tcp_syncache.c revision 174545
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 174545 2007-12-12 06:11:50Z kmacy $"); 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#ifndef DISABLE_TCP_OFFLOAD 139 void *sc_pspare[2]; /* toepcb / toe_usrreqs */ 140#endif 141#ifdef MAC 142 struct label *sc_label; /* MAC label reference */ 143#endif 144}; 145 146struct syncache_head { 147 struct mtx sch_mtx; 148 TAILQ_HEAD(sch_head, syncache) sch_bucket; 149 struct callout sch_timer; 150 int sch_nextc; 151 u_int sch_length; 152 u_int sch_oddeven; 153 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE]; 154 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE]; 155 u_int sch_reseed; /* time_uptime, seconds */ 156}; 157 158static void syncache_drop(struct syncache *, struct syncache_head *); 159static void syncache_free(struct syncache *); 160static void syncache_insert(struct syncache *, struct syncache_head *); 161struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 162static int syncache_respond(struct syncache *); 163static struct socket *syncache_socket(struct syncache *, struct socket *, 164 struct mbuf *m); 165static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, 166 int docallout); 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 SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 244#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 245#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 246 247/* 248 * Requires the syncache entry to be already removed from the bucket list. 249 */ 250static void 251syncache_free(struct syncache *sc) 252{ 253 if (sc->sc_ipopts) 254 (void) m_free(sc->sc_ipopts); 255#ifdef MAC 256 mac_syncache_destroy(&sc->sc_label); 257#endif 258 259 uma_zfree(tcp_syncache.zone, sc); 260} 261 262void 263syncache_init(void) 264{ 265 int i; 266 267 tcp_syncache.cache_count = 0; 268 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 269 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 270 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 271 tcp_syncache.hash_secret = arc4random(); 272 273 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 274 &tcp_syncache.hashsize); 275 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 276 &tcp_syncache.bucket_limit); 277 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 278 printf("WARNING: syncache hash size is not a power of 2.\n"); 279 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 280 } 281 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 282 283 /* Set limits. */ 284 tcp_syncache.cache_limit = 285 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 286 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 287 &tcp_syncache.cache_limit); 288 289 /* Allocate the hash table. */ 290 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 291 tcp_syncache.hashsize * sizeof(struct syncache_head), 292 M_SYNCACHE, M_WAITOK | M_ZERO); 293 294 /* Initialize the hash buckets. */ 295 for (i = 0; i < tcp_syncache.hashsize; i++) { 296 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 297 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 298 NULL, MTX_DEF); 299 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer, 300 &tcp_syncache.hashbase[i].sch_mtx, 0); 301 tcp_syncache.hashbase[i].sch_length = 0; 302 } 303 304 /* Create the syncache entry zone. */ 305 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 306 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 307 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 308} 309 310/* 311 * Inserts a syncache entry into the specified bucket row. 312 * Locks and unlocks the syncache_head autonomously. 313 */ 314static void 315syncache_insert(struct syncache *sc, struct syncache_head *sch) 316{ 317 struct syncache *sc2; 318 319 SCH_LOCK(sch); 320 321 /* 322 * Make sure that we don't overflow the per-bucket limit. 323 * If the bucket is full, toss the oldest element. 324 */ 325 if (sch->sch_length >= tcp_syncache.bucket_limit) { 326 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 327 ("sch->sch_length incorrect")); 328 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 329 syncache_drop(sc2, sch); 330 tcpstat.tcps_sc_bucketoverflow++; 331 } 332 333 /* Put it into the bucket. */ 334 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 335 sch->sch_length++; 336 337 /* Reinitialize the bucket row's timer. */ 338 syncache_timeout(sc, sch, 1); 339 340 SCH_UNLOCK(sch); 341 342 tcp_syncache.cache_count++; 343 tcpstat.tcps_sc_added++; 344} 345 346/* 347 * Remove and free entry from syncache bucket row. 348 * Expects locked syncache head. 349 */ 350static void 351syncache_drop(struct syncache *sc, struct syncache_head *sch) 352{ 353 354 SCH_LOCK_ASSERT(sch); 355 356 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 357 sch->sch_length--; 358 359 syncache_free(sc); 360 tcp_syncache.cache_count--; 361} 362 363/* 364 * Engage/reengage time on bucket row. 365 */ 366static void 367syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) 368{ 369 sc->sc_rxttime = ticks + 370 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]); 371 sc->sc_rxmits++; 372 if (sch->sch_nextc > sc->sc_rxttime) 373 sch->sch_nextc = sc->sc_rxttime; 374 if (!TAILQ_EMPTY(&sch->sch_bucket) && docallout) 375 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, 376 syncache_timer, (void *)sch); 377} 378 379/* 380 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 381 * If we have retransmitted an entry the maximum number of times, expire it. 382 * One separate timer for each bucket row. 383 */ 384static void 385syncache_timer(void *xsch) 386{ 387 struct syncache_head *sch = (struct syncache_head *)xsch; 388 struct syncache *sc, *nsc; 389 int tick = ticks; 390 char *s; 391 392 /* NB: syncache_head has already been locked by the callout. */ 393 SCH_LOCK_ASSERT(sch); 394 395 