tcp_syncache.c revision 184205
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 33#include <sys/cdefs.h> 34__FBSDID("$FreeBSD: head/sys/netinet/tcp_syncache.c 184205 2008-10-23 15:53:51Z des $"); 35 36#include "opt_inet.h" 37#include "opt_inet6.h" 38#include "opt_ipsec.h" 39#include "opt_mac.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/limits.h> 46#include <sys/lock.h> 47#include <sys/mutex.h> 48#include <sys/malloc.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#include <sys/syslog.h> 56#include <sys/ucred.h> 57#include <sys/vimage.h> 58 59#include <vm/uma.h> 60 61#include <net/if.h> 62#include <net/route.h> 63 64#include <netinet/in.h> 65#include <netinet/in_systm.h> 66#include <netinet/ip.h> 67#include <netinet/in_var.h> 68#include <netinet/in_pcb.h> 69#include <netinet/ip_var.h> 70#include <netinet/ip_options.h> 71#ifdef INET6 72#include <netinet/ip6.h> 73#include <netinet/icmp6.h> 74#include <netinet6/nd6.h> 75#include <netinet6/ip6_var.h> 76#include <netinet6/in6_pcb.h> 77#endif 78#include <netinet/tcp.h> 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#include <netinet/tcp_syncache.h> 84#include <netinet/tcp_offload.h> 85#ifdef INET6 86#include <netinet6/tcp6_var.h> 87#endif 88 89#ifdef IPSEC 90#include <netipsec/ipsec.h> 91#ifdef INET6 92#include <netipsec/ipsec6.h> 93#endif 94#include <netipsec/key.h> 95#endif /*IPSEC*/ 96 97#include <machine/in_cksum.h> 98 99#include <security/mac/mac_framework.h> 100 101static int tcp_syncookies = 1; 102SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 103 &tcp_syncookies, 0, 104 "Use TCP SYN cookies if the syncache overflows"); 105 106static int tcp_syncookiesonly = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW, 108 &tcp_syncookiesonly, 0, 109 "Use only TCP SYN cookies"); 110 111#ifdef TCP_OFFLOAD_DISABLE 112#define TOEPCB_ISSET(sc) (0) 113#else 114#define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL) 115#endif 116 117static void syncache_drop(struct syncache *, struct syncache_head *); 118static void syncache_free(struct syncache *); 119static void syncache_insert(struct syncache *, struct syncache_head *); 120struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 121static int syncache_respond(struct syncache *); 122static struct socket *syncache_socket(struct syncache *, struct socket *, 123 struct mbuf *m); 124static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, 125 int docallout); 126static void syncache_timer(void *); 127static void syncookie_generate(struct syncache_head *, struct syncache *, 128 u_int32_t *); 129static struct syncache 130 *syncookie_lookup(struct in_conninfo *, struct syncache_head *, 131 struct syncache *, struct tcpopt *, struct tcphdr *, 132 struct socket *); 133 134/* 135 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 136 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds, 137 * the odds are that the user has given up attempting to connect by then. 138 */ 139#define SYNCACHE_MAXREXMTS 3 140 141/* Arbitrary values */ 142#define TCP_SYNCACHE_HASHSIZE 512 143#define TCP_SYNCACHE_BUCKETLIMIT 30 144 145static struct tcp_syncache tcp_syncache; 146 147SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 148 149SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 150 bucketlimit, CTLFLAG_RDTUN, 151 tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 152 153SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 154 cachelimit, CTLFLAG_RDTUN, 155 tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 156 157SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 158 count, CTLFLAG_RD, 159 tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 160 161SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 162 hashsize, CTLFLAG_RDTUN, 163 tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 164 165SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 166 rexmtlimit, CTLFLAG_RW, 167 tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 168 169int tcp_sc_rst_sock_fail = 1; 170SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 171 rst_on_sock_fail, CTLFLAG_RW, 172 tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure"); 173 174static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 175 176#define SYNCACHE_HASH(inc, mask) \ 177 ((V_tcp_syncache.hash_secret ^ \ 178 (inc)->inc_faddr.s_addr ^ \ 179 ((inc)->inc_faddr.s_addr >> 16) ^ \ 180 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 181 182#define SYNCACHE_HASH6(inc, mask) \ 183 ((V_tcp_syncache.hash_secret ^ \ 184 (inc)->inc6_faddr.s6_addr32[0] ^ \ 185 (inc)->inc6_faddr.s6_addr32[3] ^ \ 186 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 187 188#define ENDPTS_EQ(a, b) ( \ 189 (a)->ie_fport == (b)->ie_fport && \ 190 (a)->ie_lport == (b)->ie_lport && \ 191 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 192 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 193) 194 195#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 196 197#define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 198#define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 199#define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 200 201/* 202 * Requires the syncache entry to be already removed from the bucket list. 203 */ 204static void 205syncache_free(struct syncache *sc) 206{ 207 INIT_VNET_INET(curvnet); 208 209 if (sc->sc_ipopts) 210 (void) m_free(sc->sc_ipopts); 211 if (sc->sc_cred) 212 crfree(sc->sc_cred); 213#ifdef MAC 214 mac_syncache_destroy(&sc->sc_label); 215#endif 216 217 uma_zfree(V_tcp_syncache.zone, sc); 218} 219 220void 221syncache_init(void) 222{ 223 INIT_VNET_INET(curvnet); 224 int i; 225 226 V_tcp_syncache.cache_count = 0; 227 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 228 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 229 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 230 V_tcp_syncache.hash_secret = arc4random(); 231 232 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 233 &V_tcp_syncache.hashsize); 234 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 235 &V_tcp_syncache.bucket_limit); 236 if (!powerof2(V_tcp_syncache.hashsize) || 237 V_tcp_syncache.hashsize == 0) { 238 printf("WARNING: syncache hash size is not a power of 2.