tcp_timewait.c revision 324948
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: stable/11/sys/netinet/tcp_timewait.c 324948 2017-10-24 08:56:11Z jch $"); 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_tcpdebug.h" 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/callout.h> 42#include <sys/kernel.h> 43#include <sys/sysctl.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/priv.h> 47#include <sys/proc.h> 48#include <sys/socket.h> 49#include <sys/socketvar.h> 50#ifndef INVARIANTS 51#include <sys/syslog.h> 52#endif 53#include <sys/protosw.h> 54#include <sys/random.h> 55 56#include <vm/uma.h> 57 58#include <net/route.h> 59#include <net/if.h> 60#include <net/if_var.h> 61#include <net/vnet.h> 62 63#include <netinet/in.h> 64#include <netinet/in_pcb.h> 65#include <netinet/in_systm.h> 66#include <netinet/in_var.h> 67#include <netinet/ip.h> 68#include <netinet/ip_icmp.h> 69#include <netinet/ip_var.h> 70#ifdef INET6 71#include <netinet/ip6.h> 72#include <netinet6/in6_pcb.h> 73#include <netinet6/ip6_var.h> 74#include <netinet6/scope6_var.h> 75#include <netinet6/nd6.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#ifdef INET6 83#include <netinet6/tcp6_var.h> 84#endif 85#include <netinet/tcpip.h> 86#ifdef TCPDEBUG 87#include <netinet/tcp_debug.h> 88#endif 89#ifdef INET6 90#include <netinet6/ip6protosw.h> 91#endif 92 93#include <machine/in_cksum.h> 94 95#include <security/mac/mac_framework.h> 96 97static VNET_DEFINE(uma_zone_t, tcptw_zone); 98#define V_tcptw_zone VNET(tcptw_zone) 99static int maxtcptw; 100 101/* 102 * The timed wait queue contains references to each of the TCP sessions 103 * currently in the TIME_WAIT state. The queue pointers, including the 104 * queue pointers in each tcptw structure, are protected using the global 105 * timewait lock, which must be held over queue iteration and modification. 106 * 107 * Rules on tcptw usage: 108 * - a inpcb is always freed _after_ its tcptw 109 * - a tcptw relies on its inpcb reference counting for memory stability 110 * - a tcptw is dereferenceable only while its inpcb is locked 111 */ 112static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl); 113#define V_twq_2msl VNET(twq_2msl) 114 115/* Global timewait lock */ 116static VNET_DEFINE(struct rwlock, tw_lock); 117#define V_tw_lock VNET(tw_lock) 118 119#define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0) 120#define TW_LOCK_DESTROY(tw) rw_destroy(&(tw)) 121#define TW_RLOCK(tw) rw_rlock(&(tw)) 122#define TW_WLOCK(tw) rw_wlock(&(tw)) 123#define TW_RUNLOCK(tw) rw_runlock(&(tw)) 124#define TW_WUNLOCK(tw) rw_wunlock(&(tw)) 125#define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED) 126#define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED) 127#define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED) 128#define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED) 129 130static void tcp_tw_2msl_reset(struct tcptw *, int); 131static void tcp_tw_2msl_stop(struct tcptw *, int); 132static int tcp_twrespond(struct tcptw *, int); 133 134static int 135tcptw_auto_size(void) 136{ 137 int halfrange; 138 139 /* 140 * Max out at half the ephemeral port range so that TIME_WAIT 141 * sockets don't tie up too many ephemeral ports. 142 */ 143 if (V_ipport_lastauto > V_ipport_firstauto) 144 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 145 else 146 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 147 /* Protect against goofy port ranges smaller than 32. */ 148 return (imin(imax(halfrange, 32), maxsockets / 5)); 149} 150 151static int 152sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 153{ 154 int error, new; 155 156 if (maxtcptw == 0) 157 new = tcptw_auto_size(); 158 else 159 new = maxtcptw; 160 error = sysctl_handle_int(oidp, &new, 0, req); 161 if (error == 0 && req->newptr) 162 if (new >= 32) { 163 maxtcptw = new; 164 uma_zone_set_max(V_tcptw_zone, maxtcptw); 165 } 166 