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/vnet.h> 58 59#include <netinet/in.h> 60#include <netinet/in_pcb.h> 61#include <netinet/in_systm.h> 62#include <netinet/in_var.h> 63#include <netinet/ip.h> 64#include <netinet/ip_icmp.h> 65#include <netinet/ip_var.h> 66#ifdef INET6 67#include <netinet/ip6.h> 68#include <netinet6/in6_pcb.h> 69#include <netinet6/ip6_var.h> 70#include <netinet6/scope6_var.h> 71#include <netinet6/nd6.h> 72#endif 73#include <netinet/tcp.h> 74#include <netinet/tcp_fsm.h> 75#include <netinet/tcp_seq.h> 76#include <netinet/tcp_timer.h> 77#include <netinet/tcp_var.h> 78#ifdef INET6 79#include <netinet6/tcp6_var.h> 80#endif 81#include <netinet/tcpip.h> 82#ifdef TCPDEBUG 83#include <netinet/tcp_debug.h> 84#endif 85#ifdef INET6 86#include <netinet6/ip6protosw.h> 87#endif 88 89#include <machine/in_cksum.h> 90 91#include <security/mac/mac_framework.h> 92 93static VNET_DEFINE(uma_zone_t, tcptw_zone); 94#define V_tcptw_zone VNET(tcptw_zone) 95static int maxtcptw; 96 97/* 98 * The timed wait queue contains references to each of the TCP sessions 99 * currently in the TIME_WAIT state. The queue pointers, including the 100 * queue pointers in each tcptw structure, are protected using the global 101 * tcbinfo lock, which must be held over queue iteration and modification. 102 */ 103static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl); 104#define V_twq_2msl VNET(twq_2msl) 105 106static void tcp_tw_2msl_reset(struct tcptw *, int); 107static void tcp_tw_2msl_stop(struct tcptw *); 108static int tcp_twrespond(struct tcptw *, int); 109 110static int 111tcptw_auto_size(void) 112{ 113 int halfrange; 114 115 /* 116 * Max out at half the ephemeral port range so that TIME_WAIT 117 * sockets don't tie up too many ephemeral ports. 118 */ 119 if (V_ipport_lastauto > V_ipport_firstauto) 120 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 121 else 122 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 123 /* Protect against goofy port ranges smaller than 32. */ 124 return (imin(imax(halfrange, 32), maxsockets / 5)); 125} 126 127static int 128sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 129{ 130 int error, new; 131 132 if (maxtcptw == 0) 133 new = tcptw_auto_size(); 134 else 135 new = maxtcptw; 136 error = sysctl_handle_int(oidp, &new, 0, req); 137 if (error == 0 && req->newptr) 138 if (new >= 32) { 139 maxtcptw = new; 140 uma_zone_set_max(V_tcptw_zone, maxtcptw); 141 } 142 return (error); 143} 144 145SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 146 &maxtcptw, 0, sysctl_maxtcptw, "IU", 147 "Maximum number of compressed TCP TIME_WAIT entries"); 148 149VNET_DEFINE(int, nolocaltimewait) = 0; 150#define V_nolocaltimewait VNET(nolocaltimewait) 151SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 152 &VNET_NAME(nolocaltimewait), 0, 153 "Do not create compressed TCP TIME_WAIT entries for local connections"); 154 155void 156tcp_tw_zone_change(void) 157{ 158 159 if (maxtcptw == 0) 160 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 161} 162 163void 164tcp_tw_init(void) 165{ 166 167 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 168 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 169 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 170 if (maxtcptw == 0) 171 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 172 else 173 uma_zone_set_max(V_tcptw_zone, maxtcptw); 174 TAILQ_INIT(&V_twq_2msl); 175} 176 177#ifdef VIMAGE 178void 179tcp_tw_destroy(void) 180{ 181 struct tcptw *tw; 182 183 INP_INFO_WLOCK(&V_tcbinfo); 184 while((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL) 185 tcp_twclose(tw, 0); 186 INP_INFO_WUNLOCK(&V_tcbinfo); 187 188 uma_zdestroy(V_tcptw_zone); 189} 190#endif 191 192/* 193 * Move a TCP connection into TIME_WAIT state. 194 * tcbinfo is locked. 195 * inp is locked, and is unlocked before returning. 