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 57 #include <vm/uma.h> 58 59 #include <net/if.h> 60 #include <net/route.h> 61 #include <net/vnet.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 88 #ifdef IPSEC 89 #include <netipsec/ipsec.h> 90 #ifdef INET6 91 #include <netipsec/ipsec6.h> 92 #endif 93 #include <netipsec/key.h> 94 #endif /*IPSEC*/ 95 96 #include <machine/in_cksum.h> 97 98 #include <security/mac/mac_framework.h> 99 100 static VNET_DEFINE(int, tcp_syncookies) = 1; 101 #define V_tcp_syncookies VNET(tcp_syncookies) 102 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 103 &VNET_NAME(tcp_syncookies), 0, 104 "Use TCP SYN cookies if the syncache overflows"); 105 106 static VNET_DEFINE(int, tcp_syncookiesonly) = 0; 107 #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly) 108 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW, 109 &VNET_NAME(tcp_syncookiesonly), 0, 110 "Use only TCP SYN cookies"); 111 112 #ifdef TCP_OFFLOAD_DISABLE 113 #define TOEPCB_ISSET(sc) (0) 114 #else 115 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL) 116 #endif 117 118 static void syncache_drop(struct syncache *, struct syncache_head *); 119 static void syncache_free(struct syncache *); 120 static void syncache_insert(struct syncache *, struct syncache_head *); 121 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 122 static int syncache_respond(struct syncache *); 123 static struct socket *syncache_socket(struct syncache *, struct socket *, 124 struct mbuf *m); 125 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, 126 int docallout); 127 static void syncache_timer(void *); 128 static void syncookie_generate(struct syncache_head *, struct syncache *, 129 u_int32_t *); 130 static struct syncache 131 *syncookie_lookup(struct in_conninfo *, struct syncache_head *, 132 struct syncache *, struct tcpopt *, struct tcphdr *, 133 struct socket *); 134 135 /* 136 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 137 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds, 138 * the odds are that the user has given up attempting to connect by then. 139 */ 140 #define SYNCACHE_MAXREXMTS 3 141 142 /* Arbitrary values */ 143 #define TCP_SYNCACHE_HASHSIZE 512 144 #define TCP_SYNCACHE_BUCKETLIMIT 30 145 146 static VNET_DEFINE(struct tcp_syncache, tcp_syncache); 147 #define V_tcp_syncache VNET(tcp_syncache) 148 149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 150 151 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 152 &VNET_NAME(tcp_syncache.bucket_limit), 0, 153 "Per-bucket hash limit for syncache"); 154 155 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 156 &VNET_NAME(tcp_syncache.cache_limit), 0, 157 "Overall entry limit for syncache"); 158 159 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 160 &VNET_NAME(tcp_syncache.cache_count), 0, 161 "Current number of entries in syncache"); 162 163 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 164 &VNET_NAME(tcp_syncache.hashsize), 0, 165 "Size of TCP syncache hashtable"); 166 167 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 168 &VNET_NAME(tcp_syncache.rexmt_limit), 0, 169 "Limit on SYN/ACK retransmissions"); 170 171 VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1; 172 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, 173 CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0, 174 "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 { 209 210 if (sc->sc_ipopts) 211 (void) m_free(sc->sc_ipopts); 212 if (sc->sc_cred) 213 crfree(sc->sc_cred); 214 #ifdef MAC 215 mac_syncache_destroy(&sc->sc_label); 216 #endif 217 218 uma_zfree(V_tcp_syncache.zone, sc); 219 } 220 221 void 222 syncache_init(void) 223 { 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 * 251 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO); 252 253 /* Initialize the hash buckets. */ 254 for (i = 0; i < V_tcp_syncache.hashsize; i++) { 255 #ifdef VIMAGE 256 V_tcp_syncache.hashbase[i].sch_vnet = curvnet; 257 #endif 258 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket); 259 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", 260 NULL, MTX_DEF); 261 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer, 262 &V_tcp_syncache.hashbase[i].sch_mtx, 0); 263 V_tcp_syncache.hashbase[i].sch_length = 0; 264 } 265 266 /* Create the syncache entry zone. */ 267 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), 268 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 269 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit); 270 } 271 272 #ifdef VIMAGE 273 void 274 syncache_destroy(void) 275 { 276 struct syncache_head *sch; 277 struct syncache *sc, *nsc; 278 int i; 279 280 /* Cleanup hash buckets: stop timers, free entries, destroy locks. */ 281 for (i = 0; i < V_tcp_syncache.hashsize; i++) { 282 283 sch = &V_tcp_syncache.hashbase[i]; 284 callout_drain(&sch->sch_timer); 285 286 SCH_LOCK(sch); 287 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) 288 syncache_drop(sc, sch); 289 SCH_UNLOCK(sch); 290 KASSERT(TAILQ_EMPTY(&sch->sch_bucket), 291 ("%s: sch->sch_bucket not empty", __func__)); 292 KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0", 293 __func__, sch->sch_length)); 294 mtx_destroy(&sch->sch_mtx); 295 } 296 297 KASSERT(V_tcp_syncache.cache_count == 0, ("%s: cache_count %d not 0", 298 __func__, V_tcp_syncache.cache_count)); 299 300 /* Free the allocated global resources. */ 301 uma_zdestroy(V_tcp_syncache.zone); 302 free(V_tcp_syncache.hashbase, M_SYNCACHE); 303 } 304 #endif 305 306 /* 307 * Inserts a syncache entry into the specified bucket row. 308 * Locks and unlocks the syncache_head autonomously. 309 */ 310 static void 311 syncache_insert(struct syncache *sc, struct syncache_head *sch) 312 { 313 struct syncache *sc2; 314 315 SCH_LOCK(sch); 316 317 /* 318 * Make sure that we don't overflow the per-bucket limit. 319 * If the bucket is full, toss the oldest element. 