tcp_syncache.c revision 154355
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 154355 2006-01-14 13:04:08Z glebius $ 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#include <netinet/ip_options.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 *, u_int32_t *); 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) || tcp_syncache.hashsize == 0) { 230 printf("WARNING: syncache hash size is not a power of 2.\n"); 231 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 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], NET_CALLOUT_MPSAFE); 250 } 251 252 /* 253 * Allocate the syncache entries. Allow the zone to allocate one 254 * more entry than cache limit, so a new entry can bump out an 255 * older one. 256 */ 257 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 258 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 259 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 260 tcp_syncache.cache_limit -= 1; 261} 262 263static void 264syncache_insert(sc, sch) 265 struct syncache *sc; 266 struct syncache_head *sch; 267{ 268 struct syncache *sc2; 269 int i; 270 271 INP_INFO_WLOCK_ASSERT(&tcbinfo); 272 273 /* 274 * Make sure that we don't overflow the per-bucket 275 * limit or the total cache size limit. 276 */ 277 if (sch->sch_length >= tcp_syncache.bucket_limit) { 278 /* 279 * The bucket is full, toss the oldest element. 280 */ 281 sc2 = TAILQ_FIRST(&sch->sch_bucket); 282 sc2->sc_tp->ts_recent = ticks; 283 syncache_drop(sc2, sch); 284 tcpstat.tcps_sc_bucketoverflow++; 285 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) { 286 /* 287 * The cache is full. Toss the oldest entry in the 288 * entire cache. This is the front entry in the 289 * first non-empty timer queue with the largest 290 * timeout value. 291 */ 292 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 293 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]); 294 if (sc2 != NULL) 295 break; 296 } 297 sc2->sc_tp->ts_recent = ticks; 298 syncache_drop(sc2, NULL); 299 tcpstat.tcps_sc_cacheoverflow++; 300 } 301 302 /* Initialize the entry's timer. */ 303 SYNCACHE_TIMEOUT(sc, 0); 304 305 /* Put it into the bucket. */ 306 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 307 sch->sch_length++; 308 tcp_syncache.cache_count++; 309 tcpstat.tcps_sc_added++; 310} 311 312static void 313syncache_drop(sc, sch) 314 struct syncache *sc; 315 struct syncache_head *sch; 316{ 317 INP_INFO_WLOCK_ASSERT(&tcbinfo); 318 319 if (sch == NULL) { 320#ifdef INET6 321 if (sc->sc_inc.inc_isipv6) { 322 sch = &tcp_syncache.hashbase[ 323 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 324 } else 325#endif 326 { 327 sch = &tcp_syncache.hashbase[ 328 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 329 } 330 } 331 332 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 333 sch->sch_length--; 334 tcp_syncache.cache_count--; 335 336 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq); 337 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot])) 338 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]); 339 340 syncache_free(sc); 341} 342 343/* 344 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 345 * If we have retransmitted an entry the maximum number of times, expire it. 346 */ 347static void 348syncache_timer(xslot) 349 void *xslot; 350{ 351 intptr_t slot = (intptr_t)xslot; 352 struct syncache *sc, *nsc; 353 struct inpcb *inp; 354 355 INP_INFO_WLOCK(&tcbinfo); 356 if (callout_pending(&tcp_syncache.tt_timerq[slot]) || 357 !callout_active(&tcp_syncache.tt_timerq[slot])) { 358 /* XXX can this happen? */ 359 INP_INFO_WUNLOCK(&tcbinfo); 360 return; 361 } 362 callout_deactivate(&tcp_syncache.tt_timerq[slot]); 363 364 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]); 365 while (nsc != NULL) { 366 if (ticks < nsc->sc_rxttime) 367 break; 368 sc = nsc; 369 inp = sc->sc_tp->t_inpcb; 370 if (slot == SYNCACHE_MAXREXMTS || 371 slot >= tcp_syncache.rexmt_limit || 372 inp == NULL || inp->inp_gencnt != sc->sc_inp_gencnt) { 373 nsc = TAILQ_NEXT(sc, sc_timerq); 374 syncache_drop(sc, NULL); 375 tcpstat.tcps_sc_stale++; 376 continue; 377 } 378 /* 379 * syncache_respond() may call back into the syncache to 380 * to modify another entry, so do not obtain the next 381 * entry on the timer chain until it has completed. 