tcp_syncache.c revision 172970
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 33#include <sys/cdefs.h> 34__FBSDID("$FreeBSD: head/sys/netinet/tcp_syncache.c 172970 2007-10-25 14:37:37Z rwatson $"); 35 36#include "opt_inet.h" 37#include "opt_inet6.h" 38#include "opt_ipsec.h" 39#include "opt_mac.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/kernel.h> 44#include <sys/sysctl.h> 45#include <sys/lock.h> 46#include <sys/mutex.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/md5.h> 50#include <sys/proc.h> /* for proc0 declaration */ 51#include <sys/random.h> 52#include <sys/socket.h> 53#include <sys/socketvar.h> 54#include <sys/syslog.h> 55 56#include <vm/uma.h> 57 58#include <net/if.h> 59#include <net/route.h> 60 61#include <netinet/in.h> 62#include <netinet/in_systm.h> 63#include <netinet/ip.h> 64#include <netinet/in_var.h> 65#include <netinet/in_pcb.h> 66#include <netinet/ip_var.h> 67#include <netinet/ip_options.h> 68#ifdef INET6 69#include <netinet/ip6.h> 70#include <netinet/icmp6.h> 71#include <netinet6/nd6.h> 72#include <netinet6/ip6_var.h> 73#include <netinet6/in6_pcb.h> 74#endif 75#include <netinet/tcp.h> 76#include <netinet/tcp_fsm.h> 77#include <netinet/tcp_seq.h> 78#include <netinet/tcp_timer.h> 79#include <netinet/tcp_var.h> 80#include <netinet/tcp_syncache.h> 81#ifdef INET6 82#include <netinet6/tcp6_var.h> 83#endif 84 85#ifdef IPSEC 86#include <netipsec/ipsec.h> 87#ifdef INET6 88#include <netipsec/ipsec6.h> 89#endif 90#include <netipsec/key.h> 91#endif /*IPSEC*/ 92 93#include <machine/in_cksum.h> 94 95#include <security/mac/mac_framework.h> 96 97static int tcp_syncookies = 1; 98SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 99 &tcp_syncookies, 0, 100 "Use TCP SYN cookies if the syncache overflows"); 101 102static int tcp_syncookiesonly = 0; 103SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW, 104 &tcp_syncookiesonly, 0, 105 "Use only TCP SYN cookies"); 106 107#define SYNCOOKIE_SECRET_SIZE 8 /* dwords */ 108#define SYNCOOKIE_LIFETIME 16 /* seconds */ 109 110struct syncache { 111 TAILQ_ENTRY(syncache) sc_hash; 112 struct in_conninfo sc_inc; /* addresses */ 113 u_long sc_rxttime; /* retransmit time */ 114 u_int16_t sc_rxmits; /* retransmit counter */ 115 116 u_int32_t sc_tsreflect; /* timestamp to reflect */ 117 u_int32_t sc_ts; /* our timestamp to send */ 118 u_int32_t sc_tsoff; /* ts offset w/ syncookies */ 119 u_int32_t sc_flowlabel; /* IPv6 flowlabel */ 120 tcp_seq sc_irs; /* seq from peer */ 121 tcp_seq sc_iss; /* our ISS */ 122 struct mbuf *sc_ipopts; /* source route */ 123 124 u_int16_t sc_peer_mss; /* peer's MSS */ 125 u_int16_t sc_wnd; /* advertised window */ 126 u_int8_t sc_ip_ttl; /* IPv4 TTL */ 127 u_int8_t sc_ip_tos; /* IPv4 TOS */ 128 u_int8_t sc_requested_s_scale:4, 129 sc_requested_r_scale:4; 130 u_int8_t sc_flags; 131#define SCF_NOOPT 0x01 /* no TCP options */ 132#define SCF_WINSCALE 0x02 /* negotiated window scaling */ 133#define SCF_TIMESTAMP 0x04 /* negotiated timestamps */ 134 /* MSS is implicit */ 135#define SCF_UNREACH 0x10 /* icmp unreachable received */ 136#define SCF_SIGNATURE 0x20 /* send MD5 digests */ 137#define SCF_SACK 0x80 /* send SACK option */ 138#ifdef MAC 139 struct label *sc_label; /* MAC label reference */ 140#endif 141}; 142 143struct syncache_head { 144 struct mtx sch_mtx; 145 TAILQ_HEAD(sch_head, syncache) sch_bucket; 146 struct callout sch_timer; 147 int sch_nextc; 148 u_int sch_length; 149 u_int sch_oddeven; 150 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE]; 151 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE]; 152 u_int sch_reseed; /* time_uptime, seconds */ 153}; 154 155static void syncache_drop(struct syncache *, struct syncache_head *); 156static void syncache_free(struct syncache *); 157static void syncache_insert(struct syncache *, struct syncache_head *); 158struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 159static int syncache_respond(struct syncache *); 160static struct socket *syncache_socket(struct syncache *, struct socket *, 161 struct mbuf *m); 162static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, 163 int docallout); 164static void syncache_timer(void *); 165static void syncookie_generate(struct syncache_head *, struct syncache *, 166 u_int32_t *); 167static struct syncache 168 *syncookie_lookup(struct in_conninfo *, struct syncache_head *, 169 struct syncache *, struct tcpopt *, struct tcphdr *, 170 struct socket *); 171 172/* 173 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 174 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 175 * the odds are that the user has given up attempting to connect by then. 176 */ 177#define SYNCACHE_MAXREXMTS 3 178 179/* Arbitrary values */ 180#define TCP_SYNCACHE_HASHSIZE 512 181#define TCP_SYNCACHE_BUCKETLIMIT 30 182 183struct tcp_syncache { 184 struct syncache_head *hashbase; 185 uma_zone_t zone; 186 u_int hashsize; 187 u_int hashmask; 188 u_int bucket_limit; 189 u_int cache_count; /* XXX: unprotected */ 190 u_int cache_limit; 191 u_int rexmt_limit; 192 u_int hash_secret; 193}; 194static struct tcp_syncache tcp_syncache; 195 196SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 197 198SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 199 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 200 201SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 202 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 203 204SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 205 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 206 207SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 208 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 209 210SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 211 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 212 213int tcp_sc_rst_sock_fail = 1; 214SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW, 215 &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure"); 216 217static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 218 219#define SYNCACHE_HASH(inc, mask) \ 220 ((tcp_syncache.hash_secret ^ \ 221 (inc)->inc_faddr.s_addr ^ \ 222 ((inc)->inc_faddr.s_addr >> 16) ^ \ 223 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 224 225#define SYNCACHE_HASH6(inc, mask) \ 226 ((tcp_syncache.hash_secret ^ \ 227 (inc)->inc6_faddr.s6_addr32[0] ^ \ 228 (inc)->inc6_faddr.s6_addr32[3] ^ \ 229 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 230 231#define ENDPTS_EQ(a, b) ( \ 232 (a)->ie_fport == (b)->ie_fport && \ 233 (a)->ie_lport == (b)->ie_lport && \ 234 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 