tcp_timewait.c revision 191548
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: head/sys/netinet/tcp_timewait.c 191548 2009-04-26 22:06:42Z zec $"); 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/callout.h> 43#include <sys/kernel.h> 44#include <sys/sysctl.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/priv.h> 48#include <sys/proc.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/protosw.h> 52#include <sys/random.h> 53#include <sys/vimage.h> 54 55#include <vm/uma.h> 56 57#include <net/route.h> 58#include <net/if.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/ip.h> 63#ifdef INET6 64#include <netinet/ip6.h> 65#endif 66#include <netinet/in_pcb.h> 67#ifdef INET6 68#include <netinet6/in6_pcb.h> 69#endif 70#include <netinet/in_var.h> 71#include <netinet/ip_var.h> 72#ifdef INET6 73#include <netinet6/ip6_var.h> 74#include <netinet6/scope6_var.h> 75#include <netinet6/nd6.h> 76#endif 77#include <netinet/ip_icmp.h> 78#include <netinet/tcp.h> 79#include <netinet/tcp_fsm.h> 80#include <netinet/tcp_seq.h> 81#include <netinet/tcp_timer.h> 82#include <netinet/tcp_var.h> 83#ifdef INET6 84#include <netinet6/tcp6_var.h> 85#endif 86#include <netinet/tcpip.h> 87#ifdef TCPDEBUG 88#include <netinet/tcp_debug.h> 89#endif 90#include <netinet6/ip6protosw.h> 91#include <netinet/vinet.h> 92 93#include <machine/in_cksum.h> 94 95#include <security/mac/mac_framework.h> 96 97static int maxtcptw; 98 99/* 100 * The timed wait queue contains references to each of the TCP sessions 101 * currently in the TIME_WAIT state. The queue pointers, including the 102 * queue pointers in each tcptw structure, are protected using the global 103 * tcbinfo lock, which must be held over queue iteration and modification. 104 */ 105#ifdef VIMAGE_GLOBALS 106static uma_zone_t tcptw_zone; 107static TAILQ_HEAD(, tcptw) twq_2msl; 108int nolocaltimewait; 109#endif 110 111static void tcp_tw_2msl_reset(struct tcptw *, int); 112static void tcp_tw_2msl_stop(struct tcptw *); 113 114static int 115tcptw_auto_size(void) 116{ 117 INIT_VNET_INET(curvnet); 118 int halfrange; 119 120 /* 121 * Max out at half the ephemeral port range so that TIME_WAIT 122 * sockets don't tie up too many ephemeral ports. 123 */ 124 if (V_ipport_lastauto > V_ipport_firstauto) 125 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 126 else 127 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 128 /* Protect against goofy port ranges smaller than 32. */ 129 return (imin(imax(halfrange, 32), maxsockets / 5)); 130} 131 132static int 133sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 134{ 135 INIT_VNET_INET(curvnet); 136 int error, new; 137 138 if (maxtcptw == 0) 139 new = tcptw_auto_size(); 140 else 141 new = maxtcptw; 142 error = sysctl_handle_int(oidp, &new, 0, req); 143 if (error == 0 && req->newptr) 144 if (new >= 32) { 145 maxtcptw = new; 146 uma_zone_set_max(V_tcptw_zone, maxtcptw); 147 } 148 return (error); 149} 150 151SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 152 &maxtcptw, 0, sysctl_maxtcptw, "IU", 153 "Maximum number of compressed TCP TIME_WAIT entries"); 154 155SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, nolocaltimewait, 156 CTLFLAG_RW, nolocaltimewait, 0, 157 "Do not create compressed TCP TIME_WAIT entries for local connections"); 158 159void 160tcp_tw_zone_change(void) 161{ 162 INIT_VNET_INET(curvnet); 163 164 if (maxtcptw == 0) 165 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 166} 167 168void 169tcp_tw_init(void) 170{ 171 INIT_VNET_INET(curvnet); 172 173 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 174 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 175 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 176 if (maxtcptw == 0) 177 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 178 else 179 uma_zone_set_max(V_tcptw_zone, maxtcptw); 180 TAILQ_INIT(&V_twq_2msl); 181} 182 183/* 184 * Move a TCP connection into TIME_WAIT state. 