tcp_syncache.c revision 159722
1/*- 2 * Copyright (c) 2001 McAfee, Inc. 3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Jonathan Lemon 7 * and McAfee Research, the Security Research Division of McAfee, Inc. under 8 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 9 * DARPA CHATS research program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: head/sys/netinet/tcp_syncache.c 159722 2006-06-18 11:48:03Z andre $ 33 */ 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_ipsec.h" 38#include "opt_mac.h" 39#include "opt_tcpdebug.h" 40#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/kernel.h> 45#include <sys/sysctl.h> 46#include <sys/lock.h> 47#include <sys/mutex.h> 48#include <sys/malloc.h> 49#include <sys/mac.h> 50#include <sys/mbuf.h> 51#include <sys/md5.h> 52#include <sys/proc.h> /* for proc0 declaration */ 53#include <sys/random.h> 54#include <sys/rwlock.h> 55#include <sys/socket.h> 56#include <sys/socketvar.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#ifdef TCPDEBUG 77#include <netinet/tcpip.h> 78#endif 79#include <netinet/tcp_fsm.h> 80#include <netinet/tcp_seq.h> 81#include <netinet/tcp_timer.h> 82#include <netinet/tcp_var.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif 86#ifdef INET6 87#include <netinet6/tcp6_var.h> 88#endif 89 90#ifdef IPSEC 91#include <netinet6/ipsec.h> 92#ifdef INET6 93#include <netinet6/ipsec6.h> 94#endif 95#endif /*IPSEC*/ 96 97#ifdef FAST_IPSEC 98#include <netipsec/ipsec.h> 99#ifdef INET6 100#include <netipsec/ipsec6.h> 101#endif 102#include <netipsec/key.h> 103#endif /*FAST_IPSEC*/ 104 105#include <machine/in_cksum.h> 106#include <vm/uma.h> 107 108static int tcp_syncookies = 1; 109SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 110 &tcp_syncookies, 0, 111 "Use TCP SYN cookies if the syncache overflows"); 112 113static void syncache_drop(struct syncache *, struct syncache_head *); 114static void syncache_free(struct syncache *); 115static void syncache_insert(struct syncache *, struct syncache_head *); 116struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 117static int syncache_respond(struct syncache *, struct mbuf *); 118static struct socket *syncache_socket(struct syncache *, struct socket *, 119 struct mbuf *m); 120static void syncache_timer(void *); 121static void syncookie_init(void); 122static u_int32_t syncookie_generate(struct syncache *, u_int32_t *); 123static struct syncache 124 *syncookie_lookup(struct in_conninfo *, struct tcphdr *, 125 struct socket *); 126 127/* 128 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 129 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 130 * the odds are that the user has given up attempting to connect by then. 131 */ 132#define SYNCACHE_MAXREXMTS 3 133 134/* Arbitrary values */ 135#define TCP_SYNCACHE_HASHSIZE 512 136#define TCP_SYNCACHE_BUCKETLIMIT 30 137 138struct tcp_syncache { 139 struct syncache_head *hashbase; 140 uma_zone_t zone; 141 u_int hashsize; 142 u_int hashmask; 143 u_int bucket_limit; 144 u_int cache_count; /* XXX: unprotected */ 145 u_int cache_limit; 146 u_int rexmt_limit; 147 u_int hash_secret; 148}; 149static struct tcp_syncache tcp_syncache; 150 151SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 152 153SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 154 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 155 156SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 157 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 158 159SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 160 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 161 162SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 163 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 164 165SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 166 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 167 168static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 169 170#define SYNCACHE_HASH(inc, mask) \ 171 ((tcp_syncache.hash_secret ^ \ 172 (inc)->inc_faddr.s_addr ^ \ 173 ((inc)->inc_faddr.s_addr >> 16) ^ \ 174 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 175 176#define SYNCACHE_HASH6(inc, mask) \ 177 ((tcp_syncache.hash_secret ^ \ 178 (inc)->inc6_faddr.s6_addr32[0] ^ \ 179 (inc)->inc6_faddr.s6_addr32[3] ^ \ 180 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 181 182#define ENDPTS_EQ(a, b) ( \ 183 (a)->ie_fport == (b)->ie_fport && \ 184 (a)->ie_lport == (b)->ie_lport && \ 185 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 186 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 187) 188 189#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 190 191#define SYNCACHE_TIMEOUT(sc, sch, co) do { \ 192 (sc)->sc_rxmits++; \ 193 (sc)->sc_rxttime = ticks + \ 194 TCPTV_RTOBASE * tcp_backoff[(sc)->sc_rxmits - 1]; \ 195 if ((sch)->sch_nextc > (sc)->sc_rxttime) \ 196 (sch)->sch_nextc = (sc)->sc_rxttime; \ 197 if (!TAILQ_EMPTY(&(sch)->sch_bucket) && !