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("$FreeBSD$"); 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 #include <sys/protosw.h> 51 #include <sys/random.h> 52 53 #include <vm/uma.h> 54 55 #include <net/route.h> 56 #include <net/if.h> 57 #include <net/if_var.h> 58 #include <net/vnet.h> 59 60 #include <netinet/in.h> 61 #include <netinet/in_pcb.h> 62 #include <netinet/in_systm.h> 63 #include <netinet/in_var.h> 64 #include <netinet/ip.h> 65 #include <netinet/ip_icmp.h> 66 #include <netinet/ip_var.h> 67 #ifdef INET6 68 #include <netinet/ip6.h> 69 #include <netinet6/in6_pcb.h> 70 #include <netinet6/ip6_var.h> 71 #include <netinet6/scope6_var.h> 72 #include <netinet6/nd6.h> 73 #endif 74 #include <netinet/tcp.h> 75 #include <netinet/tcp_fsm.h> 76 #include <netinet/tcp_seq.h> 77 #include <netinet/tcp_timer.h> 78 #include <netinet/tcp_var.h> 79 #ifdef INET6 80 #include <netinet6/tcp6_var.h> 81 #endif 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif 86 #ifdef INET6 87 #include <netinet6/ip6protosw.h> 88 #endif 89 90 #include <machine/in_cksum.h> 91 92 #include <security/mac/mac_framework.h> 93 94 static VNET_DEFINE(uma_zone_t, tcptw_zone); 95 #define V_tcptw_zone VNET(tcptw_zone) 96 static int maxtcptw; 97 98 /* 99 * The timed wait queue contains references to each of the TCP sessions 100 * currently in the TIME_WAIT state. The queue pointers, including the 101 * queue pointers in each tcptw structure, are protected using the global 102 * timewait lock, which must be held over queue iteration and modification. 103 * 104 * Rules on tcptw usage: 105 * - a inpcb is always freed _after_ its tcptw 106 * - a tcptw relies on its inpcb reference counting for memory stability 107 * - a tcptw is dereferenceable only while its inpcb is locked 108 */ 109 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl); 110 #define V_twq_2msl VNET(twq_2msl) 111 112 /* Global timewait lock */ 113 static VNET_DEFINE(struct rwlock, tw_lock); 114 #define V_tw_lock VNET(tw_lock) 115 116 #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0) 117 #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw)) 118 #define TW_RLOCK(tw) rw_rlock(&(tw)) 119 #define TW_WLOCK(tw) rw_wlock(&(tw)) 120 #define TW_RUNLOCK(tw) rw_runlock(&(tw)) 121 #define TW_WUNLOCK(tw) rw_wunlock(&(tw)) 122 #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED) 123 #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED) 124 #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED) 125 #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED) 126 127 static void tcp_tw_2msl_reset(struct tcptw *, int); 128 static void tcp_tw_2msl_stop(struct tcptw *, int); 129 static int tcp_twrespond(struct tcptw *, int); 130 131 static int 132 tcptw_auto_size(void) 133 { 134 int halfrange; 135 136 /* 137 * Max out at half the ephemeral port range so that TIME_WAIT 138 * sockets don't tie up too many ephemeral ports. 139 */ 140 if (V_ipport_lastauto > V_ipport_firstauto) 141 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 142 else 143 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 144 /* Protect against goofy port ranges smaller than 32. */ 145 return (imin(imax(halfrange, 32), maxsockets / 5)); 146 } 147 148 static int 149 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 150 { 151 int error, new; 152 153 if (maxtcptw == 0) 154 new = tcptw_auto_size(); 155 else 156 new = maxtcptw; 157 error = sysctl_handle_int(oidp, &new, 0, req); 158 if (error == 0 && req->newptr) 159 if (new >= 32) { 160 maxtcptw = new; 161 uma_zone_set_max(V_tcptw_zone, maxtcptw); 162 } 163 return (error); 164 } 