tcp_syncache.c revision 130398
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 130398 2004-06-13 02:50:07Z rwatson $ 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 SOCK_LOCK(so); 563 mac_set_socket_peer_from_mbuf(m, so); 564 SOCK_UNLOCK(so); 565#endif 566 567 inp = sotoinpcb(so); 568 INP_LOCK(inp); 569 570 /* 571 * Insert new socket into hash list. 572 */ 573 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 574#ifdef INET6 575 if (sc->sc_inc.inc_isipv6) { 576 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 577 } else { 578 inp->inp_vflag &= ~INP_IPV6; 579 inp->inp_vflag |= INP_IPV4; 580#endif 581 inp->inp_laddr = sc->sc_inc.inc_laddr; 582#ifdef INET6 583 } 584#endif 585 inp->inp_lport = sc->sc_inc.inc_lport; 586 if (in_pcbinshash(inp) != 0) { 587 /* 588 * Undo the assignments above if we failed to 589 * put the PCB on the hash lists. 590 */ 591#ifdef INET6 592 if (sc->sc_inc.inc_isipv6) 593 inp->in6p_laddr = in6addr_any; 594 else 595#endif 596 inp->inp_laddr.s_addr = INADDR_ANY; 597 inp->inp_lport = 0; 598 goto abort; 599 } 600#ifdef IPSEC 601 /* copy old policy into new socket's */ 602 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 603 printf("syncache_expand: could not copy policy\n"); 604#endif 605#ifdef FAST_IPSEC 606 /* copy old policy into new socket's */ 607 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 608 printf("syncache_expand: could not copy policy\n"); 609#endif 610#ifdef INET6 611 if (sc->sc_inc.inc_isipv6) { 612 struct inpcb *oinp = sotoinpcb(lso); 613 struct in6_addr laddr6; 614 struct sockaddr_in6 sin6; 615 /* 616 * Inherit socket options from the listening socket. 617 * Note that in6p_inputopts are not (and should not be) 618 * copied, since it stores previously received options and is 619 * used to detect if each new option is different than the 620 * previous one and hence should be passed to a user. 621 * If we copied in6p_inputopts, a user would not be able to 622 * receive options just after calling the accept system call. 623 */ 624 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 625 if (oinp->in6p_outputopts) 626 inp->in6p_outputopts = 627 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 628 629 sin6.sin6_family = AF_INET6; 630 sin6.sin6_len = sizeof(sin6); 631 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 632 sin6.sin6_port = sc->sc_inc.inc_fport; 633 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 634 laddr6 = inp->in6p_laddr; 635 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 636 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 637 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 638 thread0.td_ucred)) { 639 inp->in6p_laddr = laddr6; 640 goto abort; 641 } 642 } else 643#endif 644 { 645 struct in_addr laddr; 646 struct sockaddr_in sin; 647 648 inp->inp_options = ip_srcroute(); 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 if (sc->sc_flags & SCF_CC) { 694 /* 695 * Initialization of the tcpcb for transaction; 696 * set SND.WND = SEG.WND, 697 * initialize CCsend and CCrecv. 698 */ 699 tp->t_flags |= TF_REQ_CC|TF_RCVD_CC; 700 tp->cc_send = sc->sc_cc_send; 701 tp->cc_recv = sc->sc_cc_recv; 702 } 703#ifdef TCP_SIGNATURE 704 if (sc->sc_flags & SCF_SIGNATURE) 705 tp->t_flags |= TF_SIGNATURE; 706#endif 707 708 /* 709 * Set up MSS and get cached values from tcp_hostcache. 