tcp_syncache.c revision 150131
1/*- 2 * Copyright (c) 2001 McAfee, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Jonathan Lemon 6 * and McAfee Research, the Security Research Division of McAfee, Inc. under 7 * 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * $FreeBSD: head/sys/netinet/tcp_syncache.c 150131 2005-09-14 15:06:22Z andre $ 32 */ 33 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_sack.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/malloc.h> 46#include <sys/mac.h> 47#include <sys/mbuf.h> 48#include <sys/md5.h> 49#include <sys/proc.h> /* for proc0 declaration */ 50#include <sys/random.h> 51#include <sys/socket.h> 52#include <sys/socketvar.h> 53 54#include <net/if.h> 55#include <net/route.h> 56 57#include <netinet/in.h> 58#include <netinet/in_systm.h> 59#include <netinet/ip.h> 60#include <netinet/in_var.h> 61#include <netinet/in_pcb.h> 62#include <netinet/ip_var.h> 63#ifdef INET6 64#include <netinet/ip6.h> 65#include <netinet/icmp6.h> 66#include <netinet6/nd6.h> 67#include <netinet6/ip6_var.h> 68#include <netinet6/in6_pcb.h> 69#endif 70#include <netinet/tcp.h> 71#ifdef TCPDEBUG 72#include <netinet/tcpip.h> 73#endif 74#include <netinet/tcp_fsm.h> 75#include <netinet/tcp_seq.h> 76#include <netinet/tcp_timer.h> 77#include <netinet/tcp_var.h> 78#ifdef TCPDEBUG 79#include <netinet/tcp_debug.h> 80#endif 81#ifdef INET6 82#include <netinet6/tcp6_var.h> 83#endif 84 85#ifdef IPSEC 86#include <netinet6/ipsec.h> 87#ifdef INET6 88#include <netinet6/ipsec6.h> 89#endif 90#endif /*IPSEC*/ 91 92#ifdef FAST_IPSEC 93#include <netipsec/ipsec.h> 94#ifdef INET6 95#include <netipsec/ipsec6.h> 96#endif 97#include <netipsec/key.h> 98#endif /*FAST_IPSEC*/ 99 100#include <machine/in_cksum.h> 101#include <vm/uma.h> 102 103static int tcp_syncookies = 1; 104SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 105 &tcp_syncookies, 0, 106 "Use TCP SYN cookies if the syncache overflows"); 107 108static void syncache_drop(struct syncache *, struct syncache_head *); 109static void syncache_free(struct syncache *); 110static void syncache_insert(struct syncache *, struct syncache_head *); 111struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 112#ifdef TCPDEBUG 113static int syncache_respond(struct syncache *, struct mbuf *, struct socket *); 114#else 115static int syncache_respond(struct syncache *, struct mbuf *); 116#endif 117static struct socket *syncache_socket(struct syncache *, struct socket *, 118 struct mbuf *m); 119static void syncache_timer(void *); 120static u_int32_t syncookie_generate(struct syncache *, u_int32_t *); 121static struct syncache *syncookie_lookup(struct in_conninfo *, 122 struct tcphdr *, struct socket *); 123 124/* 125 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 126 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 127 * the odds are that the user has given up attempting to connect by then. 128 */ 129#define SYNCACHE_MAXREXMTS 3 130 131/* Arbitrary values */ 132#define TCP_SYNCACHE_HASHSIZE 512 133#define TCP_SYNCACHE_BUCKETLIMIT 30 134 135struct tcp_syncache { 136 struct syncache_head *hashbase; 137 uma_zone_t zone; 138 u_int hashsize; 139 u_int hashmask; 140 u_int bucket_limit; 141 u_int cache_count; 142 u_int cache_limit; 143 u_int rexmt_limit; 144 u_int hash_secret; 145 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 146 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 147}; 148static struct tcp_syncache tcp_syncache; 149 150SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 151 152SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 153 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 154 155SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 156 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 157 158SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 159 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 160 161SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 162 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 163 164SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 165 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 166 167static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 168 169#define SYNCACHE_HASH(inc, mask) \ 170 ((tcp_syncache.hash_secret ^ \ 171 (inc)->inc_faddr.s_addr ^ \ 172 ((inc)->inc_faddr.s_addr >> 16) ^ \ 173 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 174 175#define SYNCACHE_HASH6(inc, mask) \ 176 ((tcp_syncache.hash_secret ^ \ 177 (inc)->inc6_faddr.s6_addr32[0] ^ \ 178 (inc)->inc6_faddr.s6_addr32[3] ^ \ 179 