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 93 static VNET_DEFINE(uma_zone_t, tcptw_zone); 94 #define V_tcptw_zone VNET(tcptw_zone) 95 static 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 */ 103 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl); 104 #define V_twq_2msl VNET(twq_2msl) 105 106 static void tcp_tw_2msl_reset(struct tcptw *, int); 107 static void tcp_tw_2msl_stop(struct tcptw *); 108 109 static int 110 tcptw_auto_size(void) 111 { 112 int halfrange; 113 114 /* 115 * Max out at half the ephemeral port range so that TIME_WAIT 116 * sockets don't tie up too many ephemeral ports. 117 */ 118 if (V_ipport_lastauto > V_ipport_firstauto) 119 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 120 else 121 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 122 /* Protect against goofy port ranges smaller than 32. */ 123 return (imin(imax(halfrange, 32), maxsockets / 5)); 124 } 125 126 static int 127 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 128 { 129 int error, new; 130 131 if (maxtcptw == 0) 132 new = tcptw_auto_size(); 133 else 134 new = maxtcptw; 135 error = sysctl_handle_int(oidp, &new, 0, req); 136 if (error == 0 && req->newptr) 137 if (new >= 32) { 138 maxtcptw = new; 139 uma_zone_set_max(V_tcptw_zone, maxtcptw); 140 } 141 return (error); 142 } 143 144 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 145 &maxtcptw, 0, sysctl_maxtcptw, "IU", 146 "Maximum number of compressed TCP TIME_WAIT entries"); 147 148 VNET_DEFINE(int, nolocaltimewait) = 0; 149 #define V_nolocaltimewait VNET(nolocaltimewait) 150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 151 &VNET_NAME(nolocaltimewait), 0, 152 "Do not create compressed TCP TIME_WAIT entries for local connections"); 153 154 void 155 tcp_tw_zone_change(void) 156 { 157 158 if (maxtcptw == 0) 159 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 160 } 161 162 void 163 tcp_tw_init(void) 164 { 165 166 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 167 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 168 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 169 if (maxtcptw == 0) 170 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 171 else 172 uma_zone_set_max(V_tcptw_zone, maxtcptw); 173 TAILQ_INIT(&V_twq_2msl); 174 } 175 176 #ifdef VIMAGE 177 void 178 tcp_tw_destroy(void) 179 { 180 struct tcptw *tw; 181 182 INP_INFO_WLOCK(&V_tcbinfo); 183 while((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL) 184 tcp_twclose(tw, 0); 185 INP_INFO_WUNLOCK(&V_tcbinfo); 186 187 uma_zdestroy(V_tcptw_zone); 188 } 189 #endif 190 191 /* 192 * Move a TCP connection into TIME_WAIT state. 193 * tcbinfo is locked. 194 * inp is locked, and is unlocked before returning. 195 */ 196 void 197 tcp_twstart(struct tcpcb *tp) 198 { 199 struct tcptw *tw; 200 struct inpcb *inp = tp->t_inpcb; 201 int acknow; 202 struct socket *so; 203 #ifdef INET6 204 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 205 #endif 206 207 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_reset(). */ 208 INP_WLOCK_ASSERT(inp); 209 210 if (V_nolocaltimewait) { 211 int error = 0; 212 #ifdef INET6 213 if (isipv6) 214 error = in6_localaddr(&inp->in6p_faddr); 215 #endif 216 #if defined(INET6) && defined(INET) 217 else 218 #endif 219 #ifdef INET 220 error = in_localip(inp->inp_faddr); 221 #endif 222 if (error) { 223 tp = tcp_close(tp); 224 if (tp != NULL) 225 INP_WUNLOCK(inp); 226 return; 227 } 228 } 229 230 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 231 if (tw == NULL) { 232 tw = tcp_tw_2msl_scan(1); 233 if (tw == NULL) { 234 tp = tcp_close(tp); 235 if (tp != NULL) 236 INP_WUNLOCK(inp); 237 return; 238 } 239 } 240 tw->tw_inpcb = inp; 241 242 /* 243 * Recover last window size sent. 244 */ 245 KASSERT(SEQ_GEQ(tp->rcv_adv, tp->rcv_nxt), 246 ("tcp_twstart negative window: tp %p rcv_nxt %u rcv_adv %u", tp, 247 tp->rcv_nxt, tp->rcv_adv)); 248 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 249 250 /* 251 * Set t_recent if timestamps are used on the connection. 