710 * This might overwrite some of the defaults we just set. 711 */ 712 tcp_mss(tp, sc->sc_peer_mss); 713 714 /* 715 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 716 */ 717 if (sc->sc_rxtslot != 0) 718 tp->snd_cwnd = tp->t_maxseg; 719 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 720 721 INP_UNLOCK(inp); 722 723 tcpstat.tcps_accepts++; 724 return (so); 725 726abort: 727 INP_UNLOCK(inp); 728abort2: 729 if (so != NULL) 730 (void) soabort(so); 731 return (NULL); 732} 733 734/* 735 * This function gets called when we receive an ACK for a 736 * socket in the LISTEN state. We look up the connection 737 * in the syncache, and if its there, we pull it out of 738 * the cache and turn it into a full-blown connection in 739 * the SYN-RECEIVED state. 740 */ 741int 742syncache_expand(inc, th, sop, m) 743 struct in_conninfo *inc; 744 struct tcphdr *th; 745 struct socket **sop; 746 struct mbuf *m; 747{ 748 struct syncache *sc; 749 struct syncache_head *sch; 750 struct socket *so; 751 752 INP_INFO_WLOCK_ASSERT(&tcbinfo); 753 754 sc = syncache_lookup(inc, &sch); 755 if (sc == NULL) { 756 /* 757 * There is no syncache entry, so see if this ACK is 758 * a returning syncookie. To do this, first: 759 * A. See if this socket has had a syncache entry dropped in 760 * the past. We don't want to accept a bogus syncookie 761 * if we've never received a SYN. 762 * B. check that the syncookie is valid. If it is, then 763 * cobble up a fake syncache entry, and return. 764 */ 765 if (!tcp_syncookies) 766 return (0); 767 sc = syncookie_lookup(inc, th, *sop); 768 if (sc == NULL) 769 return (0); 770 sch = NULL; 771 tcpstat.tcps_sc_recvcookie++; 772 } 773 774 /* 775 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 776 */ 777 if (th->th_ack != sc->sc_iss + 1) 778 return (0); 779 780 so = syncache_socket(sc, *sop, m); 781 if (so == NULL) { 782#if 0 783resetandabort: 784 /* XXXjlemon check this - is this correct? */ 785 (void) tcp_respond(NULL, m, m, th, 786 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 787#endif 788 m_freem(m); /* XXX only needed for above */ 789 tcpstat.tcps_sc_aborted++; 790 } else 791 tcpstat.tcps_sc_completed++; 792 793 if (sch == NULL) 794 syncache_free(sc); 795 else 796 syncache_drop(sc, sch); 797 *sop = so; 798 return (1); 799} 800 801/* 802 * Given a LISTEN socket and an inbound SYN request, add 803 * this to the syn cache, and send back a segment: 804 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 805 * to the source. 806 * 807 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 808 * Doing so would require that we hold onto the data and deliver it 809 * to the application. However, if we are the target of a SYN-flood 810 * DoS attack, an attacker could send data which would eventually 811 * consume all available buffer space if it were ACKed. By not ACKing 812 * the data, we avoid this DoS scenario. 813 */ 814int 815syncache_add(inc, to, th, sop, m) 816 struct in_conninfo *inc; 817 struct tcpopt *to; 818 struct tcphdr *th; 819 struct socket **sop; 820 struct mbuf *m; 821{ 822 struct tcpcb *tp; 823 struct socket *so; 824 struct syncache *sc = NULL; 825 struct syncache_head *sch; 826 struct mbuf *ipopts = NULL; 827 struct rmxp_tao tao; 828 int i, win; 829 830 INP_INFO_WLOCK_ASSERT(&tcbinfo); 831 832 so = *sop; 833 tp = sototcpcb(so); 834 bzero(&tao, sizeof(tao)); 835 836 /* 837 * Remember the IP options, if any. 838 */ 839#ifdef INET6 840 if (!inc->inc_isipv6) 841#endif 842 ipopts = ip_srcroute(); 843 844 /* 845 * See if we already have an entry for this connection. 846 * If we do, resend the SYN,ACK, and reset the retransmit timer. 847 * 848 * XXX 849 * should the syncache be re-initialized with the contents 850 * of the new SYN here (which may have different options?) 