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 180 181#define ENDPTS_EQ(a, b) ( \ 182 (a)->ie_fport == (b)->ie_fport && \ 183 (a)->ie_lport == (b)->ie_lport && \ 184 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 185 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 186) 187 188#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 189 190#define SYNCACHE_TIMEOUT(sc, slot) do { \ 191 sc->sc_rxtslot = (slot); \ 192 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[(slot)]; \ 193 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[(slot)], sc, sc_timerq); \ 194 if (!callout_active(&tcp_syncache.tt_timerq[(slot)])) \ 195 callout_reset(&tcp_syncache.tt_timerq[(slot)], \ 196 TCPTV_RTOBASE * tcp_backoff[(slot)], \ 197 syncache_timer, (void *)((intptr_t)(slot))); \ 198} while (0) 199 200static void 201syncache_free(struct syncache *sc) 202{ 203 if (sc->sc_ipopts) 204 (void) m_free(sc->sc_ipopts); 205 206 uma_zfree(tcp_syncache.zone, sc); 207} 208 209void 210syncache_init(void) 211{ 212 int i; 213 214 tcp_syncache.cache_count = 0; 215 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 216 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 217 tcp_syncache.cache_limit = 218 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 219 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 220 tcp_syncache.hash_secret = arc4random(); 221 222 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 223 &tcp_syncache.hashsize); 224 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 225 &tcp_syncache.cache_limit); 226 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 227 &tcp_syncache.bucket_limit); 228 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 229 printf("WARNING: syncache hash size is not a power of 2.\n"); 230 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 231 } 232 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 233 234 /* Allocate the hash table. */ 235 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 236 tcp_syncache.hashsize * sizeof(struct syncache_head), 237 M_SYNCACHE, M_WAITOK); 238 239 /* Initialize the hash buckets. */ 240 for (i = 0; i < tcp_syncache.hashsize; i++) { 241 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 242 tcp_syncache.hashbase[i].sch_length = 0; 243 } 244 245 /* Initialize the timer queues. */ 246 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 247 TAILQ_INIT(&tcp_syncache.timerq[i]); 248 callout_init(&tcp_syncache.tt_timerq[i], NET_CALLOUT_MPSAFE); 249 } 250 251 /* 252 * Allocate the syncache entries. Allow the zone to allocate one 253 * more entry than cache limit, so a new entry can bump out an 254 * older one. 255 */ 256 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 257 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 258 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 259 tcp_syncache.cache_limit -= 1; 260} 261 262static void 263syncache_insert(sc, sch) 264 struct syncache *sc; 265 struct syncache_head *sch; 266{ 267 struct syncache *sc2; 268 int i; 269 270 INP_INFO_WLOCK_ASSERT(&tcbinfo); 271 272 /* 273 * Make sure that we don't overflow the per-bucket 274 * limit or the total cache size limit. 275 */ 276 if (sch->sch_length >= tcp_syncache.bucket_limit) { 277 /* 278 * The bucket is full, toss the oldest element. 279 */ 280 sc2 = TAILQ_FIRST(&sch->sch_bucket); 281 sc2->sc_tp->ts_recent = ticks; 282 syncache_drop(sc2, sch); 283 tcpstat.tcps_sc_bucketoverflow++; 284 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) { 285 /* 286 * The cache is full. Toss the oldest entry in the 287 * entire cache. This is the front entry in the 288 * first non-empty timer queue with the largest 289 * timeout value. 290 */ 291 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 292 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]); 293 if (sc2 != NULL) 294 break; 295 } 296 sc2->sc_tp->ts_recent = ticks; 297 syncache_drop(sc2, NULL); 298 tcpstat.tcps_sc_cacheoverflow++; 299 } 300 301 /* Initialize the entry's timer. */ 302 SYNCACHE_TIMEOUT(sc, 0); 303 304 /* Put it into the bucket. */ 305 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 306 sch->sch_length++; 307 tcp_syncache.cache_count++; 308 tcpstat.tcps_sc_added++; 309} 310 311static void 312syncache_drop(sc, sch) 313 struct syncache *sc; 314 struct syncache_head *sch; 315{ 316 INP_INFO_WLOCK_ASSERT(&tcbinfo); 317 318 if (sch == NULL) { 319#ifdef INET6 320 if (sc->sc_inc.inc_isipv6) { 321 sch = &tcp_syncache.hashbase[ 322 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 323 } else 324#endif 325 { 326 sch = &tcp_syncache.hashbase[ 327 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 328 } 329 } 330 331 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 332 sch->sch_length--; 333 tcp_syncache.cache_count--; 334 335 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq); 336 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot])) 337 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]); 338 339 syncache_free(sc); 340} 341 342/* 343 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 344 * If we have retransmitted an entry the maximum number of times, expire it. 345 */ 346static void 347syncache_timer(xslot) 348 void *xslot; 349{ 350 intptr_t slot = (intptr_t)xslot; 351 struct syncache *sc, *nsc; 352 struct inpcb *inp; 353 354 INP_INFO_WLOCK(&tcbinfo); 355 if (callout_pending(&tcp_syncache.tt_timerq[slot]) || 356 !callout_active(&tcp_syncache.tt_timerq[slot])) { 357 /* XXX can this happen? */ 358 INP_INFO_WUNLOCK(&tcbinfo); 359 return; 360 } 361 callout_deactivate(&tcp_syncache.tt_timerq[slot]); 362 363 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]); 364 while (nsc != NULL) { 365 if (ticks < nsc->sc_rxttime) 366 break; 367 sc = nsc; 368 inp = sc->sc_tp->t_inpcb; 369 if (slot == SYNCACHE_MAXREXMTS || 370 slot >= tcp_syncache.rexmt_limit || 371 inp == NULL || inp->inp_gencnt != sc->sc_inp_gencnt) { 372 nsc = TAILQ_NEXT(sc, sc_timerq); 373 syncache_drop(sc, NULL); 374 tcpstat.tcps_sc_stale++; 375 continue; 376 } 377 /* 378 * syncache_respond() may call back into the syncache to 379 * to modify another entry, so do not obtain the next 380 * entry on the timer chain until it has completed. 381 */ 382#ifdef TCPDEBUG 383 (void) syncache_respond(sc, NULL, NULL); 384#else 385 (void) syncache_respond(sc, NULL); 386#endif 387 nsc = TAILQ_NEXT(sc, sc_timerq); 388 tcpstat.tcps_sc_retransmitted++; 389 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq); 390 SYNCACHE_TIMEOUT(sc, slot + 1); 391 } 392 if (nsc != NULL) 393 callout_reset(&tcp_syncache.tt_timerq[slot], 394 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot)); 395 INP_INFO_WUNLOCK(&tcbinfo); 396} 397 398/* 399 * Find an entry in the syncache. 400 */ 401struct syncache * 402syncache_lookup(inc, schp) 403 struct in_conninfo *inc; 404 struct syncache_head **schp; 405{ 406 struct syncache *sc; 407 struct syncache_head *sch; 408 409 INP_INFO_WLOCK_ASSERT(&tcbinfo); 410 411#ifdef INET6 412 if (inc->inc_isipv6) { 413 sch = &tcp_syncache.hashbase[ 414 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 415 *schp = sch; 416 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 417 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 418 return (sc); 419 } 420 } else 421#endif 422 { 423 sch = &tcp_syncache.hashbase[ 424 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 425 *schp = sch; 426 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 427#ifdef INET6 428 if (sc->sc_inc.inc_isipv6) 429 continue; 430#endif 431 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 432 return (sc); 433 } 434 } 435 return (NULL); 436} 437 438/* 439 * This function is called when we get a RST for a 440 * non-existent connection, so that we can see if the 441 * connection is in the syn cache. If it is, zap it. 442 */ 443void 444syncache_chkrst(inc, th) 445 struct in_conninfo *inc; 446 struct tcphdr *th; 447{ 448 struct syncache *sc; 449 struct syncache_head *sch; 450 451 INP_INFO_WLOCK_ASSERT(&tcbinfo); 452 453 sc = syncache_lookup(inc, &sch); 454 if (sc == NULL) 455 return; 456 /* 457 * If the RST bit is set, check the sequence number to see 458 * if this is a valid reset segment. 459 * RFC 793 page 37: 460 * In all states except SYN-SENT, all reset (RST) segments 461 * are validated by checking their SEQ-fields. A reset is 462 * valid if its sequence number is in the window. 463 * 464 * The sequence number in the reset segment is normally an 465 * echo of our outgoing acknowlegement numbers, but some hosts 466 * send a reset with the sequence number at the rightmost edge 467 * of our receive window, and we have to handle this case. 468 */ 469 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 470 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 471 syncache_drop(sc, sch); 472 tcpstat.tcps_sc_reset++; 473 } 474} 475 476void 477syncache_badack(inc) 478 struct in_conninfo *inc; 479{ 480 struct syncache *sc; 481 struct syncache_head *sch; 482 483 INP_INFO_WLOCK_ASSERT(&tcbinfo); 484 485 sc = syncache_lookup(inc, &sch); 486 if (sc != NULL) { 487 syncache_drop(sc, sch); 488 tcpstat.tcps_sc_badack++; 489 } 490} 491 492void 493syncache_unreach(inc, th) 494 struct in_conninfo *inc; 495 struct tcphdr *th; 496{ 497 struct syncache *sc; 498 struct syncache_head *sch; 499 500 INP_INFO_WLOCK_ASSERT(&tcbinfo); 501 502 sc = syncache_lookup(inc, &sch); 503 if (sc == NULL) 504 return; 505 506 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 507 if (ntohl(th->th_seq) != sc->sc_iss) 508 return; 509 510 /* 511 * If we've rertransmitted 3 times and this is our second error, 512 * we remove the entry. Otherwise, we allow it to continue on. 513 * This prevents us from incorrectly nuking an entry during a 514 * spurious network outage. 