tcp_input.c revision 30052
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $Id: tcp_input.c,v 1.62 1997/09/30 16:38:09 fenner Exp $ 35 */ 36 37#include "opt_tcpdebug.h" 38 39#ifndef TUBA_INCLUDE 40#include <sys/param.h> 41#include <sys/queue.h> 42#include <sys/systm.h> 43#include <sys/kernel.h> 44#include <sys/sysctl.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/proc.h> /* for proc0 declaration */ 48#include <sys/protosw.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/errno.h> 52#include <sys/syslog.h> 53 54#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 55 56#include <net/if.h> 57#include <net/route.h> 58 59#include <netinet/in.h> 60#include <netinet/in_systm.h> 61#include <netinet/ip.h> 62#include <netinet/in_pcb.h> 63#include <netinet/ip_var.h> 64#include <netinet/tcp.h> 65#include <netinet/tcp_fsm.h> 66#include <netinet/tcp_seq.h> 67#include <netinet/tcp_timer.h> 68#include <netinet/tcp_var.h> 69#include <netinet/tcpip.h> 70#ifdef TCPDEBUG 71#include <netinet/tcp_debug.h> 72static struct tcpiphdr tcp_saveti; 73#endif 74 75static int tcprexmtthresh = 3; 76tcp_seq tcp_iss; 77tcp_cc tcp_ccgen; 78 79struct tcpstat tcpstat; 80SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 81 CTLFLAG_RD, &tcpstat , tcpstat, ""); 82 83static int log_in_vain = 0; 84SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 85 &log_in_vain, 0, ""); 86 87u_long tcp_now; 88struct inpcbhead tcb; 89struct inpcbinfo tcbinfo; 90 91static void tcp_dooptions __P((struct tcpcb *, 92 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 93static void tcp_pulloutofband __P((struct socket *, 94 struct tcpiphdr *, struct mbuf *)); 95static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 96static void tcp_xmit_timer __P((struct tcpcb *, int)); 97 98#endif /* TUBA_INCLUDE */ 99 100/* 101 * Insert segment ti into reassembly queue of tcp with 102 * control block tp. Return TH_FIN if reassembly now includes 103 * a segment with FIN. The macro form does the common case inline 104 * (segment is the next to be received on an established connection, 105 * and the queue is empty), avoiding linkage into and removal 106 * from the queue and repetition of various conversions. 107 * Set DELACK for segments received in order, but ack immediately 108 * when segments are out of order (so fast retransmit can work). 109 */ 110#ifdef TCP_ACK_HACK 111#define TCP_REASS(tp, ti, m, so, flags) { \ 112 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 113 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 114 (tp)->t_state == TCPS_ESTABLISHED) { \ 115 if (ti->ti_flags & TH_PUSH) \ 116 tp->t_flags |= TF_ACKNOW; \ 117 else \ 118 tp->t_flags |= TF_DELACK; \ 119 (tp)->rcv_nxt += (ti)->ti_len; \ 120 flags = (ti)->ti_flags & TH_FIN; \ 121 tcpstat.tcps_rcvpack++;\ 122 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 123 sbappend(&(so)->so_rcv, (m)); \ 124 sorwakeup(so); \ 125 } else { \ 126 (flags) = tcp_reass((tp), (ti), (m)); \ 127 tp->t_flags |= TF_ACKNOW; \ 128 } \ 129} 130#else 131#define TCP_REASS(tp, ti, m, so, flags) { \ 132 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 133 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 134 (tp)->t_state == TCPS_ESTABLISHED) { \ 135 tp->t_flags |= TF_DELACK; \ 136 (tp)->rcv_nxt += (ti)->ti_len; \ 137 flags = (ti)->ti_flags & TH_FIN; \ 138 tcpstat.tcps_rcvpack++;\ 139 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 140 sbappend(&(so)->so_rcv, (m)); \ 141 sorwakeup(so); \ 142 } else { \ 143 (flags) = tcp_reass((tp), (ti), (m)); \ 144 tp->t_flags |= TF_ACKNOW; \ 145 } \ 146} 147#endif 148#ifndef TUBA_INCLUDE 149 150static int 151tcp_reass(tp, ti, m) 152 register struct tcpcb *tp; 153 register struct tcpiphdr *ti; 154 struct mbuf *m; 155{ 156 register struct tcpiphdr *q; 157 struct socket *so = tp->t_inpcb->inp_socket; 158 int flags; 159 160 /* 161 * Call with ti==0 after become established to 162 * force pre-ESTABLISHED data up to user socket. 