tcp_syncache.c revision 124851
1/*- 2 * Copyright (c) 2001 Networks Associates Technology, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Jonathan Lemon 6 * and NAI Labs, the Security Research Division of Network Associates, Inc. 7 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 8 * DARPA CHATS research program. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. The name of the author may not be used to endorse or promote 19 * products derived from this software without specific prior written 20 * permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $FreeBSD: head/sys/netinet/tcp_syncache.c 124851 2004-01-23 01:39:12Z andre $ 35 */ 36 37#include "opt_inet6.h" 38#include "opt_ipsec.h" 39#include "opt_mac.h" 40#include "opt_tcpdebug.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/malloc.h> 47#include <sys/mac.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 55#include <net/if.h> 56#include <net/route.h> 57 58#include <netinet/in.h> 59#include <netinet/in_systm.h> 60#include <netinet/ip.h> 61#include <netinet/in_var.h> 62#include <netinet/in_pcb.h> 63#include <netinet/ip_var.h> 64#ifdef INET6 65#include <netinet/ip6.h> 66#include <netinet/icmp6.h> 67#include <netinet6/nd6.h> 68#include <netinet6/ip6_var.h> 69#include <netinet6/in6_pcb.h> 70#endif 71#include <netinet/tcp.h> 72#ifdef TCPDEBUG 73#include <netinet/tcpip.h> 74#endif 75#include <netinet/tcp_fsm.h> 76#include <netinet/tcp_seq.h> 77#include <netinet/tcp_timer.h> 78#include <netinet/tcp_var.h> 79#ifdef TCPDEBUG 80#include <netinet/tcp_debug.h> 81#endif 82#ifdef INET6 83#include <netinet6/tcp6_var.h> 84#endif 85 86#ifdef IPSEC 87#include <netinet6/ipsec.h> 88#ifdef INET6 89#include <netinet6/ipsec6.h> 90#endif 91#endif /*IPSEC*/ 92 93#ifdef FAST_IPSEC 94#include <netipsec/ipsec.h> 95#ifdef INET6 96#include <netipsec/ipsec6.h> 97#endif 98#include <netipsec/key.h> 99#endif /*FAST_IPSEC*/ 100 101#include <machine/in_cksum.h> 102#include <vm/uma.h> 103 104static int tcp_syncookies = 1; 105SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 106 &tcp_syncookies, 0, 107 "Use TCP SYN cookies if the syncache overflows"); 108 109static void syncache_drop(struct syncache *, struct syncache_head *); 110static void syncache_free(struct syncache *); 111static void syncache_insert(struct syncache *, struct syncache_head *); 112struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 113#ifdef TCPDEBUG 114static int syncache_respond(struct syncache *, struct mbuf *, struct socket *); 115#else 116static int syncache_respond(struct syncache *, struct mbuf *); 117#endif 118static struct socket *syncache_socket(struct syncache *, struct socket *, 119 struct mbuf *m); 120static void syncache_timer(void *); 121static u_int32_t syncookie_generate(struct syncache *); 122static struct syncache *syncookie_lookup(struct in_conninfo *, 123 struct tcphdr *, struct socket *); 124 125/* 126 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 127 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 128 * the odds are that the user has given up attempting to connect by then. 129 */ 130#define SYNCACHE_MAXREXMTS 3 131 132/* Arbitrary values */ 133#define TCP_SYNCACHE_HASHSIZE 512 134#define TCP_SYNCACHE_BUCKETLIMIT 30 135 136struct tcp_syncache { 137 struct syncache_head *hashbase; 138 uma_zone_t zone; 139 u_int hashsize; 140 u_int hashmask; 141 u_int bucket_limit; 142 u_int cache_count; 143 u_int cache_limit; 144 u_int rexmt_limit; 145 u_int hash_secret; 146 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 147 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 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, slot) do { \ 192 sc->sc_rxtslot = (slot); \ 193 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[(slot)]; \ 194 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[(slot)], sc, sc_timerq); \ 195 if (!callout_active(&tcp_syncache.tt_timerq[(slot)])) \ 196 callout_reset(&tcp_syncache.tt_timerq[(slot)], \ 197 TCPTV_RTOBASE * tcp_backoff[(slot)], \ 198 syncache_timer, (void *)((intptr_t)(slot))); \ 199} while (0) 200 201static void 202syncache_free(struct syncache *sc) 203{ 204 if (sc->sc_ipopts) 205 (void) m_free(sc->sc_ipopts); 206 207 uma_zfree(tcp_syncache.zone, sc); 208} 209 210void 211syncache_init(void) 212{ 213 int i; 214 215 tcp_syncache.cache_count = 0; 216 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 217 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 218 tcp_syncache.cache_limit = 219 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 220 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 221 tcp_syncache.hash_secret = arc4random(); 222 223 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 224 &tcp_syncache.hashsize); 225 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 226 &tcp_syncache.cache_limit); 227 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 228 &tcp_syncache.bucket_limit); 229 if (!powerof2(tcp_syncache.hashsize)) { 230 printf("WARNING: syncache hash size is not a power of 2.