1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_tcpdebug.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/priv.h> 49 #include <sys/proc.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #ifndef INVARIANTS 53 #include <sys/syslog.h> 54 #endif 55 #include <sys/protosw.h> 56 #include <sys/random.h> 57 58 #include <vm/uma.h> 59 60 #include <net/route.h> 61 #include <net/if.h> 62 #include <net/if_var.h> 63 #include <net/vnet.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_kdtrace.h> 67 #include <netinet/in_pcb.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/in_var.h> 70 #include <netinet/ip.h> 71 #include <netinet/ip_icmp.h> 72 #include <netinet/ip_var.h> 73 #ifdef INET6 74 #include <netinet/ip6.h> 75 #include <netinet6/in6_pcb.h> 76 #include <netinet6/ip6_var.h> 77 #include <netinet6/scope6_var.h> 78 #include <netinet6/nd6.h> 79 #endif 80 #include <netinet/tcp.h> 81 #include <netinet/tcp_fsm.h> 82 #include <netinet/tcp_seq.h> 83 #include <netinet/tcp_timer.h> 84 #include <netinet/tcp_var.h> 85 #include <netinet/tcp_hpts.h> 86 #ifdef INET6 87 #include <netinet6/tcp6_var.h> 88 #endif 89 #include <netinet/tcpip.h> 90 #ifdef TCPDEBUG 91 #include <netinet/tcp_debug.h> 92 #endif 93 #ifdef INET6 94 #include <netinet6/ip6protosw.h> 95 #endif 96 97 #include <netinet/udp.h> 98 #include <netinet/udp_var.h> 99 #include <machine/in_cksum.h> 100 101 #include <security/mac/mac_framework.h> 102 103 VNET_DEFINE_STATIC(uma_zone_t, tcptw_zone); 104 #define V_tcptw_zone VNET(tcptw_zone) 105 static int maxtcptw; 106 107 /* 108 * The timed wait queue contains references to each of the TCP sessions 109 * currently in the TIME_WAIT state. The queue pointers, including the 110 * queue pointers in each tcptw structure, are protected using the global 111 * timewait lock, which must be held over queue iteration and modification. 112 * 113 * Rules on tcptw usage: 114 * - a inpcb is always freed _after_ its tcptw 115 * - a tcptw relies on its inpcb reference counting for memory stability 116 * - a tcptw is dereferenceable only while its inpcb is locked 117 */ 118 VNET_DEFINE_STATIC(TAILQ_HEAD(, tcptw), twq_2msl); 119 #define V_twq_2msl VNET(twq_2msl) 120 121 /* Global timewait lock */ 122 VNET_DEFINE_STATIC(struct rwlock, tw_lock); 123 #define V_tw_lock VNET(tw_lock) 124 125 #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0) 126 #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw)) 127 #define TW_RLOCK(tw) rw_rlock(&(tw)) 128 #define TW_WLOCK(tw) rw_wlock(&(tw)) 129 #define TW_RUNLOCK(tw) rw_runlock(&(tw)) 130 #define TW_WUNLOCK(tw) rw_wunlock(&(tw)) 131 #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED) 132 #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED) 133 #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED) 134 #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED) 135 136 static void tcp_tw_2msl_reset(struct tcptw *, int); 137 static void tcp_tw_2msl_stop(struct tcptw *, int); 138 static int tcp_twrespond(struct tcptw *, int); 139 140 static int 141 tcptw_auto_size(void) 142 { 143 int halfrange; 144 145 /* 146 * Max out at half the ephemeral port range so that TIME_WAIT 147 * sockets don't tie up too many ephemeral ports. 