196 */ 197void 198tcp_twstart(struct tcpcb *tp) 199{ 200 struct tcptw *tw; 201 struct inpcb *inp = tp->t_inpcb; 202 int acknow; 203 struct socket *so; 204#ifdef INET6 205 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 206#endif 207 208 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_reset(). */ 209 INP_WLOCK_ASSERT(inp); 210 211 if (V_nolocaltimewait) { 212 int error = 0; 213#ifdef INET6 214 if (isipv6) 215 error = in6_localaddr(&inp->in6p_faddr); 216#endif 217#if defined(INET6) && defined(INET) 218 else 219#endif 220#ifdef INET 221 error = in_localip(inp->inp_faddr); 222#endif 223 if (error) { 224 tp = tcp_close(tp); 225 if (tp != NULL) 226 INP_WUNLOCK(inp); 227 return; 228 } 229 } 230 231 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 232 if (tw == NULL) { 233 tw = tcp_tw_2msl_scan(1); 234 if (tw == NULL) { 235 tp = tcp_close(tp); 236 if (tp != NULL) 237 INP_WUNLOCK(inp); 238 return; 239 } 240 } 241 tw->tw_inpcb = inp; 242 243 /* 244 * Recover last window size sent. 245 */ 246 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 247 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 248 else 249 tw->last_win = 0; 250 251 /* 252 * Set t_recent if timestamps are used on the connection. 253 */ 254 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 255 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 256 tw->t_recent = tp->ts_recent; 257 tw->ts_offset = tp->ts_offset; 258 } else { 259 tw->t_recent = 0; 260 tw->ts_offset = 0; 261 } 262 263 tw->snd_nxt = tp->snd_nxt; 264 tw->rcv_nxt = tp->rcv_nxt; 265 tw->iss = tp->iss; 266 tw->irs = tp->irs; 267 tw->t_starttime = tp->t_starttime; 268 tw->tw_time = 0; 269 270/* XXX 271 * If this code will 272 * be used for fin-wait-2 state also, then we may need 273 * a ts_recent from the last segment. 274 */ 275 acknow = tp->t_flags & TF_ACKNOW; 276 277 /* 278 * First, discard tcpcb state, which includes stopping its timers and 279 * freeing it. tcp_discardcb() used to also release the inpcb, but 280 * that work is now done in the caller. 281 * 282 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 283 * and might not be needed here any longer. 284 */ 285 tcp_discardcb(tp); 286 so = inp->inp_socket; 287 soisdisconnected(so); 288 tw->tw_cred = crhold(so->so_cred); 289 SOCK_LOCK(so); 290 tw->tw_so_options = so->so_options; 291 SOCK_UNLOCK(so); 292 if (acknow) 293 tcp_twrespond(tw, TH_ACK); 294 inp->inp_ppcb = tw; 295 inp->inp_flags |= INP_TIMEWAIT; 296 tcp_tw_2msl_reset(tw, 0); 297 298 /* 299 * If the inpcb owns the sole reference to the socket, then we can 300 * detach and free the socket as it is not needed in time wait. 301 */ 302 if (inp->inp_flags & INP_SOCKREF) { 303 KASSERT(so->so_state & SS_PROTOREF, 304 ("tcp_twstart: !SS_PROTOREF")); 305 inp->inp_flags &= ~INP_SOCKREF; 306 INP_WUNLOCK(inp); 307 ACCEPT_LOCK(); 308 SOCK_LOCK(so); 309 so->so_state &= ~SS_PROTOREF; 310 sofree(so); 311 } else 312 INP_WUNLOCK(inp); 313} 314 315#if 0 316/* 317 * The appromixate rate of ISN increase of Microsoft TCP stacks; 318 * the actual rate is slightly higher due to the addition of 319 * random positive increments. 320 * 321 * Most other new OSes use semi-randomized ISN values, so we 322 * do not need to worry about them. 323 */ 324#define MS_ISN_BYTES_PER_SECOND 250000 325 326/* 327 * Determine if the ISN we will generate has advanced beyond the last 328 * sequence number used by the previous connection. If so, indicate 329 * that it is safe to recycle this tw socket by returning 1. 330 */ 331int 332tcp_twrecycleable(struct tcptw *tw) 333{ 334 tcp_seq new_iss = tw->iss; 335 tcp_seq new_irs = tw->irs; 336 337 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 338 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 339 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 340 341 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 342 return (1); 343 else 344 return (0); 345} 346#endif 347 348/* 349 * Returns 1 if the TIME_WAIT state was killed and we should start over, 350 * looking for a pcb in the listen state. Returns 0 otherwise. 