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("$FreeBSD$"); 35 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/sysctl.h> 44 #include <sys/limits.h> 45 #include <sys/lock.h> 46 #include <sys/mutex.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/md5.h> 50 #include <sys/proc.h> /* for proc0 declaration */ 51 #include <sys/random.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/syslog.h> 55 #include <sys/ucred.h> 56 #include <sys/vimage.h> 57 58 #include <vm/uma.h> 59 60 #include <net/if.h> 61 #include <net/route.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/ip.h> 66 #include <netinet/in_var.h> 67 #include <netinet/in_pcb.h> 68 #include <netinet/ip_var.h> 69 #include <netinet/ip_options.h> 70 #ifdef INET6 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/nd6.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/in6_pcb.h> 76 #endif 77 #include <netinet/tcp.h> 78 #include <netinet/tcp_fsm.h> 79 #include <netinet/tcp_seq.h> 80 #include <netinet/tcp_timer.h> 81 #include <netinet/tcp_var.h> 82 #include <netinet/tcp_syncache.h> 83 #include <netinet/tcp_offload.h> 84 #ifdef INET6 85 #include <netinet6/tcp6_var.h> 86 #endif 87 #include <netinet/vinet.h> 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> 100 101 #ifdef VIMAGE_GLOBALS 102 static struct tcp_syncache tcp_syncache; 103 static int tcp_syncookies; 104 static int tcp_syncookiesonly; 105 int tcp_sc_rst_sock_fail; 106 #endif 107 108 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, syncookies, 109 CTLFLAG_RW, tcp_syncookies, 0, 110 "Use TCP SYN cookies if the syncache overflows"); 111 112 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, syncookies_only, 113 CTLFLAG_RW, tcp_syncookiesonly, 0, 114 "Use only TCP SYN cookies"); 115 116 #ifdef TCP_OFFLOAD_DISABLE 117 #define TOEPCB_ISSET(sc) (0) 118 #else 119 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL) 120 #endif 121 122 static void syncache_drop(struct syncache *, struct syncache_head *); 123 static void syncache_free(struct syncache *); 124 static void syncache_insert(struct syncache *, struct syncache_head *); 125 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 126 static int syncache_respond(struct syncache *); 127 static struct socket *syncache_socket(struct syncache *, struct socket *, 128 struct mbuf *m); 129 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, 130 int docallout); 131 static void syncache_timer(void *); 132 static void syncookie_generate(struct syncache_head *, struct syncache *, 133 u_int32_t *); 134 static struct syncache 135 *syncookie_lookup(struct in_conninfo *, struct syncache_head *, 136 struct syncache *, struct tcpopt *, struct tcphdr *, 137 struct socket *); 138 139 /* 140 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 141 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds, 142 * the odds are that the user has given up attempting to connect by then. 143 */ 144 #define SYNCACHE_MAXREXMTS 3 145 146 /* Arbitrary values */ 147 #define TCP_SYNCACHE_HASHSIZE 512 148 #define TCP_SYNCACHE_BUCKETLIMIT 30 149 150 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 151 152 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 153 bucketlimit, CTLFLAG_RDTUN, 154 tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 155 156 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 157 cachelimit, CTLFLAG_RDTUN, 158 tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 159 160 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 161 count, CTLFLAG_RD, 162 tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 163 164 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 165 hashsize, CTLFLAG_RDTUN, 166 tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 167 168 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 169 rexmtlimit, CTLFLAG_RW, 170 tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 171 172 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO, 173 rst_on_sock_fail, CTLFLAG_RW, 174 tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure"); 175 176 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 177 178 #define SYNCACHE_HASH(inc, mask) \ 179 ((V_tcp_syncache.hash_secret ^ \ 180 (inc)->inc_faddr.s_addr ^ \ 181 ((inc)->inc_faddr.s_addr >> 16) ^ \ 182 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 183 184 #define SYNCACHE_HASH6(inc, mask) \ 185 ((V_tcp_syncache.hash_secret ^ \ 186 (inc)->inc6_faddr.s6_addr32[0] ^ \ 187 (inc)->inc6_faddr.s6_addr32[3] ^ \ 188 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 189 190 #define ENDPTS_EQ(a, b) ( \ 191 (a)->ie_fport == (b)->ie_fport && \ 192 (a)->ie_lport == (b)->ie_lport && \ 193 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 194 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 195 ) 196 197 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 198 199 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) 200 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) 201 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) 202 203 /* 204 * Requires the syncache entry to be already removed from the bucket list. 205 */ 206 static void 207 syncache_free(struct syncache *sc) 208 { 209 INIT_VNET_INET(curvnet); 210 211 if (sc->sc_ipopts) 212 (void) m_free(sc->sc_ipopts); 213 if (sc->sc_cred) 214 crfree(sc->sc_cred); 215 #ifdef MAC 216 mac_syncache_destroy(&sc->sc_label); 217 #endif 218 219 uma_zfree(V_tcp_syncache.zone, sc); 220 } 221 222 void 223 syncache_init(void) 224 { 225 INIT_VNET_INET(curvnet); 226 int i; 227 228 V_tcp_syncookies = 1; 229 V_tcp_syncookiesonly = 0; 230 V_tcp_sc_rst_sock_fail = 1; 231 232 V_tcp_syncache.cache_count = 0; 233 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 234 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 235 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 236 V_tcp_syncache.hash_secret = arc4random(); 237 238 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 239 &V_tcp_syncache.hashsize); 240 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 241 &V_tcp_syncache.bucket_limit); 242 if (!powerof2(V_tcp_syncache.hashsize) || 243 V_tcp_syncache.hashsize == 0) { 244 printf("WARNING: syncache hash size is not a power of 2.