163 */ 164 if (ti == 0) 165 goto present; 166 167 /* 168 * Find a segment which begins after this one does. 169 */ 170 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 171 q = (struct tcpiphdr *)q->ti_next) 172 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 173 break; 174 175 /* 176 * If there is a preceding segment, it may provide some of 177 * our data already. If so, drop the data from the incoming 178 * segment. If it provides all of our data, drop us. 179 */ 180 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 181 register int i; 182 q = (struct tcpiphdr *)q->ti_prev; 183 /* conversion to int (in i) handles seq wraparound */ 184 i = q->ti_seq + q->ti_len - ti->ti_seq; 185 if (i > 0) { 186 if (i >= ti->ti_len) { 187 tcpstat.tcps_rcvduppack++; 188 tcpstat.tcps_rcvdupbyte += ti->ti_len; 189 m_freem(m); 190 /* 191 * Try to present any queued data 192 * at the left window edge to the user. 193 * This is needed after the 3-WHS 194 * completes. 195 */ 196 goto present; /* ??? */ 197 } 198 m_adj(m, i); 199 ti->ti_len -= i; 200 ti->ti_seq += i; 201 } 202 q = (struct tcpiphdr *)(q->ti_next); 203 } 204 tcpstat.tcps_rcvoopack++; 205 tcpstat.tcps_rcvoobyte += ti->ti_len; 206 REASS_MBUF(ti) = m; /* XXX */ 207 208 /* 209 * While we overlap succeeding segments trim them or, 210 * if they are completely covered, dequeue them. 211 */ 212 while (q != (struct tcpiphdr *)tp) { 213 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 214 if (i <= 0) 215 break; 216 if (i < q->ti_len) { 217 q->ti_seq += i; 218 q->ti_len -= i; 219 m_adj(REASS_MBUF(q), i); 220 break; 221 } 222 q = (struct tcpiphdr *)q->ti_next; 223 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 224 remque(q->ti_prev); 225 m_freem(m); 226 } 227 228 /* 229 * Stick new segment in its place. 230 */ 231 insque(ti, q->ti_prev); 232 233present: 234 /* 235 * Present data to user, advancing rcv_nxt through 236 * completed sequence space. 237 */ 238 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 239 return (0); 240 ti = tp->seg_next; 241 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 242 return (0); 243 do { 244 tp->rcv_nxt += ti->ti_len; 245 flags = ti->ti_flags & TH_FIN; 246 remque(ti); 247 m = REASS_MBUF(ti); 248 ti = (struct tcpiphdr *)ti->ti_next; 249 if (so->so_state & SS_CANTRCVMORE) 250 m_freem(m); 251 else 252 sbappend(&so->so_rcv, m); 253 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 254 sorwakeup(so); 255 return (flags); 256} 257 258/* 259 * TCP input routine, follows pages 65-76 of the 260 * protocol specification dated September, 1981 very closely. 261 */ 262void 263tcp_input(m, iphlen) 264 register struct mbuf *m; 265 int iphlen; 266{ 267 register struct tcpiphdr *ti; 268 register struct inpcb *inp; 269 u_char *optp = NULL; 270 int optlen = 0; 271 int len, tlen, off; 272 register struct tcpcb *tp = 0; 273 register int tiflags; 274 struct socket *so = 0; 275 int todrop, acked, ourfinisacked, needoutput = 0; 276 struct in_addr laddr; 277 int dropsocket = 0; 278 int iss = 0; 279 u_long tiwin; 280 struct tcpopt to; /* options in this segment */ 281 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 282 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 283#ifdef TCPDEBUG 284 short ostate = 0; 285#endif 286 287 