851 */ 852 sc = syncache_lookup(inc, &sch); 853 if (sc != NULL) { 854 tcpstat.tcps_sc_dupsyn++; 855 if (ipopts) { 856 /* 857 * If we were remembering a previous source route, 858 * forget it and use the new one we've been given. 859 */ 860 if (sc->sc_ipopts) 861 (void) m_free(sc->sc_ipopts); 862 sc->sc_ipopts = ipopts; 863 } 864 /* 865 * Update timestamp if present. 866 */ 867 if (sc->sc_flags & SCF_TIMESTAMP) 868 sc->sc_tsrecent = to->to_tsval; 869 /* 870 * PCB may have changed, pick up new values. 871 */ 872 sc->sc_tp = tp; 873 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 874#ifdef TCPDEBUG 875 if (syncache_respond(sc, m, so) == 0) { 876#else 877 if (syncache_respond(sc, m) == 0) { 878#endif 879 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 880 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 881 sc, sc_timerq); 882 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 883 tcpstat.tcps_sndacks++; 884 tcpstat.tcps_sndtotal++; 885 } 886 *sop = NULL; 887 return (1); 888 } 889 890 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 891 if (sc == NULL) { 892 /* 893 * The zone allocator couldn't provide more entries. 894 * Treat this as if the cache was full; drop the oldest 895 * entry and insert the new one. 896 */ 897 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */ 898 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 899 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]); 900 if (sc != NULL) 901 break; 902 } 903 sc->sc_tp->ts_recent = ticks; 904 syncache_drop(sc, NULL); 905 tcpstat.tcps_sc_zonefail++; 906 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 907 if (sc == NULL) { 908 if (ipopts) 909 (void) m_free(ipopts); 910 return (0); 911 } 912 } 913 914 /* 915 * Fill in the syncache values. 916 */ 917 bzero(sc, sizeof(*sc)); 918 sc->sc_tp = tp; 919 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 920 sc->sc_ipopts = ipopts; 921 sc->sc_inc.inc_fport = inc->inc_fport; 922 sc->sc_inc.inc_lport = inc->inc_lport; 923#ifdef INET6 924 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 925 if (inc->inc_isipv6) { 926 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 927 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 928 } else 929#endif 930 { 931 sc->sc_inc.inc_faddr = inc->inc_faddr; 932 sc->sc_inc.inc_laddr = inc->inc_laddr; 933 } 934 sc->sc_irs = th->th_seq; 935 sc->sc_flags = 0; 936 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 937 if (tcp_syncookies) 938 sc->sc_iss = syncookie_generate(sc); 939 else 940 sc->sc_iss = arc4random(); 941 942 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 943 win = sbspace(&so->so_rcv); 944 win = imax(win, 0); 945 win = imin(win, TCP_MAXWIN); 946 sc->sc_wnd = win; 947 948 if (tcp_do_rfc1323) { 949 /* 950 * A timestamp received in a SYN makes 951 * it ok to send timestamp requests and replies. 952 */ 953 if (to->to_flags & TOF_TS) { 954 sc->sc_tsrecent = to->to_tsval; 955 sc->sc_flags |= SCF_TIMESTAMP; 956 } 957 if (to->to_flags & TOF_SCALE) { 958 int wscale = 0; 959 960 /* Compute proper scaling value from buffer space */ 961 while (wscale < TCP_MAX_WINSHIFT && 962 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 963 wscale++; 964 sc->sc_request_r_scale = wscale; 965 sc->sc_requested_s_scale = to->to_requested_s_scale; 966 sc->sc_flags |= SCF_WINSCALE; 967 } 968 } 969 if (tcp_do_rfc1644) { 970 /* 971 * A CC or CC.new option received in a SYN makes 972 * it ok to send CC in subsequent segments. 