tcp_syncache.c revision 170385
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 170385 2007-06-06 22:10:12Z 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#include <sys/syslog.h> 54 55#include <vm/uma.h> 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/ip.h> 63#include <netinet/in_var.h> 64#include <netinet/in_pcb.h> 65#include <netinet/ip_var.h> 66#include <netinet/ip_options.h> 67#ifdef INET6 68#include <netinet/ip6.h> 69#include <netinet/icmp6.h> 70#include <netinet6/nd6.h> 71#include <netinet6/ip6_var.h> 72#include <netinet6/in6_pcb.h> 73#endif 74#include <netinet/tcp.h> 75#include <netinet/tcp_fsm.h> 76#include <netinet/tcp_seq.h> 77#include <netinet/tcp_timer.h> 78#include <netinet/tcp_var.h> 79#ifdef INET6 80#include <netinet6/tcp6_var.h> 81#endif 82 83#ifdef IPSEC 84#include <netinet6/ipsec.h> 85#ifdef INET6 86#include <netinet6/ipsec6.h> 87#endif 88#endif /*IPSEC*/ 89 90#ifdef FAST_IPSEC 91#include <netipsec/ipsec.h> 92#ifdef INET6 93#include <netipsec/ipsec6.h> 94#endif 95#include <netipsec/key.h> 96#endif /*FAST_IPSEC*/ 97 98#include <machine/in_cksum.h> 99 100#include <security/mac/mac_framework.h> 101 102static int tcp_syncookies = 1; 103SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 104 &tcp_syncookies, 0, 105 "Use TCP SYN cookies if the syncache overflows"); 106 107static int tcp_syncookiesonly = 0; 108SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW, 109 &tcp_syncookiesonly, 0, 110 "Use only TCP SYN cookies"); 111 112#define SYNCOOKIE_SECRET_SIZE 8 /* dwords */ 113#define SYNCOOKIE_LIFETIME 16 /* seconds */ 114 115struct syncache { 116 TAILQ_ENTRY(syncache) sc_hash; 117 struct in_conninfo sc_inc; /* addresses */ 118 u_long sc_rxttime; /* retransmit time */ 119 u_int16_t sc_rxmits; /* retransmit counter */ 120 121 u_int32_t sc_tsreflect; /* timestamp to reflect */ 122 u_int32_t sc_ts; /* our timestamp to send */ 123 u_int32_t sc_tsoff; /* ts offset w/ syncookies */ 124 u_int32_t sc_flowlabel; /* IPv6 flowlabel */ 125 tcp_seq sc_irs; /* seq from peer */ 126 tcp_seq sc_iss; /* our ISS */ 127 struct mbuf *sc_ipopts; /* source route */ 128 129 u_int16_t sc_peer_mss; /* peer's MSS */ 130 u_int16_t sc_wnd; /* advertised window */ 131 u_int8_t sc_ip_ttl; /* IPv4 TTL */ 132 u_int8_t sc_ip_tos; /* IPv4 TOS */ 133 u_int8_t sc_requested_s_scale:4, 134 sc_requested_r_scale:4; 135 u_int8_t sc_flags; 136#define SCF_NOOPT 0x01 /* no TCP options */ 137#define SCF_WINSCALE 0x02 /* negotiated window scaling */ 138#define SCF_TIMESTAMP 0x04 /* negotiated timestamps */ 139 /* MSS is implicit */ 140#define SCF_UNREACH 0x10 /* icmp unreachable received */ 141#define SCF_SIGNATURE 0x20 /* send MD5 digests */ 142#define SCF_SACK 0x80 /* send SACK option */ 143#ifdef MAC 144 struct label *sc_label; /* MAC label reference */ 145#endif 146}; 147 148struct syncache_head { 149 struct mtx sch_mtx; 150 TAILQ_HEAD(sch_head, syncache) sch_bucket; 151 struct callout sch_timer; 152 int sch_nextc; 153 u_int sch_length; 154 u_int sch_oddeven; 155 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE]; 156 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE]; 157 u_int sch_reseed; /* time_uptime, seconds */ 158}; 159 160static void syncache_drop(struct syncache *, struct syncache_head *); 161static void syncache_free(struct syncache *); 162static void syncache_insert(struct syncache *, struct syncache_head *); 163struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 164static int syncache_respond(struct syncache *); 165static struct socket *syncache_socket(struct syncache *, struct socket *, 166 struct mbuf *m); 167static void syncache_timer(void *); 168static void syncookie_generate(struct syncache_head *, struct syncache *, 169 u_int32_t *); 170static struct syncache 171 *syncookie_lookup(struct in_conninfo *, struct syncache_head *, 172 struct syncache *, struct tcpopt *, struct tcphdr *, 173 struct socket *); 174 175/* 176 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 177 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 178 * the odds are that the user has given up attempting to connect by then. 179 */ 180#define SYNCACHE_MAXREXMTS 3 181 182/* Arbitrary values */ 183#define TCP_SYNCACHE_HASHSIZE 512 184#define TCP_SYNCACHE_BUCKETLIMIT 30 185 186struct tcp_syncache { 187 struct syncache_head *hashbase; 188 uma_zone_t zone; 189 u_int hashsize; 190 u_int hashmask; 191 u_int bucket_limit; 192 u_int cache_count; /* XXX: unprotected */ 193 u_int cache_limit; 194 u_int rexmt_limit; 195 u_int hash_secret; 196}; 197static struct tcp_syncache tcp_syncache; 198 199SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 200 201SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 202 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 203 204SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 205 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 206 207SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 208 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 209 210SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 211 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 212 213SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 214 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 215 216int tcp_sc_rst_sock_fail = 1; 217SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW, 218 &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure"); 219 220static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 221 222#define SYNCACHE_HASH(inc, mask) \ 223 ((tcp_syncache.hash_secret ^ \ 224 (inc)->inc_faddr.s_addr ^ \ 225 ((inc)->inc_faddr.s_addr >> 16) ^ \ 226 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 227 228#define SYNCACHE_HASH6(inc, mask) \ 229 ((tcp_syncache.hash_secret ^ \ 230 (inc)->inc6_faddr.s6_addr32[0] ^ \ 231 (inc)->inc6_faddr.s6_addr32[3] ^ \ 232 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 233 234#define ENDPTS_EQ(a, b) ( \ 235 (a)->ie_fport == (b)->ie_fport && \ 236 (a)->ie_lport == (b)->ie_lport && \ 237 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 238 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 239) 240 241#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 242 243#define SYNCACHE_TIMEOUT(sc, sch, co) do { \ 244 (sc)->sc_rxmits++; \ 245 (sc)->sc_rxttime = ticks + \ 246 TCPTV_RTOBASE * tcp_backoff[(sc)->sc_rxmits - 1]; \ 247 if ((sch)->sch_nextc > (sc)->sc_rxttime) \ 248 (sch)->sch_nextc = (sc)->sc_rxttime; \ 249 if (!TAILQ_EMPTY(&(sch)->sch_bucket) && !