\n"); 239 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 240 } 241 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1; 242 243 /* Set limits. */ 244 V_tcp_syncache.cache_limit = 245 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit; 246 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 247 &V_tcp_syncache.cache_limit); 248 249 /* Allocate the hash table. */ 250 V_tcp_syncache.hashbase = malloc( V_tcp_syncache.hashsize * sizeof(struct syncache_head), 251 M_SYNCACHE, M_WAITOK | M_ZERO); 252 253 /* Initialize the hash buckets. */ 254 for (i = 0; i < V_tcp_syncache.hashsize; i++) { 255 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket); 256 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 257 NULL, MTX_DEF); 258 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer, 259 &V_tcp_syncache.hashbase[i].sch_mtx, 0); 260 V_tcp_syncache.hashbase[i].sch_length = 0; 261 } 262 263 /* Create the syncache entry zone. */ 264 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 265 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 266 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit); 267} 268 269/* 270 * Inserts a syncache entry into the specified bucket row. 271 * Locks and unlocks the syncache_head autonomously. 272 */ 273static void 274syncache_insert(struct syncache *sc, struct syncache_head *sch) 275{ 276 INIT_VNET_INET(sch->sch_vnet); 277 struct syncache *sc2; 278 279 SCH_LOCK(sch); 280 281 /* 282 * Make sure that we don't overflow the per-bucket limit. 283 * If the bucket is full, toss the oldest element. 284 */ 285 if (sch->sch_length >= V_tcp_syncache.bucket_limit) { 286 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 287 ("sch->sch_length incorrect")); 288 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 289 syncache_drop(sc2, sch); 290 V_tcpstat.tcps_sc_bucketoverflow++; 291 } 292 293 /* Put it into the bucket. */ 294 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 295 sch->sch_length++; 296 297 /* Reinitialize the bucket row's timer. */ 298 if (sch->sch_length == 1) 299 sch->sch_nextc = ticks + INT_MAX; 300 syncache_timeout(sc, sch, 1); 301 302 SCH_UNLOCK(sch); 303 304 V_tcp_syncache.cache_count++; 305 V_tcpstat.tcps_sc_added++; 306} 307 308/* 309 * Remove and free entry from syncache bucket row. 310 * Expects locked syncache head. 311 */ 312static void 313syncache_drop(struct syncache *sc, struct syncache_head *sch) 314{ 315 INIT_VNET_INET(sch->sch_vnet); 316 317 SCH_LOCK_ASSERT(sch); 318 319 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 320 sch->sch_length--; 321 322#ifndef TCP_OFFLOAD_DISABLE 323 if (sc->sc_tu) 324 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb); 325#endif 326 syncache_free(sc); 327 V_tcp_syncache.cache_count--; 328} 329 330/* 331 * Engage/reengage time on bucket row. 332 */ 333static void 334syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) 335{ 336 sc->sc_rxttime = ticks + 337 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]); 338 sc->sc_rxmits++; 339 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) { 340 sch->sch_nextc = sc->sc_rxttime; 341 if (docallout) 342 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, 343 syncache_timer, (void *)sch); 344 } 345} 346 347/* 348 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 349 * If we have retransmitted an entry the maximum number of times, expire it. 350 * One separate timer for each bucket row. 351 */ 352static void 353syncache_timer(void *xsch) 354{ 355 struct syncache_head *sch = (struct syncache_head *)xsch; 356 INIT_VNET_INET(sch->sch_vnet); 357 struct syncache *sc, *nsc; 358 int tick = ticks; 359 char *s; 360 361 /* NB: syncache_head has already been locked by the callout. */ 362 SCH_LOCK_ASSERT(sch); 363 364 /* 365 * In the following cycle we may remove some entries and/or 366 * advance some timeouts, so re-initialize the bucket timer. 367 */ 368 sch->sch_nextc = tick + INT_MAX; 369 370 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 371 /* 372 * We do not check if the listen socket still exists 373 * and accept the case where the listen socket may be 374 * gone by the time we resend the SYN/ACK. We do 375 * not expect this to happens often. If it does, 376 * then the RST will be sent by the time the remote 377 * host does the SYN/ACK->ACK. 378 */ 379 if (TSTMP_GT(sc->sc_rxttime, tick)) { 380 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) 381 sch->sch_nextc = sc->sc_rxttime; 382 continue; 383 } 384 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) { 385 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 386 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " 387 "giving up and removing syncache entry\n", 388 s, __func__); 389 free(s, M_TCPLOG); 390 } 391 syncache_drop(sc, sch); 392 V_tcpstat.tcps_sc_stale++; 393 continue; 394 } 395 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 396 log(LOG_DEBUG, "%s; %s: Response timeout, " 397 "retransmitting (%u) SYN|ACK\n", 398 s, __func__, sc->sc_rxmits); 399 free(s, M_TCPLOG); 400 } 401 402 (void) syncache_respond(sc); 403 V_tcpstat.tcps_sc_retransmitted++; 404 syncache_timeout(sc, sch, 0); 405 } 406 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 407 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 408 syncache_timer, (void *)(sch)); 409} 410 411/* 412 * Find an entry in the syncache. 413 * Returns always with locked syncache_head plus a matching entry or NULL. 414 */ 415struct syncache * 416syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 417{ 418 INIT_VNET_INET(curvnet); 419 struct syncache *sc; 420 struct syncache_head *sch; 421 422#ifdef INET6 423 if (inc->inc_isipv6) { 424 sch = &V_tcp_syncache.hashbase[ 425 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)]; 426 *schp = sch; 427 428 SCH_LOCK(sch); 429 430 /* Circle through bucket row to find matching entry. */ 431 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 432 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 433 return (sc); 434 } 435 } else 436#endif 437 { 438 sch = &V_tcp_syncache.hashbase[ 439 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)]; 440 *schp = sch; 441 442 SCH_LOCK(sch); 443 444 /* Circle through bucket row to find matching entry. */ 445 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 446#ifdef INET6 447 if (sc->sc_inc.inc_isipv6) 448 continue; 449#endif 450 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 451 return (sc); 452 } 453 } 454 SCH_LOCK_ASSERT(*schp); 455 return (NULL); /* always returns with locked sch */ 456} 457 458/* 459 * This function is called when we get a RST for a 460 * non-existent connection, so that we can see if the 461 * connection is in the syn cache. If it is, zap it. 462 */ 463void 464syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 465{ 466 INIT_VNET_INET(curvnet); 467 struct syncache *sc; 468 struct syncache_head *sch; 469 char *s = NULL; 470 471 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 472 SCH_LOCK_ASSERT(sch); 473 474 /* 475 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. 476 * See RFC 793 page 65, section SEGMENT ARRIVES. 