tcp_syncache.c revision 155439
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 155439 2006-02-07 19:59:46Z qingli $ 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 if (sch == NULL) 772 syncache_free(sc); 773 return (0); 774 } 775 776 so = syncache_socket(sc, *sop, m); 777 if (so == NULL) { 778#if 0 779resetandabort: 780 /* XXXjlemon check this - is this correct? */ 781 (void) tcp_respond(NULL, m, m, th, 782 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 783#endif 784 m_freem(m); /* XXX only needed for above */ 785 tcpstat.tcps_sc_aborted++; 786 } else 787 tcpstat.tcps_sc_completed++; 788 789 if (sch == NULL) 790 syncache_free(sc); 791 else 792 syncache_drop(sc, sch); 793 *sop = so; 794 return (1); 795} 796 797/* 798 * Given a LISTEN socket and an inbound SYN request, add 799 * this to the syn cache, and send back a segment: 800 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 801 * to the source. 802 * 803 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 804 * Doing so would require that we hold onto the data and deliver it 805 * to the application. However, if we are the target of a SYN-flood 806 * DoS attack, an attacker could send data which would eventually 807 * consume all available buffer space if it were ACKed. By not ACKing 808 * the data, we avoid this DoS scenario. 809 */ 810int 811syncache_add(inc, to, th, sop, m) 812 struct in_conninfo *inc; 813 struct tcpopt *to; 814 struct tcphdr *th; 815 struct socket **sop; 816 struct mbuf *m; 817{ 818 struct tcpcb *tp; 819 struct socket *so; 820 struct syncache *sc = NULL; 821 struct syncache_head *sch; 822 struct mbuf *ipopts = NULL; 823 u_int32_t flowtmp; 824 int i, win; 825 826 INP_INFO_WLOCK_ASSERT(&tcbinfo); 827 828 so = *sop; 829 tp = sototcpcb(so); 830 831 /* 832 * Remember the IP options, if any. 833 */ 834#ifdef INET6 835 if (!inc->inc_isipv6) 836#endif 837 ipopts = ip_srcroute(m); 838 839 /* 840 * See if we already have an entry for this connection. 841 * If we do, resend the SYN,ACK, and reset the retransmit timer. 842 * 843 * XXX 844 * should the syncache be re-initialized with the contents 845 * of the new SYN here (which may have different options?) 846 */ 847 sc = syncache_lookup(inc, &sch); 848 if (sc != NULL) { 849 tcpstat.tcps_sc_dupsyn++; 850 if (ipopts) { 851 /* 852 * If we were remembering a previous source route, 853 * forget it and use the new one we've been given. 854 */ 855 if (sc->sc_ipopts) 856 (void) m_free(sc->sc_ipopts); 857 sc->sc_ipopts = ipopts; 858 } 859 /* 860 * Update timestamp if present. 861 */ 862 if (sc->sc_flags & SCF_TIMESTAMP) 863 sc->sc_tsrecent = to->to_tsval; 864 /* 865 * PCB may have changed, pick up new values. 866 */ 867 sc->sc_tp = tp; 868 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 869#ifdef TCPDEBUG 870 if (syncache_respond(sc, m, so) == 0) { 871#else 872 if (syncache_respond(sc, m) == 0) { 873#endif 874 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 875 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 876 sc, sc_timerq); 877 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 878 tcpstat.tcps_sndacks++; 879 tcpstat.tcps_sndtotal++; 880 } 881 *sop = NULL; 882 return (1); 883 } 884 885 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 886 if (sc == NULL) { 887 /* 888 * The zone allocator couldn't provide more entries. 889 * Treat this as if the cache was full; drop the oldest 890 * entry and insert the new one. 891 */ 892 tcpstat.tcps_sc_zonefail++; 893 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 894 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 895 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]); 896 if (sc != NULL) 897 break; 898 } 899 if (sc == NULL) { 900 /* Generic memory failure. */ 901 if (ipopts) 902 (void) m_free(ipopts); 903 return (0); 904 } 905 sc->sc_tp->ts_recent = ticks; 906 syncache_drop(sc, NULL); 907 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 908 if (sc == NULL) { 909 if (ipopts) 910 (void) m_free(ipopts); 911 return (0); 912 } 913 } 914 915 /* 916 * Fill in the syncache values. 