185 * tcbinfo is locked. 186 * inp is locked, and is unlocked before returning. 187 */ 188void 189tcp_twstart(struct tcpcb *tp) 190{ 191#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 192 INIT_VNET_INET(tp->t_vnet); 193#endif 194 struct tcptw *tw; 195 struct inpcb *inp = tp->t_inpcb; 196 int acknow; 197 struct socket *so; 198 199 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_reset(). */ 200 INP_WLOCK_ASSERT(inp); 201 202 if (V_nolocaltimewait && in_localip(inp->inp_faddr)) { 203 tp = tcp_close(tp); 204 if (tp != NULL) 205 INP_WUNLOCK(inp); 206 return; 207 } 208 209 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 210 if (tw == NULL) { 211 tw = tcp_tw_2msl_scan(1); 212 if (tw == NULL) { 213 tp = tcp_close(tp); 214 if (tp != NULL) 215 INP_WUNLOCK(inp); 216 return; 217 } 218 } 219 tw->tw_inpcb = inp; 220 221 /* 222 * Recover last window size sent. 223 */ 224 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 225 226 /* 227 * Set t_recent if timestamps are used on the connection. 228 */ 229 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 230 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 231 tw->t_recent = tp->ts_recent; 232 tw->ts_offset = tp->ts_offset; 233 } else { 234 tw->t_recent = 0; 235 tw->ts_offset = 0; 236 } 237 238 tw->snd_nxt = tp->snd_nxt; 239 tw->rcv_nxt = tp->rcv_nxt; 240 tw->iss = tp->iss; 241 tw->irs = tp->irs; 242 tw->t_starttime = tp->t_starttime; 243 tw->tw_time = 0; 244 245/* XXX 246 * If this code will 247 * be used for fin-wait-2 state also, then we may need 248 * a ts_recent from the last segment. 249 */ 250 acknow = tp->t_flags & TF_ACKNOW; 251 252 /* 253 * First, discard tcpcb state, which includes stopping its timers and 254 * freeing it. tcp_discardcb() used to also release the inpcb, but 255 * that work is now done in the caller. 256 * 257 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 258 * and might not be needed here any longer. 259 */ 260 tcp_discardcb(tp); 261 so = inp->inp_socket; 262 soisdisconnected(so); 263 tw->tw_cred = crhold(so->so_cred); 264 SOCK_LOCK(so); 265 tw->tw_so_options = so->so_options; 266 SOCK_UNLOCK(so); 267 if (acknow) 268 tcp_twrespond(tw, TH_ACK); 269 inp->inp_ppcb = tw; 270 inp->inp_flags |= INP_TIMEWAIT; 271 tcp_tw_2msl_reset(tw, 0); 272 273 /* 274 * If the inpcb owns the sole reference to the socket, then we can 275 * detach and free the socket as it is not needed in time wait. 276 */ 277 if (inp->inp_flags & INP_SOCKREF) { 278 KASSERT(so->so_state & SS_PROTOREF, 279 ("tcp_twstart: !SS_PROTOREF")); 280 inp->inp_flags &= ~INP_SOCKREF; 281 INP_WUNLOCK(inp); 282 ACCEPT_LOCK(); 283 SOCK_LOCK(so); 284 so->so_state &= ~SS_PROTOREF; 285 sofree(so); 286 } else 287 INP_WUNLOCK(inp); 288} 289 290#if 0 291/* 292 * The appromixate rate of ISN increase of Microsoft TCP stacks; 293 * the actual rate is slightly higher due to the addition of 294 * random positive increments. 295 * 296 * Most other new OSes use semi-randomized ISN values, so we 297 * do not need to worry about them. 298 */ 299#define MS_ISN_BYTES_PER_SECOND 250000 300 301/* 302 * Determine if the ISN we will generate has advanced beyond the last 303 * sequence number used by the previous connection. If so, indicate 304 * that it is safe to recycle this tw socket by returning 1. 305 */ 306int 307tcp_twrecycleable(struct tcptw *tw) 308{ 309 INIT_VNET_INET(curvnet); 310 tcp_seq new_iss = tw->iss; 311 tcp_seq new_irs = tw->irs; 312 313 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 314 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 315 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 316 317 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 318 return (1); 319 else 320 return (0); 321} 322#endif 323 324/* 325 * Returns 1 if the TIME_WAIT state was killed and we should start over, 326 * looking for a pcb in the listen state. Returns 0 otherwise. 