(co)) \ 250 callout_reset(&(sch)->sch_timer, \ 251 (sch)->sch_nextc - ticks, \ 252 syncache_timer, (void *)(sch)); \ 253} while (0) 254 255#define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 256#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 257#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 258 259/* 260 * Requires the syncache entry to be already removed from the bucket list. 261 */ 262static void 263syncache_free(struct syncache *sc) 264{ 265 if (sc->sc_ipopts) 266 (void) m_free(sc->sc_ipopts); 267#ifdef MAC 268 mac_destroy_syncache(&sc->sc_label); 269#endif 270 271 uma_zfree(tcp_syncache.zone, sc); 272} 273 274void 275syncache_init(void) 276{ 277 int i; 278 279 tcp_syncache.cache_count = 0; 280 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 281 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 282 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 283 tcp_syncache.hash_secret = arc4random(); 284 285 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 286 &tcp_syncache.hashsize); 287 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 288 &tcp_syncache.bucket_limit); 289 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 290 printf("WARNING: syncache hash size is not a power of 2.\n"); 291 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 292 } 293 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 294 295 /* Set limits. */ 296 tcp_syncache.cache_limit = 297 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 299 &tcp_syncache.cache_limit); 300 301 /* Allocate the hash table. */ 302 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 303 tcp_syncache.hashsize * sizeof(struct syncache_head), 304 M_SYNCACHE, M_WAITOK | M_ZERO); 305 306 /* Initialize the hash buckets. */ 307 for (i = 0; i < tcp_syncache.hashsize; i++) { 308 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 309 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 310 NULL, MTX_DEF); 311 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer, 312 &tcp_syncache.hashbase[i].sch_mtx, 0); 313 tcp_syncache.hashbase[i].sch_length = 0; 314 } 315 316 /* Create the syncache entry zone. */ 317 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 318 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 319 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 320} 321 322/* 323 * Inserts a syncache entry into the specified bucket row. 324 * Locks and unlocks the syncache_head autonomously. 325 */ 326static void 327syncache_insert(struct syncache *sc, struct syncache_head *sch) 328{ 329 struct syncache *sc2; 330 331 SCH_LOCK(sch); 332 333 /* 334 * Make sure that we don't overflow the per-bucket limit. 335 * If the bucket is full, toss the oldest element. 336 */ 337 if (sch->sch_length >= tcp_syncache.bucket_limit) { 338 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 339 ("sch->sch_length incorrect")); 340 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 341 syncache_drop(sc2, sch); 342 tcpstat.tcps_sc_bucketoverflow++; 343 } 344 345 /* Put it into the bucket. */ 346 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 347 sch->sch_length++; 348 349 /* Reinitialize the bucket row's timer. */ 350 SYNCACHE_TIMEOUT(sc, sch, 1); 351 352 SCH_UNLOCK(sch); 353 354 tcp_syncache.cache_count++; 355 tcpstat.tcps_sc_added++; 356} 357 358/* 359 * Remove and free entry from syncache bucket row. 360 * Expects locked syncache head. 361 */ 362static void 363syncache_drop(struct syncache *sc, struct syncache_head *sch) 364{ 365 366 SCH_LOCK_ASSERT(sch); 367 368 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 369 sch->sch_length--; 370 371 syncache_free(sc); 372 tcp_syncache.cache_count--; 373} 374 375/* 376 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 377 * If we have retransmitted an entry the maximum number of times, expire it. 378 * One separate timer for each bucket row. 379 */ 380static void 381syncache_timer(void *xsch) 382{ 383 struct syncache_head *sch = (struct syncache_head *)xsch; 384 struct syncache *sc, *nsc; 385 int tick = ticks; 386 char *s; 387 388 /* NB: syncache_head has already been locked by the callout. */ 389 SCH_LOCK_ASSERT(sch); 390 391 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 392 /* 393 * We do not check if the listen socket still exists 394 * and accept the case where the listen socket may be 395 * gone by the time we resend the SYN/ACK. We do 396 * not expect this to happens often. If it does, 397 * then the RST will be sent by the time the remote 398 * host does the SYN/ACK->ACK. 399 */ 400 if (sc->sc_rxttime >= tick) { 401 if (sc->sc_rxttime < sch->sch_nextc) 402 sch->sch_nextc = sc->sc_rxttime; 403 continue; 404 } 405 406 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) { 407 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 408 log(LOG_DEBUG, "%s; %s: Response timeout\n", 409 s, __func__); 410 free(s, M_TCPLOG); 411 } 412 syncache_drop(sc, sch); 413 tcpstat.tcps_sc_stale++; 414 continue; 415 } 416 417 (void) syncache_respond(sc); 418 tcpstat.tcps_sc_retransmitted++; 419 SYNCACHE_TIMEOUT(sc, sch, 0); 420 } 421 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 422 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 423 syncache_timer, (void *)(sch)); 424} 425 426/* 427 * Find an entry in the syncache. 