320 */ 321 if (sch->sch_length >= V_tcp_syncache.bucket_limit) { 322 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), 323 ("sch->sch_length incorrect")); 324 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); 325 syncache_drop(sc2, sch); 326 TCPSTAT_INC(tcps_sc_bucketoverflow); 327 } 328 329 /* Put it into the bucket. */ 330 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); 331 sch->sch_length++; 332 333 /* Reinitialize the bucket row's timer. */ 334 if (sch->sch_length == 1) 335 sch->sch_nextc = ticks + INT_MAX; 336 syncache_timeout(sc, sch, 1); 337 338 SCH_UNLOCK(sch); 339 340 V_tcp_syncache.cache_count++; 341 TCPSTAT_INC(tcps_sc_added); 342 } 343 344 /* 345 * Remove and free entry from syncache bucket row. 346 * Expects locked syncache head. 347 */ 348 static void 349 syncache_drop(struct syncache *sc, struct syncache_head *sch) 350 { 351 352 SCH_LOCK_ASSERT(sch); 353 354 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 355 sch->sch_length--; 356 357 #ifndef TCP_OFFLOAD_DISABLE 358 if (sc->sc_tu) 359 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb); 360 #endif 361 syncache_free(sc); 362 V_tcp_syncache.cache_count--; 363 } 364 365 /* 366 * Engage/reengage time on bucket row. 367 */ 368 static void 369 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) 370 { 371 sc->sc_rxttime = ticks + 372 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]); 373 sc->sc_rxmits++; 374 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) { 375 sch->sch_nextc = sc->sc_rxttime; 376 if (docallout) 377 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, 378 syncache_timer, (void *)sch); 379 } 380 } 381 382 /* 383 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 384 * If we have retransmitted an entry the maximum number of times, expire it. 385 * One separate timer for each bucket row. 386 */ 387 static void 388 syncache_timer(void *xsch) 389 { 390 struct syncache_head *sch = (struct syncache_head *)xsch; 391 struct syncache *sc, *nsc; 392 int tick = ticks; 393 char *s; 394 395 CURVNET_SET(sch->sch_vnet); 396 397 /* NB: syncache_head has already been locked by the callout. */ 398 SCH_LOCK_ASSERT(sch); 399 400 /* 401 * In the following cycle we may remove some entries and/or 402 * advance some timeouts, so re-initialize the bucket timer. 403 */ 404 sch->sch_nextc = tick + INT_MAX; 405 406 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { 407 /* 408 * We do not check if the listen socket still exists 409 * and accept the case where the listen socket may be 410 * gone by the time we resend the SYN/ACK. We do 411 * not expect this to happens often. If it does, 412 * then the RST will be sent by the time the remote 413 * host does the SYN/ACK->ACK. 414 */ 415 if (TSTMP_GT(sc->sc_rxttime, tick)) { 416 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) 417 sch->sch_nextc = sc->sc_rxttime; 418 continue; 419 } 420 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) { 421 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 422 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " 423 "giving up and removing syncache entry\n", 424 s, __func__); 425 free(s, M_TCPLOG); 426 } 427 syncache_drop(sc, sch); 428 TCPSTAT_INC(tcps_sc_stale); 429 continue; 430 } 431 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 432 log(LOG_DEBUG, "%s; %s: Response timeout, " 433 "retransmitting (%u) SYN|ACK\n", 434 s, __func__, sc->sc_rxmits); 435 free(s, M_TCPLOG); 436 } 437 438 (void) syncache_respond(sc); 439 TCPSTAT_INC(tcps_sc_retransmitted); 440 syncache_timeout(sc, sch, 0); 441 } 442 if (!TAILQ_EMPTY(&(sch)->sch_bucket)) 443 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, 444 syncache_timer, (void *)(sch)); 445 CURVNET_RESTORE(); 446 } 447 448 /* 449 * Find an entry in the syncache. 450 * Returns always with locked syncache_head plus a matching entry or NULL. 451 */ 452 struct syncache * 453 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 454 { 455 struct syncache *sc; 456 struct syncache_head *sch; 457 458 #ifdef INET6 459 if (inc->inc_flags & INC_ISIPV6) { 460 sch = &V_tcp_syncache.hashbase[ 461 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)]; 462 *schp = sch; 463 464 SCH_LOCK(sch); 465 466 /* Circle through bucket row to find matching entry. */ 467 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 468 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 469 return (sc); 470 } 471 } else 472 #endif 473 { 474 sch = &V_tcp_syncache.hashbase[ 475 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)]; 476 *schp = sch; 477 478 SCH_LOCK(sch); 479 480 /* Circle through bucket row to find matching entry. */ 481 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 482 #ifdef INET6 483 if (sc->sc_inc.inc_flags & INC_ISIPV6) 484 continue; 485 #endif 486 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 487 return (sc); 488 } 489 } 490 SCH_LOCK_ASSERT(*schp); 491 return (NULL); /* always returns with locked sch */ 492 } 493 494 /* 495 * This function is called when we get a RST for a 496 * non-existent connection, so that we can see if the 497 * connection is in the syn cache. If it is, zap it. 498 */ 499 void 500 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 501 { 502 struct syncache *sc; 503 struct syncache_head *sch; 504 char *s = NULL; 505 506 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 507 SCH_LOCK_ASSERT(sch); 508 509 /* 510 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. 511 * See RFC 793 page 65, section SEGMENT ARRIVES. 512 */ 513 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { 514 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 515 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " 516 "FIN flag set, segment ignored\n", s, __func__); 517 TCPSTAT_INC(tcps_badrst); 518 goto done; 519 } 520 521 /* 522 * No corresponding connection was found in syncache. 523 * If syncookies are enabled and possibly exclusively 524 * used, or we are under memory pressure, a valid RST 525 * may not find a syncache entry. In that case we're 526 * done and no SYN|ACK retransmissions will happen. 527 * Otherwise the the RST was misdirected or spoofed. 528 */ 529 if (sc == NULL) { 530 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 531 log(LOG_DEBUG, "%s; %s: Spurious RST without matching " 532 "syncache entry (possibly syncookie only), " 533 "segment ignored\n", s, __func__); 534 TCPSTAT_INC(tcps_badrst); 535 goto done; 536 } 537 538 /* 539 * If the RST bit is set, check the sequence number to see 540 * if this is a valid reset segment. 541 * RFC 793 page 37: 542 * In all states except SYN-SENT, all reset (RST) segments 543 * are validated by checking their SEQ-fields. A reset is 544 * valid if its sequence number is in the window. 545 * 546 * The sequence number in the reset segment is normally an 547 * echo of our outgoing acknowlegement numbers, but some hosts 548 * send a reset with the sequence number at the rightmost edge 549 * of our receive window, and we have to handle this case. 550 */ 551 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 552 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 553 syncache_drop(sc, sch); 554 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 555 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " 556 "connection attempt aborted by remote endpoint\n", 557 s, __func__); 558 TCPSTAT_INC(tcps_sc_reset); 559 } else { 560 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 561 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != " 562 "IRS %u (+WND %u), segment ignored\n", 563 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd); 564 TCPSTAT_INC(tcps_badrst); 565 } 566 567 done: 568 if (s != NULL) 569 free(s, M_TCPLOG); 570 SCH_UNLOCK(sch); 571 } 572 573 void 574 syncache_badack(struct in_conninfo *inc) 575 { 576 struct syncache *sc; 577 struct syncache_head *sch; 578 579 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 580 SCH_LOCK_ASSERT(sch); 581 if (sc != NULL) { 582 syncache_drop(sc, sch); 583 TCPSTAT_INC(tcps_sc_badack); 584 } 585 SCH_UNLOCK(sch); 586 } 587 588 void 589 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 590 { 591 struct syncache *sc; 592 struct syncache_head *sch; 593 594 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 595 SCH_LOCK_ASSERT(sch); 596 if (sc == NULL) 597 goto done; 598 599 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 600 if (ntohl(th->th_seq) != sc->sc_iss) 601 goto done; 602 603 /* 604 * If we've rertransmitted 3 times and this is our second error, 605 * we remove the entry. Otherwise, we allow it to continue on. 606 * This prevents us from incorrectly nuking an entry during a 607 * spurious network outage. 608 * 609 * See tcp_notify(). 610 */ 611 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { 612 sc->sc_flags |= SCF_UNREACH; 613 goto done; 614 } 615 syncache_drop(sc, sch); 616 TCPSTAT_INC(tcps_sc_unreach); 617 done: 618 SCH_UNLOCK(sch); 619 } 620 621 /* 622 * Build a new TCP socket structure from a syncache entry. 623 */ 624 static struct socket * 625 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 626 { 627 struct inpcb *inp = NULL; 628 struct socket *so; 629 struct tcpcb *tp; 630 char *s; 631 632 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 633 634 /* 635 * Ok, create the full blown connection, and set things up 636 * as they would have been set up if we had created the 637 * connection when the SYN arrived. If we can't create 638 * the connection, abort it. 639 */ 640 so = sonewconn(lso, SS_ISCONNECTED); 641 if (so == NULL) { 642 /* 643 * Drop the connection; we will either send a RST or 644 * have the peer retransmit its SYN again after its 645 * RTO and try again. 646 */ 647 TCPSTAT_INC(tcps_listendrop); 648 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { 649 log(LOG_DEBUG, "%s; %s: Socket create failed " 650 "due to limits or memory shortage\n", 651 s, __func__); 652 free(s, M_TCPLOG); 653 } 654 goto abort2; 655 } 656 #ifdef MAC 657 mac_socketpeer_set_from_mbuf(m, so); 658 #endif 659 660 inp = sotoinpcb(so); 661 inp->inp_inc.inc_fibnum = so->so_fibnum; 662 INP_WLOCK(inp); 663 664 /* Insert new socket into PCB hash list. */ 665 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags; 666 #ifdef INET6 667 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 668 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 669 } else { 670 inp->inp_vflag &= ~INP_IPV6; 671 inp->inp_vflag |= INP_IPV4; 672 #endif 673 inp->inp_laddr = sc->sc_inc.inc_laddr; 674 #ifdef INET6 675 } 676 #endif 677 inp->inp_lport = sc->sc_inc.inc_lport; 678 if (in_pcbinshash(inp) != 0) { 679 /* 680 * Undo the assignments above if we failed to 681 * put the PCB on the hash lists. 682 */ 683 #ifdef INET6 684 if (sc->sc_inc.inc_flags & INC_ISIPV6) 685 inp->in6p_laddr = in6addr_any; 686 else 687 #endif 688 inp->inp_laddr.s_addr = INADDR_ANY; 689 inp->inp_lport = 0; 690 goto abort; 691 } 692 #ifdef IPSEC 693 /* Copy old policy into new socket's. */ 694 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 695 printf("syncache_socket: could not copy policy\n"); 696 #endif 697 #ifdef INET6 698 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 699 struct inpcb *oinp = sotoinpcb(lso); 700 struct in6_addr laddr6; 701 struct sockaddr_in6 sin6; 702 /* 703 * Inherit socket options from the listening socket. 704 * Note that in6p_inputopts are not (and should not be) 705 * copied, since it stores previously received options and is 706 * used to detect if each new option is different than the 707 * previous one and hence should be passed to a user. 708 * If we copied in6p_inputopts, a user would not be able to 709 * receive options just after calling the accept system call. 710 */ 711 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 712 if (oinp->in6p_outputopts) 713 inp->in6p_outputopts = 714 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 715 716 sin6.sin6_family = AF_INET6; 717 sin6.sin6_len = sizeof(sin6); 718 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 719 sin6.sin6_port = sc->sc_inc.inc_fport; 720 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 721 laddr6 = inp->in6p_laddr; 722 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 723 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 724 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, 725 thread0.td_ucred)) { 726 inp->in6p_laddr = laddr6; 727 goto abort; 728 } 729 /* Override flowlabel from in6_pcbconnect. */ 730 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; 731 inp->inp_flow |= sc->sc_flowlabel; 732 } else 733 #endif 734 { 735 struct in_addr laddr; 736 struct sockaddr_in sin; 737 738 inp->inp_options = (m) ? ip_srcroute(m) : NULL; 739 740 if (inp->inp_options == NULL) { 741 inp->inp_options = sc->sc_ipopts; 742 sc->sc_ipopts = NULL; 743 } 744 745 sin.sin_family = AF_INET; 746 sin.sin_len = sizeof(sin); 747 sin.sin_addr = sc->sc_inc.inc_faddr; 748 sin.sin_port = sc->sc_inc.inc_fport; 749 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); 750 laddr = inp->inp_laddr; 751 if (inp->inp_laddr.s_addr == INADDR_ANY) 752 inp->inp_laddr = sc->sc_inc.inc_laddr; 753 if (in_pcbconnect(inp, (struct sockaddr *)&sin, 754 thread0.td_ucred)) { 755 inp->inp_laddr = laddr; 756 goto abort; 757 } 758 } 759 tp = intotcpcb(inp); 760 tp->t_state = TCPS_SYN_RECEIVED; 761 tp->iss = sc->sc_iss; 762 tp->irs = sc->sc_irs; 763 tcp_rcvseqinit(tp); 764 tcp_sendseqinit(tp); 765 tp->snd_wl1 = sc->sc_irs; 766 tp->snd_max = tp->iss + 1; 767 tp->snd_nxt = tp->iss + 1; 768 tp->rcv_up = sc->sc_irs + 1; 769 tp->rcv_wnd = sc->sc_wnd; 770 tp->rcv_adv += tp->rcv_wnd; 771 tp->last_ack_sent = tp->rcv_nxt; 772 773 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 774 if (sc->sc_flags & SCF_NOOPT) 775 tp->t_flags |= TF_NOOPT; 776 else { 777 if (sc->sc_flags & SCF_WINSCALE) { 778 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 779 tp->snd_scale = sc->sc_requested_s_scale; 780 tp->request_r_scale = sc->sc_requested_r_scale; 781 } 782 if (sc->sc_flags & SCF_TIMESTAMP) { 783 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 784 tp->ts_recent = sc->sc_tsreflect; 785 tp->ts_recent_age = ticks; 786 tp->ts_offset = sc->sc_tsoff; 787 } 788 #ifdef TCP_SIGNATURE 789 if (sc->sc_flags & SCF_SIGNATURE) 790 tp->t_flags |= TF_SIGNATURE; 791 #endif 792 if (sc->sc_flags & SCF_SACK) 793 tp->t_flags |= TF_SACK_PERMIT; 794 } 795 796 if (sc->sc_flags & SCF_ECN) 797 tp->t_flags |= TF_ECN_PERMIT; 798 799 /* 800 * Set up MSS and get cached values from tcp_hostcache. 