973 */ 974 if (to->to_flags & (TOF_CC|TOF_CCNEW)) { 975 sc->sc_cc_recv = to->to_cc; 976 sc->sc_cc_send = CC_INC(tcp_ccgen); 977 sc->sc_flags |= SCF_CC; 978 } 979 } 980 if (tp->t_flags & TF_NOOPT) 981 sc->sc_flags = SCF_NOOPT; 982#ifdef TCP_SIGNATURE 983 /* 984 * If listening socket requested TCP digests, and received SYN 985 * contains the option, flag this in the syncache so that 986 * syncache_respond() will do the right thing with the SYN+ACK. 987 * XXX Currently we always record the option by default and will 988 * attempt to use it in syncache_respond(). 989 */ 990 if (to->to_flags & TOF_SIGNATURE) 991 sc->sc_flags = SCF_SIGNATURE; 992#endif 993 994 /* 995 * XXX 996 * We have the option here of not doing TAO (even if the segment 997 * qualifies) and instead fall back to a normal 3WHS via the syncache. 998 * This allows us to apply synflood protection to TAO-qualifying SYNs 999 * also. However, there should be a hueristic to determine when to 1000 * do this, and is not present at the moment. 1001 */ 1002 1003 /* 1004 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1005 * - compare SEG.CC against cached CC from the same host, if any. 1006 * - if SEG.CC > chached value, SYN must be new and is accepted 1007 * immediately: save new CC in the cache, mark the socket 1008 * connected, enter ESTABLISHED state, turn on flag to 1009 * send a SYN in the next segment. 1010 * A virtual advertised window is set in rcv_adv to 1011 * initialize SWS prevention. Then enter normal segment 1012 * processing: drop SYN, process data and FIN. 1013 * - otherwise do a normal 3-way handshake. 1014 */ 1015 if (tcp_do_rfc1644) 1016 tcp_hc_gettao(&sc->sc_inc, &tao); 1017 1018 if ((to->to_flags & TOF_CC) != 0) { 1019 if (((tp->t_flags & TF_NOPUSH) != 0) && 1020 sc->sc_flags & SCF_CC && tao.tao_cc != 0 && 1021 CC_GT(to->to_cc, tao.tao_cc)) { 1022 sc->sc_rxtslot = 0; 1023 so = syncache_socket(sc, *sop, m); 1024 if (so != NULL) { 1025 tao.tao_cc = to->to_cc; 1026 tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC, 1027 tao.tao_cc, 0); 1028 *sop = so; 1029 } 1030 syncache_free(sc); 1031 return (so != NULL); 1032 } 1033 } else { 1034 /* 1035 * No CC option, but maybe CC.NEW: invalidate cached value. 1036 */ 1037 if (tcp_do_rfc1644) { 1038 tao.tao_cc = 0; 1039 tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC, 1040 tao.tao_cc, 0); 1041 } 1042 } 1043 1044 /* 1045 * TAO test failed or there was no CC option, 1046 * do a standard 3-way handshake. 1047 */ 1048#ifdef TCPDEBUG 1049 if (syncache_respond(sc, m, so) == 0) { 1050#else 1051 if (syncache_respond(sc, m) == 0) { 1052#endif 1053 syncache_insert(sc, sch); 1054 tcpstat.tcps_sndacks++; 1055 tcpstat.tcps_sndtotal++; 1056 } else { 1057 syncache_free(sc); 1058 tcpstat.tcps_sc_dropped++; 1059 } 1060 *sop = NULL; 1061 return (1); 1062} 1063 1064#ifdef TCPDEBUG 1065static int 1066syncache_respond(sc, m, so) 1067 struct syncache *sc; 1068 struct mbuf *m; 1069 struct socket *so; 1070#else 1071static int 1072syncache_respond(sc, m) 1073 struct syncache *sc; 1074 struct mbuf *m; 1075#endif 1076{ 1077 u_int8_t *optp; 1078 int optlen, error; 1079 u_int16_t tlen, hlen, mssopt; 1080 struct ip *ip = NULL; 1081 struct tcphdr *th; 1082 struct inpcb *inp; 1083#ifdef INET6 1084 struct ip6_hdr *ip6 = NULL; 1085#endif 1086 1087 hlen = 1088#ifdef INET6 1089 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1090#endif 1091 sizeof(struct ip); 1092 1093 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1094 1095 /* Determine MSS we advertize to other end of connection */ 1096 mssopt = tcp_mssopt(&sc->sc_inc); 1097 1098 /* Compute the size of the TCP options. */ 1099 if (sc->sc_flags & SCF_NOOPT) { 1100 optlen = 0; 1101 } else { 1102 optlen = TCPOLEN_MAXSEG + 1103 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1104 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1105 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0); 1106#ifdef TCP_SIGNATURE 1107 optlen += (sc->sc_flags & SCF_SIGNATURE) ? 