148 */ 149 if (V_ipport_lastauto > V_ipport_firstauto) 150 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 151 else 152 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 153 /* Protect against goofy port ranges smaller than 32. */ 154 return (imin(imax(halfrange, 32), maxsockets / 5)); 155 } 156 157 static int 158 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 159 { 160 int error, new; 161 162 if (maxtcptw == 0) 163 new = tcptw_auto_size(); 164 else 165 new = maxtcptw; 166 error = sysctl_handle_int(oidp, &new, 0, req); 167 if (error == 0 && req->newptr) 168 if (new >= 32) { 169 maxtcptw = new; 170 uma_zone_set_max(V_tcptw_zone, maxtcptw); 171 } 172 return (error); 173 } 174 175 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, 176 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 177 &maxtcptw, 0, sysctl_maxtcptw, "IU", 178 "Maximum number of compressed TCP TIME_WAIT entries"); 179 180 VNET_DEFINE_STATIC(bool, nolocaltimewait) = true; 181 #define V_nolocaltimewait VNET(nolocaltimewait) 182 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW, 183 &VNET_NAME(nolocaltimewait), true, 184 "Do not create compressed TCP TIME_WAIT entries for local connections"); 185 186 void 187 tcp_tw_zone_change(void) 188 { 189 190 if (maxtcptw == 0) 191 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 192 } 193 194 void 195 tcp_tw_init(void) 196 { 197 198 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 200 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 201 if (maxtcptw == 0) 202 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); 203 else 204 uma_zone_set_max(V_tcptw_zone, maxtcptw); 205 TAILQ_INIT(&V_twq_2msl); 206 TW_LOCK_INIT(V_tw_lock, "tcptw"); 207 } 208 209 #ifdef VIMAGE 210 void 211 tcp_tw_destroy(void) 212 { 213 struct tcptw *tw; 214 struct epoch_tracker et; 215 216 NET_EPOCH_ENTER(et); 217 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL) 218 tcp_twclose(tw, 0); 219 NET_EPOCH_EXIT(et); 220 221 TW_LOCK_DESTROY(V_tw_lock); 222 uma_zdestroy(V_tcptw_zone); 223 } 224 #endif 225 226 /* 227 * Move a TCP connection into TIME_WAIT state. 228 * tcbinfo is locked. 229 * inp is locked, and is unlocked before returning. 230 */ 231 void 232 tcp_twstart(struct tcpcb *tp) 233 { 234 struct tcptw twlocal, *tw; 235 struct inpcb *inp = tp->t_inpcb; 236 struct socket *so; 237 uint32_t recwin; 238 bool acknow, local; 239 #ifdef INET6 240 bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 241 #endif 242 243 NET_EPOCH_ASSERT(); 244 INP_WLOCK_ASSERT(inp); 245 246 /* A dropped inp should never transition to TIME_WAIT state. */ 247 KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: " 248 "(inp->inp_flags & INP_DROPPED) != 0")); 249 250 if (V_nolocaltimewait) { 251 #ifdef INET6 252 if (isipv6) 253 local = in6_localaddr(&inp->in6p_faddr); 254 else 255 #endif 256 #ifdef INET 257 local = in_localip(inp->inp_faddr); 258 #else 259 local = false; 260 #endif 261 } else 262 local = false; 263 264 /* 265 * For use only by DTrace. We do not reference the state 266 * after this point so modifying it in place is not a problem. 267 */ 268 tcp_state_change(tp, TCPS_TIME_WAIT); 269 270 if (local) 271 tw = &twlocal; 272 else 273 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); 274 if (tw == NULL) { 275 /* 276 * Reached limit on total number of TIMEWAIT connections 277 * allowed. Remove a connection from TIMEWAIT queue in LRU 278 * fashion to make room for this connection. 279 * If that fails, use on stack tw at least to be able to 280 * run through tcp_twrespond() and standard tcpcb discard 281 * routine. 282 * 283 * XXX: Check if it possible to always have enough room 284 * in advance based on guarantees provided by uma_zalloc(). 285 */ 286 tw = tcp_tw_2msl_scan(1); 287 if (tw == NULL) { 288 tw = &twlocal; 289 local = true; 290 } 291 } 292 /* 293 * For !local case the tcptw will hold a reference on its inpcb 294 * until tcp_twclose is called. 