tcp_timewait.c revision 193511
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 193511 2009-06-05 14:55:22Z rwatson $"); 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#include <sys/vimage.h> 53 54#include <vm/uma.h> 55 56#include <net/route.h> 57#include <net/if.h> 58 59#include <netinet/in.h> 60#include <netinet/in_systm.h> 61#include <netinet/ip.h> 62#ifdef INET6 63#include <netinet/ip6.h> 64#endif 65#include <netinet/in_pcb.h> 66#ifdef INET6 67#include <netinet6/in6_pcb.h> 68#endif 69#include <netinet/in_var.h> 70#include <netinet/ip_var.h> 71#ifdef INET6 72#include <netinet6/ip6_var.h> 73#include <netinet6/scope6_var.h> 74#include <netinet6/nd6.h> 75#endif 76#include <netinet/ip_icmp.h> 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#include <netinet6/ip6protosw.h> 90#include <netinet/vinet.h> 91 92#include <machine/in_cksum.h> 93 94#include <security/mac/mac_framework.h> 95 96static 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 * tcbinfo lock, which must be held over queue iteration and modification. 103 */ 104#ifdef VIMAGE_GLOBALS 105static uma_zone_t tcptw_zone; 106static TAILQ_HEAD(, tcptw) twq_2msl; 107int nolocaltimewait; 108#endif 109 110static void tcp_tw_2msl_reset(struct tcptw *, int); 111static void tcp_tw_2msl_stop(struct tcptw *); 112 113static int 114tcptw_auto_size(void) 115{ 116 INIT_VNET_INET(curvnet); 117 int halfrange; 118 119 /* 120 * Max out at half the ephemeral port range so that TIME_WAIT 121 * sockets don't tie up too many ephemeral ports. 122 */ 123 if (V_ipport_lastauto > V_ipport_firstauto) 124 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 125 else 126 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 127 /* Protect against goofy port ranges smaller than 32. */ 128 return (imin(imax(halfrange, 32), maxsockets / 5)); 129} 130 131static int 132sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 133{ 134 INIT_VNET_INET(curvnet); 135 int error, new; 136 137 if (maxtcptw == 0) 138 new = tcptw_auto_size(); 139 else 140 new = maxtcptw; 141 error = sysctl_handle_int(oidp, &new, 0, req); 142 if (error == 0 && req->newptr) 143 if (new >= 32) { 144 maxtcptw = new; 145 uma_zone_set_max(V_tcptw_zone, maxtcptw); 146 } 147 return (error); 148} 149 150SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 151 &maxtcptw, 0, sysctl_maxtcptw, "IU", 152 "Maximum number of compressed TCP TIME_WAIT entries"); 153 154SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, nolocaltimewait, 155 CTLFLAG_RW, nolocaltimewait, 0, 156 "Do not create compressed TCP TIME_WAIT entries for local connections"); 157 158void 159tcp_tw_zone_change(void) 160{ 161 INIT_VNET_INET(curvnet); 162 163 if (maxtcptw == 0) 164 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 165} 166 167void 168tcp_tw_init(void) 169{ 170 INIT_VNET_INET(curvnet); 171 172 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 173 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 174 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 175 if (maxtcptw == 0) 176 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 177 else 178 uma_zone_set_max(V_tcptw_zone, maxtcptw); 179 TAILQ_INIT(&V_twq_2msl); 180} 181 182/* 183 * Move a TCP connection into TIME_WAIT state. 184 * tcbinfo is locked. 185 * inp is locked, and is unlocked before returning. 