165 166 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 167 &maxtcptw, 0, sysctl_maxtcptw, "IU", 168 "Maximum number of compressed TCP TIME_WAIT entries"); 169 170 VNET_DEFINE(int, nolocaltimewait) = 0; 171 #define V_nolocaltimewait VNET(nolocaltimewait) 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(nolocaltimewait), 0, 174 "Do not create compressed TCP TIME_WAIT entries for local connections"); 175 176 void 177 tcp_tw_zone_change(void) 178 { 179 180 if (maxtcptw == 0) 181 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 182 } 183 184 void 185 tcp_tw_init(void) 186 { 187 188 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 189 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 190 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 191 if (maxtcptw == 0) 192 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 193 else 194 uma_zone_set_max(V_tcptw_zone, maxtcptw); 195 TAILQ_INIT(&V_twq_2msl); 196 TW_LOCK_INIT(V_tw_lock, "tcptw"); 197 } 198 199 #ifdef VIMAGE 200 void 201 tcp_tw_destroy(void) 202 { 203 struct tcptw *tw; 204 205 INP_INFO_WLOCK(&V_tcbinfo); 206 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL) 207 tcp_twclose(tw, 0); 208 INP_INFO_WUNLOCK(&V_tcbinfo); 209 210 TW_LOCK_DESTROY(V_tw_lock); 211 uma_zdestroy(V_tcptw_zone); 212 } 213 #endif 214 215 /* 216 * Move a TCP connection into TIME_WAIT state. 217 * tcbinfo is locked. 218 * inp is locked, and is unlocked before returning. 219 */ 220 void 221 tcp_twstart(struct tcpcb *tp) 222 { 223 struct tcptw *tw; 224 struct inpcb *inp = tp->t_inpcb; 225 int acknow; 226 struct socket *so; 227 #ifdef INET6 228 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 229 #endif 230 231 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 232 INP_WLOCK_ASSERT(inp); 233 234 if (V_nolocaltimewait) { 235 int error = 0; 236 #ifdef INET6 237 if (isipv6) 238 error = in6_localaddr(&inp->in6p_faddr); 239 #endif 240 #if defined(INET6) && defined(INET) 241 else 242 #endif 243 #ifdef INET 244 error = in_localip(inp->inp_faddr); 245 #endif 246 if (error) { 247 tp = tcp_close(tp); 248 if (tp != NULL) 249 INP_WUNLOCK(inp); 250 return; 251 } 252 } 253 254 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 255 if (tw == NULL) { 256 /* 257 * Reached limit on total number of TIMEWAIT connections 258 * allowed. Remove a connection from TIMEWAIT queue in LRU 259 * fashion to make room for this connection. 260 * 261 * pcbinfo lock is needed here to prevent deadlock as 262 * two inpcb locks can be acquired simultaneously. 263 */ 264 tw = tcp_tw_2msl_scan(1); 265 if (tw == NULL) { 266 tp = tcp_close(tp); 267 if (tp != NULL) 268 INP_WUNLOCK(inp); 269 return; 270 } 271 } 272 /* 273 * The tcptw will hold a reference on its inpcb until tcp_twclose 274 * is called 275 */ 276 tw->tw_inpcb = inp; 277 in_pcbref(inp); /* Reference from tw */ 278 279 /* 280 * Recover last window size sent. 281 */ 282 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 283 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 284 else 285 tw->last_win = 0; 286 287 /* 288 * Set t_recent if timestamps are used on the connection. 289 */ 290 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 291 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 292 tw->t_recent = tp->ts_recent; 293 tw->ts_offset = tp->ts_offset; 294 } else { 295 tw->t_recent = 0; 296 tw->ts_offset = 0; 297 } 298 299 tw->snd_nxt = tp->snd_nxt; 300 tw->rcv_nxt = tp->rcv_nxt; 301 tw->iss = tp->iss; 302 tw->irs = tp->irs; 303 tw->t_starttime = tp->t_starttime; 304 tw->tw_time = 0; 305 306 /* XXX 307 * If this code will 308 * be used for fin-wait-2 state also, then we may need 309 * a ts_recent from the last segment. 