\n"); 231 tcp_syncache.hashsize = 512; /* safe default */ 232 } 233 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 234 235 /* Allocate the hash table. */ 236 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 237 tcp_syncache.hashsize * sizeof(struct syncache_head), 238 M_SYNCACHE, M_WAITOK); 239 240 /* Initialize the hash buckets. */ 241 for (i = 0; i < tcp_syncache.hashsize; i++) { 242 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 243 tcp_syncache.hashbase[i].sch_length = 0; 244 } 245 246 /* Initialize the timer queues. */ 247 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 248 TAILQ_INIT(&tcp_syncache.timerq[i]); 249 callout_init(&tcp_syncache.tt_timerq[i], 250 debug_mpsafenet ? CALLOUT_MPSAFE : 0); 251 } 252 253 /* 254 * Allocate the syncache entries. Allow the zone to allocate one 255 * more entry than cache limit, so a new entry can bump out an 256 * older one. 257 */ 258 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 259 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 260 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 261 tcp_syncache.cache_limit -= 1; 262} 263 264static void 265syncache_insert(sc, sch) 266 struct syncache *sc; 267 struct syncache_head *sch; 268{ 269 struct syncache *sc2; 270 int i; 271 272 INP_INFO_WLOCK_ASSERT(&tcbinfo); 273 274 /* 275 * Make sure that we don't overflow the per-bucket 276 * limit or the total cache size limit. 277 */ 278 if (sch->sch_length >= tcp_syncache.bucket_limit) { 279 /* 280 * The bucket is full, toss the oldest element. 281 */ 282 sc2 = TAILQ_FIRST(&sch->sch_bucket); 283 sc2->sc_tp->ts_recent = ticks; 284 syncache_drop(sc2, sch); 285 tcpstat.tcps_sc_bucketoverflow++; 286 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) { 287 /* 288 * The cache is full. Toss the oldest entry in the 289 * entire cache. This is the front entry in the 290 * first non-empty timer queue with the largest 291 * timeout value. 292 */ 293 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 294 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]); 295 if (sc2 != NULL) 296 break; 297 } 298 sc2->sc_tp->ts_recent = ticks; 299 syncache_drop(sc2, NULL); 300 tcpstat.tcps_sc_cacheoverflow++; 301 } 302 303 /* Initialize the entry's timer. */ 304 SYNCACHE_TIMEOUT(sc, 0); 305 306 /* Put it into the bucket. */ 307 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 308 sch->sch_length++; 309 tcp_syncache.cache_count++; 310 tcpstat.tcps_sc_added++; 311} 312 313static void 314syncache_drop(sc, sch) 315 struct syncache *sc; 316 struct syncache_head *sch; 317{ 318 INP_INFO_WLOCK_ASSERT(&tcbinfo); 319 320 if (sch == NULL) { 321#ifdef INET6 322 if (sc->sc_inc.inc_isipv6) { 323 sch = &tcp_syncache.hashbase[ 324 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 325 } else 326#endif 327 { 328 sch = &tcp_syncache.hashbase[ 329 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 330 } 331 } 332 333 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 334 sch->sch_length--; 335 tcp_syncache.cache_count--; 336 337 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq); 338 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot])) 339 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]); 340 341 syncache_free(sc); 342} 343 344/* 345 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 346 * If we have retransmitted an entry the maximum number of times, expire it. 347 */ 348static void 349syncache_timer(xslot) 350 void *xslot; 351{ 352 intptr_t slot = (intptr_t)xslot; 353 struct syncache *sc, *nsc; 354 struct inpcb *inp; 355 356 INP_INFO_WLOCK(&tcbinfo); 357 if (callout_pending(&tcp_syncache.tt_timerq[slot]) || 358 !callout_active(&tcp_syncache.tt_timerq[slot])) { 359 /* XXX can this happen? */ 360 INP_INFO_WUNLOCK(&tcbinfo); 361 return; 362 } 363 callout_deactivate(&tcp_syncache.tt_timerq[slot]); 364 365 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]); 366 while (nsc != NULL) { 367 if (ticks < nsc->sc_rxttime) 368 break; 369 sc = nsc; 370 inp = sc->sc_tp->t_inpcb; 371 if (slot == SYNCACHE_MAXREXMTS || 372 slot >= tcp_syncache.rexmt_limit || 373 inp == NULL || inp->inp_gencnt != sc->sc_inp_gencnt) { 374 nsc = TAILQ_NEXT(sc, sc_timerq); 375 syncache_drop(sc, NULL); 376 tcpstat.tcps_sc_stale++; 377 continue; 378 } 379 /* 380 * syncache_respond() may call back into the syncache to 381 * to modify another entry, so do not obtain the next 382 * entry on the timer chain until it has completed. 