515 * 516 * See tcp_notify(). 517 */ 518 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 519 sc->sc_flags |= SCF_UNREACH; 520 return; 521 } 522 syncache_drop(sc, sch); 523 tcpstat.tcps_sc_unreach++; 524} 525 526/* 527 * Build a new TCP socket structure from a syncache entry. 528 */ 529static struct socket * 530syncache_socket(sc, lso, m) 531 struct syncache *sc; 532 struct socket *lso; 533 struct mbuf *m; 534{ 535 struct inpcb *inp = NULL; 536 struct socket *so; 537 struct tcpcb *tp; 538 539 NET_ASSERT_GIANT(); 540 INP_INFO_WLOCK_ASSERT(&tcbinfo); 541 542 /* 543 * Ok, create the full blown connection, and set things up 544 * as they would have been set up if we had created the 545 * connection when the SYN arrived. If we can't create 546 * the connection, abort it. 547 */ 548 so = sonewconn(lso, SS_ISCONNECTED); 549 if (so == NULL) { 550 /* 551 * Drop the connection; we will send a RST if the peer 552 * retransmits the ACK, 553 */ 554 tcpstat.tcps_listendrop++; 555 goto abort2; 556 } 557#ifdef MAC 558 SOCK_LOCK(so); 559 mac_set_socket_peer_from_mbuf(m, so); 560 SOCK_UNLOCK(so); 561#endif 562 563 inp = sotoinpcb(so); 564 INP_LOCK(inp); 565 566 /* 567 * Insert new socket into hash list. 568 */ 569 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 570#ifdef INET6 571 if (sc->sc_inc.inc_isipv6) { 572 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 573 } else { 574 inp->inp_vflag &= ~INP_IPV6; 575 inp->inp_vflag |= INP_IPV4; 576#endif 577 inp->inp_laddr = sc->sc_inc.inc_laddr; 578#ifdef INET6 579 } 580#endif 581 inp->inp_lport = sc->sc_inc.inc_lport; 582 if (in_pcbinshash(inp) != 0) { 583 /* 584 * Undo the assignments above if we failed to 585 * put the PCB on the hash lists. 586 */ 587#ifdef INET6 588 if (sc->sc_inc.inc_isipv6) 589 inp->in6p_laddr = in6addr_any; 590 else 591#endif 592 inp->inp_laddr.s_addr = INADDR_ANY; 593 inp->inp_lport = 0; 594 goto abort; 595 } 596#ifdef IPSEC 597 /* copy old policy into new socket's */ 598 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 599 printf("syncache_expand: could not copy policy\n"); 600#endif 601#ifdef FAST_IPSEC 602 /* copy old policy into new socket's */ 603 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 604 printf("syncache_expand: could not copy policy\n"); 605#endif 606#ifdef INET6 607 if (sc->sc_inc.inc_isipv6) { 608 struct inpcb *oinp = sotoinpcb(lso); 609 struct in6_addr laddr6; 610 struct sockaddr_in6 sin6; 611 /* 612 * Inherit socket options from the listening socket. 613 * Note that in6p_inputopts are not (and should not be) 614 * copied, since it stores previously received options and is 615 * used to detect if each new option is different than the 616 * previous one and hence should be passed to a user. 617 * If we copied in6p_inputopts, a user would not be able to 618 * receive options just after calling the accept system call. 619 */ 620 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 621 if (oinp->in6p_outputopts) 622 inp->in6p_outputopts = 623 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 624 625 sin6.sin6_family = AF_INET6; 626 sin6.sin6_len = sizeof(sin6); 627 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 628 sin6.sin6_port = sc->sc_inc.inc_fport; 629 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 630 laddr6 = inp->in6p_laddr; 631 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 632 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 633 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 634 thread0.td_ucred)) { 635 inp->in6p_laddr = laddr6; 636 goto abort; 637 } 638 /* Override flowlabel from in6_pcbconnect. */ 639 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 640 inp->in6p_flowinfo |= sc->sc_flowlabel; 641 } else 642#endif 643 { 644 struct in_addr laddr; 645 struct sockaddr_in sin; 646 647 inp->inp_options = ip_srcroute(m); 648 if (inp->inp_options == NULL) { 649 inp->inp_options = sc->sc_ipopts; 650 sc->sc_ipopts = NULL; 651 } 652 653 sin.sin_family = AF_INET; 654 sin.sin_len = sizeof(sin); 655 sin.sin_addr = sc->sc_inc.inc_faddr; 656 sin.sin_port = sc->sc_inc.inc_fport; 657 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 658 laddr = inp->inp_laddr; 659 if (inp->inp_laddr.s_addr == INADDR_ANY) 660 inp->inp_laddr = sc->sc_inc.inc_laddr; 661 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 662 thread0.td_ucred)) { 663 inp->inp_laddr = laddr; 664 goto abort; 665 } 666 } 667 668 tp = intotcpcb(inp); 669 tp->t_state = TCPS_SYN_RECEIVED; 670 tp->iss = sc->sc_iss; 671 tp->irs = sc->sc_irs; 672 tcp_rcvseqinit(tp); 673 tcp_sendseqinit(tp); 674 tp->snd_wl1 = sc->sc_irs; 675 tp->rcv_up = sc->sc_irs + 1; 676 tp->rcv_wnd = sc->sc_wnd; 677 tp->rcv_adv += tp->rcv_wnd; 678 679 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 680 if (sc->sc_flags & SCF_NOOPT) 681 tp->t_flags |= TF_NOOPT; 682 if (sc->sc_flags & SCF_WINSCALE) { 683 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 684 tp->requested_s_scale = sc->sc_requested_s_scale; 685 tp->request_r_scale = sc->sc_request_r_scale; 686 } 687 if (sc->sc_flags & SCF_TIMESTAMP) { 688 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 689 tp->ts_recent = sc->sc_tsrecent; 690 tp->ts_recent_age = ticks; 691 } 692#ifdef TCP_SIGNATURE 693 if (sc->sc_flags & SCF_SIGNATURE) 694 tp->t_flags |= TF_SIGNATURE; 695#endif 696 if (sc->sc_flags & SCF_SACK) { 697 tp->sack_enable = 1; 698 tp->t_flags |= TF_SACK_PERMIT; 699 } 700 /* 701 * Set up MSS and get cached values from tcp_hostcache. 702 * This might overwrite some of the defaults we just set. 703 */ 704 tcp_mss(tp, sc->sc_peer_mss); 705 706 /* 707 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 708 */ 709 if (sc->sc_rxtslot != 0) 710 tp->snd_cwnd = tp->t_maxseg; 711 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 712 713 INP_UNLOCK(inp); 714 715 tcpstat.tcps_accepts++; 716 return (so); 717 718abort: 719 INP_UNLOCK(inp); 720abort2: 721 if (so != NULL) 722 (void) soabort(so); 723 return (NULL); 724} 725 726/* 727 * This function gets called when we receive an ACK for a 728 * socket in the LISTEN state. We look up the connection 729 * in the syncache, and if its there, we pull it out of 730 * the cache and turn it into a full-blown connection in 731 * the SYN-RECEIVED state. 732 */ 733int 734syncache_expand(inc, th, sop, m) 735 struct in_conninfo *inc; 736 struct tcphdr *th; 737 struct socket **sop; 738 struct mbuf *m; 739{ 740 struct syncache *sc; 741 struct syncache_head *sch; 742 struct socket *so; 743 744 INP_INFO_WLOCK_ASSERT(&tcbinfo); 745 746 sc = syncache_lookup(inc, &sch); 747 if (sc == NULL) { 748 /* 749 * There is no syncache entry, so see if this ACK is 750 * a returning syncookie. To do this, first: 751 * A. See if this socket has had a syncache entry dropped in 752 * the past. We don't want to accept a bogus syncookie 753 * if we've never received a SYN. 754 * B. check that the syncookie is valid. If it is, then 755 * cobble up a fake syncache entry, and return. 756 */ 757 if (!tcp_syncookies) 758 return (0); 759 sc = syncookie_lookup(inc, th, *sop); 760 if (sc == NULL) 761 return (0); 762 sch = NULL; 763 tcpstat.tcps_sc_recvcookie++; 764 } 765 766 /* 767 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 768 */ 769 if (th->th_ack != sc->sc_iss + 1) 770 return (0); 771 772 so = syncache_socket(sc, *sop, m); 773 if (so == NULL) { 774#if 0 775resetandabort: 776 /* XXXjlemon check this - is this correct? */ 777 (void) tcp_respond(NULL, m, m, th, 778 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 779#endif 780 m_freem(m); /* XXX only needed for above */ 781 tcpstat.tcps_sc_aborted++; 782 } else 783 tcpstat.tcps_sc_completed++; 784 785 if (sch == NULL) 786 syncache_free(sc); 787 else 788 syncache_drop(sc, sch); 789 *sop = so; 790 return (1); 791} 792 793/* 794 * Given a LISTEN socket and an inbound SYN request, add 795 * this to the syn cache, and send back a segment: 796 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 797 * to the source. 798 * 799 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 800 * Doing so would require that we hold onto the data and deliver it 801 * to the application. However, if we are the target of a SYN-flood 802 * DoS attack, an attacker could send data which would eventually 803 * consume all available buffer space if it were ACKed. By not ACKing 804 * the data, we avoid this DoS scenario. 