310 */ 311 acknow = tp->t_flags & TF_ACKNOW; 312 313 /* 314 * First, discard tcpcb state, which includes stopping its timers and 315 * freeing it. tcp_discardcb() used to also release the inpcb, but 316 * that work is now done in the caller. 317 * 318 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 319 * and might not be needed here any longer. 320 */ 321 tcp_discardcb(tp); 322 so = inp->inp_socket; 323 soisdisconnected(so); 324 tw->tw_cred = crhold(so->so_cred); 325 SOCK_LOCK(so); 326 tw->tw_so_options = so->so_options; 327 SOCK_UNLOCK(so); 328 if (acknow) 329 tcp_twrespond(tw, TH_ACK); 330 inp->inp_ppcb = tw; 331 inp->inp_flags |= INP_TIMEWAIT; 332 tcp_tw_2msl_reset(tw, 0); 333 334 /* 335 * If the inpcb owns the sole reference to the socket, then we can 336 * detach and free the socket as it is not needed in time wait. 337 */ 338 if (inp->inp_flags & INP_SOCKREF) { 339 KASSERT(so->so_state & SS_PROTOREF, 340 ("tcp_twstart: !SS_PROTOREF")); 341 inp->inp_flags &= ~INP_SOCKREF; 342 INP_WUNLOCK(inp); 343 ACCEPT_LOCK(); 344 SOCK_LOCK(so); 345 so->so_state &= ~SS_PROTOREF; 346 sofree(so); 347 } else 348 INP_WUNLOCK(inp); 349 } 350 351 /* 352 * Returns 1 if the TIME_WAIT state was killed and we should start over, 353 * looking for a pcb in the listen state. Returns 0 otherwise. 354 */ 355 int 356 tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th, 357 struct mbuf *m, int tlen) 358 { 359 struct tcptw *tw; 360 int thflags; 361 tcp_seq seq; 362 363 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 364 INP_WLOCK_ASSERT(inp); 365 366 /* 367 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 368 * still present. This is undesirable, but temporarily necessary 369 * until we work out how to handle inpcb's who's timewait state has 370 * been removed. 371 */ 372 tw = intotw(inp); 373 if (tw == NULL) 374 goto drop; 375 376 thflags = th->th_flags; 377 378 /* 379 * NOTE: for FIN_WAIT_2 (to be added later), 380 * must validate sequence number before accepting RST 381 */ 382 383 /* 384 * If the segment contains RST: 385 * Drop the segment - see Stevens, vol. 2, p. 964 and 386 * RFC 1337. 387 */ 388 if (thflags & TH_RST) 389 goto drop; 390 391 #if 0 392 /* PAWS not needed at the moment */ 393 /* 394 * RFC 1323 PAWS: If we have a timestamp reply on this segment 395 * and it's less than ts_recent, drop it. 396 */ 397 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 398 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 399 if ((thflags & TH_ACK) == 0) 400 goto drop; 401 goto ack; 402 } 403 /* 404 * ts_recent is never updated because we never accept new segments. 405 */ 406 #endif 407 408 /* 409 * If a new connection request is received 410 * while in TIME_WAIT, drop the old connection 411 * and start over if the sequence numbers 412 * are above the previous ones. 413 */ 414 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 415 tcp_twclose(tw, 0); 416 return (1); 417 } 418 419 /* 420 * Drop the segment if it does not contain an ACK. 421 */ 422 if ((thflags & TH_ACK) == 0) 423 goto drop; 424 425 /* 426 * Reset the 2MSL timer if this is a duplicate FIN. 427 */ 428 if (thflags & TH_FIN) { 429 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 430 if (seq + 1 == tw->rcv_nxt) 431 tcp_tw_2msl_reset(tw, 1); 432 } 433 434 /* 435 * Acknowledge the segment if it has data or is not a duplicate ACK. 436 */ 437 if (thflags != TH_ACK || tlen != 0 || 438 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 439 tcp_twrespond(tw, TH_ACK); 440 drop: 441 INP_WUNLOCK(inp); 442 m_freem(m); 443 return (0); 444 } 445 446 void 447 tcp_twclose(struct tcptw *tw, int reuse) 448 { 449 struct socket *so; 450 struct inpcb *inp; 451 452 /* 453 * At this point, we are in one of two situations: 454 * 455 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 456 * all state. 