383 */ 384#ifdef TCPDEBUG 385 (void) syncache_respond(sc, NULL, NULL); 386#else 387 (void) syncache_respond(sc, NULL); 388#endif 389 nsc = TAILQ_NEXT(sc, sc_timerq); 390 tcpstat.tcps_sc_retransmitted++; 391 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq); 392 SYNCACHE_TIMEOUT(sc, slot + 1); 393 } 394 if (nsc != NULL) 395 callout_reset(&tcp_syncache.tt_timerq[slot], 396 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot)); 397 INP_INFO_WUNLOCK(&tcbinfo); 398} 399 400/* 401 * Find an entry in the syncache. 402 */ 403struct syncache * 404syncache_lookup(inc, schp) 405 struct in_conninfo *inc; 406 struct syncache_head **schp; 407{ 408 struct syncache *sc; 409 struct syncache_head *sch; 410 411 INP_INFO_WLOCK_ASSERT(&tcbinfo); 412 413#ifdef INET6 414 if (inc->inc_isipv6) { 415 sch = &tcp_syncache.hashbase[ 416 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 417 *schp = sch; 418 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 419 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 420 return (sc); 421 } 422 } else 423#endif 424 { 425 sch = &tcp_syncache.hashbase[ 426 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 427 *schp = sch; 428 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 429#ifdef INET6 430 if (sc->sc_inc.inc_isipv6) 431 continue; 432#endif 433 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 434 return (sc); 435 } 436 } 437 return (NULL); 438} 439 440/* 441 * This function is called when we get a RST for a 442 * non-existent connection, so that we can see if the 443 * connection is in the syn cache. If it is, zap it. 444 */ 445void 446syncache_chkrst(inc, th) 447 struct in_conninfo *inc; 448 struct tcphdr *th; 449{ 450 struct syncache *sc; 451 struct syncache_head *sch; 452 453 INP_INFO_WLOCK_ASSERT(&tcbinfo); 454 455 sc = syncache_lookup(inc, &sch); 456 if (sc == NULL) 457 return; 458 /* 459 * If the RST bit is set, check the sequence number to see 460 * if this is a valid reset segment. 461 * RFC 793 page 37: 462 * In all states except SYN-SENT, all reset (RST) segments 463 * are validated by checking their SEQ-fields. A reset is 464 * valid if its sequence number is in the window. 465 * 466 * The sequence number in the reset segment is normally an 467 * echo of our outgoing acknowlegement numbers, but some hosts 468 * send a reset with the sequence number at the rightmost edge 469 * of our receive window, and we have to handle this case. 470 */ 471 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 472 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 473 syncache_drop(sc, sch); 474 tcpstat.tcps_sc_reset++; 475 } 476} 477 478void 479syncache_badack(inc) 480 struct in_conninfo *inc; 481{ 482 struct syncache *sc; 483 struct syncache_head *sch; 484 485 INP_INFO_WLOCK_ASSERT(&tcbinfo); 486 487 sc = syncache_lookup(inc, &sch); 488 if (sc != NULL) { 489 syncache_drop(sc, sch); 490 tcpstat.tcps_sc_badack++; 491 } 492} 493 494void 495syncache_unreach(inc, th) 496 struct in_conninfo *inc; 497 struct tcphdr *th; 498{ 499 struct syncache *sc; 500 struct syncache_head *sch; 501 502 INP_INFO_WLOCK_ASSERT(&tcbinfo); 503 504 /* we are called at splnet() here */ 505 sc = syncache_lookup(inc, &sch); 506 if (sc == NULL) 507 return; 508 509 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 510 if (ntohl(th->th_seq) != sc->sc_iss) 511 return; 512 513 /* 514 * If we've rertransmitted 3 times and this is our second error, 515 * we remove the entry. Otherwise, we allow it to continue on. 516 * This prevents us from incorrectly nuking an entry during a 517 * spurious network outage. 518 * 519 * See tcp_notify(). 520 */ 521 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 522 sc->sc_flags |= SCF_UNREACH; 523 return; 524 } 525 syncache_drop(sc, sch); 526 tcpstat.tcps_sc_unreach++; 527} 528 529/* 530 * Build a new TCP socket structure from a syncache entry. 531 */ 532static struct socket * 533syncache_socket(sc, lso, m) 534 struct syncache *sc; 535 struct socket *lso; 536 struct mbuf *m; 537{ 538 struct inpcb *inp = NULL; 539 struct socket *so; 540 struct tcpcb *tp; 541 542 GIANT_REQUIRED; /* XXX until socket locking */ 543 INP_INFO_WLOCK_ASSERT(&tcbinfo); 544 545 /* 546 * Ok, create the full blown connection, and set things up 547 * as they would have been set up if we had created the 548 * connection when the SYN arrived. If we can't create 549 * the connection, abort it. 550 */ 551 so = sonewconn(lso, SS_ISCONNECTED); 552 if (so == NULL) { 553 /* 554 * Drop the connection; we will send a RST if the peer 555 * retransmits the ACK, 556 */ 557 tcpstat.tcps_listendrop++; 558 goto abort2; 559 } 560#ifdef MAC 561 mac_set_socket_peer_from_mbuf(m, so); 562#endif 563 564 inp = sotoinpcb(so); 565 INP_LOCK(inp); 566 567 /* 568 * Insert new socket into hash list. 569 */ 570 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 571#ifdef INET6 572 if (sc->sc_inc.inc_isipv6) { 573 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 574 } else { 575 inp->inp_vflag &= ~INP_IPV6; 576 inp->inp_vflag |= INP_IPV4; 577#endif 578 inp->inp_laddr = sc->sc_inc.inc_laddr; 579#ifdef INET6 580 } 581#endif 582 inp->inp_lport = sc->sc_inc.inc_lport; 583 if (in_pcbinshash(inp) != 0) { 584 /* 585 * Undo the assignments above if we failed to 586 * put the PCB on the hash lists. 