1 /*- 2 * Copyright (c) 2001 Networks Associates Technologies, 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: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $ 35 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.2 2003/06/17 04:28:51 dillon Exp $ 36 */ 37 38 #include "opt_inet6.h" 39 #include "opt_ipsec.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/mbuf.h> 47 #include <sys/md5.h> 48 #include <sys/proc.h> /* for proc0 declaration */ 49 #include <sys/random.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 53 #include <net/if.h> 54 #include <net/route.h> 55 56 #include <netinet/in.h> 57 #include <netinet/in_systm.h> 58 #include <netinet/ip.h> 59 #include <netinet/in_var.h> 60 #include <netinet/in_pcb.h> 61 #include <netinet/ip_var.h> 62 #ifdef INET6 63 #include <netinet/ip6.h> 64 #include <netinet/icmp6.h> 65 #include <netinet6/nd6.h> 66 #include <netinet6/ip6_var.h> 67 #include <netinet6/in6_pcb.h> 68 #endif 69 #include <netinet/tcp.h> 70 #include <netinet/tcp_fsm.h> 71 #include <netinet/tcp_seq.h> 72 #include <netinet/tcp_timer.h> 73 #include <netinet/tcp_var.h> 74 #ifdef INET6 75 #include <netinet6/tcp6_var.h> 76 #endif 77 78 #ifdef IPSEC 79 #include <netinet6/ipsec.h> 80 #ifdef INET6 81 #include <netinet6/ipsec6.h> 82 #endif 83 #include <netkey/key.h> 84 #endif /*IPSEC*/ 85 86 #ifdef FAST_IPSEC 87 #include <netipsec/ipsec.h> 88 #ifdef INET6 89 #include <netipsec/ipsec6.h> 90 #endif 91 #include <netipsec/key.h> 92 #define IPSEC 93 #endif /*FAST_IPSEC*/ 94 95 #include <machine/in_cksum.h> 96 #include <vm/vm_zone.h> 97 98 static int tcp_syncookies = 1; 99 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 100 &tcp_syncookies, 0, 101 "Use TCP SYN cookies if the syncache overflows"); 102 103 static void syncache_drop(struct syncache *, struct syncache_head *); 104 static void syncache_free(struct syncache *); 105 static void syncache_insert(struct syncache *, struct syncache_head *); 106 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 107 static int syncache_respond(struct syncache *, struct mbuf *); 108 static struct socket *syncache_socket(struct syncache *, struct socket *); 109 static void syncache_timer(void *); 110 static u_int32_t syncookie_generate(struct syncache *); 111 static struct syncache *syncookie_lookup(struct in_conninfo *, 112 struct tcphdr *, struct socket *); 113 114 /* 115 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 116 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 117 * the odds are that the user has given up attempting to connect by then. 118 */ 119 #define SYNCACHE_MAXREXMTS 3 120 121 /* Arbitrary values */ 122 #define TCP_SYNCACHE_HASHSIZE 512 123 #define TCP_SYNCACHE_BUCKETLIMIT 30 124 125 struct tcp_syncache { 126 struct syncache_head *hashbase; 127 struct vm_zone *zone; 128 u_int hashsize; 129 u_int hashmask; 130 u_int bucket_limit; 131 u_int cache_count; 132 u_int cache_limit; 133 u_int rexmt_limit; 134 u_int hash_secret; 135 u_int next_reseed; 136 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 137 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 138 }; 139 static struct tcp_syncache tcp_syncache; 140 141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 142 143 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 144 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 145 146 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 147 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 148 149 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 150 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 151 152 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 153 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 154 155 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 156 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 157 158 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 159 160 #define SYNCACHE_HASH(inc, mask) \ 161 ((tcp_syncache.hash_secret ^ \ 162 (inc)->inc_faddr.s_addr ^ \ 163 ((inc)->inc_faddr.s_addr >> 16) ^ \ 164 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 165 166 #define SYNCACHE_HASH6(inc, mask) \ 167 ((tcp_syncache.hash_secret ^ \ 168 (inc)->inc6_faddr.s6_addr32[0] ^ \ 169 (inc)->inc6_faddr.s6_addr32[3] ^ \ 170 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 171 172 #define ENDPTS_EQ(a, b) ( \ 173 (a)->ie_fport == (b)->ie_fport && \ 174 (a)->ie_lport == (b)->ie_lport && \ 175 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 176 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 177 ) 178 179 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 180 181 #define SYNCACHE_TIMEOUT(sc, slot) do { \ 182 sc->sc_rxtslot = slot; \ 183 