tcp_syncache.c revision 125783
1/*- 2 * Copyright (c) 2001 Networks Associates Technology, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Jonathan Lemon 6 * and NAI Labs, the Security Research Division of Network Associates, Inc. 7 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 8 * DARPA CHATS research program. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. The name of the author may not be used to endorse or promote 19 * products derived from this software without specific prior written 20 * permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $FreeBSD: head/sys/netinet/tcp_syncache.c 125783 2004-02-13 18:21:45Z bms $ 35 */ 36 37#include "opt_inet.h" 38#include "opt_inet6.h" 39#include "opt_ipsec.h" 40#include "opt_mac.h" 41#include "opt_tcpdebug.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/malloc.h> 48#include <sys/mac.h> 49#include <sys/mbuf.h> 50#include <sys/md5.h> 51#include <sys/proc.h> /* for proc0 declaration */ 52#include <sys/random.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55 56#include <net/if.h> 57#include <net/route.h> 58 59#include <netinet/in.h> 60#include <netinet/in_systm.h> 61#include <netinet/ip.h> 62#include <netinet/in_var.h> 63#include <netinet/in_pcb.h> 64#include <netinet/ip_var.h> 65#ifdef INET6 66#include <netinet/ip6.h> 67#include <netinet/icmp6.h> 68#include <netinet6/nd6.h> 69#include <netinet6/ip6_var.h> 70#include <netinet6/in6_pcb.h> 71#endif 72#include <netinet/tcp.h> 73#ifdef TCPDEBUG 74#include <netinet/tcpip.h> 75#endif 76#include <netinet/tcp_fsm.h> 77#include <netinet/tcp_seq.h> 78#include <netinet/tcp_timer.h> 79#include <netinet/tcp_var.h> 80#ifdef TCPDEBUG 81#include <netinet/tcp_debug.h> 82#endif 83#ifdef INET6 84#include <netinet6/tcp6_var.h> 85#endif 86 87#ifdef IPSEC 88#include <netinet6/ipsec.h> 89#ifdef INET6 90#include <netinet6/ipsec6.h> 91#endif 92#endif /*IPSEC*/ 93 94#ifdef FAST_IPSEC 95#include <netipsec/ipsec.h> 96#ifdef INET6 97#include <netipsec/ipsec6.h> 98#endif 99#include <netipsec/key.h> 100#endif /*FAST_IPSEC*/ 101 102#include <machine/in_cksum.h> 103#include <vm/uma.h> 104 105static int tcp_syncookies = 1; 106SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 107 &tcp_syncookies, 0, 108 "Use TCP SYN cookies if the syncache overflows"); 109 110static void syncache_drop(struct syncache *, struct syncache_head *); 111static void syncache_free(struct syncache *); 112static void syncache_insert(struct syncache *, struct syncache_head *); 113struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 114#ifdef TCPDEBUG 115static int syncache_respond(struct syncache *, struct mbuf *, struct socket *); 116#else 117static int syncache_respond(struct syncache *, struct mbuf *); 118#endif 119static struct socket *syncache_socket(struct syncache *, struct socket *, 120 struct mbuf *m); 121static void syncache_timer(void *); 122static u_int32_t syncookie_generate(struct syncache *); 123static struct syncache *syncookie_lookup(struct in_conninfo *, 124 struct tcphdr *, struct socket *); 125 126/* 127 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 128 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 129 * the odds are that the user has given up attempting to connect by then. 130 */ 131#define SYNCACHE_MAXREXMTS 3 132 133/* Arbitrary values */ 134#define TCP_SYNCACHE_HASHSIZE 512 135#define TCP_SYNCACHE_BUCKETLIMIT 30 136 137struct tcp_syncache { 138 struct syncache_head *hashbase; 139 uma_zone_t zone; 140 u_int hashsize; 141 u_int hashmask; 142 u_int bucket_limit; 143 u_int cache_count; 144 u_int cache_limit; 145 u_int rexmt_limit; 146 u_int hash_secret; 147 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 148 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 149}; 150static struct tcp_syncache tcp_syncache; 151 152SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 153 154SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 155 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 156 157SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 158 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 159 160SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 161 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 162 163SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 164 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 165 166SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 167 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 168 169static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 170 171#define SYNCACHE_HASH(inc, mask) \ 172 ((tcp_syncache.hash_secret ^ \ 173 (inc)->inc_faddr.s_addr ^ \ 174 ((inc)->inc_faddr.s_addr >> 16) ^ \ 175 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 176 177#define SYNCACHE_HASH6(inc, mask) \ 178 ((tcp_syncache.hash_secret ^ \ 179 (inc)->inc6_faddr.s6_addr32[0] ^ \ 180 (inc)->inc6_faddr.s6_addr32[3] ^ \ 181 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 182 183#define ENDPTS_EQ(a, b) ( \ 184 (a)->ie_fport == (b)->ie_fport && \ 