587 */ 588#ifdef INET6 589 if (sc->sc_inc.inc_isipv6) 590 inp->in6p_laddr = in6addr_any; 591 else 592#endif 593 inp->inp_laddr.s_addr = INADDR_ANY; 594 inp->inp_lport = 0; 595 goto abort; 596 } 597#ifdef IPSEC 598 /* copy old policy into new socket's */ 599 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 600 printf("syncache_expand: could not copy policy\n"); 601#endif 602#ifdef FAST_IPSEC 603 /* copy old policy into new socket's */ 604 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 605 printf("syncache_expand: could not copy policy\n"); 606#endif 607#ifdef INET6 608 if (sc->sc_inc.inc_isipv6) { 609 struct inpcb *oinp = sotoinpcb(lso); 610 struct in6_addr laddr6; 611 struct sockaddr_in6 sin6; 612 /* 613 * Inherit socket options from the listening socket. 614 * Note that in6p_inputopts are not (and should not be) 615 * copied, since it stores previously received options and is 616 * used to detect if each new option is different than the 617 * previous one and hence should be passed to a user. 618 * If we copied in6p_inputopts, a user would not be able to 619 * receive options just after calling the accept system call. 620 */ 621 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 622 if (oinp->in6p_outputopts) 623 inp->in6p_outputopts = 624 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 625 626 sin6.sin6_family = AF_INET6; 627 sin6.sin6_len = sizeof(sin6); 628 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 629 sin6.sin6_port = sc->sc_inc.inc_fport; 630 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 631 laddr6 = inp->in6p_laddr; 632 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 633 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 634 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 635 inp->in6p_laddr = laddr6; 636 goto abort; 637 } 638 } else 639#endif 640 { 641 struct in_addr laddr; 642 struct sockaddr_in sin; 643 644 inp->inp_options = ip_srcroute(); 645 if (inp->inp_options == NULL) { 646 inp->inp_options = sc->sc_ipopts; 647 sc->sc_ipopts = NULL; 648 } 649 650 sin.sin_family = AF_INET; 651 sin.sin_len = sizeof(sin); 652 sin.sin_addr = sc->sc_inc.inc_faddr; 653 sin.sin_port = sc->sc_inc.inc_fport; 654 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 655 laddr = inp->inp_laddr; 656 if (inp->inp_laddr.s_addr == INADDR_ANY) 657 inp->inp_laddr = sc->sc_inc.inc_laddr; 658 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 659 inp->inp_laddr = laddr; 660 goto abort; 661 } 662 } 663 664 tp = intotcpcb(inp); 665 tp->t_state = TCPS_SYN_RECEIVED; 666 tp->iss = sc->sc_iss; 667 tp->irs = sc->sc_irs; 668 tcp_rcvseqinit(tp); 669 tcp_sendseqinit(tp); 670 tp->snd_wl1 = sc->sc_irs; 671 tp->rcv_up = sc->sc_irs + 1; 672 tp->rcv_wnd = sc->sc_wnd; 673 tp->rcv_adv += tp->rcv_wnd; 674 675 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 676 if (sc->sc_flags & SCF_NOOPT) 677 tp->t_flags |= TF_NOOPT; 678 if (sc->sc_flags & SCF_WINSCALE) { 679 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 680 tp->requested_s_scale = sc->sc_requested_s_scale; 681 tp->request_r_scale = sc->sc_request_r_scale; 682 } 683 if (sc->sc_flags & SCF_TIMESTAMP) { 684 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 685 tp->ts_recent = sc->sc_tsrecent; 686 tp->ts_recent_age = ticks; 687 } 688 if (sc->sc_flags & SCF_CC) { 689 /* 690 * Initialization of the tcpcb for transaction; 691 * set SND.WND = SEG.WND, 692 * initialize CCsend and CCrecv. 693 */ 694 tp->t_flags |= TF_REQ_CC|TF_RCVD_CC; 695 tp->cc_send = sc->sc_cc_send; 696 tp->cc_recv = sc->sc_cc_recv; 697 } 698 699 /* 700 * Set up MSS and get cached values from tcp_hostcache. 701 * This might overwrite some of the defaults we just set. 702 */ 703 tcp_mss(tp, sc->sc_peer_mss); 704 705 /* 706 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 707 */ 708 if (sc->sc_rxtslot != 0) 709 tp->snd_cwnd = tp->t_maxseg; 710 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 711 712 INP_UNLOCK(inp); 713 714 tcpstat.tcps_accepts++; 715 return (so); 716 717abort: 718 INP_UNLOCK(inp); 719abort2: 720 if (so != NULL) 721 (void) soabort(so); 722 return (NULL); 723} 724 725/* 726 * This function gets called when we receive an ACK for a 727 * socket in the LISTEN state. We look up the connection 728 * in the syncache, and if its there, we pull it out of 729 * the cache and turn it into a full-blown connection in 730 * the SYN-RECEIVED state. 