801 * This might overwrite some of the defaults we just set. 802 */ 803 tcp_mss(tp, sc->sc_peer_mss); 804 805 /* 806 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 807 * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits. 808 */ 809 if (sc->sc_rxmits > 1) 810 tp->snd_cwnd = tp->t_maxseg; 811 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 812 813 INP_WUNLOCK(inp); 814 815 TCPSTAT_INC(tcps_accepts); 816 return (so); 817 818 abort: 819 INP_WUNLOCK(inp); 820 abort2: 821 if (so != NULL) 822 soabort(so); 823 return (NULL); 824 } 825 826 /* 827 * This function gets called when we receive an ACK for a 828 * socket in the LISTEN state. We look up the connection 829 * in the syncache, and if its there, we pull it out of 830 * the cache and turn it into a full-blown connection in 831 * the SYN-RECEIVED state. 832 */ 833 int 834 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 835 struct socket **lsop, struct mbuf *m) 836 { 837 struct syncache *sc; 838 struct syncache_head *sch; 839 struct syncache scs; 840 char *s; 841 842 /* 843 * Global TCP locks are held because we manipulate the PCB lists 844 * and create a new socket. 845 */ 846 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 847 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, 848 ("%s: can handle only ACK", __func__)); 849 850 sc = syncache_lookup(inc, &sch); /* returns locked sch */ 851 SCH_LOCK_ASSERT(sch); 852 if (sc == NULL) { 853 /* 854 * There is no syncache entry, so see if this ACK is 855 * a returning syncookie. To do this, first: 856 * A. See if this socket has had a syncache entry dropped in 857 * the past. We don't want to accept a bogus syncookie 858 * if we've never received a SYN. 859 * B. check that the syncookie is valid. If it is, then 860 * cobble up a fake syncache entry, and return. 861 */ 862 if (!V_tcp_syncookies) { 863 SCH_UNLOCK(sch); 864 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 865 log(LOG_DEBUG, "%s; %s: Spurious ACK, " 866 "segment rejected (syncookies disabled)\n", 867 s, __func__); 868 goto failed; 869 } 870 bzero(&scs, sizeof(scs)); 871 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop); 872 SCH_UNLOCK(sch); 873 if (sc == NULL) { 874 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 875 log(LOG_DEBUG, "%s; %s: Segment failed " 876 "SYNCOOKIE authentication, segment rejected " 877 "(probably spoofed)\n", s, __func__); 878 goto failed; 879 } 880 } else { 881 /* Pull out the entry to unlock the bucket row. */ 882 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 883 sch->sch_length--; 884 V_tcp_syncache.cache_count--; 885 SCH_UNLOCK(sch); 886 } 887 888 /* 889 * Segment validation: 890 * ACK must match our initial sequence number + 1 (the SYN|ACK). 891 */ 892 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) { 893 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 894 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " 895 "rejected\n", s, __func__, th->th_ack, sc->sc_iss); 896 goto failed; 897 } 898 899 /* 900 * The SEQ must fall in the window starting at the received 901 * initial receive sequence number + 1 (the SYN). 902 */ 903 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) || 904 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) && 905 !TOEPCB_ISSET(sc)) { 906 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 907 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " 908 "rejected\n", s, __func__, th->th_seq, sc->sc_irs); 909 goto failed; 910 } 911 912 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { 913 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 914 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 915 "segment rejected\n", s, __func__); 916 goto failed; 917 } 918 /* 919 * If timestamps were negotiated the reflected timestamp 920 * must be equal to what we actually sent in the SYN|ACK. 921 */ 922 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts && 923 !TOEPCB_ISSET(sc)) { 924 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) 925 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, " 926 "segment rejected\n", 927 s, __func__, to->to_tsecr, sc->sc_ts); 928 goto failed; 929 } 930 931 *lsop = syncache_socket(sc, *lsop, m); 932 933 if (*lsop == NULL) 934 TCPSTAT_INC(tcps_sc_aborted); 935 else 936 TCPSTAT_INC(tcps_sc_completed); 937 938 /* how do we find the inp for the new socket? */ 939 if (sc != &scs) 940 syncache_free(sc); 941 return (1); 942 failed: 943 if (sc != NULL && sc != &scs) 944 syncache_free(sc); 945 if (s != NULL) 946 free(s, M_TCPLOG); 947 *lsop = NULL; 948 return (0); 949 } 950 951 int 952 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo, 953 struct tcphdr *th, struct socket **lsop, struct mbuf *m) 954 { 955 struct tcpopt to; 956 int rc; 957 958 bzero(&to, sizeof(struct tcpopt)); 959 to.to_mss = toeo->to_mss; 960 to.to_wscale = toeo->to_wscale; 961 to.to_flags = toeo->to_flags; 962 963 INP_INFO_WLOCK(&V_tcbinfo); 964 rc = syncache_expand(inc, &to, th, lsop, m); 965 INP_INFO_WUNLOCK(&V_tcbinfo); 966 967 return (rc); 968 } 969 970 /* 971 * Given a LISTEN socket and an inbound SYN request, add 972 * this to the syn cache, and send back a segment: 973 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 974 * to the source. 975 * 976 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 977 * Doing so would require that we hold onto the data and deliver it 978 * to the application. However, if we are the target of a SYN-flood 979 * DoS attack, an attacker could send data which would eventually 980 * consume all available buffer space if it were ACKed. By not ACKing 981 * the data, we avoid this DoS scenario. 