235 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 236) 237 238#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 239 240#define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 241#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 242#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 243 244/* 245 * Requires the syncache entry to be already removed from the bucket list. 246 */ 247static void 248syncache_free(struct syncache *sc) 249{ 250 if (sc->sc_ipopts) 251 (void) m_free(sc->sc_ipopts); 252#ifdef MAC 253 mac_syncache_destroy(&sc->sc_label); 254#endif 255 256 uma_zfree(tcp_syncache.zone, sc); 257} 258 259void 260syncache_init(void) 261{ 262 int i; 263 264 tcp_syncache.cache_count = 0; 265 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 266 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 267 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 268 tcp_syncache.hash_secret = arc4random(); 269 270 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 271 &tcp_syncache.hashsize); 272 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 273 &tcp_syncache.bucket_limit); 274 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 275 printf("WARNING: syncache hash size is not a power of 2.\n"); 276 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 277 } 278 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 279 280 /* Set limits. */ 281 tcp_syncache.cache_limit = 282 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 283 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 284 &tcp_syncache.cache_limit); 285 286 /* Allocate the hash table. */ 287 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 288 tcp_syncache.hashsize * sizeof(struct syncache_head), 289 M_SYNCACHE, M_WAITOK | M_ZERO); 290 291 /* Initialize the hash buckets. */ 292 for (i = 0; i < tcp_syncache.hashsize; i++) { 293 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 294 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 295 NULL, MTX_DEF); 296 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer, 297 &tcp_syncache.hashbase[i].sch_mtx, 0); 298 tcp_syncache.hashbase[i].sch_length = 0; 299 } 300 301 /* Create the syncache entry zone. */ 302 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 303 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 304 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 305} 306 307/* 308 * Inserts a syncache entry into the specified bucket row. 309 * Locks and unlocks the syncache_head autonomously. 310 */ 311static void 312syncache_insert(struct syncache *sc, struct syncache_head *sch) 313{ 314 struct syncache *sc2; 315 316 SCH_LOCK(sch); 317 318 /* 319 * Make sure that we don't overflow the per-bucket limit. 320 * If the bucket is full, toss the oldest element. 321 */ 322 if (sch->sch_length >= tcp_syncache.bucket_limit) { 323 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 324 ("sch->sch_length incorrect")); 325 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 326 syncache_drop(sc2, sch); 327 tcpstat.tcps_sc_bucketoverflow++; 328 } 329 330 /* Put it into the bucket. */ 331 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 332 sch->sch_length++; 333 334 /* Reinitialize the bucket row's timer. */ 335 syncache_timeout(sc, sch, 1); 336 337 SCH_UNLOCK(sch); 338 339 tcp_syncache.cache_count++; 340 tcpstat.tcps_sc_added++; 341} 342 343/* 344 * Remove and free entry from syncache bucket row. 345 * Expects locked syncache head. 346 */ 347static void 348syncache_drop(struct syncache *sc, struct syncache_head *sch) 349{ 350 351 SCH_LOCK_ASSERT(sch); 352 353 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 354 sch->sch_length--; 355 356 syncache_free(sc); 357 tcp_syncache.cache_count--; 358} 359 360/* 361 * Engage/reengage time on bucket row. 362 */ 363static void 364syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) 365{ 366 sc->sc_rxttime = ticks + 367 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]); 368 sc->sc_rxmits++; 369 if (sch->sch_nextc > sc->sc_rxttime) 370 sch->sch_nextc = sc->sc_rxttime; 371 if (!TAILQ_EMPTY(&sch->sch_bucket) && docallout) 372 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, 373 syncache_timer, (void *)sch); 374} 375 376/* 377 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 378 * If we have retransmitted an entry the maximum number of times, expire it. 379 * One separate timer for each bucket row. 380 */ 381static void 382syncache_timer(void *xsch) 383{ 384 struct syncache_head *sch = (struct syncache_head *)xsch; 385 struct syncache *sc, *nsc; 386 int tick = ticks; 387 char *s; 388 389 /* NB: syncache_head has already been locked by the callout. */ 390 SCH_LOCK_ASSERT(sch); 391 392 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 393 /* 394 * We do not check if the listen socket still exists 395 * and accept the case where the listen socket may be 396 * gone by the time we resend the SYN/ACK. We do 397 * not expect this to happens often. If it does, 398 * then the RST will be sent by the time the remote 399 * host does the SYN/ACK->ACK. 400 */ 401 if (sc->sc_rxttime > tick) { 402 if (sc->sc_rxttime < sch->sch_nextc) 403 sch->sch_nextc = sc->sc_rxttime; 404 continue; 405 } 406 407 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) { 408 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 409 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " 410 "giving up and removing syncache entry\n", 411 s, __func__); 412 free(s, M_TCPLOG); 413 } 414 syncache_drop(sc, sch); 415 tcpstat.tcps_sc_stale++; 416 continue; 417 } 418 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 419 log(LOG_DEBUG, "%s; %s: Response timeout, " 420 "retransmitting (%u) SYN|ACK\n", 421 s, __func__, sc->sc_rxmits); 422 free(s, M_TCPLOG); 423 } 424 425 (void) syncache_respond(sc); 426 tcpstat.tcps_sc_retransmitted++; 427 syncache_timeout(sc, sch, 0); 428 } 429 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 430 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 431 syncache_timer, (void *)(sch)); 432} 433 434/* 435 * Find an entry in the syncache. 