\n"); 245 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 246 } 247 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1; 248 249 /* Set limits. */ 250 V_tcp_syncache.cache_limit = 251 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit; 252 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 253 &V_tcp_syncache.cache_limit); 254 255 /* Allocate the hash table. */ 256 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize * 257 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO); 258 259 /* Initialize the hash buckets. */ 260 for (i = 0; i < V_tcp_syncache.hashsize; i++) { 261 #ifdef VIMAGE 262 V_tcp_syncache.hashbase[i].sch_vnet = curvnet; 263 #endif 264 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket); 265 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 266 NULL, MTX_DEF); 267 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer, 268 &V_tcp_syncache.hashbase[i].sch_mtx, 0); 269 V_tcp_syncache.hashbase[i].sch_length = 0; 270 } 271 272 /* Create the syncache entry zone. */ 273 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 274 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 275 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit); 276 } 277 278 #ifdef VIMAGE 279 void 280 syncache_destroy(void) 281 { 282 INIT_VNET_INET(curvnet); 283 284 /* XXX walk the cache, free remaining objects, stop timers */ 285 286 uma_zdestroy(V_tcp_syncache.zone); 287 FREE(V_tcp_syncache.hashbase, M_SYNCACHE); 288 } 289 #endif 290 291 /* 292 * Inserts a syncache entry into the specified bucket row. 293 * Locks and unlocks the syncache_head autonomously. 294 */ 295 static void 296 syncache_insert(struct syncache *sc, struct syncache_head *sch) 297 { 298 INIT_VNET_INET(sch->sch_vnet); 299 struct syncache *sc2; 300 301 SCH_LOCK(sch); 302 303 /* 304 * Make sure that we don't overflow the per-bucket limit. 305 * If the bucket is full, toss the oldest element. 306 */ 307 if (sch->sch_length >= V_tcp_syncache.bucket_limit) { 308 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 309 ("sch->sch_length incorrect")); 310 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 311 syncache_drop(sc2, sch); 312 TCPSTAT_INC(tcps_sc_bucketoverflow); 313 } 314 315 /* Put it into the bucket. */ 316 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 317 sch->sch_length++; 318 319 /* Reinitialize the bucket row's timer. */ 320 if (sch->sch_length == 1) 321 sch->sch_nextc = ticks + INT_MAX; 322 syncache_timeout(sc, sch, 1); 323 324 SCH_UNLOCK(sch); 325 326 V_tcp_syncache.cache_count++; 327 TCPSTAT_INC(tcps_sc_added); 328 } 329 330 /* 331 * Remove and free entry from syncache bucket row. 332 * Expects locked syncache head. 333 */ 334 static void 335 syncache_drop(struct syncache *sc, struct syncache_head *sch) 336 { 337 INIT_VNET_INET(sch->sch_vnet); 338 339 SCH_LOCK_ASSERT(sch); 340 341 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 342 sch->sch_length--; 343 344 #ifndef TCP_OFFLOAD_DISABLE 345 if (sc->sc_tu) 346 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb); 347 #endif 348 syncache_free(sc); 349 V_tcp_syncache.cache_count--; 350 } 351 352 /* 353 * Engage/reengage time on bucket row. 354 */ 355 static void 356 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) 357 { 358 sc->sc_rxttime = ticks + 359 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]); 360 sc->sc_rxmits++; 361 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) { 362 sch->sch_nextc = sc->sc_rxttime; 363 if (docallout) 364 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, 365 syncache_timer, (void *)sch); 366 } 367 } 368 369 /* 370 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 371 * If we have retransmitted an entry the maximum number of times, expire it. 372 * One separate timer for each bucket row. 373 */ 374 static void 375 syncache_timer(void *xsch) 376 { 377 struct syncache_head *sch = (struct syncache_head *)xsch; 378 struct syncache *sc, *nsc; 379 int tick = ticks; 380 char *s; 381 382 CURVNET_SET(sch->sch_vnet); 383 INIT_VNET_INET(sch->sch_vnet); 384 385 /* NB: syncache_head has already been locked by the callout. */ 386 SCH_LOCK_ASSERT(sch); 387 388 /* 389 * In the following cycle we may remove some entries and/or 390 * advance some timeouts, so re-initialize the bucket timer. 391 */ 392 sch->sch_nextc = tick + INT_MAX; 393 394 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 395 /* 396 * We do not check if the listen socket still exists 397 * and accept the case where the listen socket may be 398 * gone by the time we resend the SYN/ACK. We do 399 * not expect this to happens often. If it does, 400 * then the RST will be sent by the time the remote 401 * host does the SYN/ACK->ACK. 402 */ 403 if (TSTMP_GT(sc->sc_rxttime, tick)) { 404 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) 405 sch->sch_nextc = sc->sc_rxttime; 406 continue; 407 } 408 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) { 409 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 410 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " 411 "giving up and removing syncache entry\n", 412 s, __func__); 413 free(s, M_TCPLOG); 414 } 415 syncache_drop(sc, sch); 416 TCPSTAT_INC(tcps_sc_stale); 417 continue; 418 } 419 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 420 log(LOG_DEBUG, "%s; %s: Response timeout, " 421 "retransmitting (%u) SYN|ACK\n", 422 s, __func__, sc->sc_rxmits); 423 free(s, M_TCPLOG); 424 } 425 426 (void) syncache_respond(sc); 427 TCPSTAT_INC(tcps_sc_retransmitted); 428 syncache_timeout(sc, sch, 0); 429 } 430 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 431 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 432 syncache_timer, (void *)(sch)); 433 CURVNET_RESTORE(); 434 } 435 436 /* 437 * Find an entry in the syncache. 438 * Returns always with locked syncache_head plus a matching entry or NULL. 439 */ 440 struct syncache * 441 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 442 { 443 INIT_VNET_INET(curvnet); 444 struct syncache *sc; 445 struct syncache_head *sch; 446 447 #ifdef INET6 448 if (inc->inc_flags & INC_ISIPV6) { 449 sch = &V_tcp_syncache.hashbase[ 450 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)]; 451 *schp = sch; 452 453 SCH_LOCK(sch); 454 455 /* Circle through bucket row to find matching entry. */ 456 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 457 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 458 return (sc); 459 } 460 } else 461 #endif 462 { 463 sch = &V_tcp_syncache.hashbase[ 464 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)]; 465 *schp = sch; 466 467 SCH_LOCK(sch); 468 469 /* Circle through bucket row to find matching entry. */ 470 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 471 #ifdef INET6 472 if (sc->sc_inc.inc_flags & INC_ISIPV6) 473 continue; 474 #endif 475 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 476 return (sc); 477 } 478 } 479 SCH_LOCK_ASSERT(*schp); 480 return (NULL); /* always returns with locked sch */ 481 } 482 483 /* 484 * This function is called when we get a RST for a 485 * non-existent connection, so that we can see if the 486 * connection is in the syn cache. If it is, zap it. 487 */ 488 void 489 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 490 { 491 INIT_VNET_INET(curvnet); 492 struct syncache *sc; 493 struct syncache_head *sch; 494 char *s = NULL; 495 496 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 497 SCH_LOCK_ASSERT(sch); 498 499 /* 500 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. 