1108 TCPOLEN_SIGNATURE + 2 : 0; 1109#endif 1110 } 1111 tlen = hlen + sizeof(struct tcphdr) + optlen; 1112 1113 /* 1114 * XXX 1115 * assume that the entire packet will fit in a header mbuf 1116 */ 1117 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1118 1119 /* 1120 * XXX shouldn't this reuse the mbuf if possible ? 1121 * Create the IP+TCP header from scratch. 1122 */ 1123 if (m) 1124 m_freem(m); 1125 1126 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1127 if (m == NULL) 1128 return (ENOBUFS); 1129 m->m_data += max_linkhdr; 1130 m->m_len = tlen; 1131 m->m_pkthdr.len = tlen; 1132 m->m_pkthdr.rcvif = NULL; 1133 inp = sc->sc_tp->t_inpcb; 1134 INP_LOCK(inp); 1135#ifdef MAC 1136 mac_create_mbuf_from_inpcb(inp, m); 1137#endif 1138 1139#ifdef INET6 1140 if (sc->sc_inc.inc_isipv6) { 1141 ip6 = mtod(m, struct ip6_hdr *); 1142 ip6->ip6_vfc = IPV6_VERSION; 1143 ip6->ip6_nxt = IPPROTO_TCP; 1144 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1145 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1146 ip6->ip6_plen = htons(tlen - hlen); 1147 /* ip6_hlim is set after checksum */ 1148 /* ip6_flow = ??? */ 1149 1150 th = (struct tcphdr *)(ip6 + 1); 1151 } else 1152#endif 1153 { 1154 ip = mtod(m, struct ip *); 1155 ip->ip_v = IPVERSION; 1156 ip->ip_hl = sizeof(struct ip) >> 2; 1157 ip->ip_len = tlen; 1158 ip->ip_id = 0; 1159 ip->ip_off = 0; 1160 ip->ip_sum = 0; 1161 ip->ip_p = IPPROTO_TCP; 1162 ip->ip_src = sc->sc_inc.inc_laddr; 1163 ip->ip_dst = sc->sc_inc.inc_faddr; 1164 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1165 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 1166 1167 /* 1168 * See if we should do MTU discovery. Route lookups are 1169 * expensive, so we will only unset the DF bit if: 1170 * 1171 * 1) path_mtu_discovery is disabled 1172 * 2) the SCF_UNREACH flag has been set 1173 */ 1174 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1175 ip->ip_off |= IP_DF; 1176 1177 th = (struct tcphdr *)(ip + 1); 1178 } 1179 th->th_sport = sc->sc_inc.inc_lport; 1180 th->th_dport = sc->sc_inc.inc_fport; 1181 1182 th->th_seq = htonl(sc->sc_iss); 1183 th->th_ack = htonl(sc->sc_irs + 1); 1184 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1185 th->th_x2 = 0; 1186 th->th_flags = TH_SYN|TH_ACK; 1187 th->th_win = htons(sc->sc_wnd); 1188 th->th_urp = 0; 1189 1190 /* Tack on the TCP options. */ 1191 if (optlen != 0) { 1192 optp = (u_int8_t *)(th + 1); 1193 *optp++ = TCPOPT_MAXSEG; 1194 *optp++ = TCPOLEN_MAXSEG; 1195 *optp++ = (mssopt >> 8) & 0xff; 1196 *optp++ = mssopt & 0xff; 1197 1198 if (sc->sc_flags & SCF_WINSCALE) { 1199 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1200 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1201 sc->sc_request_r_scale); 1202 optp += 4; 1203 } 1204 1205 if (sc->sc_flags & SCF_TIMESTAMP) { 1206 u_int32_t *lp = (u_int32_t *)(optp); 1207 1208 /* Form timestamp option per appendix A of RFC 1323. */ 1209 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1210 *lp++ = htonl(ticks); 1211 *lp = htonl(sc->sc_tsrecent); 1212 optp += TCPOLEN_TSTAMP_APPA; 1213 } 1214 1215 /* 1216 * Send CC and CC.echo if we received CC from our peer. 1217 */ 1218 if (sc->sc_flags & SCF_CC) { 1219 u_int32_t *lp = (u_int32_t *)(optp); 1220 1221 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1222 *lp++ = htonl(sc->sc_cc_send); 1223 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1224 *lp = htonl(sc->sc_cc_recv); 1225 optp += TCPOLEN_CC_APPA * 2; 1226 } 1227 1228#ifdef TCP_SIGNATURE 1229 /* 1230 * Handle TCP-MD5 passive opener response. 