1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 /* 28 * This file contains functions related to TCP time wait processing. Also 29 * refer to the time wait handling comments in tcp_impl.h. 30 */ 31 32 #include <sys/types.h> 33 #include <sys/strsun.h> 34 #include <sys/squeue_impl.h> 35 #include <sys/squeue.h> 36 #include <sys/callo.h> 37 38 #include <inet/common.h> 39 #include <inet/ip.h> 40 #include <inet/tcp.h> 41 #include <inet/tcp_impl.h> 42 #include <inet/tcp_cluster.h> 43 44 static void tcp_timewait_close(void *, mblk_t *, void *, ip_recv_attr_t *); 45 46 /* 47 * TCP_TIME_WAIT_DELAY governs how often the time_wait_collector runs. 48 * Running it every 5 seconds seems to give the best results. 49 */ 50 #define TCP_TIME_WAIT_DELAY ((hrtime_t)5 * NANOSEC) 51 52 /* 53 * Remove a connection from the list of detached TIME_WAIT connections. 54 * It returns B_FALSE if it can't remove the connection from the list 55 * as the connection has already been removed from the list due to an 56 * earlier call to tcp_time_wait_remove(); otherwise it returns B_TRUE. 57 */ 58 boolean_t 59 tcp_time_wait_remove(tcp_t *tcp, tcp_squeue_priv_t *tcp_time_wait) 60 { 61 boolean_t locked = B_FALSE; 62 63 if (tcp_time_wait == NULL) { 64 tcp_time_wait = *((tcp_squeue_priv_t **) 65 squeue_getprivate(tcp->tcp_connp->conn_sqp, SQPRIVATE_TCP)); 66 mutex_enter(&tcp_time_wait->tcp_time_wait_lock); 67 locked = B_TRUE; 68 } else { 69 ASSERT(MUTEX_HELD(&tcp_time_wait->tcp_time_wait_lock)); 70 } 71 72 /* 0 means that the tcp_t has not been added to the time wait list. */ 73 if (tcp->tcp_time_wait_expire == 0) { 74 ASSERT(tcp->tcp_time_wait_next == NULL); 75 ASSERT(tcp->tcp_time_wait_prev == NULL); 76 if (locked) 77 mutex_exit(&tcp_time_wait->tcp_time_wait_lock); 78 return (B_FALSE); 79 } 80 ASSERT(TCP_IS_DETACHED(tcp)); 81 ASSERT(tcp->tcp_state == TCPS_TIME_WAIT); 82 83 if (tcp == tcp_time_wait->tcp_time_wait_head) { 84 ASSERT(tcp->tcp_time_wait_prev == NULL); 85 tcp_time_wait->tcp_time_wait_head = tcp->tcp_time_wait_next; 86 if (tcp_time_wait->tcp_time_wait_head != NULL) { 87 tcp_time_wait->tcp_time_wait_head->tcp_time_wait_prev = 88 NULL; 89 } else { 90 tcp_time_wait->tcp_time_wait_tail = NULL; 91 } 92 } else if (tcp == tcp_time_wait->tcp_time_wait_tail) { 93 ASSERT(tcp->tcp_time_wait_next == NULL); 94 tcp_time_wait->tcp_time_wait_tail = tcp->tcp_time_wait_prev; 95 ASSERT(tcp_time_wait->tcp_time_wait_tail != NULL); 96 tcp_time_wait->tcp_time_wait_tail->tcp_time_wait_next = NULL; 97 } else { 98 ASSERT(tcp->tcp_time_wait_prev->tcp_time_wait_next == tcp); 99 ASSERT(tcp->tcp_time_wait_next->tcp_time_wait_prev == tcp); 100 tcp->tcp_time_wait_prev->tcp_time_wait_next = 101 tcp->tcp_time_wait_next; 102 tcp->tcp_time_wait_next->tcp_time_wait_prev = 103 tcp->tcp_time_wait_prev; 104 } 105 tcp->tcp_time_wait_next = NULL; 106 tcp->tcp_time_wait_prev = NULL; 107 tcp->tcp_time_wait_expire = 0; 108 109 if (locked) 110 mutex_exit(&tcp_time_wait->tcp_time_wait_lock); 111 return (B_TRUE); 112 } 113 114 /* 115 * Add a connection to the list of detached TIME_WAIT connections 116 * and set its time to expire. 