186 */ 187void 188tcp_twstart(struct tcpcb *tp) 189{ 190 INIT_VNET_INET(tp->t_vnet); 191 struct tcptw *tw; 192 struct inpcb *inp = tp->t_inpcb; 193 int acknow; 194 struct socket *so; 195 196 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_reset(). */ 197 INP_WLOCK_ASSERT(inp); 198 199 if (V_nolocaltimewait && in_localip(inp->inp_faddr)) { 200 tp = tcp_close(tp); 201 if (tp != NULL) 202 INP_WUNLOCK(inp); 203 return; 204 } 205 206 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 207 if (tw == NULL) { 208 tw = tcp_tw_2msl_scan(1); 209 if (tw == NULL) { 210 tp = tcp_close(tp); 211 if (tp != NULL) 212 INP_WUNLOCK(inp); 213 return; 214 } 215 } 216 tw->tw_inpcb = inp; 217 218 /* 219 * Recover last window size sent. 220 */ 221 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 222 223 /* 224 * Set t_recent if timestamps are used on the connection. 225 */ 226 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 227 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 228 tw->t_recent = tp->ts_recent; 229 tw->ts_offset = tp->ts_offset; 230 } else { 231 tw->t_recent = 0; 232 tw->ts_offset = 0; 233 } 234 235 tw->snd_nxt = tp->snd_nxt; 236 tw->rcv_nxt = tp->rcv_nxt; 237 tw->iss = tp->iss; 238 tw->irs = tp->irs; 239 tw->t_starttime = tp->t_starttime; 240 tw->tw_time = 0; 241 242/* XXX 243 * If this code will 244 * be used for fin-wait-2 state also, then we may need 245 * a ts_recent from the last segment. 246 */ 247 acknow = tp->t_flags & TF_ACKNOW; 248 249 /* 250 * First, discard tcpcb state, which includes stopping its timers and 251 * freeing it. tcp_discardcb() used to also release the inpcb, but 252 * that work is now done in the caller. 253 * 254 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 255 * and might not be needed here any longer. 256 */ 257 tcp_discardcb(tp); 258 so = inp->inp_socket; 259 soisdisconnected(so); 260 tw->tw_cred = crhold(so->so_cred); 261 SOCK_LOCK(so); 262 tw->tw_so_options = so->so_options; 263 SOCK_UNLOCK(so); 264 if (acknow) 265 tcp_twrespond(tw, TH_ACK); 266 inp->inp_ppcb = tw; 267 inp->inp_flags |= INP_TIMEWAIT; 268 tcp_tw_2msl_reset(tw, 0); 269 270 /* 271 * If the inpcb owns the sole reference to the socket, then we can 272 * detach and free the socket as it is not needed in time wait. 273 */ 274 if (inp->inp_flags & INP_SOCKREF) { 275 KASSERT(so->so_state & SS_PROTOREF, 276 ("tcp_twstart: !SS_PROTOREF")); 277 inp->inp_flags &= ~INP_SOCKREF; 278 INP_WUNLOCK(inp); 279 ACCEPT_LOCK(); 280 SOCK_LOCK(so); 281 so->so_state &= ~SS_PROTOREF; 282 sofree(so); 283 } else 284 INP_WUNLOCK(inp); 285} 286 287#if 0 288/* 289 * The appromixate rate of ISN increase of Microsoft TCP stacks; 290 * the actual rate is slightly higher due to the addition of 291 * random positive increments. 292 * 293 * Most other new OSes use semi-randomized ISN values, so we 294 * do not need to worry about them. 295 */ 296#define MS_ISN_BYTES_PER_SECOND 250000 297 298/* 299 * Determine if the ISN we will generate has advanced beyond the last 300 * sequence number used by the previous connection. If so, indicate 301 * that it is safe to recycle this tw socket by returning 1. 302 */ 303int 304tcp_twrecycleable(struct tcptw *tw) 305{ 306 INIT_VNET_INET(curvnet); 307 tcp_seq new_iss = tw->iss; 308 tcp_seq new_irs = tw->irs; 309 310 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 311 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 312 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 313 314 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 315 return (1); 316 else 317 return (0); 318} 319#endif 320 321/* 322 * Returns 1 if the TIME_WAIT state was killed and we should start over, 323 * looking for a pcb in the listen state. Returns 0 otherwise. 324 */ 325int 326tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 327 struct mbuf *m, int tlen) 328{ 329#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 330 INIT_VNET_INET(curvnet); 331#endif 332 struct tcptw *tw; 333 int thflags; 334 tcp_seq seq; 335 336 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */ 337 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 338 INP_WLOCK_ASSERT(inp); 339 340 /* 341 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 342 * still present. This is undesirable, but temporarily necessary 343 * until we work out how to handle inpcb's who's timewait state has 344 * been removed. 