295 */ 296 tw->tw_inpcb = inp; 297 298 /* 299 * Recover last window size sent. 300 */ 301 so = inp->inp_socket; 302 recwin = lmin(lmax(sbspace(&so->so_rcv), 0), 303 (long)TCP_MAXWIN << tp->rcv_scale); 304 if (recwin < (so->so_rcv.sb_hiwat / 4) && 305 recwin < tp->t_maxseg) 306 recwin = 0; 307 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && 308 recwin < (tp->rcv_adv - tp->rcv_nxt)) 309 recwin = (tp->rcv_adv - tp->rcv_nxt); 310 tw->last_win = (u_short)(recwin >> tp->rcv_scale); 311 312 /* 313 * Set t_recent if timestamps are used on the connection. 314 */ 315 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 316 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 317 tw->t_recent = tp->ts_recent; 318 tw->ts_offset = tp->ts_offset; 319 } else { 320 tw->t_recent = 0; 321 tw->ts_offset = 0; 322 } 323 324 tw->snd_nxt = tp->snd_nxt; 325 tw->t_port = tp->t_port; 326 tw->rcv_nxt = tp->rcv_nxt; 327 tw->iss = tp->iss; 328 tw->irs = tp->irs; 329 tw->t_starttime = tp->t_starttime; 330 tw->tw_time = 0; 331 332 /* XXX 333 * If this code will 334 * be used for fin-wait-2 state also, then we may need 335 * a ts_recent from the last segment. 336 */ 337 acknow = tp->t_flags & TF_ACKNOW; 338 339 /* 340 * First, discard tcpcb state, which includes stopping its timers and 341 * freeing it. tcp_discardcb() used to also release the inpcb, but 342 * that work is now done in the caller. 343 * 344 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 345 * and might not be needed here any longer. 346 */ 347 #ifdef TCPHPTS 348 tcp_hpts_remove(inp, HPTS_REMOVE_ALL); 349 #endif 350 tcp_discardcb(tp); 351 soisdisconnected(so); 352 tw->tw_so_options = so->so_options; 353 inp->inp_flags |= INP_TIMEWAIT; 354 if (acknow) 355 tcp_twrespond(tw, TH_ACK); 356 if (local) 357 in_pcbdrop(inp); 358 else { 359 in_pcbref(inp); /* Reference from tw */ 360 tw->tw_cred = crhold(so->so_cred); 361 inp->inp_ppcb = tw; 362 TCPSTATES_INC(TCPS_TIME_WAIT); 363 tcp_tw_2msl_reset(tw, 0); 364 } 365 366 /* 367 * If the inpcb owns the sole reference to the socket, then we can 368 * detach and free the socket as it is not needed in time wait. 369 */ 370 if (inp->inp_flags & INP_SOCKREF) { 371 KASSERT(so->so_state & SS_PROTOREF, 372 ("tcp_twstart: !SS_PROTOREF")); 373 inp->inp_flags &= ~INP_SOCKREF; 374 INP_WUNLOCK(inp); 375 SOCK_LOCK(so); 376 so->so_state &= ~SS_PROTOREF; 377 sofree(so); 378 } else 379 INP_WUNLOCK(inp); 380 } 381 382 /* 383 * Returns 1 if the TIME_WAIT state was killed and we should start over, 384 * looking for a pcb in the listen state. Returns 0 otherwise. 385 * 386 * For pure SYN-segments the PCB shall be read-locked and the tcpopt pointer 387 * may be NULL. For the rest write-lock and valid tcpopt. 388 */ 389 int 390 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 391 struct mbuf *m, int tlen) 392 { 393 struct tcptw *tw; 394 int thflags; 395 tcp_seq seq; 396 397 NET_EPOCH_ASSERT(); 398 INP_LOCK_ASSERT(inp); 399 400 /* 401 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 402 * still present. This is undesirable, but temporarily necessary 403 * until we work out how to handle inpcb's who's timewait state has 404 * been removed. 405 */ 406 tw = intotw(inp); 407 if (tw == NULL) 408 goto drop; 409 410 thflags = th->th_flags; 411 #ifdef INVARIANTS 412 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) 413 INP_RLOCK_ASSERT(inp); 414 else { 415 INP_WLOCK_ASSERT(inp); 416 KASSERT(to != NULL, 417 ("%s: called without options on a non-SYN segment", 418 __func__)); 419 } 420 #endif 421 422 /* 423 * NOTE: for FIN_WAIT_2 (to be added later), 424 * must validate sequence number before accepting RST 425 */ 426 427 /* 428 * If the segment contains RST: 429 * Drop the segment - see Stevens, vol. 2, p. 964 and 430 * RFC 1337. 