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 396 /* 397 * We do not check if the listen socket still exists 398 * and accept the case where the listen socket may be 399 * gone by the time we resend the SYN/ACK. We do 400 * not expect this to happens often. If it does, 401 * then the RST will be sent by the time the remote 402 * host does the SYN/ACK->ACK. 403 */ 404 if (sc->sc_rxttime > tick) { 405 if (sc->sc_rxttime < sch->sch_nextc) 406 sch->sch_nextc = sc->sc_rxttime; 407 continue; 408 } 409 410 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) { 411 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 412 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " 413 "giving up and removing syncache entry\n", 414 s, __func__); 415 free(s, M_TCPLOG); 416 } 417 syncache_drop(sc, sch); 418 tcpstat.tcps_sc_stale++; 419 continue; 420 } 421 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 422 log(LOG_DEBUG, "%s; %s: Response timeout, " 423 "retransmitting (%u) SYN|ACK\n", 424 s, __func__, sc->sc_rxmits); 425 free(s, M_TCPLOG); 426 } 427 428 (void) syncache_respond(sc); 429 tcpstat.tcps_sc_retransmitted++; 430 syncache_timeout(sc, sch, 0); 431 } 432 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 433 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 434 syncache_timer, (void *)(sch)); 435} 436 437/* 438 * Find an entry in the syncache. 439 * Returns always with locked syncache_head plus a matching entry or NULL. 440 */ 441struct syncache * 442syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 443{ 444 struct syncache *sc; 445 struct syncache_head *sch; 446 447#ifdef INET6 448 if (inc->inc_isipv6) { 449 sch = &tcp_syncache.hashbase[ 450 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 451 *schp = sch; 452 453 SCH_LOCK(sch); 454 455 /* Circle through bucket row to find matching entry. */ 456 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 457 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 458 return (sc); 459 } 460 } else 461#endif 462 { 463 sch = &tcp_syncache.hashbase[ 464 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 465 *schp = sch; 466 467 SCH_LOCK(sch); 468 469 /* Circle through bucket row to find matching entry. */ 470 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 471#ifdef INET6 472 if (sc->sc_inc.inc_isipv6) 473 continue; 474#endif 475 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 476 return (sc); 477 } 478 } 479 SCH_LOCK_ASSERT(*schp); 480 return (NULL); /* always returns with locked sch */ 481} 482 483/* 484 * This function is called when we get a RST for a 485 * non-existent connection, so that we can see if the 486 * connection is in the syn cache. If it is, zap it. 487 */ 488void 489syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 490{ 491 struct syncache *sc; 492 struct syncache_head *sch; 493 char *s = NULL; 494 495 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 496 SCH_LOCK_ASSERT(sch); 497 498 /* 499 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. 500 * See RFC 793 page 65, section SEGMENT ARRIVES. 501 */ 502 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { 503 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 504 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " 505 "FIN flag set, segment ignored\n", s, __func__); 506 tcpstat.tcps_badrst++; 507 goto done; 508 } 509 510 /* 511 * No corresponding connection was found in syncache. 512 * If syncookies are enabled and possibly exclusively 513 * used, or we are under memory pressure, a valid RST 514 * may not find a syncache entry. In that case we're 515 * done and no SYN|ACK retransmissions will happen. 516 * Otherwise the the RST was misdirected or spoofed. 517 */ 518 if (sc == NULL) { 519 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 520 log(LOG_DEBUG, "%s; %s: Spurious RST without matching " 521 "syncache entry (possibly syncookie only), " 522 "segment ignored\n", s, __func__); 523 tcpstat.tcps_badrst++; 524 goto done; 525 } 526 527 /* 528 * If the RST bit is set, check the sequence number to see 529 * if this is a valid reset segment. 530 * RFC 793 page 37: 531 * In all states except SYN-SENT, all reset (RST) segments 532 * are validated by checking their SEQ-fields. A reset is 533 * valid if its sequence number is in the window. 534 * 535 * The sequence number in the reset segment is normally an 536 * echo of our outgoing acknowlegement numbers, but some hosts 537 * send a reset with the sequence number at the rightmost edge 538 * of our receive window, and we have to handle this case. 539 */ 540 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 541 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 542 syncache_drop(sc, sch); 543 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 544 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " 545 "connection attempt aborted by remote endpoint\n", 546 s, __func__); 547 tcpstat.tcps_sc_reset++; 548 } else if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 549 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != IRS %u " 550 "(+WND %u), segment ignored\n", 551 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd); 552 tcpstat.tcps_badrst++; 553 } 554 555done: 556 if (s != NULL) 557 free(s, M_TCPLOG); 558 SCH_UNLOCK(sch); 559} 560 561void 562syncache_badack(struct in_conninfo *inc) 563{ 564 struct syncache *sc; 565 struct syncache_head *sch; 566 567 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 568 SCH_LOCK_ASSERT(sch); 569 if (sc != NULL) { 570 syncache_drop(sc, sch); 571 tcpstat.tcps_sc_badack++; 572 } 573 SCH_UNLOCK(sch); 574} 575 576void 577syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 578{ 579 struct syncache *sc; 580 struct syncache_head *sch; 581 582 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 583 SCH_LOCK_ASSERT(sch); 584 if (sc == NULL) 585 goto done; 586 587 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 588 if (ntohl(th->th_seq) != sc->sc_iss) 589 goto done; 590 591 /* 592 * If we've rertransmitted 3 times and this is our second error, 593 * we remove the entry. Otherwise, we allow it to continue on. 594 * This prevents us from incorrectly nuking an entry during a 595 * spurious network outage. 