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; \ 184 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[slot], sc, sc_timerq); \ 185 if (!callout_active(&tcp_syncache.tt_timerq[slot])) \ 186 callout_reset(&tcp_syncache.tt_timerq[slot], \ 187 TCPTV_RTOBASE * tcp_backoff[slot], \ 188 syncache_timer, (void *)((intptr_t)slot)); \ 189 } while (0) 190 191 static void 192 syncache_free(struct syncache *sc) 193 { 194 struct rtentry *rt; 195 196 if (sc->sc_ipopts) 197 (void) m_free(sc->sc_ipopts); 198 #ifdef INET6 199 if (sc->sc_inc.inc_isipv6) 200 rt = sc->sc_route6.ro_rt; 201 else 202 #endif 203 rt = sc->sc_route.ro_rt; 204 if (rt != NULL) { 205 /* 206 * If this is the only reference to a protocol cloned 207 * route, remove it immediately. 208 */ 209 if (rt->rt_flags & RTF_WASCLONED && 210 (sc->sc_flags & SCF_KEEPROUTE) == 0 && 211 rt->rt_refcnt == 1) 212 rtrequest(RTM_DELETE, rt_key(rt), 213 rt->rt_gateway, rt_mask(rt), 214 rt->rt_flags, NULL); 215 RTFREE(rt); 216 } 217 zfree(tcp_syncache.zone, sc); 218 } 219 220 void 221 syncache_init(void) 222 { 223 int i; 224 225 tcp_syncache.cache_count = 0; 226 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 227 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 228 tcp_syncache.cache_limit = 229 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 230 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 231 tcp_syncache.next_reseed = 0; 232 tcp_syncache.hash_secret = arc4random(); 233 234 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 235 &tcp_syncache.hashsize); 236 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 237 &tcp_syncache.cache_limit); 238 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 239 &tcp_syncache.bucket_limit); 240 if (!powerof2(tcp_syncache.hashsize)) { 241 printf("WARNING: syncache hash size is not a power of 2.\n"); 242 tcp_syncache.hashsize = 512; /* safe default */ 243 } 244 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 245 246 /* Allocate the hash table. */ 247 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 248 tcp_syncache.hashsize * sizeof(struct syncache_head), 249 M_SYNCACHE, M_WAITOK); 250 251 /* Initialize the hash buckets. */ 252 for (i = 0; i < tcp_syncache.hashsize; i++) { 253 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 254 tcp_syncache.hashbase[i].sch_length = 0; 255 } 256 257 /* Initialize the timer queues. */ 258 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 259 TAILQ_INIT(&tcp_syncache.timerq[i]); 260 callout_init(&tcp_syncache.tt_timerq[i]); 261 } 262 263 /* 264 * Allocate the syncache entries. Allow the zone to allocate one 265 * more entry than cache limit, so a new entry can bump out an 266 * older one. 267 */ 268 tcp_syncache.cache_limit -= 1; 269 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 270 tcp_syncache.cache_limit, ZONE_INTERRUPT, 0); 271 } 272 273 static void 274 syncache_insert(sc, sch) 275 struct syncache *sc; 276 struct syncache_head *sch; 277 { 278 struct syncache *sc2; 279 int s, i; 280 281 /* 282 * Make sure that we don't overflow the per-bucket 283 * limit or the total cache size limit. 284 */ 285 s = splnet(); 286 if (sch->sch_length >= tcp_syncache.bucket_limit) { 287 /* 288 * The bucket is full, toss the oldest element. 289 */ 290 sc2 = TAILQ_FIRST(&sch->sch_bucket); 291 sc2->sc_tp->ts_recent = ticks; 292 syncache_drop(sc2, sch); 293 tcpstat.tcps_sc_bucketoverflow++; 294 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) { 295 /* 296 * The cache is full. Toss the oldest entry in the 297 * entire cache. This is the front entry in the 298 * first non-empty timer queue with the largest 299 * timeout value. 300 */ 301 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 302 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]); 303 if (sc2 != NULL) 304 break; 305 } 306 sc2->sc_tp->ts_recent = ticks; 307 syncache_drop(sc2, NULL); 308 tcpstat.tcps_sc_cacheoverflow++; 309 } 310 311 /* Initialize the entry's timer. */ 312 SYNCACHE_TIMEOUT(sc, 0); 313 314 /* Put it into the bucket. */ 315 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 316 sch->sch_length++; 317 tcp_syncache.cache_count++; 318 tcpstat.tcps_sc_added++; 319 splx(s); 320 } 321 322 static void 323 syncache_drop(sc, sch) 324 struct syncache *sc; 325 struct syncache_head *sch; 326 { 327 int s; 328 329 if (sch == NULL) { 330 #ifdef INET6 331 if (sc->sc_inc.inc_isipv6) { 332 sch = &tcp_syncache.hashbase[ 333 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 334 } else 335 #endif 336 { 337 sch = &tcp_syncache.hashbase[ 338 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 339 } 340 } 341 342 s = splnet(); 343 344 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 345 