805 */ 806int 807syncache_add(inc, to, th, sop, m) 808 struct in_conninfo *inc; 809 struct tcpopt *to; 810 struct tcphdr *th; 811 struct socket **sop; 812 struct mbuf *m; 813{ 814 struct tcpcb *tp; 815 struct socket *so; 816 struct syncache *sc = NULL; 817 struct syncache_head *sch; 818 struct mbuf *ipopts = NULL; 819 u_int32_t flowtmp; 820 int i, win; 821 822 INP_INFO_WLOCK_ASSERT(&tcbinfo); 823 824 so = *sop; 825 tp = sototcpcb(so); 826 827 /* 828 * Remember the IP options, if any. 829 */ 830#ifdef INET6 831 if (!inc->inc_isipv6) 832#endif 833 ipopts = ip_srcroute(m); 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 sc->sc_flowlabel = 0; 929 if (tcp_syncookies) { 930 sc->sc_iss = syncookie_generate(sc, &flowtmp); 931#ifdef INET6 932 if (inc->inc_isipv6 && 933 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 934 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK; 935 } 936#endif 937 } else { 938 sc->sc_iss = arc4random(); 939#ifdef INET6 940 if (inc->inc_isipv6 && 941 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 942 sc->sc_flowlabel = 943 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 944 } 945#endif 946 } 947 948 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 949 win = sbspace(&so->so_rcv); 950 win = imax(win, 0); 951 win = imin(win, TCP_MAXWIN); 952 sc->sc_wnd = win; 953 954 if (tcp_do_rfc1323) { 955 /* 956 * A timestamp received in a SYN makes 957 * it ok to send timestamp requests and replies. 958 */ 959 if (to->to_flags & TOF_TS) { 960 sc->sc_tsrecent = to->to_tsval; 961 sc->sc_flags |= SCF_TIMESTAMP; 962 } 963 if (to->to_flags & TOF_SCALE) { 964 int wscale = 0; 965 966 /* Compute proper scaling value from buffer space */ 967 while (wscale < TCP_MAX_WINSHIFT && 968 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 969 wscale++; 970 sc->sc_request_r_scale = wscale; 971 sc->sc_requested_s_scale = to->to_requested_s_scale; 972 sc->sc_flags |= SCF_WINSCALE; 973 } 974 } 975 if (tp->t_flags & TF_NOOPT) 976 sc->sc_flags = SCF_NOOPT; 977#ifdef TCP_SIGNATURE 978 /* 979 * If listening socket requested TCP digests, and received SYN 980 * contains the option, flag this in the syncache so that 981 * syncache_respond() will do the right thing with the SYN+ACK. 982 * XXX Currently we always record the option by default and will 983 * attempt to use it in syncache_respond(). 984 */ 985 if (to->to_flags & TOF_SIGNATURE) 986 sc->sc_flags |= SCF_SIGNATURE; 987#endif 988 989 if (to->to_flags & TOF_SACK) 990 sc->sc_flags |= SCF_SACK; 991 992 /* 993 * Do a standard 3-way handshake. 994 */ 995#ifdef TCPDEBUG 996 if (syncache_respond(sc, m, so) == 0) { 997#else 998 if (syncache_respond(sc, m) == 0) { 999#endif 1000 syncache_insert(sc, sch); 1001 tcpstat.tcps_sndacks++; 1002 tcpstat.tcps_sndtotal++; 1003 } else { 1004 syncache_free(sc); 1005 tcpstat.tcps_sc_dropped++; 1006 } 1007 *sop = NULL; 1008 return (1); 1009} 1010 1011#ifdef TCPDEBUG 1012static int 1013syncache_respond(sc, m, so) 1014 struct syncache *sc; 1015 struct mbuf *m; 1016 struct socket *so; 1017#else 1018static int 1019syncache_respond(sc, m) 1020 struct syncache *sc; 1021 struct mbuf *m; 1022#endif 1023{ 1024 u_int8_t *optp; 1025 int optlen, error; 1026 u_int16_t tlen, hlen, mssopt; 1027 struct ip *ip = NULL; 1028 struct tcphdr *th; 1029 struct inpcb *inp; 1030#ifdef INET6 1031 struct ip6_hdr *ip6 = NULL; 1032#endif 1033 1034 hlen = 1035#ifdef INET6 1036 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1037#endif 1038 sizeof(struct ip); 1039 1040 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1041 1042 /* Determine MSS we advertize to other end of connection */ 1043 mssopt = tcp_mssopt(&sc->sc_inc); 1044 1045 /* Compute the size of the TCP options. */ 1046 if (sc->sc_flags & SCF_NOOPT) { 1047 optlen = 0; 1048 } else { 1049 optlen = TCPOLEN_MAXSEG + 1050 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1051 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 1052#ifdef TCP_SIGNATURE 1053 if (sc->sc_flags & SCF_SIGNATURE) 1054 optlen += TCPOLEN_SIGNATURE; 1055#endif 1056 if (sc->sc_flags & SCF_SACK) 1057 optlen += TCPOLEN_SACK_PERMITTED; 1058 optlen = roundup2(optlen, 4); 1059 } 1060 tlen = hlen + sizeof(struct tcphdr) + optlen; 1061 1062 /* 1063 * XXX 1064 * assume that the entire packet will fit in a header mbuf 1065 */ 1066 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1067 1068 /* 1069 * XXX shouldn't this reuse the mbuf if possible ? 1070 * Create the IP+TCP header from scratch. 