477 */ 478 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { 479 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 480 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " 481 "FIN flag set, segment ignored\n", s, __func__); 482 V_tcpstat.tcps_badrst++; 483 goto done; 484 } 485 486 /* 487 * No corresponding connection was found in syncache. 488 * If syncookies are enabled and possibly exclusively 489 * used, or we are under memory pressure, a valid RST 490 * may not find a syncache entry. In that case we're 491 * done and no SYN|ACK retransmissions will happen. 492 * Otherwise the the RST was misdirected or spoofed. 493 */ 494 if (sc == NULL) { 495 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 496 log(LOG_DEBUG, "%s; %s: Spurious RST without matching " 497 "syncache entry (possibly syncookie only), " 498 "segment ignored\n", s, __func__); 499 V_tcpstat.tcps_badrst++; 500 goto done; 501 } 502 503 /* 504 * If the RST bit is set, check the sequence number to see 505 * if this is a valid reset segment. 506 * RFC 793 page 37: 507 * In all states except SYN-SENT, all reset (RST) segments 508 * are validated by checking their SEQ-fields. A reset is 509 * valid if its sequence number is in the window. 510 * 511 * The sequence number in the reset segment is normally an 512 * echo of our outgoing acknowlegement numbers, but some hosts 513 * send a reset with the sequence number at the rightmost edge 514 * of our receive window, and we have to handle this case. 515 */ 516 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 517 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 518 syncache_drop(sc, sch); 519 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 520 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " 521 "connection attempt aborted by remote endpoint\n", 522 s, __func__); 523 V_tcpstat.tcps_sc_reset++; 524 } else { 525 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 526 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != " 527 "IRS %u (+WND %u), segment ignored\n", 528 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd); 529 V_tcpstat.tcps_badrst++; 530 } 531 532done: 533 if (s != NULL) 534 free(s, M_TCPLOG); 535 SCH_UNLOCK(sch); 536} 537 538void 539syncache_badack(struct in_conninfo *inc) 540{ 541 INIT_VNET_INET(curvnet); 542 struct syncache *sc; 543 struct syncache_head *sch; 544 545 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 546 SCH_LOCK_ASSERT(sch); 547 if (sc != NULL) { 548 syncache_drop(sc, sch); 549 V_tcpstat.tcps_sc_badack++; 550 } 551 SCH_UNLOCK(sch); 552} 553 554void 555syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 556{ 557 INIT_VNET_INET(curvnet); 558 struct syncache *sc; 559 struct syncache_head *sch; 560 561 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 562 SCH_LOCK_ASSERT(sch); 563 if (sc == NULL) 564 goto done; 565 566 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 567 if (ntohl(th->th_seq) != sc->sc_iss) 568 goto done; 569 570 /* 571 * If we've rertransmitted 3 times and this is our second error, 572 * we remove the entry. Otherwise, we allow it to continue on. 573 * This prevents us from incorrectly nuking an entry during a 574 * spurious network outage. 575 * 576 * See tcp_notify(). 577 */ 578 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 579 sc->sc_flags |= SCF_UNREACH; 580 goto done; 581 } 582 syncache_drop(sc, sch); 583 V_tcpstat.tcps_sc_unreach++; 584done: 585 SCH_UNLOCK(sch); 586} 587 588/* 589 * Build a new TCP socket structure from a syncache entry. 590 */ 591static struct socket * 592syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 593{ 594 INIT_VNET_INET(lso->so_vnet); 595 struct inpcb *inp = NULL; 596 struct socket *so; 597 struct tcpcb *tp; 598 char *s; 599 600 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 601 602 /* 603 * Ok, create the full blown connection, and set things up 604 * as they would have been set up if we had created the 605 * connection when the SYN arrived. If we can't create 606 * the connection, abort it. 607 */ 608 so = sonewconn(lso, SS_ISCONNECTED); 609 if (so == NULL) { 610 /* 611 * Drop the connection; we will either send a RST or 612 * have the peer retransmit its SYN again after its 613 * RTO and try again. 614 */ 615 V_tcpstat.tcps_listendrop++; 616 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 617 log(LOG_DEBUG, "%s; %s: Socket create failed " 618 "due to limits or memory shortage\n", 619 s, __func__); 620 free(s, M_TCPLOG); 621 } 622 goto abort2; 623 } 624#ifdef MAC 625 SOCK_LOCK(so); 626 mac_socketpeer_set_from_mbuf(m, so); 627 SOCK_UNLOCK(so); 628#endif 629 630 inp = sotoinpcb(so); 631 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum; 632 so->so_fibnum = sc->sc_inc.inc_fibnum; 633 INP_WLOCK(inp); 634 635 /* Insert new socket into PCB hash list. */ 636 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 637#ifdef INET6 638 if (sc->sc_inc.inc_isipv6) { 639 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 640 } else { 641 inp->inp_vflag &= ~INP_IPV6; 642 inp->inp_vflag |= INP_IPV4; 643#endif 644 inp->inp_laddr = sc->sc_inc.inc_laddr; 645#ifdef INET6 646 } 647#endif 648 inp->inp_lport = sc->sc_inc.inc_lport; 649 if (in_pcbinshash(inp) != 0) { 650 /* 651 * Undo the assignments above if we failed to 652 * put the PCB on the hash lists. 653 */ 654#ifdef INET6 655 if (sc->sc_inc.inc_isipv6) 656 inp->in6p_laddr = in6addr_any; 657 else 658#endif 659 inp->inp_laddr.s_addr = INADDR_ANY; 660 inp->inp_lport = 0; 661 goto abort; 662 } 663#ifdef IPSEC 664 /* Copy old policy into new socket's. */ 665 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 666 printf("syncache_socket: could not copy policy\n"); 667#endif 668#ifdef INET6 669 if (sc->sc_inc.inc_isipv6) { 670 struct inpcb *oinp = sotoinpcb(lso); 671 struct in6_addr laddr6; 672 struct sockaddr_in6 sin6; 673 /* 674 * Inherit socket options from the listening socket. 