252 */ 253 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 254 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 255 tw->t_recent = tp->ts_recent; 256 tw->ts_offset = tp->ts_offset; 257 } else { 258 tw->t_recent = 0; 259 tw->ts_offset = 0; 260 } 261 262 tw->snd_nxt = tp->snd_nxt; 263 tw->rcv_nxt = tp->rcv_nxt; 264 tw->iss = tp->iss; 265 tw->irs = tp->irs; 266 tw->t_starttime = tp->t_starttime; 267 tw->tw_time = 0; 268 269 /* XXX 270 * If this code will 271 * be used for fin-wait-2 state also, then we may need 272 * a ts_recent from the last segment. 273 */ 274 acknow = tp->t_flags & TF_ACKNOW; 275 276 /* 277 * First, discard tcpcb state, which includes stopping its timers and 278 * freeing it. tcp_discardcb() used to also release the inpcb, but 279 * that work is now done in the caller. 280 * 281 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 282 * and might not be needed here any longer. 283 */ 284 tcp_discardcb(tp); 285 so = inp->inp_socket; 286 soisdisconnected(so); 287 tw->tw_cred = crhold(so->so_cred); 288 SOCK_LOCK(so); 289 tw->tw_so_options = so->so_options; 290 SOCK_UNLOCK(so); 291 if (acknow) 292 tcp_twrespond(tw, TH_ACK); 293 inp->inp_ppcb = tw; 294 inp->inp_flags |= INP_TIMEWAIT; 295 tcp_tw_2msl_reset(tw, 0); 296 297 /* 298 * If the inpcb owns the sole reference to the socket, then we can 299 * detach and free the socket as it is not needed in time wait. 300 */ 301 if (inp->inp_flags & INP_SOCKREF) { 302 KASSERT(so->so_state & SS_PROTOREF, 303 ("tcp_twstart: !SS_PROTOREF")); 304 inp->inp_flags &= ~INP_SOCKREF; 305 INP_WUNLOCK(inp); 306 ACCEPT_LOCK(); 307 SOCK_LOCK(so); 308 so->so_state &= ~SS_PROTOREF; 309 sofree(so); 310 } else 311 INP_WUNLOCK(inp); 312 } 313 314 #if 0 315 /* 316 * The appromixate rate of ISN increase of Microsoft TCP stacks; 317 * the actual rate is slightly higher due to the addition of 318 * random positive increments. 319 * 320 * Most other new OSes use semi-randomized ISN values, so we 321 * do not need to worry about them. 322 */ 323 #define MS_ISN_BYTES_PER_SECOND 250000 324 325 /* 326 * Determine if the ISN we will generate has advanced beyond the last 327 * sequence number used by the previous connection. If so, indicate 328 * that it is safe to recycle this tw socket by returning 1. 329 */ 330 int 331 tcp_twrecycleable(struct tcptw *tw) 332 { 333 tcp_seq new_iss = tw->iss; 334 tcp_seq new_irs = tw->irs; 335 336 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 337 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 338 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 339 340 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 341 return (1); 342 else 343 return (0); 344 } 345 #endif 346 347 /* 348 * Returns 1 if the TIME_WAIT state was killed and we should start over, 349 * looking for a pcb in the listen state. Returns 0 otherwise. 350 */ 351 int 352 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 353 struct mbuf *m, int tlen) 354 { 355 struct tcptw *tw; 356 int thflags; 357 tcp_seq seq; 358 359 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */ 360 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 361 INP_WLOCK_ASSERT(inp); 362 363 /* 364 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 365 * still present. This is undesirable, but temporarily necessary 366 * until we work out how to handle inpcb's who's timewait state has 367 * been removed. 