982 */ 983 static void 984 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 985 struct inpcb *inp, struct socket **lsop, struct mbuf *m, 986 struct toe_usrreqs *tu, void *toepcb) 987 { 988 struct tcpcb *tp; 989 struct socket *so; 990 struct syncache *sc = NULL; 991 struct syncache_head *sch; 992 struct mbuf *ipopts = NULL; 993 u_int32_t flowtmp; 994 int win, sb_hiwat, ip_ttl, ip_tos, noopt; 995 char *s; 996 #ifdef INET6 997 int autoflowlabel = 0; 998 #endif 999 #ifdef MAC 1000 struct label *maclabel; 1001 #endif 1002 struct syncache scs; 1003 struct ucred *cred; 1004 1005 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1006 INP_WLOCK_ASSERT(inp); /* listen socket */ 1007 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, 1008 ("%s: unexpected tcp flags", __func__)); 1009 1010 /* 1011 * Combine all so/tp operations very early to drop the INP lock as 1012 * soon as possible. 1013 */ 1014 so = *lsop; 1015 tp = sototcpcb(so); 1016 cred = crhold(so->so_cred); 1017 1018 #ifdef INET6 1019 if ((inc->inc_flags & INC_ISIPV6) && 1020 (inp->inp_flags & IN6P_AUTOFLOWLABEL)) 1021 autoflowlabel = 1; 1022 #endif 1023 ip_ttl = inp->inp_ip_ttl; 1024 ip_tos = inp->inp_ip_tos; 1025 win = sbspace(&so->so_rcv); 1026 sb_hiwat = so->so_rcv.sb_hiwat; 1027 noopt = (tp->t_flags & TF_NOOPT); 1028 1029 /* By the time we drop the lock these should no longer be used. */ 1030 so = NULL; 1031 tp = NULL; 1032 1033 #ifdef MAC 1034 if (mac_syncache_init(&maclabel) != 0) { 1035 INP_WUNLOCK(inp); 1036 INP_INFO_WUNLOCK(&V_tcbinfo); 1037 goto done; 1038 } else 1039 mac_syncache_create(maclabel, inp); 1040 #endif 1041 INP_WUNLOCK(inp); 1042 INP_INFO_WUNLOCK(&V_tcbinfo); 1043 1044 /* 1045 * Remember the IP options, if any. 1046 */ 1047 #ifdef INET6 1048 if (!(inc->inc_flags & INC_ISIPV6)) 1049 #endif 1050 ipopts = (m) ? ip_srcroute(m) : NULL; 1051 1052 /* 1053 * See if we already have an entry for this connection. 1054 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1055 * 1056 * XXX: should the syncache be re-initialized with the contents 1057 * of the new SYN here (which may have different options?) 1058 * 1059 * XXX: We do not check the sequence number to see if this is a 1060 * real retransmit or a new connection attempt. The question is 1061 * how to handle such a case; either ignore it as spoofed, or 1062 * drop the current entry and create a new one? 1063 */ 1064 sc = syncache_lookup(inc, &sch); /* returns locked entry */ 1065 SCH_LOCK_ASSERT(sch); 1066 if (sc != NULL) { 1067 #ifndef TCP_OFFLOAD_DISABLE 1068 if (sc->sc_tu) 1069 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT, 1070 sc->sc_toepcb); 1071 #endif 1072 TCPSTAT_INC(tcps_sc_dupsyn); 1073 if (ipopts) { 1074 /* 1075 * If we were remembering a previous source route, 1076 * forget it and use the new one we've been given. 1077 */ 1078 if (sc->sc_ipopts) 1079 (void) m_free(sc->sc_ipopts); 1080 sc->sc_ipopts = ipopts; 1081 } 1082 /* 1083 * Update timestamp if present. 1084 */ 1085 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) 1086 sc->sc_tsreflect = to->to_tsval; 1087 else 1088 sc->sc_flags &= ~SCF_TIMESTAMP; 1089 #ifdef MAC 1090 /* 1091 * Since we have already unconditionally allocated label 1092 * storage, free it up. The syncache entry will already 1093 * have an initialized label we can use. 1094 */ 1095 mac_syncache_destroy(&maclabel); 1096 #endif 1097 /* Retransmit SYN|ACK and reset retransmit count. */ 1098 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { 1099 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " 1100 "resetting timer and retransmitting SYN|ACK\n", 1101 s, __func__); 1102 free(s, M_TCPLOG); 1103 } 1104 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) { 1105 sc->sc_rxmits = 0; 1106 syncache_timeout(sc, sch, 1); 1107 TCPSTAT_INC(tcps_sndacks); 1108 TCPSTAT_INC(tcps_sndtotal); 1109 } 1110 SCH_UNLOCK(sch); 1111 goto done; 1112 } 1113 1114 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); 1115 if (sc == NULL) { 1116 /* 1117 * The zone allocator couldn't provide more entries. 1118 * Treat this as if the cache was full; drop the oldest 1119 * entry and insert the new one. 1120 */ 1121 TCPSTAT_INC(tcps_sc_zonefail); 1122 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) 1123 syncache_drop(sc, sch); 1124 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); 1125 if (sc == NULL) { 1126 if (V_tcp_syncookies) { 1127 bzero(&scs, sizeof(scs)); 1128 sc = &scs; 1129 } else { 1130 SCH_UNLOCK(sch); 1131 if (ipopts) 1132 (void) m_free(ipopts); 1133 goto done; 1134 } 1135 } 1136 } 1137 1138 /* 1139 * Fill in the syncache values. 1140 */ 1141 #ifdef MAC 1142 sc->sc_label = maclabel; 1143 #endif 1144 sc->sc_cred = cred; 1145 cred = NULL; 1146 sc->sc_ipopts = ipopts; 1147 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1148 #ifdef INET6 1149 if (!(inc->inc_flags & INC_ISIPV6)) 1150 #endif 1151 { 1152 sc->sc_ip_tos = ip_tos; 1153 sc->sc_ip_ttl = ip_ttl; 1154 } 1155 #ifndef TCP_OFFLOAD_DISABLE 1156 sc->sc_tu = tu; 1157 sc->sc_toepcb = toepcb; 1158 #endif 1159 sc->sc_irs = th->th_seq; 1160 sc->sc_iss = arc4random(); 1161 sc->sc_flags = 0; 1162 sc->sc_flowlabel = 0; 1163 1164 /* 1165 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. 1166 * win was derived from socket earlier in the function. 1167 */ 1168 win = imax(win, 0); 1169 win = imin(win, TCP_MAXWIN); 1170 sc->sc_wnd = win; 1171 1172 if (V_tcp_do_rfc1323) { 1173 /* 1174 * A timestamp received in a SYN makes 1175 * it ok to send timestamp requests and replies. 1176 */ 1177 if (to->to_flags & TOF_TS) { 1178 sc->sc_tsreflect = to->to_tsval; 1179 sc->sc_ts = ticks; 1180 sc->sc_flags |= SCF_TIMESTAMP; 1181 } 1182 if (to->to_flags & TOF_SCALE) { 1183 int wscale = 0; 1184 1185 /* 1186 * Pick the smallest possible scaling factor that 1187 * will still allow us to scale up to sb_max, aka 1188 * kern.ipc.maxsockbuf. 1189 * 1190 * We do this because there are broken firewalls that 1191 * will corrupt the window scale option, leading to 1192 * the other endpoint believing that our advertised 1193 * window is unscaled. At scale factors larger than 1194 * 5 the unscaled window will drop below 1500 bytes, 1195 * leading to serious problems when traversing these 1196 * broken firewalls. 1197 * 1198 * With the default maxsockbuf of 256K, a scale factor 1199 * of 3 will be chosen by this algorithm. Those who 1200 * choose a larger maxsockbuf should watch out 1201 * for the compatiblity problems mentioned above. 1202 * 1203 * RFC1323: The Window field in a SYN (i.e., a <SYN> 1204 * or <SYN,ACK>) segment itself is never scaled. 