sch->sch_length--; 346 tcp_syncache.cache_count--; 347 348 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq); 349 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot])) 350 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]); 351 splx(s); 352 353 syncache_free(sc); 354 } 355 356 /* 357 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 358 * If we have retransmitted an entry the maximum number of times, expire it. 359 */ 360 static void 361 syncache_timer(xslot) 362 void *xslot; 363 { 364 intptr_t slot = (intptr_t)xslot; 365 struct syncache *sc, *nsc; 366 struct inpcb *inp; 367 int s; 368 369 s = splnet(); 370 if (callout_pending(&tcp_syncache.tt_timerq[slot]) || 371 !callout_active(&tcp_syncache.tt_timerq[slot])) { 372 splx(s); 373 return; 374 } 375 callout_deactivate(&tcp_syncache.tt_timerq[slot]); 376 377 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]); 378 while (nsc != NULL) { 379 if (ticks < nsc->sc_rxttime) 380 break; 381 sc = nsc; 382 inp = sc->sc_tp->t_inpcb; 383 if (slot == SYNCACHE_MAXREXMTS || 384 slot >= tcp_syncache.rexmt_limit || 385 inp->inp_gencnt != sc->sc_inp_gencnt) { 386 nsc = TAILQ_NEXT(sc, sc_timerq); 387 syncache_drop(sc, NULL); 388 tcpstat.tcps_sc_stale++; 389 continue; 390 } 391 /* 392 * syncache_respond() may call back into the syncache to 393 * to modify another entry, so do not obtain the next 394 * entry on the timer chain until it has completed. 395 */ 396 (void) syncache_respond(sc, NULL); 397 nsc = TAILQ_NEXT(sc, sc_timerq); 398 tcpstat.tcps_sc_retransmitted++; 399 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq); 400 SYNCACHE_TIMEOUT(sc, slot + 1); 401 } 402 if (nsc != NULL) 403 callout_reset(&tcp_syncache.tt_timerq[slot], 404 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot)); 405 splx(s); 406 } 407 408 /* 409 * Find an entry in the syncache. 410 */ 411 struct syncache * 412 syncache_lookup(inc, schp) 413 struct in_conninfo *inc; 414 struct syncache_head **schp; 415 { 416 struct syncache *sc; 417 struct syncache_head *sch; 418 int s; 419 420 #ifdef INET6 421 if (inc->inc_isipv6) { 422 sch = &tcp_syncache.hashbase[ 423 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 424 *schp = sch; 425 s = splnet(); 426 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 427 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) { 428 splx(s); 429 return (sc); 430 } 431 } 432 splx(s); 433 } else 434 #endif 435 { 436 sch = &tcp_syncache.hashbase[ 437 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 438 *schp = sch; 439 s = splnet(); 440 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 441 #ifdef INET6 442 if (sc->sc_inc.inc_isipv6) 443 continue; 444 #endif 445 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) { 446 splx(s); 447 return (sc); 448 } 449 } 450 splx(s); 451 } 452 return (NULL); 453 } 454 455 /* 456 * This function is called when we get a RST for a 457 * non-existent connection, so that we can see if the 458 * connection is in the syn cache. If it is, zap it. 459 */ 460 void 461 syncache_chkrst(inc, th) 462 struct in_conninfo *inc; 463 struct tcphdr *th; 464 { 465 struct syncache *sc; 466 struct syncache_head *sch; 467 468 sc = syncache_lookup(inc, &sch); 469 if (sc == NULL) 470 return; 471 /* 472 * If the RST bit is set, check the sequence number to see 473 * if this is a valid reset segment. 474 * RFC 793 page 37: 475 * In all states except SYN-SENT, all reset (RST) segments 476 * are validated by checking their SEQ-fields. A reset is 477 * valid if its sequence number is in the window. 478 * 479 * The sequence number in the reset segment is normally an 480 * echo of our outgoing acknowlegement numbers, but some hosts 481 * send a reset with the sequence number at the rightmost edge 482 * of our receive window, and we have to handle this case. 483 */ 484 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 485 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 486 syncache_drop(sc, sch); 487 tcpstat.tcps_sc_reset++; 488 } 489 } 490 491 void 492 syncache_badack(inc) 493 struct in_conninfo *inc; 494 { 495 struct syncache *sc; 496 struct syncache_head *sch; 497 498 sc = syncache_lookup(inc, &sch); 499 if (sc != NULL) { 500 syncache_drop(sc, sch); 501 tcpstat.tcps_sc_badack++; 502 } 503 } 504 505 void 506 syncache_unreach(inc, th) 507 struct in_conninfo *inc; 508 struct tcphdr *th; 509 { 510 struct syncache *sc; 511 struct syncache_head *sch; 512 513 /* we are called at splnet() here */ 514 sc = syncache_lookup(inc, &sch); 515 if (sc == NULL) 516 return; 517 518 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 519 if (ntohl(th->th_seq) != sc->sc_iss) 520 return; 521 522 /* 523 * If we've rertransmitted 3 times and this is our second error, 524 * we remove the entry. Otherwise, we allow it to continue on. 525 * This prevents us from incorrectly nuking an entry during a 526 * spurious network outage. 