return (error); 167} 168 169SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 170 &maxtcptw, 0, sysctl_maxtcptw, "IU", 171 "Maximum number of compressed TCP TIME_WAIT entries"); 172 173VNET_DEFINE(int, nolocaltimewait) = 0; 174#define V_nolocaltimewait VNET(nolocaltimewait) 175SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(nolocaltimewait), 0, 177 "Do not create compressed TCP TIME_WAIT entries for local connections"); 178 179void 180tcp_tw_zone_change(void) 181{ 182 183 if (maxtcptw == 0) 184 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 185} 186 187void 188tcp_tw_init(void) 189{ 190 191 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 192 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 193 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 194 if (maxtcptw == 0) 195 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 196 else 197 uma_zone_set_max(V_tcptw_zone, maxtcptw); 198 TAILQ_INIT(&V_twq_2msl); 199 TW_LOCK_INIT(V_tw_lock, "tcptw"); 200} 201 202#ifdef VIMAGE 203void 204tcp_tw_destroy(void) 205{ 206 struct tcptw *tw; 207 208 INP_INFO_RLOCK(&V_tcbinfo); 209 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL) 210 tcp_twclose(tw, 0); 211 INP_INFO_RUNLOCK(&V_tcbinfo); 212 213 TW_LOCK_DESTROY(V_tw_lock); 214 uma_zdestroy(V_tcptw_zone); 215} 216#endif 217 218/* 219 * Move a TCP connection into TIME_WAIT state. 220 * tcbinfo is locked. 221 * inp is locked, and is unlocked before returning. 222 */ 223void 224tcp_twstart(struct tcpcb *tp) 225{ 226 struct tcptw *tw; 227 struct inpcb *inp = tp->t_inpcb; 228 int acknow; 229 struct socket *so; 230#ifdef INET6 231 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 232#endif 233 234 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 235 INP_WLOCK_ASSERT(inp); 236 237 /* A dropped inp should never transition to TIME_WAIT state. */ 238 KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: " 239 "(inp->inp_flags & INP_DROPPED) != 0")); 240 241 if (V_nolocaltimewait) { 242 int error = 0; 243#ifdef INET6 244 if (isipv6) 245 error = in6_localaddr(&inp->in6p_faddr); 246#endif 247#if defined(INET6) && defined(INET) 248 else 249#endif 250#ifdef INET 251 error = in_localip(inp->inp_faddr); 252#endif 253 if (error) { 254 tp = tcp_close(tp); 255 if (tp != NULL) 256 INP_WUNLOCK(inp); 257 return; 258 } 259 } 260 261 262 /* 263 * For use only by DTrace. We do not reference the state 264 * after this point so modifying it in place is not a problem. 265 */ 266 tcp_state_change(tp, TCPS_TIME_WAIT); 267 268 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 269 if (tw == NULL) { 270 /* 271 * Reached limit on total number of TIMEWAIT connections 272 * allowed. Remove a connection from TIMEWAIT queue in LRU 273 * fashion to make room for this connection. 274 * 275 * XXX: Check if it possible to always have enough room 276 * in advance based on guarantees provided by uma_zalloc(). 277 */ 278 tw = tcp_tw_2msl_scan(1); 279 if (tw == NULL) { 280 tp = tcp_close(tp); 281 if (tp != NULL) 282 INP_WUNLOCK(inp); 283 return; 284 } 285 } 286 /* 287 * The tcptw will hold a reference on its inpcb until tcp_twclose 288 * is called 289 */ 290 tw->tw_inpcb = inp; 291 in_pcbref(inp); /* Reference from tw */ 292 293 /* 294 * Recover last window size sent. 295 */ 296 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 297 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 298 else 299 tw->last_win = 0; 300 301 /* 302 * Set t_recent if timestamps are used on the connection. 303 */ 304 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 305 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 306 tw->t_recent = tp->ts_recent; 307 tw->ts_offset = tp->ts_offset; 308 } else { 309 tw->t_recent = 0; 310 tw->ts_offset = 0; 311 } 312 313 tw->snd_nxt = tp->snd_nxt; 314 tw->rcv_nxt = tp->rcv_nxt; 315 tw->iss = tp->iss; 316 tw->irs = tp->irs; 317 tw->t_starttime = tp->t_starttime; 318 tw->tw_time = 0; 319 320/* XXX 321 * If this code will 322 * be used for fin-wait-2 state also, then we may need 323 * a ts_recent from the last segment. 