431 */ 432 if (thflags & TH_RST) 433 goto drop; 434 435 #if 0 436 /* PAWS not needed at the moment */ 437 /* 438 * RFC 1323 PAWS: If we have a timestamp reply on this segment 439 * and it's less than ts_recent, drop it. 440 */ 441 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 442 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 443 if ((thflags & TH_ACK) == 0) 444 goto drop; 445 goto ack; 446 } 447 /* 448 * ts_recent is never updated because we never accept new segments. 449 */ 450 #endif 451 452 /* 453 * If a new connection request is received 454 * while in TIME_WAIT, drop the old connection 455 * and start over if the sequence numbers 456 * are above the previous ones. 457 * Allow UDP port number changes in this case. 458 */ 459 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 460 /* 461 * In case we can't upgrade our lock just pretend we have 462 * lost this packet. 463 */ 464 if (((thflags & (TH_SYN | TH_ACK)) == TH_SYN) && 465 INP_TRY_UPGRADE(inp) == 0) 466 goto drop; 467 tcp_twclose(tw, 0); 468 return (1); 469 } 470 471 /* 472 * Send RST if UDP port numbers don't match 473 */ 474 if (tw->t_port != m->m_pkthdr.tcp_tun_port) { 475 if (th->th_flags & TH_ACK) { 476 tcp_respond(NULL, mtod(m, void *), th, m, 477 (tcp_seq)0, th->th_ack, TH_RST); 478 } else { 479 if (th->th_flags & TH_SYN) 480 tlen++; 481 if (th->th_flags & TH_FIN) 482 tlen++; 483 tcp_respond(NULL, mtod(m, void *), th, m, 484 th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK); 485 } 486 INP_UNLOCK(inp); 487 return (0); 488 } 489 490 /* 491 * Drop the segment if it does not contain an ACK. 492 */ 493 if ((thflags & TH_ACK) == 0) 494 goto drop; 495 496 INP_WLOCK_ASSERT(inp); 497 498 /* 499 * If timestamps were negotiated during SYN/ACK and a 500 * segment without a timestamp is received, silently drop 501 * the segment, unless the missing timestamps are tolerated. 502 * See section 3.2 of RFC 7323. 503 */ 504 if (((to->to_flags & TOF_TS) == 0) && (tw->t_recent != 0) && 505 (V_tcp_tolerate_missing_ts == 0)) { 506 goto drop; 507 } 508 509 /* 510 * Reset the 2MSL timer if this is a duplicate FIN. 511 */ 512 if (thflags & TH_FIN) { 513 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 514 if (seq + 1 == tw->rcv_nxt) 515 tcp_tw_2msl_reset(tw, 1); 516 } 517 518 /* 519 * Acknowledge the segment if it has data or is not a duplicate ACK. 520 */ 521 if (thflags != TH_ACK || tlen != 0 || 522 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) { 523 TCP_PROBE5(receive, NULL, NULL, m, NULL, th); 524 tcp_twrespond(tw, TH_ACK); 525 goto dropnoprobe; 526 } 527 drop: 528 TCP_PROBE5(receive, NULL, NULL, m, NULL, th); 529 dropnoprobe: 530 INP_UNLOCK(inp); 531 m_freem(m); 532 return (0); 533 } 534 535 void 536 tcp_twclose(struct tcptw *tw, int reuse) 537 { 538 struct socket *so; 539 struct inpcb *inp; 540 541 /* 542 * At this point, we are in one of two situations: 543 * 544 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 545 * all state. 546 * 547 * (2) We have a socket -- if we own a reference, release it and 548 * notify the socket layer. 549 */ 550 inp = tw->tw_inpcb; 551 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 552 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 553 NET_EPOCH_ASSERT(); 554 INP_WLOCK_ASSERT(inp); 555 556 tcp_tw_2msl_stop(tw, reuse); 557 inp->inp_ppcb = NULL; 558 in_pcbdrop(inp); 559 560 so = inp->inp_socket; 561 if (so != NULL) { 562 /* 563 * If there's a socket, handle two cases: first, we own a 564 * strong reference, which we will now release, or we don't 565 * in which case another reference exists (XXXRW: think 566 * about this more), and we don't need to take action. 