382 */ 383#ifdef TCPDEBUG 384 (void) syncache_respond(sc, NULL, NULL); 385#else 386 (void) syncache_respond(sc, NULL); 387#endif 388 nsc = TAILQ_NEXT(sc, sc_timerq); 389 tcpstat.tcps_sc_retransmitted++; 390 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq); 391 SYNCACHE_TIMEOUT(sc, slot + 1); 392 } 393 if (nsc != NULL) 394 callout_reset(&tcp_syncache.tt_timerq[slot], 395 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot)); 396 INP_INFO_WUNLOCK(&tcbinfo); 397} 398 399/* 400 * Find an entry in the syncache. 401 */ 402struct syncache * 403syncache_lookup(inc, schp) 404 struct in_conninfo *inc; 405 struct syncache_head **schp; 406{ 407 struct syncache *sc; 408 struct syncache_head *sch; 409 410 INP_INFO_WLOCK_ASSERT(&tcbinfo); 411 412#ifdef INET6 413 if (inc->inc_isipv6) { 414 sch = &tcp_syncache.hashbase[ 415 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 416 *schp = sch; 417 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 418 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 419 return (sc); 420 } 421 } else 422#endif 423 { 424 sch = &tcp_syncache.hashbase[ 425 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 426 *schp = sch; 427 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 428#ifdef INET6 429 if (sc->sc_inc.inc_isipv6) 430 continue; 431#endif 432 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 433 return (sc); 434 } 435 } 436 return (NULL); 437} 438 439/* 440 * This function is called when we get a RST for a 441 * non-existent connection, so that we can see if the 442 * connection is in the syn cache. If it is, zap it. 443 */ 444void 445syncache_chkrst(inc, th) 446 struct in_conninfo *inc; 447 struct tcphdr *th; 448{ 449 struct syncache *sc; 450 struct syncache_head *sch; 451 452 INP_INFO_WLOCK_ASSERT(&tcbinfo); 453 454 sc = syncache_lookup(inc, &sch); 455 if (sc == NULL) 456 return; 457 /* 458 * If the RST bit is set, check the sequence number to see 459 * if this is a valid reset segment. 460 * RFC 793 page 37: 461 * In all states except SYN-SENT, all reset (RST) segments 462 * are validated by checking their SEQ-fields. A reset is 463 * valid if its sequence number is in the window. 464 * 465 * The sequence number in the reset segment is normally an 466 * echo of our outgoing acknowlegement numbers, but some hosts 467 * send a reset with the sequence number at the rightmost edge 468 * of our receive window, and we have to handle this case. 469 */ 470 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 471 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 472 syncache_drop(sc, sch); 473 tcpstat.tcps_sc_reset++; 474 } 475} 476 477void 478syncache_badack(inc) 479 struct in_conninfo *inc; 480{ 481 struct syncache *sc; 482 struct syncache_head *sch; 483 484 INP_INFO_WLOCK_ASSERT(&tcbinfo); 485 486 sc = syncache_lookup(inc, &sch); 487 if (sc != NULL) { 488 syncache_drop(sc, sch); 489 tcpstat.tcps_sc_badack++; 490 } 491} 492 493void 494syncache_unreach(inc, th) 495 struct in_conninfo *inc; 496 struct tcphdr *th; 497{ 498 struct syncache *sc; 499 struct syncache_head *sch; 500 501 INP_INFO_WLOCK_ASSERT(&tcbinfo); 502 503 sc = syncache_lookup(inc, &sch); 504 if (sc == NULL) 505 return; 506 507 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 508 if (ntohl(th->th_seq) != sc->sc_iss) 509 return; 510 511 /* 512 * If we've rertransmitted 3 times and this is our second error, 513 * we remove the entry. Otherwise, we allow it to continue on. 514 * This prevents us from incorrectly nuking an entry during a 515 * spurious network outage. 