731 */ 732int 733syncache_expand(inc, th, sop, m) 734 struct in_conninfo *inc; 735 struct tcphdr *th; 736 struct socket **sop; 737 struct mbuf *m; 738{ 739 struct syncache *sc; 740 struct syncache_head *sch; 741 struct socket *so; 742 743 INP_INFO_WLOCK_ASSERT(&tcbinfo); 744 745 sc = syncache_lookup(inc, &sch); 746 if (sc == NULL) { 747 /* 748 * There is no syncache entry, so see if this ACK is 749 * a returning syncookie. To do this, first: 750 * A. See if this socket has had a syncache entry dropped in 751 * the past. We don't want to accept a bogus syncookie 752 * if we've never received a SYN. 753 * B. check that the syncookie is valid. If it is, then 754 * cobble up a fake syncache entry, and return. 755 */ 756 if (!tcp_syncookies) 757 return (0); 758 sc = syncookie_lookup(inc, th, *sop); 759 if (sc == NULL) 760 return (0); 761 sch = NULL; 762 tcpstat.tcps_sc_recvcookie++; 763 } 764 765 /* 766 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 767 */ 768 if (th->th_ack != sc->sc_iss + 1) 769 return (0); 770 771 so = syncache_socket(sc, *sop, m); 772 if (so == NULL) { 773#if 0 774resetandabort: 775 /* XXXjlemon check this - is this correct? */ 776 (void) tcp_respond(NULL, m, m, th, 777 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 778#endif 779 m_freem(m); /* XXX only needed for above */ 780 tcpstat.tcps_sc_aborted++; 781 } else 782 tcpstat.tcps_sc_completed++; 783 784 if (sch == NULL) 785 syncache_free(sc); 786 else 787 syncache_drop(sc, sch); 788 *sop = so; 789 return (1); 790} 791 792/* 793 * Given a LISTEN socket and an inbound SYN request, add 794 * this to the syn cache, and send back a segment: 795 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 796 * to the source. 797 * 798 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 799 * Doing so would require that we hold onto the data and deliver it 800 * to the application. However, if we are the target of a SYN-flood 801 * DoS attack, an attacker could send data which would eventually 802 * consume all available buffer space if it were ACKed. By not ACKing 803 * the data, we avoid this DoS scenario. 804 */ 805int 806syncache_add(inc, to, th, sop, m) 807 struct in_conninfo *inc; 808 struct tcpopt *to; 809 struct tcphdr *th; 810 struct socket **sop; 811 struct mbuf *m; 812{ 813 struct tcpcb *tp; 814 struct socket *so; 815 struct syncache *sc = NULL; 816 struct syncache_head *sch; 817 struct mbuf *ipopts = NULL; 818 struct rmxp_tao tao; 819 int i, win; 820 821 INP_INFO_WLOCK_ASSERT(&tcbinfo); 822 823 so = *sop; 824 tp = sototcpcb(so); 825 bzero(&tao, sizeof(tao)); 826 827 /* 828 * Remember the IP options, if any. 829 */ 830#ifdef INET6 831 if (!inc->inc_isipv6) 832#endif 833 ipopts = ip_srcroute(); 834 835 /* 836 * See if we already have an entry for this connection. 837 * If we do, resend the SYN,ACK, and reset the retransmit timer. 838 * 839 * XXX 840 * should the syncache be re-initialized with the contents 841 * of the new SYN here (which may have different options?) 842 */ 843 sc = syncache_lookup(inc, &sch); 844 if (sc != NULL) { 845 tcpstat.tcps_sc_dupsyn++; 846 if (ipopts) { 847 /* 848 * If we were remembering a previous source route, 849 * forget it and use the new one we've been given. 850 */ 851 if (sc->sc_ipopts) 852 (void) m_free(sc->sc_ipopts); 853 sc->sc_ipopts = ipopts; 854 } 855 /* 856 * Update timestamp if present. 857 */ 858 if (sc->sc_flags & SCF_TIMESTAMP) 859 sc->sc_tsrecent = to->to_tsval; 860 /* 861 * PCB may have changed, pick up new values. 862 */ 863 sc->sc_tp = tp; 864 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 865#ifdef TCPDEBUG 866 if (syncache_respond(sc, m, so) == 0) { 867#else 868 if (syncache_respond(sc, m) == 0) { 869#endif 870 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 871 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 872 sc, sc_timerq); 873 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 874 tcpstat.tcps_sndacks++; 875 tcpstat.tcps_sndtotal++; 876 } 877 *sop = NULL; 878 return (1); 879 } 880 881 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 882 if (sc == NULL) { 883 /* 884 * The zone allocator couldn't provide more entries. 885 * Treat this as if the cache was full; drop the oldest 886 * entry and insert the new one. 887 */ 888 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 889 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 890 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]); 891 if (sc != NULL) 892 break; 893 } 894 sc->sc_tp->ts_recent = ticks; 895 syncache_drop(sc, NULL); 896 tcpstat.tcps_sc_zonefail++; 897 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 898 if (sc == NULL) { 899 if (ipopts) 900 (void) m_free(ipopts); 901 return (0); 902 } 903 } 904 905 /* 906 * Fill in the syncache values. 