tcp_syncache.c revision 159922
1/*- 2 * Copyright (c) 2001 McAfee, Inc. 3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Jonathan Lemon 7 * and McAfee Research, the Security Research Division of McAfee, Inc. under 8 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 9 * DARPA CHATS research program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: head/sys/netinet/tcp_syncache.c 159922 2006-06-25 11:11:33Z andre $ 33 */ 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_ipsec.h" 38#include "opt_mac.h" 39#include "opt_tcpdebug.h" 40#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/kernel.h> 45#include <sys/sysctl.h> 46#include <sys/lock.h> 47#include <sys/mutex.h> 48#include <sys/malloc.h> 49#include <sys/mac.h> 50#include <sys/mbuf.h> 51#include <sys/md5.h> 52#include <sys/proc.h> /* for proc0 declaration */ 53#include <sys/random.h> 54#include <sys/rwlock.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57 58#include <net/if.h> 59#include <net/route.h> 60 61#include <netinet/in.h> 62#include <netinet/in_systm.h> 63#include <netinet/ip.h> 64#include <netinet/in_var.h> 65#include <netinet/in_pcb.h> 66#include <netinet/ip_var.h> 67#include <netinet/ip_options.h> 68#ifdef INET6 69#include <netinet/ip6.h> 70#include <netinet/icmp6.h> 71#include <netinet6/nd6.h> 72#include <netinet6/ip6_var.h> 73#include <netinet6/in6_pcb.h> 74#endif 75#include <netinet/tcp.h> 76#ifdef TCPDEBUG 77#include <netinet/tcpip.h> 78#endif 79#include <netinet/tcp_fsm.h> 80#include <netinet/tcp_seq.h> 81#include <netinet/tcp_timer.h> 82#include <netinet/tcp_var.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif 86#ifdef INET6 87#include <netinet6/tcp6_var.h> 88#endif 89 90#ifdef IPSEC 91#include <netinet6/ipsec.h> 92#ifdef INET6 93#include <netinet6/ipsec6.h> 94#endif 95#endif /*IPSEC*/ 96 97#ifdef FAST_IPSEC 98#include <netipsec/ipsec.h> 99#ifdef INET6 100#include <netipsec/ipsec6.h> 101#endif 102#include <netipsec/key.h> 103#endif /*FAST_IPSEC*/ 104 105#include <machine/in_cksum.h> 106#include <vm/uma.h> 107 108static int tcp_syncookies = 1; 109SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 110 &tcp_syncookies, 0, 111 "Use TCP SYN cookies if the syncache overflows"); 112 113struct syncache { 114 TAILQ_ENTRY(syncache) sc_hash; 115 struct in_conninfo sc_inc; /* addresses */ 116 u_long sc_rxttime; /* retransmit time */ 117 u_int16_t sc_rxmits; /* retransmit counter */ 118 119 u_int32_t sc_tsrecent; 120 u_int32_t sc_flowlabel; /* IPv6 flowlabel */ 121 tcp_seq sc_irs; /* seq from peer */ 122 tcp_seq sc_iss; /* our ISS */ 123 struct mbuf *sc_ipopts; /* source route */ 124 125 u_int16_t sc_peer_mss; /* peer's MSS */ 126 u_int16_t sc_wnd; /* advertised window */ 127 u_int8_t sc_ip_ttl; /* IPv4 TTL */ 128 u_int8_t sc_ip_tos; /* IPv4 TOS */ 129 u_int8_t sc_requested_s_scale:4, 130 sc_request_r_scale:4; 131 u_int8_t sc_flags; 132#define SCF_NOOPT 0x01 /* no TCP options */ 133#define SCF_WINSCALE 0x02 /* negotiated window scaling */ 134#define SCF_TIMESTAMP 0x04 /* negotiated timestamps */ 135#define SCF_UNREACH 0x10 /* icmp unreachable received */ 136#define SCF_SIGNATURE 0x20 /* send MD5 digests */ 137#define SCF_SACK 0x80 /* send SACK option */ 138}; 139 140struct syncache_head { 141 struct mtx sch_mtx; 142 TAILQ_HEAD(sch_head, syncache) sch_bucket; 143 struct callout sch_timer; 144 int sch_nextc; 145 u_int sch_length; 146}; 147 148static void syncache_drop(struct syncache *, struct syncache_head *); 149static void syncache_free(struct syncache *); 150static void syncache_insert(struct syncache *, struct syncache_head *); 151struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 152static int syncache_respond(struct syncache *, struct mbuf *); 153static struct socket *syncache_socket(struct syncache *, struct socket *, 154 struct mbuf *m); 155static void syncache_timer(void *); 156static void syncookie_init(void); 157static u_int32_t syncookie_generate(struct syncache *, u_int32_t *); 158static struct syncache 159 *syncookie_lookup(struct in_conninfo *, struct tcphdr *, 160 struct socket *); 161 162/* 163 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 164 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 165 * the odds are that the user has given up attempting to connect by then. 166 */ 167#define SYNCACHE_MAXREXMTS 3 168 169/* Arbitrary values */ 170#define TCP_SYNCACHE_HASHSIZE 512 171#define TCP_SYNCACHE_BUCKETLIMIT 30 172 173struct tcp_syncache { 174 struct syncache_head *hashbase; 175 uma_zone_t zone; 176 u_int hashsize; 177 u_int hashmask; 178 u_int bucket_limit; 179 u_int cache_count; /* XXX: unprotected */ 180 u_int cache_limit; 181 u_int rexmt_limit; 182 u_int hash_secret; 183}; 184static struct tcp_syncache tcp_syncache; 185 186SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 187 188SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 189 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 190 191SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 192 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 193 194SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 195 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 