567 */ 568 if (inp->inp_flags & INP_SOCKREF) { 569 inp->inp_flags &= ~INP_SOCKREF; 570 INP_WUNLOCK(inp); 571 SOCK_LOCK(so); 572 KASSERT(so->so_state & SS_PROTOREF, 573 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 574 so->so_state &= ~SS_PROTOREF; 575 sofree(so); 576 } else { 577 /* 578 * If we don't own the only reference, the socket and 579 * inpcb need to be left around to be handled by 580 * tcp_usr_detach() later. 581 */ 582 INP_WUNLOCK(inp); 583 } 584 } else { 585 /* 586 * The socket has been already cleaned-up for us, only free the 587 * inpcb. 588 */ 589 in_pcbfree(inp); 590 } 591 TCPSTAT_INC(tcps_closed); 592 } 593 594 static int 595 tcp_twrespond(struct tcptw *tw, int flags) 596 { 597 struct inpcb *inp = tw->tw_inpcb; 598 #if defined(INET6) || defined(INET) 599 struct tcphdr *th = NULL; 600 #endif 601 struct mbuf *m; 602 #ifdef INET 603 struct ip *ip = NULL; 604 #endif 605 u_int hdrlen, optlen, ulen; 606 int error = 0; /* Keep compiler happy */ 607 struct tcpopt to; 608 #ifdef INET6 609 struct ip6_hdr *ip6 = NULL; 610 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 611 #endif 612 struct udphdr *udp = NULL; 613 hdrlen = 0; /* Keep compiler happy */ 614 615 INP_WLOCK_ASSERT(inp); 616 617 m = m_gethdr(M_NOWAIT, MT_DATA); 618 if (m == NULL) 619 return (ENOBUFS); 620 m->m_data += max_linkhdr; 621 622 #ifdef MAC 623 mac_inpcb_create_mbuf(inp, m); 624 #endif 625 626 #ifdef INET6 627 if (isipv6) { 628 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 629 ip6 = mtod(m, struct ip6_hdr *); 630 if (tw->t_port) { 631 udp = (struct udphdr *)(ip6 + 1); 632 hdrlen += sizeof(struct udphdr); 633 udp->uh_sport = htons(V_tcp_udp_tunneling_port); 634 udp->uh_dport = tw->t_port; 635 ulen = (hdrlen - sizeof(struct ip6_hdr)); 636 th = (struct tcphdr *)(udp + 1); 637 } else 638 th = (struct tcphdr *)(ip6 + 1); 639 tcpip_fillheaders(inp, tw->t_port, ip6, th); 640 } 641 #endif 642 #if defined(INET6) && defined(INET) 643 else 644 #endif 645 #ifdef INET 646 { 647 hdrlen = sizeof(struct tcpiphdr); 648 ip = mtod(m, struct ip *); 649 if (tw->t_port) { 650 udp = (struct udphdr *)(ip + 1); 651 hdrlen += sizeof(struct udphdr); 652 udp->uh_sport = htons(V_tcp_udp_tunneling_port); 653 udp->uh_dport = tw->t_port; 654 ulen = (hdrlen - sizeof(struct ip)); 655 th = (struct tcphdr *)(udp + 1); 656 } else 657 th = (struct tcphdr *)(ip + 1); 658 tcpip_fillheaders(inp, tw->t_port, ip, th); 659 } 660 #endif 661 to.to_flags = 0; 662 663 /* 664 * Send a timestamp and echo-reply if both our side and our peer 665 * have sent timestamps in our SYN's and this is not a RST. 666 */ 667 if (tw->t_recent && flags == TH_ACK) { 668 to.to_flags |= TOF_TS; 669 to.to_tsval = tcp_ts_getticks() + tw->ts_offset; 670 to.to_tsecr = tw->t_recent; 671 } 672 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 673 674 if (udp) { 675 ulen += optlen; 676 udp->uh_ulen = htons(ulen); 677 } 678 m->m_len = hdrlen + optlen; 679 m->m_pkthdr.len = m->m_len; 680 681 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 682 683 th->th_seq = htonl(tw->snd_nxt); 684 th->th_ack = htonl(tw->rcv_nxt); 685 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 686 th->th_flags = flags; 687 th->th_win = htons(tw->last_win); 688 689 #ifdef INET6 690 if (isipv6) { 691 if (tw->t_port) { 692 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 693 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 694 udp->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 695 th->th_sum = htons(0); 696 } else { 697 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 698 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 699 th->th_sum = in6_cksum_pseudo(ip6, 700 