516 * 517 * See tcp_notify(). 518 */ 519 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 520 sc->sc_flags |= SCF_UNREACH; 521 return; 522 } 523 syncache_drop(sc, sch); 524 tcpstat.tcps_sc_unreach++; 525} 526 527/* 528 * Build a new TCP socket structure from a syncache entry. 529 */ 530static struct socket * 531syncache_socket(sc, lso, m) 532 struct syncache *sc; 533 struct socket *lso; 534 struct mbuf *m; 535{ 536 struct inpcb *inp = NULL; 537 struct socket *so; 538 struct tcpcb *tp; 539 540 NET_ASSERT_GIANT(); 541 INP_INFO_WLOCK_ASSERT(&tcbinfo); 542 543 /* 544 * Ok, create the full blown connection, and set things up 545 * as they would have been set up if we had created the 546 * connection when the SYN arrived. If we can't create 547 * the connection, abort it. 548 */ 549 so = sonewconn(lso, SS_ISCONNECTED); 550 if (so == NULL) { 551 /* 552 * Drop the connection; we will send a RST if the peer 553 * retransmits the ACK, 554 */ 555 tcpstat.tcps_listendrop++; 556 goto abort2; 557 } 558#ifdef MAC 559 SOCK_LOCK(so); 560 mac_set_socket_peer_from_mbuf(m, so); 561 SOCK_UNLOCK(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, 635 thread0.td_ucred)) { 636 inp->in6p_laddr = laddr6; 637 goto abort; 638 } 639 /* Override flowlabel from in6_pcbconnect. */ 640 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 641 inp->in6p_flowinfo |= sc->sc_flowlabel; 642 } else 643#endif 644 { 645 struct in_addr laddr; 646 struct sockaddr_in sin; 647 648 inp->inp_options = ip_srcroute(m); 649 if (inp->inp_options == NULL) { 650 inp->inp_options = sc->sc_ipopts; 651 sc->sc_ipopts = NULL; 652 } 653 654 sin.sin_family = AF_INET; 655 sin.sin_len = sizeof(sin); 656 sin.sin_addr = sc->sc_inc.inc_faddr; 657 sin.sin_port = sc->sc_inc.inc_fport; 658 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 659 laddr = inp->inp_laddr; 660 if (inp->inp_laddr.s_addr == INADDR_ANY) 661 inp->inp_laddr = sc->sc_inc.inc_laddr; 662 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 663 thread0.td_ucred)) { 664 inp->inp_laddr = laddr; 665 goto abort; 666 } 667 } 668 669 tp = intotcpcb(inp); 670 tp->t_state = TCPS_SYN_RECEIVED; 671 tp->iss = sc->sc_iss; 672 tp->irs = sc->sc_irs; 673 tcp_rcvseqinit(tp); 674 tcp_sendseqinit(tp); 675 tp->snd_wl1 = sc->sc_irs; 676 tp->rcv_up = sc->sc_irs + 1; 677 tp->rcv_wnd = sc->sc_wnd; 678 tp->rcv_adv += tp->rcv_wnd; 679 680 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 681 if (sc->sc_flags & SCF_NOOPT) 682 tp->t_flags |= TF_NOOPT; 683 if (sc->sc_flags & SCF_WINSCALE) { 684 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 685 tp->requested_s_scale = sc->sc_requested_s_scale; 686 tp->request_r_scale = sc->sc_request_r_scale; 687 } 688 if (sc->sc_flags & SCF_TIMESTAMP) { 689 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 690 tp->ts_recent = sc->sc_tsrecent; 691 tp->ts_recent_age = ticks; 692 } 693#ifdef TCP_SIGNATURE 694 if (sc->sc_flags & SCF_SIGNATURE) 695 tp->t_flags |= TF_SIGNATURE; 696#endif 697 if (sc->sc_flags & SCF_SACK) { 698 tp->sack_enable = 1; 699 tp->t_flags |= TF_SACK_PERMIT; 700 } 701 /* 702 * Set up MSS and get cached values from tcp_hostcache. 703 * This might overwrite some of the defaults we just set. 704 */ 705 tcp_mss(tp, sc->sc_peer_mss); 706 707 /* 708 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 709 */ 710 if (sc->sc_rxtslot != 0) 711 tp->snd_cwnd = tp->t_maxseg; 712 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 713 714 INP_UNLOCK(inp); 715 716 tcpstat.tcps_accepts++; 717 return (so); 718 719abort: 720 INP_UNLOCK(inp); 721abort2: 722 if (so != NULL) 723 (void) soabort(so); 724 return (NULL); 725} 726 727/* 728 * This function gets called when we receive an ACK for a 729 * socket in the LISTEN state. We look up the connection 730 * in the syncache, and if its there, we pull it out of 731 * the cache and turn it into a full-blown connection in 732 * the SYN-RECEIVED state. 