907 */ 908 bzero(sc, sizeof(*sc)); 909 sc->sc_tp = tp; 910 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 911 sc->sc_ipopts = ipopts; 912 sc->sc_inc.inc_fport = inc->inc_fport; 913 sc->sc_inc.inc_lport = inc->inc_lport; 914#ifdef INET6 915 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 916 if (inc->inc_isipv6) { 917 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 918 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 919 } else 920#endif 921 { 922 sc->sc_inc.inc_faddr = inc->inc_faddr; 923 sc->sc_inc.inc_laddr = inc->inc_laddr; 924 } 925 sc->sc_irs = th->th_seq; 926 sc->sc_flags = 0; 927 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 928 if (tcp_syncookies) 929 sc->sc_iss = syncookie_generate(sc); 930 else 931 sc->sc_iss = arc4random(); 932 933 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 934 win = sbspace(&so->so_rcv); 935 win = imax(win, 0); 936 win = imin(win, TCP_MAXWIN); 937 sc->sc_wnd = win; 938 939 if (tcp_do_rfc1323) { 940 /* 941 * A timestamp received in a SYN makes 942 * it ok to send timestamp requests and replies. 943 */ 944 if (to->to_flags & TOF_TS) { 945 sc->sc_tsrecent = to->to_tsval; 946 sc->sc_flags |= SCF_TIMESTAMP; 947 } 948 if (to->to_flags & TOF_SCALE) { 949 int wscale = 0; 950 951 /* Compute proper scaling value from buffer space */ 952 while (wscale < TCP_MAX_WINSHIFT && 953 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 954 wscale++; 955 sc->sc_request_r_scale = wscale; 956 sc->sc_requested_s_scale = to->to_requested_s_scale; 957 sc->sc_flags |= SCF_WINSCALE; 958 } 959 } 960 if (tcp_do_rfc1644) { 961 /* 962 * A CC or CC.new option received in a SYN makes 963 * it ok to send CC in subsequent segments. 964 */ 965 if (to->to_flags & (TOF_CC|TOF_CCNEW)) { 966 sc->sc_cc_recv = to->to_cc; 967 sc->sc_cc_send = CC_INC(tcp_ccgen); 968 sc->sc_flags |= SCF_CC; 969 } 970 } 971 if (tp->t_flags & TF_NOOPT) 972 sc->sc_flags = SCF_NOOPT; 973 974 /* 975 * XXX 976 * We have the option here of not doing TAO (even if the segment 977 * qualifies) and instead fall back to a normal 3WHS via the syncache. 978 * This allows us to apply synflood protection to TAO-qualifying SYNs 979 * also. However, there should be a hueristic to determine when to 980 * do this, and is not present at the moment. 981 */ 982 983 /* 984 * Perform TAO test on incoming CC (SEG.CC) option, if any. 985 * - compare SEG.CC against cached CC from the same host, if any. 986 * - if SEG.CC > chached value, SYN must be new and is accepted 987 * immediately: save new CC in the cache, mark the socket 988 * connected, enter ESTABLISHED state, turn on flag to 989 * send a SYN in the next segment. 990 * A virtual advertised window is set in rcv_adv to 991 * initialize SWS prevention. Then enter normal segment 992 * processing: drop SYN, process data and FIN. 993 * - otherwise do a normal 3-way handshake. 994 */ 995 if (tcp_do_rfc1644) 996 tcp_hc_gettao(&sc->sc_inc, &tao); 997 998 if ((to->to_flags & TOF_CC) != 0) { 999 if (((tp->t_flags & TF_NOPUSH) != 0) && 1000 sc->sc_flags & SCF_CC && tao.tao_cc != 0 && 1001 CC_GT(to->to_cc, tao.tao_cc)) { 1002 sc->sc_rxtslot = 0; 1003 so = syncache_socket(sc, *sop, m); 1004 if (so != NULL) { 1005 tao.tao_cc = to->to_cc; 1006 tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC, 1007 tao.tao_cc, 0); 1008 *sop = so; 1009 } 1010 syncache_free(sc); 1011 return (so != NULL); 1012 } 1013 } else { 1014 /* 1015 * No CC option, but maybe CC.NEW: invalidate cached value. 1016 */ 1017 if (tcp_do_rfc1644) { 1018 tao.tao_cc = 0; 1019 tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC, 1020 tao.tao_cc, 0); 1021 } 1022 } 1023 1024 /* 1025 * TAO test failed or there was no CC option, 1026 * do a standard 3-way handshake. 1027 */ 1028#ifdef TCPDEBUG 1029 if (syncache_respond(sc, m, so) == 0) { 1030#else 1031 if (syncache_respond(sc, m) == 0) { 1032#endif 1033 syncache_insert(sc, sch); 1034 tcpstat.tcps_sndacks++; 1035 tcpstat.tcps_sndtotal++; 1036 } else { 1037 syncache_free(sc); 1038 tcpstat.tcps_sc_dropped++; 1039 } 1040 *sop = NULL; 1041 return (1); 1042} 1043 1044#ifdef TCPDEBUG 1045static int 1046syncache_respond(sc, m, so) 1047 struct syncache *sc; 1048 struct mbuf *m; 1049 struct socket *so; 1050#else 1051static int 1052syncache_respond(sc, m) 1053 struct syncache *sc; 1054 struct mbuf *m; 1055#endif 1056{ 1057 u_int8_t *optp; 1058 int optlen, error; 1059 u_int16_t tlen, hlen, mssopt; 1060 struct ip *ip = NULL; 1061 struct tcphdr *th; 1062 struct inpcb *inp; 1063#ifdef INET6 1064 struct ip6_hdr *ip6 = NULL; 1065#endif 1066 1067 hlen = 1068#ifdef INET6 1069 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1070#endif 1071 sizeof(struct ip); 1072 1073 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1074 1075 /* Determine MSS we advertize to other end of connection */ 1076 mssopt = tcp_mssopt(&sc->sc_inc); 1077 1078 /* Compute the size of the TCP options. */ 1079 if (sc->sc_flags & SCF_NOOPT) { 1080 optlen = 0; 1081 } else { 1082 optlen = TCPOLEN_MAXSEG + 1083 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1084 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1085 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0); 1086 } 1087 tlen = hlen + sizeof(struct tcphdr) + optlen; 1088 1089 /* 1090 * XXX 1091 * assume that the entire packet will fit in a header mbuf 1092 */ 1093 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1094 1095 /* 1096 * XXX shouldn't this reuse the mbuf if possible ? 