185 (a)->ie_lport == (b)->ie_lport && \ 186 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 187 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 188) 189 190#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 191 192#define SYNCACHE_TIMEOUT(sc, slot) do { \ 193 sc->sc_rxtslot = (slot); \ 194 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[(slot)]; \ 195 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[(slot)], sc, sc_timerq); \ 196 if (!callout_active(&tcp_syncache.tt_timerq[(slot)])) \ 197 callout_reset(&tcp_syncache.tt_timerq[(slot)], \ 198 TCPTV_RTOBASE * tcp_backoff[(slot)], \ 199 syncache_timer, (void *)((intptr_t)(slot))); \ 200} while (0) 201 202static void 203syncache_free(struct syncache *sc) 204{ 205 if (sc->sc_ipopts) 206 (void) m_free(sc->sc_ipopts); 207 208 uma_zfree(tcp_syncache.zone, sc); 209} 210 211void 212syncache_init(void) 213{ 214 int i; 215 216 tcp_syncache.cache_count = 0; 217 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 218 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 219 tcp_syncache.cache_limit = 220 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 221 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 222 tcp_syncache.hash_secret = arc4random(); 223 224 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 225 &tcp_syncache.hashsize); 226 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 227 &tcp_syncache.cache_limit); 228 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 229 &tcp_syncache.bucket_limit); 230 if (!powerof2(tcp_syncache.hashsize)) { 231 printf("WARNING: syncache hash size is not a power of 2.\n"); 232 tcp_syncache.hashsize = 512; /* safe default */ 233 } 234 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 235 236 /* Allocate the hash table. */ 237 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 238 tcp_syncache.hashsize * sizeof(struct syncache_head), 239 M_SYNCACHE, M_WAITOK); 240 241 /* Initialize the hash buckets. */ 242 for (i = 0; i < tcp_syncache.hashsize; i++) { 243 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 244 tcp_syncache.hashbase[i].sch_length = 0; 245 } 246 247 /* Initialize the timer queues. */ 248 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 249 TAILQ_INIT(&tcp_syncache.timerq[i]); 250 callout_init(&tcp_syncache.tt_timerq[i], 251 debug_mpsafenet ? CALLOUT_MPSAFE : 0); 252 } 253 254 /* 255 * Allocate the syncache entries. Allow the zone to allocate one 256 * more entry than cache limit, so a new entry can bump out an 257 * older one. 258 */ 259 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 260 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 261 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 262 tcp_syncache.cache_limit -= 1; 263} 264 265static void 266syncache_insert(sc, sch) 267 struct syncache *sc; 268 struct syncache_head *sch; 269{ 270 struct syncache *sc2; 271 int i; 272 273 INP_INFO_WLOCK_ASSERT(&tcbinfo); 274 275 /* 276 * Make sure that we don't overflow the per-bucket 277 * limit or the total cache size limit. 278 */ 279 if (sch->sch_length >= tcp_syncache.bucket_limit) { 280 /* 281 * The bucket is full, toss the oldest element. 282 */ 283 sc2 = TAILQ_FIRST(&sch->sch_bucket); 284 sc2->sc_tp->ts_recent = ticks; 285 syncache_drop(sc2, sch); 286 tcpstat.tcps_sc_bucketoverflow++; 287 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) { 288 /* 289 * The cache is full. Toss the oldest entry in the 290 * entire cache. This is the front entry in the 291 * first non-empty timer queue with the largest 292 * timeout value. 293 */ 294 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 295 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]); 296 if (sc2 != NULL) 297 break; 298 } 299 sc2->sc_tp->ts_recent = ticks; 300 syncache_drop(sc2, NULL); 301 tcpstat.tcps_sc_cacheoverflow++; 302 } 303 304 /* Initialize the entry's timer. */ 305 SYNCACHE_TIMEOUT(sc, 0); 306 307 /* Put it into the bucket. */ 308 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 309 sch->sch_length++; 310 tcp_syncache.cache_count++; 311 tcpstat.tcps_sc_added++; 312} 313 314static void 315syncache_drop(sc, sch) 316 struct syncache *sc; 317 struct syncache_head *sch; 318{ 319 INP_INFO_WLOCK_ASSERT(&tcbinfo); 320 321 if (sch == NULL) { 322#ifdef INET6 323 if (sc->sc_inc.inc_isipv6) { 324 sch = &tcp_syncache.hashbase[ 325 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 326 } else 327#endif 328 { 329 sch = &tcp_syncache.hashbase[ 330 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 331 } 332 } 333 334 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 335 sch->sch_length--; 336 tcp_syncache.cache_count--; 337 338 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq); 339 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot])) 340 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]); 341 342 syncache_free(sc); 343} 344 345/* 346 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 347 * If we have retransmitted an entry the maximum number of times, expire it. 348 */ 349static void 350syncache_timer(xslot) 351 void *xslot; 352{ 353 intptr_t slot = (intptr_t)xslot; 354 struct syncache *sc, *nsc; 355 struct inpcb *inp; 356 357 INP_INFO_WLOCK(&tcbinfo); 358 if (callout_pending(&tcp_syncache.tt_timerq[slot]) || 359 !callout_active(&tcp_syncache.tt_timerq[slot])) { 360 /* XXX can this happen? */ 361 INP_INFO_WUNLOCK(&tcbinfo); 362 return; 363 } 364 callout_deactivate(&tcp_syncache.tt_timerq[slot]); 365 366 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]); 367 while (nsc != NULL) { 368 if (ticks < nsc->sc_rxttime) 369 break; 370 sc = nsc; 371 inp = sc->sc_tp->t_inpcb; 372 if (slot == SYNCACHE_MAXREXMTS || 373 slot >= tcp_syncache.rexmt_limit || 374 inp == NULL || inp->inp_gencnt != sc->sc_inp_gencnt) { 375 nsc = TAILQ_NEXT(sc, sc_timerq); 376 syncache_drop(sc, NULL); 377 tcpstat.tcps_sc_stale++; 378 continue; 379 } 380 /* 381 * syncache_respond() may call back into the syncache to 382 * to modify another entry, so do not obtain the next 383 * entry on the timer chain until it has completed. 384 */ 385#ifdef TCPDEBUG 386 (void) syncache_respond(sc, NULL, NULL); 387#else 388 (void) syncache_respond(sc, NULL); 389#endif 390 nsc = TAILQ_NEXT(sc, sc_timerq); 391 tcpstat.tcps_sc_retransmitted++; 392 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq); 393 SYNCACHE_TIMEOUT(sc, slot + 1); 394 } 395 if (nsc != NULL) 396 callout_reset(&tcp_syncache.tt_timerq[slot], 397 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot)); 398 INP_INFO_WUNLOCK(&tcbinfo); 399} 400 401/* 402 * Find an entry in the syncache. 403 */ 404struct syncache * 405syncache_lookup(inc, schp) 406 struct in_conninfo *inc; 407 struct syncache_head **schp; 408{ 409 struct syncache *sc; 410 struct syncache_head *sch; 411 412 INP_INFO_WLOCK_ASSERT(&tcbinfo); 413 414#ifdef INET6 415 if (inc->inc_isipv6) { 416 sch = &tcp_syncache.hashbase[ 417 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 418 *schp = sch; 419 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 420 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 421 return (sc); 422 } 423 } else 424#endif 425 { 426 sch = &tcp_syncache.hashbase[ 427 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 428 *schp = sch; 429 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 430#ifdef INET6 431 if (sc->sc_inc.inc_isipv6) 432 continue; 433#endif 434 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 435 return (sc); 436 } 437 } 438 return (NULL); 439} 440 441/* 442 * This function is called when we get a RST for a 443 * non-existent connection, so that we can see if the 444 * connection is in the syn cache. If it is, zap it. 445 */ 446void 447syncache_chkrst(inc, th) 448 struct in_conninfo *inc; 449 struct tcphdr *th; 450{ 451 struct syncache *sc; 452 struct syncache_head *sch; 453 454 INP_INFO_WLOCK_ASSERT(&tcbinfo); 455 456 sc = syncache_lookup(inc, &sch); 457 if (sc == NULL) 458 return; 459 /* 460 * If the RST bit is set, check the sequence number to see 461 * if this is a valid reset segment. 462 * RFC 793 page 37: 463 * In all states except SYN-SENT, all reset (RST) segments 464 * are validated by checking their SEQ-fields. A reset is 465 * valid if its sequence number is in the window. 466 * 467 * The sequence number in the reset segment is normally an 468 * echo of our outgoing acknowlegement numbers, but some hosts 469 * send a reset with the sequence number at the rightmost edge 470 * of our receive window, and we have to handle this case. 471 */ 472 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 473 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 474 syncache_drop(sc, sch); 475 tcpstat.tcps_sc_reset++; 476 } 477} 478 479void 480syncache_badack(inc) 481 struct in_conninfo *inc; 482{ 483 struct syncache *sc; 484 struct syncache_head *sch; 485 486 INP_INFO_WLOCK_ASSERT(&tcbinfo); 487 488 sc = syncache_lookup(inc, &sch); 489 if (sc != NULL) { 490 syncache_drop(sc, sch); 491 tcpstat.tcps_sc_badack++; 492 } 493} 494 495void 496syncache_unreach(inc, th) 497 struct in_conninfo *inc; 498 struct tcphdr *th; 499{ 500 struct syncache *sc; 501 struct syncache_head *sch; 502 503 INP_INFO_WLOCK_ASSERT(&tcbinfo); 504 505 /* we are called at splnet() here */ 506 sc = syncache_lookup(inc, &sch); 507 if (sc == NULL) 508 return; 509 510 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 511 if (ntohl(th->th_seq) != sc->sc_iss) 512 return; 513 514 /* 515 * If we've rertransmitted 3 times and this is our second error, 516 * we remove the entry. Otherwise, we allow it to continue on. 517 * This prevents us from incorrectly nuking an entry during a 518 * spurious network outage. 519 * 520 * See tcp_notify(). 521 */ 522 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 523 sc->sc_flags |= SCF_UNREACH; 524 return; 525 } 526 syncache_drop(sc, sch); 527 tcpstat.tcps_sc_unreach++; 528} 529 530/* 531 * Build a new TCP socket structure from a syncache entry. 