196 197SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 198 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 199 200SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 201 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 202 203static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 204 205#define SYNCACHE_HASH(inc, mask) \ 206 ((tcp_syncache.hash_secret ^ \ 207 (inc)->inc_faddr.s_addr ^ \ 208 ((inc)->inc_faddr.s_addr >> 16) ^ \ 209 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 210 211#define SYNCACHE_HASH6(inc, mask) \ 212 ((tcp_syncache.hash_secret ^ \ 213 (inc)->inc6_faddr.s6_addr32[0] ^ \ 214 (inc)->inc6_faddr.s6_addr32[3] ^ \ 215 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 216 217#define ENDPTS_EQ(a, b) ( \ 218 (a)->ie_fport == (b)->ie_fport && \ 219 (a)->ie_lport == (b)->ie_lport && \ 220 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 221 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 222) 223 224#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 225 226#define SYNCACHE_TIMEOUT(sc, sch, co) do { \ 227 (sc)->sc_rxmits++; \ 228 (sc)->sc_rxttime = ticks + \ 229 TCPTV_RTOBASE * tcp_backoff[(sc)->sc_rxmits - 1]; \ 230 if ((sch)->sch_nextc > (sc)->sc_rxttime) \ 231 (sch)->sch_nextc = (sc)->sc_rxttime; \ 232 if (!TAILQ_EMPTY(&(sch)->sch_bucket) && !(co)) \ 233 callout_reset(&(sch)->sch_timer, \ 234 (sch)->sch_nextc - ticks, \ 235 syncache_timer, (void *)(sch)); \ 236} while (0) 237 238#define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 239#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 240#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 241 242/* 243 * Requires the syncache entry to be already removed from the bucket list. 244 */ 245static void 246syncache_free(struct syncache *sc) 247{ 248 if (sc->sc_ipopts) 249 (void) m_free(sc->sc_ipopts); 250 251 uma_zfree(tcp_syncache.zone, sc); 252} 253 254void 255syncache_init(void) 256{ 257 int i; 258 259 tcp_syncache.cache_count = 0; 260 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 261 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 262 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 263 tcp_syncache.hash_secret = arc4random(); 264 265 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 266 &tcp_syncache.hashsize); 267 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 268 &tcp_syncache.bucket_limit); 269 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) { 270 printf("WARNING: syncache hash size is not a power of 2.\n"); 271 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 272 } 273 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 274 275 /* Set limits. */ 276 tcp_syncache.cache_limit = 277 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 278 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 279 &tcp_syncache.cache_limit); 280 281 /* Allocate the hash table. */ 282 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 283 tcp_syncache.hashsize * sizeof(struct syncache_head), 284 M_SYNCACHE, M_WAITOK | M_ZERO); 285 286 /* Initialize the hash buckets. */ 287 for (i = 0; i < tcp_syncache.hashsize; i++) { 288 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 289 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 290 NULL, MTX_DEF); 291 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer, 292 &tcp_syncache.hashbase[i].sch_mtx, 0); 293 tcp_syncache.hashbase[i].sch_length = 0; 294 } 295 296 syncookie_init(); 297 298 /* Create the syncache entry zone. */ 299 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 300 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 301 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit); 302} 303 304/* 305 * Inserts a syncache entry into the specified bucket row. 306 * Locks and unlocks the syncache_head autonomously. 307 */ 308static void 309syncache_insert(struct syncache *sc, struct syncache_head *sch) 310{ 311 struct syncache *sc2; 312 313 SCH_LOCK(sch); 314 315 /* 316 * Make sure that we don't overflow the per-bucket limit. 317 * If the bucket is full, toss the oldest element. 318 */ 319 if (sch->sch_length >= tcp_syncache.bucket_limit) { 320 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 321 ("sch->sch_length incorrect")); 322 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 323 syncache_drop(sc2, sch); 324 tcpstat.tcps_sc_bucketoverflow++; 325 } 326 327 /* Put it into the bucket. */ 328 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 329 sch->sch_length++; 330 331 /* Reinitialize the bucket row's timer. */ 332 SYNCACHE_TIMEOUT(sc, sch, 1); 333 334 SCH_UNLOCK(sch); 335 336 tcp_syncache.cache_count++; 337 tcpstat.tcps_sc_added++; 338} 339 340/* 341 * Remove and free entry from syncache bucket row. 342 * Expects locked syncache head. 343 */ 344static void 345syncache_drop(struct syncache *sc, struct syncache_head *sch) 346{ 347 348 SCH_LOCK_ASSERT(sch); 349 350 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 351 sch->sch_length--; 352 353 syncache_free(sc); 354 tcp_syncache.cache_count--; 355} 356 357/* 358 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 359 * If we have retransmitted an entry the maximum number