345 */ 346 tw = intotw(inp); 347 if (tw == NULL) 348 goto drop; 349 350 thflags = th->th_flags; 351 352 /* 353 * NOTE: for FIN_WAIT_2 (to be added later), 354 * must validate sequence number before accepting RST 355 */ 356 357 /* 358 * If the segment contains RST: 359 * Drop the segment - see Stevens, vol. 2, p. 964 and 360 * RFC 1337. 361 */ 362 if (thflags & TH_RST) 363 goto drop; 364 365#if 0 366/* PAWS not needed at the moment */ 367 /* 368 * RFC 1323 PAWS: If we have a timestamp reply on this segment 369 * and it's less than ts_recent, drop it. 370 */ 371 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 372 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 373 if ((thflags & TH_ACK) == 0) 374 goto drop; 375 goto ack; 376 } 377 /* 378 * ts_recent is never updated because we never accept new segments. 379 */ 380#endif 381 382 /* 383 * If a new connection request is received 384 * while in TIME_WAIT, drop the old connection 385 * and start over if the sequence numbers 386 * are above the previous ones. 387 */ 388 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 389 tcp_twclose(tw, 0); 390 return (1); 391 } 392 393 /* 394 * Drop the the segment if it does not contain an ACK. 395 */ 396 if ((thflags & TH_ACK) == 0) 397 goto drop; 398 399 /* 400 * Reset the 2MSL timer if this is a duplicate FIN. 401 */ 402 if (thflags & TH_FIN) { 403 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 404 if (seq + 1 == tw->rcv_nxt) 405 tcp_tw_2msl_reset(tw, 1); 406 } 407 408 /* 409 * Acknowledge the segment if it has data or is not a duplicate ACK. 410 */ 411 if (thflags != TH_ACK || tlen != 0 || 412 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 413 tcp_twrespond(tw, TH_ACK); 414drop: 415 INP_WUNLOCK(inp); 416 m_freem(m); 417 return (0); 418} 419 420void 421tcp_twclose(struct tcptw *tw, int reuse) 422{ 423 INIT_VNET_INET(curvnet); 424 struct socket *so; 425 struct inpcb *inp; 426 427 /* 428 * At this point, we are in one of two situations: 429 * 430 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 431 * all state. 432 * 433 * (2) We have a socket -- if we own a reference, release it and 434 * notify the socket layer. 435 */ 436 inp = tw->tw_inpcb; 437 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 438 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 439 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_stop(). */ 440 INP_WLOCK_ASSERT(inp); 441 442 tw->tw_inpcb = NULL; 443 tcp_tw_2msl_stop(tw); 444 inp->inp_ppcb = NULL; 445 in_pcbdrop(inp); 446 447 so = inp->inp_socket; 448 if (so != NULL) { 449 /* 450 * If there's a socket, handle two cases: first, we own a 451 * strong reference, which we will now release, or we don't 452 * in which case another reference exists (XXXRW: think 453 * about this more), and we don't need to take action. 454 */ 455 if (inp->inp_flags & INP_SOCKREF) { 456 inp->inp_flags &= ~INP_SOCKREF; 457 INP_WUNLOCK(inp); 458 ACCEPT_LOCK(); 459 SOCK_LOCK(so); 460 KASSERT(so->so_state & SS_PROTOREF, 461 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 462 so->so_state &= ~SS_PROTOREF; 463 sofree(so); 464 } else { 465 /* 466 * If we don't own the only reference, the socket and 467 * inpcb need to be left around to be handled by 468 * tcp_usr_detach() later. 