501 * See RFC 793 page 65, section SEGMENT ARRIVES. 502 */ 503 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { 504 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 505 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " 506 "FIN flag set, segment ignored\n", s, __func__); 507 TCPSTAT_INC(tcps_badrst); 508 goto done; 509 } 510 511 /* 512 * No corresponding connection was found in syncache. 513 * If syncookies are enabled and possibly exclusively 514 * used, or we are under memory pressure, a valid RST 515 * may not find a syncache entry. In that case we're 516 * done and no SYN|ACK retransmissions will happen. 517 * Otherwise the the RST was misdirected or spoofed. 518 */ 519 if (sc == NULL) { 520 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 521 log(LOG_DEBUG, "%s; %s: Spurious RST without matching " 522 "syncache entry (possibly syncookie only), " 523 "segment ignored\n", s, __func__); 524 TCPSTAT_INC(tcps_badrst); 525 goto done; 526 } 527 528 /* 529 * If the RST bit is set, check the sequence number to see 530 * if this is a valid reset segment. 531 * RFC 793 page 37: 532 * In all states except SYN-SENT, all reset (RST) segments 533 * are validated by checking their SEQ-fields. A reset is 534 * valid if its sequence number is in the window. 535 * 536 * The sequence number in the reset segment is normally an 537 * echo of our outgoing acknowlegement numbers, but some hosts 538 * send a reset with the sequence number at the rightmost edge 539 * of our receive window, and we have to handle this case. 540 */ 541 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 542 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 543 syncache_drop(sc, sch); 544 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 545 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " 546 "connection attempt aborted by remote endpoint\n", 547 s, __func__); 548 TCPSTAT_INC(tcps_sc_reset); 549 } else { 550 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 551 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != " 552 "IRS %u (+WND %u), segment ignored\n", 553 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd); 554 TCPSTAT_INC(tcps_badrst); 555 } 556 557 done: 558 if (s != NULL) 559 free(s, M_TCPLOG); 560 SCH_UNLOCK(sch); 561 } 562 563 void 564 syncache_badack(struct in_conninfo *inc) 565 { 566 INIT_VNET_INET(curvnet); 567 struct syncache *sc; 568 struct syncache_head *sch; 569 570 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 571 SCH_LOCK_ASSERT(sch); 572 if (sc != NULL) { 573 syncache_drop(sc, sch); 574 TCPSTAT_INC(tcps_sc_badack); 575 } 576 SCH_UNLOCK(sch); 577 } 578 579 void 580 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 581 { 582 INIT_VNET_INET(curvnet); 583 struct syncache *sc; 584 struct syncache_head *sch; 585 586 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 587 SCH_LOCK_ASSERT(sch); 588 if (sc == NULL) 589 goto done; 590 591 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 592 if (ntohl(th->th_seq) != sc->sc_iss) 593 goto done; 594 595 /* 596 * If we've rertransmitted 3 times and this is our second error, 597 * we remove the entry. Otherwise, we allow it to continue on. 598 * This prevents us from incorrectly nuking an entry during a 599 * spurious network outage. 600 * 601 * See tcp_notify(). 602 */ 603 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 604 sc->sc_flags |= SCF_UNREACH; 605 goto done; 606 } 607 syncache_drop(sc, sch); 608 TCPSTAT_INC(tcps_sc_unreach); 609 done: 610 SCH_UNLOCK(sch); 611 } 612 613 /* 614 * Build a new TCP socket structure from a syncache entry. 615 */ 616 static struct socket * 617 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 618 { 619 INIT_VNET_INET(lso->so_vnet); 620 struct inpcb *inp = NULL; 621 struct socket *so; 622 struct tcpcb *tp; 623 char *s; 624 625 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 626 627 /* 628 * Ok, create the full blown connection, and set things up 629 * as they would have been set up if we had created the 630 * connection when the SYN arrived. If we can't create 631 * the connection, abort it. 632 */ 633 so = sonewconn(lso, SS_ISCONNECTED); 634 if (so == NULL) { 635 /* 636 * Drop the connection; we will either send a RST or 637 * have the peer retransmit its SYN again after its 638 * RTO and try again. 639 */ 640 TCPSTAT_INC(tcps_listendrop); 641 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 642 log(LOG_DEBUG, "%s; %s: Socket create failed " 643 "due to limits or memory shortage\n", 644 s, __func__); 645 free(s, M_TCPLOG); 646 } 647 goto abort2; 648 } 649 #ifdef MAC 650 mac_socketpeer_set_from_mbuf(m, so); 651 #endif 652 653 inp = sotoinpcb(so); 654 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum; 655 so->so_fibnum = sc->sc_inc.inc_fibnum; 656 INP_WLOCK(inp); 657 658 /* Insert new socket into PCB hash list. */ 659 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags; 660 #ifdef INET6 661 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 662 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 663 } else { 664 inp->inp_vflag &= ~INP_IPV6; 665 inp->inp_vflag |= INP_IPV4; 666 #endif 667 inp->inp_laddr = sc->sc_inc.inc_laddr; 668 #ifdef INET6 669 } 670 #endif 671 inp->inp_lport = sc->sc_inc.inc_lport; 672 if (in_pcbinshash(inp) != 0) { 673 /* 674 * Undo the assignments above if we failed to 675 * put the PCB on the hash lists. 676 */ 677 #ifdef INET6 678 if (sc->sc_inc.inc_flags & INC_ISIPV6) 679 inp->in6p_laddr = in6addr_any; 680 else 681 #endif 682 inp->inp_laddr.s_addr = INADDR_ANY; 683 inp->inp_lport = 0; 684 goto abort; 685 } 686 #ifdef IPSEC 687 /* Copy old policy into new socket's. */ 688 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 689 printf("syncache_socket: could not copy policy\n"); 690 #endif 691 #ifdef INET6 692 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 693 struct inpcb *oinp = sotoinpcb(lso); 694 struct in6_addr laddr6; 695 struct sockaddr_in6 sin6; 696 /* 697 * Inherit socket options from the listening socket. 