428 * Returns always with locked syncache_head plus a matching entry or NULL. 429 */ 430struct syncache * 431syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 432{ 433 struct syncache *sc; 434 struct syncache_head *sch; 435 436#ifdef INET6 437 if (inc->inc_isipv6) { 438 sch = &tcp_syncache.hashbase[ 439 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 440 *schp = sch; 441 442 SCH_LOCK(sch); 443 444 /* Circle through bucket row to find matching entry. */ 445 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 446 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 447 return (sc); 448 } 449 } else 450#endif 451 { 452 sch = &tcp_syncache.hashbase[ 453 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 454 *schp = sch; 455 456 SCH_LOCK(sch); 457 458 /* Circle through bucket row to find matching entry. */ 459 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 460#ifdef INET6 461 if (sc->sc_inc.inc_isipv6) 462 continue; 463#endif 464 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 465 return (sc); 466 } 467 } 468 SCH_LOCK_ASSERT(*schp); 469 return (NULL); /* always returns with locked sch */ 470} 471 472/* 473 * This function is called when we get a RST for a 474 * non-existent connection, so that we can see if the 475 * connection is in the syn cache. If it is, zap it. 476 */ 477void 478syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 479{ 480 struct syncache *sc; 481 struct syncache_head *sch; 482 483 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 484 SCH_LOCK_ASSERT(sch); 485 if (sc == NULL) 486 goto done; 487 488 /* 489 * If the RST bit is set, check the sequence number to see 490 * if this is a valid reset segment. 491 * RFC 793 page 37: 492 * In all states except SYN-SENT, all reset (RST) segments 493 * are validated by checking their SEQ-fields. A reset is 494 * valid if its sequence number is in the window. 495 * 496 * The sequence number in the reset segment is normally an 497 * echo of our outgoing acknowlegement numbers, but some hosts 498 * send a reset with the sequence number at the rightmost edge 499 * of our receive window, and we have to handle this case. 500 */ 501 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 502 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 503 syncache_drop(sc, sch); 504 tcpstat.tcps_sc_reset++; 505 } 506done: 507 SCH_UNLOCK(sch); 508} 509 510void 511syncache_badack(struct in_conninfo *inc) 512{ 513 struct syncache *sc; 514 struct syncache_head *sch; 515 516 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 517 SCH_LOCK_ASSERT(sch); 518 if (sc != NULL) { 519 syncache_drop(sc, sch); 520 tcpstat.tcps_sc_badack++; 521 } 522 SCH_UNLOCK(sch); 523} 524 525void 526syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 527{ 528 struct syncache *sc; 529 struct syncache_head *sch; 530 531 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 532 SCH_LOCK_ASSERT(sch); 533 if (sc == NULL) 534 goto done; 535 536 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 537 if (ntohl(th->th_seq) != sc->sc_iss) 538 goto done; 539 540 /* 541 * If we've rertransmitted 3 times and this is our second error, 542 * we remove the entry. Otherwise, we allow it to continue on. 543 * This prevents us from incorrectly nuking an entry during a 544 * spurious network outage. 545 * 546 * See tcp_notify(). 547 */ 548 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 549 sc->sc_flags |= SCF_UNREACH; 550 goto done; 551 } 552 syncache_drop(sc, sch); 553 tcpstat.tcps_sc_unreach++; 554done: 555 SCH_UNLOCK(sch); 556} 557 558/* 559 * Build a new TCP socket structure from a syncache entry. 560 */ 561static struct socket * 562syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 563{ 564 struct inpcb *inp = NULL; 565 struct socket *so; 566 struct tcpcb *tp; 567 char *s; 568 569 NET_ASSERT_GIANT(); 570 INP_INFO_WLOCK_ASSERT(&tcbinfo); 571 572 /* 573 * Ok, create the full blown connection, and set things up 574 * as they would have been set up if we had created the 575 * connection when the SYN arrived. If we can't create 576 * the connection, abort it. 577 */ 578 so = sonewconn(lso, SS_ISCONNECTED); 579 if (so == NULL) { 580 /* 581 * Drop the connection; we will either send a RST or 582 * have the peer retransmit its SYN again after its 583 * RTO and try again. 584 */ 585 tcpstat.tcps_listendrop++; 586 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 587 log(LOG_DEBUG, "%s; %s: Socket create failed " 588 "due to limits or memory shortage\n", 589 s, __func__); 590 free(s, M_TCPLOG); 591 } 592 goto abort2; 593 } 594#ifdef MAC 595 SOCK_LOCK(so); 596 mac_set_socket_peer_from_mbuf(m, so); 597 SOCK_UNLOCK(so); 598#endif 599 600 inp = sotoinpcb(so); 601 INP_LOCK(inp); 602 603 /* Insert new socket into PCB hash list. */ 604 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 605#ifdef INET6 606 if (sc->sc_inc.inc_isipv6) { 607 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 608 } else { 609 inp->inp_vflag &= ~INP_IPV6; 610 inp->inp_vflag |= INP_IPV4; 611#endif 612 inp->inp_laddr = sc->sc_inc.inc_laddr; 613#ifdef INET6 614 } 615#endif 616 inp->inp_lport = sc->sc_inc.inc_lport; 617 if (in_pcbinshash(inp) != 0) { 618 /* 619 * Undo the assignments above if we failed to 620 * put the PCB on the hash lists. 621 */ 622#ifdef INET6 623 if (sc->sc_inc.inc_isipv6) 624 inp->in6p_laddr = in6addr_any; 625 else 626#endif 627 inp->inp_laddr.s_addr = INADDR_ANY; 628 inp->inp_lport = 0; 629 goto abort; 630 } 631#ifdef IPSEC 632 /* Copy old policy into new socket's. */ 633 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 634 printf("syncache_socket: could not copy policy\n"); 635#endif 636#ifdef FAST_IPSEC 637 /* Copy old policy into new socket's. */ 638 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 639 printf("syncache_socket: could not copy policy\n"); 640#endif 641#ifdef INET6 642 if (sc->sc_inc.inc_isipv6) { 643 struct inpcb *oinp = sotoinpcb(lso); 644 struct in6_addr laddr6; 645 struct sockaddr_in6 sin6; 646 /* 647 * Inherit socket options from the listening socket. 