596 * 597 * See tcp_notify(). 598 */ 599 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 600 sc->sc_flags |= SCF_UNREACH; 601 goto done; 602 } 603 syncache_drop(sc, sch); 604 tcpstat.tcps_sc_unreach++; 605done: 606 SCH_UNLOCK(sch); 607} 608 609/* 610 * Build a new TCP socket structure from a syncache entry. 611 */ 612static struct socket * 613syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 614{ 615 struct inpcb *inp = NULL; 616 struct socket *so; 617 struct tcpcb *tp; 618 char *s; 619 620 INP_INFO_WLOCK_ASSERT(&tcbinfo); 621 622 /* 623 * Ok, create the full blown connection, and set things up 624 * as they would have been set up if we had created the 625 * connection when the SYN arrived. If we can't create 626 * the connection, abort it. 627 */ 628 so = sonewconn(lso, SS_ISCONNECTED); 629 if (so == NULL) { 630 /* 631 * Drop the connection; we will either send a RST or 632 * have the peer retransmit its SYN again after its 633 * RTO and try again. 634 */ 635 tcpstat.tcps_listendrop++; 636 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 637 log(LOG_DEBUG, "%s; %s: Socket create failed " 638 "due to limits or memory shortage\n", 639 s, __func__); 640 free(s, M_TCPLOG); 641 } 642 goto abort2; 643 } 644#ifdef MAC 645 SOCK_LOCK(so); 646 mac_socketpeer_set_from_mbuf(m, so); 647 SOCK_UNLOCK(so); 648#endif 649 650 inp = sotoinpcb(so); 651 INP_LOCK(inp); 652 653 /* Insert new socket into PCB hash list. */ 654 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 655#ifdef INET6 656 if (sc->sc_inc.inc_isipv6) { 657 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 658 } else { 659 inp->inp_vflag &= ~INP_IPV6; 660 inp->inp_vflag |= INP_IPV4; 661#endif 662 inp->inp_laddr = sc->sc_inc.inc_laddr; 663#ifdef INET6 664 } 665#endif 666 inp->inp_lport = sc->sc_inc.inc_lport; 667 if (in_pcbinshash(inp) != 0) { 668 /* 669 * Undo the assignments above if we failed to 670 * put the PCB on the hash lists. 671 */ 672#ifdef INET6 673 if (sc->sc_inc.inc_isipv6) 674 inp->in6p_laddr = in6addr_any; 675 else 676#endif 677 inp->inp_laddr.s_addr = INADDR_ANY; 678 inp->inp_lport = 0; 679 goto abort; 680 } 681#ifdef IPSEC 682 /* Copy old policy into new socket's. */ 683 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 684 printf("syncache_socket: could not copy policy\n"); 685#endif 686#ifdef INET6 687 if (sc->sc_inc.inc_isipv6) { 688 struct inpcb *oinp = sotoinpcb(lso); 689 struct in6_addr laddr6; 690 struct sockaddr_in6 sin6; 691 /* 692 * Inherit socket options from the listening socket. 693 * Note that in6p_inputopts are not (and should not be) 694 * copied, since it stores previously received options and is 695 * used to detect if each new option is different than the 696 * previous one and hence should be passed to a user. 697 * If we copied in6p_inputopts, a user would not be able to 698 * receive options just after calling the accept system call. 699 */ 700 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 701 if (oinp->in6p_outputopts) 702 inp->in6p_outputopts = 703 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 704 705 sin6.sin6_family = AF_INET6; 706 sin6.sin6_len = sizeof(sin6); 707 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 708 sin6.sin6_port = sc->sc_inc.inc_fport; 709 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 710 laddr6 = inp->in6p_laddr; 711 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 712 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 713 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 714 thread0.td_ucred)) { 715 inp->in6p_laddr = laddr6; 716 goto abort; 717 } 718 /* Override flowlabel from in6_pcbconnect. */ 719 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 720 inp->in6p_flowinfo |= sc->sc_flowlabel; 721 } else 722#endif 723 { 724 struct in_addr laddr; 725 struct sockaddr_in sin; 726 727 inp->inp_options = ip_srcroute(m); 728 if (inp->inp_options == NULL) { 729 inp->inp_options = sc->sc_ipopts; 730 sc->sc_ipopts = NULL; 731 } 732 733 sin.sin_family = AF_INET; 734 sin.sin_len = sizeof(sin); 735 sin.sin_addr = sc->sc_inc.inc_faddr; 736 sin.sin_port = sc->sc_inc.inc_fport; 737 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 738 laddr = inp->inp_laddr; 739 if (inp->inp_laddr.s_addr == INADDR_ANY) 740 inp->inp_laddr = sc->sc_inc.inc_laddr; 741 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 742 thread0.td_ucred)) { 743 inp->inp_laddr = laddr; 744 goto abort; 745 } 746 } 747 tp = intotcpcb(inp); 748 tp->t_state = TCPS_SYN_RECEIVED; 749 tp->iss = sc->sc_iss; 750 tp->irs = sc->sc_irs; 751 tcp_rcvseqinit(tp); 752 tcp_sendseqinit(tp); 753 tp->snd_wl1 = sc->sc_irs; 754 tp->snd_max = tp->iss + 1; 755 tp->snd_nxt = tp->iss + 1; 756 tp->rcv_up = sc->sc_irs + 1; 757 tp->rcv_wnd = sc->sc_wnd; 758 tp->rcv_adv += tp->rcv_wnd; 759 tp->last_ack_sent = tp->rcv_nxt; 760 761 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 762 if (sc->sc_flags & SCF_NOOPT) 763 tp->t_flags |= TF_NOOPT; 764 else { 765 if (sc->sc_flags & SCF_WINSCALE) { 766 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 767 tp->snd_scale = sc->sc_requested_s_scale; 768 tp->request_r_scale = sc->sc_requested_r_scale; 769 } 770 if (sc->sc_flags & SCF_TIMESTAMP) { 771 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 772 tp->ts_recent = sc->sc_tsreflect; 773 tp->ts_recent_age = ticks; 774 tp->ts_offset = sc->sc_tsoff; 775 } 776#ifdef TCP_SIGNATURE 777 if (sc->sc_flags & SCF_SIGNATURE) 778 tp->t_flags |= TF_SIGNATURE; 779#endif 780 if (sc->sc_flags & SCF_SACK) 781 tp->t_flags |= TF_SACK_PERMIT; 782 } 783 784 /* 785 * Set up MSS and get cached values from tcp_hostcache. 