1071 */ 1072 if (m) 1073 m_freem(m); 1074 1075 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1076 if (m == NULL) 1077 return (ENOBUFS); 1078 m->m_data += max_linkhdr; 1079 m->m_len = tlen; 1080 m->m_pkthdr.len = tlen; 1081 m->m_pkthdr.rcvif = NULL; 1082 inp = sc->sc_tp->t_inpcb; 1083 INP_LOCK(inp); 1084#ifdef MAC 1085 mac_create_mbuf_from_inpcb(inp, m); 1086#endif 1087 1088#ifdef INET6 1089 if (sc->sc_inc.inc_isipv6) { 1090 ip6 = mtod(m, struct ip6_hdr *); 1091 ip6->ip6_vfc = IPV6_VERSION; 1092 ip6->ip6_nxt = IPPROTO_TCP; 1093 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1094 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1095 ip6->ip6_plen = htons(tlen - hlen); 1096 /* ip6_hlim is set after checksum */ 1097 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1098 ip6->ip6_flow |= sc->sc_flowlabel; 1099 1100 th = (struct tcphdr *)(ip6 + 1); 1101 } else 1102#endif 1103 { 1104 ip = mtod(m, struct ip *); 1105 ip->ip_v = IPVERSION; 1106 ip->ip_hl = sizeof(struct ip) >> 2; 1107 ip->ip_len = tlen; 1108 ip->ip_id = 0; 1109 ip->ip_off = 0; 1110 ip->ip_sum = 0; 1111 ip->ip_p = IPPROTO_TCP; 1112 ip->ip_src = sc->sc_inc.inc_laddr; 1113 ip->ip_dst = sc->sc_inc.inc_faddr; 1114 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1115 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 1116 1117 /* 1118 * See if we should do MTU discovery. Route lookups are 1119 * expensive, so we will only unset the DF bit if: 1120 * 1121 * 1) path_mtu_discovery is disabled 1122 * 2) the SCF_UNREACH flag has been set 1123 */ 1124 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1125 ip->ip_off |= IP_DF; 1126 1127 th = (struct tcphdr *)(ip + 1); 1128 } 1129 th->th_sport = sc->sc_inc.inc_lport; 1130 th->th_dport = sc->sc_inc.inc_fport; 1131 1132 th->th_seq = htonl(sc->sc_iss); 1133 th->th_ack = htonl(sc->sc_irs + 1); 1134 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1135 th->th_x2 = 0; 1136 th->th_flags = TH_SYN|TH_ACK; 1137 th->th_win = htons(sc->sc_wnd); 1138 th->th_urp = 0; 1139 1140 /* Tack on the TCP options. */ 1141 if (optlen != 0) { 1142 optp = (u_int8_t *)(th + 1); 1143 *optp++ = TCPOPT_MAXSEG; 1144 *optp++ = TCPOLEN_MAXSEG; 1145 *optp++ = (mssopt >> 8) & 0xff; 1146 *optp++ = mssopt & 0xff; 1147 1148 if (sc->sc_flags & SCF_WINSCALE) { 1149 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1150 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1151 sc->sc_request_r_scale); 1152 optp += 4; 1153 } 1154 1155 if (sc->sc_flags & SCF_TIMESTAMP) { 1156 u_int32_t *lp = (u_int32_t *)(optp); 1157 1158 /* Form timestamp option per appendix A of RFC 1323. */ 1159 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1160 *lp++ = htonl(ticks); 1161 *lp = htonl(sc->sc_tsrecent); 1162 optp += TCPOLEN_TSTAMP_APPA; 1163 } 1164 1165#ifdef TCP_SIGNATURE 1166 /* 1167 * Handle TCP-MD5 passive opener response. 1168 */ 1169 if (sc->sc_flags & SCF_SIGNATURE) { 1170 u_int8_t *bp = optp; 1171 int i; 1172 1173 *bp++ = TCPOPT_SIGNATURE; 1174 *bp++ = TCPOLEN_SIGNATURE; 1175 for (i = 0; i < TCP_SIGLEN; i++) 1176 *bp++ = 0; 1177 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1178 optp + 2, IPSEC_DIR_OUTBOUND); 1179 optp += TCPOLEN_SIGNATURE; 1180 } 1181#endif /* TCP_SIGNATURE */ 1182 1183 if (sc->sc_flags & SCF_SACK) { 1184 *optp++ = TCPOPT_SACK_PERMITTED; 1185 *optp++ = TCPOLEN_SACK_PERMITTED; 1186 } 1187 1188 { 1189 /* Pad TCP options to a 4 byte boundary */ 1190 int padlen = optlen - (optp - (u_int8_t *)(th + 1)); 1191 while (padlen-- > 0) 1192 *optp++ = TCPOPT_EOL; 1193 } 1194 } 1195 1196#ifdef INET6 1197 if (sc->sc_inc.inc_isipv6) { 1198 th->th_sum = 0; 1199 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1200 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1201 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1202 } else 1203#endif 1204 { 1205 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1206 htons(tlen - hlen + IPPROTO_TCP)); 1207 m->m_pkthdr.csum_flags = CSUM_TCP; 1208 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1209#ifdef TCPDEBUG 1210 /* 1211 * Trace. 1212 */ 1213 if (so != NULL && so->so_options & SO_DEBUG) { 1214 struct tcpcb *tp = sototcpcb(so); 1215 tcp_trace(TA_OUTPUT, tp->t_state, tp, 1216 mtod(m, void *), th, 0); 1217 } 1218#endif 1219 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp); 1220 } 1221 INP_UNLOCK(inp); 1222 return (error); 1223} 1224 1225/* 1226 * cookie layers: 1227 * 1228 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1229 * | peer iss | 1230 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1231 * | 0 |(A)| | 1232 * (A): peer mss index 1233 */ 1234 1235/* 1236 * The values below are chosen to minimize the size of the tcp_secret 1237 * table, as well as providing roughly a 16 second lifetime for the cookie. 