of times, expire it. 360 * One separate timer for each bucket row. 361 */ 362static void 363syncache_timer(void *xsch) 364{ 365 struct syncache_head *sch = (struct syncache_head *)xsch; 366 struct syncache *sc, *nsc; 367 int tick = ticks; 368 369 /* NB: syncache_head has already been locked by the callout. */ 370 SCH_LOCK_ASSERT(sch); 371 372 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 373 /* 374 * We do not check if the listen socket still exists 375 * and accept the case where the listen socket may be 376 * gone by the time we resend the SYN/ACK. We do 377 * not expect this to happens often. If it does, 378 * then the RST will be sent by the time the remote 379 * host does the SYN/ACK->ACK. 380 */ 381 if (sc->sc_rxttime >= tick) { 382 if (sc->sc_rxttime < sch->sch_nextc) 383 sch->sch_nextc = sc->sc_rxttime; 384 continue; 385 } 386 387 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) { 388 syncache_drop(sc, sch); 389 tcpstat.tcps_sc_stale++; 390 continue; 391 } 392 393 (void) syncache_respond(sc, NULL); 394 tcpstat.tcps_sc_retransmitted++; 395 SYNCACHE_TIMEOUT(sc, sch, 0); 396 } 397 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 398 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 399 syncache_timer, (void *)(sch)); 400} 401 402/* 403 * Find an entry in the syncache. 404 * Returns always with locked syncache_head plus a matching entry or NULL. 405 */ 406struct syncache * 407syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 408{ 409 struct syncache *sc; 410 struct syncache_head *sch; 411 412#ifdef INET6 413 if (inc->inc_isipv6) { 414 sch = &tcp_syncache.hashbase[ 415 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 416 *schp = sch; 417 418 SCH_LOCK(sch); 419 420 /* Circle through bucket row to find matching entry. */ 421 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 422 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 423 return (sc); 424 } 425 } else 426#endif 427 { 428 sch = &tcp_syncache.hashbase[ 429 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 430 *schp = sch; 431 432 SCH_LOCK(sch); 433 434 /* Circle through bucket row to find matching entry. */ 435 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 436#ifdef INET6 437 if (sc->sc_inc.inc_isipv6) 438 continue; 439#endif 440 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 441 return (sc); 442 } 443 } 444 SCH_LOCK_ASSERT(*schp); 445 return (NULL); /* always returns with locked sch */ 446} 447 448/* 449 * This function is called when we get a RST for a 450 * non-existent connection, so that we can see if the 451 * connection is in the syn cache. If it is, zap it. 452 */ 453void 454syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 455{ 456 struct syncache *sc; 457 struct syncache_head *sch; 458 459 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 460 SCH_LOCK_ASSERT(sch); 461 if (sc == NULL) 462 goto done; 463 464 /* 465 * If the RST bit is set, check the sequence number to see 466 * if this is a valid reset segment. 467 * RFC 793 page 37: 468 * In all states except SYN-SENT, all reset (RST) segments 469 * are validated by checking their SEQ-fields. A reset is 470 * valid if its sequence number is in the window. 471 * 472 * The sequence number in the reset segment is normally an 473 * echo of our outgoing acknowlegement numbers, but some hosts 474 * send a reset with the sequence number at the rightmost edge 475 * of our receive window, and we have to handle this case. 476 */ 477 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 478 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 479 syncache_drop(sc, sch); 480 tcpstat.tcps_sc_reset++; 481 } 482done: 483 SCH_UNLOCK(sch); 484} 485 486void 487syncache_badack(struct in_conninfo *inc) 488{ 489 struct syncache *sc; 490 struct syncache_head *sch; 491 492 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 493 SCH_LOCK_ASSERT(sch); 494 if (sc != NULL) { 495 syncache_drop(sc, sch); 496 tcpstat.tcps_sc_badack++; 497 } 498 SCH_UNLOCK(sch); 499} 500 501void 502syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 503{ 504 struct syncache *sc; 505 struct syncache_head *sch; 506 507 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 508 SCH_LOCK_ASSERT(sch); 509 if (sc == NULL) 510 goto done; 511 512 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 513 if (ntohl(th->th_seq) != sc->sc_iss) 514 goto done; 515 516 /* 517 * If we've rertransmitted 3 times and this is our second error, 518 * we remove the entry. Otherwise, we allow it to continue on. 519 * This prevents us from incorrectly nuking an entry during a 520 * spurious network outage. 521 * 522 * See tcp_notify(). 523 */ 524 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 525 sc->sc_flags |= SCF_UNREACH; 526 goto done; 527 } 528 syncache_drop(sc, sch); 529 tcpstat.tcps_sc_unreach++; 530done: 531 SCH_UNLOCK(sch); 532} 533 534/* 535 * Build a new TCP socket structure from a syncache entry. 