786 * This might overwrite some of the defaults we just set. 787 */ 788 tcp_mss(tp, sc->sc_peer_mss); 789 790 /* 791 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 792 */ 793 if (sc->sc_rxmits) 794 tp->snd_cwnd = tp->t_maxseg; 795 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 796 797 INP_UNLOCK(inp); 798 799 tcpstat.tcps_accepts++; 800 return (so); 801 802abort: 803 INP_UNLOCK(inp); 804abort2: 805 if (so != NULL) 806 soabort(so); 807 return (NULL); 808} 809 810/* 811 * This function gets called when we receive an ACK for a 812 * socket in the LISTEN state. We look up the connection 813 * in the syncache, and if its there, we pull it out of 814 * the cache and turn it into a full-blown connection in 815 * the SYN-RECEIVED state. 816 */ 817int 818syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 819 struct socket **lsop, struct mbuf *m) 820{ 821 struct syncache *sc; 822 struct syncache_head *sch; 823 struct syncache scs; 824 char *s; 825 826 /* 827 * Global TCP locks are held because we manipulate the PCB lists 828 * and create a new socket. 829 */ 830 INP_INFO_WLOCK_ASSERT(&tcbinfo); 831 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, 832 ("%s: can handle only ACK", __func__)); 833 834 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 835 SCH_LOCK_ASSERT(sch); 836 if (sc == NULL) { 837 /* 838 * There is no syncache entry, so see if this ACK is 839 * a returning syncookie. To do this, first: 840 * A. See if this socket has had a syncache entry dropped in 841 * the past. We don't want to accept a bogus syncookie 842 * if we've never received a SYN. 843 * B. check that the syncookie is valid. If it is, then 844 * cobble up a fake syncache entry, and return. 845 */ 846 if (!tcp_syncookies) { 847 SCH_UNLOCK(sch); 848 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 849 log(LOG_DEBUG, "%s; %s: Spurious ACK, " 850 "segment rejected (syncookies disabled)\n", 851 s, __func__); 852 goto failed; 853 } 854 bzero(&scs, sizeof(scs)); 855 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 856 SCH_UNLOCK(sch); 857 if (sc == NULL) { 858 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 859 log(LOG_DEBUG, "%s; %s: Segment failed " 860 "SYNCOOKIE authentication, segment rejected " 861 "(probably spoofed)\n", s, __func__); 862 goto failed; 863 } 864 } else { 865 /* Pull out the entry to unlock the bucket row. */ 866 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 867 sch->sch_length--; 868 tcp_syncache.cache_count--; 869 SCH_UNLOCK(sch); 870 } 871 872 /* 873 * Segment validation: 874 * ACK must match our initial sequence number + 1 (the SYN|ACK). 875 */ 876 if (th->th_ack != sc->sc_iss + 1) { 877 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 878 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " 879 "rejected\n", s, __func__, th->th_ack, sc->sc_iss); 880 goto failed; 881 } 882 /* 883 * The SEQ must match the received initial receive sequence 884 * number + 1 (the SYN) because we didn't ACK any data that 885 * may have come with the SYN. 886 */ 887 if (th->th_seq != sc->sc_irs + 1) { 888 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 889 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " 890 "rejected\n", s, __func__, th->th_seq, sc->sc_irs); 891 goto failed; 892 } 893 894 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { 895 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 896 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 897 "segment rejected\n", s, __func__); 898 goto failed; 899 } 900 /* 901 * If timestamps were negotiated the reflected timestamp 902 * must be equal to what we actually sent in the SYN|ACK. 903 */ 904 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) { 905 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 906 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, " 907 "segment rejected\n", 908 s, __func__, to->to_tsecr, sc->sc_ts); 909 goto failed; 910 } 911 912 *lsop = syncache_socket(sc, *lsop, m); 913 914 if (*lsop == NULL) 915 tcpstat.tcps_sc_aborted++; 916 else 917 tcpstat.tcps_sc_completed++; 918 919 if (sc != &scs) 920 syncache_free(sc); 921 return (1); 922failed: 923 if (sc != NULL && sc != &scs) 924 syncache_free(sc); 925 if (s != NULL) 926 free(s, M_TCPLOG); 927 *lsop = NULL; 928 return (0); 929} 930 931/* 932 * Given a LISTEN socket and an inbound SYN request, add 933 * this to the syn cache, and send back a segment: 934 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 935 * to the source. 936 * 937 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 938 * Doing so would require that we hold onto the data and deliver it 939 * to the application. However, if we are the target of a SYN-flood 940 * DoS attack, an attacker could send data which would eventually 941 * consume all available buffer space if it were ACKed. By not ACKing 942 * the data, we avoid this DoS scenario. 943 */ 944void 945syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 946 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 947{ 948 struct tcpcb *tp; 949 struct socket *so; 950 struct syncache *sc = NULL; 951 struct syncache_head *sch; 952 struct mbuf *ipopts = NULL; 953 u_int32_t flowtmp; 954 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 955 char *s; 956#ifdef INET6 957 int autoflowlabel = 0; 958#endif 959#ifdef MAC 960 struct label *maclabel; 961#endif 962 struct syncache scs; 963 964 INP_INFO_WLOCK_ASSERT(&tcbinfo); 965 INP_LOCK_ASSERT(inp); /* listen socket */ 966 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, 967 ("%s: unexpected tcp flags", __func__)); 968 969 /* 970 * Combine all so/tp operations very early to drop the INP lock as 971 * soon as possible. 