917 */ 918 bzero(sc, sizeof(*sc)); 919 sc->sc_tp = tp; 920 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 921 sc->sc_ipopts = ipopts; 922 sc->sc_inc.inc_fport = inc->inc_fport; 923 sc->sc_inc.inc_lport = inc->inc_lport; 924#ifdef INET6 925 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 926 if (inc->inc_isipv6) { 927 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 928 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 929 } else 930#endif 931 { 932 sc->sc_inc.inc_faddr = inc->inc_faddr; 933 sc->sc_inc.inc_laddr = inc->inc_laddr; 934 } 935 sc->sc_irs = th->th_seq; 936 sc->sc_flags = 0; 937 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 938 sc->sc_flowlabel = 0; 939 if (tcp_syncookies) { 940 sc->sc_iss = syncookie_generate(sc, &flowtmp); 941#ifdef INET6 942 if (inc->inc_isipv6 && 943 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 944 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK; 945 } 946#endif 947 } else { 948 sc->sc_iss = arc4random(); 949#ifdef INET6 950 if (inc->inc_isipv6 && 951 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 952 sc->sc_flowlabel = 953 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 954 } 955#endif 956 } 957 958 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 959 win = sbspace(&so->so_rcv); 960 win = imax(win, 0); 961 win = imin(win, TCP_MAXWIN); 962 sc->sc_wnd = win; 963 964 if (tcp_do_rfc1323) { 965 /* 966 * A timestamp received in a SYN makes 967 * it ok to send timestamp requests and replies. 968 */ 969 if (to->to_flags & TOF_TS) { 970 sc->sc_tsrecent = to->to_tsval; 971 sc->sc_flags |= SCF_TIMESTAMP; 972 } 973 if (to->to_flags & TOF_SCALE) { 974 int wscale = 0; 975 976 /* Compute proper scaling value from buffer space */ 977 while (wscale < TCP_MAX_WINSHIFT && 978 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 979 wscale++; 980 sc->sc_request_r_scale = wscale; 981 sc->sc_requested_s_scale = to->to_requested_s_scale; 982 sc->sc_flags |= SCF_WINSCALE; 983 } 984 } 985 if (tp->t_flags & TF_NOOPT) 986 sc->sc_flags = SCF_NOOPT; 987#ifdef TCP_SIGNATURE 988 /* 989 * If listening socket requested TCP digests, and received SYN 990 * contains the option, flag this in the syncache so that 991 * syncache_respond() will do the right thing with the SYN+ACK. 992 * XXX Currently we always record the option by default and will 993 * attempt to use it in syncache_respond(). 994 */ 995 if (to->to_flags & TOF_SIGNATURE) 996 sc->sc_flags |= SCF_SIGNATURE; 997#endif 998 999 if (to->to_flags & TOF_SACK) 1000 sc->sc_flags |= SCF_SACK; 1001 1002 /* 1003 * Do a standard 3-way handshake. 1004 */ 1005#ifdef TCPDEBUG 1006 if (syncache_respond(sc, m, so) == 0) { 1007#else 1008 if (syncache_respond(sc, m) == 0) { 1009#endif 1010 syncache_insert(sc, sch); 1011 tcpstat.tcps_sndacks++; 1012 tcpstat.tcps_sndtotal++; 1013 } else { 1014 syncache_free(sc); 1015 tcpstat.tcps_sc_dropped++; 1016 } 1017 *sop = NULL; 1018 return (1); 1019} 1020 1021#ifdef TCPDEBUG 1022static int 1023syncache_respond(sc, m, so) 1024 struct syncache *sc; 1025 struct mbuf *m; 1026 struct socket *so; 1027#else 1028static int 1029syncache_respond(sc, m) 1030 struct syncache *sc; 1031 struct mbuf *m; 1032#endif 1033{ 1034 u_int8_t *optp; 1035 int optlen, error; 1036 u_int16_t tlen, hlen, mssopt; 1037 struct ip *ip = NULL; 1038 struct tcphdr *th; 1039 struct inpcb *inp; 1040#ifdef INET6 1041 struct ip6_hdr *ip6 = NULL; 