed one to ensure the 1309 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1310 * have been tampered with but this isn't different from supplying bogus 1311 * values in the SYN in the first place. 1312 * 1313 * Some problems with SYN cookies remain however: 1314 * Consider the problem of a recreated (and retransmitted) cookie. If the 1315 * original SYN was accepted, the connection is established. The second 1316 * SYN is inflight, and if it arrives with an ISN that falls within the 1317 * receive window, the connection is killed. 1318 * 1319 * Notes: 1320 * A heuristic to determine when to accept syn cookies is not necessary. 1321 * An ACK flood would cause the syncookie verification to be attempted, 1322 * but a SYN flood causes syncookies to be generated. Both are of equal 1323 * cost, so there's no point in trying to optimize the ACK flood case. 1324 * Also, if you don't process certain ACKs for some reason, then all someone 1325 * would have to do is launch a SYN and ACK flood at the same time, which 1326 * would stop cookie verification and defeat the entire purpose of syncookies. 1327 */ 1328static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1329 1330static void 1331syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1332 u_int32_t *flowlabel) 1333{ 1334 MD5_CTX ctx; 1335 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1336 u_int32_t data; 1337 u_int32_t *secbits; 1338 u_int off, pmss, mss; 1339 int i; 1340 1341 SCH_LOCK_ASSERT(sch); 1342 1343 /* Which of the two secrets to use. */ 1344 secbits = sch->sch_oddeven ? 1345 sch->sch_secbits_odd : sch->sch_secbits_even; 1346 1347 /* Reseed secret if too old. */ 1348 if (sch->sch_reseed < time_uptime) { 1349 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1350 secbits = sch->sch_oddeven ? 1351 sch->sch_secbits_odd : sch->sch_secbits_even; 1352 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1353 secbits[i] = arc4random(); 1354 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1355 } 1356 1357 /* Secret rotation offset. */ 1358 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1359 1360 /* Maximum segment size calculation. */ 1361 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss); 1362 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1363 if (tcp_sc_msstab[mss] <= pmss) 1364 break; 1365 1366 /* Fold parameters and MD5 digest into the ISN we will send. */ 1367 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1368 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1369 data |= mss << 4; /* mss, 3 bits */ 1370 1371 MD5Init(&ctx); 1372 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1373 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1374 MD5Update(&ctx, secbits, off); 1375 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1376 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1377 MD5Update(&ctx, &data, sizeof(data)); 1378 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1379 1380 data |= (md5_buffer[0] << 7); 1381 sc->sc_iss = data; 1382 1383#ifdef INET6 1384 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1385#endif 1386 1387 /* Additional parameters are stored in the timestamp if present. */ 1388 if (sc->sc_flags & SCF_TIMESTAMP) { 1389 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1390 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1391 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1392 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1393 data |= md5_buffer[2] << 10; /* more digest bits */ 1394 data ^= md5_buffer[3]; 1395 sc->sc_ts = data; 1396 sc->sc_tsoff = data - ticks; /* after XOR */ 1397 } else 1398 sc->sc_ts = 0; 1399 1400 return; 1401} 1402 1403static struct syncache * 1404syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1405 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1406 struct socket *so) 1407{ 1408 MD5_CTX ctx; 1409 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1410 u_int32_t data = 0; 1411 u_int32_t *secbits; 1412 tcp_seq ack, seq; 1413 int off, mss, wnd, flags; 1414 1415 SCH_LOCK_ASSERT(sch); 1416 1417 /* 1418 * Pull information out of SYN-ACK/ACK and 1419 * revert sequence number advances. 1420 */ 1421 ack = th->th_ack - 1; 1422 seq = th->th_seq - 1; 1423 off = (ack >> 1) & 0x7; 1424 mss = (ack >> 4) & 0x7; 1425 flags = ack & 0x7f; 1426 1427 /* Which of the two secrets to use. */ 1428 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1429 1430 /* 1431 * The secret wasn't updated for the lifetime of a syncookie, 1432 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1433 */ 1434 if (sch->sch_reseed < time_uptime) { 1435 return (NULL); 1436 } 1437 1438 /* Recompute the digest so we can compare it. */ 1439 MD5Init(&ctx); 1440 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1441 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1442 MD5Update(&ctx, secbits, off); 1443 MD5Update(&ctx, inc, sizeof(*inc)); 1444 MD5Update(&ctx, &seq, sizeof(seq)); 1445 MD5Update(&ctx, &flags, sizeof(flags)); 1446 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1447 1448 /* Does the digest part of or ACK'ed ISS match? */ 1449 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1450 return (NULL); 1451 1452 /* Does the digest part of our reflected timestamp match? */ 1453 if (to->to_flags & TOF_TS) { 1454 data = md5_buffer[3] ^ to->to_tsecr; 1455 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1456 return (NULL); 1457 } 1458 1459 /* Fill in the syncache values. */ 1460 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1461 sc->sc_ipopts = NULL; 1462 1463 sc->sc_irs = seq; 1464 sc->sc_iss = ack; 1465 1466#ifdef INET6 1467 if (inc->inc_isipv6) { 1468 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1469 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1470 } else 1471#endif 1472 { 1473 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1474 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1475 } 1476 1477 /* Additional parameters that were encoded in the timestamp. */ 1478 if (data) { 1479 sc->sc_flags |= SCF_TIMESTAMP; 1480 sc->sc_tsreflect = to->to_tsval; 1481 sc->sc_tsoff = to->to_tsecr - ticks; 1482 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1483 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1484 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1485 TCP_MAX_WINSHIFT); 1486 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1487 TCP_MAX_WINSHIFT); 1488 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1489 sc->sc_flags |= SCF_WINSCALE; 1490 } else 1491 sc->sc_flags |= SCF_NOOPT; 1492 1493 wnd = sbspace(&so->so_rcv); 1494 wnd = imax(wnd, 0); 1495 wnd = imin(wnd, TCP_MAXWIN); 1496 sc->sc_wnd = wnd; 1497 1498 sc->sc_rxmits = 0; 1499 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1500 1501 return (sc); 1502} 1503