972 */ 973 so = *lsop; 974 tp = sototcpcb(so); 975 976#ifdef INET6 977 if (inc->inc_isipv6 && 978 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 979 autoflowlabel = 1; 980#endif 981 ip_ttl = inp->inp_ip_ttl; 982 ip_tos = inp->inp_ip_tos; 983 win = sbspace(&so->so_rcv); 984 sb_hiwat = so->so_rcv.sb_hiwat; 985 noopt = (tp->t_flags & TF_NOOPT); 986 987 so = NULL; 988 tp = NULL; 989 990#ifdef MAC 991 if (mac_syncache_init(&maclabel) != 0) { 992 INP_UNLOCK(inp); 993 INP_INFO_WUNLOCK(&tcbinfo); 994 goto done; 995 } else 996 mac_syncache_create(maclabel, inp); 997#endif 998 INP_UNLOCK(inp); 999 INP_INFO_WUNLOCK(&tcbinfo); 1000 1001 /* 1002 * Remember the IP options, if any. 1003 */ 1004#ifdef INET6 1005 if (!inc->inc_isipv6) 1006#endif 1007 ipopts = ip_srcroute(m); 1008 1009 /* 1010 * See if we already have an entry for this connection. 1011 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1012 * 1013 * XXX: should the syncache be re-initialized with the contents 1014 * of the new SYN here (which may have different options?) 1015 * 1016 * XXX: We do not check the sequence number to see if this is a 1017 * real retransmit or a new connection attempt. The question is 1018 * how to handle such a case; either ignore it as spoofed, or 1019 * drop the current entry and create a new one? 1020 */ 1021 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 1022 SCH_LOCK_ASSERT(sch); 1023 if (sc != NULL) { 1024 tcpstat.tcps_sc_dupsyn++; 1025 if (ipopts) { 1026 /* 1027 * If we were remembering a previous source route, 1028 * forget it and use the new one we've been given. 1029 */ 1030 if (sc->sc_ipopts) 1031 (void) m_free(sc->sc_ipopts); 1032 sc->sc_ipopts = ipopts; 1033 } 1034 /* 1035 * Update timestamp if present. 1036 */ 1037 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 1038 sc->sc_tsreflect = to->to_tsval; 1039 else 1040 sc->sc_flags &= ~SCF_TIMESTAMP; 1041#ifdef MAC 1042 /* 1043 * Since we have already unconditionally allocated label 1044 * storage, free it up. The syncache entry will already 1045 * have an initialized label we can use. 1046 */ 1047 mac_syncache_destroy(&maclabel); 1048 KASSERT(sc->sc_label != NULL, 1049 ("%s: label not initialized", __func__)); 1050#endif 1051 /* Retransmit SYN|ACK and reset retransmit count. */ 1052 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { 1053 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " 1054 "resetting timer and retransmitting SYN|ACK\n", 1055 s, __func__); 1056 free(s, M_TCPLOG); 1057 } 1058 if (syncache_respond(sc) == 0) { 1059 sc->sc_rxmits = 0; 1060 syncache_timeout(sc, sch, 1); 1061 tcpstat.tcps_sndacks++; 1062 tcpstat.tcps_sndtotal++; 1063 } 1064 SCH_UNLOCK(sch); 1065 goto done; 1066 } 1067 1068 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1069 if (sc == NULL) { 1070 /* 1071 * The zone allocator couldn't provide more entries. 1072 * Treat this as if the cache was full; drop the oldest 1073 * entry and insert the new one. 1074 */ 1075 tcpstat.tcps_sc_zonefail++; 1076 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 1077 syncache_drop(sc, sch); 1078 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1079 if (sc == NULL) { 1080 if (tcp_syncookies) { 1081 bzero(&scs, sizeof(scs)); 1082 sc = &scs; 1083 } else { 1084 SCH_UNLOCK(sch); 1085 if (ipopts) 1086 (void) m_free(ipopts); 1087 goto done; 1088 } 1089 } 1090 } 1091 1092 /* 1093 * Fill in the syncache values. 1094 */ 1095#ifdef MAC 1096 sc->sc_label = maclabel; 1097#endif 1098 sc->sc_ipopts = ipopts; 1099 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1100#ifdef INET6 1101 if (!inc->inc_isipv6) 1102#endif 1103 { 1104 sc->sc_ip_tos = ip_tos; 1105 sc->sc_ip_ttl = ip_ttl; 1106 } 1107 1108 sc->sc_irs = th->th_seq; 1109 sc->sc_iss = arc4random(); 1110 sc->sc_flags = 0; 1111 sc->sc_flowlabel = 0; 1112 1113 /* 1114 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1115 * win was derived from socket earlier in the function. 1116 */ 1117 win = imax(win, 0); 1118 win = imin(win, TCP_MAXWIN); 1119 sc->sc_wnd = win; 1120 1121 if (tcp_do_rfc1323) { 1122 /* 1123 * A timestamp received in a SYN makes 1124 * it ok to send timestamp requests and replies. 1125 */ 1126 if (to->to_flags & TOF_TS) { 1127 sc->sc_tsreflect = to->to_tsval; 1128 sc->sc_ts = ticks; 1129 sc->sc_flags |= SCF_TIMESTAMP; 1130 } 1131 if (to->to_flags & TOF_SCALE) { 1132 int wscale = 0; 1133 1134 /* 1135 * Pick the smallest possible scaling factor that 1136 * will still allow us to scale up to sb_max, aka 1137 * kern.ipc.maxsockbuf. 1138 * 1139 * We do this because there are broken firewalls that 1140 * will corrupt the window scale option, leading to 1141 * the other endpoint believing that our advertised 1142 * window is unscaled. At scale factors larger than 1143 * 5 the unscaled window will drop below 1500 bytes, 1144 * leading to serious problems when traversing these 1145 * broken firewalls. 1146 * 1147 * With the default maxsockbuf of 256K, a scale factor 1148 * of 3 will be chosen by this algorithm. Those who 1149 * choose a larger maxsockbuf should watch out 1150 * for the compatiblity problems mentioned above. 1151 * 1152 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1153 * or <SYN,ACK>) segment itself is never scaled. 1154 */ 1155 while (wscale < TCP_MAX_WINSHIFT && 1156 (TCP_MAXWIN << wscale) < sb_max) 1157 wscale++; 1158 sc->sc_requested_r_scale = wscale; 1159 sc->sc_requested_s_scale = to->to_wscale; 1160 sc->sc_flags |= SCF_WINSCALE; 1161 } 1162 } 1163#ifdef TCP_SIGNATURE 1164 /* 1165 * If listening socket requested TCP digests, and received SYN 1166 * contains the option, flag this in the syncache so that 1167 * syncache_respond() will do the right thing with the SYN+ACK. 1168 * XXX: Currently we always record the option by default and will 1169 * attempt to use it in syncache_respond(). 