324 */ 325 acknow = tp->t_flags & TF_ACKNOW; 326 327 /* 328 * First, discard tcpcb state, which includes stopping its timers and 329 * freeing it. tcp_discardcb() used to also release the inpcb, but 330 * that work is now done in the caller. 331 * 332 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 333 * and might not be needed here any longer. 334 */ 335 tcp_discardcb(tp); 336 so = inp->inp_socket; 337 soisdisconnected(so); 338 tw->tw_cred = crhold(so->so_cred); 339 SOCK_LOCK(so); 340 tw->tw_so_options = so->so_options; 341 SOCK_UNLOCK(so); 342 if (acknow) 343 tcp_twrespond(tw, TH_ACK); 344 inp->inp_ppcb = tw; 345 inp->inp_flags |= INP_TIMEWAIT; 346 TCPSTATES_INC(TCPS_TIME_WAIT); 347 tcp_tw_2msl_reset(tw, 0); 348 349 /* 350 * If the inpcb owns the sole reference to the socket, then we can 351 * detach and free the socket as it is not needed in time wait. 352 */ 353 if (inp->inp_flags & INP_SOCKREF) { 354 KASSERT(so->so_state & SS_PROTOREF, 355 ("tcp_twstart: !SS_PROTOREF")); 356 inp->inp_flags &= ~INP_SOCKREF; 357 INP_WUNLOCK(inp); 358 ACCEPT_LOCK(); 359 SOCK_LOCK(so); 360 so->so_state &= ~SS_PROTOREF; 361 sofree(so); 362 } else 363 INP_WUNLOCK(inp); 364} 365 366/* 367 * Returns 1 if the TIME_WAIT state was killed and we should start over, 368 * looking for a pcb in the listen state. Returns 0 otherwise. 369 */ 370int 371tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th, 372 struct mbuf *m, int tlen) 373{ 374 struct tcptw *tw; 375 int thflags; 376 tcp_seq seq; 377 378 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 379 INP_WLOCK_ASSERT(inp); 380 381 /* 382 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 383 * still present. This is undesirable, but temporarily necessary 384 * until we work out how to handle inpcb's who's timewait state has 385 * been removed. 386 */ 387 tw = intotw(inp); 388 if (tw == NULL) 389 goto drop; 390 391 thflags = th->th_flags; 392 393 /* 394 * NOTE: for FIN_WAIT_2 (to be added later), 395 * must validate sequence number before accepting RST 396 */ 397 398 /* 399 * If the segment contains RST: 400 * Drop the segment - see Stevens, vol. 2, p. 964 and 401 * RFC 1337. 402 */ 403 if (thflags & TH_RST) 404 goto drop; 405 406#if 0 407/* PAWS not needed at the moment */ 408 /* 409 * RFC 1323 PAWS: If we have a timestamp reply on this segment 410 * and it's less than ts_recent, drop it. 411 */ 412 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 413 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 414 if ((thflags & TH_ACK) == 0) 415 goto drop; 416 goto ack; 417 } 418 /* 419 * ts_recent is never updated because we never accept new segments. 420 */ 421#endif 422 423 /* 424 * If a new connection request is received 425 * while in TIME_WAIT, drop the old connection 426 * and start over if the sequence numbers 427 * are above the previous ones. 428 */ 429 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 430 tcp_twclose(tw, 0); 431 return (1); 432 } 433 434 /* 435 * Drop the segment if it does not contain an ACK. 436 */ 437 if ((thflags & TH_ACK) == 0) 438 goto drop; 439 440 /* 441 * Reset the 2MSL timer if this is a duplicate FIN. 442 */ 443 if (thflags & TH_FIN) { 444 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 445 if (seq + 1 == tw->rcv_nxt) 446 tcp_tw_2msl_reset(tw, 1); 447 } 448 449 /* 450 * Acknowledge the segment if it has data or is not a duplicate ACK. 451 */ 452 if (thflags != TH_ACK || tlen != 0 || 453 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 454 tcp_twrespond(tw, TH_ACK); 455drop: 456 INP_WUNLOCK(inp); 457 m_freem(m); 458 return (0); 459} 460 461void 462tcp_twclose(struct tcptw *tw, int reuse) 463{ 464 struct socket *so; 465 struct inpcb *inp; 466 467 /* 468 * At this point, we are in one of two situations: 469 * 470 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 471 * all state. 