733 */ 734int 735syncache_expand(inc, th, sop, m) 736 struct in_conninfo *inc; 737 struct tcphdr *th; 738 struct socket **sop; 739 struct mbuf *m; 740{ 741 struct syncache *sc; 742 struct syncache_head *sch; 743 struct socket *so; 744 745 INP_INFO_WLOCK_ASSERT(&tcbinfo); 746 747 sc = syncache_lookup(inc, &sch); 748 if (sc == NULL) { 749 /* 750 * There is no syncache entry, so see if this ACK is 751 * a returning syncookie. To do this, first: 752 * A. See if this socket has had a syncache entry dropped in 753 * the past. We don't want to accept a bogus syncookie 754 * if we've never received a SYN. 755 * B. check that the syncookie is valid. If it is, then 756 * cobble up a fake syncache entry, and return. 757 */ 758 if (!tcp_syncookies) 759 return (0); 760 sc = syncookie_lookup(inc, th, *sop); 761 if (sc == NULL) 762 return (0); 763 sch = NULL; 764 tcpstat.tcps_sc_recvcookie++; 765 } 766 767 /* 768 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 769 */ 770 if (th->th_ack != sc->sc_iss + 1) 771 return (0); 772 773 so = syncache_socket(sc, *sop, m); 774 if (so == NULL) { 775#if 0 776resetandabort: 777 /* XXXjlemon check this - is this correct? */ 778 (void) tcp_respond(NULL, m, m, th, 779 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 780#endif 781 m_freem(m); /* XXX only needed for above */ 782 tcpstat.tcps_sc_aborted++; 783 } else 784 tcpstat.tcps_sc_completed++; 785 786 if (sch == NULL) 787 syncache_free(sc); 788 else 789 syncache_drop(sc, sch); 790 *sop = so; 791 return (1); 792} 793 794/* 795 * Given a LISTEN socket and an inbound SYN request, add 796 * this to the syn cache, and send back a segment: 797 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 798 * to the source. 799 * 800 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 801 * Doing so would require that we hold onto the data and deliver it 802 * to the application. However, if we are the target of a SYN-flood 803 * DoS attack, an attacker could send data which would eventually 804 * consume all available buffer space if it were ACKed. By not ACKing 805 * the data, we avoid this DoS scenario. 806 */ 807int 808syncache_add(inc, to, th, sop, m) 809 struct in_conninfo *inc; 810 struct tcpopt *to; 811 struct tcphdr *th; 812 struct socket **sop; 813 struct mbuf *m; 814{ 815 struct tcpcb *tp; 816 struct socket *so; 817 struct syncache *sc = NULL; 818 struct syncache_head *sch; 819 struct mbuf *ipopts = NULL; 820 u_int32_t flowtmp; 821 int i, win; 822 823 INP_INFO_WLOCK_ASSERT(&tcbinfo); 824 825 so = *sop; 826 tp = sototcpcb(so); 827 828 /* 829 * Remember the IP options, if any. 830 */ 831#ifdef INET6 832 if (!inc->inc_isipv6) 833#endif 834 ipopts = ip_srcroute(m); 835 836 /* 837 * See if we already have an entry for this connection. 838 * If we do, resend the SYN,ACK, and reset the retransmit timer. 839 * 840 * XXX 841 * should the syncache be re-initialized with the contents 842 * of the new SYN here (which may have different options?) 843 */ 844 sc = syncache_lookup(inc, &sch); 845 if (sc != NULL) { 846 tcpstat.tcps_sc_dupsyn++; 847 if (ipopts) { 848 /* 849 * If we were remembering a previous source route, 850 * forget it and use the new one we've been given. 851 */ 852 if (sc->sc_ipopts) 853 (void) m_free(sc->sc_ipopts); 854 sc->sc_ipopts = ipopts; 855 } 856 /* 857 * Update timestamp if present. 858 */ 859 if (sc->sc_flags & SCF_TIMESTAMP) 860 sc->sc_tsrecent = to->to_tsval; 861 /* 862 * PCB may have changed, pick up new values. 