436 * Returns always with locked syncache_head plus a matching entry or NULL. 437 */ 438struct syncache * 439syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 440{ 441 struct syncache *sc; 442 struct syncache_head *sch; 443 444#ifdef INET6 445 if (inc->inc_isipv6) { 446 sch = &tcp_syncache.hashbase[ 447 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 448 *schp = sch; 449 450 SCH_LOCK(sch); 451 452 /* Circle through bucket row to find matching entry. */ 453 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 454 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 455 return (sc); 456 } 457 } else 458#endif 459 { 460 sch = &tcp_syncache.hashbase[ 461 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 462 *schp = sch; 463 464 SCH_LOCK(sch); 465 466 /* Circle through bucket row to find matching entry. */ 467 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 468#ifdef INET6 469 if (sc->sc_inc.inc_isipv6) 470 continue; 471#endif 472 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 473 return (sc); 474 } 475 } 476 SCH_LOCK_ASSERT(*schp); 477 return (NULL); /* always returns with locked sch */ 478} 479 480/* 481 * This function is called when we get a RST for a 482 * non-existent connection, so that we can see if the 483 * connection is in the syn cache. If it is, zap it. 484 */ 485void 486syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 487{ 488 struct syncache *sc; 489 struct syncache_head *sch; 490 char *s = NULL; 491 492 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 493 SCH_LOCK_ASSERT(sch); 494 495 /* 496 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. 497 * See RFC 793 page 65, section SEGMENT ARRIVES. 498 */ 499 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { 500 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 501 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " 502 "FIN flag set, segment ignored\n", s, __func__); 503 tcpstat.tcps_badrst++; 504 goto done; 505 } 506 507 /* 508 * No corresponding connection was found in syncache. 509 * If syncookies are enabled and possibly exclusively 510 * used, or we are under memory pressure, a valid RST 511 * may not find a syncache entry. In that case we're 512 * done and no SYN|ACK retransmissions will happen. 513 * Otherwise the the RST was misdirected or spoofed. 514 */ 515 if (sc == NULL) { 516 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 517 log(LOG_DEBUG, "%s; %s: Spurious RST without matching " 518 "syncache entry (possibly syncookie only), " 519 "segment ignored\n", s, __func__); 520 tcpstat.tcps_badrst++; 521 goto done; 522 } 523 524 /* 525 * If the RST bit is set, check the sequence number to see 526 * if this is a valid reset segment. 527 * RFC 793 page 37: 528 * In all states except SYN-SENT, all reset (RST) segments 529 * are validated by checking their SEQ-fields. A reset is 530 * valid if its sequence number is in the window. 531 * 532 * The sequence number in the reset segment is normally an 533 * echo of our outgoing acknowlegement numbers, but some hosts 534 * send a reset with the sequence number at the rightmost edge 535 * of our receive window, and we have to handle this case. 536 */ 537 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 538 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 539 syncache_drop(sc, sch); 540 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 541 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " 542 "connection attempt aborted by remote endpoint\n", 543 s, __func__); 544 tcpstat.tcps_sc_reset++; 545 } else if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 546 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != IRS %u " 547 "(+WND %u), segment ignored\n", 548 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd); 549 tcpstat.tcps_badrst++; 550 } 551 552done: 553 if (s != NULL) 554 free(s, M_TCPLOG); 555 SCH_UNLOCK(sch); 556} 557 558void 559syncache_badack(struct in_conninfo *inc) 560{ 561 struct syncache *sc; 562 struct syncache_head *sch; 563 564 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 565 SCH_LOCK_ASSERT(sch); 566 if (sc != NULL) { 567 syncache_drop(sc, sch); 568 tcpstat.tcps_sc_badack++; 569 } 570 SCH_UNLOCK(sch); 571} 572 573void 574syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 575{ 576 struct syncache *sc; 577 struct syncache_head *sch; 578 579 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 580 SCH_LOCK_ASSERT(sch); 581 if (sc == NULL) 582 goto done; 583 584 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 585 if (ntohl(th->th_seq) != sc->sc_iss) 586 goto done; 587 588 /* 589 * If we've rertransmitted 3 times and this is our second error, 590 * we remove the entry. Otherwise, we allow it to continue on. 591 * This prevents us from incorrectly nuking an entry during a 592 * spurious network outage. 593 * 594 * See tcp_notify(). 595 */ 596 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 597 sc->sc_flags |= SCF_UNREACH; 598 goto done; 599 } 600 syncache_drop(sc, sch); 601 tcpstat.tcps_sc_unreach++; 602done: 603 SCH_UNLOCK(sch); 604} 605 606/* 607 * Build a new TCP socket structure from a syncache entry. 608 */ 609static struct socket * 610syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 611{ 612 struct inpcb *inp = NULL; 613 struct socket *so; 614 struct tcpcb *tp; 615 char *s; 616 617 INP_INFO_WLOCK_ASSERT(&tcbinfo); 618 619 /* 620 * Ok, create the full blown connection, and set things up 621 * as they would have been set up if we had created the 622 * connection when the SYN arrived. If we can't create 623 * the connection, abort it. 624 */ 625 so = sonewconn(lso, SS_ISCONNECTED); 626 if (so == NULL) { 627 /* 628 * Drop the connection; we will either send a RST or 629 * have the peer retransmit its SYN again after its 630 * RTO and try again. 631 */ 632 tcpstat.tcps_listendrop++; 633 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 634 log(LOG_DEBUG, "%s; %s: Socket create failed " 635 "due to limits or memory shortage\n", 636 s, __func__); 637 free(s, M_TCPLOG); 638 } 639 goto abort2; 640 } 641#ifdef MAC 642 SOCK_LOCK(so); 643 mac_socketpeer_set_from_mbuf(m, so); 644 SOCK_UNLOCK(so); 645#endif 646 647 inp = sotoinpcb(so); 648 INP_LOCK(inp); 649 650 /* Insert new socket into PCB hash list. */ 651 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 652#ifdef INET6 653 if (sc->sc_inc.inc_isipv6) { 654 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 655 } else { 656 inp->inp_vflag &= ~INP_IPV6; 657 inp->inp_vflag |= INP_IPV4; 658#endif 659 inp->inp_laddr = sc->sc_inc.inc_laddr; 660#ifdef INET6 661 } 662#endif 663 inp->inp_lport = sc->sc_inc.inc_lport; 664 if (in_pcbinshash(inp) != 0) { 665 /* 666 * Undo the assignments above if we failed to 667 * put the PCB on the hash lists. 