648 * Note that in6p_inputopts are not (and should not be) 649 * copied, since it stores previously received options and is 650 * used to detect if each new option is different than the 651 * previous one and hence should be passed to a user. 652 * If we copied in6p_inputopts, a user would not be able to 653 * receive options just after calling the accept system call. 654 */ 655 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 656 if (oinp->in6p_outputopts) 657 inp->in6p_outputopts = 658 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 659 660 sin6.sin6_family = AF_INET6; 661 sin6.sin6_len = sizeof(sin6); 662 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 663 sin6.sin6_port = sc->sc_inc.inc_fport; 664 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 665 laddr6 = inp->in6p_laddr; 666 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 667 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 668 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 669 thread0.td_ucred)) { 670 inp->in6p_laddr = laddr6; 671 goto abort; 672 } 673 /* Override flowlabel from in6_pcbconnect. */ 674 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 675 inp->in6p_flowinfo |= sc->sc_flowlabel; 676 } else 677#endif 678 { 679 struct in_addr laddr; 680 struct sockaddr_in sin; 681 682 inp->inp_options = ip_srcroute(m); 683 if (inp->inp_options == NULL) { 684 inp->inp_options = sc->sc_ipopts; 685 sc->sc_ipopts = NULL; 686 } 687 688 sin.sin_family = AF_INET; 689 sin.sin_len = sizeof(sin); 690 sin.sin_addr = sc->sc_inc.inc_faddr; 691 sin.sin_port = sc->sc_inc.inc_fport; 692 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 693 laddr = inp->inp_laddr; 694 if (inp->inp_laddr.s_addr == INADDR_ANY) 695 inp->inp_laddr = sc->sc_inc.inc_laddr; 696 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 697 thread0.td_ucred)) { 698 inp->inp_laddr = laddr; 699 goto abort; 700 } 701 } 702 tp = intotcpcb(inp); 703 tp->t_state = TCPS_SYN_RECEIVED; 704 tp->iss = sc->sc_iss; 705 tp->irs = sc->sc_irs; 706 tcp_rcvseqinit(tp); 707 tcp_sendseqinit(tp); 708 tp->snd_wl1 = sc->sc_irs; 709 tp->snd_max = tp->iss + 1; 710 tp->snd_nxt = tp->iss + 1; 711 tp->rcv_up = sc->sc_irs + 1; 712 tp->rcv_wnd = sc->sc_wnd; 713 tp->rcv_adv += tp->rcv_wnd; 714 tp->last_ack_sent = tp->rcv_nxt; 715 716 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 717 if (sc->sc_flags & SCF_NOOPT) 718 tp->t_flags |= TF_NOOPT; 719 else { 720 if (sc->sc_flags & SCF_WINSCALE) { 721 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 722 tp->snd_scale = sc->sc_requested_s_scale; 723 tp->request_r_scale = sc->sc_requested_r_scale; 724 } 725 if (sc->sc_flags & SCF_TIMESTAMP) { 726 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 727 tp->ts_recent = sc->sc_tsreflect; 728 tp->ts_recent_age = ticks; 729 tp->ts_offset = sc->sc_tsoff; 730 } 731#ifdef TCP_SIGNATURE 732 if (sc->sc_flags & SCF_SIGNATURE) 733 tp->t_flags |= TF_SIGNATURE; 734#endif 735 if (sc->sc_flags & SCF_SACK) 736 tp->t_flags |= TF_SACK_PERMIT; 737 } 738 739 /* 740 * Set up MSS and get cached values from tcp_hostcache. 741 * This might overwrite some of the defaults we just set. 742 */ 743 tcp_mss(tp, sc->sc_peer_mss); 744 745 /* 746 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 747 */ 748 if (sc->sc_rxmits > 1) 749 tp->snd_cwnd = tp->t_maxseg; 750 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 751 752 INP_UNLOCK(inp); 753 754 tcpstat.tcps_accepts++; 755 return (so); 756 757abort: 758 INP_UNLOCK(inp); 759abort2: 760 if (so != NULL) 761 soabort(so); 762 return (NULL); 763} 764 765/* 766 * This function gets called when we receive an ACK for a 767 * socket in the LISTEN state. We look up the connection 768 * in the syncache, and if its there, we pull it out of 769 * the cache and turn it into a full-blown connection in 770 * the SYN-RECEIVED state. 771 */ 772int 773syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 774 struct socket **lsop, struct mbuf *m) 775{ 776 struct syncache *sc; 777 struct syncache_head *sch; 778 struct syncache scs; 779 char *s; 780 781 /* 782 * Global TCP locks are held because we manipulate the PCB lists 783 * and create a new socket. 784 */ 785 INP_INFO_WLOCK_ASSERT(&tcbinfo); 786 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, 787 ("%s: can handle only ACK", __func__)); 788 789 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 790 SCH_LOCK_ASSERT(sch); 791 if (sc == NULL) { 792 /* 793 * There is no syncache entry, so see if this ACK is 794 * a returning syncookie. To do this, first: 795 * A. See if this socket has had a syncache entry dropped in 796 * the past. We don't want to accept a bogus syncookie 797 * if we've never received a SYN. 798 * B. check that the syncookie is valid. If it is, then 799 * cobble up a fake syncache entry, and return. 800 */ 801 if (!tcp_syncookies) { 802 SCH_UNLOCK(sch); 803 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 804 log(LOG_DEBUG, "%s; %s: Spurious ACK, " 805 "segment rejected (syncookies disabled)\n", 806 s, __func__); 807 goto failed; 808 } 809 bzero(&scs, sizeof(scs)); 810 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 811 SCH_UNLOCK(sch); 812 if (sc == NULL) { 813 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 814 log(LOG_DEBUG, "%s; %s: Segment failed " 815 "SYNCOOKIE authentication, segment rejected " 816 "(probably spoofed)\n", s, __func__); 817 goto failed; 818 } 819 tcpstat.tcps_sc_recvcookie++; 820 } else { 821 /* Pull out the entry to unlock the bucket row. */ 822 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 823 sch->sch_length--; 824 tcp_syncache.cache_count--; 825 SCH_UNLOCK(sch); 826 } 827 828 /* 829 * Segment validation: 830 * ACK must match our initial sequence number + 1 (the SYN|ACK). 