1205 */ 1206 while (wscale < TCP_MAX_WINSHIFT && 1207 (TCP_MAXWIN << wscale) < sb_max) 1208 wscale++; 1209 sc->sc_requested_r_scale = wscale; 1210 sc->sc_requested_s_scale = to->to_wscale; 1211 sc->sc_flags |= SCF_WINSCALE; 1212 } 1213 } 1214 #ifdef TCP_SIGNATURE 1215 /* 1216 * If listening socket requested TCP digests, and received SYN 1217 * contains the option, flag this in the syncache so that 1218 * syncache_respond() will do the right thing with the SYN+ACK. 1219 * XXX: Currently we always record the option by default and will 1220 * attempt to use it in syncache_respond(). 1221 */ 1222 if (to->to_flags & TOF_SIGNATURE) 1223 sc->sc_flags |= SCF_SIGNATURE; 1224 #endif 1225 if (to->to_flags & TOF_SACKPERM) 1226 sc->sc_flags |= SCF_SACK; 1227 if (to->to_flags & TOF_MSS) 1228 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ 1229 if (noopt) 1230 sc->sc_flags |= SCF_NOOPT; 1231 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn) 1232 sc->sc_flags |= SCF_ECN; 1233 1234 if (V_tcp_syncookies) { 1235 syncookie_generate(sch, sc, &flowtmp); 1236 #ifdef INET6 1237 if (autoflowlabel) 1238 sc->sc_flowlabel = flowtmp; 1239 #endif 1240 } else { 1241 #ifdef INET6 1242 if (autoflowlabel) 1243 sc->sc_flowlabel = 1244 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1245 #endif 1246 } 1247 SCH_UNLOCK(sch); 1248 1249 /* 1250 * Do a standard 3-way handshake. 1251 */ 1252 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) { 1253 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs) 1254 syncache_free(sc); 1255 else if (sc != &scs) 1256 syncache_insert(sc, sch); /* locks and unlocks sch */ 1257 TCPSTAT_INC(tcps_sndacks); 1258 TCPSTAT_INC(tcps_sndtotal); 1259 } else { 1260 if (sc != &scs) 1261 syncache_free(sc); 1262 TCPSTAT_INC(tcps_sc_dropped); 1263 } 1264 1265 done: 1266 if (cred != NULL) 1267 crfree(cred); 1268 #ifdef MAC 1269 if (sc == &scs) 1270 mac_syncache_destroy(&maclabel); 1271 #endif 1272 if (m) { 1273 1274 *lsop = NULL; 1275 m_freem(m); 1276 } 1277 } 1278 1279 static int 1280 syncache_respond(struct syncache *sc) 1281 { 1282 struct ip *ip = NULL; 1283 struct mbuf *m; 1284 struct tcphdr *th; 1285 int optlen, error; 1286 u_int16_t hlen, tlen, mssopt; 1287 struct tcpopt to; 1288 #ifdef INET6 1289 struct ip6_hdr *ip6 = NULL; 1290 #endif 1291 1292 hlen = 1293 #ifdef INET6 1294 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) : 1295 #endif 1296 sizeof(struct ip); 1297 tlen = hlen + sizeof(struct tcphdr); 1298 1299 /* Determine MSS we advertize to other end of connection. */ 1300 mssopt = tcp_mssopt(&sc->sc_inc); 1301 if (sc->sc_peer_mss) 1302 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss); 1303 1304 /* XXX: Assume that the entire packet will fit in a header mbuf. */ 1305 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, 1306 ("syncache: mbuf too small")); 1307 1308 /* Create the IP+TCP header from scratch. */ 1309 m = m_gethdr(M_DONTWAIT, MT_DATA); 1310 if (m == NULL) 1311 return (ENOBUFS); 1312 #ifdef MAC 1313 mac_syncache_create_mbuf(sc->sc_label, m); 1314 #endif 1315 m->m_data += max_linkhdr; 1316 m->m_len = tlen; 1317 m->m_pkthdr.len = tlen; 1318 m->m_pkthdr.rcvif = NULL; 1319 1320 #ifdef INET6 1321 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 1322 ip6 = mtod(m, struct ip6_hdr *); 1323 ip6->ip6_vfc = IPV6_VERSION; 1324 ip6->ip6_nxt = IPPROTO_TCP; 1325 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1326 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1327 ip6->ip6_plen = htons(tlen - hlen); 1328 /* ip6_hlim is set after checksum */ 1329 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1330 ip6->ip6_flow |= sc->sc_flowlabel; 1331 1332 th = (struct tcphdr *)(ip6 + 1); 1333 } else 1334 #endif 1335 { 1336 ip = mtod(m, struct ip *); 1337 ip->ip_v = IPVERSION; 1338 ip->ip_hl = sizeof(struct ip) >> 2; 1339 ip->ip_len = tlen; 1340 ip->ip_id = 0; 1341 ip->ip_off = 0; 1342 ip->ip_sum = 0; 1343 ip->ip_p = IPPROTO_TCP; 1344 ip->ip_src = sc->sc_inc.inc_laddr; 1345 ip->ip_dst = sc->sc_inc.inc_faddr; 1346 ip->ip_ttl = sc->sc_ip_ttl; 1347 ip->ip_tos = sc->sc_ip_tos; 1348 1349 /* 1350 * See if we should do MTU discovery. Route lookups are 1351 * expensive, so we will only unset the DF bit if: 1352 * 1353 * 1) path_mtu_discovery is disabled 1354 * 2) the SCF_UNREACH flag has been set 1355 */ 1356 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) 1357 ip->ip_off |= IP_DF; 1358 1359 th = (struct tcphdr *)(ip + 1); 1360 } 1361 th->th_sport = sc->sc_inc.inc_lport; 1362 th->th_dport = sc->sc_inc.inc_fport; 1363 1364 th->th_seq = htonl(sc->sc_iss); 1365 th->th_ack = htonl(sc->sc_irs + 1); 1366 th->th_off = sizeof(struct tcphdr) >> 2; 1367 th->th_x2 = 0; 1368 th->th_flags = TH_SYN|TH_ACK; 1369 th->th_win = htons(sc->sc_wnd); 1370 th->th_urp = 0; 1371 1372 if (sc->sc_flags & SCF_ECN) { 1373 th->th_flags |= TH_ECE; 1374 TCPSTAT_INC(tcps_ecn_shs); 1375 } 1376 1377 /* Tack on the TCP options. */ 1378 if ((sc->sc_flags & SCF_NOOPT) == 0) { 1379 to.to_flags = 0; 1380 1381 to.to_mss = mssopt; 1382 to.to_flags = TOF_MSS; 1383 if (sc->sc_flags & SCF_WINSCALE) { 1384 to.to_wscale = sc->sc_requested_r_scale; 1385 to.to_flags |= TOF_SCALE; 1386 } 1387 if (sc->sc_flags & SCF_TIMESTAMP) { 1388 /* Virgin timestamp or TCP cookie enhanced one. */ 1389 to.to_tsval = sc->sc_ts; 1390 to.to_tsecr = sc->sc_tsreflect; 1391 to.to_flags |= TOF_TS; 1392 } 1393 if (sc->sc_flags & SCF_SACK) 1394 to.to_flags |= TOF_SACKPERM; 1395 #ifdef TCP_SIGNATURE 1396 if (sc->sc_flags & SCF_SIGNATURE) 1397 to.to_flags |= TOF_SIGNATURE; 1398 #endif 1399 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1400 1401 /* Adjust headers by option size. */ 1402 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1403 m->m_len += optlen; 1404 m->m_pkthdr.len += optlen; 1405 1406 #ifdef TCP_SIGNATURE 1407 if (sc->sc_flags & SCF_SIGNATURE) 1408 tcp_signature_compute(m, 0, 0, optlen, 1409 to.to_signature, IPSEC_DIR_OUTBOUND); 1410 #endif 1411 #ifdef INET6 1412 if (sc->sc_inc.inc_flags & INC_ISIPV6) 1413 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); 1414 else 1415 #endif 1416 ip->ip_len += optlen; 1417 } else 1418 optlen = 0; 1419 1420 M_SETFIB(m, sc->sc_inc.inc_fibnum); 1421 #ifdef INET6 1422 if (sc->sc_inc.inc_flags & INC_ISIPV6) { 1423 th->th_sum = 0; 1424 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, 1425 tlen + optlen - hlen); 1426 ip6->ip6_hlim = in6_selecthlim(NULL, NULL); 1427 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1428 } else 1429 #endif 1430 { 1431 