536 */ 537static struct socket * 538syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 539{ 540 struct inpcb *inp = NULL; 541 struct socket *so; 542 struct tcpcb *tp; 543 544 NET_ASSERT_GIANT(); 545 INP_INFO_WLOCK_ASSERT(&tcbinfo); 546 547 /* 548 * Ok, create the full blown connection, and set things up 549 * as they would have been set up if we had created the 550 * connection when the SYN arrived. If we can't create 551 * the connection, abort it. 552 */ 553 so = sonewconn(lso, SS_ISCONNECTED); 554 if (so == NULL) { 555 /* 556 * Drop the connection; we will send a RST if the peer 557 * retransmits the ACK, 558 */ 559 tcpstat.tcps_listendrop++; 560 goto abort2; 561 } 562#ifdef MAC 563 SOCK_LOCK(so); 564 mac_set_socket_peer_from_mbuf(m, so); 565 SOCK_UNLOCK(so); 566#endif 567 568 inp = sotoinpcb(so); 569 INP_LOCK(inp); 570 571 /* Insert new socket into PCB hash list. */ 572 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 573#ifdef INET6 574 if (sc->sc_inc.inc_isipv6) { 575 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 576 } else { 577 inp->inp_vflag &= ~INP_IPV6; 578 inp->inp_vflag |= INP_IPV4; 579#endif 580 inp->inp_laddr = sc->sc_inc.inc_laddr; 581#ifdef INET6 582 } 583#endif 584 inp->inp_lport = sc->sc_inc.inc_lport; 585 if (in_pcbinshash(inp) != 0) { 586 /* 587 * Undo the assignments above if we failed to 588 * put the PCB on the hash lists. 589 */ 590#ifdef INET6 591 if (sc->sc_inc.inc_isipv6) 592 inp->in6p_laddr = in6addr_any; 593 else 594#endif 595 inp->inp_laddr.s_addr = INADDR_ANY; 596 inp->inp_lport = 0; 597 goto abort; 598 } 599#ifdef IPSEC 600 /* Copy old policy into new socket's. */ 601 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 602 printf("syncache_expand: could not copy policy\n"); 603#endif 604#ifdef FAST_IPSEC 605 /* Copy old policy into new socket's. */ 606 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 607 printf("syncache_expand: could not copy policy\n"); 608#endif 609#ifdef INET6 610 if (sc->sc_inc.inc_isipv6) { 611 struct inpcb *oinp = sotoinpcb(lso); 612 struct in6_addr laddr6; 613 struct sockaddr_in6 sin6; 614 /* 615 * Inherit socket options from the listening socket. 616 * Note that in6p_inputopts are not (and should not be) 617 * copied, since it stores previously received options and is 618 * used to detect if each new option is different than the 619 * previous one and hence should be passed to a user. 620 * If we copied in6p_inputopts, a user would not be able to 621 * receive options just after calling the accept system call. 622 */ 623 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 624 if (oinp->in6p_outputopts) 625 inp->in6p_outputopts = 626 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 627 628 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 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 633 laddr6 = inp->in6p_laddr; 634 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 635 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 636 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 637 thread0.td_ucred)) { 638 inp->in6p_laddr = laddr6; 639 goto abort; 640 } 641 /* Override flowlabel from in6_pcbconnect. */ 642 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 643 inp->in6p_flowinfo |= sc->sc_flowlabel; 644 } else 645#endif 646 { 647 struct in_addr laddr; 648 struct sockaddr_in sin; 649 650 inp->inp_options = ip_srcroute(m); 651 if (inp->inp_options == NULL) { 652 inp->inp_options = sc->sc_ipopts; 653 sc->sc_ipopts = NULL; 654 } 655 656 sin.sin_family = AF_INET; 657 sin.sin_len = sizeof(sin); 658 sin.sin_addr = sc->sc_inc.inc_faddr; 659 sin.sin_port = sc->sc_inc.inc_fport; 660 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 661 laddr = inp->inp_laddr; 662 if (inp->inp_laddr.s_addr == INADDR_ANY) 663 inp->inp_laddr = sc->sc_inc.inc_laddr; 664 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 665 thread0.td_ucred)) { 666 inp->inp_laddr = laddr; 667 goto abort; 668 } 669 } 670 tp = intotcpcb(inp); 671 tp->t_state = TCPS_SYN_RECEIVED; 672 tp->iss = sc->sc_iss; 673 tp->irs = sc->sc_irs; 674 tcp_rcvseqinit(tp); 675 tcp_sendseqinit(tp); 676 tp->snd_wl1 = sc->sc_irs; 677 tp->rcv_up = sc->sc_irs + 1; 678 tp->rcv_wnd = sc->sc_wnd; 679 tp->rcv_adv += tp->rcv_wnd; 680 681 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 682 if (sc->sc_flags & SCF_NOOPT) 683 tp->t_flags |= TF_NOOPT; 684 if (sc->sc_flags & SCF_WINSCALE) { 685 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 686 tp->snd_scale = sc->sc_requested_s_scale; 687 tp->request_r_scale = sc->sc_request_r_scale; 688 } 689 if (sc->sc_flags & SCF_TIMESTAMP) { 690 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 691 tp->ts_recent = sc->sc_tsrecent; 692 tp->ts_recent_age = ticks; 693 } 694#ifdef TCP_SIGNATURE 695 if (sc->sc_flags & SCF_SIGNATURE) 696 tp->t_flags |= TF_SIGNATURE; 697#endif 698 if (sc->sc_flags & SCF_SACK) { 699 tp->sack_enable = 1; 700 tp->t_flags |= TF_SACK_PERMIT; 701 } 702 703 /* 704 * Set up MSS and get cached values from tcp_hostcache. 