(co)) \ 198 callout_reset(&(sch)->sch_timer, \ 199 (sch)->sch_nextc - ticks, \ 200 syncache_timer, (void *)(sch)); \ 201} while (0) 202 203#define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 204#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 205#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 206 207/* 208 * Requires the syncache entry to be already removed from the bucket list. 209 */ 210static void 211syncache_free(struct syncache *sc) 212{ 213 if (sc->sc_ipopts) 214 (void) m_free(sc->sc_ipopts); 215 216 uma_zfree(tcp_syncache.zone, sc); 217} 218 219void 220syncache_init(void) 221{ 222 int i; 223 224 tcp_syncache.cache_count = 0; 225 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 226 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 227 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 228 tcp_syncache.hash_secret = arc4random(); 229 230 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 231 &tcp_syncache.hashsize); 232 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 233 &tcp_syncache.bucket_limit); 234 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 235 printf("WARNING: syncache hash size is not a power of 2.\n"); 236 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 237 } 238 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 239 240 /* Set limits. */ 241 tcp_syncache.cache_limit = 242 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 243 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 244 &tcp_syncache.cache_limit); 245 246 /* Allocate the hash table. */ 247 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 248 tcp_syncache.hashsize * sizeof(struct syncache_head), 249 M_SYNCACHE, M_WAITOK); 250 251 /* Initialize the hash buckets. */ 252 for (i = 0; i < tcp_syncache.hashsize; i++) { 253 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 254 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 255 NULL, MTX_DEF); 256 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer, 257 &tcp_syncache.hashbase[i].sch_mtx, 0); 258 tcp_syncache.hashbase[i].sch_length = 0; 259 } 260 261 syncookie_init(); 262 263 /* Create the syncache entry zone. */ 264 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 265 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 266 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 267} 268 269/* 270 * Inserts a syncache entry into the specified bucket row. 271 * Locks and unlocks the syncache_head autonomously. 272 */ 273static void 274syncache_insert(struct syncache *sc, struct syncache_head *sch) 275{ 276 struct syncache *sc2; 277 278 SCH_LOCK(sch); 279 280 /* 281 * Make sure that we don't overflow the per-bucket limit. 282 * If the bucket is full, toss the oldest element. 283 */ 284 if (sch->sch_length >= tcp_syncache.bucket_limit) { 285 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 286 ("sch->sch_length incorrect")); 287 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 288 syncache_drop(sc2, sch); 289 tcpstat.tcps_sc_bucketoverflow++; 290 } 291 292 /* Put it into the bucket. */ 293 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 294 sch->sch_length++; 295 296 /* Reinitialize the bucket row's timer. */ 297 SYNCACHE_TIMEOUT(sc, sch, 1); 298 299 SCH_UNLOCK(sch); 300 301 tcp_syncache.cache_count++; 302 tcpstat.tcps_sc_added++; 303} 304 305/* 306 * Remove and free entry from syncache bucket row. 307 * Expects locked syncache head. 308 */ 309static void 310syncache_drop(struct syncache *sc, struct syncache_head *sch) 311{ 312 313 SCH_LOCK_ASSERT(sch); 314 315 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 316 sch->sch_length--; 317 318 syncache_free(sc); 319 tcp_syncache.cache_count--; 320} 321 322/* 323 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 324 * If we have retransmitted an entry the maximum number of times, expire it. 325 * One separate timer for each bucket row. 326 */ 327static void 328syncache_timer(void *xsch) 329{ 330 struct syncache_head *sch = (struct syncache_head *)xsch; 331 struct syncache *sc, *nsc; 332 int tick = ticks; 333 334 /* NB: syncache_head has already been locked by the callout. */ 335 SCH_LOCK_ASSERT(sch); 336 337 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 338 /* 339 * We do not check if the listen socket still exists 340 * and accept the case where the listen socket may be 341 * gone by the time we resend the SYN/ACK. We do 342 * not expect this to happens often. If it does, 343 * then the RST will be sent by the time the remote 344 * host does the SYN/ACK->ACK. 345 */ 346 if (sc->sc_rxttime >= tick) { 347 if (sc->sc_rxttime < sch->sch_nextc) 348 sch->sch_nextc = sc->sc_rxttime; 349 continue; 350 } 351 352 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) { 353 syncache_drop(sc, sch); 354 tcpstat.tcps_sc_stale++; 355 continue; 356 } 357 358 (void) syncache_respond(sc, NULL); 359 tcpstat.tcps_sc_retransmitted++; 360 SYNCACHE_TIMEOUT(sc, sch, 0); 361 } 362 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 363 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 364 syncache_timer, (void *)(sch)); 365} 366 367/* 368 * Find an entry in the syncache. 