668 */ 669#ifdef INET6 670 if (sc->sc_inc.inc_isipv6) 671 inp->in6p_laddr = in6addr_any; 672 else 673#endif 674 inp->inp_laddr.s_addr = INADDR_ANY; 675 inp->inp_lport = 0; 676 goto abort; 677 } 678#ifdef IPSEC 679 /* Copy old policy into new socket's. */ 680 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 681 printf("syncache_socket: could not copy policy\n"); 682#endif 683#ifdef INET6 684 if (sc->sc_inc.inc_isipv6) { 685 struct inpcb *oinp = sotoinpcb(lso); 686 struct in6_addr laddr6; 687 struct sockaddr_in6 sin6; 688 /* 689 * Inherit socket options from the listening socket. 690 * Note that in6p_inputopts are not (and should not be) 691 * copied, since it stores previously received options and is 692 * used to detect if each new option is different than the 693 * previous one and hence should be passed to a user. 694 * If we copied in6p_inputopts, a user would not be able to 695 * receive options just after calling the accept system call. 696 */ 697 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 698 if (oinp->in6p_outputopts) 699 inp->in6p_outputopts = 700 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 701 702 sin6.sin6_family = AF_INET6; 703 sin6.sin6_len = sizeof(sin6); 704 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 705 sin6.sin6_port = sc->sc_inc.inc_fport; 706 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 707 laddr6 = inp->in6p_laddr; 708 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 709 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 710 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 711 thread0.td_ucred)) { 712 inp->in6p_laddr = laddr6; 713 goto abort; 714 } 715 /* Override flowlabel from in6_pcbconnect. */ 716 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 717 inp->in6p_flowinfo |= sc->sc_flowlabel; 718 } else 719#endif 720 { 721 struct in_addr laddr; 722 struct sockaddr_in sin; 723 724 inp->inp_options = ip_srcroute(m); 725 if (inp->inp_options == NULL) { 726 inp->inp_options = sc->sc_ipopts; 727 sc->sc_ipopts = NULL; 728 } 729 730 sin.sin_family = AF_INET; 731 sin.sin_len = sizeof(sin); 732 sin.sin_addr = sc->sc_inc.inc_faddr; 733 sin.sin_port = sc->sc_inc.inc_fport; 734 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 735 laddr = inp->inp_laddr; 736 if (inp->inp_laddr.s_addr == INADDR_ANY) 737 inp->inp_laddr = sc->sc_inc.inc_laddr; 738 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 739 thread0.td_ucred)) { 740 inp->inp_laddr = laddr; 741 goto abort; 742 } 743 } 744 tp = intotcpcb(inp); 745 tp->t_state = TCPS_SYN_RECEIVED; 746 tp->iss = sc->sc_iss; 747 tp->irs = sc->sc_irs; 748 tcp_rcvseqinit(tp); 749 tcp_sendseqinit(tp); 750 tp->snd_wl1 = sc->sc_irs; 751 tp->snd_max = tp->iss + 1; 752 tp->snd_nxt = tp->iss + 1; 753 tp->rcv_up = sc->sc_irs + 1; 754 tp->rcv_wnd = sc->sc_wnd; 755 tp->rcv_adv += tp->rcv_wnd; 756 tp->last_ack_sent = tp->rcv_nxt; 757 758 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 759 if (sc->sc_flags & SCF_NOOPT) 760 tp->t_flags |= TF_NOOPT; 761 else { 762 if (sc->sc_flags & SCF_WINSCALE) { 763 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 764 tp->snd_scale = sc->sc_requested_s_scale; 765 tp->request_r_scale = sc->sc_requested_r_scale; 766 } 767 if (sc->sc_flags & SCF_TIMESTAMP) { 768 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 769 tp->ts_recent = sc->sc_tsreflect; 770 tp->ts_recent_age = ticks; 771 tp->ts_offset = sc->sc_tsoff; 772 } 773#ifdef TCP_SIGNATURE 774 if (sc->sc_flags & SCF_SIGNATURE) 775 tp->t_flags |= TF_SIGNATURE; 776#endif 777 if (sc->sc_flags & SCF_SACK) 778 tp->t_flags |= TF_SACK_PERMIT; 779 } 780 781 /* 782 * Set up MSS and get cached values from tcp_hostcache. 783 * This might overwrite some of the defaults we just set. 784 */ 785 tcp_mss(tp, sc->sc_peer_mss); 786 787 /* 788 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 789 */ 790 if (sc->sc_rxmits) 791 tp->snd_cwnd = tp->t_maxseg; 792 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 793 794 INP_UNLOCK(inp); 795 796 tcpstat.tcps_accepts++; 797 return (so); 798 799abort: 800 INP_UNLOCK(inp); 801abort2: 802 if (so != NULL) 803 soabort(so); 804 return (NULL); 805} 806 807/* 808 * This function gets called when we receive an ACK for a 809 * socket in the LISTEN state. We look up the connection 810 * in the syncache, and if its there, we pull it out of 811 * the cache and turn it into a full-blown connection in 812 * the SYN-RECEIVED state. 813 */ 814int 815syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 816 struct socket **lsop, struct mbuf *m) 817{ 818 struct syncache *sc; 819 struct syncache_head *sch; 820 struct syncache scs; 821 char *s; 822 823 /* 824 * Global TCP locks are held because we manipulate the PCB lists 825 * and create a new socket. 826 */ 827 INP_INFO_WLOCK_ASSERT(&tcbinfo); 828 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, 829 ("%s: can handle only ACK", __func__)); 830 831 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 832 SCH_LOCK_ASSERT(sch); 833 if (sc == NULL) { 834 /* 835 * There is no syncache entry, so see if this ACK is 836 * a returning syncookie. To do this, first: 837 * A. See if this socket has had a syncache entry dropped in 838 * the past. We don't want to accept a bogus syncookie 839 * if we've never received a SYN. 840 * B. check that the syncookie is valid. If it is, then 841 * cobble up a fake syncache entry, and return. 842 */ 843 if (!tcp_syncookies) { 844 SCH_UNLOCK(sch); 845 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 846 log(LOG_DEBUG, "%s; %s: Spurious ACK, " 847 "segment rejected (syncookies disabled)\n", 848 s, __func__); 849 goto failed; 850 } 851 bzero(&scs, sizeof(scs)); 852 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 853 SCH_UNLOCK(sch); 854 if (sc == NULL) { 855 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 856 log(LOG_DEBUG, "%s; %s: Segment failed " 857 "SYNCOOKIE authentication, segment rejected " 858 "(probably spoofed)\n", s, __func__); 859 goto failed; 860 } 861 } else { 862 /* Pull out the entry to unlock the bucket row. */ 863 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 864 sch->sch_length--; 865 tcp_syncache.cache_count--; 866 SCH_UNLOCK(sch); 867 } 868 869 /* 870 * Segment validation: 871 * ACK must match our initial sequence number + 1 (the SYN|ACK). 872 */ 873 if (th->th_ack != sc->sc_iss + 1) { 874 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 875 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " 876 "rejected\n", s, __func__, th->th_ack, sc->sc_iss); 877 goto failed; 878 } 879 /* 880 * The SEQ must match the received initial receive sequence 881 * number + 1 (the SYN) because we didn't ACK any data that 882 * may have come with the SYN. 883 */ 884 if (th->th_seq != sc->sc_irs + 1) { 885 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 886 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " 887 "rejected\n", s, __func__, th->th_seq, sc->sc_irs); 888 goto failed; 889 } 890 /* 891 * If timestamps were present in the SYN and we accepted 892 * them in our SYN|ACK we require them to be present from 893 * now on. And vice versa. 