698 * Note that in6p_inputopts are not (and should not be) 699 * copied, since it stores previously received options and is 700 * used to detect if each new option is different than the 701 * previous one and hence should be passed to a user. 702 * If we copied in6p_inputopts, a user would not be able to 703 * receive options just after calling the accept system call. 704 */ 705 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 706 if (oinp->in6p_outputopts) 707 inp->in6p_outputopts = 708 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 709 710 sin6.sin6_family = AF_INET6; 711 sin6.sin6_len = sizeof(sin6); 712 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 713 sin6.sin6_port = sc->sc_inc.inc_fport; 714 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 715 laddr6 = inp->in6p_laddr; 716 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 717 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 718 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 719 thread0.td_ucred)) { 720 inp->in6p_laddr = laddr6; 721 goto abort; 722 } 723 /* Override flowlabel from in6_pcbconnect. */ 724 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; 725 inp->inp_flow |= sc->sc_flowlabel; 726 } else 727 #endif 728 { 729 struct in_addr laddr; 730 struct sockaddr_in sin; 731 732 inp->inp_options = (m) ? ip_srcroute(m) : NULL; 733 734 if (inp->inp_options == NULL) { 735 inp->inp_options = sc->sc_ipopts; 736 sc->sc_ipopts = NULL; 737 } 738 739 sin.sin_family = AF_INET; 740 sin.sin_len = sizeof(sin); 741 sin.sin_addr = sc->sc_inc.inc_faddr; 742 sin.sin_port = sc->sc_inc.inc_fport; 743 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 744 laddr = inp->inp_laddr; 745 if (inp->inp_laddr.s_addr == INADDR_ANY) 746 inp->inp_laddr = sc->sc_inc.inc_laddr; 747 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 748 thread0.td_ucred)) { 749 inp->inp_laddr = laddr; 750 goto abort; 751 } 752 } 753 tp = intotcpcb(inp); 754 tp->t_state = TCPS_SYN_RECEIVED; 755 tp->iss = sc->sc_iss; 756 tp->irs = sc->sc_irs; 757 tcp_rcvseqinit(tp); 758 tcp_sendseqinit(tp); 759 tp->snd_wl1 = sc->sc_irs; 760 tp->snd_max = tp->iss + 1; 761 tp->snd_nxt = tp->iss + 1; 762 tp->rcv_up = sc->sc_irs + 1; 763 tp->rcv_wnd = sc->sc_wnd; 764 tp->rcv_adv += tp->rcv_wnd; 765 tp->last_ack_sent = tp->rcv_nxt; 766 767 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 768 if (sc->sc_flags & SCF_NOOPT) 769 tp->t_flags |= TF_NOOPT; 770 else { 771 if (sc->sc_flags & SCF_WINSCALE) { 772 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 773 tp->snd_scale = sc->sc_requested_s_scale; 774 tp->request_r_scale = sc->sc_requested_r_scale; 775 } 776 if (sc->sc_flags & SCF_TIMESTAMP) { 777 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 778 tp->ts_recent = sc->sc_tsreflect; 779 tp->ts_recent_age = ticks; 780 tp->ts_offset = sc->sc_tsoff; 781 } 782 #ifdef TCP_SIGNATURE 783 if (sc->sc_flags & SCF_SIGNATURE) 784 tp->t_flags |= TF_SIGNATURE; 785 #endif 786 if (sc->sc_flags & SCF_SACK) 787 tp->t_flags |= TF_SACK_PERMIT; 788 } 789 790 if (sc->sc_flags & SCF_ECN) 791 tp->t_flags |= TF_ECN_PERMIT; 792 793 /* 794 * Set up MSS and get cached values from tcp_hostcache. 795 * This might overwrite some of the defaults we just set. 796 */ 797 tcp_mss(tp, sc->sc_peer_mss); 798 799 /* 800 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 801 */ 802 if (sc->sc_rxmits) 803 tp->snd_cwnd = tp->t_maxseg; 804 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 805 806 INP_WUNLOCK(inp); 807 808 TCPSTAT_INC(tcps_accepts); 809 return (so); 810 811 abort: 812 INP_WUNLOCK(inp); 813 abort2: 814 if (so != NULL) 815 soabort(so); 816 return (NULL); 817 } 818 819 /* 820 * This function gets called when we receive an ACK for a 821 * socket in the LISTEN state. We look up the connection 822 * in the syncache, and if its there, we pull it out of 823 * the cache and turn it into a full-blown connection in 824 * the SYN-RECEIVED state. 825 */ 826 int 827 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 828 struct socket **lsop, struct mbuf *m) 829 { 830 INIT_VNET_INET(curvnet); 831 struct syncache *sc; 832 struct syncache_head *sch; 833 struct syncache scs; 834 char *s; 835 836 /* 837 * Global TCP locks are held because we manipulate the PCB lists 838 * and create a new socket. 839 */ 840 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 841 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, 842 ("%s: can handle only ACK", __func__)); 843 844 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 845 SCH_LOCK_ASSERT(sch); 846 if (sc == NULL) { 847 /* 848 * There is no syncache entry, so see if this ACK is 849 * a returning syncookie. To do this, first: 850 * A. See if this socket has had a syncache entry dropped in 851 * the past. We don't want to accept a bogus syncookie 852 * if we've never received a SYN. 853 * B. check that the syncookie is valid. If it is, then 854 * cobble up a fake syncache entry, and return. 855 */ 856 if (!V_tcp_syncookies) { 857 SCH_UNLOCK(sch); 858 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 859 log(LOG_DEBUG, "%s; %s: Spurious ACK, " 860 "segment rejected (syncookies disabled)\n", 861 s, __func__); 862 goto failed; 863 } 864 bzero(&scs, sizeof(scs)); 865 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 866 SCH_UNLOCK(sch); 867 if (sc == NULL) { 868 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 869 log(LOG_DEBUG, "%s; %s: Segment failed " 870 "SYNCOOKIE authentication, segment rejected " 871 "(probably spoofed)\n", s, __func__); 872 goto failed; 873 } 874 } else { 875 /* Pull out the entry to unlock the bucket row. */ 876 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 877 sch->sch_length--; 878 V_tcp_syncache.cache_count--; 879 SCH_UNLOCK(sch); 880 } 881 882 /* 883 * Segment validation: 884 * ACK must match our initial sequence number + 1 (the SYN|ACK). 885 */ 886 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) { 887 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 888 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " 889 "rejected\n", s, __func__, th->th_ack, sc->sc_iss); 890 goto failed; 891 } 892 893 /* 894 * The SEQ must fall in the window starting at the received 895 * initial receive sequence number + 1 (the SYN). 896 */ 897 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) || 898 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) && 899 !TOEPCB_ISSET(sc)) { 900 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 901 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " 902 "rejected\n", s, __func__, th->th_seq, sc->sc_irs); 903 goto failed; 904 } 905 906 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { 907 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 908 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 909 "segment rejected\n", s, __func__); 910 goto failed; 911 } 912 /* 913 * If timestamps were negotiated the reflected timestamp 914 * must be equal to what we actually sent in the SYN|ACK. 915 */ 916 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts && 917 !TOEPCB_ISSET(sc)) { 918 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 919 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, " 920 "segment rejected\n", 921 s, __func__, to->to_tsecr, sc->sc_ts); 922 goto failed; 923 } 924 925 *lsop = syncache_socket(sc, *lsop, m); 926 927 if (*lsop == NULL) 928 TCPSTAT_INC(tcps_sc_aborted); 929 else 930 TCPSTAT_INC(tcps_sc_completed); 931 932 /* how do we find the inp for the new socket? */ 933 if (sc != &scs) 934 syncache_free(sc); 935 return (1); 936 failed: 937 if (sc != NULL && sc != &scs) 938 syncache_free(sc); 939 if (s != NULL) 940 free(s, M_TCPLOG); 941 *lsop = NULL; 942 return (0); 943 } 944 945 int 946 tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to, 947 struct tcphdr *th, struct socket **lsop, struct mbuf *m) 948 { 949 INIT_VNET_INET(curvnet); 950 int rc; 951 952 INP_INFO_WLOCK(&V_tcbinfo); 953 rc = syncache_expand(inc, to, th, lsop, m); 954 INP_INFO_WUNLOCK(&V_tcbinfo); 955 956 return (rc); 957 } 958 959 /* 960 * Given a LISTEN socket and an inbound SYN request, add 961 * this to the syn cache, and send back a segment: 962 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 963 * to the source. 