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0); 701 } 702 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 703 TCP_PROBE5(send, NULL, NULL, ip6, NULL, th); 704 error = ip6_output(m, inp->in6p_outputopts, NULL, 705 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 706 } 707 #endif 708 #if defined(INET6) && defined(INET) 709 else 710 #endif 711 #ifdef INET 712 { 713 if (tw->t_port) { 714 m->m_pkthdr.csum_flags = CSUM_UDP; 715 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 716 udp->uh_sum = in_pseudo(ip->ip_src.s_addr, 717 ip->ip_dst.s_addr, htons(ulen + IPPROTO_UDP)); 718 th->th_sum = htons(0); 719 } else { 720 m->m_pkthdr.csum_flags = CSUM_TCP; 721 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 722 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 723 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 724 } 725 ip->ip_len = htons(m->m_pkthdr.len); 726 if (V_path_mtu_discovery) 727 ip->ip_off |= htons(IP_DF); 728 TCP_PROBE5(send, NULL, NULL, ip, NULL, th); 729 error = ip_output(m, inp->inp_options, NULL, 730 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 731 NULL, inp); 732 } 733 #endif 734 if (flags & TH_ACK) 735 TCPSTAT_INC(tcps_sndacks); 736 else 737 TCPSTAT_INC(tcps_sndctrl); 738 TCPSTAT_INC(tcps_sndtotal); 739 return (error); 740 } 741 742 static void 743 tcp_tw_2msl_reset(struct tcptw *tw, int rearm) 744 { 745 746 NET_EPOCH_ASSERT(); 747 INP_WLOCK_ASSERT(tw->tw_inpcb); 748 749 TW_WLOCK(V_tw_lock); 750 if (rearm) 751 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 752 tw->tw_time = ticks + 2 * tcp_msl; 753 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 754 TW_WUNLOCK(V_tw_lock); 755 } 756 757 static void 758 tcp_tw_2msl_stop(struct tcptw *tw, int reuse) 759 { 760 struct ucred *cred; 761 struct inpcb *inp; 762 int released __unused; 763 764 NET_EPOCH_ASSERT(); 765 766 TW_WLOCK(V_tw_lock); 767 inp = tw->tw_inpcb; 768 tw->tw_inpcb = NULL; 769 770 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 771 cred = tw->tw_cred; 772 tw->tw_cred = NULL; 773 TW_WUNLOCK(V_tw_lock); 774 775 if (cred != NULL) 776 crfree(cred); 777 778 released = in_pcbrele_wlocked(inp); 779 KASSERT(!released, ("%s: inp should not be released here", __func__)); 780 781 if (!reuse) 782 uma_zfree(V_tcptw_zone, tw); 783 TCPSTATES_DEC(TCPS_TIME_WAIT); 784 } 785 786 struct tcptw * 787 tcp_tw_2msl_scan(int reuse) 788 { 789 struct tcptw *tw; 790 struct inpcb *inp; 791 792 NET_EPOCH_ASSERT(); 793 794 for (;;) { 795 TW_RLOCK(V_tw_lock); 796 tw = TAILQ_FIRST(&V_twq_2msl); 797 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) { 798 TW_RUNLOCK(V_tw_lock); 799 break; 800 } 801 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL", 802 __func__)); 803 804 inp = tw->tw_inpcb; 805 in_pcbref(inp); 806 TW_RUNLOCK(V_tw_lock); 807 808 INP_WLOCK(inp); 809 tw = intotw(inp); 810 if (in_pcbrele_wlocked(inp)) { 811 if (__predict_true(tw == NULL)) { 812 continue; 813 } else { 814 /* This should not happen as in TIMEWAIT 815 * state the inp should not be destroyed 816 * before its tcptw. If INVARIANTS is 817 * defined panic. 818 */ 819 #ifdef INVARIANTS 820 panic("%s: Panic before an infinite " 821 "loop: INP_TIMEWAIT && (INP_FREED " 822 "|| inp last reference) && tw != " 823 "NULL", __func__); 824 #else 825 log(LOG_ERR, "%s: Avoid an infinite " 826 "loop: INP_TIMEWAIT && (INP_FREED " 827 "|| inp last reference) && tw != " 828 "NULL", __func__); 829 #endif 830 break; 831 } 832 } 833 834 if (tw == NULL) { 835 /* tcp_twclose() has already been called */ 836 INP_WUNLOCK(inp); 837 continue; 838 } 839 840 tcp_twclose(tw, reuse); 841 if (reuse) 842 return tw; 843 } 844 845 return NULL; 846 } 847