894 */ 895 if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) { 896 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 897 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 898 "segment rejected\n", s, __func__); 899 goto failed; 900 } 901 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { 902 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 903 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 904 "segment rejected\n", s, __func__); 905 goto failed; 906 } 907 /* 908 * If timestamps were negotiated the reflected timestamp 909 * must be equal to what we actually sent in the SYN|ACK. 910 */ 911 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) { 912 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 913 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, " 914 "segment rejected\n", 915 s, __func__, to->to_tsecr, sc->sc_ts); 916 goto failed; 917 } 918 919 *lsop = syncache_socket(sc, *lsop, m); 920 921 if (*lsop == NULL) 922 tcpstat.tcps_sc_aborted++; 923 else 924 tcpstat.tcps_sc_completed++; 925 926 if (sc != &scs) 927 syncache_free(sc); 928 return (1); 929failed: 930 if (sc != NULL && sc != &scs) 931 syncache_free(sc); 932 if (s != NULL) 933 free(s, M_TCPLOG); 934 *lsop = NULL; 935 return (0); 936} 937 938/* 939 * Given a LISTEN socket and an inbound SYN request, add 940 * this to the syn cache, and send back a segment: 941 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 942 * to the source. 943 * 944 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 945 * Doing so would require that we hold onto the data and deliver it 946 * to the application. However, if we are the target of a SYN-flood 947 * DoS attack, an attacker could send data which would eventually 948 * consume all available buffer space if it were ACKed. By not ACKing 949 * the data, we avoid this DoS scenario. 950 */ 951void 952syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 953 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 954{ 955 struct tcpcb *tp; 956 struct socket *so; 957 struct syncache *sc = NULL; 958 struct syncache_head *sch; 959 struct mbuf *ipopts = NULL; 960 u_int32_t flowtmp; 961 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 962 char *s; 963#ifdef INET6 964 int autoflowlabel = 0; 965#endif 966#ifdef MAC 967 struct label *maclabel; 968#endif 969 struct syncache scs; 970 971 INP_INFO_WLOCK_ASSERT(&tcbinfo); 972 INP_LOCK_ASSERT(inp); /* listen socket */ 973 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, 974 ("%s: unexpected tcp flags", __func__)); 975 976 /* 977 * Combine all so/tp operations very early to drop the INP lock as 978 * soon as possible. 979 */ 980 so = *lsop; 981 tp = sototcpcb(so); 982 983#ifdef INET6 984 if (inc->inc_isipv6 && 985 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 986 autoflowlabel = 1; 987#endif 988 ip_ttl = inp->inp_ip_ttl; 989 ip_tos = inp->inp_ip_tos; 990 win = sbspace(&so->so_rcv); 991 sb_hiwat = so->so_rcv.sb_hiwat; 992 noopt = (tp->t_flags & TF_NOOPT); 993 994 so = NULL; 995 tp = NULL; 996 997#ifdef MAC 998 if (mac_syncache_init(&maclabel) != 0) { 999 INP_UNLOCK(inp); 1000 INP_INFO_WUNLOCK(&tcbinfo); 1001 goto done; 1002 } else 1003 mac_syncache_create(maclabel, inp); 1004#endif 1005 INP_UNLOCK(inp); 1006 INP_INFO_WUNLOCK(&tcbinfo); 1007 1008 /* 1009 * Remember the IP options, if any. 1010 */ 1011#ifdef INET6 1012 if (!inc->inc_isipv6) 1013#endif 1014 ipopts = ip_srcroute(m); 1015 1016 /* 1017 * See if we already have an entry for this connection. 1018 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1019 * 1020 * XXX: should the syncache be re-initialized with the contents 1021 * of the new SYN here (which may have different options?) 1022 * 1023 * XXX: We do not check the sequence number to see if this is a 1024 * real retransmit or a new connection attempt. The question is 1025 * how to handle such a case; either ignore it as spoofed, or 1026 * drop the current entry and create a new one? 1027 */ 1028 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 1029 SCH_LOCK_ASSERT(sch); 1030 if (sc != NULL) { 1031 tcpstat.tcps_sc_dupsyn++; 1032 if (ipopts) { 1033 /* 1034 * If we were remembering a previous source route, 1035 * forget it and use the new one we've been given. 1036 */ 1037 if (sc->sc_ipopts) 1038 (void) m_free(sc->sc_ipopts); 1039 sc->sc_ipopts = ipopts; 1040 } 1041 /* 1042 * Update timestamp if present. 1043 */ 1044 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 1045 sc->sc_tsreflect = to->to_tsval; 1046 else 1047 sc->sc_flags &= ~SCF_TIMESTAMP; 1048#ifdef MAC 1049 /* 1050 * Since we have already unconditionally allocated label 1051 * storage, free it up. The syncache entry will already 1052 * have an initialized label we can use. 1053 */ 1054 mac_syncache_destroy(&maclabel); 1055 KASSERT(sc->sc_label != NULL, 1056 ("%s: label not initialized", __func__)); 1057#endif 1058 /* Retransmit SYN|ACK and reset retransmit count. */ 1059 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { 1060 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " 1061 "resetting timer and retransmitting SYN|ACK\n", 1062 s, __func__); 1063 free(s, M_TCPLOG); 1064 } 1065 if (syncache_respond(sc) == 0) { 1066 sc->sc_rxmits = 0; 1067 syncache_timeout(sc, sch, 1); 1068 tcpstat.tcps_sndacks++; 1069 tcpstat.tcps_sndtotal++; 1070 } 1071 SCH_UNLOCK(sch); 1072 goto done; 1073 } 1074 1075 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1076 if (sc == NULL) { 1077 /* 1078 * The zone allocator couldn't provide more entries. 1079 * Treat this as if the cache was full; drop the oldest 1080 * entry and insert the new one. 1081 */ 1082 tcpstat.tcps_sc_zonefail++; 1083 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 1084 syncache_drop(sc, sch); 1085 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1086 if (sc == NULL) { 1087 if (tcp_syncookies) { 1088 bzero(&scs, sizeof(scs)); 1089 sc = &scs; 1090 } else { 1091 SCH_UNLOCK(sch); 1092 if (ipopts) 1093 (void) m_free(ipopts); 1094 goto done; 1095 } 1096 } 1097 } 1098 1099 /* 1100 * Fill in the syncache values. 