1042#endif 1043 1044 hlen = 1045#ifdef INET6 1046 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1047#endif 1048 sizeof(struct ip); 1049 1050 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1051 1052 /* Determine MSS we advertize to other end of connection */ 1053 mssopt = tcp_mssopt(&sc->sc_inc); 1054 1055 /* Compute the size of the TCP options. */ 1056 if (sc->sc_flags & SCF_NOOPT) { 1057 optlen = 0; 1058 } else { 1059 optlen = TCPOLEN_MAXSEG + 1060 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1061 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 1062#ifdef TCP_SIGNATURE 1063 if (sc->sc_flags & SCF_SIGNATURE) 1064 optlen += TCPOLEN_SIGNATURE; 1065#endif 1066 if (sc->sc_flags & SCF_SACK) 1067 optlen += TCPOLEN_SACK_PERMITTED; 1068 optlen = roundup2(optlen, 4); 1069 } 1070 tlen = hlen + sizeof(struct tcphdr) + optlen; 1071 1072 /* 1073 * XXX 1074 * assume that the entire packet will fit in a header mbuf 1075 */ 1076 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1077 1078 /* 1079 * XXX shouldn't this reuse the mbuf if possible ? 1080 * Create the IP+TCP header from scratch. 1081 */ 1082 if (m) 1083 m_freem(m); 1084 1085 m = m_gethdr(M_DONTWAIT, MT_DATA); 1086 if (m == NULL) 1087 return (ENOBUFS); 1088 m->m_data += max_linkhdr; 1089 m->m_len = tlen; 1090 m->m_pkthdr.len = tlen; 1091 m->m_pkthdr.rcvif = NULL; 1092 inp = sc->sc_tp->t_inpcb; 1093 INP_LOCK(inp); 1094#ifdef MAC 1095 mac_create_mbuf_from_inpcb(inp, m); 1096#endif 1097 1098#ifdef INET6 1099 if (sc->sc_inc.inc_isipv6) { 1100 ip6 = mtod(m, struct ip6_hdr *); 1101 ip6->ip6_vfc = IPV6_VERSION; 1102 ip6->ip6_nxt = IPPROTO_TCP; 1103 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1104 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1105 ip6->ip6_plen = htons(tlen - hlen); 1106 /* ip6_hlim is set after checksum */ 1107 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1108 ip6->ip6_flow |= sc->sc_flowlabel; 1109 1110 th = (struct tcphdr *)(ip6 + 1); 1111 } else 1112#endif 1113 { 1114 ip = mtod(m, struct ip *); 1115 ip->ip_v = IPVERSION; 1116 ip->ip_hl = sizeof(struct ip) >> 2; 1117 ip->ip_len = tlen; 1118 ip->ip_id = 0; 1119 ip->ip_off = 0; 1120 ip->ip_sum = 0; 1121 ip->ip_p = IPPROTO_TCP; 1122 ip->ip_src = sc->sc_inc.inc_laddr; 1123 ip->ip_dst = sc->sc_inc.inc_faddr; 1124 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1125 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 1126 1127 /* 1128 * See if we should do MTU discovery. Route lookups are 1129 * expensive, so we will only unset the DF bit if: 1130 * 1131 * 1) path_mtu_discovery is disabled 1132 * 2) the SCF_UNREACH flag has been set 1133 */ 1134 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1135 ip->ip_off |= IP_DF; 1136 1137 th = (struct tcphdr *)(ip + 1); 1138 } 1139 th->th_sport = sc->sc_inc.inc_lport; 1140 th->th_dport = sc->sc_inc.inc_fport; 1141 1142 th->th_seq = htonl(sc->sc_iss); 1143 th->th_ack = htonl(sc->sc_irs + 1); 1144 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1145 th->th_x2 = 0; 1146 th->th_flags = TH_SYN|TH_ACK; 1147 th->th_win = htons(sc->sc_wnd); 1148 th->th_urp = 0; 1149 1150 /* Tack on the TCP options. */ 1151 if (optlen != 0) { 1152 optp = (u_int8_t *)(th + 1); 1153 *optp++ = TCPOPT_MAXSEG; 1154 *optp++ = TCPOLEN_MAXSEG; 1155 *optp++ = (mssopt >> 8) & 0xff; 1156 *optp++ = mssopt & 0xff; 1157 1158 if (sc->sc_flags & SCF_WINSCALE) { 1159 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1160 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1161 sc->sc_request_r_scale); 1162 optp += 4; 1163 } 1164 1165 if (sc->sc_flags & SCF_TIMESTAMP) { 1166 u_int32_t *lp = (u_int32_t *)(optp); 1167 1168 /* Form timestamp option per appendix A of RFC 1323. */ 1169 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1170 *lp++ = htonl(ticks); 1171 *lp = htonl(sc->sc_tsrecent); 1172 optp += TCPOLEN_TSTAMP_APPA; 1173 } 1174 1175#ifdef TCP_SIGNATURE 1176 /* 1177 * Handle TCP-MD5 passive opener response. 