527 * 528 * See tcp_notify(). 529 */ 530 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 531 sc->sc_flags |= SCF_UNREACH; 532 return; 533 } 534 syncache_drop(sc, sch); 535 tcpstat.tcps_sc_unreach++; 536 } 537 538 /* 539 * Build a new TCP socket structure from a syncache entry. 540 */ 541 static struct socket * 542 syncache_socket(sc, lso) 543 struct syncache *sc; 544 struct socket *lso; 545 { 546 struct inpcb *inp = NULL; 547 struct socket *so; 548 struct tcpcb *tp; 549 550 /* 551 * Ok, create the full blown connection, and set things up 552 * as they would have been set up if we had created the 553 * connection when the SYN arrived. If we can't create 554 * the connection, abort it. 555 */ 556 so = sonewconn(lso, SS_ISCONNECTED); 557 if (so == NULL) { 558 /* 559 * Drop the connection; we will send a RST if the peer 560 * retransmits the ACK, 561 */ 562 tcpstat.tcps_listendrop++; 563 goto abort; 564 } 565 566 inp = sotoinpcb(so); 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_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 601 printf("syncache_expand: could not copy policy\n"); 602 #endif 603 #ifdef INET6 604 if (sc->sc_inc.inc_isipv6) { 605 struct inpcb *oinp = sotoinpcb(lso); 606 struct in6_addr laddr6; 607 struct sockaddr_in6 *sin6; 608 /* 609 * Inherit socket options from the listening socket. 610 * Note that in6p_inputopts are not (and should not be) 611 * copied, since it stores previously received options and is 612 * used to detect if each new option is different than the 613 * previous one and hence should be passed to a user. 614 * If we copied in6p_inputopts, a user would not be able to 615 * receive options just after calling the accept system call. 616 */ 617 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 618 if (oinp->in6p_outputopts) 619 inp->in6p_outputopts = 620 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 621 inp->in6p_route = sc->sc_route6; 622 sc->sc_route6.ro_rt = NULL; 623 624 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, 625 M_SONAME, M_NOWAIT | M_ZERO); 626 if (sin6 == NULL) 627 goto abort; 628 sin6->sin6_family = AF_INET6; 629 sin6->sin6_len = sizeof(*sin6); 630 sin6->sin6_addr = sc->sc_inc.inc6_faddr; 631 sin6->sin6_port = sc->sc_inc.inc_fport; 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, &proc0)) { 636 inp->in6p_laddr = laddr6; 637 FREE(sin6, M_SONAME); 638 goto abort; 639 } 640 FREE(sin6, M_SONAME); 641 } else 642 #endif 643 { 644 struct in_addr laddr; 645 struct sockaddr_in *sin; 646 647 inp->inp_options = ip_srcroute(); 648 if (inp->inp_options == NULL) { 649 inp->inp_options = sc->sc_ipopts; 650 sc->sc_ipopts = NULL; 651 } 652 inp->inp_route = sc->sc_route; 653 sc->sc_route.ro_rt = NULL; 654 655 MALLOC(sin, struct sockaddr_in *, sizeof *sin, 656 M_SONAME, M_NOWAIT | M_ZERO); 657 if (sin == NULL) 658 goto abort; 659 sin->sin_family = AF_INET; 660 sin->sin_len = sizeof(*sin); 661 sin->sin_addr = sc->sc_inc.inc_faddr; 662 sin->sin_port = sc->sc_inc.inc_fport; 663 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 664 laddr = inp->inp_laddr; 665 if (inp->inp_laddr.s_addr == INADDR_ANY) 666 inp->inp_laddr = sc->sc_inc.inc_laddr; 667 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 668 inp->inp_laddr = laddr; 669 FREE(sin, M_SONAME); 670 goto abort; 671 } 672 FREE(sin, M_SONAME); 673 } 674 675 tp = intotcpcb(inp); 676 tp->t_state = TCPS_SYN_RECEIVED; 677 tp->iss = sc->sc_iss; 678 tp->irs = sc->sc_irs; 679 tcp_rcvseqinit(tp); 680 tcp_sendseqinit(tp); 681 tp->snd_wl1 = sc->sc_irs; 682 tp->rcv_up = sc->sc_irs + 1; 683 tp->rcv_wnd = sc->sc_wnd; 684 tp->rcv_adv += tp->rcv_wnd; 685 686 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 687 if (sc->sc_flags & SCF_NOOPT) 688 tp->t_flags |= TF_NOOPT; 689 if (sc->sc_flags & SCF_WINSCALE) { 690 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 691 tp->requested_s_scale = sc->sc_requested_s_scale; 692 tp->request_r_scale = sc->sc_request_r_scale; 693 } 694 if (sc->sc_flags & SCF_TIMESTAMP) { 695 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 696 tp->ts_recent = sc->sc_tsrecent; 697 tp->ts_recent_age = ticks; 698 } 699 if (sc->sc_flags & SCF_CC) { 700 /* 701 * Initialization of the tcpcb for transaction; 702 * set SND.WND = SEG.WND, 703 * initialize CCsend and CCrecv. 