457 * 458 * (2) We have a socket -- if we own a reference, release it and 459 * notify the socket layer. 460 */ 461 inp = tw->tw_inpcb; 462 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 463 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 464 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */ 465 INP_WLOCK_ASSERT(inp); 466 467 tcp_tw_2msl_stop(tw, reuse); 468 inp->inp_ppcb = NULL; 469 in_pcbdrop(inp); 470 471 so = inp->inp_socket; 472 if (so != NULL) { 473 /* 474 * If there's a socket, handle two cases: first, we own a 475 * strong reference, which we will now release, or we don't 476 * in which case another reference exists (XXXRW: think 477 * about this more), and we don't need to take action. 478 */ 479 if (inp->inp_flags & INP_SOCKREF) { 480 inp->inp_flags &= ~INP_SOCKREF; 481 INP_WUNLOCK(inp); 482 ACCEPT_LOCK(); 483 SOCK_LOCK(so); 484 KASSERT(so->so_state & SS_PROTOREF, 485 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 486 so->so_state &= ~SS_PROTOREF; 487 sofree(so); 488 } else { 489 /* 490 * If we don't own the only reference, the socket and 491 * inpcb need to be left around to be handled by 492 * tcp_usr_detach() later. 493 */ 494 INP_WUNLOCK(inp); 495 } 496 } else { 497 /* 498 * The socket has been already cleaned-up for us, only free the 499 * inpcb. 500 */ 501 in_pcbfree(inp); 502 } 503 TCPSTAT_INC(tcps_closed); 504 } 505 506 static int 507 tcp_twrespond(struct tcptw *tw, int flags) 508 { 509 struct inpcb *inp = tw->tw_inpcb; 510 #if defined(INET6) || defined(INET) 511 struct tcphdr *th = NULL; 512 #endif 513 struct mbuf *m; 514 #ifdef INET 515 struct ip *ip = NULL; 516 #endif 517 u_int hdrlen, optlen; 518 int error = 0; /* Keep compiler happy */ 519 struct tcpopt to; 520 #ifdef INET6 521 struct ip6_hdr *ip6 = NULL; 522 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 523 #endif 524 hdrlen = 0; /* Keep compiler happy */ 525 526 INP_WLOCK_ASSERT(inp); 527 528 m = m_gethdr(M_NOWAIT, MT_DATA); 529 if (m == NULL) 530 return (ENOBUFS); 531 m->m_data += max_linkhdr; 532 533 #ifdef MAC 534 mac_inpcb_create_mbuf(inp, m); 535 #endif 536 537 #ifdef INET6 538 if (isipv6) { 539 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 540 ip6 = mtod(m, struct ip6_hdr *); 541 th = (struct tcphdr *)(ip6 + 1); 542 tcpip_fillheaders(inp, ip6, th); 543 } 544 #endif 545 #if defined(INET6) && defined(INET) 546 else 547 #endif 548 #ifdef INET 549 { 550 hdrlen = sizeof(struct tcpiphdr); 551 ip = mtod(m, struct ip *); 552 th = (struct tcphdr *)(ip + 1); 553 tcpip_fillheaders(inp, ip, th); 554 } 555 #endif 556 to.to_flags = 0; 557 558 /* 559 * Send a timestamp and echo-reply if both our side and our peer 560 * have sent timestamps in our SYN's and this is not a RST. 561 */ 562 if (tw->t_recent && flags == TH_ACK) { 563 to.to_flags |= TOF_TS; 564 to.to_tsval = tcp_ts_getticks() + tw->ts_offset; 565 to.to_tsecr = tw->t_recent; 566 } 567 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 568 569 m->m_len = hdrlen + optlen; 570 m->m_pkthdr.len = m->m_len; 571 572 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 573 574 th->th_seq = htonl(tw->snd_nxt); 575 th->th_ack = htonl(tw->rcv_nxt); 576 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 577 th->th_flags = flags; 578 th->th_win = htons(tw->last_win); 579 580 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 