369 * Returns always with locked syncache_head plus a matching entry or NULL. 370 */ 371struct syncache * 372syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 373{ 374 struct syncache *sc; 375 struct syncache_head *sch; 376 377#ifdef INET6 378 if (inc->inc_isipv6) { 379 sch = &tcp_syncache.hashbase[ 380 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 381 *schp = sch; 382 383 SCH_LOCK(sch); 384 385 /* Circle through bucket row to find matching entry. */ 386 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 387 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 388 return (sc); 389 } 390 } else 391#endif 392 { 393 sch = &tcp_syncache.hashbase[ 394 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 395 *schp = sch; 396 397 SCH_LOCK(sch); 398 399 /* Circle through bucket row to find matching entry. */ 400 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 401#ifdef INET6 402 if (sc->sc_inc.inc_isipv6) 403 continue; 404#endif 405 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 406 return (sc); 407 } 408 } 409 SCH_LOCK_ASSERT(*schp); 410 return (NULL); /* always returns with locked sch */ 411} 412 413/* 414 * This function is called when we get a RST for a 415 * non-existent connection, so that we can see if the 416 * connection is in the syn cache. If it is, zap it. 417 */ 418void 419syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 420{ 421 struct syncache *sc; 422 struct syncache_head *sch; 423 424 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 425 SCH_LOCK_ASSERT(sch); 426 if (sc == NULL) 427 goto done; 428 429 /* 430 * If the RST bit is set, check the sequence number to see 431 * if this is a valid reset segment. 432 * RFC 793 page 37: 433 * In all states except SYN-SENT, all reset (RST) segments 434 * are validated by checking their SEQ-fields. A reset is 435 * valid if its sequence number is in the window. 436 * 437 * The sequence number in the reset segment is normally an 438 * echo of our outgoing acknowlegement numbers, but some hosts 439 * send a reset with the sequence number at the rightmost edge 440 * of our receive window, and we have to handle this case. 441 */ 442 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 443 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 444 syncache_drop(sc, sch); 445 tcpstat.tcps_sc_reset++; 446 } 447done: 448 SCH_UNLOCK(sch); 449} 450 451void 452syncache_badack(struct in_conninfo *inc) 453{ 454 struct syncache *sc; 455 struct syncache_head *sch; 456 457 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 458 SCH_LOCK_ASSERT(sch); 459 if (sc != NULL) { 460 syncache_drop(sc, sch); 461 tcpstat.tcps_sc_badack++; 462 } 463 SCH_UNLOCK(sch); 464} 465 466void 467syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 468{ 469 struct syncache *sc; 470 struct syncache_head *sch; 471 472 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 473 SCH_LOCK_ASSERT(sch); 474 if (sc == NULL) 475 goto done; 476 477 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 478 if (ntohl(th->th_seq) != sc->sc_iss) 479 goto done; 480 481 /* 482 * If we've rertransmitted 3 times and this is our second error, 483 * we remove the entry. Otherwise, we allow it to continue on. 484 * This prevents us from incorrectly nuking an entry during a 485 * spurious network outage. 486 * 487 * See tcp_notify(). 488 */ 489 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 490 sc->sc_flags |= SCF_UNREACH; 491 goto done; 492 } 493 syncache_drop(sc, sch); 494 tcpstat.tcps_sc_unreach++; 495done: 496 SCH_UNLOCK(sch); 497} 498 499/* 500 * Build a new TCP socket structure from a syncache entry. 501 */ 502static struct socket * 503syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 504{ 505 struct inpcb *inp = NULL; 506 struct socket *so; 507 struct tcpcb *tp; 508 509 NET_ASSERT_GIANT(); 510 INP_INFO_WLOCK_ASSERT(&tcbinfo); 511 512 /* 513 * Ok, create the full blown connection, and set things up 514 * as they would have been set up if we had created the 515 * connection when the SYN arrived. If we can't create 516 * the connection, abort it. 517 */ 518 so = sonewconn(lso, SS_ISCONNECTED); 519 if (so == NULL) { 520 /* 521 * Drop the connection; we will send a RST if the peer 522 * retransmits the ACK, 523 */ 524 tcpstat.tcps_listendrop++; 525 goto abort2; 526 } 527#ifdef MAC 528 SOCK_LOCK(so); 529 mac_set_socket_peer_from_mbuf(m, so); 530 SOCK_UNLOCK(so); 531#endif 532 533 inp = sotoinpcb(so); 534 INP_LOCK(inp); 535 536 /* Insert new socket into PCB hash list. */ 537 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 538#ifdef INET6 539 if (sc->sc_inc.inc_isipv6) { 540 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 541 } else { 542 inp->inp_vflag &= ~INP_IPV6; 543 inp->inp_vflag |= INP_IPV4; 544#endif 545 inp->inp_laddr = sc->sc_inc.inc_laddr; 546#ifdef INET6 547 } 548#endif 549 inp->inp_lport = sc->sc_inc.inc_lport; 550 if (in_pcbinshash(inp) != 0) { 551 /* 552 * Undo the assignments above if we failed to 553 * put the PCB on the hash lists. 