1170 */ 1171 if (to->to_flags & TOF_SIGNATURE) 1172 sc->sc_flags |= SCF_SIGNATURE; 1173#endif 1174 if (to->to_flags & TOF_SACKPERM) 1175 sc->sc_flags |= SCF_SACK; 1176 if (to->to_flags & TOF_MSS) 1177 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1178 if (noopt) 1179 sc->sc_flags |= SCF_NOOPT; 1180 1181 if (tcp_syncookies) { 1182 syncookie_generate(sch, sc, &flowtmp); 1183#ifdef INET6 1184 if (autoflowlabel) 1185 sc->sc_flowlabel = flowtmp; 1186#endif 1187 } else { 1188#ifdef INET6 1189 if (autoflowlabel) 1190 sc->sc_flowlabel = 1191 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1192#endif 1193 } 1194 SCH_UNLOCK(sch); 1195 1196 /* 1197 * Do a standard 3-way handshake. 1198 */ 1199 if (syncache_respond(sc) == 0) { 1200 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs) 1201 syncache_free(sc); 1202 else if (sc != &scs) 1203 syncache_insert(sc, sch); /* locks and unlocks sch */ 1204 tcpstat.tcps_sndacks++; 1205 tcpstat.tcps_sndtotal++; 1206 } else { 1207 if (sc != &scs) 1208 syncache_free(sc); 1209 tcpstat.tcps_sc_dropped++; 1210 } 1211 1212done: 1213#ifdef MAC 1214 if (sc == &scs) 1215 mac_syncache_destroy(&maclabel); 1216#endif 1217 *lsop = NULL; 1218 m_freem(m); 1219 return; 1220} 1221 1222static int 1223syncache_respond(struct syncache *sc) 1224{ 1225 struct ip *ip = NULL; 1226 struct mbuf *m; 1227 struct tcphdr *th; 1228 int optlen, error; 1229 u_int16_t hlen, tlen, mssopt; 1230 struct tcpopt to; 1231#ifdef INET6 1232 struct ip6_hdr *ip6 = NULL; 1233#endif 1234 1235 hlen = 1236#ifdef INET6 1237 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1238#endif 1239 sizeof(struct ip); 1240 tlen = hlen + sizeof(struct tcphdr); 1241 1242 /* Determine MSS we advertize to other end of connection. */ 1243 mssopt = tcp_mssopt(&sc->sc_inc); 1244 if (sc->sc_peer_mss) 1245 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss); 1246 1247 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1248 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, 1249 ("syncache: mbuf too small")); 1250 1251 /* Create the IP+TCP header from scratch. */ 1252 m = m_gethdr(M_DONTWAIT, MT_DATA); 1253 if (m == NULL) 1254 return (ENOBUFS); 1255#ifdef MAC 1256 mac_syncache_create_mbuf(sc->sc_label, m); 1257#endif 1258 m->m_data += max_linkhdr; 1259 m->m_len = tlen; 1260 m->m_pkthdr.len = tlen; 1261 m->m_pkthdr.rcvif = NULL; 1262 1263#ifdef INET6 1264 if (sc->sc_inc.inc_isipv6) { 1265 ip6 = mtod(m, struct ip6_hdr *); 1266 ip6->ip6_vfc = IPV6_VERSION; 1267 ip6->ip6_nxt = IPPROTO_TCP; 1268 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1269 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1270 ip6->ip6_plen = htons(tlen - hlen); 1271 /* ip6_hlim is set after checksum */ 1272 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1273 ip6->ip6_flow |= sc->sc_flowlabel; 1274 1275 th = (struct tcphdr *)(ip6 + 1); 1276 } else 1277#endif 1278 { 1279 ip = mtod(m, struct ip *); 1280 ip->ip_v = IPVERSION; 1281 ip->ip_hl = sizeof(struct ip) >> 2; 1282 ip->ip_len = tlen; 1283 ip->ip_id = 0; 1284 ip->ip_off = 0; 1285 ip->ip_sum = 0; 1286 ip->ip_p = IPPROTO_TCP; 1287 ip->ip_src = sc->sc_inc.inc_laddr; 1288 ip->ip_dst = sc->sc_inc.inc_faddr; 1289 ip->ip_ttl = sc->sc_ip_ttl; 1290 ip->ip_tos = sc->sc_ip_tos; 1291 1292 /* 1293 * See if we should do MTU discovery. Route lookups are 1294 * expensive, so we will only unset the DF bit if: 1295 * 1296 * 1) path_mtu_discovery is disabled 1297 * 2) the SCF_UNREACH flag has been set 1298 */ 1299 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1300 ip->ip_off |= IP_DF; 1301 1302 th = (struct tcphdr *)(ip + 1); 1303 } 1304 th->th_sport = sc->sc_inc.inc_lport; 1305 th->th_dport = sc->sc_inc.inc_fport; 1306 1307 th->th_seq = htonl(sc->sc_iss); 1308 th->th_ack = htonl(sc->sc_irs + 1); 1309 th->th_off = sizeof(struct tcphdr) >> 2; 1310 th->th_x2 = 0; 1311 th->th_flags = TH_SYN|TH_ACK; 1312 th->th_win = htons(sc->sc_wnd); 1313 th->th_urp = 0; 1314 1315 /* Tack on the TCP options. */ 1316 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1317 to.to_flags = 0; 1318 1319 to.to_mss = mssopt; 1320 to.to_flags = TOF_MSS; 1321 if (sc->sc_flags & SCF_WINSCALE) { 1322 to.to_wscale = sc->sc_requested_r_scale; 1323 to.to_flags |= TOF_SCALE; 1324 } 1325 if (sc->sc_flags & SCF_TIMESTAMP) { 1326 /* Virgin timestamp or TCP cookie enhanced one. */ 1327 to.to_tsval = sc->sc_ts; 1328 to.to_tsecr = sc->sc_tsreflect; 1329 to.to_flags |= TOF_TS; 1330 } 1331 if (sc->sc_flags & SCF_SACK) 1332 to.to_flags |= TOF_SACKPERM; 1333#ifdef TCP_SIGNATURE 1334 if (sc->sc_flags & SCF_SIGNATURE) 1335 to.to_flags |= TOF_SIGNATURE; 1336#endif 1337 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1338 1339 /* Adjust headers by option size. */ 1340 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1341 m->m_len += optlen; 1342 m->m_pkthdr.len += optlen; 1343 1344#ifdef TCP_SIGNATURE 1345 if (sc->sc_flags & SCF_SIGNATURE) 1346 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1347 to.to_signature, IPSEC_DIR_OUTBOUND); 1348#endif 1349#ifdef INET6 1350 if (sc->sc_inc.inc_isipv6) 1351 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1352 else 1353#endif 1354 ip->ip_len += optlen; 1355 } else 1356 optlen = 0; 1357 1358#ifdef INET6 1359 if (sc->sc_inc.inc_isipv6) { 1360 th->th_sum = 0; 1361 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1362 tlen + optlen - hlen); 1363 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1364 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1365 } else 1366#endif 1367 { 1368 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1369 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1370 m->m_pkthdr.csum_flags = CSUM_TCP; 1371 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1372 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1373 } 1374 return (error); 1375} 1376 1377/* 1378 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1379 * receive