1097 * Create the IP+TCP header from scratch. 1098 */ 1099 if (m) 1100 m_freem(m); 1101 1102 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1103 if (m == NULL) 1104 return (ENOBUFS); 1105 m->m_data += max_linkhdr; 1106 m->m_len = tlen; 1107 m->m_pkthdr.len = tlen; 1108 m->m_pkthdr.rcvif = NULL; 1109 inp = sc->sc_tp->t_inpcb; 1110 INP_LOCK(inp); 1111#ifdef MAC 1112 mac_create_mbuf_from_socket(inp->inp_socket, m); 1113#endif 1114 1115#ifdef INET6 1116 if (sc->sc_inc.inc_isipv6) { 1117 ip6 = mtod(m, struct ip6_hdr *); 1118 ip6->ip6_vfc = IPV6_VERSION; 1119 ip6->ip6_nxt = IPPROTO_TCP; 1120 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1121 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1122 ip6->ip6_plen = htons(tlen - hlen); 1123 /* ip6_hlim is set after checksum */ 1124 /* ip6_flow = ??? */ 1125 1126 th = (struct tcphdr *)(ip6 + 1); 1127 } else 1128#endif 1129 { 1130 ip = mtod(m, struct ip *); 1131 ip->ip_v = IPVERSION; 1132 ip->ip_hl = sizeof(struct ip) >> 2; 1133 ip->ip_len = tlen; 1134 ip->ip_id = 0; 1135 ip->ip_off = 0; 1136 ip->ip_sum = 0; 1137 ip->ip_p = IPPROTO_TCP; 1138 ip->ip_src = sc->sc_inc.inc_laddr; 1139 ip->ip_dst = sc->sc_inc.inc_faddr; 1140 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1141 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 1142 1143 /* 1144 * See if we should do MTU discovery. Route lookups are 1145 * expensive, so we will only unset the DF bit if: 1146 * 1147 * 1) path_mtu_discovery is disabled 1148 * 2) the SCF_UNREACH flag has been set 1149 */ 1150 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1151 ip->ip_off |= IP_DF; 1152 1153 th = (struct tcphdr *)(ip + 1); 1154 } 1155 th->th_sport = sc->sc_inc.inc_lport; 1156 th->th_dport = sc->sc_inc.inc_fport; 1157 1158 th->th_seq = htonl(sc->sc_iss); 1159 th->th_ack = htonl(sc->sc_irs + 1); 1160 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1161 th->th_x2 = 0; 1162 th->th_flags = TH_SYN|TH_ACK; 1163 th->th_win = htons(sc->sc_wnd); 1164 th->th_urp = 0; 1165 1166 /* Tack on the TCP options. */ 1167 if (optlen != 0) { 1168 optp = (u_int8_t *)(th + 1); 1169 *optp++ = TCPOPT_MAXSEG; 1170 *optp++ = TCPOLEN_MAXSEG; 1171 *optp++ = (mssopt >> 8) & 0xff; 1172 *optp++ = mssopt & 0xff; 1173 1174 if (sc->sc_flags & SCF_WINSCALE) { 1175 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1176 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1177 sc->sc_request_r_scale); 1178 optp += 4; 1179 } 1180 1181 if (sc->sc_flags & SCF_TIMESTAMP) { 1182 u_int32_t *lp = (u_int32_t *)(optp); 1183 1184 /* Form timestamp option per appendix A of RFC 1323. */ 1185 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1186 *lp++ = htonl(ticks); 1187 *lp = htonl(sc->sc_tsrecent); 1188 optp += TCPOLEN_TSTAMP_APPA; 1189 } 1190 1191 /* 1192 * Send CC and CC.echo if we received CC from our peer. 1193 */ 1194 if (sc->sc_flags & SCF_CC) { 1195 u_int32_t *lp = (u_int32_t *)(optp); 1196 1197 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1198 *lp++ = htonl(sc->sc_cc_send); 1199 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1200 *lp = htonl(sc->sc_cc_recv); 1201 optp += TCPOLEN_CC_APPA * 2; 1202 } 1203 } 1204 1205#ifdef INET6 1206 if (sc->sc_inc.inc_isipv6) { 1207 th->th_sum = 0; 1208 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1209 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1210 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1211 } else 1212#endif 1213 { 1214 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1215 htons(tlen - hlen + IPPROTO_TCP)); 1216 m->m_pkthdr.csum_flags = CSUM_TCP; 1217 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1218#ifdef TCPDEBUG 1219 /* 1220 * Trace. 1221 */ 1222 if (so != NULL && so->so_options & SO_DEBUG) { 1223 struct tcpcb *tp = sototcpcb(so); 1224 tcp_trace(TA_OUTPUT, tp->t_state, tp, 1225 mtod(m, void *), th, 0); 1226 } 1227#endif 1228 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp); 1229 } 1230 INP_UNLOCK(inp); 1231 return (error); 1232} 1233 1234/* 1235 * cookie layers: 1236 * 1237 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1238 * | peer iss | 1239 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1240 * | 0 |(A)| | 1241 * (A): peer mss index 1242 */ 1243 1244/* 1245 * The values below are chosen to minimize the size of the tcp_secret 1246 * table, as well as providing roughly a 16 second lifetime for the cookie. 