351 */ 352int 353tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th, 354 struct mbuf *m, int tlen) 355{ 356 struct tcptw *tw; 357 int thflags; 358 tcp_seq seq; 359 360 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */ 361 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 362 INP_WLOCK_ASSERT(inp); 363 364 /* 365 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 366 * still present. This is undesirable, but temporarily necessary 367 * until we work out how to handle inpcb's who's timewait state has 368 * been removed. 369 */ 370 tw = intotw(inp); 371 if (tw == NULL) 372 goto drop; 373 374 thflags = th->th_flags; 375 376 /* 377 * NOTE: for FIN_WAIT_2 (to be added later), 378 * must validate sequence number before accepting RST 379 */ 380 381 /* 382 * If the segment contains RST: 383 * Drop the segment - see Stevens, vol. 2, p. 964 and 384 * RFC 1337. 385 */ 386 if (thflags & TH_RST) 387 goto drop; 388 389#if 0 390/* PAWS not needed at the moment */ 391 /* 392 * RFC 1323 PAWS: If we have a timestamp reply on this segment 393 * and it's less than ts_recent, drop it. 394 */ 395 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 396 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 397 if ((thflags & TH_ACK) == 0) 398 goto drop; 399 goto ack; 400 } 401 /* 402 * ts_recent is never updated because we never accept new segments. 403 */ 404#endif 405 406 /* 407 * If a new connection request is received 408 * while in TIME_WAIT, drop the old connection 409 * and start over if the sequence numbers 410 * are above the previous ones. 411 */ 412 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 413 tcp_twclose(tw, 0); 414 return (1); 415 } 416 417 /* 418 * Drop the segment if it does not contain an ACK. 419 */ 420 if ((thflags & TH_ACK) == 0) 421 goto drop; 422 423 /* 424 * Reset the 2MSL timer if this is a duplicate FIN. 425 */ 426 if (thflags & TH_FIN) { 427 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 428 if (seq + 1 == tw->rcv_nxt) 429 tcp_tw_2msl_reset(tw, 1); 430 } 431 432 /* 433 * Acknowledge the segment if it has data or is not a duplicate ACK. 434 */ 435 if (thflags != TH_ACK || tlen != 0 || 436 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 437 tcp_twrespond(tw, TH_ACK); 438drop: 439 INP_WUNLOCK(inp); 440 m_freem(m); 441 return (0); 442} 443 444void 445tcp_twclose(struct tcptw *tw, int reuse) 446{ 447 struct socket *so; 448 struct inpcb *inp; 449 450 /* 451 * At this point, we are in one of two situations: 452 * 453 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 454 * all state. 455 * 456 * (2) We have a socket -- if we own a reference, release it and 457 * notify the socket layer. 458 */ 459 inp = tw->tw_inpcb; 460 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 461 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 462 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_stop(). */ 463 INP_WLOCK_ASSERT(inp); 464 465 tw->tw_inpcb = NULL; 466 tcp_tw_2msl_stop(tw); 467 inp->inp_ppcb = NULL; 468 in_pcbdrop(inp); 469 470 so = inp->inp_socket; 471 if (so != NULL) { 472 /* 473 * If there's a socket, handle two cases: first, we own a 474 * strong reference, which we will now release, or we don't 475 * in which case another reference exists (XXXRW: think 476 * about this more), and we don't need to take action. 477 */ 478 if (inp->inp_flags & INP_SOCKREF) { 479 inp->inp_flags &= ~INP_SOCKREF; 480 INP_WUNLOCK(inp); 481 ACCEPT_LOCK(); 482 SOCK_LOCK(so); 483 KASSERT(so->so_state & SS_PROTOREF, 484 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 485 so->so_state &= ~SS_PROTOREF; 486 sofree(so); 487 } else { 488 /* 489 * If we don't own the only reference, the socket and 490 * inpcb need to be left around to be handled by 491 * tcp_usr_detach() later. 