705 * This might overwrite some of the defaults we just set. 706 */ 707 tcp_mss(tp, sc->sc_peer_mss); 708 709 /* 710 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 711 */ 712 if (sc->sc_rxmits > 1) 713 tp->snd_cwnd = tp->t_maxseg; 714 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 715 716 INP_UNLOCK(inp); 717 718 tcpstat.tcps_accepts++; 719 return (so); 720 721abort: 722 INP_UNLOCK(inp); 723abort2: 724 if (so != NULL) 725 soabort(so); 726 return (NULL); 727} 728 729/* 730 * This function gets called when we receive an ACK for a 731 * socket in the LISTEN state. We look up the connection 732 * in the syncache, and if its there, we pull it out of 733 * the cache and turn it into a full-blown connection in 734 * the SYN-RECEIVED state. 735 */ 736int 737syncache_expand(struct in_conninfo *inc, struct tcphdr *th, 738 struct socket **lsop, struct mbuf *m) 739{ 740 struct syncache *sc; 741 struct syncache_head *sch; 742 struct socket *so; 743 744 /* 745 * Global TCP locks are held because we manipulate the PCB lists 746 * and create a new socket. 747 */ 748 INP_INFO_WLOCK_ASSERT(&tcbinfo); 749 750 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 751 SCH_LOCK_ASSERT(sch); 752 if (sc == NULL) { 753 /* 754 * There is no syncache entry, so see if this ACK is 755 * a returning syncookie. To do this, first: 756 * A. See if this socket has had a syncache entry dropped in 757 * the past. We don't want to accept a bogus syncookie 758 * if we've never received a SYN. 759 * B. check that the syncookie is valid. If it is, then 760 * cobble up a fake syncache entry, and return. 761 */ 762 SCH_UNLOCK(sch); 763 sch = NULL; 764 765 if (!tcp_syncookies) 766 goto failed; 767 sc = syncookie_lookup(inc, th, *lsop); 768 if (sc == NULL) 769 goto failed; 770 tcpstat.tcps_sc_recvcookie++; 771 } else { 772 /* Pull out the entry to unlock the bucket row. */ 773 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 774 sch->sch_length--; 775 tcp_syncache.cache_count--; 776 SCH_UNLOCK(sch); 777 } 778 779 /* 780 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 781 */ 782 if (th->th_ack != sc->sc_iss + 1) 783 goto failed; 784 785 so = syncache_socket(sc, *lsop, m); 786 787 if (so == NULL) { 788#if 0 789resetandabort: 790 /* XXXjlemon check this - is this correct? */ 791 (void) tcp_respond(NULL, m, m, th, 792 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 793#endif 794 m_freem(m); /* XXX: only needed for above */ 795 tcpstat.tcps_sc_aborted++; 796 if (sch != NULL) { 797 syncache_insert(sc, sch); /* try again later */ 798 sc = NULL; 799 } 800 goto failed; 801 } else 802 tcpstat.tcps_sc_completed++; 803 *lsop = so; 804 805 syncache_free(sc); 806 return (1); 807failed: 808 if (sc != NULL) 809 syncache_free(sc); 810 return (0); 811} 812 813/* 814 * Given a LISTEN socket and an inbound SYN request, add 815 * this to the syn cache, and send back a segment: 816 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 817 * to the source. 818 * 819 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 820 * Doing so would require that we hold onto the data and deliver it 821 * to the application. However, if we are the target of a SYN-flood 822 * DoS attack, an attacker could send data which would eventually 823 * consume all available buffer space if it were ACKed. By not ACKing 824 * the data, we avoid this DoS scenario. 825 */ 826int 827syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 828 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 829{ 830 struct tcpcb *tp; 831 struct socket *so; 832 struct syncache *sc = NULL; 833 struct syncache_head *sch; 834 struct mbuf *ipopts = NULL; 835 u_int32_t flowtmp; 836 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 837#ifdef INET6 838 int autoflowlabel = 0; 839#endif 840 841 INP_INFO_WLOCK_ASSERT(&tcbinfo); 842 INP_LOCK_ASSERT(inp); /* listen socket */ 843 844 /* 845 * Combine all so/tp operations very early to drop the INP lock as 846 * soon as possible. 847 */ 848 so = *lsop; 849 tp = sototcpcb(so); 850 851#ifdef INET6 852 if (inc->inc_isipv6 && 853 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 854 autoflowlabel = 1; 855#endif 856 ip_ttl = inp->inp_ip_ttl; 857 ip_tos = inp->inp_ip_tos; 858 win = sbspace(&so->so_rcv); 859 sb_hiwat = so->so_rcv.sb_hiwat; 860 noopt = (tp->t_flags & TF_NOOPT); 861 862 so = NULL; 863 tp = NULL; 864 865 INP_UNLOCK(inp); 866 INP_INFO_WUNLOCK(&tcbinfo); 867 868 /* 869 * Remember the IP options, if any. 870 */ 871#ifdef INET6 872 if (!inc->inc_isipv6) 873#endif 874 ipopts = ip_srcroute(m); 875 876 /* 877 * See if we already have an entry for this connection. 878 * If we do, resend the SYN,ACK, and reset the retransmit timer. 879 * 880 * XXX: should the syncache be re-initialized with the contents 881 * of the new SYN here (which may have different options?) 882 */ 883 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 884 SCH_LOCK_ASSERT(sch); 885 if (sc != NULL) { 886 tcpstat.tcps_sc_dupsyn++; 887 if (ipopts) { 888 /* 889 * If we were remembering a previous source route, 890 * forget it and use the new one we've been given. 891 */ 892 if (sc->sc_ipopts) 893 (void) m_free(sc->sc_ipopts); 894 sc->sc_ipopts = ipopts; 895 } 896 /* 897 * Update timestamp if present. 898 */ 899 if (sc->sc_flags & SCF_TIMESTAMP) 900 sc->sc_tsrecent = to->to_tsval; 901 if (syncache_respond(sc, m) == 0) { 902 SYNCACHE_TIMEOUT(sc, sch, 1); 903 tcpstat.tcps_sndacks++; 904 tcpstat.tcps_sndtotal++; 905 } 906 SCH_UNLOCK(sch); 907 goto done; 908 } 909 910 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 911 if (sc == NULL) { 912 /* 913 * The zone allocator couldn't provide more entries. 