532 */ 533static struct socket * 534syncache_socket(sc, lso, m) 535 struct syncache *sc; 536 struct socket *lso; 537 struct mbuf *m; 538{ 539 struct inpcb *inp = NULL; 540 struct socket *so; 541 struct tcpcb *tp; 542 543 GIANT_REQUIRED; /* XXX until socket locking */ 544 INP_INFO_WLOCK_ASSERT(&tcbinfo); 545 546 /* 547 * Ok, create the full blown connection, and set things up 548 * as they would have been set up if we had created the 549 * connection when the SYN arrived. If we can't create 550 * the connection, abort it. 551 */ 552 so = sonewconn(lso, SS_ISCONNECTED); 553 if (so == NULL) { 554 /* 555 * Drop the connection; we will send a RST if the peer 556 * retransmits the ACK, 557 */ 558 tcpstat.tcps_listendrop++; 559 goto abort2; 560 } 561#ifdef MAC 562 mac_set_socket_peer_from_mbuf(m, so); 563#endif 564 565 inp = sotoinpcb(so); 566 INP_LOCK(inp); 567 568 /* 569 * Insert new socket into hash list. 570 */ 571 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 572#ifdef INET6 573 if (sc->sc_inc.inc_isipv6) { 574 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 575 } else { 576 inp->inp_vflag &= ~INP_IPV6; 577 inp->inp_vflag |= INP_IPV4; 578#endif 579 inp->inp_laddr = sc->sc_inc.inc_laddr; 580#ifdef INET6 581 } 582#endif 583 inp->inp_lport = sc->sc_inc.inc_lport; 584 if (in_pcbinshash(inp) != 0) { 585 /* 586 * Undo the assignments above if we failed to 587 * put the PCB on the hash lists. 588 */ 589#ifdef INET6 590 if (sc->sc_inc.inc_isipv6) 591 inp->in6p_laddr = in6addr_any; 592 else 593#endif 594 inp->inp_laddr.s_addr = INADDR_ANY; 595 inp->inp_lport = 0; 596 goto abort; 597 } 598#ifdef IPSEC 599 /* copy old policy into new socket's */ 600 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 601 printf("syncache_expand: could not copy policy\n"); 602#endif 603#ifdef FAST_IPSEC 604 /* copy old policy into new socket's */ 605 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 606 printf("syncache_expand: could not copy policy\n"); 607#endif 608#ifdef INET6 609 if (sc->sc_inc.inc_isipv6) { 610 struct inpcb *oinp = sotoinpcb(lso); 611 struct in6_addr laddr6; 612 struct sockaddr_in6 sin6; 613 /* 614 * Inherit socket options from the listening socket. 615 * Note that in6p_inputopts are not (and should not be) 616 * copied, since it stores previously received options and is 617 * used to detect if each new option is different than the 618 * previous one and hence should be passed to a user. 619 * If we copied in6p_inputopts, a user would not be able to 620 * receive options just after calling the accept system call. 621 */ 622 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 623 if (oinp->in6p_outputopts) 624 inp->in6p_outputopts = 625 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 626 627 sin6.sin6_family = AF_INET6; 628 sin6.sin6_len = sizeof(sin6); 629 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 630 sin6.sin6_port = sc->sc_inc.inc_fport; 631 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 632 laddr6 = inp->in6p_laddr; 633 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 634 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 635 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 636 inp->in6p_laddr = laddr6; 637 goto abort; 638 } 639 } else 640#endif 641 { 642 struct in_addr laddr; 643 struct sockaddr_in sin; 644 645 inp->inp_options = ip_srcroute(); 646 if (inp->inp_options == NULL) { 647 inp->inp_options = sc->sc_ipopts; 648 sc->sc_ipopts = NULL; 649 } 650 651 sin.sin_family = AF_INET; 652 sin.sin_len = sizeof(sin); 653 sin.sin_addr = sc->sc_inc.inc_faddr; 654 sin.sin_port = sc->sc_inc.inc_fport; 655 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 656 laddr = inp->inp_laddr; 657 if (inp->inp_laddr.s_addr == INADDR_ANY) 658 inp->inp_laddr = sc->sc_inc.inc_laddr; 659 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 660 inp->inp_laddr = laddr; 661 goto abort; 662 } 663 } 664 665 tp = intotcpcb(inp); 666 tp->t_state = TCPS_SYN_RECEIVED; 667 tp->iss = sc->sc_iss; 668 tp->irs = sc->sc_irs; 669 tcp_rcvseqinit(tp); 670 tcp_sendseqinit(tp); 671 tp->snd_wl1 = sc->sc_irs; 672 tp->rcv_up = sc->sc_irs + 1; 673 tp->rcv_wnd = sc->sc_wnd; 674 tp->rcv_adv += tp->rcv_wnd; 675 676 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 677 if (sc->sc_flags & SCF_NOOPT) 678 tp->t_flags |= TF_NOOPT; 679 if (sc->sc_flags & SCF_WINSCALE) { 680 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 681 tp->requested_s_scale = sc->sc_requested_s_scale; 682 tp->request_r_scale = sc->sc_request_r_scale; 683 } 684 if (sc->sc_flags & SCF_TIMESTAMP) { 685 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 686 tp->ts_recent = sc->sc_tsrecent; 687 tp->ts_recent_age = ticks; 688 } 689 if (sc->sc_flags & SCF_CC) { 690 /* 691 * Initialization of the tcpcb for transaction; 692 * set SND.WND = SEG.WND, 693 * initialize CCsend and CCrecv. 694 */ 695 tp->t_flags |= TF_REQ_CC|TF_RCVD_CC; 696 tp->cc_send = sc->sc_cc_send; 697 tp->cc_recv = sc->sc_cc_recv; 698 } 699#ifdef TCP_SIGNATURE 700 if (sc->sc_flags & SCF_SIGNATURE) 701 tp->t_flags |= TF_SIGNATURE; 702#endif 703 704 /* 705 * Set up MSS and get cached values from tcp_hostcache. 