117 */ 118 void 119 tcp_time_wait_append(tcp_t *tcp) 120 { 121 tcp_stack_t *tcps = tcp->tcp_tcps; 122 tcp_squeue_priv_t *tcp_time_wait = 123 *((tcp_squeue_priv_t **)squeue_getprivate(tcp->tcp_connp->conn_sqp, 124 SQPRIVATE_TCP)); 125 126 tcp_timers_stop(tcp); 127 128 /* Freed above */ 129 ASSERT(tcp->tcp_timer_tid == 0); 130 ASSERT(tcp->tcp_ack_tid == 0); 131 132 /* must have happened at the time of detaching the tcp */ 133 ASSERT(tcp->tcp_ptpahn == NULL); 134 ASSERT(tcp->tcp_flow_stopped == 0); 135 ASSERT(tcp->tcp_time_wait_next == NULL); 136 ASSERT(tcp->tcp_time_wait_prev == NULL); 137 ASSERT(tcp->tcp_time_wait_expire == NULL); 138 ASSERT(tcp->tcp_listener == NULL); 139 140 tcp->tcp_time_wait_expire = ddi_get_lbolt(); 141 /* 142 * The value computed below in tcp->tcp_time_wait_expire may 143 * appear negative or wrap around. That is ok since our 144 * interest is only in the difference between the current lbolt 145 * value and tcp->tcp_time_wait_expire. But the value should not 146 * be zero, since it means the tcp is not in the TIME_WAIT list. 147 * The corresponding comparison in tcp_time_wait_collector() uses 148 * modular arithmetic. 149 */ 150 tcp->tcp_time_wait_expire += MSEC_TO_TICK( 151 tcps->tcps_time_wait_interval); 152 if (tcp->tcp_time_wait_expire == 0) 153 tcp->tcp_time_wait_expire = 1; 154 155 ASSERT(TCP_IS_DETACHED(tcp)); 156 ASSERT(tcp->tcp_state == TCPS_TIME_WAIT); 157 ASSERT(tcp->tcp_time_wait_next == NULL); 158 ASSERT(tcp->tcp_time_wait_prev == NULL); 159 TCP_DBGSTAT(tcps, tcp_time_wait); 160 161 mutex_enter(&tcp_time_wait->tcp_time_wait_lock); 162 if (tcp_time_wait->tcp_time_wait_head == NULL) { 163 ASSERT(tcp_time_wait->tcp_time_wait_tail == NULL); 164 tcp_time_wait->tcp_time_wait_head = tcp; 165 } else { 166 ASSERT(tcp_time_wait->tcp_time_wait_tail != NULL); 167 ASSERT(tcp_time_wait->tcp_time_wait_tail->tcp_state == 168 TCPS_TIME_WAIT); 169 tcp_time_wait->tcp_time_wait_tail->tcp_time_wait_next = tcp; 170 tcp->tcp_time_wait_prev = tcp_time_wait->tcp_time_wait_tail; 171 } 172 tcp_time_wait->tcp_time_wait_tail = tcp; 173 mutex_exit(&tcp_time_wait->tcp_time_wait_lock); 174 } 175 176 /* 177 * Wrapper to call tcp_close_detached() via squeue to clean up TIME-WAIT 178 * tcp_t. Used in tcp_time_wait_collector(). 179 */ 180 /* ARGSUSED */ 181 static void 182 tcp_timewait_close(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy) 183 { 184 conn_t *connp = (conn_t *)arg; 185 tcp_t *tcp = connp->conn_tcp; 186 187 ASSERT(tcp != NULL); 188 if (tcp->tcp_state == TCPS_CLOSED) { 189 return; 190 } 191 192 ASSERT((connp->conn_family == AF_INET && 193 connp->conn_ipversion == IPV4_VERSION) || 194 (connp->conn_family == AF_INET6 && 195 (connp->conn_ipversion == IPV4_VERSION || 196 connp->conn_ipversion == IPV6_VERSION))); 197 ASSERT(!tcp->tcp_listener); 198 199 ASSERT(TCP_IS_DETACHED(tcp)); 200 201 /* 202 * Because they have no upstream client to rebind or tcp_close() 203 * them later, we axe the connection here and now. 204 */ 205 tcp_close_detached(tcp); 206 } 207 208 /* 209 * Blows away all tcps whose TIME_WAIT has expired. List traversal 210 * is done forwards from the head. 211 * This walks all stack instances since 212 * tcp_time_wait remains global across all stacks. 