675 * Note that in6p_inputopts are not (and should not be) 676 * copied, since it stores previously received options and is 677 * used to detect if each new option is different than the 678 * previous one and hence should be passed to a user. 679 * If we copied in6p_inputopts, a user would not be able to 680 * receive options just after calling the accept system call. 681 */ 682 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 683 if (oinp->in6p_outputopts) 684 inp->in6p_outputopts = 685 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 686 687 sin6.sin6_family = AF_INET6; 688 sin6.sin6_len = sizeof(sin6); 689 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 690 sin6.sin6_port = sc->sc_inc.inc_fport; 691 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 692 laddr6 = inp->in6p_laddr; 693 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 694 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 695 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 696 thread0.td_ucred)) { 697 inp->in6p_laddr = laddr6; 698 goto abort; 699 } 700 /* Override flowlabel from in6_pcbconnect. */ 701 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 702 inp->in6p_flowinfo |= sc->sc_flowlabel; 703 } else 704#endif 705 { 706 struct in_addr laddr; 707 struct sockaddr_in sin; 708 709 inp->inp_options = (m) ? ip_srcroute(m) : NULL; 710 711 if (inp->inp_options == NULL) { 712 inp->inp_options = sc->sc_ipopts; 713 sc->sc_ipopts = NULL; 714 } 715 716 sin.sin_family = AF_INET; 717 sin.sin_len = sizeof(sin); 718 sin.sin_addr = sc->sc_inc.inc_faddr; 719 sin.sin_port = sc->sc_inc.inc_fport; 720 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 721 laddr = inp->inp_laddr; 722 if (inp->inp_laddr.s_addr == INADDR_ANY) 723 inp->inp_laddr = sc->sc_inc.inc_laddr; 724 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 725 thread0.td_ucred)) { 726 inp->inp_laddr = laddr; 727 goto abort; 728 } 729 } 730 tp = intotcpcb(inp); 731 tp->t_state = TCPS_SYN_RECEIVED; 732 tp->iss = sc->sc_iss; 733 tp->irs = sc->sc_irs; 734 tcp_rcvseqinit(tp); 735 tcp_sendseqinit(tp); 736 tp->snd_wl1 = sc->sc_irs; 737 tp->snd_max = tp->iss + 1; 738 tp->snd_nxt = tp->iss + 1; 739 tp->rcv_up = sc->sc_irs + 1; 740 tp->rcv_wnd = sc->sc_wnd; 741 tp->rcv_adv += tp->rcv_wnd; 742 tp->last_ack_sent = tp->rcv_nxt; 743 744 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 745 if (sc->sc_flags & SCF_NOOPT) 746 tp->t_flags |= TF_NOOPT; 747 else { 748 if (sc->sc_flags & SCF_WINSCALE) { 749 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 750 tp->snd_scale = sc->sc_requested_s_scale; 751 tp->request_r_scale = sc->sc_requested_r_scale; 752 } 753 if (sc->sc_flags & SCF_TIMESTAMP) { 754 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 755 tp->ts_recent = sc->sc_tsreflect; 756 tp->ts_recent_age = ticks; 757 tp->ts_offset = sc->sc_tsoff; 758 } 759#ifdef TCP_SIGNATURE 760 if (sc->sc_flags & SCF_SIGNATURE) 761 tp->t_flags |= TF_SIGNATURE; 762#endif 763 if (sc->sc_flags & SCF_SACK) 764 tp->t_flags |= TF_SACK_PERMIT; 765 } 766 767 if (sc->sc_flags & SCF_ECN) 768 tp->t_flags |= TF_ECN_PERMIT; 769 770 /* 771 * Set up MSS and get cached values from tcp_hostcache. 772 * This might overwrite some of the defaults we just set. 773 */ 774 tcp_mss(tp, sc->sc_peer_mss); 775 776 /* 777 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 778 */ 779 if (sc->sc_rxmits) 780 tp->snd_cwnd = tp->t_maxseg; 781 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 782 783 INP_WUNLOCK(inp); 784 785 V_tcpstat.tcps_accepts++; 786 return (so); 787 788abort: 789 INP_WUNLOCK(inp); 790abort2: 791 if (so != NULL) 792 soabort(so); 793 return (NULL); 794} 795 796/* 797 * This function gets called when we receive an ACK for a 798 * socket in the LISTEN state. We look up the connection 799 * in the syncache, and if its there, we pull it out of 800 * the cache and turn it into a full-blown connection in 801 * the SYN-RECEIVED state. 802 */ 803int 804syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 805 struct socket **lsop, struct mbuf *m) 806{ 807 INIT_VNET_INET(curvnet); 808 struct syncache *sc; 809 struct syncache_head *sch; 810 struct syncache scs; 811 char *s; 812 813 /* 814 * Global TCP locks are held because we manipulate the PCB lists 815 * and create a new socket. 816 */ 817 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 818 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, 819 ("%s: can handle only ACK", __func__)); 820 821 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 822 SCH_LOCK_ASSERT(sch); 823 if (sc == NULL) { 824 /* 825 * There is no syncache entry, so see if this ACK is 826 * a returning syncookie. To do this, first: 827 * A. See if this socket has had a syncache entry dropped in 828 * the past. We don't want to accept a bogus syncookie 829 * if we've never received a SYN. 830 * B. check that the syncookie is valid. If it is, then 831 * cobble up a fake syncache entry, and return. 832 */ 833 if (!tcp_syncookies) { 834 SCH_UNLOCK(sch); 835 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 836 log(LOG_DEBUG, "%s; %s: Spurious ACK, " 837 "segment rejected (syncookies disabled)\n", 838 s, __func__); 839 goto failed; 840 } 841 bzero(&scs, sizeof(scs)); 842 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 843 SCH_UNLOCK(sch); 844 if (sc == NULL) { 845 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 846 log(LOG_DEBUG, "%s; %s: Segment failed " 847 "SYNCOOKIE authentication, segment rejected " 848 "(probably spoofed)\n", s, __func__); 849 goto failed; 850 } 851 } else { 852 /* Pull out the entry to unlock the bucket row. */ 853 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 854 sch->sch_length--; 855 V_tcp_syncache.cache_count--; 856 SCH_UNLOCK(sch); 857 } 858 859 /* 860 * Segment validation: 861 * ACK must match our initial sequence number + 1 (the SYN|ACK). 862 */ 863 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) { 864 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 865 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " 866 "rejected\n", s, __func__, th->th_ack, sc->sc_iss); 867 goto failed; 868 } 869 870 /* 871 * The SEQ must fall in the window starting at the received 872 * initial receive sequence number + 1 (the SYN). 873 */ 874 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) || 875 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) && 876 !TOEPCB_ISSET(sc)) { 877 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 878 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " 879 "rejected\n", s, __func__, th->th_seq, sc->sc_irs); 880 goto failed; 881 } 882 883 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { 884 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 885 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 886 "segment rejected\n", s, __func__); 887 goto failed; 888 } 889 /* 890 * If timestamps were negotiated the reflected timestamp 891 * must be equal to what we actually sent in the SYN|ACK. 892 */ 893 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts && 894 !TOEPCB_ISSET(sc)) { 895 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 896 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, " 897 "segment rejected\n", 898 s, __func__, to->to_tsecr, sc->sc_ts); 899 goto failed; 900 } 901 902 *lsop = syncache_socket(sc, *lsop, m); 903 904 if (*lsop == NULL) 905 V_tcpstat.tcps_sc_aborted++; 906 else 907 V_tcpstat.tcps_sc_completed++; 908 909/* how do we find the inp for the new socket? */ 910 if (sc != &scs) 911 syncache_free(sc); 912 return (1); 913failed: 914 if (sc != NULL && sc != &scs) 915 syncache_free(sc); 916 if (s != NULL) 917 free(s, M_TCPLOG); 918 *lsop = NULL; 919 return (0); 920} 921 922int 923tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to, 924 struct tcphdr *th, struct socket **lsop, struct mbuf *m) 925{ 926 int rc; 927 928 INP_INFO_WLOCK(&V_tcbinfo); 929 rc = syncache_expand(inc, to, th, lsop, m); 930 INP_INFO_WUNLOCK(&V_tcbinfo); 931 932 return (rc); 933} 934 935/* 936 * Given a LISTEN socket and an inbound SYN request, add 937 * this to the syn cache, and send back a segment: 938 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 939 * to the source. 