964 * 965 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 966 * Doing so would require that we hold onto the data and deliver it 967 * to the application. However, if we are the target of a SYN-flood 968 * DoS attack, an attacker could send data which would eventually 969 * consume all available buffer space if it were ACKed. By not ACKing 970 * the data, we avoid this DoS scenario. 971 */ 972 static void 973 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 974 struct inpcb *inp, struct socket **lsop, struct mbuf *m, 975 struct toe_usrreqs *tu, void *toepcb) 976 { 977 INIT_VNET_INET(inp->inp_vnet); 978 struct tcpcb *tp; 979 struct socket *so; 980 struct syncache *sc = NULL; 981 struct syncache_head *sch; 982 struct mbuf *ipopts = NULL; 983 u_int32_t flowtmp; 984 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 985 char *s; 986 #ifdef INET6 987 int autoflowlabel = 0; 988 #endif 989 #ifdef MAC 990 struct label *maclabel; 991 #endif 992 struct syncache scs; 993 struct ucred *cred; 994 995 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 996 INP_WLOCK_ASSERT(inp); /* listen socket */ 997 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, 998 ("%s: unexpected tcp flags", __func__)); 999 1000 /* 1001 * Combine all so/tp operations very early to drop the INP lock as 1002 * soon as possible. 1003 */ 1004 so = *lsop; 1005 tp = sototcpcb(so); 1006 cred = crhold(so->so_cred); 1007 1008 #ifdef INET6 1009 if ((inc->inc_flags & INC_ISIPV6) && 1010 (inp->inp_flags & IN6P_AUTOFLOWLABEL)) 1011 autoflowlabel = 1; 1012 #endif 1013 ip_ttl = inp->inp_ip_ttl; 1014 ip_tos = inp->inp_ip_tos; 1015 win = sbspace(&so->so_rcv); 1016 sb_hiwat = so->so_rcv.sb_hiwat; 1017 noopt = (tp->t_flags & TF_NOOPT); 1018 1019 /* By the time we drop the lock these should no longer be used. */ 1020 so = NULL; 1021 tp = NULL; 1022 1023 #ifdef MAC 1024 if (mac_syncache_init(&maclabel) != 0) { 1025 INP_WUNLOCK(inp); 1026 INP_INFO_WUNLOCK(&V_tcbinfo); 1027 goto done; 1028 } else 1029 mac_syncache_create(maclabel, inp); 1030 #endif 1031 INP_WUNLOCK(inp); 1032 INP_INFO_WUNLOCK(&V_tcbinfo); 1033 1034 /* 1035 * Remember the IP options, if any. 1036 */ 1037 #ifdef INET6 1038 if (!(inc->inc_flags & INC_ISIPV6)) 1039 #endif 1040 ipopts = (m) ? ip_srcroute(m) : NULL; 1041 1042 /* 1043 * See if we already have an entry for this connection. 1044 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1045 * 1046 * XXX: should the syncache be re-initialized with the contents 1047 * of the new SYN here (which may have different options?) 1048 * 1049 * XXX: We do not check the sequence number to see if this is a 1050 * real retransmit or a new connection attempt. The question is 1051 * how to handle such a case; either ignore it as spoofed, or 1052 * drop the current entry and create a new one? 1053 */ 1054 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 1055 SCH_LOCK_ASSERT(sch); 1056 if (sc != NULL) { 1057 #ifndef TCP_OFFLOAD_DISABLE 1058 if (sc->sc_tu) 1059 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT, 1060 sc->sc_toepcb); 1061 #endif 1062 TCPSTAT_INC(tcps_sc_dupsyn); 1063 if (ipopts) { 1064 /* 1065 * If we were remembering a previous source route, 1066 * forget it and use the new one we've been given. 1067 */ 1068 if (sc->sc_ipopts) 1069 (void) m_free(sc->sc_ipopts); 1070 sc->sc_ipopts = ipopts; 1071 } 1072 /* 1073 * Update timestamp if present. 1074 */ 1075 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 1076 sc->sc_tsreflect = to->to_tsval; 1077 else 1078 sc->sc_flags &= ~SCF_TIMESTAMP; 1079 #ifdef MAC 1080 /* 1081 * Since we have already unconditionally allocated label 1082 * storage, free it up. The syncache entry will already 1083 * have an initialized label we can use. 1084 */ 1085 mac_syncache_destroy(&maclabel); 1086 #endif 1087 /* Retransmit SYN|ACK and reset retransmit count. */ 1088 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { 1089 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " 1090 "resetting timer and retransmitting SYN|ACK\n", 1091 s, __func__); 1092 free(s, M_TCPLOG); 1093 } 1094 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) { 1095 sc->sc_rxmits = 0; 1096 syncache_timeout(sc, sch, 1); 1097 TCPSTAT_INC(tcps_sndacks); 1098 TCPSTAT_INC(tcps_sndtotal); 1099 } 1100 SCH_UNLOCK(sch); 1101 goto done; 1102 } 1103 1104 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); 1105 if (sc == NULL) { 1106 /* 1107 * The zone allocator couldn't provide more entries. 1108 * Treat this as if the cache was full; drop the oldest 1109 * entry and insert the new one. 1110 */ 1111 TCPSTAT_INC(tcps_sc_zonefail); 1112 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 1113 syncache_drop(sc, sch); 1114 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); 1115 if (sc == NULL) { 1116 if (V_tcp_syncookies) { 1117 bzero(&scs, sizeof(scs)); 1118 sc = &scs; 1119 } else { 1120 SCH_UNLOCK(sch); 1121 if (ipopts) 1122 (void) m_free(ipopts); 1123 goto done; 1124 } 1125 } 1126 } 1127 1128 /* 1129 * Fill in the syncache values. 1130 */ 1131 #ifdef MAC 1132 sc->sc_label = maclabel; 1133 #endif 1134 sc->sc_cred = cred; 1135 cred = NULL; 1136 sc->sc_ipopts = ipopts; 1137 /* XXX-BZ this fib assignment is just useless. */ 1138 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum; 1139 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1140 #ifdef INET6 1141 if (!(inc->inc_flags & INC_ISIPV6)) 1142 #endif 1143 { 1144 sc->sc_ip_tos = ip_tos; 1145 sc->sc_ip_ttl = ip_ttl; 1146 } 1147 #ifndef TCP_OFFLOAD_DISABLE 1148 sc->sc_tu = tu; 1149 sc->sc_toepcb = toepcb; 1150 #endif 1151 sc->sc_irs = th->th_seq; 1152 sc->sc_iss = arc4random(); 1153 sc->sc_flags = 0; 1154 sc->sc_flowlabel = 0; 1155 1156 /* 1157 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1158 * win was derived from socket earlier in the function. 1159 */ 1160 win = imax(win, 0); 1161 win = imin(win, TCP_MAXWIN); 1162 sc->sc_wnd = win; 1163 1164 if (V_tcp_do_rfc1323) { 1165 /* 1166 * A timestamp received in a SYN makes 1167 * it ok to send timestamp requests and replies. 