831 */ 832 if (th->th_ack != sc->sc_iss + 1) { 833 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 834 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " 835 "rejected\n", s, __func__, th->th_ack, sc->sc_iss); 836 goto failed; 837 } 838 /* 839 * The SEQ must match the received initial receive sequence 840 * number + 1 (the SYN) because we didn't ACK any data that 841 * may have come with the SYN. 842 */ 843 if (th->th_seq != sc->sc_irs + 1) { 844 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 845 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " 846 "rejected\n", s, __func__, th->th_seq, sc->sc_irs); 847 goto failed; 848 } 849 /* 850 * If timestamps were present in the SYN and we accepted 851 * them in our SYN|ACK we require them to be present from 852 * now on. And vice versa. 853 */ 854 if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) { 855 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 856 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 857 "segment rejected\n", s, __func__); 858 goto failed; 859 } 860 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { 861 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 862 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 863 "segment rejected\n", s, __func__); 864 goto failed; 865 } 866 /* 867 * If timestamps were negotiated the reflected timestamp 868 * must be equal to what we actually sent in the SYN|ACK. 869 */ 870 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) { 871 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 872 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, " 873 "segment rejected\n", 874 s, __func__, to->to_tsecr, sc->sc_ts); 875 goto failed; 876 } 877 878 *lsop = syncache_socket(sc, *lsop, m); 879 880 if (*lsop == NULL) 881 tcpstat.tcps_sc_aborted++; 882 else 883 tcpstat.tcps_sc_completed++; 884 885 if (sc != &scs) 886 syncache_free(sc); 887 return (1); 888failed: 889 if (sc != NULL && sc != &scs) 890 syncache_free(sc); 891 if (s != NULL) 892 free(s, M_TCPLOG); 893 *lsop = NULL; 894 return (0); 895} 896 897/* 898 * Given a LISTEN socket and an inbound SYN request, add 899 * this to the syn cache, and send back a segment: 900 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 901 * to the source. 902 * 903 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 904 * Doing so would require that we hold onto the data and deliver it 905 * to the application. However, if we are the target of a SYN-flood 906 * DoS attack, an attacker could send data which would eventually 907 * consume all available buffer space if it were ACKed. By not ACKing 908 * the data, we avoid this DoS scenario. 909 */ 910void 911syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 912 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 913{ 914 struct tcpcb *tp; 915 struct socket *so; 916 struct syncache *sc = NULL; 917 struct syncache_head *sch; 918 struct mbuf *ipopts = NULL; 919 u_int32_t flowtmp; 920 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 921#ifdef INET6 922 int autoflowlabel = 0; 923#endif 924#ifdef MAC 925 struct label *maclabel; 926#endif 927 struct syncache scs; 928 929 INP_INFO_WLOCK_ASSERT(&tcbinfo); 930 INP_LOCK_ASSERT(inp); /* listen socket */ 931 932 /* 933 * Combine all so/tp operations very early to drop the INP lock as 934 * soon as possible. 935 */ 936 so = *lsop; 937 tp = sototcpcb(so); 938 939#ifdef INET6 940 if (inc->inc_isipv6 && 941 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 942 autoflowlabel = 1; 943#endif 944 ip_ttl = inp->inp_ip_ttl; 945 ip_tos = inp->inp_ip_tos; 946 win = sbspace(&so->so_rcv); 947 sb_hiwat = so->so_rcv.sb_hiwat; 948 noopt = (tp->t_flags & TF_NOOPT); 949 950 so = NULL; 951 tp = NULL; 952 953#ifdef MAC 954 if (mac_init_syncache(&maclabel) != 0) { 955 INP_UNLOCK(inp); 956 INP_INFO_WUNLOCK(&tcbinfo); 957 goto done; 958 } else 959 mac_init_syncache_from_inpcb(maclabel, inp); 960#endif 961 INP_UNLOCK(inp); 962 INP_INFO_WUNLOCK(&tcbinfo); 963 964 /* 965 * Remember the IP options, if any. 966 */ 967#ifdef INET6 968 if (!inc->inc_isipv6) 969#endif 970 ipopts = ip_srcroute(m); 971 972 /* 973 * See if we already have an entry for this connection. 974 * If we do, resend the SYN,ACK, and reset the retransmit timer. 975 * 976 * XXX: should the syncache be re-initialized with the contents 977 * of the new SYN here (which may have different options?) 978 */ 979 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 980 SCH_LOCK_ASSERT(sch); 981 if (sc != NULL) { 982 tcpstat.tcps_sc_dupsyn++; 983 if (ipopts) { 984 /* 985 * If we were remembering a previous source route, 986 * forget it and use the new one we've been given. 987 */ 988 if (sc->sc_ipopts) 989 (void) m_free(sc->sc_ipopts); 990 sc->sc_ipopts = ipopts; 991 } 992 /* 993 * Update timestamp if present. 994 */ 995 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 996 sc->sc_tsreflect = to->to_tsval; 997 else 998 sc->sc_flags &= ~SCF_TIMESTAMP; 999#ifdef MAC 1000 /* 1001 * Since we have already unconditionally allocated label 1002 * storage, free it up. The syncache entry will already 1003 * have an initialized label we can use. 1004 */ 1005 mac_destroy_syncache(&maclabel); 1006 KASSERT(sc->sc_label != NULL, 1007 ("%s: label not initialized", __func__)); 1008#endif 1009 if (syncache_respond(sc) == 0) { 1010 SYNCACHE_TIMEOUT(sc, sch, 1); 1011 tcpstat.tcps_sndacks++; 1012 tcpstat.tcps_sndtotal++; 1013 } 1014 SCH_UNLOCK(sch); 1015 goto done; 1016 } 1017 1018 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1019 if (sc == NULL) { 1020 /* 1021 * The zone allocator couldn't provide more entries. 