554 */ 555#ifdef INET6 556 if (sc->sc_inc.inc_isipv6) 557 inp->in6p_laddr = in6addr_any; 558 else 559#endif 560 inp->inp_laddr.s_addr = INADDR_ANY; 561 inp->inp_lport = 0; 562 goto abort; 563 } 564#ifdef IPSEC 565 /* Copy old policy into new socket's. */ 566 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 567 printf("syncache_expand: could not copy policy\n"); 568#endif 569#ifdef FAST_IPSEC 570 /* Copy old policy into new socket's. */ 571 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 572 printf("syncache_expand: could not copy policy\n"); 573#endif 574#ifdef INET6 575 if (sc->sc_inc.inc_isipv6) { 576 struct inpcb *oinp = sotoinpcb(lso); 577 struct in6_addr laddr6; 578 struct sockaddr_in6 sin6; 579 /* 580 * Inherit socket options from the listening socket. 581 * Note that in6p_inputopts are not (and should not be) 582 * copied, since it stores previously received options and is 583 * used to detect if each new option is different than the 584 * previous one and hence should be passed to a user. 585 * If we copied in6p_inputopts, a user would not be able to 586 * receive options just after calling the accept system call. 587 */ 588 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 589 if (oinp->in6p_outputopts) 590 inp->in6p_outputopts = 591 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 592 593 sin6.sin6_family = AF_INET6; 594 sin6.sin6_len = sizeof(sin6); 595 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 596 sin6.sin6_port = sc->sc_inc.inc_fport; 597 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 598 laddr6 = inp->in6p_laddr; 599 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 600 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 601 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 602 thread0.td_ucred)) { 603 inp->in6p_laddr = laddr6; 604 goto abort; 605 } 606 /* Override flowlabel from in6_pcbconnect. */ 607 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 608 inp->in6p_flowinfo |= sc->sc_flowlabel; 609 } else 610#endif 611 { 612 struct in_addr laddr; 613 struct sockaddr_in sin; 614 615 inp->inp_options = ip_srcroute(m); 616 if (inp->inp_options == NULL) { 617 inp->inp_options = sc->sc_ipopts; 618 sc->sc_ipopts = NULL; 619 } 620 621 sin.sin_family = AF_INET; 622 sin.sin_len = sizeof(sin); 623 sin.sin_addr = sc->sc_inc.inc_faddr; 624 sin.sin_port = sc->sc_inc.inc_fport; 625 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 626 laddr = inp->inp_laddr; 627 if (inp->inp_laddr.s_addr == INADDR_ANY) 628 inp->inp_laddr = sc->sc_inc.inc_laddr; 629 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 630 thread0.td_ucred)) { 631 inp->inp_laddr = laddr; 632 goto abort; 633 } 634 } 635 tp = intotcpcb(inp); 636 tp->t_state = TCPS_SYN_RECEIVED; 637 tp->iss = sc->sc_iss; 638 tp->irs = sc->sc_irs; 639 tcp_rcvseqinit(tp); 640 tcp_sendseqinit(tp); 641 tp->snd_wl1 = sc->sc_irs; 642 tp->rcv_up = sc->sc_irs + 1; 643 tp->rcv_wnd = sc->sc_wnd; 644 tp->rcv_adv += tp->rcv_wnd; 645 646 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 647 if (sc->sc_flags & SCF_NOOPT) 648 tp->t_flags |= TF_NOOPT; 649 if (sc->sc_flags & SCF_WINSCALE) { 650 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 651 tp->snd_scale = sc->sc_requested_s_scale; 652 tp->request_r_scale = sc->sc_request_r_scale; 653 } 654 if (sc->sc_flags & SCF_TIMESTAMP) { 655 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 656 tp->ts_recent = sc->sc_tsrecent; 657 tp->ts_recent_age = ticks; 658 } 659#ifdef TCP_SIGNATURE 660 if (sc->sc_flags & SCF_SIGNATURE) 661 tp->t_flags |= TF_SIGNATURE; 662#endif 663 if (sc->sc_flags & SCF_SACK) { 664 tp->sack_enable = 1; 665 tp->t_flags |= TF_SACK_PERMIT; 666 } 667 668 /* 669 * Set up MSS and get cached values from tcp_hostcache. 670 * This might overwrite some of the defaults we just set. 671 */ 672 tcp_mss(tp, sc->sc_peer_mss); 673 674 /* 675 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 676 */ 677 if (sc->sc_rxmits > 1) 678 tp->snd_cwnd = tp->t_maxseg; 679 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 680 681 INP_UNLOCK(inp); 682 683 tcpstat.tcps_accepts++; 684 return (so); 685 686abort: 687 INP_UNLOCK(inp); 688abort2: 689 if (so != NULL) 690 soabort(so); 691 return (NULL); 692} 693 694/* 695 * This function gets called when we receive an ACK for a 696 * socket in the LISTEN state. We look up the connection 697 * in the syncache, and if its there, we pull it out of 698 * the cache and turn it into a full-blown connection in 699 * the SYN-RECEIVED state. 700 */ 701int 702syncache_expand(struct in_conninfo *inc, struct tcphdr *th, 703 struct socket **lsop, struct mbuf *m) 704{ 705 struct syncache *sc; 706 struct syncache_head *sch; 707 struct socket *so; 708 709 /* 710 * Global TCP locks are held because we manipulate the PCB lists 711 * and create a new socket. 712 */ 713 INP_INFO_WLOCK_ASSERT(&tcbinfo); 714 715 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 716 SCH_LOCK_ASSERT(sch); 717 if (sc == NULL) { 718 /* 719 * There is no syncache entry, so see if this ACK is 720 * a returning syncookie. To do this, first: 721 * A. See if this socket has had a syncache entry dropped in 722 * the past. We don't want to accept a bogus syncookie 723 * if we've never received a SYN. 724 * B. check that the syncookie is valid. If it is, then 725 * cobble up a fake syncache entry, and return. 