940 * 941 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 942 * Doing so would require that we hold onto the data and deliver it 943 * to the application. However, if we are the target of a SYN-flood 944 * DoS attack, an attacker could send data which would eventually 945 * consume all available buffer space if it were ACKed. By not ACKing 946 * the data, we avoid this DoS scenario. 947 */ 948static void 949_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 950 struct inpcb *inp, struct socket **lsop, struct mbuf *m, 951 struct toe_usrreqs *tu, void *toepcb) 952{ 953 INIT_VNET_INET(inp->inp_vnet); 954 struct tcpcb *tp; 955 struct socket *so; 956 struct syncache *sc = NULL; 957 struct syncache_head *sch; 958 struct mbuf *ipopts = NULL; 959 u_int32_t flowtmp; 960 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 961 char *s; 962#ifdef INET6 963 int autoflowlabel = 0; 964#endif 965#ifdef MAC 966 struct label *maclabel; 967#endif 968 struct syncache scs; 969 struct ucred *cred; 970 971 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 972 INP_WLOCK_ASSERT(inp); /* listen socket */ 973 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, 974 ("%s: unexpected tcp flags", __func__)); 975 976 /* 977 * Combine all so/tp operations very early to drop the INP lock as 978 * soon as possible. 979 */ 980 so = *lsop; 981 tp = sototcpcb(so); 982 cred = crhold(so->so_cred); 983 984#ifdef INET6 985 if (inc->inc_isipv6 && 986 (inp->in6p_flags & IN6P_AUTOFLOWLABEL)) 987 autoflowlabel = 1; 988#endif 989 ip_ttl = inp->inp_ip_ttl; 990 ip_tos = inp->inp_ip_tos; 991 win = sbspace(&so->so_rcv); 992 sb_hiwat = so->so_rcv.sb_hiwat; 993 noopt = (tp->t_flags & TF_NOOPT); 994 995 /* By the time we drop the lock these should no longer be used. */ 996 so = NULL; 997 tp = NULL; 998 999#ifdef MAC 1000 if (mac_syncache_init(&maclabel) != 0) { 1001 INP_WUNLOCK(inp); 1002 INP_INFO_WUNLOCK(&V_tcbinfo); 1003 goto done; 1004 } else 1005 mac_syncache_create(maclabel, inp); 1006#endif 1007 INP_WUNLOCK(inp); 1008 INP_INFO_WUNLOCK(&V_tcbinfo); 1009 1010 /* 1011 * Remember the IP options, if any. 1012 */ 1013#ifdef INET6 1014 if (!inc->inc_isipv6) 1015#endif 1016 ipopts = (m) ? ip_srcroute(m) : NULL; 1017 1018 /* 1019 * See if we already have an entry for this connection. 1020 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1021 * 1022 * XXX: should the syncache be re-initialized with the contents 1023 * of the new SYN here (which may have different options?) 1024 * 1025 * XXX: We do not check the sequence number to see if this is a 1026 * real retransmit or a new connection attempt. The question is 1027 * how to handle such a case; either ignore it as spoofed, or 1028 * drop the current entry and create a new one? 1029 */ 1030 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 1031 SCH_LOCK_ASSERT(sch); 1032 if (sc != NULL) { 1033#ifndef TCP_OFFLOAD_DISABLE 1034 if (sc->sc_tu) 1035 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT, 1036 sc->sc_toepcb); 1037#endif 1038 V_tcpstat.tcps_sc_dupsyn++; 1039 if (ipopts) { 1040 /* 1041 * If we were remembering a previous source route, 1042 * forget it and use the new one we've been given. 1043 */ 1044 if (sc->sc_ipopts) 1045 (void) m_free(sc->sc_ipopts); 1046 sc->sc_ipopts = ipopts; 1047 } 1048 /* 1049 * Update timestamp if present. 1050 */ 1051 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 1052 sc->sc_tsreflect = to->to_tsval; 1053 else 1054 sc->sc_flags &= ~SCF_TIMESTAMP; 1055#ifdef MAC 1056 /* 1057 * Since we have already unconditionally allocated label 1058 * storage, free it up. The syncache entry will already 1059 * have an initialized label we can use. 1060 */ 1061 mac_syncache_destroy(&maclabel); 1062 KASSERT(sc->sc_label != NULL, 1063 ("%s: label not initialized", __func__)); 1064#endif 1065 /* Retransmit SYN|ACK and reset retransmit count. */ 1066 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { 1067 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " 1068 "resetting timer and retransmitting SYN|ACK\n", 1069 s, __func__); 1070 free(s, M_TCPLOG); 1071 } 1072 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) { 1073 sc->sc_rxmits = 0; 1074 syncache_timeout(sc, sch, 1); 1075 V_tcpstat.tcps_sndacks++; 1076 V_tcpstat.tcps_sndtotal++; 1077 } 1078 SCH_UNLOCK(sch); 1079 goto done; 1080 } 1081 1082 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); 1083 if (sc == NULL) { 1084 /* 1085 * The zone allocator couldn't provide more entries. 1086 * Treat this as if the cache was full; drop the oldest 1087 * entry and insert the new one. 1088 */ 1089 V_tcpstat.tcps_sc_zonefail++; 1090 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 1091 syncache_drop(sc, sch); 1092 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); 1093 if (sc == NULL) { 1094 if (tcp_syncookies) { 1095 bzero(&scs, sizeof(scs)); 1096 sc = &scs; 1097 } else { 1098 SCH_UNLOCK(sch); 1099 if (ipopts) 1100 (void) m_free(ipopts); 1101 goto done; 1102 } 1103 } 1104 } 1105 1106 /* 1107 * Fill in the syncache values. 1108 */ 1109#ifdef MAC 1110 sc->sc_label = maclabel; 1111#endif 1112 sc->sc_cred = cred; 1113 cred = NULL; 1114 sc->sc_ipopts = ipopts; 1115 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum; 1116 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1117#ifdef INET6 1118 if (!inc->inc_isipv6) 1119#endif 1120 { 1121 sc->sc_ip_tos = ip_tos; 1122 sc->sc_ip_ttl = ip_ttl; 1123 } 1124#ifndef TCP_OFFLOAD_DISABLE 1125 sc->sc_tu = tu; 1126 sc->sc_toepcb = toepcb; 1127#endif 1128 sc->sc_irs = th->th_seq; 1129 sc->sc_iss = arc4random(); 1130 sc->sc_flags = 0; 1131 sc->sc_flowlabel = 0; 1132 1133 /* 1134 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1135 * win was derived from socket earlier in the function. 1136 */ 1137 win = imax(win, 0); 1138 win = imin(win, TCP_MAXWIN); 1139 sc->sc_wnd = win; 1140 1141 if (V_tcp_do_rfc1323) { 1142 /* 1143 * A timestamp received in a SYN makes 1144 * it ok to send timestamp requests and replies. 1145 */ 1146 if (to->to_flags & TOF_TS) { 1147 sc->sc_tsreflect = to->to_tsval; 1148 sc->sc_ts = ticks; 1149 sc->sc_flags |= SCF_TIMESTAMP; 1150 } 1151 if (to->to_flags & TOF_SCALE) { 1152 int wscale = 0; 1153 1154 /* 1155 * Pick the smallest possible scaling factor that 1156 * will still allow us to scale up to sb_max, aka 1157 * kern.ipc.maxsockbuf. 