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