1101 */ 1102#ifdef MAC 1103 sc->sc_label = maclabel; 1104#endif 1105 sc->sc_ipopts = ipopts; 1106 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1107#ifdef INET6 1108 if (!inc->inc_isipv6) 1109#endif 1110 { 1111 sc->sc_ip_tos = ip_tos; 1112 sc->sc_ip_ttl = ip_ttl; 1113 } 1114 1115 sc->sc_irs = th->th_seq; 1116 sc->sc_iss = arc4random(); 1117 sc->sc_flags = 0; 1118 sc->sc_flowlabel = 0; 1119 1120 /* 1121 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1122 * win was derived from socket earlier in the function. 1123 */ 1124 win = imax(win, 0); 1125 win = imin(win, TCP_MAXWIN); 1126 sc->sc_wnd = win; 1127 1128 if (tcp_do_rfc1323) { 1129 /* 1130 * A timestamp received in a SYN makes 1131 * it ok to send timestamp requests and replies. 1132 */ 1133 if (to->to_flags & TOF_TS) { 1134 sc->sc_tsreflect = to->to_tsval; 1135 sc->sc_ts = ticks; 1136 sc->sc_flags |= SCF_TIMESTAMP; 1137 } 1138 if (to->to_flags & TOF_SCALE) { 1139 int wscale = 0; 1140 1141 /* 1142 * Pick the smallest possible scaling factor that 1143 * will still allow us to scale up to sb_max, aka 1144 * kern.ipc.maxsockbuf. 1145 * 1146 * We do this because there are broken firewalls that 1147 * will corrupt the window scale option, leading to 1148 * the other endpoint believing that our advertised 1149 * window is unscaled. At scale factors larger than 1150 * 5 the unscaled window will drop below 1500 bytes, 1151 * leading to serious problems when traversing these 1152 * broken firewalls. 1153 * 1154 * With the default maxsockbuf of 256K, a scale factor 1155 * of 3 will be chosen by this algorithm. Those who 1156 * choose a larger maxsockbuf should watch out 1157 * for the compatiblity problems mentioned above. 1158 * 1159 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1160 * or <SYN,ACK>) segment itself is never scaled. 1161 */ 1162 while (wscale < TCP_MAX_WINSHIFT && 1163 (TCP_MAXWIN << wscale) < sb_max) 1164 wscale++; 1165 sc->sc_requested_r_scale = wscale; 1166 sc->sc_requested_s_scale = to->to_wscale; 1167 sc->sc_flags |= SCF_WINSCALE; 1168 } 1169 } 1170#ifdef TCP_SIGNATURE 1171 /* 1172 * If listening socket requested TCP digests, and received SYN 1173 * contains the option, flag this in the syncache so that 1174 * syncache_respond() will do the right thing with the SYN+ACK. 1175 * XXX: Currently we always record the option by default and will 1176 * attempt to use it in syncache_respond(). 1177 */ 1178 if (to->to_flags & TOF_SIGNATURE) 1179 sc->sc_flags |= SCF_SIGNATURE; 1180#endif 1181 if (to->to_flags & TOF_SACK) 1182 sc->sc_flags |= SCF_SACK; 1183 if (to->to_flags & TOF_MSS) 1184 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1185 if (noopt) 1186 sc->sc_flags |= SCF_NOOPT; 1187 1188 if (tcp_syncookies) { 1189 syncookie_generate(sch, sc, &flowtmp); 1190#ifdef INET6 1191 if (autoflowlabel) 1192 sc->sc_flowlabel = flowtmp; 1193#endif 1194 } else { 1195#ifdef INET6 1196 if (autoflowlabel) 1197 sc->sc_flowlabel = 1198 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1199#endif 1200 } 1201 SCH_UNLOCK(sch); 1202 1203 /* 1204 * Do a standard 3-way handshake. 1205 */ 1206 if (syncache_respond(sc) == 0) { 1207 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs) 1208 syncache_free(sc); 1209 else if (sc != &scs) 1210 syncache_insert(sc, sch); /* locks and unlocks sch */ 1211 tcpstat.tcps_sndacks++; 1212 tcpstat.tcps_sndtotal++; 1213 } else { 1214 if (sc != &scs) 1215 syncache_free(sc); 1216 tcpstat.tcps_sc_dropped++; 1217 } 1218 1219done: 1220#ifdef MAC 1221 if (sc == &scs) 1222 mac_syncache_destroy(&maclabel); 1223#endif 1224 *lsop = NULL; 1225 m_freem(m); 1226 return; 1227} 1228 1229static int 1230syncache_respond(struct syncache *sc) 1231{ 1232 struct ip *ip = NULL; 1233 struct mbuf *m; 1234 struct tcphdr *th; 1235 int optlen, error; 1236 u_int16_t hlen, tlen, mssopt; 1237 struct tcpopt to; 1238#ifdef INET6 1239 struct ip6_hdr *ip6 = NULL; 1240#endif 1241 1242 hlen = 1243#ifdef INET6 1244 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1245#endif 1246 sizeof(struct ip); 1247 tlen = hlen + sizeof(struct tcphdr); 1248 1249 /* Determine MSS we advertize to other end of connection. */ 1250 mssopt = tcp_mssopt(&sc->sc_inc); 1251 if (sc->sc_peer_mss) 1252 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss); 1253 1254 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1255 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, 1256 ("syncache: mbuf too small")); 1257 1258 /* Create the IP+TCP header from scratch. */ 1259 m = m_gethdr(M_DONTWAIT, MT_DATA); 1260 if (m == NULL) 1261 return (ENOBUFS); 1262#ifdef MAC 1263 mac_syncache_create_mbuf(sc->sc_label, m); 1264#endif 1265 m->m_data += max_linkhdr; 1266 m->m_len = tlen; 1267 m->m_pkthdr.len = tlen; 1268 m->m_pkthdr.rcvif = NULL; 1269 1270#ifdef INET6 1271 if (sc->sc_inc.inc_isipv6) { 1272 ip6 = mtod(m, struct ip6_hdr *); 1273 ip6->ip6_vfc = IPV6_VERSION; 1274 ip6->ip6_nxt = IPPROTO_TCP; 1275 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1276 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1277 ip6->ip6_plen = htons(tlen - hlen); 1278 /* ip6_hlim is set after checksum */ 1279 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1280 ip6->ip6_flow |= sc->sc_flowlabel; 1281 1282 th = (struct tcphdr *)(ip6 + 1); 1283 } else 1284#endif 1285 { 1286 ip = mtod(m, struct ip *); 1287 ip->ip_v = IPVERSION; 1288 ip->ip_hl = sizeof(struct ip) >> 2; 1289 ip->ip_len = tlen; 1290 ip->ip_id = 0; 1291 ip->ip_off = 0; 1292 ip->ip_sum = 0; 1293 ip->ip_p = IPPROTO_TCP; 1294 ip->ip_src = sc->sc_inc.inc_laddr; 1295 ip->ip_dst = sc->sc_inc.inc_faddr; 1296 ip->ip_ttl = sc->sc_ip_ttl; 1297 ip->ip_tos = sc->sc_ip_tos; 1298 1299 /* 1300 * See if we should do MTU discovery. Route lookups are 1301 * expensive, so we will only unset the DF bit if: 1302 * 1303 * 1) path_mtu_discovery is disabled 1304 * 2) the SCF_UNREACH flag has been set 1305 */ 1306 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1307 ip->ip_off |= IP_DF; 1308 1309 th = (struct tcphdr *)(ip + 1); 1310 } 1311 th->th_sport = sc->sc_inc.inc_lport; 1312 th->th_dport = sc->sc_inc.inc_fport; 1313 1314 th->th_seq = htonl(sc->sc_iss); 1315 th->th_ack = htonl(sc->sc_irs + 1); 1316 th->th_off = sizeof(struct tcphdr) >> 2; 1317 th->th_x2 = 0; 1318 th->th_flags = TH_SYN|TH_ACK; 1319 th->th_win = htons(sc->sc_wnd); 1320 th->th_urp = 0; 1321 1322 /* Tack on the TCP options. */ 1323 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1324 to.to_flags = 0; 1325 1326 to.to_mss = mssopt; 1327 to.to_flags = TOF_MSS; 1328 if (sc->sc_flags & SCF_WINSCALE) { 1329 to.to_wscale = sc->sc_requested_r_scale; 1330 to.to_flags |= TOF_SCALE; 1331 } 1332 if (sc->sc_flags & SCF_TIMESTAMP) { 1333 /* Virgin timestamp or TCP cookie enhanced one. */ 1334 to.to_tsval = sc->sc_ts; 1335 to.to_tsecr = sc->sc_tsreflect; 1336 