863 */ 864 sc->sc_tp = tp; 865 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 866#ifdef TCPDEBUG 867 if (syncache_respond(sc, m, so) == 0) { 868#else 869 if (syncache_respond(sc, m) == 0) { 870#endif 871 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 872 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 873 sc, sc_timerq); 874 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 875 tcpstat.tcps_sndacks++; 876 tcpstat.tcps_sndtotal++; 877 } 878 *sop = NULL; 879 return (1); 880 } 881 882 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 883 if (sc == NULL) { 884 /* 885 * The zone allocator couldn't provide more entries. 886 * Treat this as if the cache was full; drop the oldest 887 * entry and insert the new one. 888 */ 889 tcpstat.tcps_sc_zonefail++; 890 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 891 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 892 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]); 893 if (sc != NULL) 894 break; 895 } 896 if (sc == NULL) { 897 /* Generic memory failure. */ 898 if (ipopts) 899 (void) m_free(ipopts); 900 return (0); 901 } 902 sc->sc_tp->ts_recent = ticks; 903 syncache_drop(sc, NULL); 904 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 905 if (sc == NULL) { 906 if (ipopts) 907 (void) m_free(ipopts); 908 return (0); 909 } 910 } 911 912 /* 913 * Fill in the syncache values. 914 */ 915 bzero(sc, sizeof(*sc)); 916 sc->sc_tp = tp; 917 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 918 sc->sc_ipopts = ipopts; 919 sc->sc_inc.inc_fport = inc->inc_fport; 920 sc->sc_inc.inc_lport = inc->inc_lport; 921#ifdef INET6 922 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 923 if (inc->inc_isipv6) { 924 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 925 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 926 } else 927#endif 928 { 929 sc->sc_inc.inc_faddr = inc->inc_faddr; 930 sc->sc_inc.inc_laddr = inc->inc_laddr; 931 } 932 sc->sc_irs = th->th_seq; 933 sc->sc_flags = 0; 934 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 935 sc->sc_flowlabel = 0; 936 if (tcp_syncookies) { 937 sc->sc_iss = syncookie_generate(sc, &flowtmp); 938#ifdef INET6 939 if (inc->inc_isipv6 && 940 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 941 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK; 942 } 943#endif 944 } else { 945 sc->sc_iss = arc4random(); 946#ifdef INET6 947 if (inc->inc_isipv6 && 948 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 949 sc->sc_flowlabel = 950 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 951 } 952#endif 953 } 954 955 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 956 win = sbspace(&so->so_rcv); 957 win = imax(win, 0); 958 win = imin(win, TCP_MAXWIN); 959 sc->sc_wnd = win; 960 961 if (tcp_do_rfc1323) { 962 /* 963 * A timestamp received in a SYN makes 964 * it ok to send timestamp requests and replies. 965 */ 966 if (to->to_flags & TOF_TS) { 967 sc->sc_tsrecent = to->to_tsval; 968 sc->sc_flags |= SCF_TIMESTAMP; 969 } 970 if (to->to_flags & TOF_SCALE) { 971 int wscale = 0; 972 973 /* Compute proper scaling value from buffer space */ 974 while (wscale < TCP_MAX_WINSHIFT && 975 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 976 wscale++; 977 sc->sc_request_r_scale = wscale; 978 sc->sc_requested_s_scale = to->to_requested_s_scale; 979 sc->sc_flags |= SCF_WINSCALE; 980 } 981 } 982 if (tp->t_flags & TF_NOOPT) 983 sc->sc_flags = SCF_NOOPT; 984#ifdef TCP_SIGNATURE 985 /* 986 * If listening socket requested TCP digests, and received SYN 987 * contains the option, flag this in the syncache so that 988 * syncache_respond() will do the right thing with the SYN+ACK. 989 * XXX Currently we always record the option by default and will 990 * attempt to use it in syncache_respond(). 991 */ 992 if (to->to_flags & TOF_SIGNATURE) 993 sc->sc_flags |= SCF_SIGNATURE; 994#endif 995 996 if (to->to_flags & TOF_SACK) 997 sc->sc_flags |= SCF_SACK; 998 999 /* 1000 * Do a standard 3-way handshake. 