327 */ 328int 329tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 330 struct mbuf *m, int tlen) 331{ 332#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 333 INIT_VNET_INET(curvnet); 334#endif 335 struct tcptw *tw; 336 int thflags; 337 tcp_seq seq; 338 339 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */ 340 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 341 INP_WLOCK_ASSERT(inp); 342 343 /* 344 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 345 * still present. This is undesirable, but temporarily necessary 346 * until we work out how to handle inpcb's who's timewait state has 347 * been removed. 348 */ 349 tw = intotw(inp); 350 if (tw == NULL) 351 goto drop; 352 353 thflags = th->th_flags; 354 355 /* 356 * NOTE: for FIN_WAIT_2 (to be added later), 357 * must validate sequence number before accepting RST 358 */ 359 360 /* 361 * If the segment contains RST: 362 * Drop the segment - see Stevens, vol. 2, p. 964 and 363 * RFC 1337. 364 */ 365 if (thflags & TH_RST) 366 goto drop; 367 368#if 0 369/* PAWS not needed at the moment */ 370 /* 371 * RFC 1323 PAWS: If we have a timestamp reply on this segment 372 * and it's less than ts_recent, drop it. 373 */ 374 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 375 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 376 if ((thflags & TH_ACK) == 0) 377 goto drop; 378 goto ack; 379 } 380 /* 381 * ts_recent is never updated because we never accept new segments. 382 */ 383#endif 384 385 /* 386 * If a new connection request is received 387 * while in TIME_WAIT, drop the old connection 388 * and start over if the sequence numbers 389 * are above the previous ones. 390 */ 391 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 392 tcp_twclose(tw, 0); 393 return (1); 394 } 395 396 /* 397 * Drop the the segment if it does not contain an ACK. 398 */ 399 if ((thflags & TH_ACK) == 0) 400 goto drop; 401 402 /* 403 * Reset the 2MSL timer if this is a duplicate FIN. 404 */ 405 if (thflags & TH_FIN) { 406 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 407 if (seq + 1 == tw->rcv_nxt) 408 tcp_tw_2msl_reset(tw, 1); 409 } 410 411 /* 412 * Acknowledge the segment if it has data or is not a duplicate ACK. 413 */ 414 if (thflags != TH_ACK || tlen != 0 || 415 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 416 tcp_twrespond(tw, TH_ACK); 417drop: 418 INP_WUNLOCK(inp); 419 m_freem(m); 420 return (0); 421} 422 423void 424tcp_twclose(struct tcptw *tw, int reuse) 425{ 426 INIT_VNET_INET(curvnet); 427 struct socket *so; 428 struct inpcb *inp; 429 430 /* 431 * At this point, we are in one of two situations: 432 * 433 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 434 * all state. 435 * 436 * (2) We have a socket -- if we own a reference, release it and 437 * notify the socket layer. 438 */ 439 inp = tw->tw_inpcb; 440 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 441 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 442 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_stop(). */ 443 INP_WLOCK_ASSERT(inp); 444 445 tw->tw_inpcb = NULL; 446 tcp_tw_2msl_stop(tw); 447 inp->inp_ppcb = NULL; 448 in_pcbdrop(inp); 449 450 so = inp->inp_socket; 451 if (so != NULL) { 452 /* 453 * If there's a socket, handle two cases: first, we own a 454 * strong reference, which we will now release, or we don't 455 * in which case another reference exists (XXXRW: think 456 * about this more), and we don't need to take action. 457 */ 458 if (inp->inp_flags & INP_SOCKREF) { 459 inp->inp_flags &= ~INP_SOCKREF; 460 INP_WUNLOCK(inp); 461 ACCEPT_LOCK(); 462 SOCK_LOCK(so); 463 KASSERT(so->so_state & SS_PROTOREF, 464 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 465 so->so_state &= ~SS_PROTOREF; 466 sofree(so); 467 } else { 468 /* 469 * If we don't own the only reference, the socket and 470 * inpcb need to be left around to be handled by 471 * tcp_usr_detach() later. 