704 */ 705 tp->t_flags |= TF_REQ_CC|TF_RCVD_CC; 706 tp->cc_send = sc->sc_cc_send; 707 tp->cc_recv = sc->sc_cc_recv; 708 } 709 710 tcp_mss(tp, sc->sc_peer_mss); 711 712 /* 713 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 714 */ 715 if (sc->sc_rxtslot != 0) 716 tp->snd_cwnd = tp->t_maxseg; 717 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 718 719 tcpstat.tcps_accepts++; 720 return (so); 721 722 abort: 723 if (so != NULL) 724 (void) soabort(so); 725 return (NULL); 726 } 727 728 /* 729 * This function gets called when we receive an ACK for a 730 * socket in the LISTEN state. We look up the connection 731 * in the syncache, and if its there, we pull it out of 732 * the cache and turn it into a full-blown connection in 733 * the SYN-RECEIVED state. 734 */ 735 int 736 syncache_expand(inc, th, sop, m) 737 struct in_conninfo *inc; 738 struct tcphdr *th; 739 struct socket **sop; 740 struct mbuf *m; 741 { 742 struct syncache *sc; 743 struct syncache_head *sch; 744 struct socket *so; 745 746 sc = syncache_lookup(inc, &sch); 747 if (sc == NULL) { 748 /* 749 * There is no syncache entry, so see if this ACK is 750 * a returning syncookie. To do this, first: 751 * A. See if this socket has had a syncache entry dropped in 752 * the past. We don't want to accept a bogus syncookie 753 * if we've never received a SYN. 754 * B. check that the syncookie is valid. If it is, then 755 * cobble up a fake syncache entry, and return. 756 */ 757 if (!tcp_syncookies) 758 return (0); 759 sc = syncookie_lookup(inc, th, *sop); 760 if (sc == NULL) 761 return (0); 762 sch = NULL; 763 tcpstat.tcps_sc_recvcookie++; 764 } 765 766 /* 767 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 768 */ 769 if (th->th_ack != sc->sc_iss + 1) 770 return (0); 771 772 so = syncache_socket(sc, *sop); 773 if (so == NULL) { 774 #if 0 775 resetandabort: 776 /* XXXjlemon check this - is this correct? */ 777 (void) tcp_respond(NULL, m, m, th, 778 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 779 #endif 780 m_freem(m); /* XXX only needed for above */ 781 tcpstat.tcps_sc_aborted++; 782 } else { 783 sc->sc_flags |= SCF_KEEPROUTE; 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 */ 807 int 808 syncache_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 struct rmxp_tao *taop; 821 int i, s, win; 822 823 so = *sop; 824 tp = sototcpcb(so); 825 826 /* 827 * Remember the IP options, if any. 828 */ 829 #ifdef INET6 830 if (!inc->inc_isipv6) 831 #endif 832 ipopts = ip_srcroute(); 833 834 /* 835 * See if we already have an entry for this connection. 836 * If we do, resend the SYN,ACK, and reset the retransmit timer. 837 * 838 * XXX 839 * should the syncache be re-initialized with the contents 840 * of the new SYN here (which may have different options?) 841 */ 842 sc = syncache_lookup(inc, &sch); 843 if (sc != NULL) { 844 tcpstat.tcps_sc_dupsyn++; 845 if (ipopts) { 846 /* 847 * If we were remembering a previous source route, 848 * forget it and use the new one we've been given. 849 */ 850 if (sc->sc_ipopts) 851 (void) m_free(sc->sc_ipopts); 852 sc->sc_ipopts = ipopts; 853 } 854 /* 855 * Update timestamp if present. 856 */ 857 if (sc->sc_flags & SCF_TIMESTAMP) 858 sc->sc_tsrecent = to->to_tsval; 859 /* 860 * PCB may have changed, pick up new values. 861 */ 862 sc->sc_tp = tp; 863 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 864 if (syncache_respond(sc, m) == 0) { 865 s = splnet(); 866 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 867 sc, sc_timerq); 868 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 869 splx(s); 870 tcpstat.tcps_sndacks++; 871 tcpstat.tcps_sndtotal++; 872 } 873 *sop = NULL; 874 return (1); 875 } 876 877 sc = zalloc(tcp_syncache.zone); 878 if (sc == NULL) { 879 /* 880 * The zone allocator couldn't provide more entries. 881 * Treat this as if the cache was full; drop the oldest 882 * entry and insert the new one. 883 */ 884 s = splnet(); 885 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 886 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]); 887 if (sc != NULL) 888 break; 889 } 890 sc->sc_tp->ts_recent = ticks; 891 syncache_drop(sc, NULL); 892 splx(s); 893 tcpstat.tcps_sc_zonefail++; 894 sc = zalloc(tcp_syncache.zone); 895 if (sc == NULL) { 896 if (ipopts) 897 (void) m_free(ipopts); 898 return (0); 899 } 900 } 901 902 /* 903 * Fill in the syncache values. 