213 */ 214 /* ARGSUSED */ 215 void 216 tcp_time_wait_collector(void *arg) 217 { 218 tcp_t *tcp; 219 clock_t now; 220 mblk_t *mp; 221 conn_t *connp; 222 kmutex_t *lock; 223 boolean_t removed; 224 extern void (*cl_inet_disconnect)(netstackid_t, uint8_t, sa_family_t, 225 uint8_t *, in_port_t, uint8_t *, in_port_t, void *); 226 227 squeue_t *sqp = (squeue_t *)arg; 228 tcp_squeue_priv_t *tcp_time_wait = 229 *((tcp_squeue_priv_t **)squeue_getprivate(sqp, SQPRIVATE_TCP)); 230 231 mutex_enter(&tcp_time_wait->tcp_time_wait_lock); 232 tcp_time_wait->tcp_time_wait_tid = 0; 233 234 if (tcp_time_wait->tcp_free_list != NULL && 235 tcp_time_wait->tcp_free_list->tcp_in_free_list == B_TRUE) { 236 TCP_G_STAT(tcp_freelist_cleanup); 237 while ((tcp = tcp_time_wait->tcp_free_list) != NULL) { 238 tcp_time_wait->tcp_free_list = tcp->tcp_time_wait_next; 239 tcp->tcp_time_wait_next = NULL; 240 tcp_time_wait->tcp_free_list_cnt--; 241 ASSERT(tcp->tcp_tcps == NULL); 242 CONN_DEC_REF(tcp->tcp_connp); 243 } 244 ASSERT(tcp_time_wait->tcp_free_list_cnt == 0); 245 } 246 247 /* 248 * In order to reap time waits reliably, we should use a 249 * source of time that is not adjustable by the user -- hence 250 * the call to ddi_get_lbolt(). 251 */ 252 now = ddi_get_lbolt(); 253 while ((tcp = tcp_time_wait->tcp_time_wait_head) != NULL) { 254 /* 255 * Compare times using modular arithmetic, since 256 * lbolt can wrapover. 257 */ 258 if ((now - tcp->tcp_time_wait_expire) < 0) { 259 break; 260 } 261 262 removed = tcp_time_wait_remove(tcp, tcp_time_wait); 263 ASSERT(removed); 264 265 connp = tcp->tcp_connp; 266 ASSERT(connp->conn_fanout != NULL); 267 lock = &connp->conn_fanout->connf_lock; 268 /* 269 * This is essentially a TW reclaim fast path optimization for 270 * performance where the timewait collector checks under the 271 * fanout lock (so that no one else can get access to the 272 * conn_t) that the refcnt is 2 i.e. one for TCP and one for 273 * the classifier hash list. If ref count is indeed 2, we can 274 * just remove the conn under the fanout lock and avoid 275 * cleaning up the conn under the squeue, provided that 276 * clustering callbacks are not enabled. If clustering is 277 * enabled, we need to make the clustering callback before 278 * setting the CONDEMNED flag and after dropping all locks and 279 * so we forego this optimization and fall back to the slow 280 * path. Also please see the comments in tcp_closei_local 281 * regarding the refcnt logic. 282 * 283 * Since we are holding the tcp_time_wait_lock, its better 284 * not to block on the fanout_lock because other connections 285 * can't add themselves to time_wait list. So we do a 286 * tryenter instead of mutex_enter. 287 */ 288 if (mutex_tryenter(lock)) { 289 mutex_enter(&connp->conn_lock); 290 if ((connp->conn_ref == 2) && 291 (cl_inet_disconnect == NULL)) { 292 ipcl_hash_remove_locked(connp, 293 connp->conn_fanout); 294 /* 295 * Set the CONDEMNED flag now itself so that 296 * the refcnt cannot increase due to any 297 * walker. 