1178 */ 1179 if (sc->sc_flags & SCF_SIGNATURE) { 1180 u_int8_t *bp = optp; 1181 int i; 1182 1183 *bp++ = TCPOPT_SIGNATURE; 1184 *bp++ = TCPOLEN_SIGNATURE; 1185 for (i = 0; i < TCP_SIGLEN; i++) 1186 *bp++ = 0; 1187 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1188 optp + 2, IPSEC_DIR_OUTBOUND); 1189 optp += TCPOLEN_SIGNATURE; 1190 } 1191#endif /* TCP_SIGNATURE */ 1192 1193 if (sc->sc_flags & SCF_SACK) { 1194 *optp++ = TCPOPT_SACK_PERMITTED; 1195 *optp++ = TCPOLEN_SACK_PERMITTED; 1196 } 1197 1198 { 1199 /* Pad TCP options to a 4 byte boundary */ 1200 int padlen = optlen - (optp - (u_int8_t *)(th + 1)); 1201 while (padlen-- > 0) 1202 *optp++ = TCPOPT_EOL; 1203 } 1204 } 1205 1206#ifdef INET6 1207 if (sc->sc_inc.inc_isipv6) { 1208 th->th_sum = 0; 1209 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1210 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1211 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1212 } else 1213#endif 1214 { 1215 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1216 htons(tlen - hlen + IPPROTO_TCP)); 1217 m->m_pkthdr.csum_flags = CSUM_TCP; 1218 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1219#ifdef TCPDEBUG 1220 /* 1221 * Trace. 1222 */ 1223 if (so != NULL && so->so_options & SO_DEBUG) { 1224 struct tcpcb *tp = sototcpcb(so); 1225 tcp_trace(TA_OUTPUT, tp->t_state, tp, 1226 mtod(m, void *), th, 0); 1227 } 1228#endif 1229 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp); 1230 } 1231 INP_UNLOCK(inp); 1232 return (error); 1233} 1234 1235/* 1236 * cookie layers: 1237 * 1238 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1239 * | peer iss | 1240 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1241 * | 0 |(A)| | 1242 * (A): peer mss index 1243 */ 1244 1245/* 1246 * The values below are chosen to minimize the size of the tcp_secret 1247 * table, as well as providing roughly a 16 second lifetime for the cookie. 1248 */ 1249 1250#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1251#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1252 1253#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1254#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1255#define SYNCOOKIE_TIMEOUT \ 1256 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1257#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1258 1259static struct { 1260 u_int32_t ts_secbits[4]; 1261 u_int ts_expire; 1262} tcp_secret[SYNCOOKIE_NSECRETS]; 1263 1264static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1265 1266static MD5_CTX syn_ctx; 1267 1268#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1269 1270struct md5_add { 1271 u_int32_t laddr, faddr; 1272 u_int32_t secbits[4]; 1273 u_int16_t lport, fport; 1274}; 1275 1276#ifdef CTASSERT 1277CTASSERT(sizeof(struct md5_add) == 28); 1278#endif 1279 1280/* 1281 * Consider the problem of a recreated (and retransmitted) cookie. If the 1282 * original SYN was accepted, the connection is established. The second 1283 * SYN is inflight, and if it arrives with an ISN that falls within the 1284 * receive window, the connection is killed. 1285 * 1286 * However, since cookies have other problems, this may not be worth 1287 * worrying about. 