1022 * Treat this as if the cache was full; drop the oldest 1023 * entry and insert the new one. 1024 */ 1025 tcpstat.tcps_sc_zonefail++; 1026 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 1027 syncache_drop(sc, sch); 1028 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1029 if (sc == NULL) { 1030 if (tcp_syncookies) { 1031 bzero(&scs, sizeof(scs)); 1032 sc = &scs; 1033 } else { 1034 SCH_UNLOCK(sch); 1035 if (ipopts) 1036 (void) m_free(ipopts); 1037 goto done; 1038 } 1039 } 1040 } 1041 1042 /* 1043 * Fill in the syncache values. 1044 */ 1045#ifdef MAC 1046 sc->sc_label = maclabel; 1047#endif 1048 sc->sc_ipopts = ipopts; 1049 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1050#ifdef INET6 1051 if (!inc->inc_isipv6) 1052#endif 1053 { 1054 sc->sc_ip_tos = ip_tos; 1055 sc->sc_ip_ttl = ip_ttl; 1056 } 1057 1058 sc->sc_irs = th->th_seq; 1059 sc->sc_iss = arc4random(); 1060 sc->sc_flags = 0; 1061 sc->sc_flowlabel = 0; 1062 1063 /* 1064 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1065 * win was derived from socket earlier in the function. 1066 */ 1067 win = imax(win, 0); 1068 win = imin(win, TCP_MAXWIN); 1069 sc->sc_wnd = win; 1070 1071 if (tcp_do_rfc1323) { 1072 /* 1073 * A timestamp received in a SYN makes 1074 * it ok to send timestamp requests and replies. 1075 */ 1076 if (to->to_flags & TOF_TS) { 1077 sc->sc_tsreflect = to->to_tsval; 1078 sc->sc_ts = ticks; 1079 sc->sc_flags |= SCF_TIMESTAMP; 1080 } 1081 if (to->to_flags & TOF_SCALE) { 1082 int wscale = 0; 1083 1084 /* 1085 * Compute proper scaling value from buffer space. 1086 * Leave enough room for the socket buffer to grow 1087 * with auto sizing. This allows us to scale the 1088 * receive buffer over a wide range while not losing 1089 * any efficiency or fine granularity. 1090 * 1091 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1092 * or <SYN,ACK>) segment itself is never scaled. 1093 */ 1094 while (wscale < TCP_MAX_WINSHIFT && 1095 (0x1 << wscale) < tcp_minmss) 1096 wscale++; 1097 sc->sc_requested_r_scale = wscale; 1098 sc->sc_requested_s_scale = to->to_wscale; 1099 sc->sc_flags |= SCF_WINSCALE; 1100 } 1101 } 1102#ifdef TCP_SIGNATURE 1103 /* 1104 * If listening socket requested TCP digests, and received SYN 1105 * contains the option, flag this in the syncache so that 1106 * syncache_respond() will do the right thing with the SYN+ACK. 1107 * XXX: Currently we always record the option by default and will 1108 * attempt to use it in syncache_respond(). 1109 */ 1110 if (to->to_flags & TOF_SIGNATURE) 1111 sc->sc_flags |= SCF_SIGNATURE; 1112#endif 1113 if (to->to_flags & TOF_SACK) 1114 sc->sc_flags |= SCF_SACK; 1115 if (to->to_flags & TOF_MSS) 1116 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1117 if (noopt) 1118 sc->sc_flags |= SCF_NOOPT; 1119 1120 if (tcp_syncookies) { 1121 syncookie_generate(sch, sc, &flowtmp); 1122#ifdef INET6 1123 if (autoflowlabel) 1124 sc->sc_flowlabel = flowtmp; 1125#endif 1126 } else { 1127#ifdef INET6 1128 if (autoflowlabel) 1129 sc->sc_flowlabel = 1130 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1131#endif 1132 } 1133 SCH_UNLOCK(sch); 1134 1135 /* 1136 * Do a standard 3-way handshake. 1137 */ 1138 if (syncache_respond(sc) == 0) { 1139 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs) 1140 syncache_free(sc); 1141 else if (sc != &scs) 1142 syncache_insert(sc, sch); /* locks and unlocks sch */ 1143 tcpstat.tcps_sndacks++; 1144 tcpstat.tcps_sndtotal++; 1145 } else { 1146 if (sc != &scs) 1147 syncache_free(sc); 1148 tcpstat.tcps_sc_dropped++; 1149 } 1150 1151done: 1152#ifdef MAC 1153 if (sc == &scs) 1154 mac_destroy_syncache(&maclabel); 1155#endif 1156 *lsop = NULL; 1157 m_freem(m); 1158 return; 1159} 1160 1161static int 1162syncache_respond(struct syncache *sc) 1163{ 1164 struct ip *ip = NULL; 1165 struct mbuf *m; 1166 struct tcphdr *th; 1167 int optlen, error; 1168 u_int16_t hlen, tlen, mssopt; 1169 struct tcpopt to; 1170#ifdef INET6 1171 struct ip6_hdr *ip6 = NULL; 1172#endif 1173 1174 hlen = 1175#ifdef INET6 1176 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1177#endif 1178 sizeof(struct ip); 1179 tlen = hlen + sizeof(struct tcphdr); 1180 1181 /* Determine MSS we advertize to other end of connection. */ 1182 mssopt = tcp_mssopt(&sc->sc_inc); 1183 if (sc->sc_peer_mss) 1184 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss); 1185 1186 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1187 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, 1188 ("syncache: mbuf too small")); 1189 1190 /* Create the IP+TCP header from scratch. */ 1191 m = m_gethdr(M_DONTWAIT, MT_DATA); 1192 if (m == NULL) 1193 return (ENOBUFS); 1194#ifdef MAC 1195 mac_create_mbuf_from_syncache(sc->sc_label, m); 1196#endif 1197 m->m_data += max_linkhdr; 1198 m->m_len = tlen; 1199 m->m_pkthdr.len = tlen; 1200 m->m_pkthdr.rcvif = NULL; 1201 1202#ifdef INET6 1203 if (sc->sc_inc.inc_isipv6) { 1204 ip6 = mtod(m, struct ip6_hdr *); 1205 ip6->ip6_vfc = IPV6_VERSION; 1206 ip6->ip6_nxt = IPPROTO_TCP; 1207 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1208 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1209 ip6->ip6_plen = htons(tlen - hlen); 1210 /* ip6_hlim is set after checksum */ 1211 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1212 ip6->ip6_flow |= sc->sc_flowlabel; 1213 1214 th = (struct tcphdr *)(ip6 + 1); 1215 } else 1216#endif 1217 { 1218 ip = mtod(m, struct ip *); 1219 ip->ip_v = IPVERSION; 1220 ip->ip_hl = sizeof(struct ip) >> 2; 1221 ip->ip_len = tlen; 1222 ip->ip_id = 0; 1223 ip->ip_off = 0; 1224 ip->ip_sum = 0; 1225 ip->ip_p = IPPROTO_TCP; 1226 ip->ip_src = sc->sc_inc.inc_laddr; 1227 ip->ip_dst = sc->sc_inc.inc_faddr; 1228 ip->ip_ttl = sc->sc_ip_ttl; 1229 ip->ip_tos = sc->sc_ip_tos; 1230 1231 /* 1232 * See if we should do MTU discovery. Route lookups are 1233 * expensive, so we will only unset the DF bit if: 1234 * 1235 * 1) path_mtu_discovery is disabled 1236 * 2) the SCF_UNREACH flag has been set 1237 */ 1238 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1239 ip->ip_off |= IP_DF; 1240 1241 th = (struct tcphdr *)(ip + 1); 1242 } 1243 