1001 */ 1002#ifdef TCPDEBUG 1003 if (syncache_respond(sc, m, so) == 0) { 1004#else 1005 if (syncache_respond(sc, m) == 0) { 1006#endif 1007 syncache_insert(sc, sch); 1008 tcpstat.tcps_sndacks++; 1009 tcpstat.tcps_sndtotal++; 1010 } else { 1011 syncache_free(sc); 1012 tcpstat.tcps_sc_dropped++; 1013 } 1014 *sop = NULL; 1015 return (1); 1016} 1017 1018#ifdef TCPDEBUG 1019static int 1020syncache_respond(sc, m, so) 1021 struct syncache *sc; 1022 struct mbuf *m; 1023 struct socket *so; 1024#else 1025static int 1026syncache_respond(sc, m) 1027 struct syncache *sc; 1028 struct mbuf *m; 1029#endif 1030{ 1031 u_int8_t *optp; 1032 int optlen, error; 1033 u_int16_t tlen, hlen, mssopt; 1034 struct ip *ip = NULL; 1035 struct tcphdr *th; 1036 struct inpcb *inp; 1037#ifdef INET6 1038 struct ip6_hdr *ip6 = NULL; 1039#endif 1040 1041 hlen = 1042#ifdef INET6 1043 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1044#endif 1045 sizeof(struct ip); 1046 1047 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1048 1049 /* Determine MSS we advertize to other end of connection */ 1050 mssopt = tcp_mssopt(&sc->sc_inc); 1051 1052 /* Compute the size of the TCP options. */ 1053 if (sc->sc_flags & SCF_NOOPT) { 1054 optlen = 0; 1055 } else { 1056 optlen = TCPOLEN_MAXSEG + 1057 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1058 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 1059#ifdef TCP_SIGNATURE 1060 if (sc->sc_flags & SCF_SIGNATURE) 1061 optlen += TCPOLEN_SIGNATURE; 1062#endif 1063 if (sc->sc_flags & SCF_SACK) 1064 optlen += TCPOLEN_SACK_PERMITTED; 1065 optlen = roundup2(optlen, 4); 1066 } 1067 tlen = hlen + sizeof(struct tcphdr) + optlen; 1068 1069 /* 1070 * XXX 1071 * assume that the entire packet will fit in a header mbuf 1072 */ 1073 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1074 1075 /* 1076 * XXX shouldn't this reuse the mbuf if possible ? 1077 * Create the IP+TCP header from scratch. 1078 */ 1079 if (m) 1080 m_freem(m); 1081 1082 m = m_gethdr(M_DONTWAIT, MT_DATA); 1083 if (m == NULL) 1084 return (ENOBUFS); 1085 m->m_data += max_linkhdr; 1086 m->m_len = tlen; 1087 m->m_pkthdr.len = tlen; 1088 m->m_pkthdr.rcvif = NULL; 1089 inp = sc->sc_tp->t_inpcb; 1090 INP_LOCK(inp); 1091#ifdef MAC 1092 mac_create_mbuf_from_inpcb(inp, m); 1093#endif 1094 1095#ifdef INET6 1096 if (sc->sc_inc.inc_isipv6) { 1097 ip6 = mtod(m, struct ip6_hdr *); 1098 ip6->ip6_vfc = IPV6_VERSION; 1099 ip6->ip6_nxt = IPPROTO_TCP; 1100 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1101 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1102 ip6->ip6_plen = htons(tlen - hlen); 1103 /* ip6_hlim is set after checksum */ 1104 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1105 ip6->ip6_flow |= sc->sc_flowlabel; 1106 1107 th = (struct tcphdr *)(ip6 + 1); 1108 } else 1109#endif 1110 { 1111 ip = mtod(m, struct ip *); 1112 ip->ip_v = IPVERSION; 1113 ip->ip_hl = sizeof(struct ip) >> 2; 1114 ip->ip_len = tlen; 1115 ip->ip_id = 0; 1116 ip->ip_off = 0; 1117 ip->ip_sum = 0; 1118 ip->ip_p = IPPROTO_TCP; 1119 ip->ip_src = sc->sc_inc.inc_laddr; 1120 ip->ip_dst = sc->sc_inc.inc_faddr; 1121 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1122 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 1123 1124 /* 1125 * See if we should do MTU discovery. Route lookups are 1126 * expensive, so we will only unset the DF bit if: 1127 * 1128 * 1) path_mtu_discovery is disabled 1129 * 2) the SCF_UNREACH flag has been set 1130 */ 1131 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1132 ip->ip_off |= IP_DF; 1133 1134 th = (struct tcphdr *)(ip + 1); 1135 } 1136 th->th_sport = sc->sc_inc.inc_lport; 1137 th->th_dport = sc->sc_inc.inc_fport; 1138 1139 th->th_seq = htonl(sc->sc_iss); 1140 th->th_ack = htonl(sc->sc_irs + 1); 1141 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1142 th->th_x2 = 0; 1143 th->th_flags = TH_SYN|TH_ACK; 1144 th->th_win = htons(sc->sc_wnd); 1145 th->th_urp = 0; 1146 1147 /* Tack on the TCP options. */ 1148 if (optlen != 0) { 1149 optp = (u_int8_t *)(th + 1); 1150 *optp++ = TCPOPT_MAXSEG; 1151 *optp++ = TCPOLEN_MAXSEG; 1152 *optp++ = (mssopt >> 8) & 0xff; 1153 *optp++ = mssopt & 0xff; 1154 1155 if (sc->sc_flags & SCF_WINSCALE) { 1156 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1157 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1158 sc->sc_request_r_scale); 1159 optp += 4; 1160 } 1161 1162 if (sc->sc_flags & SCF_TIMESTAMP) { 1163 u_int32_t *lp = (u_int32_t *)(optp); 1164 1165 /* Form timestamp option per appendix A of RFC 1323. */ 1166 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1167 *lp++ = htonl(ticks); 1168 *lp = htonl(sc->sc_tsrecent); 1169 optp += TCPOLEN_TSTAMP_APPA; 1170 } 1171 1172#ifdef TCP_SIGNATURE 1173 /* 1174 * Handle TCP-MD5 passive opener response. 1175 */ 1176 if (sc->sc_flags & SCF_SIGNATURE) { 1177 u_int8_t *bp = optp; 1178 int i; 1179 1180 *bp++ = TCPOPT_SIGNATURE; 1181 *bp++ = TCPOLEN_SIGNATURE; 1182 for (i = 0; i < TCP_SIGLEN; i++) 1183 *bp++ = 0; 1184 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1185 optp + 2, IPSEC_DIR_OUTBOUND); 1186 optp += TCPOLEN_SIGNATURE; 1187 } 1188#endif /* TCP_SIGNATURE */ 1189 1190 if (sc->sc_flags & SCF_SACK) { 1191 *optp++ = TCPOPT_SACK_PERMITTED; 1192 *optp++ = TCPOLEN_SACK_PERMITTED; 1193 } 1194 1195 { 1196 /* Pad TCP options to a 4 byte boundary */ 1197 int padlen = optlen - (optp - (u_int8_t *)(th + 1)); 1198 while (padlen-- > 0) 1199 *optp++ = TCPOPT_EOL; 1200 } 1201 } 1202 1203#ifdef INET6 1204 if (sc->sc_inc.inc_isipv6) { 1205 th->th_sum = 0; 1206 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1207 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1208 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1209 } else 1210#endif 1211 { 1212 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1213 htons(tlen - hlen + IPPROTO_TCP)); 1214 m->m_pkthdr.csum_flags = CSUM_TCP; 1215 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1216#ifdef TCPDEBUG 1217 /* 1218 * Trace. 1219 */ 1220 if (so != NULL && so->so_options & SO_DEBUG) { 1221 struct tcpcb *tp = sototcpcb(so); 1222 tcp_trace(TA_OUTPUT, tp->t_state, tp, 1223 mtod(m, void *), th, 0); 1224 } 1225#endif 1226 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp); 1227 } 1228 INP_UNLOCK(inp); 1229 return (error); 1230} 1231 1232/* 1233 * cookie layers: 1234 * 1235 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1236 * | peer iss | 1237 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1238 * | 0 |(A)| | 1239 * (A): peer mss index 1240 */ 1241 1242/* 1243 * The values below are chosen to minimize the size of the tcp_secret 1244 * table, as well as providing roughly a 16 second lifetime for the cookie. 1245 */ 1246 1247#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1248#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1249 1250#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1251#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1252#define SYNCOOKIE_TIMEOUT \ 1253 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1254#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1255 1256static struct { 1257 u_int32_t ts_secbits[4]; 1258 u_int ts_expire; 1259} tcp_secret[SYNCOOKIE_NSECRETS]; 1260 1261static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1262 1263static MD5_CTX syn_ctx; 1264 1265#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1266 1267struct md5_add { 1268 u_int32_t laddr, faddr; 1269 u_int32_t secbits[4]; 1270 u_int16_t lport, fport; 1271}; 1272 1273#ifdef CTASSERT 1274CTASSERT(sizeof(struct md5_add) == 28); 1275#endif 1276 1277/* 1278 * Consider the problem of a recreated (and retransmitted) cookie. If the 1279 * original SYN was accepted, the connection is established. The second 1280 * SYN is inflight, and if it arrives with an ISN that falls within the 1281 * receive window, the connection is killed. 1282 * 1283 * However, since cookies have other problems, this may not be worth 1284 * worrying about. 