706 * This might overwrite some of the defaults we just set. 707 */ 708 tcp_mss(tp, sc->sc_peer_mss); 709 710 /* 711 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 712 */ 713 if (sc->sc_rxtslot != 0) 714 tp->snd_cwnd = tp->t_maxseg; 715 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 716 717 INP_UNLOCK(inp); 718 719 tcpstat.tcps_accepts++; 720 return (so); 721 722abort: 723 INP_UNLOCK(inp); 724abort2: 725 if (so != NULL) 726 (void) soabort(so); 727 return (NULL); 728} 729 730/* 731 * This function gets called when we receive an ACK for a 732 * socket in the LISTEN state. We look up the connection 733 * in the syncache, and if its there, we pull it out of 734 * the cache and turn it into a full-blown connection in 735 * the SYN-RECEIVED state. 736 */ 737int 738syncache_expand(inc, th, sop, m) 739 struct in_conninfo *inc; 740 struct tcphdr *th; 741 struct socket **sop; 742 struct mbuf *m; 743{ 744 struct syncache *sc; 745 struct syncache_head *sch; 746 struct socket *so; 747 748 INP_INFO_WLOCK_ASSERT(&tcbinfo); 749 750 sc = syncache_lookup(inc, &sch); 751 if (sc == NULL) { 752 /* 753 * There is no syncache entry, so see if this ACK is 754 * a returning syncookie. To do this, first: 755 * A. See if this socket has had a syncache entry dropped in 756 * the past. We don't want to accept a bogus syncookie 757 * if we've never received a SYN. 758 * B. check that the syncookie is valid. If it is, then 759 * cobble up a fake syncache entry, and return. 760 */ 761 if (!tcp_syncookies) 762 return (0); 763 sc = syncookie_lookup(inc, th, *sop); 764 if (sc == NULL) 765 return (0); 766 sch = NULL; 767 tcpstat.tcps_sc_recvcookie++; 768 } 769 770 /* 771 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 772 */ 773 if (th->th_ack != sc->sc_iss + 1) 774 return (0); 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 struct rmxp_tao tao; 824 int i, win; 825 826 INP_INFO_WLOCK_ASSERT(&tcbinfo); 827 828 so = *sop; 829 tp = sototcpcb(so); 830 bzero(&tao, sizeof(tao)); 831 832 /* 833 * Remember the IP options, if any. 834 */ 835#ifdef INET6 836 if (!inc->inc_isipv6) 837#endif 838 ipopts = ip_srcroute(); 839 840 /* 841 * See if we already have an entry for this connection. 842 * If we do, resend the SYN,ACK, and reset the retransmit timer. 843 * 844 * XXX 845 * should the syncache be re-initialized with the contents 846 * of the new SYN here (which may have different options?) 847 */ 848 sc = syncache_lookup(inc, &sch); 849 if (sc != NULL) { 850 tcpstat.tcps_sc_dupsyn++; 851 if (ipopts) { 852 /* 853 * If we were remembering a previous source route, 854 * forget it and use the new one we've been given. 855 */ 856 if (sc->sc_ipopts) 857 (void) m_free(sc->sc_ipopts); 858 sc->sc_ipopts = ipopts; 859 } 860 /* 861 * Update timestamp if present. 862 */ 863 if (sc->sc_flags & SCF_TIMESTAMP) 864 sc->sc_tsrecent = to->to_tsval; 865 /* 866 * PCB may have changed, pick up new values. 867 */ 868 sc->sc_tp = tp; 869 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 870#ifdef TCPDEBUG 871 if (syncache_respond(sc, m, so) == 0) { 872#else 873 if (syncache_respond(sc, m) == 0) { 874#endif 875 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 876 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 877 sc, sc_timerq); 878 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 879 tcpstat.tcps_sndacks++; 880 tcpstat.tcps_sndtotal++; 881 } 882 *sop = NULL; 883 return (1); 884 } 885 886 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 887 if (sc == NULL) { 888 /* 889 * The zone allocator couldn't provide more entries. 890 * Treat this as if the cache was full; drop the oldest 891 * entry and insert the new one. 892 */ 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 sc->sc_tp->ts_recent = ticks; 900 syncache_drop(sc, NULL); 901 tcpstat.tcps_sc_zonefail++; 902 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 903 if (sc == NULL) { 904 if (ipopts) 905 (void) m_free(ipopts); 906 return (0); 907 } 908 } 909 910 /* 911 * Fill in the syncache values. 912 */ 913 bzero(sc, sizeof(*sc)); 914 sc->sc_tp = tp; 915 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 916 sc->sc_ipopts = ipopts; 917 sc->sc_inc.inc_fport = inc->inc_fport; 918 sc->sc_inc.inc_lport = inc->inc_lport; 919#ifdef INET6 920 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 921 if (inc->inc_isipv6) { 922 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 923 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 924 } else 925#endif 926 { 927 sc->sc_inc.inc_faddr = inc->inc_faddr; 928 sc->sc_inc.inc_laddr = inc->inc_laddr; 929 } 930 sc->sc_irs = th->th_seq; 931 sc->sc_flags = 0; 932 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 933 if (tcp_syncookies) 934 sc->sc_iss = syncookie_generate(sc); 935 else 936 sc->sc_iss = arc4random(); 937 938 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 939 win = sbspace(&so->so_rcv); 940 win = imax(win, 0); 941 win = imin(win, TCP_MAXWIN); 942 sc->sc_wnd = win; 943 944 if (tcp_do_rfc1323) { 945 /* 946 * A timestamp received in a SYN makes 947 * it ok to send timestamp requests and replies. 