th->th_sport = sc->sc_inc.inc_lport; 1244 th->th_dport = sc->sc_inc.inc_fport; 1245 1246 th->th_seq = htonl(sc->sc_iss); 1247 th->th_ack = htonl(sc->sc_irs + 1); 1248 th->th_off = sizeof(struct tcphdr) >> 2; 1249 th->th_x2 = 0; 1250 th->th_flags = TH_SYN|TH_ACK; 1251 th->th_win = htons(sc->sc_wnd); 1252 th->th_urp = 0; 1253 1254 /* Tack on the TCP options. */ 1255 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1256 to.to_flags = 0; 1257 1258 to.to_mss = mssopt; 1259 to.to_flags = TOF_MSS; 1260 if (sc->sc_flags & SCF_WINSCALE) { 1261 to.to_wscale = sc->sc_requested_r_scale; 1262 to.to_flags |= TOF_SCALE; 1263 } 1264 if (sc->sc_flags & SCF_TIMESTAMP) { 1265 /* Virgin timestamp or TCP cookie enhanced one. */ 1266 to.to_tsval = sc->sc_ts; 1267 to.to_tsecr = sc->sc_tsreflect; 1268 to.to_flags |= TOF_TS; 1269 } 1270 if (sc->sc_flags & SCF_SACK) 1271 to.to_flags |= TOF_SACKPERM; 1272#ifdef TCP_SIGNATURE 1273 if (sc->sc_flags & SCF_SIGNATURE) 1274 to.to_flags |= TOF_SIGNATURE; 1275#endif 1276 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1277 1278#ifdef TCP_SIGNATURE 1279 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1280 to.to_signature, IPSEC_DIR_OUTBOUND); 1281#endif 1282 1283 /* Adjust headers by option size. */ 1284 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1285 m->m_len += optlen; 1286 m->m_pkthdr.len += optlen; 1287#ifdef INET6 1288 if (sc->sc_inc.inc_isipv6) 1289 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1290 else 1291#endif 1292 ip->ip_len += optlen; 1293 } else 1294 optlen = 0; 1295 1296#ifdef INET6 1297 if (sc->sc_inc.inc_isipv6) { 1298 th->th_sum = 0; 1299 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1300 tlen + optlen - hlen); 1301 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1302 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1303 } else 1304#endif 1305 { 1306 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1307 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1308 m->m_pkthdr.csum_flags = CSUM_TCP; 1309 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1310 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1311 } 1312 return (error); 1313} 1314 1315/* 1316 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1317 * receive and to be able to handle SYN floods from bogus source addresses 1318 * (where we will never receive any reply). SYN floods try to exhaust all 1319 * our memory and available slots in the SYN cache table to cause a denial 1320 * of service to legitimate users of the local host. 1321 * 1322 * The idea of SYN cookies is to encode and include all necessary information 1323 * about the connection setup state within the SYN-ACK we send back and thus 1324 * to get along without keeping any local state until the ACK to the SYN-ACK 1325 * arrives (if ever). Everything we need to know should be available from 1326 * the information we encoded in the SYN-ACK. 1327 * 1328 * More information about the theory behind SYN cookies and its first 1329 * discussion and specification can be found at: 1330 * http://cr.yp.to/syncookies.html (overview) 1331 * http://cr.yp.to/syncookies/archive (gory details) 1332 * 1333 * This implementation extends the orginal idea and first implementation 1334 * of FreeBSD by using not only the initial sequence number field to store 1335 * information but also the timestamp field if present. This way we can 1336 * keep track of the entire state we need to know to recreate the session in 1337 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1338 * these days. For those that do not we still have to live with the known 1339 * shortcomings of the ISN only SYN cookies. 1340 * 1341 * Cookie layers: 1342 * 1343 * Initial sequence number we send: 1344 * 31|................................|0 1345 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1346 * D = MD5 Digest (first dword) 1347 * M = MSS index 1348 * R = Rotation of secret 1349 * P = Odd or Even secret 1350 * 1351 * The MD5 Digest is computed with over following parameters: 1352 * a) randomly rotated secret 1353 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1354 * c) the received initial sequence number from remote host 1355 * d) the rotation offset and odd/even bit 1356 * 1357 * Timestamp we send: 1358 * 31|................................|0 1359 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1360 * D = MD5 Digest (third dword) (only as filler) 1361 * S = Requested send window scale 1362 * R = Requested receive window scale 1363 * A = SACK allowed 1364 * 5 = TCP-MD5 enabled (not implemented yet) 1365 * XORed with MD5 Digest (forth dword) 1366 * 1367 * The timestamp isn't cryptographically secure and doesn't need to be. 1368 * The double use of the MD5 digest dwords ties it to a specific remote/ 1369 * local host/port, remote initial sequence number and our local time 1370 * limited secret. A received timestamp is reverted (XORed) and then 1371 * the contained MD5 dword is compared to the computed one to ensure the 1372 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1373 * have been tampered with but this isn't different from supplying bogus 1374 * values in the SYN in the first place. 1375 * 1376 * Some problems with SYN cookies remain however: 1377 * Consider the problem of a recreated (and retransmitted) cookie. If the 1378 * original SYN was accepted, the connection is established. The second 1379 * SYN is inflight, and if it arrives with an ISN that falls within the 1380 * receive window, the connection is killed. 1381 * 1382 * Notes: 1383 * A heuristic to determine when to accept syn cookies is not necessary. 1384 * An ACK flood would cause the syncookie verification to be attempted, 1385 * but a SYN flood causes syncookies to be generated. Both are of equal 1386 * cost, so there's no point in trying to optimize the ACK flood case. 1387 * Also, if you don't process certain ACKs for some reason, then all someone 1388 * would have to do is launch a SYN and ACK flood at the same time, which 1389 * would stop cookie verification and defeat the entire purpose of syncookies. 