914 * Treat this as if the cache was full; drop the oldest 915 * entry and insert the new one. 916 */ 917 tcpstat.tcps_sc_zonefail++; 918 sc = TAILQ_LAST(&sch->sch_bucket, sch_head); 919 syncache_drop(sc, sch); 920 SCH_UNLOCK(sch); 921 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO); 922 if (sc == NULL) { 923 if (ipopts) 924 (void) m_free(ipopts); 925 goto done; 926 } 927 } else 928 SCH_UNLOCK(sch); 929 930 /* 931 * Fill in the syncache values. 932 */ 933 sc->sc_ipopts = ipopts; 934 sc->sc_inc.inc_fport = inc->inc_fport; 935 sc->sc_inc.inc_lport = inc->inc_lport; 936#ifdef INET6 937 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 938 if (inc->inc_isipv6) { 939 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 940 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 941 } else 942#endif 943 { 944 sc->sc_inc.inc_faddr = inc->inc_faddr; 945 sc->sc_inc.inc_laddr = inc->inc_laddr; 946 sc->sc_ip_tos = ip_tos; 947 sc->sc_ip_ttl = ip_ttl; 948 } 949 sc->sc_irs = th->th_seq; 950 sc->sc_flags = 0; 951 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 952 sc->sc_flowlabel = 0; 953 if (tcp_syncookies) { 954 sc->sc_iss = syncookie_generate(sc, &flowtmp); 955#ifdef INET6 956 if (autoflowlabel) 957 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK; 958#endif 959 } else { 960 sc->sc_iss = arc4random(); 961#ifdef INET6 962 if (autoflowlabel) 963 sc->sc_flowlabel = 964 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 965#endif 966 } 967 968 /* 969 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 970 * win was derived from socket earlier in the function. 971 */ 972 win = imax(win, 0); 973 win = imin(win, TCP_MAXWIN); 974 sc->sc_wnd = win; 975 976 if (tcp_do_rfc1323) { 977 /* 978 * A timestamp received in a SYN makes 979 * it ok to send timestamp requests and replies. 980 */ 981 if (to->to_flags & TOF_TS) { 982 sc->sc_tsrecent = to->to_tsval; 983 sc->sc_flags |= SCF_TIMESTAMP; 984 } 985 if (to->to_flags & TOF_SCALE) { 986 int wscale = 0; 987 988 /* Compute proper scaling value from buffer space */ 989 while (wscale < TCP_MAX_WINSHIFT && 990 (TCP_MAXWIN << wscale) < sb_hiwat) 991 wscale++; 992 sc->sc_request_r_scale = wscale; 993 sc->sc_requested_s_scale = to->to_requested_s_scale; 994 sc->sc_flags |= SCF_WINSCALE; 995 } 996 } 997#ifdef TCP_SIGNATURE 998 /* 999 * If listening socket requested TCP digests, and received SYN 1000 * contains the option, flag this in the syncache so that 1001 * syncache_respond() will do the right thing with the SYN+ACK. 1002 * XXX: Currently we always record the option by default and will 1003 * attempt to use it in syncache_respond(). 1004 */ 1005 if (to->to_flags & TOF_SIGNATURE) 1006 sc->sc_flags |= SCF_SIGNATURE; 1007#endif 1008 1009 if (to->to_flags & TOF_SACK) 1010 sc->sc_flags |= SCF_SACK; 1011 if (noopt) 1012 sc->sc_flags |= SCF_NOOPT; 1013 1014 /* 1015 * Do a standard 3-way handshake. 1016 */ 1017 if (syncache_respond(sc, m) == 0) { 1018 syncache_insert(sc, sch); /* locks and unlocks sch */ 1019 tcpstat.tcps_sndacks++; 1020 tcpstat.tcps_sndtotal++; 1021 } else { 1022 syncache_free(sc); 1023 tcpstat.tcps_sc_dropped++; 1024 } 1025 1026done: 1027 *lsop = NULL; 1028 return (1); 1029} 1030 1031static int 1032syncache_respond(struct syncache *sc, struct mbuf *m) 1033{ 1034 u_int8_t *optp; 1035 int optlen, error; 1036 u_int16_t tlen, hlen, mssopt; 1037 struct ip *ip = NULL; 1038 struct tcphdr *th; 1039#ifdef INET6 1040 struct ip6_hdr *ip6 = NULL; 1041#endif 1042#ifdef MAC 1043 struct inpcb *inp = NULL; 1044#endif 1045 1046 hlen = 1047#ifdef INET6 1048 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1049#endif 1050 sizeof(struct ip); 1051 1052 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer")); 1053 1054 /* Determine MSS we advertize to other end of connection. */ 1055 mssopt = tcp_mssopt(&sc->sc_inc); 1056 1057 /* Compute the size of the TCP options. */ 1058 if (sc->sc_flags & SCF_NOOPT) { 1059 optlen = 0; 1060 } else { 1061 optlen = TCPOLEN_MAXSEG + 1062 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1063 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 1064#ifdef TCP_SIGNATURE 1065 if (sc->sc_flags & SCF_SIGNATURE) 1066 optlen += TCPOLEN_SIGNATURE; 1067#endif 1068 if (sc->sc_flags & SCF_SACK) 1069 optlen += TCPOLEN_SACK_PERMITTED; 1070 optlen = roundup2(optlen, 4); 1071 } 1072 tlen = hlen + sizeof(struct tcphdr) + optlen; 1073 1074 /* 1075 * XXX: Assume that the entire packet will fit in a header mbuf. 1076 */ 1077 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1078 1079 /* Create the IP+TCP header from scratch. */ 1080 if (m) 1081 m_freem(m); 1082 1083 m = m_gethdr(M_DONTWAIT, MT_DATA); 1084 if (m == NULL) 1085 return (ENOBUFS); 1086 m->m_data += max_linkhdr; 1087 m->m_len = tlen; 1088 m->m_pkthdr.len = tlen; 1089 m->m_pkthdr.rcvif = NULL; 1090 1091#ifdef MAC 1092 /* 1093 * For MAC look up the inpcb to get access to the label information. 