1288 */ 1289 1290static u_int32_t 1291syncookie_generate(struct syncache *sc, u_int32_t *flowid) 1292{ 1293 u_int32_t md5_buffer[4]; 1294 u_int32_t data; 1295 int idx, i; 1296 struct md5_add add; 1297 1298 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1299 1300 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1301 if (tcp_secret[idx].ts_expire < ticks) { 1302 for (i = 0; i < 4; i++) 1303 tcp_secret[idx].ts_secbits[i] = arc4random(); 1304 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1305 } 1306 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1307 if (tcp_msstab[data] <= sc->sc_peer_mss) 1308 break; 1309 data = (data << SYNCOOKIE_WNDBITS) | idx; 1310 data ^= sc->sc_irs; /* peer's iss */ 1311 MD5Init(&syn_ctx); 1312#ifdef INET6 1313 if (sc->sc_inc.inc_isipv6) { 1314 MD5Add(sc->sc_inc.inc6_laddr); 1315 MD5Add(sc->sc_inc.inc6_faddr); 1316 add.laddr = 0; 1317 add.faddr = 0; 1318 } else 1319#endif 1320 { 1321 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1322 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1323 } 1324 add.lport = sc->sc_inc.inc_lport; 1325 add.fport = sc->sc_inc.inc_fport; 1326 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1327 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1328 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1329 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1330 MD5Add(add); 1331 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1332 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1333 *flowid = md5_buffer[1]; 1334 return (data); 1335} 1336 1337static struct syncache * 1338syncookie_lookup(inc, th, so) 1339 struct in_conninfo *inc; 1340 struct tcphdr *th; 1341 struct socket *so; 1342{ 1343 u_int32_t md5_buffer[4]; 1344 struct syncache *sc; 1345 u_int32_t data; 1346 int wnd, idx; 1347 struct md5_add add; 1348 1349 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1350 1351 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1352 idx = data & SYNCOOKIE_WNDMASK; 1353 if (tcp_secret[idx].ts_expire < ticks || 1354 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1355 return (NULL); 1356 MD5Init(&syn_ctx); 1357#ifdef INET6 1358 if (inc->inc_isipv6) { 1359 MD5Add(inc->inc6_laddr); 1360 MD5Add(inc->inc6_faddr); 1361 add.laddr = 0; 1362 add.faddr = 0; 1363 } else 1364#endif 1365 { 1366 add.laddr = inc->inc_laddr.s_addr; 1367 add.faddr = inc->inc_faddr.s_addr; 1368 } 1369 add.lport = inc->inc_lport; 1370 add.fport = inc->inc_fport; 1371 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1372 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1373 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1374 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1375 MD5Add(add); 1376 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1377 data ^= md5_buffer[0]; 1378 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1379 return (NULL); 1380 data = data >> SYNCOOKIE_WNDBITS; 1381 1382 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 1383 if (sc == NULL) 1384 return (NULL); 1385 /* 1386 * Fill in the syncache values. 1387 * XXX duplicate code from syncache_add 1388 */ 1389 bzero(sc, sizeof(*sc)); 1390 sc->sc_ipopts = NULL; 1391 sc->sc_inc.inc_fport = inc->inc_fport; 1392 sc->sc_inc.inc_lport = inc->inc_lport; 1393 sc->sc_tp = sototcpcb(so); 1394#ifdef INET6 1395 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1396 if (inc->inc_isipv6) { 1397 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1398 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1399 if (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL) 1400 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1401 } else 1402#endif 1403 { 1404 sc->sc_inc.inc_faddr = inc->inc_faddr; 1405 sc->sc_inc.inc_laddr = inc->inc_laddr; 1406 } 1407 sc->sc_irs = th->th_seq - 1; 1408 sc->sc_iss = th->th_ack - 1; 1409 wnd = sbspace(&so->so_rcv); 1410 wnd = imax(wnd, 0); 1411 wnd = imin(wnd, TCP_MAXWIN); 1412 sc->sc_wnd = wnd; 1413 sc->sc_flags = 0; 1414 sc->sc_rxtslot = 0; 1415 sc->sc_peer_mss = tcp_msstab[data]; 1416 return (sc); 1417} 1418