to.to_flags |= TOF_TS; 1337 } 1338 if (sc->sc_flags & SCF_SACK) 1339 to.to_flags |= TOF_SACKPERM; 1340#ifdef TCP_SIGNATURE 1341 if (sc->sc_flags & SCF_SIGNATURE) 1342 to.to_flags |= TOF_SIGNATURE; 1343#endif 1344 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1345 1346#ifdef TCP_SIGNATURE 1347 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1348 to.to_signature, IPSEC_DIR_OUTBOUND); 1349#endif 1350 1351 /* Adjust headers by option size. */ 1352 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1353 m->m_len += optlen; 1354 m->m_pkthdr.len += optlen; 1355#ifdef INET6 1356 if (sc->sc_inc.inc_isipv6) 1357 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1358 else 1359#endif 1360 ip->ip_len += optlen; 1361 } else 1362 optlen = 0; 1363 1364#ifdef INET6 1365 if (sc->sc_inc.inc_isipv6) { 1366 th->th_sum = 0; 1367 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1368 tlen + optlen - hlen); 1369 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1370 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1371 } else 1372#endif 1373 { 1374 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1375 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1376 m->m_pkthdr.csum_flags = CSUM_TCP; 1377 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1378 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1379 } 1380 return (error); 1381} 1382 1383/* 1384 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1385 * receive and to be able to handle SYN floods from bogus source addresses 1386 * (where we will never receive any reply). SYN floods try to exhaust all 1387 * our memory and available slots in the SYN cache table to cause a denial 1388 * of service to legitimate users of the local host. 1389 * 1390 * The idea of SYN cookies is to encode and include all necessary information 1391 * about the connection setup state within the SYN-ACK we send back and thus 1392 * to get along without keeping any local state until the ACK to the SYN-ACK 1393 * arrives (if ever). Everything we need to know should be available from 1394 * the information we encoded in the SYN-ACK. 1395 * 1396 * More information about the theory behind SYN cookies and its first 1397 * discussion and specification can be found at: 1398 * http://cr.yp.to/syncookies.html (overview) 1399 * http://cr.yp.to/syncookies/archive (gory details) 1400 * 1401 * This implementation extends the orginal idea and first implementation 1402 * of FreeBSD by using not only the initial sequence number field to store 1403 * information but also the timestamp field if present. This way we can 1404 * keep track of the entire state we need to know to recreate the session in 1405 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1406 * these days. For those that do not we still have to live with the known 1407 * shortcomings of the ISN only SYN cookies. 1408 * 1409 * Cookie layers: 1410 * 1411 * Initial sequence number we send: 1412 * 31|................................|0 1413 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1414 * D = MD5 Digest (first dword) 1415 * M = MSS index 1416 * R = Rotation of secret 1417 * P = Odd or Even secret 1418 * 1419 * The MD5 Digest is computed with over following parameters: 1420 * a) randomly rotated secret 1421 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1422 * c) the received initial sequence number from remote host 1423 * d) the rotation offset and odd/even bit 1424 * 1425 * Timestamp we send: 1426 * 31|................................|0 1427 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1428 * D = MD5 Digest (third dword) (only as filler) 1429 * S = Requested send window scale 1430 * R = Requested receive window scale 1431 * A = SACK allowed 1432 * 5 = TCP-MD5 enabled (not implemented yet) 1433 * XORed with MD5 Digest (forth dword) 1434 * 1435 * The timestamp isn't cryptographically secure and doesn't need to be. 1436 * The double use of the MD5 digest dwords ties it to a specific remote/ 1437 * local host/port, remote initial sequence number and our local time 1438 * limited secret. A received timestamp is reverted (XORed) and then 1439 * the contained MD5 dword is compared to the computed one to ensure the 1440 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1441 * have been tampered with but this isn't different from supplying bogus 1442 * values in the SYN in the first place. 1443 * 1444 * Some problems with SYN cookies remain however: 1445 * Consider the problem of a recreated (and retransmitted) cookie. If the 1446 * original SYN was accepted, the connection is established. The second 1447 * SYN is inflight, and if it arrives with an ISN that falls within the 1448 * receive window, the connection is killed. 1449 * 1450 * Notes: 1451 * A heuristic to determine when to accept syn cookies is not necessary. 1452 * An ACK flood would cause the syncookie verification to be attempted, 1453 * but a SYN flood causes syncookies to be generated. Both are of equal 1454 * cost, so there's no point in trying to optimize the ACK flood case. 1455 * Also, if you don't process certain ACKs for some reason, then all someone 1456 * would have to do is launch a SYN and ACK flood at the same time, which 1457 * would stop cookie verification and defeat the entire purpose of syncookies. 1458 */ 1459static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1460 1461static void 1462syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1463 u_int32_t *flowlabel) 1464{ 1465 MD5_CTX ctx; 1466 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1467 u_int32_t data; 1468 u_int32_t *secbits; 1469 u_int off, pmss, mss; 1470 int i; 1471 1472 SCH_LOCK_ASSERT(sch); 1473 1474 /* Which of the two secrets to use. */ 1475 secbits = sch->sch_oddeven ? 1476 sch->sch_secbits_odd : sch->sch_secbits_even; 1477 1478 /* Reseed secret if too old. */ 1479 if (sch->sch_reseed < time_uptime) { 1480 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1481 secbits = sch->sch_oddeven ? 