1168 */ 1169 if (to->to_flags & TOF_TS) { 1170 sc->sc_tsreflect = to->to_tsval; 1171 sc->sc_ts = ticks; 1172 sc->sc_flags |= SCF_TIMESTAMP; 1173 } 1174 if (to->to_flags & TOF_SCALE) { 1175 int wscale = 0; 1176 1177 /* 1178 * Pick the smallest possible scaling factor that 1179 * will still allow us to scale up to sb_max, aka 1180 * kern.ipc.maxsockbuf. 1181 * 1182 * We do this because there are broken firewalls that 1183 * will corrupt the window scale option, leading to 1184 * the other endpoint believing that our advertised 1185 * window is unscaled. At scale factors larger than 1186 * 5 the unscaled window will drop below 1500 bytes, 1187 * leading to serious problems when traversing these 1188 * broken firewalls. 1189 * 1190 * With the default maxsockbuf of 256K, a scale factor 1191 * of 3 will be chosen by this algorithm. Those who 1192 * choose a larger maxsockbuf should watch out 1193 * for the compatiblity problems mentioned above. 1194 * 1195 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1196 * or <SYN,ACK>) segment itself is never scaled. 1197 */ 1198 while (wscale < TCP_MAX_WINSHIFT && 1199 (TCP_MAXWIN << wscale) < sb_max) 1200 wscale++; 1201 sc->sc_requested_r_scale = wscale; 1202 sc->sc_requested_s_scale = to->to_wscale; 1203 sc->sc_flags |= SCF_WINSCALE; 1204 } 1205 } 1206 #ifdef TCP_SIGNATURE 1207 /* 1208 * If listening socket requested TCP digests, and received SYN 1209 * contains the option, flag this in the syncache so that 1210 * syncache_respond() will do the right thing with the SYN+ACK. 1211 * XXX: Currently we always record the option by default and will 1212 * attempt to use it in syncache_respond(). 1213 */ 1214 if (to->to_flags & TOF_SIGNATURE) 1215 sc->sc_flags |= SCF_SIGNATURE; 1216 #endif 1217 if (to->to_flags & TOF_SACKPERM) 1218 sc->sc_flags |= SCF_SACK; 1219 if (to->to_flags & TOF_MSS) 1220 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1221 if (noopt) 1222 sc->sc_flags |= SCF_NOOPT; 1223 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn) 1224 sc->sc_flags |= SCF_ECN; 1225 1226 if (V_tcp_syncookies) { 1227 syncookie_generate(sch, sc, &flowtmp); 1228 #ifdef INET6 1229 if (autoflowlabel) 1230 sc->sc_flowlabel = flowtmp; 1231 #endif 1232 } else { 1233 #ifdef INET6 1234 if (autoflowlabel) 1235 sc->sc_flowlabel = 1236 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1237 #endif 1238 } 1239 SCH_UNLOCK(sch); 1240 1241 /* 1242 * Do a standard 3-way handshake. 1243 */ 1244 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) { 1245 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs) 1246 syncache_free(sc); 1247 else if (sc != &scs) 1248 syncache_insert(sc, sch); /* locks and unlocks sch */ 1249 TCPSTAT_INC(tcps_sndacks); 1250 TCPSTAT_INC(tcps_sndtotal); 1251 } else { 1252 if (sc != &scs) 1253 syncache_free(sc); 1254 TCPSTAT_INC(tcps_sc_dropped); 1255 } 1256 1257 done: 1258 if (cred != NULL) 1259 crfree(cred); 1260 #ifdef MAC 1261 if (sc == &scs) 1262 mac_syncache_destroy(&maclabel); 1263 #endif 1264 if (m) { 1265 1266 *lsop = NULL; 1267 m_freem(m); 1268 } 1269 } 1270 1271 static int 1272 syncache_respond(struct syncache *sc) 1273 { 1274 INIT_VNET_INET(curvnet); 1275 struct ip *ip = NULL; 1276 struct mbuf *m; 1277 struct tcphdr *th; 1278 int optlen, error; 1279 u_int16_t hlen, tlen, mssopt; 1280 struct tcpopt to; 1281 #ifdef INET6 1282 struct ip6_hdr *ip6 = NULL; 1283 #endif 1284 1285 hlen = 1286 #ifdef INET6 1287 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) : 1288 #endif 1289 sizeof(struct ip); 1290 tlen = hlen + sizeof(struct tcphdr); 1291 1292 /* Determine MSS we advertize to other end of connection. */ 1293 mssopt = tcp_mssopt(&sc->sc_inc); 1294 if (sc->sc_peer_mss) 1295 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss); 1296 1297 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1298 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, 1299 ("syncache: mbuf too small")); 1300 1301 /* Create the IP+TCP header from scratch. */ 1302 m = m_gethdr(M_DONTWAIT, MT_DATA); 1303 if (m == NULL) 1304 return (ENOBUFS); 1305 #ifdef MAC 1306 mac_syncache_create_mbuf(sc->sc_label, m); 1307 #endif 1308 m->m_data += max_linkhdr; 1309 m->m_len = tlen; 1310 m->m_pkthdr.len = tlen; 1311 m->m_pkthdr.rcvif = NULL; 1312 1313 #ifdef INET6 1314 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 1315 ip6 = mtod(m, struct ip6_hdr *); 1316 ip6->ip6_vfc = IPV6_VERSION; 1317 ip6->ip6_nxt = IPPROTO_TCP; 1318 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1319 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1320 ip6->ip6_plen = htons(tlen - hlen); 1321 /* ip6_hlim is set after checksum */ 1322 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1323 ip6->ip6_flow |= sc->sc_flowlabel; 1324 1325 th = (struct tcphdr *)(ip6 + 1); 1326 } else 1327 #endif 1328 { 1329 ip = mtod(m, struct ip *); 1330 ip->ip_v = IPVERSION; 1331 ip->ip_hl = sizeof(struct ip) >> 2; 1332 ip->ip_len = tlen; 1333 ip->ip_id = 0; 1334 ip->ip_off = 0; 1335 ip->ip_sum = 0; 1336 ip->ip_p = IPPROTO_TCP; 1337 ip->ip_src = sc->sc_inc.inc_laddr; 1338 ip->ip_dst = sc->sc_inc.inc_faddr; 1339 ip->ip_ttl = sc->sc_ip_ttl; 1340 ip->ip_tos = sc->sc_ip_tos; 1341 1342 /* 1343 * See if we should do MTU discovery. Route lookups are 1344 * expensive, so we will only unset the DF bit if: 1345 * 1346 * 1) path_mtu_discovery is disabled 1347 * 2) the SCF_UNREACH flag has been set 1348 */ 1349 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1350 ip->ip_off |= IP_DF; 1351 1352 th = (struct tcphdr *)(ip + 1); 1353 } 1354 th->th_sport = sc->sc_inc.inc_lport; 1355 th->th_dport = sc->sc_inc.inc_fport; 1356 1357 th->th_seq = htonl(sc->sc_iss); 1358 th->th_ack = htonl(sc->sc_irs + 1); 1359 th->th_off = sizeof(struct tcphdr) >> 2; 1360 th->th_x2 = 0; 1361 th->th_flags = TH_SYN|TH_ACK; 1362 th->th_win = htons(sc->sc_wnd); 1363 th->th_urp = 0; 1364 1365 if (sc->sc_flags & SCF_ECN) { 1366 th->th_flags |= TH_ECE; 1367 TCPSTAT_INC(tcps_ecn_shs); 1368 } 1369 1370 /* Tack on the TCP options. */ 1371 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1372 to.to_flags = 0; 1373 1374 to.to_mss = mssopt; 1375 to.to_flags = TOF_MSS; 1376 if (sc->sc_flags & SCF_WINSCALE) { 1377 to.to_wscale = sc->sc_requested_r_scale; 1378 to.to_flags |= TOF_SCALE; 1379 } 1380 if (sc->sc_flags & SCF_TIMESTAMP) { 1381 /* Virgin timestamp or TCP cookie enhanced one. */ 1382 to.to_tsval = sc->sc_ts; 1383 to.to_tsecr = sc->sc_tsreflect; 1384 to.to_flags |= TOF_TS; 1385 } 1386 if (sc->sc_flags & SCF_SACK) 1387 to.to_flags |= TOF_SACKPERM; 1388 #ifdef TCP_SIGNATURE 1389 if (sc->sc_flags & SCF_SIGNATURE) 1390 to.to_flags |= TOF_SIGNATURE; 1391 #endif 1392 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1393 1394 /* Adjust headers by option size. */ 1395 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1396 m->m_len += optlen; 1397 m->m_pkthdr.len += optlen; 1398 1399 #ifdef TCP_SIGNATURE 1400 if (sc->sc_flags & SCF_SIGNATURE) 1401 tcp_signature_compute(m, 0, 0, optlen, 1402 to.to_signature, IPSEC_DIR_OUTBOUND); 1403 #endif 1404 #ifdef INET6 1405 if (sc->sc_inc.inc_flags & INC_ISIPV6) 1406 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1407 else 1408 #endif 1409 ip->ip_len += optlen; 1410 } else 1411 optlen = 0; 1412 1413 #ifdef INET6 1414 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 1415 th->th_sum = 0; 1416 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1417 tlen + optlen - hlen); 1418 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1419 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1420 } else 1421 #endif 1422 { 1423 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1424 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1425 m->m_pkthdr.csum_flags = CSUM_TCP; 1426 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1427 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1428 } 1429 return (error); 1430 } 1431 1432 void 1433 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 1434 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 1435 { 1436 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL); 1437 } 1438 1439 void 1440 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to, 1441 struct tcphdr *th, struct inpcb *inp, struct socket **lsop, 1442 struct toe_usrreqs *tu, void *toepcb) 1443 { 1444 INIT_VNET_INET(curvnet); 1445 1446 INP_INFO_WLOCK(&V_tcbinfo); 1447 INP_WLOCK(inp); 1448 _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb); 1449 } 1450 1451 /* 1452 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1453 * receive and to be able to handle SYN floods from bogus source addresses 1454 * (where we will never receive any reply). SYN floods try to exhaust all 1455 * our memory and available slots in the SYN cache table to cause a denial 1456 * of service to legitimate users of the local host. 1457 * 1458 * The idea of SYN cookies is to encode and include all necessary information 1459 * about the connection setup state within the SYN-ACK we send back and thus 1460 * to get along without keeping any local state until the ACK to the SYN-ACK 1461 * arrives (if ever). Everything we need to know should be available from 1462 * the information we encoded in the SYN-ACK. 1463 * 1464 * More information about the theory behind SYN cookies and its first 1465 * discussion and specification can be found at: 1466 * http://cr.yp.to/syncookies.html (overview) 1467 * http://cr.yp.to/syncookies/archive (gory details) 1468 * 1469 * This implementation extends the orginal idea and first implementation 1470 * of FreeBSD by using not only the initial sequence number field to store 1471 * information but also the timestamp field if present. This way we can 1472 * keep track of the entire state we need to know to recreate the session in 1473 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1474 * these days. For those that do not we still have to live with the known 1475 * shortcomings of the ISN only SYN cookies. 1476 * 1477 * Cookie layers: 1478 * 1479 * Initial sequence number we send: 1480 * 31|................................|0 1481 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1482 * D = MD5 Digest (first dword) 1483 * M = MSS index 1484 * R = Rotation of secret 1485 * P = Odd or Even secret 1486 * 1487 * The MD5 Digest is computed with over following parameters: 1488 * a) randomly rotated secret 1489 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1490 * c) the received initial sequence number from remote host 1491 * d) the rotation offset and odd/even bit 1492 * 1493 * Timestamp we send: 1494 * 31|................................|0 1495 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1496 * D = MD5 Digest (third dword) (only as filler) 1497 * S = Requested send window scale 1498 * R = Requested receive window scale 1499 * A = SACK allowed 1500 * 5 = TCP-MD5 enabled (not implemented yet) 1501 * XORed with MD5 Digest (forth dword) 1502 * 1503 * The timestamp isn't cryptographically secure and doesn't need to be. 1504 * The double use of the MD5 digest dwords ties it to a specific remote/ 1505 * local host/port, remote initial sequence number and our local time 1506 * limited secret. A received timestamp is reverted (XORed) and then 1507 * the contained MD5 dword is compared to the computed one to ensure the 1508 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1509 * have been tampered with but this isn't different from supplying bogus 1510 * values in the SYN in the first place. 1511 * 1512 * Some problems with SYN cookies remain however: 1513 * Consider the problem of a recreated (and retransmitted) cookie. If the 1514 * original SYN was accepted, the connection is established. The second 1515 * SYN is inflight, and if it arrives with an ISN that falls within the 1516 * receive window, the connection is killed. 1517 * 1518 * Notes: 1519 * A heuristic to determine when to accept syn cookies is not necessary. 1520 * An ACK flood would cause the syncookie verification to be attempted, 1521 * but a SYN flood causes syncookies to be generated. Both are of equal 1522 * cost, so there's no point in trying to optimize the ACK flood case. 1523 * Also, if you don't process certain ACKs for some reason, then all someone 1524 * would have to do is launch a SYN and ACK flood at the same time, which 1525 * would stop cookie verification and defeat the entire purpose of syncookies. 1526 */ 1527 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1528 1529 static void 1530 syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1531 u_int32_t *flowlabel) 1532 { 1533 INIT_VNET_INET(curvnet); 1534 MD5_CTX ctx; 1535 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1536 u_int32_t data; 1537 u_int32_t *secbits; 1538 u_int off, pmss, mss; 1539 int i; 1540 1541 SCH_LOCK_ASSERT(sch); 1542 1543 /* Which of the two secrets to use. */ 1544 secbits = sch->sch_oddeven ? 1545 sch->sch_secbits_odd : sch->sch_secbits_even; 1546 1547 /* Reseed secret if too old. */ 1548 if (sch->sch_reseed < time_uptime) { 1549 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1550 secbits = sch->sch_oddeven ? 