948 */ 949 if (to->to_flags & TOF_TS) { 950 sc->sc_tsrecent = to->to_tsval; 951 sc->sc_flags |= SCF_TIMESTAMP; 952 } 953 if (to->to_flags & TOF_SCALE) { 954 int wscale = 0; 955 956 /* Compute proper scaling value from buffer space */ 957 while (wscale < TCP_MAX_WINSHIFT && 958 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 959 wscale++; 960 sc->sc_request_r_scale = wscale; 961 sc->sc_requested_s_scale = to->to_requested_s_scale; 962 sc->sc_flags |= SCF_WINSCALE; 963 } 964 } 965 if (tcp_do_rfc1644) { 966 /* 967 * A CC or CC.new option received in a SYN makes 968 * it ok to send CC in subsequent segments. 969 */ 970 if (to->to_flags & (TOF_CC|TOF_CCNEW)) { 971 sc->sc_cc_recv = to->to_cc; 972 sc->sc_cc_send = CC_INC(tcp_ccgen); 973 sc->sc_flags |= SCF_CC; 974 } 975 } 976 if (tp->t_flags & TF_NOOPT) 977 sc->sc_flags = SCF_NOOPT; 978#ifdef TCP_SIGNATURE 979 /* 980 * If listening socket requested TCP digests, and received SYN 981 * contains the option, flag this in the syncache so that 982 * syncache_respond() will do the right thing with the SYN+ACK. 983 * XXX Currently we always record the option by default and will 984 * attempt to use it in syncache_respond(). 985 */ 986 if (to->to_flags & TOF_SIGNATURE) 987 sc->sc_flags = SCF_SIGNATURE; 988#endif 989 990 /* 991 * XXX 992 * We have the option here of not doing TAO (even if the segment 993 * qualifies) and instead fall back to a normal 3WHS via the syncache. 994 * This allows us to apply synflood protection to TAO-qualifying SYNs 995 * also. However, there should be a hueristic to determine when to 996 * do this, and is not present at the moment. 997 */ 998 999 /* 1000 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1001 * - compare SEG.CC against cached CC from the same host, if any. 1002 * - if SEG.CC > chached value, SYN must be new and is accepted 1003 * immediately: save new CC in the cache, mark the socket 1004 * connected, enter ESTABLISHED state, turn on flag to 1005 * send a SYN in the next segment. 1006 * A virtual advertised window is set in rcv_adv to 1007 * initialize SWS prevention. Then enter normal segment 1008 * processing: drop SYN, process data and FIN. 1009 * - otherwise do a normal 3-way handshake. 1010 */ 1011 if (tcp_do_rfc1644) 1012 tcp_hc_gettao(&sc->sc_inc, &tao); 1013 1014 if ((to->to_flags & TOF_CC) != 0) { 1015 if (((tp->t_flags & TF_NOPUSH) != 0) && 1016 sc->sc_flags & SCF_CC && tao.tao_cc != 0 && 1017 CC_GT(to->to_cc, tao.tao_cc)) { 1018 sc->sc_rxtslot = 0; 1019 so = syncache_socket(sc, *sop, m); 1020 if (so != NULL) { 1021 tao.tao_cc = to->to_cc; 1022 tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC, 1023 tao.tao_cc, 0); 1024 *sop = so; 1025 } 1026 syncache_free(sc); 1027 return (so != NULL); 1028 } 1029 } else { 1030 /* 1031 * No CC option, but maybe CC.NEW: invalidate cached value. 1032 */ 1033 if (tcp_do_rfc1644) { 1034 tao.tao_cc = 0; 1035 tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC, 1036 tao.tao_cc, 0); 1037 } 1038 } 1039 1040 /* 1041 * TAO test failed or there was no CC option, 1042 * do a standard 3-way handshake. 1043 */ 1044#ifdef TCPDEBUG 1045 if (syncache_respond(sc, m, so) == 0) { 1046#else 1047 if (syncache_respond(sc, m) == 0) { 1048#endif 1049 syncache_insert(sc, sch); 1050 tcpstat.tcps_sndacks++; 1051 tcpstat.tcps_sndtotal++; 1052 } else { 1053 syncache_free(sc); 1054 tcpstat.tcps_sc_dropped++; 1055 } 1056 *sop = NULL; 1057 return (1); 1058} 1059 1060#ifdef TCPDEBUG 1061static int 1062syncache_respond(sc, m, so) 1063 struct syncache *sc; 1064 struct mbuf *m; 1065 struct socket *so; 1066#else 1067static int 1068syncache_respond(sc, m) 1069 struct syncache *sc; 1070 struct mbuf *m; 1071#endif 1072{ 1073 u_int8_t *optp; 1074 int optlen, error; 1075 u_int16_t tlen, hlen, mssopt; 1076 struct ip *ip = NULL; 1077 struct tcphdr *th; 1078 struct inpcb *inp; 1079#ifdef INET6 1080 struct ip6_hdr *ip6 = NULL; 1081#endif 1082 1083 hlen = 1084#ifdef INET6 1085 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1086#endif 1087 sizeof(struct ip); 1088 1089 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1090 1091 /* Determine MSS we advertize to other end of connection */ 1092 mssopt = tcp_mssopt(&sc->sc_inc); 1093 1094 /* Compute the size of the TCP options. */ 1095 if (sc->sc_flags & SCF_NOOPT) { 1096 optlen = 0; 1097 } else { 1098 optlen = TCPOLEN_MAXSEG + 1099 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1100 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1101 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0); 1102#ifdef TCP_SIGNATURE 1103 optlen += (sc->sc_flags & SCF_SIGNATURE) ? 1104 (TCPOLEN_SIGNATURE + 2) : 0; 1105#endif 1106 } 1107 tlen = hlen + sizeof(struct tcphdr) + optlen; 1108 1109 /* 1110 * XXX 1111 * assume that the entire packet will fit in a header mbuf 1112 */ 1113 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1114 1115 /* 1116 * XXX shouldn't this reuse the mbuf if possible ? 