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1432 htons(tlen + optlen - hlen + IPPROTO_TCP)); 1433 m->m_pkthdr.csum_flags = CSUM_TCP; 1434 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1435 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); 1436 } 1437 return (error); 1438 } 1439 1440 void 1441 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 1442 struct inpcb *inp, struct socket **lsop, struct mbuf *m) 1443 { 1444 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL); 1445 } 1446 1447 void 1448 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo, 1449 struct tcphdr *th, struct inpcb *inp, struct socket **lsop, 1450 struct toe_usrreqs *tu, void *toepcb) 1451 { 1452 struct tcpopt to; 1453 1454 bzero(&to, sizeof(struct tcpopt)); 1455 to.to_mss = toeo->to_mss; 1456 to.to_wscale = toeo->to_wscale; 1457 to.to_flags = toeo->to_flags; 1458 1459 INP_INFO_WLOCK(&V_tcbinfo); 1460 INP_WLOCK(inp); 1461 1462 _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb); 1463 } 1464 1465 /* 1466 * The purpose of SYN cookies is to avoid keeping track of all SYN's we 1467 * receive and to be able to handle SYN floods from bogus source addresses 1468 * (where we will never receive any reply). SYN floods try to exhaust all 1469 * our memory and available slots in the SYN cache table to cause a denial 1470 * of service to legitimate users of the local host. 1471 * 1472 * The idea of SYN cookies is to encode and include all necessary information 1473 * about the connection setup state within the SYN-ACK we send back and thus 1474 * to get along without keeping any local state until the ACK to the SYN-ACK 1475 * arrives (if ever). Everything we need to know should be available from 1476 * the information we encoded in the SYN-ACK. 1477 * 1478 * More information about the theory behind SYN cookies and its first 1479 * discussion and specification can be found at: 1480 * http://cr.yp.to/syncookies.html (overview) 1481 * http://cr.yp.to/syncookies/archive (gory details) 1482 * 1483 * This implementation extends the orginal idea and first implementation 1484 * of FreeBSD by using not only the initial sequence number field to store 1485 * information but also the timestamp field if present. This way we can 1486 * keep track of the entire state we need to know to recreate the session in 1487 * its original form. Almost all TCP speakers implement RFC1323 timestamps 1488 * these days. For those that do not we still have to live with the known 1489 * shortcomings of the ISN only SYN cookies. 1490 * 1491 * Cookie layers: 1492 * 1493 * Initial sequence number we send: 1494 * 31|................................|0 1495 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP 1496 * D = MD5 Digest (first dword) 1497 * M = MSS index 1498 * R = Rotation of secret 1499 * P = Odd or Even secret 1500 * 1501 * The MD5 Digest is computed with over following parameters: 1502 * a) randomly rotated secret 1503 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6) 1504 * c) the received initial sequence number from remote host 1505 * d) the rotation offset and odd/even bit 1506 * 1507 * Timestamp we send: 1508 * 31|................................|0 1509 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5 1510 * D = MD5 Digest (third dword) (only as filler) 1511 * S = Requested send window scale 1512 * R = Requested receive window scale 1513 * A = SACK allowed 1514 * 5 = TCP-MD5 enabled (not implemented yet) 1515 * XORed with MD5 Digest (forth dword) 1516 * 1517 * The timestamp isn't cryptographically secure and doesn't need to be. 1518 * The double use of the MD5 digest dwords ties it to a specific remote/ 1519 * local host/port, remote initial sequence number and our local time 1520 * limited secret. A received timestamp is reverted (XORed) and then 1521 * the contained MD5 dword is compared to the computed one to ensure the 1522 * timestamp belongs to the SYN-ACK we sent. The other parameters may 1523 * have been tampered with but this isn't different from supplying bogus 1524 * values in the SYN in the first place. 1525 * 1526 * Some problems with SYN cookies remain however: 1527 * Consider the problem of a recreated (and retransmitted) cookie. If the 1528 * original SYN was accepted, the connection is established. The second 1529 * SYN is inflight, and if it arrives with an ISN that falls within the 1530 * receive window, the connection is killed. 1531 * 1532 * Notes: 1533 * A heuristic to determine when to accept syn cookies is not necessary. 1534 * An ACK flood would cause the syncookie verification to be attempted, 1535 * but a SYN flood causes syncookies to be generated. Both are of equal 1536 * cost, so there's no point in trying to optimize the ACK flood case. 1537 * Also, if you don't process certain ACKs for some reason, then all someone 1538 * would have to do is launch a SYN and ACK flood at the same time, which 1539 * would stop cookie verification and defeat the entire purpose of syncookies. 1540 */ 1541 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 }; 1542 1543 static void 1544 syncookie_generate(struct syncache_head *sch, struct syncache *sc, 1545 u_int32_t *flowlabel) 1546 { 1547 MD5_CTX ctx; 1548 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1549 u_int32_t data; 1550 u_int32_t *secbits; 1551 u_int off, pmss, mss; 1552 int i; 1553 1554 SCH_LOCK_ASSERT(sch); 1555 1556 /* Which of the two secrets to use. */ 1557 secbits = sch->sch_oddeven ? 1558 sch->sch_secbits_odd : sch->sch_secbits_even; 1559 1560 /* Reseed secret if too old. */ 1561 if (sch->sch_reseed < time_uptime) { 1562 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */ 1563 secbits = sch->sch_oddeven ? 