492 */ 493 INP_WUNLOCK(inp); 494 } 495 } else 496 in_pcbfree(inp); 497 TCPSTAT_INC(tcps_closed); 498 crfree(tw->tw_cred); 499 tw->tw_cred = NULL; 500 if (reuse) 501 return; 502 uma_zfree(V_tcptw_zone, tw); 503} 504 505static int 506tcp_twrespond(struct tcptw *tw, int flags) 507{ 508 struct inpcb *inp = tw->tw_inpcb; 509#if defined(INET6) || defined(INET) 510 struct tcphdr *th = NULL; 511#endif 512 struct mbuf *m; 513#ifdef INET 514 struct ip *ip = NULL; 515#endif 516 u_int hdrlen, optlen; 517 int error = 0; /* Keep compiler happy */ 518 struct tcpopt to; 519#ifdef INET6 520 struct ip6_hdr *ip6 = NULL; 521 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 522#endif 523 hdrlen = 0; /* Keep compiler happy */ 524 525 INP_WLOCK_ASSERT(inp); 526 527 m = m_gethdr(M_NOWAIT, MT_DATA); 528 if (m == NULL) 529 return (ENOBUFS); 530 m->m_data += max_linkhdr; 531 532#ifdef MAC 533 mac_inpcb_create_mbuf(inp, m); 534#endif 535 536#ifdef INET6 537 if (isipv6) { 538 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 539 ip6 = mtod(m, struct ip6_hdr *); 540 th = (struct tcphdr *)(ip6 + 1); 541 tcpip_fillheaders(inp, ip6, th); 542 } 543#endif 544#if defined(INET6) && defined(INET) 545 else 546#endif 547#ifdef INET 548 { 549 hdrlen = sizeof(struct tcpiphdr); 550 ip = mtod(m, struct ip *); 551 th = (struct tcphdr *)(ip + 1); 552 tcpip_fillheaders(inp, ip, th); 553 } 554#endif 555 to.to_flags = 0; 556 557 /* 558 * Send a timestamp and echo-reply if both our side and our peer 559 * have sent timestamps in our SYN's and this is not a RST. 560 */ 561 if (tw->t_recent && flags == TH_ACK) { 562 to.to_flags |= TOF_TS; 563 to.to_tsval = tcp_ts_getticks() + tw->ts_offset; 564 to.to_tsecr = tw->t_recent; 565 } 566 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 567 568 m->m_len = hdrlen + optlen; 569 m->m_pkthdr.len = m->m_len; 570 571 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 572 573 th->th_seq = htonl(tw->snd_nxt); 574 th->th_ack = htonl(tw->rcv_nxt); 575 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 576 th->th_flags = flags; 577 th->th_win = htons(tw->last_win); 578 579 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 580#ifdef INET6 581 if (isipv6) { 582 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 583 th->th_sum = in6_cksum_pseudo(ip6, 584 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0); 585 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 586 error = ip6_output(m, inp->in6p_outputopts, NULL, 587 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 588 } 589#endif 590#if defined(INET6) && defined(INET) 591 else 592#endif 593#ifdef INET 594 { 595 m->m_pkthdr.csum_flags = CSUM_TCP; 596 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 597 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 598 ip->ip_len = htons(m->m_pkthdr.len); 599 if (V_path_mtu_discovery) 600 ip->ip_off |= htons(IP_DF); 601 error = ip_output(m, inp->inp_options, NULL, 602 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 603 NULL, inp); 604 } 605#endif 606 if (flags & TH_ACK) 607 TCPSTAT_INC(tcps_sndacks); 608 else 609 TCPSTAT_INC(tcps_sndctrl); 610 TCPSTAT_INC(tcps_sndtotal); 611 return (error); 612} 613 614static void 615tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 616{ 617 618 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 619 INP_WLOCK_ASSERT(tw->tw_inpcb); 620 if (rearm) 621 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 622 tw->tw_time = ticks + 2 * tcp_msl; 623 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 624} 625 626static void 627tcp_tw_2msl_stop(struct tcptw *tw) 628{ 629 630 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 631 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 632} 633 634struct tcptw * 635tcp_tw_2msl_scan(int reuse) 636{ 637 struct tcptw *tw; 638 639 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 640 for (;;) { 641 tw = TAILQ_FIRST(&V_twq_2msl); 642 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) 643 break; 644 INP_WLOCK(tw->tw_inpcb); 645 tcp_twclose(tw, reuse); 646 if (reuse) 647 return (tw); 648 } 649 return (NULL); 650} 651