1231 */ 1232 if (sc->sc_flags & SCF_SIGNATURE) { 1233 u_int8_t *bp = optp; 1234 int i; 1235 1236 *bp++ = TCPOPT_SIGNATURE; 1237 *bp++ = TCPOLEN_SIGNATURE; 1238 for (i = 0; i < TCP_SIGLEN; i++) 1239 *bp++ = 0; 1240 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1241 optp + 2, IPSEC_DIR_OUTBOUND); 1242 *bp++ = TCPOPT_NOP; 1243 *bp++ = TCPOPT_EOL; 1244 optp += TCPOLEN_SIGNATURE + 2; 1245 } 1246#endif /* TCP_SIGNATURE */ 1247 } 1248 1249#ifdef INET6 1250 if (sc->sc_inc.inc_isipv6) { 1251 th->th_sum = 0; 1252 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1253 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1254 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1255 } else 1256#endif 1257 { 1258 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1259 htons(tlen - hlen + IPPROTO_TCP)); 1260 m->m_pkthdr.csum_flags = CSUM_TCP; 1261 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1262#ifdef TCPDEBUG 1263 /* 1264 * Trace. 1265 */ 1266 if (so != NULL && so->so_options & SO_DEBUG) { 1267 struct tcpcb *tp = sototcpcb(so); 1268 tcp_trace(TA_OUTPUT, tp->t_state, tp, 1269 mtod(m, void *), th, 0); 1270 } 1271#endif 1272 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp); 1273 } 1274 INP_UNLOCK(inp); 1275 return (error); 1276} 1277 1278/* 1279 * cookie layers: 1280 * 1281 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1282 * | peer iss | 1283 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1284 * | 0 |(A)| | 1285 * (A): peer mss index 1286 */ 1287 1288/* 1289 * The values below are chosen to minimize the size of the tcp_secret 1290 * table, as well as providing roughly a 16 second lifetime for the cookie. 1291 */ 1292 1293#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1294#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1295 1296#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1297#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1298#define SYNCOOKIE_TIMEOUT \ 1299 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1300#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1301 1302static struct { 1303 u_int32_t ts_secbits[4]; 1304 u_int ts_expire; 1305} tcp_secret[SYNCOOKIE_NSECRETS]; 1306 1307static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1308 1309static MD5_CTX syn_ctx; 1310 1311#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1312 1313struct md5_add { 1314 u_int32_t laddr, faddr; 1315 u_int32_t secbits[4]; 1316 u_int16_t lport, fport; 1317}; 1318 1319#ifdef CTASSERT 1320CTASSERT(sizeof(struct md5_add) == 28); 1321#endif 1322 1323/* 1324 * Consider the problem of a recreated (and retransmitted) cookie. If the 1325 * original SYN was accepted, the connection is established. The second 1326 * SYN is inflight, and if it arrives with an ISN that falls within the 1327 * receive window, the connection is killed. 1328 * 1329 * However, since cookies have other problems, this may not be worth 1330 * worrying about. 