1094 * We don't store the inpcb pointer in struct syncache to make locking 1095 * less complicated and to save locking operations. However for MAC 1096 * this gives a slight overhead as we have to do a full pcblookup here. 1097 */ 1098 INP_INFO_RLOCK(&tcbinfo); 1099 if (inp == NULL) { 1100#ifdef INET6 /* && MAC */ 1101 if (sc->sc_inc.inc_isipv6) 1102 inp = in6_pcblookup_hash(&tcbinfo, 1103 &sc->sc_inc.inc6_laddr, sc->sc_inc.inc_lport, 1104 &sc->sc_inc.inc6_faddr, sc->sc_inc.inc_fport, 1105 1, NULL); 1106 else 1107#endif /* INET6 */ 1108 inp = in_pcblookup_hash(&tcbinfo, 1109 sc->sc_inc.inc_laddr, sc->sc_inc.inc_lport, 1110 sc->sc_inc.inc_faddr, sc->sc_inc.inc_fport, 1111 1, NULL); 1112 if (inp == NULL) { 1113 m_freem(m); 1114 INP_INFO_RUNLOCK(&tcbinfo); 1115 return (ESHUTDOWN); 1116 } 1117 } 1118 INP_LOCK(inp); 1119 if (!inp->inp_socket->so_options & SO_ACCEPTCONN) { 1120 m_freem(m); 1121 INP_UNLOCK(inp); 1122 INP_INFO_RUNLOCK(&tcbinfo); 1123 return (ESHUTDOWN); 1124 } 1125 mac_create_mbuf_from_inpcb(inp, m); 1126 INP_UNLOCK(inp); 1127 INP_INFO_RUNLOCK(&tcbinfo); 1128#endif /* MAC */ 1129 1130#ifdef INET6 1131 if (sc->sc_inc.inc_isipv6) { 1132 ip6 = mtod(m, struct ip6_hdr *); 1133 ip6->ip6_vfc = IPV6_VERSION; 1134 ip6->ip6_nxt = IPPROTO_TCP; 1135 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1136 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1137 ip6->ip6_plen = htons(tlen - hlen); 1138 /* ip6_hlim is set after checksum */ 1139 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1140 ip6->ip6_flow |= sc->sc_flowlabel; 1141 1142 th = (struct tcphdr *)(ip6 + 1); 1143 } else 1144#endif 1145 { 1146 ip = mtod(m, struct ip *); 1147 ip->ip_v = IPVERSION; 1148 ip->ip_hl = sizeof(struct ip) >> 2; 1149 ip->ip_len = tlen; 1150 ip->ip_id = 0; 1151 ip->ip_off = 0; 1152 ip->ip_sum = 0; 1153 ip->ip_p = IPPROTO_TCP; 1154 ip->ip_src = sc->sc_inc.inc_laddr; 1155 ip->ip_dst = sc->sc_inc.inc_faddr; 1156 ip->ip_ttl = sc->sc_ip_ttl; 1157 ip->ip_tos = sc->sc_ip_tos; 1158 1159 /* 1160 * See if we should do MTU discovery. Route lookups are 1161 * expensive, so we will only unset the DF bit if: 1162 * 1163 * 1) path_mtu_discovery is disabled 1164 * 2) the SCF_UNREACH flag has been set 1165 */ 1166 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1167 ip->ip_off |= IP_DF; 1168 1169 th = (struct tcphdr *)(ip + 1); 1170 } 1171 th->th_sport = sc->sc_inc.inc_lport; 1172 th->th_dport = sc->sc_inc.inc_fport; 1173 1174 th->th_seq = htonl(sc->sc_iss); 1175 th->th_ack = htonl(sc->sc_irs + 1); 1176 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1177 th->th_x2 = 0; 1178 th->th_flags = TH_SYN|TH_ACK; 1179 th->th_win = htons(sc->sc_wnd); 1180 th->th_urp = 0; 1181 1182 /* Tack on the TCP options. */ 1183 if (optlen != 0) { 1184 optp = (u_int8_t *)(th + 1); 1185 *optp++ = TCPOPT_MAXSEG; 1186 *optp++ = TCPOLEN_MAXSEG; 1187 *optp++ = (mssopt >> 8) & 0xff; 1188 *optp++ = mssopt & 0xff; 1189 1190 if (sc->sc_flags & SCF_WINSCALE) { 1191 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1192 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1193 sc->sc_request_r_scale); 1194 optp += 4; 1195 } 1196 1197 if (sc->sc_flags & SCF_TIMESTAMP) { 1198 u_int32_t *lp = (u_int32_t *)(optp); 1199 1200 /* Form timestamp option per appendix A of RFC 1323. */ 1201 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1202 *lp++ = htonl(ticks); 1203 *lp = htonl(sc->sc_tsrecent); 1204 optp += TCPOLEN_TSTAMP_APPA; 1205 } 1206 1207#ifdef TCP_SIGNATURE 1208 /* 1209 * Handle TCP-MD5 passive opener response. 1210 */ 1211 if (sc->sc_flags & SCF_SIGNATURE) { 1212 u_int8_t *bp = optp; 1213 int i; 1214 1215 *bp++ = TCPOPT_SIGNATURE; 1216 *bp++ = TCPOLEN_SIGNATURE; 1217 for (i = 0; i < TCP_SIGLEN; i++) 1218 *bp++ = 0; 1219 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1220 optp + 2, IPSEC_DIR_OUTBOUND); 1221 optp += TCPOLEN_SIGNATURE; 1222 } 1223#endif /* TCP_SIGNATURE */ 1224 1225 if (sc->sc_flags & SCF_SACK) { 1226 *optp++ = TCPOPT_SACK_PERMITTED; 1227 *optp++ = TCPOLEN_SACK_PERMITTED; 1228 } 1229 1230 { 1231 /* Pad TCP options to a 4 byte boundary */ 1232 int padlen = optlen - (optp - (u_int8_t *)(th + 1)); 1233 while (padlen-- > 0) 1234 *optp++ = TCPOPT_EOL; 1235 } 1236 } 1237 1238#ifdef INET6 1239 if (sc->sc_inc.inc_isipv6) { 1240 th->th_sum = 0; 1241 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1242 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1243 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1244 } else 1245#endif 1246 { 1247 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1248 htons(tlen - hlen + IPPROTO_TCP)); 1249 m->m_pkthdr.csum_flags = CSUM_TCP; 1250 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1251 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1252 } 1253 return (error); 1254} 1255 1256/* 1257 * cookie layers: 1258 * 1259 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1260 * | peer iss | 1261 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1262 * | 0 |(A)| | 1263 * (A): peer mss index 1264 */ 1265 1266/* 1267 * The values below are chosen to minimize the size of the tcp_secret 1268 * table, as well as providing roughly a 16 second lifetime for the cookie. 