298 */ 299 connp->conn_state_flags |= CONN_CONDEMNED; 300 mutex_exit(lock); 301 mutex_exit(&connp->conn_lock); 302 if (tcp_time_wait->tcp_free_list_cnt < 303 tcp_free_list_max_cnt) { 304 /* Add to head of tcp_free_list */ 305 mutex_exit( 306 &tcp_time_wait->tcp_time_wait_lock); 307 tcp_cleanup(tcp); 308 ASSERT(connp->conn_latch == NULL); 309 ASSERT(connp->conn_policy == NULL); 310 ASSERT(tcp->tcp_tcps == NULL); 311 ASSERT(connp->conn_netstack == NULL); 312 313 mutex_enter( 314 &tcp_time_wait->tcp_time_wait_lock); 315 tcp->tcp_time_wait_next = 316 tcp_time_wait->tcp_free_list; 317 tcp_time_wait->tcp_free_list = tcp; 318 tcp_time_wait->tcp_free_list_cnt++; 319 continue; 320 } else { 321 /* Do not add to tcp_free_list */ 322 mutex_exit( 323 &tcp_time_wait->tcp_time_wait_lock); 324 tcp_bind_hash_remove(tcp); 325 ixa_cleanup(tcp->tcp_connp->conn_ixa); 326 tcp_ipsec_cleanup(tcp); 327 CONN_DEC_REF(tcp->tcp_connp); 328 } 329 } else { 330 CONN_INC_REF_LOCKED(connp); 331 mutex_exit(lock); 332 mutex_exit(&tcp_time_wait->tcp_time_wait_lock); 333 mutex_exit(&connp->conn_lock); 334 /* 335 * We can reuse the closemp here since conn has 336 * detached (otherwise we wouldn't even be in 337 * time_wait list). tcp_closemp_used can safely 338 * be changed without taking a lock as no other 339 * thread can concurrently access it at this 340 * point in the connection lifecycle. 341 */ 342 343 if (tcp->tcp_closemp.b_prev == NULL) 344 tcp->tcp_closemp_used = B_TRUE; 345 else 346 cmn_err(CE_PANIC, 347 "tcp_timewait_collector: " 348 "concurrent use of tcp_closemp: " 349 "connp %p tcp %p\n", (void *)connp, 350 (void *)tcp); 351 352 TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15); 353 mp = &tcp->tcp_closemp; 354 SQUEUE_ENTER_ONE(connp->conn_sqp, mp, 355 tcp_timewait_close, connp, NULL, 356 SQ_FILL, SQTAG_TCP_TIMEWAIT); 357 } 358 } else { 359 mutex_enter(&connp->conn_lock); 360 CONN_INC_REF_LOCKED(connp); 361 mutex_exit(&tcp_time_wait->tcp_time_wait_lock); 362 mutex_exit(&connp->conn_lock); 363 /* 364 * We can reuse the closemp here since conn has 365 * detached (otherwise we wouldn't even be in 366 * time_wait list). tcp_closemp_used can safely 367 * be changed without taking a lock as no other 368 * thread can concurrently access it at this 369 * point in the connection lifecycle. 370 */ 371 372 if (tcp->tcp_closemp.b_prev == NULL) 373 tcp->tcp_closemp_used = B_TRUE; 374 else 375 cmn_err(CE_PANIC, "tcp_timewait_collector: " 376 "concurrent use of tcp_closemp: " 377 "connp %p tcp %p\n", (void *)connp, 378 (void *)tcp); 379 380 TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15); 381 mp = &tcp->tcp_closemp; 382 SQUEUE_ENTER_ONE(connp->conn_sqp, mp, 383 tcp_timewait_close, connp, NULL, 384 SQ_FILL, SQTAG_TCP_TIMEWAIT); 385 } 386 mutex_enter(&tcp_time_wait->tcp_time_wait_lock); 387 } 388 389 if (tcp_time_wait->tcp_free_list != NULL) 390 tcp_time_wait->tcp_free_list->tcp_in_free_list = B_TRUE; 391 392 tcp_time_wait->tcp_time_wait_tid = 393 timeout_generic(CALLOUT_NORMAL, tcp_time_wait_collector, sqp, 394 TCP_TIME_WAIT_DELAY, CALLOUT_TCP_RESOLUTION, 395 CALLOUT_FLAG_ROUNDUP); 396 mutex_exit(&tcp_time_wait->tcp_time_wait_lock); 397 } 398 399 /* 400 * tcp_time_wait_processing() handles processing of incoming packets when 401 * the tcp_t is in the TIME_WAIT state. 