472 * 473 * (2) We have a socket -- if we own a reference, release it and 474 * notify the socket layer. 475 */ 476 inp = tw->tw_inpcb; 477 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 478 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 479 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */ 480 INP_WLOCK_ASSERT(inp); 481 482 tcp_tw_2msl_stop(tw, reuse); 483 inp->inp_ppcb = NULL; 484 in_pcbdrop(inp); 485 486 so = inp->inp_socket; 487 if (so != NULL) { 488 /* 489 * If there's a socket, handle two cases: first, we own a 490 * strong reference, which we will now release, or we don't 491 * in which case another reference exists (XXXRW: think 492 * about this more), and we don't need to take action. 493 */ 494 if (inp->inp_flags & INP_SOCKREF) { 495 inp->inp_flags &= ~INP_SOCKREF; 496 INP_WUNLOCK(inp); 497 ACCEPT_LOCK(); 498 SOCK_LOCK(so); 499 KASSERT(so->so_state & SS_PROTOREF, 500 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 501 so->so_state &= ~SS_PROTOREF; 502 sofree(so); 503 } else { 504 /* 505 * If we don't own the only reference, the socket and 506 * inpcb need to be left around to be handled by 507 * tcp_usr_detach() later. 508 */ 509 INP_WUNLOCK(inp); 510 } 511 } else { 512 /* 513 * The socket has been already cleaned-up for us, only free the 514 * inpcb. 515 */ 516 in_pcbfree(inp); 517 } 518 TCPSTAT_INC(tcps_closed); 519} 520 521static int 522tcp_twrespond(struct tcptw *tw, int flags) 523{ 524 struct inpcb *inp = tw->tw_inpcb; 525#if defined(INET6) || defined(INET) 526 struct tcphdr *th = NULL; 527#endif 528 struct mbuf *m; 529#ifdef INET 530 struct ip *ip = NULL; 531#endif 532 u_int hdrlen, optlen; 533 int error = 0; /* Keep compiler happy */ 534 struct tcpopt to; 535#ifdef INET6 536 struct ip6_hdr *ip6 = NULL; 537 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 538#endif 539 hdrlen = 0; /* Keep compiler happy */ 540 541 INP_WLOCK_ASSERT(inp); 542 543 m = m_gethdr(M_NOWAIT, MT_DATA); 544 if (m == NULL) 545 return (ENOBUFS); 546 m->m_data += max_linkhdr; 547 548#ifdef MAC 549 mac_inpcb_create_mbuf(inp, m); 550#endif 551 552#ifdef INET6 553 if (isipv6) { 554 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 555 ip6 = mtod(m, struct ip6_hdr *); 556 th = (struct tcphdr *)(ip6 + 1); 557 tcpip_fillheaders(inp, ip6, th); 558 } 559#endif 560#if defined(INET6) && defined(INET) 561 else 562#endif 563#ifdef INET 564 { 565 hdrlen = sizeof(struct tcpiphdr); 566 ip = mtod(m, struct ip *); 567 th = (struct tcphdr *)(ip + 1); 568 tcpip_fillheaders(inp, ip, th); 569 } 570#endif 571 to.to_flags = 0; 572 573 /* 574 * Send a timestamp and echo-reply if both our side and our peer 575 * have sent timestamps in our SYN's and this is not a RST. 576 */ 577 if (tw->t_recent && flags == TH_ACK) { 578 to.to_flags |= TOF_TS; 579 to.to_tsval = tcp_ts_getticks() + tw->ts_offset; 580 to.to_tsecr = tw->t_recent; 581 } 582 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 583 584 m->m_len = hdrlen + optlen; 585 m->m_pkthdr.len = m->m_len; 586 587 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 588 589 th->th_seq = htonl(tw->snd_nxt); 590 th->th_ack = htonl(tw->rcv_nxt); 591 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 592 th->th_flags = flags; 593 th->th_win = htons(tw->last_win); 594 595 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 596#ifdef INET6 597 if (isipv6) { 598 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 599 th->th_sum = in6_cksum_pseudo(ip6, 600 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0); 601 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 602 error = ip6_output(m, inp->in6p_outputopts, NULL, 603 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 604 } 605#endif 606#if defined(INET6) && defined(INET) 607 else 