726 */ 727 SCH_UNLOCK(sch); 728 sch = NULL; 729 730 if (!tcp_syncookies) 731 goto failed; 732 sc = syncookie_lookup(inc, th, *lsop); 733 if (sc == NULL) 734 goto failed; 735 tcpstat.tcps_sc_recvcookie++; 736 } else { 737 /* Pull out the entry to unlock the bucket row. */ 738 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 739 sch->sch_length--; 740 SCH_UNLOCK(sch); 741 } 742 743 /* 744 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 745 */ 746 if (th->th_ack != sc->sc_iss + 1) 747 goto failed; 748 749 so = syncache_socket(sc, *lsop, m); 750 751 if (so == NULL) { 752#if 0 753resetandabort: 754 /* XXXjlemon check this - is this correct? */ 755 (void) tcp_respond(NULL, m, m, th, 756 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 757#endif 758 m_freem(m); /* XXX: only needed for above */ 759 tcpstat.tcps_sc_aborted++; 760 if (sch != NULL) { 761 syncache_insert(sc, sch); /* try again later */ 762 sc = NULL; 763 } 764 goto failed; 765 } else 766 tcpstat.tcps_sc_completed++; 767 *lsop = so; 768 769 syncache_free(sc); 770 return (1); 771failed: 772 if (sc != NULL) 773 syncache_free(sc); 774 return (0); 775} 776 777/* 778 * Given a LISTEN socket and an inbound SYN request, add 779 * this to the syn cache, and send back a segment: 780 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 781 * to the source. 782 * 783 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 784 * Doing so would require that we hold onto the data and deliver it 785 * to the application. However, if we are the target of a SYN-flood 786 * DoS attack, an attacker could send data which would eventually 787 * consume all available buffer space if it were ACKed. By not ACKing 788 * the data, we avoid this DoS scenario. 789 */ 790int 791syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 792 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 793{ 794 struct tcpcb *tp; 795 struct socket *so; 796 struct syncache *sc = NULL; 797 struct syncache_head *sch; 798 struct mbuf *ipopts = NULL; 799 u_int32_t flowtmp; 800 int win, sb_hiwat, ip_ttl, ip_tos; 801#ifdef INET6 802 int autoflowlabel = 0; 803#endif 804 805 INP_INFO_WLOCK_ASSERT(&tcbinfo); 806 INP_LOCK_ASSERT(inp); /* listen socket */ 807 808 /* 809 * Combine all so/tp operations very early to drop the INP lock as 810 * soon as possible. 811 */ 812 so = *lsop; 813 tp = sototcpcb(so); 814 815#ifdef INET6 816 if (inc->inc_isipv6 && 817 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 818 autoflowlabel = 1; 819#endif 820 ip_ttl = inp->inp_ip_ttl; 821 ip_tos = inp->inp_ip_tos; 822 win = sbspace(&so->so_rcv); 823 sb_hiwat = so->so_rcv.sb_hiwat; 824 if (tp->t_flags & TF_NOOPT) 825 sc->sc_flags = SCF_NOOPT; 826 827 so = NULL; 828 tp = NULL; 829 830 INP_UNLOCK(inp); 831 INP_INFO_WUNLOCK(&tcbinfo); 832 833 /* 834 * Remember the IP options, if any. 835 */ 836#ifdef INET6 837 if (!inc->inc_isipv6) 838#endif 839 ipopts = ip_srcroute(m); 840 841 /* 842 * See if we already have an entry for this connection. 843 * If we do, resend the SYN,ACK, and reset the retransmit timer. 844 * 845 * XXX: should the syncache be re-initialized with the contents 846 * of the new SYN here (which may have different options?) 847 */ 848 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 849 SCH_LOCK_ASSERT(sch); 850 if (sc != NULL) { 851 tcpstat.tcps_sc_dupsyn++; 852 if (ipopts) { 853 /* 854 * If we were remembering a previous source route, 855 * forget it and use the new one we've been given. 856 */ 857 if (sc->sc_ipopts) 858 (void) m_free(sc->sc_ipopts); 859 sc->sc_ipopts = ipopts; 860 } 861 /* 862 * Update timestamp if present. 863 */ 864 if (sc->sc_flags & SCF_TIMESTAMP) 865 sc->sc_tsrecent = to->to_tsval; 866 if (syncache_respond(sc, m) == 0) { 867 SYNCACHE_TIMEOUT(sc, sch, 1); 868 tcpstat.tcps_sndacks++; 869 tcpstat.tcps_sndtotal++; 870 } 871 SCH_UNLOCK(sch); 872 goto done; 873 } 874 875 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 876 if (sc == NULL) { 877 /* 878 * The zone allocator couldn't provide more entries. 879 * Treat this as if the cache was full; drop the oldest 880 * entry and insert the new one. 881 */ 882 tcpstat.tcps_sc_zonefail++; 883 sc = TAILQ_LAST(&sch->sch_bucket, sch_head); 884 syncache_drop(sc, sch); 885 SCH_UNLOCK(sch); 886 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 887 if (sc == NULL) { 888 if (ipopts) 889 (void) m_free(ipopts); 890 goto done; 891 } 892 } else 893 SCH_UNLOCK(sch); 894 895 /* 896 * Fill in the syncache values. 