1158 * 1159 * We do this because there are broken firewalls that 1160 * will corrupt the window scale option, leading to 1161 * the other endpoint believing that our advertised 1162 * window is unscaled. At scale factors larger than 1163 * 5 the unscaled window will drop below 1500 bytes, 1164 * leading to serious problems when traversing these 1165 * broken firewalls. 1166 * 1167 * With the default maxsockbuf of 256K, a scale factor 1168 * of 3 will be chosen by this algorithm. Those who 1169 * choose a larger maxsockbuf should watch out 1170 * for the compatiblity problems mentioned above. 1171 * 1172 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1173 * or <SYN,ACK>) segment itself is never scaled. 1174 */ 1175 while (wscale < TCP_MAX_WINSHIFT && 1176 (TCP_MAXWIN << wscale) < sb_max) 1177 wscale++; 1178 sc->sc_requested_r_scale = wscale; 1179 sc->sc_requested_s_scale = to->to_wscale; 1180 sc->sc_flags |= SCF_WINSCALE; 1181 } 1182 } 1183#ifdef TCP_SIGNATURE 1184 /* 1185 * If listening socket requested TCP digests, and received SYN 1186 * contains the option, flag this in the syncache so that 1187 * syncache_respond() will do the right thing with the SYN+ACK. 1188 * XXX: Currently we always record the option by default and will 1189 * attempt to use it in syncache_respond(). 1190 */ 1191 if (to->to_flags & TOF_SIGNATURE) 1192 sc->sc_flags |= SCF_SIGNATURE; 1193#endif 1194 if (to->to_flags & TOF_SACKPERM) 1195 sc->sc_flags |= SCF_SACK; 1196 if (to->to_flags & TOF_MSS) 1197 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1198 if (noopt) 1199 sc->sc_flags |= SCF_NOOPT; 1200 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn) 1201 sc->sc_flags |= SCF_ECN; 1202 1203 if (tcp_syncookies) { 1204 syncookie_generate(sch, sc, &flowtmp); 1205#ifdef INET6 1206 if (autoflowlabel) 1207 sc->sc_flowlabel = flowtmp; 1208#endif 1209 } else { 1210#ifdef INET6 1211 if (autoflowlabel) 1212 sc->sc_flowlabel = 1213 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1214#endif 1215 } 1216 SCH_UNLOCK(sch); 1217 1218 /* 1219 * Do a standard 3-way handshake. 1220 */ 1221 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) { 1222 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs) 1223 syncache_free(sc); 1224 else if (sc != &scs) 1225 syncache_insert(sc, sch); /* locks and unlocks sch */ 1226 V_tcpstat.tcps_sndacks++; 1227 V_tcpstat.tcps_sndtotal++; 1228 } else { 1229 if (sc != &scs) 1230 syncache_free(sc); 1231 V_tcpstat.tcps_sc_dropped++; 1232 } 1233 1234done: 1235 if (cred != NULL) 1236 crfree(cred); 1237#ifdef MAC 1238 if (sc == &scs) 1239 mac_syncache_destroy(&maclabel); 1240#endif 1241 if (m) { 1242 1243 *lsop = NULL; 1244 m_freem(m); 1245 } 1246 return; 1247} 1248 1249static int 1250syncache_respond(struct syncache *sc) 1251{ 1252 INIT_VNET_INET(curvnet); 1253 struct ip *ip = NULL; 1254 struct mbuf *m; 1255 struct tcphdr *th; 1256 int optlen, error; 1257 u_int16_t hlen, tlen, mssopt; 1258 struct tcpopt to; 1259#ifdef INET6 1260 struct ip6_hdr *ip6 = NULL; 1261#endif 1262 1263 hlen = 1264#ifdef INET6 1265 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 1266#endif 1267 sizeof(struct ip); 1268 tlen = hlen + sizeof(struct tcphdr); 1269 1270 /* Determine MSS we advertize to other end of connection. */ 1271 mssopt = tcp_mssopt(&sc->sc_inc); 1272 if (sc->sc_peer_mss) 1273 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss); 1274 1275 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1276 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, 1277 ("syncache: mbuf too small")); 1278 1279 /* Create the IP+TCP header from scratch. */ 1280 m = m_gethdr(M_DONTWAIT, MT_DATA); 1281 if (m == NULL) 1282 return (ENOBUFS); 1283#ifdef MAC 1284 mac_syncache_create_mbuf(sc->sc_label, m); 1285#endif 1286 m->m_data += max_linkhdr; 1287 m->m_len = tlen; 1288 m->m_pkthdr.len = tlen; 1289 m->m_pkthdr.rcvif = NULL; 1290 1291#ifdef INET6 1292 if (sc->sc_inc.inc_isipv6) { 1293 ip6 = mtod(m, struct ip6_hdr *); 1294 ip6->ip6_vfc = IPV6_VERSION; 1295 ip6->ip6_nxt = IPPROTO_TCP; 1296 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1297 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1298 ip6->ip6_plen = htons(tlen - hlen); 1299 /* ip6_hlim is set after checksum */ 1300 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1301 ip6->ip6_flow |= sc->sc_flowlabel; 1302 1303 th = (struct tcphdr *)(ip6 + 1); 1304 } else 1305#endif 1306 { 1307 ip = mtod(m, struct ip *); 1308 ip->ip_v = IPVERSION; 1309 ip->ip_hl = sizeof(struct ip) >> 2; 1310 ip->ip_len = tlen; 1311 ip->ip_id = 0; 1312 ip->ip_off = 0; 1313 ip->ip_sum = 0; 1314 ip->ip_p = IPPROTO_TCP; 1315 ip->ip_src = sc->sc_inc.inc_laddr; 1316 ip->ip_dst = sc->sc_inc.inc_faddr; 1317 ip->ip_ttl = sc->sc_ip_ttl; 1318 ip->ip_tos = sc->sc_ip_tos; 1319 1320 /* 1321 * See if we should do MTU discovery. Route lookups are 1322 * expensive, so we will only unset the DF bit if: 1323 * 1324 * 1) path_mtu_discovery is disabled 1325 * 2) the SCF_UNREACH flag has been set 1326 */ 1327 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1328 ip->ip_off |= IP_DF; 1329 1330 th = (struct tcphdr *)(ip + 1); 1331 } 1332 th->th_sport = sc->sc_inc.inc_lport; 1333 th->th_dport = sc->sc_inc.inc_fport; 1334 1335 th->th_seq = htonl(sc->sc_iss); 1336 th->th_ack = htonl(sc->sc_irs + 1); 1337 th->th_off = sizeof(struct tcphdr) >> 2; 1338 th->th_x2 = 0; 1339 th->th_flags = TH_SYN|TH_ACK; 1340 th->th_win = htons(sc->sc_wnd); 1341 th->th_urp = 0; 1342 1343 if (sc->sc_flags & SCF_ECN) { 1344 th->th_flags |= TH_ECE; 1345 V_tcpstat.tcps_ecn_shs++; 1346 } 1347 1348 /* Tack on the TCP options. */ 1349 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1350 to.to_flags = 0; 1351 1352 to.to_mss = mssopt; 1353 to.to_flags = TOF_MSS; 1354 if (sc->sc_flags & SCF_WINSCALE) { 1355 to.to_wscale = sc->sc_requested_r_scale; 1356 to.to_flags |= TOF_SCALE; 1357 } 1358 if (sc->sc_flags & SCF_TIMESTAMP) { 1359 /* Virgin timestamp or TCP cookie enhanced one. */ 1360 to.to_tsval = sc->sc_ts; 1361 to.to_tsecr = sc->sc_tsreflect; 1362 to.to_flags |= TOF_TS; 1363 } 1364 if (sc->sc_flags & SCF_SACK) 1365 to.to_flags |= TOF_SACKPERM; 1366#ifdef TCP_SIGNATURE 1367 if (sc->sc_flags & SCF_SIGNATURE) 1368 to.to_flags |= TOF_SIGNATURE; 1369#endif 1370 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1371 1372 /* Adjust headers by option size. */ 1373 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1374 m->m_len += optlen; 1375 m->m_pkthdr.len += optlen; 1376 1377#ifdef TCP_SIGNATURE 1378 if (sc->sc_flags & SCF_SIGNATURE) 1379 tcp_signature_compute(m, 