1247 */ 1248 1249#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1250#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1251 1252#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1253#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1254#define SYNCOOKIE_TIMEOUT \ 1255 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1256#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1257 1258static struct { 1259 u_int32_t ts_secbits[4]; 1260 u_int ts_expire; 1261} tcp_secret[SYNCOOKIE_NSECRETS]; 1262 1263static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1264 1265static MD5_CTX syn_ctx; 1266 1267#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1268 1269struct md5_add { 1270 u_int32_t laddr, faddr; 1271 u_int32_t secbits[4]; 1272 u_int16_t lport, fport; 1273}; 1274 1275#ifdef CTASSERT 1276CTASSERT(sizeof(struct md5_add) == 28); 1277#endif 1278 1279/* 1280 * Consider the problem of a recreated (and retransmitted) cookie. If the 1281 * original SYN was accepted, the connection is established. The second 1282 * SYN is inflight, and if it arrives with an ISN that falls within the 1283 * receive window, the connection is killed. 1284 * 1285 * However, since cookies have other problems, this may not be worth 1286 * worrying about. 1287 */ 1288 1289static u_int32_t 1290syncookie_generate(struct syncache *sc) 1291{ 1292 u_int32_t md5_buffer[4]; 1293 u_int32_t data; 1294 int idx, i; 1295 struct md5_add add; 1296 1297 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1298 1299 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1300 if (tcp_secret[idx].ts_expire < ticks) { 1301 for (i = 0; i < 4; i++) 1302 tcp_secret[idx].ts_secbits[i] = arc4random(); 1303 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1304 } 1305 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1306 if (tcp_msstab[data] <= sc->sc_peer_mss) 1307 break; 1308 data = (data << SYNCOOKIE_WNDBITS) | idx; 1309 data ^= sc->sc_irs; /* peer's iss */ 1310 MD5Init(&syn_ctx); 1311#ifdef INET6 1312 if (sc->sc_inc.inc_isipv6) { 1313 MD5Add(sc->sc_inc.inc6_laddr); 1314 MD5Add(sc->sc_inc.inc6_faddr); 1315 add.laddr = 0; 1316 add.faddr = 0; 1317 } else 1318#endif 1319 { 1320 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1321 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1322 } 1323 add.lport = sc->sc_inc.inc_lport; 1324 add.fport = sc->sc_inc.inc_fport; 1325 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1326 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1327 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1328 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1329 MD5Add(add); 1330 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1331 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1332 return (data); 1333} 1334 1335static struct syncache * 1336syncookie_lookup(inc, th, so) 1337 struct in_conninfo *inc; 1338 struct tcphdr *th; 1339 struct socket *so; 1340{ 1341 u_int32_t md5_buffer[4]; 1342 struct syncache *sc; 1343 u_int32_t data; 1344 int wnd, idx; 1345 struct md5_add add; 1346 1347 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1348 1349 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1350 idx = data & SYNCOOKIE_WNDMASK; 1351 if (tcp_secret[idx].ts_expire < ticks || 1352 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1353 return (NULL); 1354 MD5Init(&syn_ctx); 1355#ifdef INET6 1356 if (inc->inc_isipv6) { 1357 MD5Add(inc->inc6_laddr); 1358 MD5Add(inc->inc6_faddr); 1359 add.laddr = 0; 1360 add.faddr = 0; 1361 } else 1362#endif 1363 { 1364 add.laddr = inc->inc_laddr.s_addr; 1365 add.faddr = inc->inc_faddr.s_addr; 1366 } 1367 add.lport = inc->inc_lport; 1368 add.fport = inc->inc_fport; 1369 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1370 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1371 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1372 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1373 MD5Add(add); 1374 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1375 data ^= md5_buffer[0]; 1376 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1377 return (NULL); 1378 data = data >> SYNCOOKIE_WNDBITS; 1379 1380 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 1381 if (sc == NULL) 1382 return (NULL); 1383 /* 1384 * Fill in the syncache values. 1385 * XXX duplicate code from syncache_add 1386 */ 1387 sc->sc_ipopts = NULL; 1388 sc->sc_inc.inc_fport = inc->inc_fport; 1389 sc->sc_inc.inc_lport = inc->inc_lport; 1390#ifdef INET6 1391 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1392 if (inc->inc_isipv6) { 1393 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1394 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1395 } else 1396#endif 1397 { 1398 sc->sc_inc.inc_faddr = inc->inc_faddr; 1399 sc->sc_inc.inc_laddr = inc->inc_laddr; 1400 } 1401 sc->sc_irs = th->th_seq - 1; 1402 sc->sc_iss = th->th_ack - 1; 1403 wnd = sbspace(&so->so_rcv); 1404 wnd = imax(wnd, 0); 1405 wnd = imin(wnd, TCP_MAXWIN); 1406 sc->sc_wnd = wnd; 1407 sc->sc_flags = 0; 1408 sc->sc_rxtslot = 0; 1409 sc->sc_peer_mss = tcp_msstab[data]; 1410 return (sc); 1411} 1412