897 */ 898 sc->sc_ipopts = ipopts; 899 sc->sc_inc.inc_fport = inc->inc_fport; 900 sc->sc_inc.inc_lport = inc->inc_lport; 901#ifdef INET6 902 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 903 if (inc->inc_isipv6) { 904 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 905 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 906 } else 907#endif 908 { 909 sc->sc_inc.inc_faddr = inc->inc_faddr; 910 sc->sc_inc.inc_laddr = inc->inc_laddr; 911 sc->sc_ip_tos = ip_tos; 912 sc->sc_ip_ttl = ip_ttl; 913 } 914 sc->sc_irs = th->th_seq; 915 sc->sc_flags = 0; 916 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 917 sc->sc_flowlabel = 0; 918 if (tcp_syncookies) { 919 sc->sc_iss = syncookie_generate(sc, &flowtmp); 920#ifdef INET6 921 if (autoflowlabel) 922 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK; 923#endif 924 } else { 925 sc->sc_iss = arc4random(); 926#ifdef INET6 927 if (autoflowlabel) 928 sc->sc_flowlabel = 929 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 930#endif 931 } 932 933 /* 934 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 935 * win was derived from socket earlier in the function. 936 */ 937 win = imax(win, 0); 938 win = imin(win, TCP_MAXWIN); 939 sc->sc_wnd = win; 940 941 if (tcp_do_rfc1323) { 942 /* 943 * A timestamp received in a SYN makes 944 * it ok to send timestamp requests and replies. 945 */ 946 if (to->to_flags & TOF_TS) { 947 sc->sc_tsrecent = to->to_tsval; 948 sc->sc_flags |= SCF_TIMESTAMP; 949 } 950 if (to->to_flags & TOF_SCALE) { 951 int wscale = 0; 952 953 /* Compute proper scaling value from buffer space */ 954 while (wscale < TCP_MAX_WINSHIFT && 955 (TCP_MAXWIN << wscale) < sb_hiwat) 956 wscale++; 957 sc->sc_request_r_scale = wscale; 958 sc->sc_requested_s_scale = to->to_requested_s_scale; 959 sc->sc_flags |= SCF_WINSCALE; 960 } 961 } 962#ifdef TCP_SIGNATURE 963 /* 964 * If listening socket requested TCP digests, and received SYN 965 * contains the option, flag this in the syncache so that 966 * syncache_respond() will do the right thing with the SYN+ACK. 967 * XXX: Currently we always record the option by default and will 968 * attempt to use it in syncache_respond(). 969 */ 970 if (to->to_flags & TOF_SIGNATURE) 971 sc->sc_flags |= SCF_SIGNATURE; 972#endif 973 974 if (to->to_flags & TOF_SACK) 975 sc->sc_flags |= SCF_SACK; 976 977 /* 978 * Do a standard 3-way handshake. 979 */ 980 if (syncache_respond(sc, m) == 0) { 981 syncache_insert(sc, sch); /* locks and unlocks sch */ 982 tcpstat.tcps_sndacks++; 983 tcpstat.tcps_sndtotal++; 984 } else { 985 syncache_free(sc); 986 tcpstat.tcps_sc_dropped++; 987 } 988 989done: 990 *lsop = NULL; 991 return (1); 992} 993 994static int 995syncache_respond(struct syncache *sc, struct mbuf *m) 996{ 997 u_int8_t *optp; 998 int optlen, error; 999 u_int16_t tlen, hlen, mssopt; 1000 struct ip *ip = NULL; 1001 struct tcphdr *th; 1002#ifdef INET6 1003 struct ip6_hdr *ip6 = NULL; 1004#endif 1005#ifdef MAC 1006 struct inpcb *inp = NULL; 1007#endif 1008 1009 hlen = 1010#ifdef INET6 1011 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1012#endif 1013 sizeof(struct ip); 1014 1015 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1016 1017 /* Determine MSS we advertize to other end of connection. */ 1018 mssopt = tcp_mssopt(&sc->sc_inc); 1019 1020 /* Compute the size of the TCP options. */ 1021 if (sc->sc_flags & SCF_NOOPT) { 1022 optlen = 0; 1023 } else { 1024 optlen = TCPOLEN_MAXSEG + 1025 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1026 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 1027#ifdef TCP_SIGNATURE 1028 if (sc->sc_flags & SCF_SIGNATURE) 1029 optlen += TCPOLEN_SIGNATURE; 1030#endif 1031 if (sc->sc_flags & SCF_SACK) 1032 optlen += TCPOLEN_SACK_PERMITTED; 1033 optlen = roundup2(optlen, 4); 1034 } 1035 tlen = hlen + sizeof(struct tcphdr) + optlen; 1036 1037 /* 1038 * XXX: Assume that the entire packet will fit in a header mbuf. 1039 */ 1040 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1041 1042 /* Create the IP+TCP header from scratch. */ 1043 if (m) 1044 m_freem(m); 1045 1046 m = m_gethdr(M_DONTWAIT, MT_DATA); 1047 if (m == NULL) 1048 return (ENOBUFS); 1049 m->m_data += max_linkhdr; 1050 m->m_len = tlen; 1051 m->m_pkthdr.len = tlen; 1052 m->m_pkthdr.rcvif = NULL; 1053 1054#ifdef MAC 1055 /* 1056 * For MAC look up the inpcb to get access to the label information. 1057 * We don't store the inpcb pointer in struct syncache to make locking 1058 * less complicated and to save locking operations. However for MAC 1059 * this gives a slight overhead as we have to do a full pcblookup here. 1060 */ 1061 INP_INFO_RLOCK(&tcbinfo); 1062 if (inp == NULL) { 1063#ifdef INET6 /* && MAC */ 1064 if (sc->sc_inc.inc_isipv6) 1065 inp = in6_pcblookup_hash(&tcbinfo, 1066 &sc->sc_inc.inc6_laddr, sc->sc_inc.inc_lport, 1067 &sc->sc_inc.inc6_faddr, sc->sc_inc.inc_fport, 1068 1, NULL); 1069 else 1070#endif /* INET6 */ 1071 inp = in_pcblookup_hash(&tcbinfo, 1072 sc->sc_inc.inc_laddr, sc->sc_inc.inc_lport, 1073 sc->sc_inc.inc_faddr, sc->sc_inc.inc_fport, 1074 1, NULL); 1075 if (inp == NULL) { 1076 m_freem(m); 1077 INP_INFO_RUNLOCK(&tcbinfo); 1078 return (ESHUTDOWN); 1079 } 1080 } 1081 INP_LOCK(inp); 1082 if (!inp->inp_socket->so_options & SO_ACCEPTCONN) { 1083 m_freem(m); 1084 INP_UNLOCK(inp); 1085 INP_INFO_RUNLOCK(&tcbinfo); 1086 return (ESHUTDOWN); 1087 } 1088 mac_create_mbuf_from_inpcb(inp, m); 1089 INP_UNLOCK(inp); 1090 INP_INFO_RUNLOCK(&tcbinfo); 1091#endif /* MAC */ 1092 1093#ifdef