608#endif 609#ifdef INET 610 { 611 m->m_pkthdr.csum_flags = CSUM_TCP; 612 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 613 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 614 ip->ip_len = htons(m->m_pkthdr.len); 615 if (V_path_mtu_discovery) 616 ip->ip_off |= htons(IP_DF); 617 error = ip_output(m, inp->inp_options, NULL, 618 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 619 NULL, inp); 620 } 621#endif 622 if (flags & TH_ACK) 623 TCPSTAT_INC(tcps_sndacks); 624 else 625 TCPSTAT_INC(tcps_sndctrl); 626 TCPSTAT_INC(tcps_sndtotal); 627 return (error); 628} 629 630static void 631tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 632{ 633 634 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 635 INP_WLOCK_ASSERT(tw->tw_inpcb); 636 637 TW_WLOCK(V_tw_lock); 638 if (rearm) 639 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 640 tw->tw_time = ticks + 2 * tcp_msl; 641 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 642 TW_WUNLOCK(V_tw_lock); 643} 644 645static void 646tcp_tw_2msl_stop(struct tcptw *tw, int reuse) 647{ 648 struct ucred *cred; 649 struct inpcb *inp; 650 int released; 651 652 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 653 654 TW_WLOCK(V_tw_lock); 655 inp = tw->tw_inpcb; 656 tw->tw_inpcb = NULL; 657 658 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 659 cred = tw->tw_cred; 660 tw->tw_cred = NULL; 661 TW_WUNLOCK(V_tw_lock); 662 663 if (cred != NULL) 664 crfree(cred); 665 666 released = in_pcbrele_wlocked(inp); 667 KASSERT(!released, ("%s: inp should not be released here", __func__)); 668 669 if (!reuse) 670 uma_zfree(V_tcptw_zone, tw); 671 TCPSTATES_DEC(TCPS_TIME_WAIT); 672} 673 674struct tcptw * 675tcp_tw_2msl_scan(int reuse) 676{ 677 struct tcptw *tw; 678 struct inpcb *inp; 679 680#ifdef INVARIANTS 681 if (reuse) { 682 /* 683 * Exclusive pcbinfo lock is not required in reuse case even if 684 * two inpcb locks can be acquired simultaneously: 685 * - the inpcb transitioning to TIME_WAIT state in 686 * tcp_tw_start(), 687 * - the inpcb closed by tcp_twclose(). 688 * 689 * It is because only inpcbs in FIN_WAIT2 or CLOSING states can 690 * transition in TIME_WAIT state. Then a pcbcb cannot be in 691 * TIME_WAIT list and transitioning to TIME_WAIT state at same 692 * time. 693 */ 694 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 695 } 696#endif 697 698 for (;;) { 699 TW_RLOCK(V_tw_lock); 700 tw = TAILQ_FIRST(&V_twq_2msl); 701 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) { 702 TW_RUNLOCK(V_tw_lock); 703 break; 704 } 705 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL", 706 __func__)); 707 708 inp = tw->tw_inpcb; 709 in_pcbref(inp); 710 TW_RUNLOCK(V_tw_lock); 711 712 if (INP_INFO_TRY_RLOCK(&V_tcbinfo)) { 713 714 INP_WLOCK(inp); 715 tw = intotw(inp); 716 if (in_pcbrele_wlocked(inp)) { 717 if (__predict_true(tw == NULL)) { 718 INP_INFO_RUNLOCK(&V_tcbinfo); 719 continue; 720 } else { 721 /* This should not happen as in TIMEWAIT 722 * state the inp should not be destroyed 723 * before its tcptw. If INVARIANTS is 724 * defined panic. 725 */ 726#ifdef INVARIANTS 727 panic("%s: Panic before an infinite " 728 "loop: INP_TIMEWAIT && (INP_FREED " 729 "|| inp last reference) && tw != " 730 "NULL", __func__); 731#else 732 log(LOG_ERR, "%s: Avoid an infinite " 733 "loop: INP_TIMEWAIT && (INP_FREED " 734 "|| inp last reference) && tw != " 735 "NULL", __func__); 736#endif 737 INP_INFO_RUNLOCK(&V_tcbinfo); 738 break; 739 } 740 } 741 742 if (tw == NULL) { 743 /* tcp_twclose() has already been called */ 744 INP_WUNLOCK(inp); 745 INP_INFO_RUNLOCK(&V_tcbinfo); 746 continue; 747 } 748 749 tcp_twclose(tw, reuse); 750 INP_INFO_RUNLOCK(&V_tcbinfo); 751 if (reuse) 752 return tw; 753 } else { 754 /* INP_INFO lock is busy, continue later. */ 755 INP_WLOCK(inp); 756 if (!in_pcbrele_wlocked(inp)) 757 INP_WUNLOCK(inp); 758 break; 759 } 760 } 761 762 return NULL; 763} 764