and to be able to handle SYN floods from bogus source addresses 1380 * (where we will never receive any reply). SYN floods try to exhaust all 1381 * our memory and available slots in the SYN cache table to cause a denial 1382 * of service to legitimate users of the local host. 1383 * 1384 * The idea of SYN cookies is to encode and include all necessary information 1385 * about the connection setup state within the SYN-ACK we send back and thus 1386 * to get along without keeping any local state until the ACK to the SYN-ACK 1387 * arrives (if ever). Everything we need to know should be available from 1388 * the information we encoded in the SYN-ACK. 1389 * 1390 * More information about the theory behind SYN cookies and its first 1391 * discussion and specification can be found at: 1392 * http://cr.yp.to/syncookies.html (overview) 1393 * http://cr.yp.to/syncookies/archive (gory details) 1394 * 1395 * This implementation extends the orginal idea and first implementation 1396 * of FreeBSD by using not only the initial sequence number field to store 1397 * information but also the timestamp field if present. This way we can 1398 * keep track of the entire state we need to know to recreate the session in 1399 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1400 * these days. For those that do not we still have to live with the known 1401 * shortcomings of the ISN only SYN cookies. 1402 * 1403 * Cookie layers: 1404 * 1405 * Initial sequence number we send: 1406 * 31|................................|0 1407 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1408 * D = MD5 Digest (first dword) 1409 * M = MSS index 1410 * R = Rotation of secret 1411 * P = Odd or Even secret 1412 * 1413 * The MD5 Digest is computed with over following parameters: 1414 * a) randomly rotated secret 1415 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1416 * c) the received initial sequence number from remote host 1417 * d) the rotation offset and odd/even bit 1418 * 1419 * Timestamp we send: 1420 * 31|................................|0 1421 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1422 * D = MD5 Digest (third dword) (only as filler) 1423 * S = Requested send window scale 1424 * R = Requested receive window scale 1425 * A = SACK allowed 1426 * 5 = TCP-MD5 enabled (not implemented yet) 1427 * XORed with MD5 Digest (forth dword) 1428 * 1429 * The timestamp isn't cryptographically secure and doesn't need to be. 1430 * The double use of the MD5 digest dwords ties it to a specific remote/ 1431 * local host/port, remote initial sequence number and our local time 1432 * limited secret. A received timestamp is reverted (XORed) and then 1433 * the contained MD5 dword is compared to the computed one to ensure the 1434 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1435 * have been tampered with but this isn't different from supplying bogus 1436 * values in the SYN in the first place. 1437 * 1438 * Some problems with SYN cookies remain however: 1439 * Consider the problem of a recreated (and retransmitted) cookie. If the 1440 * original SYN was accepted, the connection is established. The second 1441 * SYN is inflight, and if it arrives with an ISN that falls within the 1442 * receive window, the connection is killed. 1443 * 1444 * Notes: 1445 * A heuristic to determine when to accept syn cookies is not necessary. 1446 * An ACK flood would cause the syncookie verification to be attempted, 1447 * but a SYN flood causes syncookies to be generated. Both are of equal 1448 * cost, so there's no point in trying to optimize the ACK flood case. 1449 * Also, if you don't process certain ACKs for some reason, then all someone 1450 * would have to do is launch a SYN and ACK flood at the same time, which 1451 * would stop cookie verification and defeat the entire purpose of syncookies. 1452 */ 1453static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1454 1455static void 1456syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1457 u_int32_t *flowlabel) 1458{ 1459 MD5_CTX ctx; 1460 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1461 u_int32_t data; 1462 u_int32_t *secbits; 1463 u_int off, pmss, mss; 1464 int i; 1465 1466 SCH_LOCK_ASSERT(sch); 1467 1468 /* Which of the two secrets to use. */ 1469 secbits = sch->sch_oddeven ? 1470 sch->sch_secbits_odd : sch->sch_secbits_even; 1471 1472 /* Reseed secret if too old. */ 1473 if (sch->sch_reseed < time_uptime) { 1474 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1475 secbits = sch->sch_oddeven ? 1476 sch->sch_secbits_odd : sch->sch_secbits_even; 1477 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1478 secbits[i] = arc4random(); 1479 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1480 } 1481 1482 /* Secret rotation offset. */ 1483 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1484 1485 /* Maximum segment size calculation. */ 1486 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss); 1487 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1488 if (tcp_sc_msstab[mss] <= pmss) 1489 break; 1490 1491 /* Fold parameters and MD5 digest into the ISN we will send. */ 1492 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1493 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1494 data |= mss << 4; /* mss, 3 bits */ 1495 1496 MD5Init(&ctx); 1497 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1498 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1499 MD5Update(&ctx, secbits, off); 1500 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1501 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1502 MD5Update(&ctx, &data, sizeof(data)); 1503 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1504 1505 data |= (md5_buffer[0] << 7); 1506 sc->sc_iss = data; 1507 1508#ifdef INET6 1509 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1510#endif 1511 1512 /* Additional parameters are stored in the timestamp if present. */ 1513 if (sc->sc_flags & SCF_TIMESTAMP) { 1514 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1515 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1516 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1517 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1518 data |= md5_buffer[2] << 10; /* more digest bits */ 1519 data ^= md5_buffer[3]; 1520 sc->sc_ts = data; 1521 sc->sc_tsoff = data - ticks; /* after XOR */ 1522 } 1523 1524 tcpstat.tcps_sc_sendcookie++; 1525 return; 1526} 1527 1528static struct syncache * 1529syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1530 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1531 struct socket *so) 1532{ 1533 MD5_CTX ctx; 1534 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1535 u_int32_t data = 0; 1536 u_int32_t *secbits; 1537 tcp_seq ack, seq; 1538 int off, mss, wnd, flags; 1539 1540 SCH_LOCK_ASSERT(sch); 1541 1542 /* 1543 * Pull information out of SYN-ACK/ACK and 1544 * revert sequence number advances. 1545 */ 1546 ack = th->th_ack - 1; 1547 seq = th->th_seq - 1; 1548 off = (ack >> 1) & 0x7; 1549 mss = (ack >> 4) & 0x7; 1550 flags = ack & 0x7f; 1551 1552 /* Which of the two secrets to use. */ 1553 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1554 1555 /* 1556 * The secret wasn't updated for the lifetime of a syncookie, 1557 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1558 */ 1559 if (sch->sch_reseed < time_uptime) { 1560 return (NULL); 1561 } 1562 1563 /* Recompute the digest so we can compare it. */ 1564 MD5Init(&ctx); 1565 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1566 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1567 MD5Update(&ctx, secbits, off); 1568 MD5Update(&ctx, inc, sizeof(*inc)); 1569 MD5Update(&ctx, &seq, sizeof(seq)); 1570 MD5Update(&ctx, &flags, sizeof(flags)); 1571 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1572 1573 /* Does the digest part of or ACK'ed ISS match? */ 1574 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1575 return (NULL); 1576 1577 /* Does the digest part of our reflected timestamp match? */ 1578 if (to->to_flags & TOF_TS) { 1579 data = md5_buffer[3] ^ to->to_tsecr; 1580 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1581 return (NULL); 1582 } 1583 1584 /* Fill in the syncache values. */ 1585 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1586 sc->sc_ipopts = NULL; 1587 1588 sc->sc_irs = seq; 1589 sc->sc_iss = ack; 1590 1591#ifdef INET6 1592 if (inc->inc_isipv6) { 1593 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1594 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1595 } else 1596#endif 1597 { 1598 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1599 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1600 } 1601 1602 /* Additional parameters that were encoded in the timestamp. */ 1603 if (data) { 1604 sc->sc_flags |= SCF_TIMESTAMP; 1605 sc->sc_tsreflect = to->to_tsval; 1606 sc->sc_ts = to->to_tsecr; 1607 sc->sc_tsoff = to->to_tsecr - ticks; 1608 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1609 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1610 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1611 TCP_MAX_WINSHIFT); 1612 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1613 TCP_MAX_WINSHIFT); 1614 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1615 sc->sc_flags |= SCF_WINSCALE; 1616 } else 1617 sc->sc_flags |= SCF_NOOPT; 1618 1619 wnd = sbspace(&so->so_rcv); 1620 wnd = imax(wnd, 0); 1621 wnd = imin(wnd, TCP_MAXWIN); 1622 sc->sc_wnd = wnd; 1623 1624 sc->sc_rxmits = 0; 1625 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1626 1627 tcpstat.tcps_sc_recvcookie++; 1628 return (sc); 1629} 1630 1631/* 1632 * Returns the current number of syncache entries. This number 1633 * will probably change before you get around to calling 1634 * syncache_pcblist. 1635 */ 1636 1637int 1638syncache_pcbcount(void) 1639{ 1640 struct syncache_head *sch; 1641 int count, i; 1642 1643 for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) { 1644 /* No need to lock for a read. */ 1645 sch = &tcp_syncache.hashbase[i]; 1646 count += sch->sch_length; 1647 } 1648 return count; 1649} 1650 1651/* 1652 * Exports the syncache entries to userland so that netstat can display 1653 * them alongside the other sockets. This function is intended to be 1654 * called only from tcp_pcblist. 1655 * 1656 * Due to concurrency on an active system, the number of pcbs exported 1657 * may have no relation to max_pcbs. max_pcbs merely indicates the 1658 * amount of space the caller allocated for this function to use. 1659 */ 1660int 1661syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported) 1662{ 1663 struct xtcpcb xt; 1664 struct syncache *sc; 1665 struct syncache_head *sch; 1666 int count, error, i; 1667 1668 for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) { 1669 sch = &tcp_syncache.hashbase[i]; 1670 SCH_LOCK(sch); 1671 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 1672 if (count >= max_pcbs) { 1673 SCH_UNLOCK(sch); 1674 goto exit; 1675 } 1676 bzero(&xt, sizeof(xt)); 1677 xt.xt_len = sizeof(xt); 1678 if (sc->sc_inc.inc_isipv6) 1679 xt.xt_inp.inp_vflag = INP_IPV6; 1680 else 1681 xt.xt_inp.inp_vflag = INP_IPV4; 1682 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); 1683 xt.xt_tp.t_inpcb = &xt.xt_inp; 1684 xt.xt_tp.t_state = TCPS_SYN_RECEIVED; 1685 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1686 xt.xt_socket.xso_len = sizeof (struct xsocket); 1687 xt.xt_socket.so_type = SOCK_STREAM; 1688 xt.xt_socket.so_state = SS_ISCONNECTING; 1689 error = SYSCTL_OUT(req, &xt, sizeof xt); 1690 if (error) { 1691 SCH_UNLOCK(sch); 1692 goto exit; 1693 } 1694 count++; 1695 } 1696 SCH_UNLOCK(sch); 1697 } 1698exit: 1699 *pcbs_exported = count; 1700 return error; 1701} 1702 1703