INET6 1094 if (sc->sc_inc.inc_isipv6) { 1095 ip6 = mtod(m, struct ip6_hdr *); 1096 ip6->ip6_vfc = IPV6_VERSION; 1097 ip6->ip6_nxt = IPPROTO_TCP; 1098 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1099 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1100 ip6->ip6_plen = htons(tlen - hlen); 1101 /* ip6_hlim is set after checksum */ 1102 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1103 ip6->ip6_flow |= sc->sc_flowlabel; 1104 1105 th = (struct tcphdr *)(ip6 + 1); 1106 } else 1107#endif 1108 { 1109 ip = mtod(m, struct ip *); 1110 ip->ip_v = IPVERSION; 1111 ip->ip_hl = sizeof(struct ip) >> 2; 1112 ip->ip_len = tlen; 1113 ip->ip_id = 0; 1114 ip->ip_off = 0; 1115 ip->ip_sum = 0; 1116 ip->ip_p = IPPROTO_TCP; 1117 ip->ip_src = sc->sc_inc.inc_laddr; 1118 ip->ip_dst = sc->sc_inc.inc_faddr; 1119 ip->ip_ttl = sc->sc_ip_ttl; 1120 ip->ip_tos = sc->sc_ip_tos; 1121 1122 /* 1123 * See if we should do MTU discovery. Route lookups are 1124 * expensive, so we will only unset the DF bit if: 1125 * 1126 * 1) path_mtu_discovery is disabled 1127 * 2) the SCF_UNREACH flag has been set 1128 */ 1129 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1130 ip->ip_off |= IP_DF; 1131 1132 th = (struct tcphdr *)(ip + 1); 1133 } 1134 th->th_sport = sc->sc_inc.inc_lport; 1135 th->th_dport = sc->sc_inc.inc_fport; 1136 1137 th->th_seq = htonl(sc->sc_iss); 1138 th->th_ack = htonl(sc->sc_irs + 1); 1139 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1140 th->th_x2 = 0; 1141 th->th_flags = TH_SYN|TH_ACK; 1142 th->th_win = htons(sc->sc_wnd); 1143 th->th_urp = 0; 1144 1145 /* Tack on the TCP options. */ 1146 if (optlen != 0) { 1147 optp = (u_int8_t *)(th + 1); 1148 *optp++ = TCPOPT_MAXSEG; 1149 *optp++ = TCPOLEN_MAXSEG; 1150 *optp++ = (mssopt >> 8) & 0xff; 1151 *optp++ = mssopt & 0xff; 1152 1153 if (sc->sc_flags & SCF_WINSCALE) { 1154 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1155 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1156 sc->sc_request_r_scale); 1157 optp += 4; 1158 } 1159 1160 if (sc->sc_flags & SCF_TIMESTAMP) { 1161 u_int32_t *lp = (u_int32_t *)(optp); 1162 1163 /* Form timestamp option per appendix A of RFC 1323. */ 1164 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1165 *lp++ = htonl(ticks); 1166 *lp = htonl(sc->sc_tsrecent); 1167 optp += TCPOLEN_TSTAMP_APPA; 1168 } 1169 1170#ifdef TCP_SIGNATURE 1171 /* 1172 * Handle TCP-MD5 passive opener response. 1173 */ 1174 if (sc->sc_flags & SCF_SIGNATURE) { 1175 u_int8_t *bp = optp; 1176 int i; 1177 1178 *bp++ = TCPOPT_SIGNATURE; 1179 *bp++ = TCPOLEN_SIGNATURE; 1180 for (i = 0; i < TCP_SIGLEN; i++) 1181 *bp++ = 0; 1182 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1183 optp + 2, IPSEC_DIR_OUTBOUND); 1184 optp += TCPOLEN_SIGNATURE; 1185 } 1186#endif /* TCP_SIGNATURE */ 1187 1188 if (sc->sc_flags & SCF_SACK) { 1189 *optp++ = TCPOPT_SACK_PERMITTED; 1190 *optp++ = TCPOLEN_SACK_PERMITTED; 1191 } 1192 1193 { 1194 /* Pad TCP options to a 4 byte boundary */ 1195 int padlen = optlen - (optp - (u_int8_t *)(th + 1)); 1196 while (padlen-- > 0) 1197 *optp++ = TCPOPT_EOL; 1198 } 1199 } 1200 1201#ifdef INET6 1202 if (sc->sc_inc.inc_isipv6) { 1203 th->th_sum = 0; 1204 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1205 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1206 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1207 } else 1208#endif 1209 { 1210 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1211 htons(tlen - hlen + IPPROTO_TCP)); 1212 m->m_pkthdr.csum_flags = CSUM_TCP; 1213 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1214 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1215 } 1216 return (error); 1217} 1218 1219/* 1220 * cookie layers: 1221 * 1222 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1223 * | peer iss | 1224 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1225 * | 0 |(A)| | 1226 * (A): peer mss index 1227 */ 1228 1229/* 1230 * The values below are chosen to minimize the size of the tcp_secret 1231 * table, as well as providing roughly a 16 second lifetime for the cookie. 1232 */ 1233 1234#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1235#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1236 1237#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1238#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1239#define SYNCOOKIE_TIMEOUT \ 1240 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1241#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1242 1243#define SYNCOOKIE_RLOCK(ts) (rw_rlock(&(ts).ts_rwmtx)) 1244#define SYNCOOKIE_RUNLOCK(ts) (rw_runlock(&(ts).ts_rwmtx)) 1245#define SYNCOOKIE_TRY_UPGRADE(ts) (rw_try_upgrade(&(ts).ts_rwmtx)) 1246#define SYNCOOKIE_DOWNGRADE(ts) (rw_downgrade(&(ts).ts_rwmtx)) 1247 1248static struct { 1249 struct rwlock ts_rwmtx; 1250 u_int ts_expire; /* ticks */ 1251 u_int32_t ts_secbits[4]; 1252} tcp_secret[SYNCOOKIE_NSECRETS]; 1253 1254static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1255 1256static MD5_CTX syn_ctx; 1257 1258#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1259 1260struct md5_add { 1261 u_int32_t laddr, faddr; 1262 u_int32_t secbits[4]; 1263 u_int16_t lport, fport; 1264}; 1265 1266#ifdef CTASSERT 1267CTASSERT(sizeof(struct md5_add) == 28); 1268#endif 1269 1270/* 1271 * Consider the problem of a recreated (and retransmitted) cookie. If the 1272 * original SYN was accepted, the connection is established. The second 1273 * SYN is inflight, and if it arrives with an ISN that falls within the 1274 * receive window, the connection is killed. 