1564 sch->sch_secbits_odd : sch->sch_secbits_even; 1565 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++) 1566 secbits[i] = arc4random(); 1567 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME; 1568 } 1569 1570 /* Secret rotation offset. */ 1571 off = sc->sc_iss & 0x7; /* iss was randomized before */ 1572 1573 /* Maximum segment size calculation. */ 1574 pmss = 1575 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss); 1576 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--) 1577 if (tcp_sc_msstab[mss] <= pmss) 1578 break; 1579 1580 /* Fold parameters and MD5 digest into the ISN we will send. */ 1581 data = sch->sch_oddeven;/* odd or even secret, 1 bit */ 1582 data |= off << 1; /* secret offset, derived from iss, 3 bits */ 1583 data |= mss << 4; /* mss, 3 bits */ 1584 1585 MD5Init(&ctx); 1586 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1587 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1588 MD5Update(&ctx, secbits, off); 1589 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc)); 1590 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs)); 1591 MD5Update(&ctx, &data, sizeof(data)); 1592 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1593 1594 data |= (md5_buffer[0] << 7); 1595 sc->sc_iss = data; 1596 1597 #ifdef INET6 1598 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1599 #endif 1600 1601 /* Additional parameters are stored in the timestamp if present. */ 1602 if (sc->sc_flags & SCF_TIMESTAMP) { 1603 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */ 1604 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */ 1605 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */ 1606 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */ 1607 data |= md5_buffer[2] << 10; /* more digest bits */ 1608 data ^= md5_buffer[3]; 1609 sc->sc_ts = data; 1610 sc->sc_tsoff = data - ticks; /* after XOR */ 1611 } 1612 1613 TCPSTAT_INC(tcps_sc_sendcookie); 1614 } 1615 1616 static struct syncache * 1617 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 1618 struct syncache *sc, struct tcpopt *to, struct tcphdr *th, 1619 struct socket *so) 1620 { 1621 MD5_CTX ctx; 1622 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)]; 1623 u_int32_t data = 0; 1624 u_int32_t *secbits; 1625 tcp_seq ack, seq; 1626 int off, mss, wnd, flags; 1627 1628 SCH_LOCK_ASSERT(sch); 1629 1630 /* 1631 * Pull information out of SYN-ACK/ACK and 1632 * revert sequence number advances. 1633 */ 1634 ack = th->th_ack - 1; 1635 seq = th->th_seq - 1; 1636 off = (ack >> 1) & 0x7; 1637 mss = (ack >> 4) & 0x7; 1638 flags = ack & 0x7f; 1639 1640 /* Which of the two secrets to use. */ 1641 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even; 1642 1643 /* 1644 * The secret wasn't updated for the lifetime of a syncookie, 1645 * so this SYN-ACK/ACK is either too old (replay) or totally bogus. 1646 */ 1647 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) { 1648 return (NULL); 1649 } 1650 1651 /* Recompute the digest so we can compare it. */ 1652 MD5Init(&ctx); 1653 MD5Update(&ctx, ((u_int8_t *)secbits) + off, 1654 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off); 1655 MD5Update(&ctx, secbits, off); 1656 MD5Update(&ctx, inc, sizeof(*inc)); 1657 MD5Update(&ctx, &seq, sizeof(seq)); 1658 MD5Update(&ctx, &flags, sizeof(flags)); 1659 MD5Final((u_int8_t *)&md5_buffer, &ctx); 1660 1661 /* Does the digest part of or ACK'ed ISS match? */ 1662 if ((ack & (~0x7f)) != (md5_buffer[0] << 7)) 1663 return (NULL); 1664 1665 /* Does the digest part of our reflected timestamp match? */ 1666 if (to->to_flags & TOF_TS) { 1667 data = md5_buffer[3] ^ to->to_tsecr; 1668 if ((data & (~0x3ff)) != (md5_buffer[2] << 10)) 1669 return (NULL); 1670 } 1671 1672 /* Fill in the syncache values. */ 1673 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); 1674 sc->sc_ipopts = NULL; 1675 1676 sc->sc_irs = seq; 1677 sc->sc_iss = ack; 1678 1679 #ifdef INET6 1680 if (inc->inc_flags & INC_ISIPV6) { 1681 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL) 1682 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1683 } else 1684 #endif 1685 { 1686 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl; 1687 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos; 1688 } 1689 1690 /* Additional parameters that were encoded in the timestamp. */ 1691 if (data) { 1692 sc->sc_flags |= SCF_TIMESTAMP; 1693 sc->sc_tsreflect = to->to_tsval; 1694 sc->sc_ts = to->to_tsecr; 1695 sc->sc_tsoff = to->to_tsecr - ticks; 1696 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0; 1697 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0; 1698 sc->sc_requested_s_scale = min((data >> 2) & 0xf, 1699 TCP_MAX_WINSHIFT); 1700 sc->sc_requested_r_scale = min((data >> 6) & 0xf, 1701 TCP_MAX_WINSHIFT); 1702 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale) 1703 sc->sc_flags |= SCF_WINSCALE; 1704 } else 1705 sc->sc_flags |= SCF_NOOPT; 1706 1707 wnd = sbspace(&so->so_rcv); 1708 wnd = imax(wnd, 0); 1709 wnd = imin(wnd, TCP_MAXWIN); 1710 sc->sc_wnd = wnd; 1711 1712 sc->sc_rxmits = 0; 1713 sc->sc_peer_mss = tcp_sc_msstab[mss]; 1714 1715 TCPSTAT_INC(tcps_sc_recvcookie); 1716 return (sc); 1717 } 1718 1719 /* 1720 * Returns the current number of syncache entries. This number 1721 * will probably change before you get around to calling 1722 * syncache_pcblist. 1723 */ 1724 1725 int 1726 syncache_pcbcount(void) 1727 { 1728 struct syncache_head *sch; 1729 int count, i; 1730 1731 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { 1732 /* No need to lock for a read. */ 1733 sch = &V_tcp_syncache.hashbase[i]; 1734 count += sch->sch_length; 1735 } 1736 return count; 1737 } 1738 1739 /* 1740 * Exports the syncache entries to userland so that netstat can display 1741 * them alongside the other sockets. This function is intended to be 1742 * called only from tcp_pcblist. 1743 * 1744 * Due to concurrency on an active system, the number of pcbs exported 1745 * may have no relation to max_pcbs. max_pcbs merely indicates the 1746 * amount of space the caller allocated for this function to use. 1747 */ 1748 int 1749 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported) 1750 { 1751 struct xtcpcb xt; 1752 struct syncache *sc; 1753 struct syncache_head *sch; 1754 int count, error, i; 1755 1756 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { 1757 sch = &V_tcp_syncache.hashbase[i]; 1758 SCH_LOCK(sch); 1759 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 1760 if (count >= max_pcbs) { 1761 SCH_UNLOCK(sch); 1762 goto exit; 1763 } 1764 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0) 1765 continue; 1766 bzero(&xt, sizeof(xt)); 1767 xt.xt_len = sizeof(xt); 1768 if (sc->sc_inc.inc_flags & INC_ISIPV6) 1769 xt.xt_inp.inp_vflag = INP_IPV6; 1770 else 1771 xt.xt_inp.inp_vflag = INP_IPV4; 1772 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); 1773 xt.xt_tp.t_inpcb = &xt.xt_inp; 1774 xt.xt_tp.t_state = TCPS_SYN_RECEIVED; 1775 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1776 xt.xt_socket.xso_len = sizeof (struct xsocket); 1777 xt.xt_socket.so_type = SOCK_STREAM; 1778 xt.xt_socket.so_state = SS_ISCONNECTING; 1779 error = SYSCTL_OUT(req, &xt, sizeof xt); 1780 if (error) { 1781 SCH_UNLOCK(sch); 1782 goto exit; 1783 } 1784 count++; 1785 } 1786 SCH_UNLOCK(sch); 1787 } 1788 exit: 1789 *pcbs_exported = count; 1790 return error; 1791 } 1792