1285 */ 1286 1287static u_int32_t 1288syncookie_generate(struct syncache *sc, u_int32_t *flowid) 1289{ 1290 u_int32_t md5_buffer[4]; 1291 u_int32_t data; 1292 int idx, i; 1293 struct md5_add add; 1294 1295 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1296 1297 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1298 if (tcp_secret[idx].ts_expire < ticks) { 1299 for (i = 0; i < 4; i++) 1300 tcp_secret[idx].ts_secbits[i] = arc4random(); 1301 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1302 } 1303 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1304 if (tcp_msstab[data] <= sc->sc_peer_mss) 1305 break; 1306 data = (data << SYNCOOKIE_WNDBITS) | idx; 1307 data ^= sc->sc_irs; /* peer's iss */ 1308 MD5Init(&syn_ctx); 1309#ifdef INET6 1310 if (sc->sc_inc.inc_isipv6) { 1311 MD5Add(sc->sc_inc.inc6_laddr); 1312 MD5Add(sc->sc_inc.inc6_faddr); 1313 add.laddr = 0; 1314 add.faddr = 0; 1315 } else 1316#endif 1317 { 1318 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1319 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1320 } 1321 add.lport = sc->sc_inc.inc_lport; 1322 add.fport = sc->sc_inc.inc_fport; 1323 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1324 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1325 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1326 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1327 MD5Add(add); 1328 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1329 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1330 *flowid = md5_buffer[1]; 1331 return (data); 1332} 1333 1334static struct syncache * 1335syncookie_lookup(inc, th, so) 1336 struct in_conninfo *inc; 1337 struct tcphdr *th; 1338 struct socket *so; 1339{ 1340 u_int32_t md5_buffer[4]; 1341 struct syncache *sc; 1342 u_int32_t data; 1343 int wnd, idx; 1344 struct md5_add add; 1345 1346 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1347 1348 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1349 idx = data & SYNCOOKIE_WNDMASK; 1350 if (tcp_secret[idx].ts_expire < ticks || 1351 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1352 return (NULL); 1353 MD5Init(&syn_ctx); 1354#ifdef INET6 1355 if (inc->inc_isipv6) { 1356 MD5Add(inc->inc6_laddr); 1357 MD5Add(inc->inc6_faddr); 1358 add.laddr = 0; 1359 add.faddr = 0; 1360 } else 1361#endif 1362 { 1363 add.laddr = inc->inc_laddr.s_addr; 1364 add.faddr = inc->inc_faddr.s_addr; 1365 } 1366 add.lport = inc->inc_lport; 1367 add.fport = inc->inc_fport; 1368 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1369 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1370 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1371 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1372 MD5Add(add); 1373 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1374 data ^= md5_buffer[0]; 1375 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1376 return (NULL); 1377 data = data >> SYNCOOKIE_WNDBITS; 1378 1379 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 1380 if (sc == NULL) 1381 return (NULL); 1382 /* 1383 * Fill in the syncache values. 1384 * XXX duplicate code from syncache_add 1385 */ 1386 sc->sc_ipopts = NULL; 1387 sc->sc_inc.inc_fport = inc->inc_fport; 1388 sc->sc_inc.inc_lport = inc->inc_lport; 1389 sc->sc_tp = sototcpcb(so); 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 if (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL) 1396 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1397 } else 1398#endif 1399 { 1400 sc->sc_inc.inc_faddr = inc->inc_faddr; 1401 sc->sc_inc.inc_laddr = inc->inc_laddr; 1402 } 1403 sc->sc_irs = th->th_seq - 1; 1404 sc->sc_iss = th->th_ack - 1; 1405 wnd = sbspace(&so->so_rcv); 1406 wnd = imax(wnd, 0); 1407 wnd = imin(wnd, TCP_MAXWIN); 1408 sc->sc_wnd = wnd; 1409 sc->sc_flags = 0; 1410 sc->sc_rxtslot = 0; 1411 sc->sc_peer_mss = tcp_msstab[data]; 1412 return (sc); 1413} 1414