0, 0, optlen, 1380 to.to_signature, IPSEC_DIR_OUTBOUND); 1381#endif 1382#ifdef INET6 1383 if (sc->sc_inc.inc_isipv6) 1384 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1385 else 1386#endif 1387 ip->ip_len += optlen; 1388 } else 1389 optlen = 0; 1390 1391#ifdef INET6 1392 if (sc->sc_inc.inc_isipv6) { 1393 th->th_sum = 0; 1394 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1395 tlen + optlen - hlen); 1396 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1397 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1398 } else 1399#endif 1400 { 1401 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1402 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1403 m->m_pkthdr.csum_flags = CSUM_TCP; 1404 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1405 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1406 } 1407 return (error); 1408} 1409 1410void 1411syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 1412 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 1413{ 1414 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL); 1415} 1416 1417void 1418tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to, 1419 struct tcphdr *th, struct inpcb *inp, struct socket **lsop, 1420 struct toe_usrreqs *tu, void *toepcb) 1421{ 1422 INIT_VNET_INET(curvnet); 1423 1424 INP_INFO_WLOCK(&V_tcbinfo); 1425 INP_WLOCK(inp); 1426 _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb); 1427} 1428 1429/* 1430 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1431 * receive and to be able to handle SYN floods from bogus source addresses 1432 * (where we will never receive any reply). SYN floods try to exhaust all 1433 * our memory and available slots in the SYN cache table to cause a denial 1434 * of service to legitimate users of the local host. 1435 * 1436 * The idea of SYN cookies is to encode and include all necessary information 1437 * about the connection setup state within the SYN-ACK we send back and thus 1438 * to get along without keeping any local state until the ACK to the SYN-ACK 1439 * arrives (if ever). Everything we need to know should be available from 1440 * the information we encoded in the SYN-ACK. 1441 * 1442 * More information about the theory behind SYN cookies and its first 1443 * discussion and specification can be found at: 1444 * http://cr.yp.to/syncookies.html (overview) 1445 * http://cr.yp.to/syncookies/archive (gory details) 1446 * 1447 * This implementation extends the orginal idea and first implementation 1448 * of FreeBSD by using not only the initial sequence number field to store 1449 * information but also the timestamp field if present. This way we can 1450 * keep track of the entire state we need to know to recreate the session in 1451 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1452 * these days. For those that do not we still have to live with the known 1453 * shortcomings of the ISN only SYN cookies. 1454 * 1455 * Cookie layers: 1456 * 1457 * Initial sequence number we send: 1458 * 31|................................|0 1459 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1460 * D = MD5 Digest (first dword) 1461 * M = MSS index 1462 * R = Rotation of secret 1463 * P = Odd or Even secret 1464 * 1465 * The MD5 Digest is computed with over following parameters: 1466 * a) randomly rotated secret 1467 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1468 * c) the received initial sequence number from remote host 1469 * d) the rotation offset and odd/even bit 1470 * 1471 * Timestamp we send: 1472 * 31|................................|0 1473 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1474 * D = MD5 Digest (third dword) (only as filler) 1475 * S = Requested send window scale 1476 * R = Requested receive window scale 1477 * A = SACK allowed 1478 * 5 = TCP-MD5 enabled (not implemented yet) 1479 * XORed with MD5 Digest (forth dword) 1480 * 1481 * The timestamp isn't cryptographically secure and doesn't need to be. 1482 * The double use of the MD5 digest dwords ties it to a specific remote/ 1483 * local host/port, remote initial sequence number and our local time 1484 * limited secret. A received timestamp is reverted (XORed) and then 1485 * the contained MD5 dword is compared to the computed one to ensure the 1486 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1487 * have been tampered with but this isn't different from supplying bogus 1488 * values in the SYN in the first place. 1489 * 1490 * Some problems with SYN cookies remain however: 1491 * Consider the problem of a recreated (and retransmitted) cookie. If the 1492 * original SYN was accepted, the connection is established. The second 1493 * SYN is inflight, and if it arrives with an ISN that falls within the 1494 * receive window, the connection is killed. 1495 * 1496 * Notes: 1497 * A heuristic to determine when to accept syn cookies is not necessary. 1498 * An ACK flood would cause the syncookie verification to be attempted, 1499 * but a SYN flood causes syncookies to be generated. Both are of equal 1500 * cost, so there's no point in trying to optimize the ACK flood case. 1501 * Also, if you don't process certain ACKs for some reason, then all someone 1502 * would have to do is launch a SYN and ACK flood at the same time, which 1503 * would stop cookie verification and defeat the entire purpose of syncookies. 1504 */ 1505static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1506 1507static void 1508syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1509 u_int32_t *flowlabel) 1510{ 1511 INIT_VNET_INET(curvnet); 1512 MD5_CTX ctx; 1513 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1514 u_int32_t data; 1515 u_int32_t *secbits; 1516 u_int off, pmss, mss; 1517 int i; 1518 1519 SCH_LOCK_ASSERT(sch); 1520 1521 /* Which of the two secrets to use. */ 1522 secbits = sch->sch_oddeven ? 1523 sch->sch_secbits_odd : sch->sch_secbits_even; 1524 1525 /* Reseed secret if too old. */ 1526 if (sch->sch_reseed < time_uptime) { 1527 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1528 secbits = sch->sch_oddeven ? 