1269 */ 1270 1271#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1272#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1273 1274#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1275#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1276#define SYNCOOKIE_TIMEOUT \ 1277 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1278#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1279 1280#define SYNCOOKIE_RLOCK(ts) (rw_rlock(&(ts).ts_rwmtx)) 1281#define SYNCOOKIE_RUNLOCK(ts) (rw_runlock(&(ts).ts_rwmtx)) 1282#define SYNCOOKIE_TRY_UPGRADE(ts) (rw_try_upgrade(&(ts).ts_rwmtx)) 1283#define SYNCOOKIE_DOWNGRADE(ts) (rw_downgrade(&(ts).ts_rwmtx)) 1284 1285static struct { 1286 struct rwlock ts_rwmtx; 1287 u_int ts_expire; /* ticks */ 1288 u_int32_t ts_secbits[4]; 1289} tcp_secret[SYNCOOKIE_NSECRETS]; 1290 1291static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1292 1293#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1294 1295struct md5_add { 1296 u_int32_t laddr, faddr; 1297 u_int32_t secbits[4]; 1298 u_int16_t lport, fport; 1299}; 1300 1301#ifdef CTASSERT 1302CTASSERT(sizeof(struct md5_add) == 28); 1303#endif 1304 1305/* 1306 * Consider the problem of a recreated (and retransmitted) cookie. If the 1307 * original SYN was accepted, the connection is established. The second 1308 * SYN is inflight, and if it arrives with an ISN that falls within the 1309 * receive window, the connection is killed. 1310 * 1311 * However, since cookies have other problems, this may not be worth 1312 * worrying about. 1313 */ 1314 1315static void 1316syncookie_init(void) { 1317 int idx; 1318 1319 for (idx = 0; idx < SYNCOOKIE_NSECRETS; idx++) { 1320 rw_init(&(tcp_secret[idx].ts_rwmtx), "tcp_secret"); 1321 } 1322} 1323 1324static u_int32_t 1325syncookie_generate(struct syncache *sc, u_int32_t *flowid) 1326{ 1327 u_int32_t md5_buffer[4]; 1328 u_int32_t data; 1329 int idx, i; 1330 struct md5_add add; 1331 MD5_CTX syn_ctx; 1332 1333 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1334 SYNCOOKIE_RLOCK(tcp_secret[idx]); 1335 if (tcp_secret[idx].ts_expire < time_uptime && 1336 SYNCOOKIE_TRY_UPGRADE(tcp_secret[idx]) ) { 1337 /* need write access */ 1338 for (i = 0; i < 4; i++) 1339 tcp_secret[idx].ts_secbits[i] = arc4random(); 1340 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1341 SYNCOOKIE_DOWNGRADE(tcp_secret[idx]); 1342 } 1343 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1344 if (tcp_msstab[data] <= sc->sc_peer_mss) 1345 break; 1346 data = (data << SYNCOOKIE_WNDBITS) | idx; 1347 data ^= sc->sc_irs; /* peer's iss */ 1348 MD5Init(&syn_ctx); 1349#ifdef INET6 1350 if (sc->sc_inc.inc_isipv6) { 1351 MD5Add(sc->sc_inc.inc6_laddr); 1352 MD5Add(sc->sc_inc.inc6_faddr); 1353 add.laddr = 0; 1354 add.faddr = 0; 1355 } else 1356#endif 1357 { 1358 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1359 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1360 } 1361 add.lport = sc->sc_inc.inc_lport; 1362 add.fport = sc->sc_inc.inc_fport; 1363 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1364 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1365 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1366 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1367 SYNCOOKIE_RUNLOCK(tcp_secret[idx]); 1368 MD5Add(add); 1369 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1370 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1371 *flowid = md5_buffer[1]; 1372 return (data); 1373} 1374 1375static struct syncache * 1376syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1377{ 1378 u_int32_t md5_buffer[4]; 1379 struct syncache *sc; 1380 u_int32_t data; 1381 int wnd, idx; 1382 struct md5_add add; 1383 MD5_CTX syn_ctx; 1384 1385 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1386 idx = data & SYNCOOKIE_WNDMASK; 1387 SYNCOOKIE_RLOCK(tcp_secret[idx]); 1388 if (tcp_secret[idx].ts_expire < ticks || 1389 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) { 1390 SYNCOOKIE_RUNLOCK(tcp_secret[idx]); 1391 return (NULL); 1392 } 1393 MD5Init(&syn_ctx); 1394#ifdef INET6 1395 if (inc->inc_isipv6) { 1396 MD5Add(inc->inc6_laddr); 1397 MD5Add(inc->inc6_faddr); 1398 add.laddr = 0; 1399 add.faddr = 0; 1400 } else 1401#endif 1402 { 1403 add.laddr = inc->inc_laddr.s_addr; 1404 add.faddr = inc->inc_faddr.s_addr; 1405 } 1406 add.lport = inc->inc_lport; 1407 add.fport = inc->inc_fport; 1408 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1409 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1410 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1411 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1412 SYNCOOKIE_RUNLOCK(tcp_secret[idx]); 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 | M_ZERO); 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 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1436 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1437 } else 1438#endif 1439 { 1440 sc->sc_inc.inc_faddr = inc->inc_faddr; 1441 sc->sc_inc.inc_laddr = inc->inc_laddr; 1442 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1443 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 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_rxmits = 0; 1453 sc->sc_peer_mss = tcp_msstab[data]; 1454 return (sc); 1455} 1456