904 */ 905 bzero(sc, sizeof(*sc)); 906 sc->sc_tp = tp; 907 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 908 sc->sc_ipopts = ipopts; 909 sc->sc_inc.inc_fport = inc->inc_fport; 910 sc->sc_inc.inc_lport = inc->inc_lport; 911 #ifdef INET6 912 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 913 if (inc->inc_isipv6) { 914 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 915 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 916 sc->sc_route6.ro_rt = NULL; 917 } else 918 #endif 919 { 920 sc->sc_inc.inc_faddr = inc->inc_faddr; 921 sc->sc_inc.inc_laddr = inc->inc_laddr; 922 sc->sc_route.ro_rt = NULL; 923 } 924 sc->sc_irs = th->th_seq; 925 sc->sc_flags = 0; 926 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 927 if (tcp_syncookies) 928 sc->sc_iss = syncookie_generate(sc); 929 else 930 sc->sc_iss = arc4random(); 931 932 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 933 win = sbspace(&so->so_rcv); 934 win = imax(win, 0); 935 win = imin(win, TCP_MAXWIN); 936 sc->sc_wnd = win; 937 938 if (tcp_do_rfc1323) { 939 /* 940 * A timestamp received in a SYN makes 941 * it ok to send timestamp requests and replies. 942 */ 943 if (to->to_flags & TOF_TS) { 944 sc->sc_tsrecent = to->to_tsval; 945 sc->sc_flags |= SCF_TIMESTAMP; 946 } 947 if (to->to_flags & TOF_SCALE) { 948 int wscale = 0; 949 950 /* Compute proper scaling value from buffer space */ 951 while (wscale < TCP_MAX_WINSHIFT && 952 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 953 wscale++; 954 sc->sc_request_r_scale = wscale; 955 sc->sc_requested_s_scale = to->to_requested_s_scale; 956 sc->sc_flags |= SCF_WINSCALE; 957 } 958 } 959 if (tcp_do_rfc1644) { 960 /* 961 * A CC or CC.new option received in a SYN makes 962 * it ok to send CC in subsequent segments. 963 */ 964 if (to->to_flags & (TOF_CC|TOF_CCNEW)) { 965 sc->sc_cc_recv = to->to_cc; 966 sc->sc_cc_send = CC_INC(tcp_ccgen); 967 sc->sc_flags |= SCF_CC; 968 } 969 } 970 if (tp->t_flags & TF_NOOPT) 971 sc->sc_flags = SCF_NOOPT; 972 973 /* 974 * XXX 975 * We have the option here of not doing TAO (even if the segment 976 * qualifies) and instead fall back to a normal 3WHS via the syncache. 977 * This allows us to apply synflood protection to TAO-qualifying SYNs 978 * also. However, there should be a hueristic to determine when to 979 * do this, and is not present at the moment. 980 */ 981 982 /* 983 * Perform TAO test on incoming CC (SEG.CC) option, if any. 984 * - compare SEG.CC against cached CC from the same host, if any. 985 * - if SEG.CC > chached value, SYN must be new and is accepted 986 * immediately: save new CC in the cache, mark the socket 987 * connected, enter ESTABLISHED state, turn on flag to 988 * send a SYN in the next segment. 989 * A virtual advertised window is set in rcv_adv to 990 * initialize SWS prevention. Then enter normal segment 991 * processing: drop SYN, process data and FIN. 992 * - otherwise do a normal 3-way handshake. 993 */ 994 taop = tcp_gettaocache(&sc->sc_inc); 995 if ((to->to_flags & TOF_CC) != 0) { 996 if (((tp->t_flags & TF_NOPUSH) != 0) && 997 sc->sc_flags & SCF_CC && 998 taop != NULL && taop->tao_cc != 0 && 999 CC_GT(to->to_cc, taop->tao_cc)) { 1000 sc->sc_rxtslot = 0; 1001 so = syncache_socket(sc, *sop); 1002 if (so != NULL) { 1003 sc->sc_flags |= SCF_KEEPROUTE; 1004 taop->tao_cc = to->to_cc; 1005 *sop = so; 1006 } 1007 syncache_free(sc); 1008 return (so != NULL); 1009 } 1010 } else { 1011 /* 1012 * No CC option, but maybe CC.NEW: invalidate cached value. 1013 */ 1014 if (taop != NULL) 1015 taop->tao_cc = 0; 1016 } 1017 /* 1018 * TAO test failed or there was no CC option, 1019 * do a standard 3-way handshake. 1020 */ 1021 if (syncache_respond(sc, m) == 0) { 1022 syncache_insert(sc, sch); 1023 tcpstat.tcps_sndacks++; 1024 tcpstat.tcps_sndtotal++; 1025 } else { 1026 syncache_free(sc); 1027 tcpstat.tcps_sc_dropped++; 1028 } 1029 *sop = NULL; 1030 return (1); 1031 } 1032 1033 static int 1034 syncache_respond(sc, m) 1035 struct syncache *sc; 1036 struct mbuf *m; 1037 { 1038 u_int8_t *optp; 1039 int optlen, error; 1040 u_int16_t tlen, hlen, mssopt; 1041 struct ip *ip = NULL; 1042 struct rtentry *rt; 1043 struct tcphdr *th; 1044 #ifdef INET6 1045 struct ip6_hdr *ip6 = NULL; 1046 #endif 1047 1048 #ifdef INET6 1049 if (sc->sc_inc.inc_isipv6) { 1050 rt = tcp_rtlookup6(&sc->sc_inc); 1051 if (rt != NULL) 1052 mssopt = rt->rt_ifp->if_mtu - 1053 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1054 else 1055 mssopt = tcp_v6mssdflt; 1056 hlen = sizeof(struct ip6_hdr); 1057 } else 1058 #endif 1059 { 1060 rt = tcp_rtlookup(&sc->sc_inc); 1061 if (rt != NULL) 1062 mssopt = rt->rt_ifp->if_mtu - 1063 (sizeof(struct ip) + sizeof(struct tcphdr)); 1064 else 1065 mssopt = tcp_mssdflt; 1066 hlen = sizeof(struct ip); 1067 } 1068 1069 /* Compute the size of the TCP options. */ 1070 if (sc->sc_flags & SCF_NOOPT) { 1071 optlen = 0; 1072 } else { 1073 optlen = TCPOLEN_MAXSEG + 1074 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1075 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1076 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0); 1077 } 1078 tlen = hlen + sizeof(struct tcphdr) + optlen; 1079 1080 /* 1081 * XXX 1082 * assume that the entire packet will fit in a header mbuf 1083 */ 1084 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1085 1086 /* 1087 * XXX shouldn't this reuse the mbuf if possible ? 