1117 * Create the IP+TCP header from scratch. 1118 */ 1119 if (m) 1120 m_freem(m); 1121 1122 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1123 if (m == NULL) 1124 return (ENOBUFS); 1125 m->m_data += max_linkhdr; 1126 m->m_len = tlen; 1127 m->m_pkthdr.len = tlen; 1128 m->m_pkthdr.rcvif = NULL; 1129 inp = sc->sc_tp->t_inpcb; 1130 INP_LOCK(inp); 1131#ifdef MAC 1132 mac_create_mbuf_from_socket(inp->inp_socket, m); 1133#endif 1134 1135#ifdef INET6 1136 if (sc->sc_inc.inc_isipv6) { 1137 ip6 = mtod(m, struct ip6_hdr *); 1138 ip6->ip6_vfc = IPV6_VERSION; 1139 ip6->ip6_nxt = IPPROTO_TCP; 1140 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1141 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1142 ip6->ip6_plen = htons(tlen - hlen); 1143 /* ip6_hlim is set after checksum */ 1144 /* ip6_flow = ??? */ 1145 1146 th = (struct tcphdr *)(ip6 + 1); 1147 } else 1148#endif 1149 { 1150 ip = mtod(m, struct ip *); 1151 ip->ip_v = IPVERSION; 1152 ip->ip_hl = sizeof(struct ip) >> 2; 1153 ip->ip_len = tlen; 1154 ip->ip_id = 0; 1155 ip->ip_off = 0; 1156 ip->ip_sum = 0; 1157 ip->ip_p = IPPROTO_TCP; 1158 ip->ip_src = sc->sc_inc.inc_laddr; 1159 ip->ip_dst = sc->sc_inc.inc_faddr; 1160 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1161 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 1162 1163 /* 1164 * See if we should do MTU discovery. Route lookups are 1165 * expensive, so we will only unset the DF bit if: 1166 * 1167 * 1) path_mtu_discovery is disabled 1168 * 2) the SCF_UNREACH flag has been set 1169 */ 1170 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1171 ip->ip_off |= IP_DF; 1172 1173 th = (struct tcphdr *)(ip + 1); 1174 } 1175 th->th_sport = sc->sc_inc.inc_lport; 1176 th->th_dport = sc->sc_inc.inc_fport; 1177 1178 th->th_seq = htonl(sc->sc_iss); 1179 th->th_ack = htonl(sc->sc_irs + 1); 1180 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1181 th->th_x2 = 0; 1182 th->th_flags = TH_SYN|TH_ACK; 1183 th->th_win = htons(sc->sc_wnd); 1184 th->th_urp = 0; 1185 1186 /* Tack on the TCP options. */ 1187 if (optlen != 0) { 1188 optp = (u_int8_t *)(th + 1); 1189 *optp++ = TCPOPT_MAXSEG; 1190 *optp++ = TCPOLEN_MAXSEG; 1191 *optp++ = (mssopt >> 8) & 0xff; 1192 *optp++ = mssopt & 0xff; 1193 1194 if (sc->sc_flags & SCF_WINSCALE) { 1195 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1196 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1197 sc->sc_request_r_scale); 1198 optp += 4; 1199 } 1200 1201 if (sc->sc_flags & SCF_TIMESTAMP) { 1202 u_int32_t *lp = (u_int32_t *)(optp); 1203 1204 /* Form timestamp option per appendix A of RFC 1323. */ 1205 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1206 *lp++ = htonl(ticks); 1207 *lp = htonl(sc->sc_tsrecent); 1208 optp += TCPOLEN_TSTAMP_APPA; 1209 } 1210 1211 /* 1212 * Send CC and CC.echo if we received CC from our peer. 1213 */ 1214 if (sc->sc_flags & SCF_CC) { 1215 u_int32_t *lp = (u_int32_t *)(optp); 1216 1217 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1218 *lp++ = htonl(sc->sc_cc_send); 1219 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1220 *lp = htonl(sc->sc_cc_recv); 1221 optp += TCPOLEN_CC_APPA * 2; 1222 } 1223 1224#ifdef TCP_SIGNATURE 1225 /* 1226 * Handle TCP-MD5 passive opener response. 1227 */ 1228 if (sc->sc_flags & SCF_SIGNATURE) { 1229 u_int8_t *bp = optp; 1230 int i; 1231 1232 *bp++ = TCPOPT_SIGNATURE; 1233 *bp++ = TCPOLEN_SIGNATURE; 1234 for (i = 0; i < TCP_SIGLEN; i++) 1235 *bp++ = 0; 1236 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1237 optp + 2, IPSEC_DIR_OUTBOUND); 1238 *bp++ = TCPOPT_NOP; 1239 *bp++ = TCPOPT_EOL; 1240 optp += TCPOLEN_SIGNATURE + 2; 1241 } 1242#endif /* TCP_SIGNATURE */ 1243 } 1244 1245#ifdef INET6 1246 if (sc->sc_inc.inc_isipv6) { 1247 th->th_sum = 0; 1248 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1249 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1250 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1251 } else 1252#endif 1253 { 1254 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1255 htons(tlen - hlen + IPPROTO_TCP)); 1256 m->m_pkthdr.csum_flags = CSUM_TCP; 1257 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1258#ifdef TCPDEBUG 1259 /* 1260 * Trace. 1261 */ 1262 if (so != NULL && so->so_options & SO_DEBUG) { 1263 struct tcpcb *tp = sototcpcb(so); 1264 tcp_trace(TA_OUTPUT, tp->t_state, tp, 1265 mtod(m, void *), th, 0); 1266 } 1267#endif 1268 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp); 1269 } 1270 INP_UNLOCK(inp); 1271 return (error); 1272} 1273 1274/* 1275 * cookie layers: 1276 * 1277 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1278 * | peer iss | 1279 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1280 * | 0 |(A)| | 1281 * (A): peer mss index 1282 */ 1283 1284/* 1285 * The values below are chosen to minimize the size of the tcp_secret 1286 * table, as well as providing roughly a 16 second lifetime for the cookie. 