368 */ 369 tw = intotw(inp); 370 if (tw == NULL) 371 goto drop; 372 373 thflags = th->th_flags; 374 375 /* 376 * NOTE: for FIN_WAIT_2 (to be added later), 377 * must validate sequence number before accepting RST 378 */ 379 380 /* 381 * If the segment contains RST: 382 * Drop the segment - see Stevens, vol. 2, p. 964 and 383 * RFC 1337. 384 */ 385 if (thflags & TH_RST) 386 goto drop; 387 388 #if 0 389 /* PAWS not needed at the moment */ 390 /* 391 * RFC 1323 PAWS: If we have a timestamp reply on this segment 392 * and it's less than ts_recent, drop it. 393 */ 394 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 395 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 396 if ((thflags & TH_ACK) == 0) 397 goto drop; 398 goto ack; 399 } 400 /* 401 * ts_recent is never updated because we never accept new segments. 402 */ 403 #endif 404 405 /* 406 * If a new connection request is received 407 * while in TIME_WAIT, drop the old connection 408 * and start over if the sequence numbers 409 * are above the previous ones. 410 */ 411 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 412 tcp_twclose(tw, 0); 413 return (1); 414 } 415 416 /* 417 * Drop the segment if it does not contain an ACK. 418 */ 419 if ((thflags & TH_ACK) == 0) 420 goto drop; 421 422 /* 423 * Reset the 2MSL timer if this is a duplicate FIN. 424 */ 425 if (thflags & TH_FIN) { 426 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 427 if (seq + 1 == tw->rcv_nxt) 428 tcp_tw_2msl_reset(tw, 1); 429 } 430 431 /* 432 * Acknowledge the segment if it has data or is not a duplicate ACK. 433 */ 434 if (thflags != TH_ACK || tlen != 0 || 435 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 436 tcp_twrespond(tw, TH_ACK); 437 drop: 438 INP_WUNLOCK(inp); 439 m_freem(m); 440 return (0); 441 } 442 443 void 444 tcp_twclose(struct tcptw *tw, int reuse) 445 { 446 struct socket *so; 447 struct inpcb *inp; 448 449 /* 450 * At this point, we are in one of two situations: 451 * 452 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 453 * all state. 454 * 455 * (2) We have a socket -- if we own a reference, release it and 456 * notify the socket layer. 457 */ 458 inp = tw->tw_inpcb; 459 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 460 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 461 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_stop(). */ 462 INP_WLOCK_ASSERT(inp); 463 464 tw->tw_inpcb = NULL; 465 tcp_tw_2msl_stop(tw); 466 inp->inp_ppcb = NULL; 467 in_pcbdrop(inp); 468 469 so = inp->inp_socket; 470 if (so != NULL) { 471 /* 472 * If there's a socket, handle two cases: first, we own a 473 * strong reference, which we will now release, or we don't 474 * in which case another reference exists (XXXRW: think 475 * about this more), and we don't need to take action. 476 */ 477 if (inp->inp_flags & INP_SOCKREF) { 478 inp->inp_flags &= ~INP_SOCKREF; 479 INP_WUNLOCK(inp); 480 ACCEPT_LOCK(); 481 SOCK_LOCK(so); 482 KASSERT(so->so_state & SS_PROTOREF, 483 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 484 so->so_state &= ~SS_PROTOREF; 485 sofree(so); 486 } else { 487 /* 488 * If we don't own the only reference, the socket and 489 * inpcb need to be left around to be handled by 490 * tcp_usr_detach() later. 