1482 sch->sch_secbits_odd : sch->sch_secbits_even; 1483 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1484 secbits[i] = arc4random(); 1485 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1486 } 1487 1488 /* Secret rotation offset. */ 1489 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1490 1491 /* Maximum segment size calculation. */ 1492 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss); 1493 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1494 if (tcp_sc_msstab[mss] <= pmss) 1495 break; 1496 1497 /* Fold parameters and MD5 digest into the ISN we will send. */ 1498 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1499 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1500 data |= mss << 4; /* mss, 3 bits */ 1501 1502 MD5Init(&ctx); 1503 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1504 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1505 MD5Update(&ctx, secbits, off); 1506 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1507 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1508 MD5Update(&ctx, &data, sizeof(data)); 1509 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1510 1511 data |= (md5_buffer[0] << 7); 1512 sc->sc_iss = data; 1513 1514#ifdef INET6 1515 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1516#endif 1517 1518 /* Additional parameters are stored in the timestamp if present. */ 1519 if (sc->sc_flags & SCF_TIMESTAMP) { 1520 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1521 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1522 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1523 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1524 data |= md5_buffer[2] << 10; /* more digest bits */ 1525 data ^= md5_buffer[3]; 1526 sc->sc_ts = data; 1527 sc->sc_tsoff = data - ticks; /* after XOR */ 1528 } 1529 1530 tcpstat.tcps_sc_sendcookie++; 1531 return; 1532} 1533 1534static struct syncache * 1535syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1536 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1537 struct socket *so) 1538{ 1539 MD5_CTX ctx; 1540 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1541 u_int32_t data = 0; 1542 u_int32_t *secbits; 1543 tcp_seq ack, seq; 1544 int off, mss, wnd, flags; 1545 1546 SCH_LOCK_ASSERT(sch); 1547 1548 /* 1549 * Pull information out of SYN-ACK/ACK and 1550 * revert sequence number advances. 1551 */ 1552 ack = th->th_ack - 1; 1553 seq = th->th_seq - 1; 1554 off = (ack >> 1) & 0x7; 1555 mss = (ack >> 4) & 0x7; 1556 flags = ack & 0x7f; 1557 1558 /* Which of the two secrets to use. */ 1559 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1560 1561 /* 1562 * The secret wasn't updated for the lifetime of a syncookie, 1563 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1564 */ 1565 if (sch->sch_reseed < time_uptime) { 1566 return (NULL); 1567 } 1568 1569 /* Recompute the digest so we can compare it. */ 1570 MD5Init(&ctx); 1571 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1572 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1573 MD5Update(&ctx, secbits, off); 1574 MD5Update(&ctx, inc, sizeof(*inc)); 1575 MD5Update(&ctx, &seq, sizeof(seq)); 1576 MD5Update(&ctx, &flags, sizeof(flags)); 1577 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1578 1579 /* Does the digest part of or ACK'ed ISS match? */ 1580 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1581 return (NULL); 1582 1583 /* Does the digest part of our reflected timestamp match? */ 1584 if (to->to_flags & TOF_TS) { 1585 data = md5_buffer[3] ^ to->to_tsecr; 1586 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1587 return (NULL); 1588 } 1589 1590 /* Fill in the syncache values. */ 1591 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1592 sc->sc_ipopts = NULL; 1593 1594 sc->sc_irs = seq; 1595 sc->sc_iss = ack; 1596 1597#ifdef INET6 1598 if (inc->inc_isipv6) { 1599 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1600 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1601 } else 1602#endif 1603 { 1604 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1605 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1606 } 1607 1608 /* Additional parameters that were encoded in the timestamp. */ 1609 if (data) { 1610 sc->sc_flags |= SCF_TIMESTAMP; 1611 sc->sc_tsreflect = to->to_tsval; 1612 sc->sc_ts = to->to_tsecr; 1613 sc->sc_tsoff = to->to_tsecr - ticks; 1614 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1615 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1616 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1617 TCP_MAX_WINSHIFT); 1618 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1619 TCP_MAX_WINSHIFT); 1620 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1621 sc->sc_flags |= SCF_WINSCALE; 1622 } else 1623 sc->sc_flags |= SCF_NOOPT; 1624 1625 wnd = sbspace(&so->so_rcv); 1626 wnd = imax(wnd, 0); 1627 wnd = imin(wnd, TCP_MAXWIN); 1628 sc->sc_wnd = wnd; 1629 1630 sc->sc_rxmits = 0; 1631 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1632 1633 tcpstat.tcps_sc_recvcookie++; 1634 return (sc); 1635} 1636 1637/* 1638 * Returns the current number of syncache entries. This number 1639 * will probably change before you get around to calling 1640 * syncache_pcblist. 1641 */ 1642 1643int 1644syncache_pcbcount(void) 1645{ 1646 struct syncache_head *sch; 1647 int count, i; 1648 1649 for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) { 1650 /* No need to lock for a read. */ 1651 sch = &tcp_syncache.hashbase[i]; 1652 count += sch->sch_length; 1653 } 1654 return count; 1655} 1656 1657/* 1658 * Exports the syncache entries to userland so that netstat can display 1659 * them alongside the other sockets. This function is intended to be 1660 * called only from tcp_pcblist. 1661 * 1662 * Due to concurrency on an active system, the number of pcbs exported 1663 * may have no relation to max_pcbs. max_pcbs merely indicates the 1664 * amount of space the caller allocated for this function to use. 1665 */ 1666int 1667syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported) 1668{ 1669 struct xtcpcb xt; 1670 struct syncache *sc; 1671 struct syncache_head *sch; 1672 int count, error, i; 1673 1674 for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) { 1675 sch = &tcp_syncache.hashbase[i]; 1676 SCH_LOCK(sch); 1677 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 1678 if (count >= max_pcbs) { 1679 SCH_UNLOCK(sch); 1680 goto exit; 1681 } 1682 bzero(&xt, sizeof(xt)); 1683 xt.xt_len = sizeof(xt); 1684 if (sc->sc_inc.inc_isipv6) 1685 xt.xt_inp.inp_vflag = INP_IPV6; 1686 else 1687 xt.xt_inp.inp_vflag = INP_IPV4; 1688 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); 1689 xt.xt_tp.t_inpcb = &xt.xt_inp; 1690 xt.xt_tp.t_state = TCPS_SYN_RECEIVED; 1691 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1692 xt.xt_socket.xso_len = sizeof (struct xsocket); 1693 xt.xt_socket.so_type = SOCK_STREAM; 1694 xt.xt_socket.so_state = SS_ISCONNECTING; 1695 error = SYSCTL_OUT(req, &xt, sizeof xt); 1696 if (error) { 1697 SCH_UNLOCK(sch); 1698 goto exit; 1699 } 1700 count++; 1701 } 1702 SCH_UNLOCK(sch); 1703 } 1704exit: 1705 *pcbs_exported = count; 1706 return error; 1707} 1708 1709