469 */ 470 INP_WUNLOCK(inp); 471 } 472 } else 473 in_pcbfree(inp); 474 TCPSTAT_INC(tcps_closed); 475 crfree(tw->tw_cred); 476 tw->tw_cred = NULL; 477 if (reuse) 478 return; 479 uma_zfree(V_tcptw_zone, tw); 480} 481 482int 483tcp_twrespond(struct tcptw *tw, int flags) 484{ 485 INIT_VNET_INET(curvnet); 486 struct inpcb *inp = tw->tw_inpcb; 487 struct tcphdr *th; 488 struct mbuf *m; 489 struct ip *ip = NULL; 490 u_int hdrlen, optlen; 491 int error; 492 struct tcpopt to; 493#ifdef INET6 494 struct ip6_hdr *ip6 = NULL; 495 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 496#endif 497 498 INP_WLOCK_ASSERT(inp); 499 500 m = m_gethdr(M_DONTWAIT, MT_DATA); 501 if (m == NULL) 502 return (ENOBUFS); 503 m->m_data += max_linkhdr; 504 505#ifdef MAC 506 mac_inpcb_create_mbuf(inp, m); 507#endif 508 509#ifdef INET6 510 if (isipv6) { 511 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 512 ip6 = mtod(m, struct ip6_hdr *); 513 th = (struct tcphdr *)(ip6 + 1); 514 tcpip_fillheaders(inp, ip6, th); 515 } else 516#endif 517 { 518 hdrlen = sizeof(struct tcpiphdr); 519 ip = mtod(m, struct ip *); 520 th = (struct tcphdr *)(ip + 1); 521 tcpip_fillheaders(inp, ip, th); 522 } 523 to.to_flags = 0; 524 525 /* 526 * Send a timestamp and echo-reply if both our side and our peer 527 * have sent timestamps in our SYN's and this is not a RST. 528 */ 529 if (tw->t_recent && flags == TH_ACK) { 530 to.to_flags |= TOF_TS; 531 to.to_tsval = ticks + tw->ts_offset; 532 to.to_tsecr = tw->t_recent; 533 } 534 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 535 536 m->m_len = hdrlen + optlen; 537 m->m_pkthdr.len = m->m_len; 538 539 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 540 541 th->th_seq = htonl(tw->snd_nxt); 542 th->th_ack = htonl(tw->rcv_nxt); 543 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 544 th->th_flags = flags; 545 th->th_win = htons(tw->last_win); 546 547#ifdef INET6 548 if (isipv6) { 549 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 550 sizeof(struct tcphdr) + optlen); 551 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 552 error = ip6_output(m, inp->in6p_outputopts, NULL, 553 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 554 } else 555#endif 556 { 557 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 558 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 559 m->m_pkthdr.csum_flags = CSUM_TCP; 560 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 561 ip->ip_len = m->m_pkthdr.len; 562 if (V_path_mtu_discovery) 563 ip->ip_off |= IP_DF; 564 error = ip_output(m, inp->inp_options, NULL, 565 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 566 NULL, inp); 567 } 568 if (flags & TH_ACK) 569 TCPSTAT_INC(tcps_sndacks); 570 else 571 TCPSTAT_INC(tcps_sndctrl); 572 TCPSTAT_INC(tcps_sndtotal); 573 return (error); 574} 575 576static void 577tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 578{ 579 INIT_VNET_INET(curvnet); 580 581 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 582 INP_WLOCK_ASSERT(tw->tw_inpcb); 583 if (rearm) 584 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 585 tw->tw_time = ticks + 2 * tcp_msl; 586 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 587} 588 589static void 590tcp_tw_2msl_stop(struct tcptw *tw) 591{ 592 INIT_VNET_INET(curvnet); 593 594 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 595 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 596} 597 598struct tcptw * 599tcp_tw_2msl_scan(int reuse) 600{ 601 INIT_VNET_INET(curvnet); 602 struct tcptw *tw; 603 604 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 605 for (;;) { 606 tw = TAILQ_FIRST(&V_twq_2msl); 607 if (tw == NULL || (!reuse && tw->tw_time > ticks)) 608 break; 609 INP_WLOCK(tw->tw_inpcb); 610 tcp_twclose(tw, reuse); 611 if (reuse) 612 return (tw); 613 } 614 return (NULL); 615} 616