1529 sch->sch_secbits_odd : sch->sch_secbits_even; 1530 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1531 secbits[i] = arc4random(); 1532 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1533 } 1534 1535 /* Secret rotation offset. */ 1536 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1537 1538 /* Maximum segment size calculation. */ 1539 pmss = 1540 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss); 1541 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1542 if (tcp_sc_msstab[mss] <= pmss) 1543 break; 1544 1545 /* Fold parameters and MD5 digest into the ISN we will send. */ 1546 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1547 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1548 data |= mss << 4; /* mss, 3 bits */ 1549 1550 MD5Init(&ctx); 1551 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1552 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1553 MD5Update(&ctx, secbits, off); 1554 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1555 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1556 MD5Update(&ctx, &data, sizeof(data)); 1557 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1558 1559 data |= (md5_buffer[0] << 7); 1560 sc->sc_iss = data; 1561 1562#ifdef INET6 1563 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1564#endif 1565 1566 /* Additional parameters are stored in the timestamp if present. */ 1567 if (sc->sc_flags & SCF_TIMESTAMP) { 1568 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1569 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1570 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1571 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1572 data |= md5_buffer[2] << 10; /* more digest bits */ 1573 data ^= md5_buffer[3]; 1574 sc->sc_ts = data; 1575 sc->sc_tsoff = data - ticks; /* after XOR */ 1576 } 1577 1578 V_tcpstat.tcps_sc_sendcookie++; 1579 return; 1580} 1581 1582static struct syncache * 1583syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1584 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1585 struct socket *so) 1586{ 1587 INIT_VNET_INET(curvnet); 1588 MD5_CTX ctx; 1589 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1590 u_int32_t data = 0; 1591 u_int32_t *secbits; 1592 tcp_seq ack, seq; 1593 int off, mss, wnd, flags; 1594 1595 SCH_LOCK_ASSERT(sch); 1596 1597 /* 1598 * Pull information out of SYN-ACK/ACK and 1599 * revert sequence number advances. 1600 */ 1601 ack = th->th_ack - 1; 1602 seq = th->th_seq - 1; 1603 off = (ack >> 1) & 0x7; 1604 mss = (ack >> 4) & 0x7; 1605 flags = ack & 0x7f; 1606 1607 /* Which of the two secrets to use. */ 1608 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1609 1610 /* 1611 * The secret wasn't updated for the lifetime of a syncookie, 1612 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1613 */ 1614 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) { 1615 return (NULL); 1616 } 1617 1618 /* Recompute the digest so we can compare it. */ 1619 MD5Init(&ctx); 1620 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1621 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1622 MD5Update(&ctx, secbits, off); 1623 MD5Update(&ctx, inc, sizeof(*inc)); 1624 MD5Update(&ctx, &seq, sizeof(seq)); 1625 MD5Update(&ctx, &flags, sizeof(flags)); 1626 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1627 1628 /* Does the digest part of or ACK'ed ISS match? */ 1629 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1630 return (NULL); 1631 1632 /* Does the digest part of our reflected timestamp match? */ 1633 if (to->to_flags & TOF_TS) { 1634 data = md5_buffer[3] ^ to->to_tsecr; 1635 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1636 return (NULL); 1637 } 1638 1639 /* Fill in the syncache values. */ 1640 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1641 sc->sc_ipopts = NULL; 1642 1643 sc->sc_irs = seq; 1644 sc->sc_iss = ack; 1645 1646#ifdef INET6 1647 if (inc->inc_isipv6) { 1648 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL) 1649 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1650 } else 1651#endif 1652 { 1653 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1654 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1655 } 1656 1657 /* Additional parameters that were encoded in the timestamp. */ 1658 if (data) { 1659 sc->sc_flags |= SCF_TIMESTAMP; 1660 sc->sc_tsreflect = to->to_tsval; 1661 sc->sc_ts = to->to_tsecr; 1662 sc->sc_tsoff = to->to_tsecr - ticks; 1663 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1664 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1665 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1666 TCP_MAX_WINSHIFT); 1667 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1668 TCP_MAX_WINSHIFT); 1669 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1670 sc->sc_flags |= SCF_WINSCALE; 1671 } else 1672 sc->sc_flags |= SCF_NOOPT; 1673 1674 wnd = sbspace(&so->so_rcv); 1675 wnd = imax(wnd, 0); 1676 wnd = imin(wnd, TCP_MAXWIN); 1677 sc->sc_wnd = wnd; 1678 1679 sc->sc_rxmits = 0; 1680 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1681 1682 V_tcpstat.tcps_sc_recvcookie++; 1683 return (sc); 1684} 1685 1686/* 1687 * Returns the current number of syncache entries. This number 1688 * will probably change before you get around to calling 1689 * syncache_pcblist. 1690 */ 1691 1692int 1693syncache_pcbcount(void) 1694{ 1695 INIT_VNET_INET(curvnet); 1696 struct syncache_head *sch; 1697 int count, i; 1698 1699 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { 1700 /* No need to lock for a read. */ 1701 sch = &V_tcp_syncache.hashbase[i]; 1702 count += sch->sch_length; 1703 } 1704 return count; 1705} 1706 1707/* 1708 * Exports the syncache entries to userland so that netstat can display 1709 * them alongside the other sockets. This function is intended to be 1710 * called only from tcp_pcblist. 1711 * 1712 * Due to concurrency on an active system, the number of pcbs exported 1713 * may have no relation to max_pcbs. max_pcbs merely indicates the 1714 * amount of space the caller allocated for this function to use. 1715 */ 1716int 1717syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported) 1718{ 1719 INIT_VNET_INET(curvnet); 1720 struct xtcpcb xt; 1721 struct syncache *sc; 1722 struct syncache_head *sch; 1723 int count, error, i; 1724 1725 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { 1726 sch = &V_tcp_syncache.hashbase[i]; 1727 SCH_LOCK(sch); 1728 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 1729 if (count >= max_pcbs) { 1730 SCH_UNLOCK(sch); 1731 goto exit; 1732 } 1733 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0) 1734 continue; 1735 bzero(&xt, sizeof(xt)); 1736 xt.xt_len = sizeof(xt); 1737 if (sc->sc_inc.inc_isipv6) 1738 xt.xt_inp.inp_vflag = INP_IPV6; 1739 else 1740 xt.xt_inp.inp_vflag = INP_IPV4; 1741 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); 1742 xt.xt_tp.t_inpcb = &xt.xt_inp; 1743 xt.xt_tp.t_state = TCPS_SYN_RECEIVED; 1744 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1745 xt.xt_socket.xso_len = sizeof (struct xsocket); 1746 xt.xt_socket.so_type = SOCK_STREAM; 1747 xt.xt_socket.so_state = SS_ISCONNECTING; 1748 error = SYSCTL_OUT(req, &xt, sizeof xt); 1749 if (error) { 1750 SCH_UNLOCK(sch); 1751 goto exit; 1752 } 1753 count++; 1754 } 1755 SCH_UNLOCK(sch); 1756 } 1757exit: 1758 *pcbs_exported = count; 1759 return error; 1760} 1761