402 * 403 * A TIME_WAIT tcp_t that has an associated open TCP end point (not in 404 * detached state) is never put on the time wait list. 405 */ 406 void 407 tcp_time_wait_processing(tcp_t *tcp, mblk_t *mp, uint32_t seg_seq, 408 uint32_t seg_ack, int seg_len, tcpha_t *tcpha, ip_recv_attr_t *ira) 409 { 410 int32_t bytes_acked; 411 int32_t gap; 412 int32_t rgap; 413 tcp_opt_t tcpopt; 414 uint_t flags; 415 uint32_t new_swnd = 0; 416 conn_t *nconnp; 417 conn_t *connp = tcp->tcp_connp; 418 tcp_stack_t *tcps = tcp->tcp_tcps; 419 420 BUMP_LOCAL(tcp->tcp_ibsegs); 421 DTRACE_PROBE2(tcp__trace__recv, mblk_t *, mp, tcp_t *, tcp); 422 423 flags = (unsigned int)tcpha->tha_flags & 0xFF; 424 new_swnd = ntohs(tcpha->tha_win) << 425 ((tcpha->tha_flags & TH_SYN) ? 0 : tcp->tcp_snd_ws); 426 if (tcp->tcp_snd_ts_ok) { 427 if (!tcp_paws_check(tcp, tcpha, &tcpopt)) { 428 tcp_xmit_ctl(NULL, tcp, tcp->tcp_snxt, 429 tcp->tcp_rnxt, TH_ACK); 430 goto done; 431 } 432 } 433 gap = seg_seq - tcp->tcp_rnxt; 434 rgap = tcp->tcp_rwnd - (gap + seg_len); 435 if (gap < 0) { 436 TCPS_BUMP_MIB(tcps, tcpInDataDupSegs); 437 TCPS_UPDATE_MIB(tcps, tcpInDataDupBytes, 438 (seg_len > -gap ? -gap : seg_len)); 439 seg_len += gap; 440 if (seg_len < 0 || (seg_len == 0 && !(flags & TH_FIN))) { 441 if (flags & TH_RST) { 442 goto done; 443 } 444 if ((flags & TH_FIN) && seg_len == -1) { 445 /* 446 * When TCP receives a duplicate FIN in 447 * TIME_WAIT state, restart the 2 MSL timer. 448 * See page 73 in RFC 793. Make sure this TCP 449 * is already on the TIME_WAIT list. If not, 450 * just restart the timer. 451 */ 452 if (TCP_IS_DETACHED(tcp)) { 453 if (tcp_time_wait_remove(tcp, NULL) == 454 B_TRUE) { 455 tcp_time_wait_append(tcp); 456 TCP_DBGSTAT(tcps, 457 tcp_rput_time_wait); 458 } 459 } else { 460 ASSERT(tcp != NULL); 461 TCP_TIMER_RESTART(tcp, 462 tcps->tcps_time_wait_interval); 463 } 464 tcp_xmit_ctl(NULL, tcp, tcp->tcp_snxt, 465 tcp->tcp_rnxt, TH_ACK); 466 goto done; 467 } 468 flags |= TH_ACK_NEEDED; 469 seg_len = 0; 470 goto process_ack; 471 } 472 473 /* Fix seg_seq, and chew the gap off the front. */ 474 seg_seq = tcp->tcp_rnxt; 475 } 476 477 if ((flags & TH_SYN) && gap > 0 && rgap < 0) { 478 /* 479 * Make sure that when we accept the connection, pick 480 * an ISS greater than (tcp_snxt + ISS_INCR/2) for the 481 * old connection. 482 * 483 * The next ISS generated is equal to tcp_iss_incr_extra 484 * + ISS_INCR/2 + other components depending on the 485 * value of tcp_strong_iss. We pre-calculate the new 486 * ISS here and compare with tcp_snxt to determine if 487 * we need to make adjustment to tcp_iss_incr_extra. 488 * 489 * The above calculation is ugly and is a 490 * waste of CPU cycles... 491 */ 492 uint32_t new_iss = tcps->tcps_iss_incr_extra; 493 int32_t adj; 494 ip_stack_t *ipst = tcps->tcps_netstack->netstack_ip; 495 496 switch (tcps->tcps_strong_iss) { 497 case 2: { 498 /* Add time and MD5 components. */ 499 uint32_t answer[4]; 500 struct { 501 uint32_t ports; 502 in6_addr_t src; 503 in6_addr_t dst; 504 } arg; 505 MD5_CTX context; 506 507 mutex_enter(&tcps->tcps_iss_key_lock); 508 context = tcps->tcps_iss_key; 509 mutex_exit(&tcps->tcps_iss_key_lock); 510 arg.ports = connp->conn_ports; 511 /* We use MAPPED addresses in tcp_iss_init */ 512 arg.src = connp->conn_laddr_v6; 513 arg.dst = connp->conn_faddr_v6; 514 MD5Update(&context, (uchar_t *)&arg, 515 sizeof (arg)); 516 MD5Final((uchar_t *)answer, &context); 517 