1331 */ 1332 1333static u_int32_t 1334syncookie_generate(struct syncache *sc) 1335{ 1336 u_int32_t md5_buffer[4]; 1337 u_int32_t data; 1338 int idx, i; 1339 struct md5_add add; 1340 1341 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1342 1343 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1344 if (tcp_secret[idx].ts_expire < ticks) { 1345 for (i = 0; i < 4; i++) 1346 tcp_secret[idx].ts_secbits[i] = arc4random(); 1347 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1348 } 1349 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1350 if (tcp_msstab[data] <= sc->sc_peer_mss) 1351 break; 1352 data = (data << SYNCOOKIE_WNDBITS) | idx; 1353 data ^= sc->sc_irs; /* peer's iss */ 1354 MD5Init(&syn_ctx); 1355#ifdef INET6 1356 if (sc->sc_inc.inc_isipv6) { 1357 MD5Add(sc->sc_inc.inc6_laddr); 1358 MD5Add(sc->sc_inc.inc6_faddr); 1359 add.laddr = 0; 1360 add.faddr = 0; 1361 } else 1362#endif 1363 { 1364 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1365 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1366 } 1367 add.lport = sc->sc_inc.inc_lport; 1368 add.fport = sc->sc_inc.inc_fport; 1369 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1370 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1371 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1372 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1373 MD5Add(add); 1374 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1375 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1376 return (data); 1377} 1378 1379static struct syncache * 1380syncookie_lookup(inc, th, so) 1381 struct in_conninfo *inc; 1382 struct tcphdr *th; 1383 struct socket *so; 1384{ 1385 u_int32_t md5_buffer[4]; 1386 struct syncache *sc; 1387 u_int32_t data; 1388 int wnd, idx; 1389 struct md5_add add; 1390 1391 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */ 1392 1393 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1394 idx = data & SYNCOOKIE_WNDMASK; 1395 if (tcp_secret[idx].ts_expire < ticks || 1396 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1397 return (NULL); 1398 MD5Init(&syn_ctx); 1399#ifdef INET6 1400 if (inc->inc_isipv6) { 1401 MD5Add(inc->inc6_laddr); 1402 MD5Add(inc->inc6_faddr); 1403 add.laddr = 0; 1404 add.faddr = 0; 1405 } else 1406#endif 1407 { 1408 add.laddr = inc->inc_laddr.s_addr; 1409 add.faddr = inc->inc_faddr.s_addr; 1410 } 1411 add.lport = inc->inc_lport; 1412 add.fport = inc->inc_fport; 1413 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1414 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1415 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1416 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1417 MD5Add(add); 1418 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1419 data ^= md5_buffer[0]; 1420 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1421 return (NULL); 1422 data = data >> SYNCOOKIE_WNDBITS; 1423 1424 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT); 1425 if (sc == NULL) 1426 return (NULL); 1427 /* 1428 * Fill in the syncache values. 1429 * XXX duplicate code from syncache_add 1430 */ 1431 sc->sc_ipopts = NULL; 1432 sc->sc_inc.inc_fport = inc->inc_fport; 1433 sc->sc_inc.inc_lport = inc->inc_lport; 1434#ifdef INET6 1435 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1436 if (inc->inc_isipv6) { 1437 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1438 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1439 } else 1440#endif 1441 { 1442 sc->sc_inc.inc_faddr = inc->inc_faddr; 1443 sc->sc_inc.inc_laddr = inc->inc_laddr; 1444 } 1445 sc->sc_irs = th->th_seq - 1; 1446 sc->sc_iss = th->th_ack - 1; 1447 wnd = sbspace(&so->so_rcv); 1448 wnd = imax(wnd, 0); 1449 wnd = imin(wnd, TCP_MAXWIN); 1450 sc->sc_wnd = wnd; 1451 sc->sc_flags = 0; 1452 sc->sc_rxtslot = 0; 1453 sc->sc_peer_mss = tcp_msstab[data]; 1454 return (sc); 1455} 1456