1551 sch->sch_secbits_odd : sch->sch_secbits_even; 1552 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1553 secbits[i] = arc4random(); 1554 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1555 } 1556 1557 /* Secret rotation offset. */ 1558 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1559 1560 /* Maximum segment size calculation. */ 1561 pmss = 1562 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss); 1563 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1564 if (tcp_sc_msstab[mss] <= pmss) 1565 break; 1566 1567 /* Fold parameters and MD5 digest into the ISN we will send. */ 1568 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1569 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1570 data |= mss << 4; /* mss, 3 bits */ 1571 1572 MD5Init(&ctx); 1573 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1574 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1575 MD5Update(&ctx, secbits, off); 1576 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1577 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1578 MD5Update(&ctx, &data, sizeof(data)); 1579 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1580 1581 data |= (md5_buffer[0] << 7); 1582 sc->sc_iss = data; 1583 1584 #ifdef INET6 1585 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1586 #endif 1587 1588 /* Additional parameters are stored in the timestamp if present. */ 1589 if (sc->sc_flags & SCF_TIMESTAMP) { 1590 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1591 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1592 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1593 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1594 data |= md5_buffer[2] << 10; /* more digest bits */ 1595 data ^= md5_buffer[3]; 1596 sc->sc_ts = data; 1597 sc->sc_tsoff = data - ticks; /* after XOR */ 1598 } 1599 1600 TCPSTAT_INC(tcps_sc_sendcookie); 1601 } 1602 1603 static struct syncache * 1604 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1605 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1606 struct socket *so) 1607 { 1608 INIT_VNET_INET(curvnet); 1609 MD5_CTX ctx; 1610 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1611 u_int32_t data = 0; 1612 u_int32_t *secbits; 1613 tcp_seq ack, seq; 1614 int off, mss, wnd, flags; 1615 1616 SCH_LOCK_ASSERT(sch); 1617 1618 /* 1619 * Pull information out of SYN-ACK/ACK and 1620 * revert sequence number advances. 1621 */ 1622 ack = th->th_ack - 1; 1623 seq = th->th_seq - 1; 1624 off = (ack >> 1) & 0x7; 1625 mss = (ack >> 4) & 0x7; 1626 flags = ack & 0x7f; 1627 1628 /* Which of the two secrets to use. */ 1629 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1630 1631 /* 1632 * The secret wasn't updated for the lifetime of a syncookie, 1633 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1634 */ 1635 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) { 1636 return (NULL); 1637 } 1638 1639 /* Recompute the digest so we can compare it. */ 1640 MD5Init(&ctx); 1641 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1642 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1643 MD5Update(&ctx, secbits, off); 1644 MD5Update(&ctx, inc, sizeof(*inc)); 1645 MD5Update(&ctx, &seq, sizeof(seq)); 1646 MD5Update(&ctx, &flags, sizeof(flags)); 1647 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1648 1649 /* Does the digest part of or ACK'ed ISS match? */ 1650 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1651 return (NULL); 1652 1653 /* Does the digest part of our reflected timestamp match? */ 1654 if (to->to_flags & TOF_TS) { 1655 data = md5_buffer[3] ^ to->to_tsecr; 1656 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1657 return (NULL); 1658 } 1659 1660 /* Fill in the syncache values. */ 1661 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1662 sc->sc_ipopts = NULL; 1663 1664 sc->sc_irs = seq; 1665 sc->sc_iss = ack; 1666 1667 #ifdef INET6 1668 if (inc->inc_flags & INC_ISIPV6) { 1669 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL) 1670 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1671 } else 1672 #endif 1673 { 1674 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1675 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1676 } 1677 1678 /* Additional parameters that were encoded in the timestamp. */ 1679 if (data) { 1680 sc->sc_flags |= SCF_TIMESTAMP; 1681 sc->sc_tsreflect = to->to_tsval; 1682 sc->sc_ts = to->to_tsecr; 1683 sc->sc_tsoff = to->to_tsecr - ticks; 1684 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1685 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1686 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1687 TCP_MAX_WINSHIFT); 1688 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1689 TCP_MAX_WINSHIFT); 1690 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1691 sc->sc_flags |= SCF_WINSCALE; 1692 } else 1693 sc->sc_flags |= SCF_NOOPT; 1694 1695 wnd = sbspace(&so->so_rcv); 1696 wnd = imax(wnd, 0); 1697 wnd = imin(wnd, TCP_MAXWIN); 1698 sc->sc_wnd = wnd; 1699 1700 sc->sc_rxmits = 0; 1701 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1702 1703 TCPSTAT_INC(tcps_sc_recvcookie); 1704 return (sc); 1705 } 1706 1707 /* 1708 * Returns the current number of syncache entries. This number 1709 * will probably change before you get around to calling 1710 * syncache_pcblist. 1711 */ 1712 1713 int 1714 syncache_pcbcount(void) 1715 { 1716 INIT_VNET_INET(curvnet); 1717 struct syncache_head *sch; 1718 int count, i; 1719 1720 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { 1721 /* No need to lock for a read. */ 1722 sch = &V_tcp_syncache.hashbase[i]; 1723 count += sch->sch_length; 1724 } 1725 return count; 1726 } 1727 1728 /* 1729 * Exports the syncache entries to userland so that netstat can display 1730 * them alongside the other sockets. This function is intended to be 1731 * called only from tcp_pcblist. 1732 * 1733 * Due to concurrency on an active system, the number of pcbs exported 1734 * may have no relation to max_pcbs. max_pcbs merely indicates the 1735 * amount of space the caller allocated for this function to use. 1736 */ 1737 int 1738 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported) 1739 { 1740 INIT_VNET_INET(curvnet); 1741 struct xtcpcb xt; 1742 struct syncache *sc; 1743 struct syncache_head *sch; 1744 int count, error, i; 1745 1746 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { 1747 sch = &V_tcp_syncache.hashbase[i]; 1748 SCH_LOCK(sch); 1749 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 1750 if (count >= max_pcbs) { 1751 SCH_UNLOCK(sch); 1752 goto exit; 1753 } 1754 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0) 1755 continue; 1756 bzero(&xt, sizeof(xt)); 1757 xt.xt_len = sizeof(xt); 1758 if (sc->sc_inc.inc_flags & INC_ISIPV6) 1759 xt.xt_inp.inp_vflag = INP_IPV6; 1760 else 1761 xt.xt_inp.inp_vflag = INP_IPV4; 1762 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); 1763 xt.xt_tp.t_inpcb = &xt.xt_inp; 1764 xt.xt_tp.t_state = TCPS_SYN_RECEIVED; 1765 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1766 xt.xt_socket.xso_len = sizeof (struct xsocket); 1767 xt.xt_socket.so_type = SOCK_STREAM; 1768 xt.xt_socket.so_state = SS_ISCONNECTING; 1769 error = SYSCTL_OUT(req, &xt, sizeof xt); 1770 if (error) { 1771 SCH_UNLOCK(sch); 1772 goto exit; 1773 } 1774 count++; 1775 } 1776 SCH_UNLOCK(sch); 1777 } 1778 exit: 1779 *pcbs_exported = count; 1780 return error; 1781 } 1782