581 #ifdef INET6 582 if (isipv6) { 583 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 584 th->th_sum = in6_cksum_pseudo(ip6, 585 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0); 586 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 587 error = ip6_output(m, inp->in6p_outputopts, NULL, 588 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 589 } 590 #endif 591 #if defined(INET6) && defined(INET) 592 else 593 #endif 594 #ifdef INET 595 { 596 m->m_pkthdr.csum_flags = CSUM_TCP; 597 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 598 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 599 ip->ip_len = htons(m->m_pkthdr.len); 600 if (V_path_mtu_discovery) 601 ip->ip_off |= htons(IP_DF); 602 error = ip_output(m, inp->inp_options, NULL, 603 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 604 NULL, inp); 605 } 606 #endif 607 if (flags & TH_ACK) 608 TCPSTAT_INC(tcps_sndacks); 609 else 610 TCPSTAT_INC(tcps_sndctrl); 611 TCPSTAT_INC(tcps_sndtotal); 612 return (error); 613 } 614 615 static void 616 tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 617 { 618 619 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 620 INP_WLOCK_ASSERT(tw->tw_inpcb); 621 622 TW_WLOCK(V_tw_lock); 623 if (rearm) 624 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 625 tw->tw_time = ticks + 2 * tcp_msl; 626 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 627 TW_WUNLOCK(V_tw_lock); 628 } 629 630 static void 631 tcp_tw_2msl_stop(struct tcptw *tw, int reuse) 632 { 633 struct ucred *cred; 634 struct inpcb *inp; 635 int released; 636 637 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 638 639 TW_WLOCK(V_tw_lock); 640 inp = tw->tw_inpcb; 641 tw->tw_inpcb = NULL; 642 643 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 644 cred = tw->tw_cred; 645 tw->tw_cred = NULL; 646 TW_WUNLOCK(V_tw_lock); 647 648 if (cred != NULL) 649 crfree(cred); 650 651 released = in_pcbrele_wlocked(inp); 652 KASSERT(!released, ("%s: inp should not be released here", __func__)); 653 654 if (!reuse) 655 uma_zfree(V_tcptw_zone, tw); 656 } 657 658 struct tcptw * 659 tcp_tw_2msl_scan(int reuse) 660 { 661 struct tcptw *tw; 662 struct inpcb *inp; 663 664 #ifdef INVARIANTS 665 if (reuse) { 666 /* 667 * pcbinfo lock is needed in reuse case to prevent deadlock 668 * as two inpcb locks can be acquired simultaneously: 669 * - the inpcb transitioning to TIME_WAIT state in 670 * tcp_tw_start(), 671 * - the inpcb closed by tcp_twclose(). 672 */ 673 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 674 } 675 #endif 676 677 for (;;) { 678 TW_RLOCK(V_tw_lock); 679 tw = TAILQ_FIRST(&V_twq_2msl); 680 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) { 681 TW_RUNLOCK(V_tw_lock); 682 break; 683 } 684 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL", 685 __func__)); 686 687 inp = tw->tw_inpcb; 688 in_pcbref(inp); 689 TW_RUNLOCK(V_tw_lock); 690 691 if (INP_INFO_TRY_WLOCK(&V_tcbinfo)) { 692 693 INP_WLOCK(inp); 694 tw = intotw(inp); 695 if (in_pcbrele_wlocked(inp)) { 696 KASSERT(tw == NULL, ("%s: held last inp " 697 "reference but tw not NULL", __func__)); 698 INP_INFO_WUNLOCK(&V_tcbinfo); 699 continue; 700 } 701 702 if (tw == NULL) { 703 /* tcp_twclose() has already been called */ 704 INP_WUNLOCK(inp); 705 INP_INFO_WUNLOCK(&V_tcbinfo); 706 continue; 707 } 708 709 tcp_twclose(tw, reuse); 710 INP_INFO_WUNLOCK(&V_tcbinfo); 711 if (reuse) 712 return tw; 713 } else { 714 /* INP_INFO lock is busy, continue later. */ 715 INP_WLOCK(inp); 716 if (!in_pcbrele_wlocked(inp)) 717 INP_WUNLOCK(inp); 718 break; 719 } 720 } 721 722 return NULL; 723 } 724