1275 * 1276 * However, since cookies have other problems, this may not be worth 1277 * worrying about. 1278 */ 1279 1280static void 1281syncookie_init(void) { 1282 int idx; 1283 1284 for (idx = 0; idx < SYNCOOKIE_NSECRETS; idx++) { 1285 rw_init(&(tcp_secret[idx].ts_rwmtx), "tcp_secret"); 1286 } 1287} 1288 1289static u_int32_t 1290syncookie_generate(struct syncache *sc, u_int32_t *flowid) 1291{ 1292 u_int32_t md5_buffer[4]; 1293 u_int32_t data; 1294 int idx, i; 1295 struct md5_add add; 1296 1297 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1298 SYNCOOKIE_RLOCK(tcp_secret[idx]); 1299 if (tcp_secret[idx].ts_expire < time_uptime && 1300 SYNCOOKIE_TRY_UPGRADE(tcp_secret[idx]) ) { 1301 /* need write access */ 1302 for (i = 0; i < 4; i++) 1303 tcp_secret[idx].ts_secbits[i] = arc4random(); 1304 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1305 SYNCOOKIE_DOWNGRADE(tcp_secret[idx]); 1306 } 1307 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1308 if (tcp_msstab[data] <= sc->sc_peer_mss) 1309 break; 1310 data = (data << SYNCOOKIE_WNDBITS) | idx; 1311 data ^= sc->sc_irs; /* peer's iss */ 1312 MD5Init(&syn_ctx); 1313#ifdef INET6 1314 if (sc->sc_inc.inc_isipv6) { 1315 MD5Add(sc->sc_inc.inc6_laddr); 1316 MD5Add(sc->sc_inc.inc6_faddr); 1317 add.laddr = 0; 1318 add.faddr = 0; 1319 } else 1320#endif 1321 { 1322 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1323 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1324 } 1325 add.lport = sc->sc_inc.inc_lport; 1326 add.fport = sc->sc_inc.inc_fport; 1327 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1328 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1329 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1330 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1331 SYNCOOKIE_RUNLOCK(tcp_secret[idx]); 1332 MD5Add(add); 1333 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1334 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1335 *flowid = md5_buffer[1]; 1336 return (data); 1337} 1338 1339static struct syncache * 1340syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1341{ 1342 u_int32_t md5_buffer[4]; 1343 struct syncache *sc; 1344 u_int32_t data; 1345 int wnd, idx; 1346 struct md5_add add; 1347 1348 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1349 idx = data & SYNCOOKIE_WNDMASK; 1350 SYNCOOKIE_RLOCK(tcp_secret[idx]); 1351 if (tcp_secret[idx].ts_expire < ticks || 1352 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) { 1353 SYNCOOKIE_RUNLOCK(tcp_secret[idx]); 1354 return (NULL); 1355 } 1356 MD5Init(&syn_ctx); 1357#ifdef INET6 1358 if (inc->inc_isipv6) { 1359 MD5Add(inc->inc6_laddr); 1360 MD5Add(inc->inc6_faddr); 1361 add.laddr = 0; 1362 add.faddr = 0; 1363 } else 1364#endif 1365 { 1366 add.laddr = inc->inc_laddr.s_addr; 1367 add.faddr = inc->inc_faddr.s_addr; 1368 } 1369 add.lport = inc->inc_lport; 1370 add.fport = inc->inc_fport; 1371 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1372 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1373 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1374 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1375 SYNCOOKIE_RUNLOCK(tcp_secret[idx]); 1376 MD5Add(add); 1377 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1378 data ^= md5_buffer[0]; 1379 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1380 return (NULL); 1381 data = data >> SYNCOOKIE_WNDBITS; 1382 1383 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 1384 if (sc == NULL) 1385 return (NULL); 1386 /* 1387 * Fill in the syncache values. 1388 * XXX: duplicate code from syncache_add 1389 */ 1390 sc->sc_ipopts = NULL; 1391 sc->sc_inc.inc_fport = inc->inc_fport; 1392 sc->sc_inc.inc_lport = inc->inc_lport; 1393#ifdef INET6 1394 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1395 if (inc->inc_isipv6) { 1396 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1397 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1398 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1399 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1400 } else 1401#endif 1402 { 1403 sc->sc_inc.inc_faddr = inc->inc_faddr; 1404 sc->sc_inc.inc_laddr = inc->inc_laddr; 1405 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1406 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1407 } 1408 sc->sc_irs = th->th_seq - 1; 1409 sc->sc_iss = th->th_ack - 1; 1410 wnd = sbspace(&so->so_rcv); 1411 wnd = imax(wnd, 0); 1412 wnd = imin(wnd, TCP_MAXWIN); 1413 sc->sc_wnd = wnd; 1414 sc->sc_flags = 0; 1415 sc->sc_rxmits = 0; 1416 sc->sc_peer_mss = tcp_msstab[data]; 1417 return (sc); 1418} 1419