1287 */ 1288 1289#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1290#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1291 1292#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1293#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1294#define SYNCOOKIE_TIMEOUT \ 1295 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1296#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1297 1298static struct { 1299 u_int32_t ts_secbits[4]; 1300 u_int ts_expire; 1301} tcp_secret[SYNCOOKIE_NSECRETS]; 1302 1303static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1304 1305static MD5_CTX syn_ctx; 1306 1307#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1308 1309struct md5_add { 1310 u_int32_t laddr, faddr; 1311 u_int32_t secbits[4]; 1312 u_int16_t lport, fport; 1313}; 1314 1315#ifdef CTASSERT 1316CTASSERT(sizeof(struct md5_add) == 28); 1317#endif 1318 1319/* 1320 * Consider the problem of a recreated (and retransmitted) cookie. If the 1321 * original SYN was accepted, the connection is established. The second 1322 * SYN is inflight, and if it arrives with an ISN that falls within the 1323 * receive window, the connection is killed. 1324 * 1325 * However, since cookies have other problems, this may not be worth 1326 * worrying about. 1327 */ 1328 1329static u_int32_t 1330syncookie_generate(struct syncache *sc) 1331{ 1332 u_int32_t md5_buffer[4]; 1333 u_int32_t data; 1334 int idx, i; 1335 struct md5_add add; 1336 1337 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1338 1339 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1340 if (tcp_secret[idx].ts_expire < ticks) { 1341 for (i = 0; i < 4; i++) 1342 tcp_secret[idx].ts_secbits[i] = arc4random(); 1343 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1344 } 1345 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1346 if (tcp_msstab[data] <= sc->sc_peer_mss) 1347 break; 1348 data = (data << SYNCOOKIE_WNDBITS) | idx; 1349 data ^= sc->sc_irs; /* peer's iss */ 1350 MD5Init(&syn_ctx); 1351#ifdef INET6 1352 if (sc->sc_inc.inc_isipv6) { 1353 MD5Add(sc->sc_inc.inc6_laddr); 1354 MD5Add(sc->sc_inc.inc6_faddr); 1355 add.laddr = 0; 1356 add.faddr = 0; 1357 } else 1358#endif 1359 { 1360 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1361 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1362 } 1363 add.lport = sc->sc_inc.inc_lport; 1364 add.fport = sc->sc_inc.inc_fport; 1365 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1366 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1367 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1368 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1369 MD5Add(add); 1370 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1371 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1372 return (data); 1373} 1374 1375static struct syncache * 1376syncookie_lookup(inc, th, so) 1377 struct in_conninfo *inc; 1378 struct tcphdr *th; 1379 struct socket *so; 1380{ 1381 u_int32_t md5_buffer[4]; 1382 struct syncache *sc; 1383 u_int32_t data; 1384 int wnd, idx; 1385 struct md5_add add; 1386 1387 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1388 1389 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1390 idx = data & SYNCOOKIE_WNDMASK; 1391 if (tcp_secret[idx].ts_expire < ticks || 1392 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1393 return (NULL); 1394 MD5Init(&syn_ctx); 1395#ifdef INET6 1396 if (inc->inc_isipv6) { 1397 MD5Add(inc->inc6_laddr); 1398 MD5Add(inc->inc6_faddr); 1399 add.laddr = 0; 1400 add.faddr = 0; 1401 } else 1402#endif 1403 { 1404 add.laddr = inc->inc_laddr.s_addr; 1405 add.faddr = inc->inc_faddr.s_addr; 1406 } 1407 add.lport = inc->inc_lport; 1408 add.fport = inc->inc_fport; 1409 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1410 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1411 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1412 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1413 MD5Add(add); 1414 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1415 data ^= md5_buffer[0]; 1416 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1417 return (NULL); 1418 data = data >> SYNCOOKIE_WNDBITS; 1419 1420 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 1421 if (sc == NULL) 1422 return (NULL); 1423 /* 1424 * Fill in the syncache values. 1425 * XXX duplicate code from syncache_add 1426 */ 1427 sc->sc_ipopts = NULL; 1428 sc->sc_inc.inc_fport = inc->inc_fport; 1429 sc->sc_inc.inc_lport = inc->inc_lport; 1430#ifdef INET6 1431 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1432 if (inc->inc_isipv6) { 1433 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1434 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1435 } else 1436#endif 1437 { 1438 sc->sc_inc.inc_faddr = inc->inc_faddr; 1439 sc->sc_inc.inc_laddr = inc->inc_laddr; 1440 } 1441 sc->sc_irs = th->th_seq - 1; 1442 sc->sc_iss = th->th_ack - 1; 1443 wnd = sbspace(&so->so_rcv); 1444 wnd = imax(wnd, 0); 1445 wnd = imin(wnd, TCP_MAXWIN); 1446 sc->sc_wnd = wnd; 1447 sc->sc_flags = 0; 1448 sc->sc_rxtslot = 0; 1449 sc->sc_peer_mss = tcp_msstab[data]; 1450 return (sc); 1451} 1452