1088 * Create the IP+TCP header from scratch. 1089 */ 1090 if (m) 1091 m_freem(m); 1092 1093 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1094 if (m == NULL) 1095 return (ENOBUFS); 1096 m->m_data += max_linkhdr; 1097 m->m_len = tlen; 1098 m->m_pkthdr.len = tlen; 1099 m->m_pkthdr.rcvif = NULL; 1100 1101 #ifdef INET6 1102 if (sc->sc_inc.inc_isipv6) { 1103 ip6 = mtod(m, struct ip6_hdr *); 1104 ip6->ip6_vfc = IPV6_VERSION; 1105 ip6->ip6_nxt = IPPROTO_TCP; 1106 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1107 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1108 ip6->ip6_plen = htons(tlen - hlen); 1109 /* ip6_hlim is set after checksum */ 1110 /* ip6_flow = ??? */ 1111 1112 th = (struct tcphdr *)(ip6 + 1); 1113 } else 1114 #endif 1115 { 1116 ip = mtod(m, struct ip *); 1117 ip->ip_v = IPVERSION; 1118 ip->ip_hl = sizeof(struct ip) >> 2; 1119 ip->ip_len = tlen; 1120 ip->ip_id = 0; 1121 ip->ip_off = 0; 1122 ip->ip_sum = 0; 1123 ip->ip_p = IPPROTO_TCP; 1124 ip->ip_src = sc->sc_inc.inc_laddr; 1125 ip->ip_dst = sc->sc_inc.inc_faddr; 1126 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1127 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1128 1129 /* 1130 * See if we should do MTU discovery. Route lookups are expensive, 1131 * so we will only unset the DF bit if: 1132 * 1133 * 1) path_mtu_discovery is disabled 1134 * 2) the SCF_UNREACH flag has been set 1135 */ 1136 if (path_mtu_discovery 1137 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1138 ip->ip_off |= IP_DF; 1139 } 1140 1141 th = (struct tcphdr *)(ip + 1); 1142 } 1143 th->th_sport = sc->sc_inc.inc_lport; 1144 th->th_dport = sc->sc_inc.inc_fport; 1145 1146 th->th_seq = htonl(sc->sc_iss); 1147 th->th_ack = htonl(sc->sc_irs + 1); 1148 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1149 th->th_x2 = 0; 1150 th->th_flags = TH_SYN|TH_ACK; 1151 th->th_win = htons(sc->sc_wnd); 1152 th->th_urp = 0; 1153 1154 /* Tack on the TCP options. */ 1155 if (optlen == 0) 1156 goto no_options; 1157 optp = (u_int8_t *)(th + 1); 1158 *optp++ = TCPOPT_MAXSEG; 1159 *optp++ = TCPOLEN_MAXSEG; 1160 *optp++ = (mssopt >> 8) & 0xff; 1161 *optp++ = mssopt & 0xff; 1162 1163 if (sc->sc_flags & SCF_WINSCALE) { 1164 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1165 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1166 sc->sc_request_r_scale); 1167 optp += 4; 1168 } 1169 1170 if (sc->sc_flags & SCF_TIMESTAMP) { 1171 u_int32_t *lp = (u_int32_t *)(optp); 1172 1173 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1174 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1175 *lp++ = htonl(ticks); 1176 *lp = htonl(sc->sc_tsrecent); 1177 optp += TCPOLEN_TSTAMP_APPA; 1178 } 1179 1180 /* 1181 * Send CC and CC.echo if we received CC from our peer. 1182 */ 1183 if (sc->sc_flags & SCF_CC) { 1184 u_int32_t *lp = (u_int32_t *)(optp); 1185 1186 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1187 *lp++ = htonl(sc->sc_cc_send); 1188 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1189 *lp = htonl(sc->sc_cc_recv); 1190 optp += TCPOLEN_CC_APPA * 2; 1191 } 1192 no_options: 1193 1194 #ifdef INET6 1195 if (sc->sc_inc.inc_isipv6) { 1196 struct route_in6 *ro6 = &sc->sc_route6; 1197 1198 th->th_sum = 0; 1199 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1200 ip6->ip6_hlim = in6_selecthlim(NULL, 1201 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1202 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1203 sc->sc_tp->t_inpcb); 1204 } else 1205 #endif 1206 { 1207 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1208 htons(tlen - hlen + IPPROTO_TCP)); 1209 m->m_pkthdr.csum_flags = CSUM_TCP; 1210 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1211 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1212 sc->sc_tp->t_inpcb); 1213 } 1214 return (error); 1215 } 1216 1217 /* 1218 * cookie layers: 1219 * 1220 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1221 * | peer iss | 1222 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1223 * | 0 |(A)| | 1224 * (A): peer mss index 1225 */ 1226 1227 /* 1228 * The values below are chosen to minimize the size of the tcp_secret 1229 * table, as well as providing roughly a 16 second lifetime for the cookie. 