1390 */ 1391static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1392 1393static void 1394syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1395 u_int32_t *flowlabel) 1396{ 1397 MD5_CTX ctx; 1398 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1399 u_int32_t data; 1400 u_int32_t *secbits; 1401 u_int off, pmss, mss; 1402 int i; 1403 1404 SCH_LOCK_ASSERT(sch); 1405 1406 /* Which of the two secrets to use. */ 1407 secbits = sch->sch_oddeven ? 1408 sch->sch_secbits_odd : sch->sch_secbits_even; 1409 1410 /* Reseed secret if too old. */ 1411 if (sch->sch_reseed < time_uptime) { 1412 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1413 secbits = sch->sch_oddeven ? 1414 sch->sch_secbits_odd : sch->sch_secbits_even; 1415 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1416 secbits[i] = arc4random(); 1417 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1418 } 1419 1420 /* Secret rotation offset. */ 1421 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1422 1423 /* Maximum segment size calculation. */ 1424 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss); 1425 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1426 if (tcp_sc_msstab[mss] <= pmss) 1427 break; 1428 1429 /* Fold parameters and MD5 digest into the ISN we will send. */ 1430 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1431 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1432 data |= mss << 4; /* mss, 3 bits */ 1433 1434 MD5Init(&ctx); 1435 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1436 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1437 MD5Update(&ctx, secbits, off); 1438 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1439 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1440 MD5Update(&ctx, &data, sizeof(data)); 1441 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1442 1443 data |= (md5_buffer[0] << 7); 1444 sc->sc_iss = data; 1445 1446#ifdef INET6 1447 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1448#endif 1449 1450 /* Additional parameters are stored in the timestamp if present. */ 1451 if (sc->sc_flags & SCF_TIMESTAMP) { 1452 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1453 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1454 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1455 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1456 data |= md5_buffer[2] << 10; /* more digest bits */ 1457 data ^= md5_buffer[3]; 1458 sc->sc_ts = data; 1459 sc->sc_tsoff = data - ticks; /* after XOR */ 1460 } 1461 1462 return; 1463} 1464 1465static struct syncache * 1466syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1467 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1468 struct socket *so) 1469{ 1470 MD5_CTX ctx; 1471 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1472 u_int32_t data = 0; 1473 u_int32_t *secbits; 1474 tcp_seq ack, seq; 1475 int off, mss, wnd, flags; 1476 1477 SCH_LOCK_ASSERT(sch); 1478 1479 /* 1480 * Pull information out of SYN-ACK/ACK and 1481 * revert sequence number advances. 1482 */ 1483 ack = th->th_ack - 1; 1484 seq = th->th_seq - 1; 1485 off = (ack >> 1) & 0x7; 1486 mss = (ack >> 4) & 0x7; 1487 flags = ack & 0x7f; 1488 1489 /* Which of the two secrets to use. */ 1490 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1491 1492 /* 1493 * The secret wasn't updated for the lifetime of a syncookie, 1494 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1495 */ 1496 if (sch->sch_reseed < time_uptime) { 1497 return (NULL); 1498 } 1499 1500 /* Recompute the digest so we can compare it. */ 1501 MD5Init(&ctx); 1502 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1503 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1504 MD5Update(&ctx, secbits, off); 1505 MD5Update(&ctx, inc, sizeof(*inc)); 1506 MD5Update(&ctx, &seq, sizeof(seq)); 1507 MD5Update(&ctx, &flags, sizeof(flags)); 1508 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1509 1510 /* Does the digest part of or ACK'ed ISS match? */ 1511 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1512 return (NULL); 1513 1514 /* Does the digest part of our reflected timestamp match? */ 1515 if (to->to_flags & TOF_TS) { 1516 data = md5_buffer[3] ^ to->to_tsecr; 1517 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1518 return (NULL); 1519 } 1520 1521 /* Fill in the syncache values. */ 1522 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1523 sc->sc_ipopts = NULL; 1524 1525 sc->sc_irs = seq; 1526 sc->sc_iss = ack; 1527 1528#ifdef INET6 1529 if (inc->inc_isipv6) { 1530 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1531 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1532 } else 1533#endif 1534 { 1535 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1536 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1537 } 1538 1539 /* Additional parameters that were encoded in the timestamp. */ 1540 if (data) { 1541 sc->sc_flags |= SCF_TIMESTAMP; 1542 sc->sc_tsreflect = to->to_tsval; 1543 sc->sc_ts = to->to_tsecr; 1544 sc->sc_tsoff = to->to_tsecr - ticks; 1545 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1546 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1547 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1548 TCP_MAX_WINSHIFT); 1549 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1550 TCP_MAX_WINSHIFT); 1551 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1552 sc->sc_flags |= SCF_WINSCALE; 1553 } else 1554 sc->sc_flags |= SCF_NOOPT; 1555 1556 wnd = sbspace(&so->so_rcv); 1557 wnd = imax(wnd, 0); 1558 wnd = imin(wnd, TCP_MAXWIN); 1559 sc->sc_wnd = wnd; 1560 1561 sc->sc_rxmits = 0; 1562 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1563 1564 return (sc); 1565} 1566