472 */ 473 INP_WUNLOCK(inp); 474 } 475 } else 476 in_pcbfree(inp); 477 TCPSTAT_INC(tcps_closed); 478 crfree(tw->tw_cred); 479 tw->tw_cred = NULL; 480 if (reuse) 481 return; 482 uma_zfree(V_tcptw_zone, tw); 483} 484 485int 486tcp_twrespond(struct tcptw *tw, int flags) 487{ 488 INIT_VNET_INET(curvnet); 489 struct inpcb *inp = tw->tw_inpcb; 490 struct tcphdr *th; 491 struct mbuf *m; 492 struct ip *ip = NULL; 493 u_int hdrlen, optlen; 494 int error; 495 struct tcpopt to; 496#ifdef INET6 497 struct ip6_hdr *ip6 = NULL; 498 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 499#endif 500 501 INP_WLOCK_ASSERT(inp); 502 503 m = m_gethdr(M_DONTWAIT, MT_DATA); 504 if (m == NULL) 505 return (ENOBUFS); 506 m->m_data += max_linkhdr; 507 508#ifdef MAC 509 mac_inpcb_create_mbuf(inp, m); 510#endif 511 512#ifdef INET6 513 if (isipv6) { 514 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 515 ip6 = mtod(m, struct ip6_hdr *); 516 th = (struct tcphdr *)(ip6 + 1); 517 tcpip_fillheaders(inp, ip6, th); 518 } else 519#endif 520 { 521 hdrlen = sizeof(struct tcpiphdr); 522 ip = mtod(m, struct ip *); 523 th = (struct tcphdr *)(ip + 1); 524 tcpip_fillheaders(inp, ip, th); 525 } 526 to.to_flags = 0; 527 528 /* 529 * Send a timestamp and echo-reply if both our side and our peer 530 * have sent timestamps in our SYN's and this is not a RST. 531 */ 532 if (tw->t_recent && flags == TH_ACK) { 533 to.to_flags |= TOF_TS; 534 to.to_tsval = ticks + tw->ts_offset; 535 to.to_tsecr = tw->t_recent; 536 } 537 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 538 539 m->m_len = hdrlen + optlen; 540 m->m_pkthdr.len = m->m_len; 541 542 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 543 544 th->th_seq = htonl(tw->snd_nxt); 545 th->th_ack = htonl(tw->rcv_nxt); 546 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 547 th->th_flags = flags; 548 th->th_win = htons(tw->last_win); 549 550#ifdef INET6 551 if (isipv6) { 552 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 553 sizeof(struct tcphdr) + optlen); 554 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 555 error = ip6_output(m, inp->in6p_outputopts, NULL, 556 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 557 } else 558#endif 559 { 560 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 561 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 562 m->m_pkthdr.csum_flags = CSUM_TCP; 563 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 564 ip->ip_len = m->m_pkthdr.len; 565 if (V_path_mtu_discovery) 566 ip->ip_off |= IP_DF; 567 error = ip_output(m, inp->inp_options, NULL, 568 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 569 NULL, inp); 570 } 571 if (flags & TH_ACK) 572 TCPSTAT_INC(tcps_sndacks); 573 else 574 TCPSTAT_INC(tcps_sndctrl); 575 TCPSTAT_INC(tcps_sndtotal); 576 return (error); 577} 578 579static void 580tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 581{ 582 INIT_VNET_INET(curvnet); 583 584 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 585 INP_WLOCK_ASSERT(tw->tw_inpcb); 586 if (rearm) 587 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 588 tw->tw_time = ticks + 2 * tcp_msl; 589 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 590} 591 592static void 593tcp_tw_2msl_stop(struct tcptw *tw) 594{ 595 INIT_VNET_INET(curvnet); 596 597 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 598 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 599} 600 601struct tcptw * 602tcp_tw_2msl_scan(int reuse) 603{ 604 INIT_VNET_INET(curvnet); 605 struct tcptw *tw; 606 607 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 608 for (;;) { 609 tw = TAILQ_FIRST(&V_twq_2msl); 610 if (tw == NULL || (!reuse && tw->tw_time > ticks)) 611 break; 612 INP_WLOCK(tw->tw_inpcb); 613 tcp_twclose(tw, reuse); 614 if (reuse) 615 return (tw); 616 } 617 return (NULL); 618} 619