1230 */ 1231 1232 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1233 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1234 1235 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1236 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1237 #define SYNCOOKIE_TIMEOUT \ 1238 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1239 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1240 1241 static struct { 1242 u_int32_t ts_secbits[4]; 1243 u_int ts_expire; 1244 } tcp_secret[SYNCOOKIE_NSECRETS]; 1245 1246 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1247 1248 static MD5_CTX syn_ctx; 1249 1250 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1251 1252 struct md5_add { 1253 u_int32_t laddr, faddr; 1254 u_int32_t secbits[4]; 1255 u_int16_t lport, fport; 1256 }; 1257 1258 #ifdef CTASSERT 1259 CTASSERT(sizeof(struct md5_add) == 28); 1260 #endif 1261 1262 /* 1263 * Consider the problem of a recreated (and retransmitted) cookie. If the 1264 * original SYN was accepted, the connection is established. The second 1265 * SYN is inflight, and if it arrives with an ISN that falls within the 1266 * receive window, the connection is killed. 1267 * 1268 * However, since cookies have other problems, this may not be worth 1269 * worrying about. 1270 */ 1271 1272 static u_int32_t 1273 syncookie_generate(struct syncache *sc) 1274 { 1275 u_int32_t md5_buffer[4]; 1276 u_int32_t data; 1277 int idx, i; 1278 struct md5_add add; 1279 1280 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1281 if (tcp_secret[idx].ts_expire < ticks) { 1282 for (i = 0; i < 4; i++) 1283 tcp_secret[idx].ts_secbits[i] = arc4random(); 1284 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1285 } 1286 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1287 if (tcp_msstab[data] <= sc->sc_peer_mss) 1288 break; 1289 data = (data << SYNCOOKIE_WNDBITS) | idx; 1290 data ^= sc->sc_irs; /* peer's iss */ 1291 MD5Init(&syn_ctx); 1292 #ifdef INET6 1293 if (sc->sc_inc.inc_isipv6) { 1294 MD5Add(sc->sc_inc.inc6_laddr); 1295 MD5Add(sc->sc_inc.inc6_faddr); 1296 add.laddr = 0; 1297 add.faddr = 0; 1298 } else 1299 #endif 1300 { 1301 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1302 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1303 } 1304 add.lport = sc->sc_inc.inc_lport; 1305 add.fport = sc->sc_inc.inc_fport; 1306 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1307 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1308 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1309 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1310 MD5Add(add); 1311 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1312 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1313 return (data); 1314 } 1315 1316 static struct syncache * 1317 syncookie_lookup(inc, th, so) 1318 struct in_conninfo *inc; 1319 struct tcphdr *th; 1320 struct socket *so; 1321 { 1322 u_int32_t md5_buffer[4]; 1323 struct syncache *sc; 1324 u_int32_t data; 1325 int wnd, idx; 1326 struct md5_add add; 1327 1328 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1329 idx = data & SYNCOOKIE_WNDMASK; 1330 if (tcp_secret[idx].ts_expire < ticks || 1331 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1332 return (NULL); 1333 MD5Init(&syn_ctx); 1334 #ifdef INET6 1335 if (inc->inc_isipv6) { 1336 MD5Add(inc->inc6_laddr); 1337 MD5Add(inc->inc6_faddr); 1338 add.laddr = 0; 1339 add.faddr = 0; 1340 } else 1341 #endif 1342 { 1343 add.laddr = inc->inc_laddr.s_addr; 1344 add.faddr = inc->inc_faddr.s_addr; 1345 } 1346 add.lport = inc->inc_lport; 1347 add.fport = inc->inc_fport; 1348 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1349 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1350 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1351 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1352 MD5Add(add); 1353 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1354 data ^= md5_buffer[0]; 1355 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1356 return (NULL); 1357 data = data >> SYNCOOKIE_WNDBITS; 1358 1359 sc = zalloc(tcp_syncache.zone); 1360 if (sc == NULL) 1361 return (NULL); 1362 /* 1363 * Fill in the syncache values. 1364 * XXX duplicate code from syncache_add 1365 */ 1366 sc->sc_ipopts = NULL; 1367 sc->sc_inc.inc_fport = inc->inc_fport; 1368 sc->sc_inc.inc_lport = inc->inc_lport; 1369 #ifdef INET6 1370 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1371 if (inc->inc_isipv6) { 1372 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1373 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1374 sc->sc_route6.ro_rt = NULL; 1375 } else 1376 #endif 1377 { 1378 sc->sc_inc.inc_faddr = inc->inc_faddr; 1379 sc->sc_inc.inc_laddr = inc->inc_laddr; 1380 sc->sc_route.ro_rt = NULL; 1381 } 1382 sc->sc_irs = th->th_seq - 1; 1383 sc->sc_iss = th->th_ack - 1; 1384 wnd = sbspace(&so->so_rcv); 1385 wnd = imax(wnd, 0); 1386 wnd = imin(wnd, TCP_MAXWIN); 1387 sc->sc_wnd = wnd; 1388 sc->sc_flags = 0; 1389 sc->sc_rxtslot = 0; 1390 sc->sc_peer_mss = tcp_msstab[data]; 1391 return (sc); 1392 } 1393