1238 */ 1239 1240#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1241#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1242 1243#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1244#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1245#define SYNCOOKIE_TIMEOUT \ 1246 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1247#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1248 1249static struct { 1250 u_int32_t ts_secbits[4]; 1251 u_int ts_expire; 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 u_int32_t 1281syncookie_generate(struct syncache *sc, u_int32_t *flowid) 1282{ 1283 u_int32_t md5_buffer[4]; 1284 u_int32_t data; 1285 int idx, i; 1286 struct md5_add add; 1287 1288 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1289 1290 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1291 if (tcp_secret[idx].ts_expire < ticks) { 1292 for (i = 0; i < 4; i++) 1293 tcp_secret[idx].ts_secbits[i] = arc4random(); 1294 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1295 } 1296 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1297 if (tcp_msstab[data] <= sc->sc_peer_mss) 1298 break; 1299 data = (data << SYNCOOKIE_WNDBITS) | idx; 1300 data ^= sc->sc_irs; /* peer's iss */ 1301 MD5Init(&syn_ctx); 1302#ifdef INET6 1303 if (sc->sc_inc.inc_isipv6) { 1304 MD5Add(sc->sc_inc.inc6_laddr); 1305 MD5Add(sc->sc_inc.inc6_faddr); 1306 add.laddr = 0; 1307 add.faddr = 0; 1308 } else 1309#endif 1310 { 1311 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1312 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1313 } 1314 add.lport = sc->sc_inc.inc_lport; 1315 add.fport = sc->sc_inc.inc_fport; 1316 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1317 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1318 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1319 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1320 MD5Add(add); 1321 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1322 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1323 *flowid = md5_buffer[1]; 1324 return (data); 1325} 1326 1327static struct syncache * 1328syncookie_lookup(inc, th, so) 1329 struct in_conninfo *inc; 1330 struct tcphdr *th; 1331 struct socket *so; 1332{ 1333 u_int32_t md5_buffer[4]; 1334 struct syncache *sc; 1335 u_int32_t data; 1336 int wnd, idx; 1337 struct md5_add add; 1338 1339 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1340 1341 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1342 idx = data & SYNCOOKIE_WNDMASK; 1343 if (tcp_secret[idx].ts_expire < ticks || 1344 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1345 return (NULL); 1346 MD5Init(&syn_ctx); 1347#ifdef INET6 1348 if (inc->inc_isipv6) { 1349 MD5Add(inc->inc6_laddr); 1350 MD5Add(inc->inc6_faddr); 1351 add.laddr = 0; 1352 add.faddr = 0; 1353 } else 1354#endif 1355 { 1356 add.laddr = inc->inc_laddr.s_addr; 1357 add.faddr = inc->inc_faddr.s_addr; 1358 } 1359 add.lport = inc->inc_lport; 1360 add.fport = inc->inc_fport; 1361 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1362 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1363 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1364 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1365 MD5Add(add); 1366 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1367 data ^= md5_buffer[0]; 1368 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1369 return (NULL); 1370 data = data >> SYNCOOKIE_WNDBITS; 1371 1372 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 1373 if (sc == NULL) 1374 return (NULL); 1375 /* 1376 * Fill in the syncache values. 1377 * XXX duplicate code from syncache_add 1378 */ 1379 sc->sc_ipopts = NULL; 1380 sc->sc_inc.inc_fport = inc->inc_fport; 1381 sc->sc_inc.inc_lport = inc->inc_lport; 1382 sc->sc_tp = sototcpcb(so); 1383#ifdef INET6 1384 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1385 if (inc->inc_isipv6) { 1386 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1387 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1388 if (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL) 1389 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1390 } else 1391#endif 1392 { 1393 sc->sc_inc.inc_faddr = inc->inc_faddr; 1394 sc->sc_inc.inc_laddr = inc->inc_laddr; 1395 } 1396 sc->sc_irs = th->th_seq - 1; 1397 sc->sc_iss = th->th_ack - 1; 1398 wnd = sbspace(&so->so_rcv); 1399 wnd = imax(wnd, 0); 1400 wnd = imin(wnd, TCP_MAXWIN); 1401 sc->sc_wnd = wnd; 1402 sc->sc_flags = 0; 1403 sc->sc_rxtslot = 0; 1404 sc->sc_peer_mss = tcp_msstab[data]; 1405 return (sc); 1406} 1407