answer[0] ^= answer[1] ^ answer[2] ^ answer[3]; 518 new_iss += (gethrtime() >> ISS_NSEC_SHT) + answer[0]; 519 break; 520 } 521 case 1: 522 /* Add time component and min random (i.e. 1). */ 523 new_iss += (gethrtime() >> ISS_NSEC_SHT) + 1; 524 break; 525 default: 526 /* Add only time component. */ 527 new_iss += (uint32_t)gethrestime_sec() * ISS_INCR; 528 break; 529 } 530 if ((adj = (int32_t)(tcp->tcp_snxt - new_iss)) > 0) { 531 /* 532 * New ISS not guaranteed to be ISS_INCR/2 533 * ahead of the current tcp_snxt, so add the 534 * difference to tcp_iss_incr_extra. 535 */ 536 tcps->tcps_iss_incr_extra += adj; 537 } 538 /* 539 * If tcp_clean_death() can not perform the task now, 540 * drop the SYN packet and let the other side re-xmit. 541 * Otherwise pass the SYN packet back in, since the 542 * old tcp state has been cleaned up or freed. 543 */ 544 if (tcp_clean_death(tcp, 0) == -1) 545 goto done; 546 nconnp = ipcl_classify(mp, ira, ipst); 547 if (nconnp != NULL) { 548 TCP_STAT(tcps, tcp_time_wait_syn_success); 549 /* Drops ref on nconnp */ 550 tcp_reinput(nconnp, mp, ira, ipst); 551 return; 552 } 553 goto done; 554 } 555 556 /* 557 * rgap is the amount of stuff received out of window. A negative 558 * value is the amount out of window. 559 */ 560 if (rgap < 0) { 561 TCPS_BUMP_MIB(tcps, tcpInDataPastWinSegs); 562 TCPS_UPDATE_MIB(tcps, tcpInDataPastWinBytes, -rgap); 563 /* Fix seg_len and make sure there is something left. */ 564 seg_len += rgap; 565 if (seg_len <= 0) { 566 if (flags & TH_RST) { 567 goto done; 568 } 569 flags |= TH_ACK_NEEDED; 570 seg_len = 0; 571 goto process_ack; 572 } 573 } 574 /* 575 * Check whether we can update tcp_ts_recent. This test is 576 * NOT the one in RFC 1323 3.4. It is from Braden, 1993, "TCP 577 * Extensions for High Performance: An Update", Internet Draft. 578 */ 579 if (tcp->tcp_snd_ts_ok && 580 TSTMP_GEQ(tcpopt.tcp_opt_ts_val, tcp->tcp_ts_recent) && 581 SEQ_LEQ(seg_seq, tcp->tcp_rack)) { 582 tcp->tcp_ts_recent = tcpopt.tcp_opt_ts_val; 583 tcp->tcp_last_rcv_lbolt = ddi_get_lbolt64(); 584 } 585 586 if (seg_seq != tcp->tcp_rnxt && seg_len > 0) { 587 /* Always ack out of order packets */ 588 flags |= TH_ACK_NEEDED; 589 seg_len = 0; 590 } else if (seg_len > 0) { 591 TCPS_BUMP_MIB(tcps, tcpInClosed); 592 TCPS_BUMP_MIB(tcps, tcpInDataInorderSegs); 593 TCPS_UPDATE_MIB(tcps, tcpInDataInorderBytes, seg_len); 594 } 595 if (flags & TH_RST) { 596 (void) tcp_clean_death(tcp, 0); 597 goto done; 598 } 599 if (flags & TH_SYN) { 600 tcp_xmit_ctl("TH_SYN", tcp, seg_ack, seg_seq + 1, 601 TH_RST|TH_ACK); 602 /* 603 * Do not delete the TCP structure if it is in 604 * TIME_WAIT state. Refer to RFC 1122, 4.2.2.13. 605 */ 606 goto done; 607 } 608 process_ack: 609 if (flags & TH_ACK) { 610 bytes_acked = (int)(seg_ack - tcp->tcp_suna); 611 if (bytes_acked <= 0) { 612 if (bytes_acked == 0 && seg_len == 0 && 613 new_swnd == tcp->tcp_swnd) 614 TCPS_BUMP_MIB(tcps, tcpInDupAck); 615 } else { 616 /* Acks something not sent */ 617 flags |= TH_ACK_NEEDED; 618 } 619 } 620 if (flags & TH_ACK_NEEDED) { 621 /* 622 * Time to send an ack for some reason. 623 */ 624 tcp_xmit_ctl(NULL, tcp, tcp->tcp_snxt, 625 tcp->tcp_rnxt, TH_ACK); 626 } 627 done: 628 freemsg(mp); 629 } 630