491 */ 492 INP_WUNLOCK(inp); 493 } 494 } else 495 in_pcbfree(inp); 496 TCPSTAT_INC(tcps_closed); 497 crfree(tw->tw_cred); 498 tw->tw_cred = NULL; 499 if (reuse) 500 return; 501 uma_zfree(V_tcptw_zone, tw); 502 } 503 504 int 505 tcp_twrespond(struct tcptw *tw, int flags) 506 { 507 struct inpcb *inp = tw->tw_inpcb; 508 #if defined(INET6) || defined(INET) 509 struct tcphdr *th = NULL; 510 #endif 511 struct mbuf *m; 512 #ifdef INET 513 struct ip *ip = NULL; 514 #endif 515 u_int hdrlen, optlen; 516 int error = 0; /* Keep compiler happy */ 517 struct tcpopt to; 518 #ifdef INET6 519 struct ip6_hdr *ip6 = NULL; 520 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 521 #endif 522 523 INP_WLOCK_ASSERT(inp); 524 525 m = m_gethdr(M_DONTWAIT, MT_DATA); 526 if (m == NULL) 527 return (ENOBUFS); 528 m->m_data += max_linkhdr; 529 530 #ifdef MAC 531 mac_inpcb_create_mbuf(inp, m); 532 #endif 533 534 #ifdef INET6 535 if (isipv6) { 536 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 537 ip6 = mtod(m, struct ip6_hdr *); 538 th = (struct tcphdr *)(ip6 + 1); 539 tcpip_fillheaders(inp, ip6, th); 540 } 541 #endif 542 #if defined(INET6) && defined(INET) 543 else 544 #endif 545 #ifdef INET 546 { 547 hdrlen = sizeof(struct tcpiphdr); 548 ip = mtod(m, struct ip *); 549 th = (struct tcphdr *)(ip + 1); 550 tcpip_fillheaders(inp, ip, th); 551 } 552 #endif 553 to.to_flags = 0; 554 555 /* 556 * Send a timestamp and echo-reply if both our side and our peer 557 * have sent timestamps in our SYN's and this is not a RST. 558 */ 559 if (tw->t_recent && flags == TH_ACK) { 560 to.to_flags |= TOF_TS; 561 to.to_tsval = ticks + tw->ts_offset; 562 to.to_tsecr = tw->t_recent; 563 } 564 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 565 566 m->m_len = hdrlen + optlen; 567 m->m_pkthdr.len = m->m_len; 568 569 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 570 571 th->th_seq = htonl(tw->snd_nxt); 572 th->th_ack = htonl(tw->rcv_nxt); 573 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 574 th->th_flags = flags; 575 th->th_win = htons(tw->last_win); 576 577 #ifdef INET6 578 if (isipv6) { 579 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 580 sizeof(struct tcphdr) + optlen); 581 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 582 error = ip6_output(m, inp->in6p_outputopts, NULL, 583 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 584 } 585 #endif 586 #if defined(INET6) && defined(INET) 587 else 588 #endif 589 #ifdef INET 590 { 591 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 592 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 593 m->m_pkthdr.csum_flags = CSUM_TCP; 594 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 595 ip->ip_len = m->m_pkthdr.len; 596 if (V_path_mtu_discovery) 597 ip->ip_off |= IP_DF; 598 error = ip_output(m, inp->inp_options, NULL, 599 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 600 NULL, inp); 601 } 602 #endif 603 if (flags & TH_ACK) 604 TCPSTAT_INC(tcps_sndacks); 605 else 606 TCPSTAT_INC(tcps_sndctrl); 607 TCPSTAT_INC(tcps_sndtotal); 608 return (error); 609 } 610 611 static void 612 tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 613 { 614 615 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 616 INP_WLOCK_ASSERT(tw->tw_inpcb); 617 if (rearm) 618 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 619 tw->tw_time = ticks + 2 * tcp_msl; 620 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 621 } 622 623 static void 624 tcp_tw_2msl_stop(struct tcptw *tw) 625 { 626 627 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 628 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 629 } 630 631 struct tcptw * 632 tcp_tw_2msl_scan(int reuse) 633 { 634 struct tcptw *tw; 635 636 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 637 for (;;) { 638 tw = TAILQ_FIRST(&V_twq_2msl); 639 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) 640 break; 641 INP_WLOCK(tw->tw_inpcb); 642 tcp_twclose(tw, reuse); 643 if (reuse) 644 return (tw); 645 } 646 return (NULL); 647 } 648