bzero((char *)&to, sizeof(to)); 288 289 tcpstat.tcps_rcvtotal++; 290 /* 291 * Get IP and TCP header together in first mbuf. 292 * Note: IP leaves IP header in first mbuf. 293 */ 294 ti = mtod(m, struct tcpiphdr *); 295 if (iphlen > sizeof (struct ip)) 296 ip_stripoptions(m, (struct mbuf *)0); 297 if (m->m_len < sizeof (struct tcpiphdr)) { 298 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 299 tcpstat.tcps_rcvshort++; 300 return; 301 } 302 ti = mtod(m, struct tcpiphdr *); 303 } 304 305 /* 306 * Checksum extended TCP header and data. 307 */ 308 tlen = ((struct ip *)ti)->ip_len; 309 len = sizeof (struct ip) + tlen; 310 ti->ti_next = ti->ti_prev = 0; 311 ti->ti_x1 = 0; 312 ti->ti_len = (u_short)tlen; 313 HTONS(ti->ti_len); 314 ti->ti_sum = in_cksum(m, len); 315 if (ti->ti_sum) { 316 tcpstat.tcps_rcvbadsum++; 317 goto drop; 318 } 319#endif /* TUBA_INCLUDE */ 320 321 /* 322 * Check that TCP offset makes sense, 323 * pull out TCP options and adjust length. XXX 324 */ 325 off = ti->ti_off << 2; 326 if (off < sizeof (struct tcphdr) || off > tlen) { 327 tcpstat.tcps_rcvbadoff++; 328 goto drop; 329 } 330 tlen -= off; 331 ti->ti_len = tlen; 332 if (off > sizeof (struct tcphdr)) { 333 if (m->m_len < sizeof(struct ip) + off) { 334 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 335 tcpstat.tcps_rcvshort++; 336 return; 337 } 338 ti = mtod(m, struct tcpiphdr *); 339 } 340 optlen = off - sizeof (struct tcphdr); 341 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 342 } 343 tiflags = ti->ti_flags; 344 345 /* 346 * Convert TCP protocol specific fields to host format. 347 */ 348 NTOHL(ti->ti_seq); 349 NTOHL(ti->ti_ack); 350 NTOHS(ti->ti_win); 351 NTOHS(ti->ti_urp); 352 353 /* 354 * Drop TCP, IP headers and TCP options. 355 */ 356 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 357 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 358 359 /* 360 * Locate pcb for segment. 361 */ 362findpcb: 363 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 364 ti->ti_dst, ti->ti_dport, 1); 365 366 /* 367 * If the state is CLOSED (i.e., TCB does not exist) then 368 * all data in the incoming segment is discarded. 369 * If the TCB exists but is in CLOSED state, it is embryonic, 370 * but should either do a listen or a connect soon. 371 */ 372 if (inp == NULL) { 373 if (log_in_vain && tiflags & TH_SYN) { 374 char buf[4*sizeof "123"]; 375 376 strcpy(buf, inet_ntoa(ti->ti_dst)); 377 log(LOG_INFO, "Connection attempt to TCP %s:%d" 378 " from %s:%d\n", 379 buf, ntohs(ti->ti_dport), 380 inet_ntoa(ti->ti_src), ntohs(ti->ti_sport)); 381 } 382 goto dropwithreset; 383 } 384 tp = intotcpcb(inp); 385 if (tp == 0) 386 goto dropwithreset; 387 if (tp->t_state == TCPS_CLOSED) 388 goto drop; 389 390 /* Unscale the window into a 32-bit value. */ 391 if ((tiflags & TH_SYN) == 0) 392 tiwin = ti->ti_win << tp->snd_scale; 393 else 394 tiwin = ti->ti_win; 395 396 so = inp->inp_socket; 397 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 398#ifdef TCPDEBUG 399 if (so->so_options & SO_DEBUG) { 400 ostate = tp->t_state; 401 tcp_saveti = *ti; 402 } 403#endif 404 if (so->so_options & SO_ACCEPTCONN) { 405 register struct tcpcb *tp0 = tp; 406 struct socket *so2; 407 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 408 /* 409 * Note: dropwithreset makes sure we don't 410 * send a RST in response to a RST. 411 */ 412 if (tiflags & TH_ACK) { 413 tcpstat.tcps_badsyn++; 414 goto dropwithreset; 415 } 416 goto drop; 417 } 418 so2 = sonewconn(so, 0); 419 if (so2 == 0) { 420 tcpstat.tcps_listendrop++; 421 so2 = sodropablereq(so); 422 if (so2) { 423 tcp_drop(sototcpcb(so2), ETIMEDOUT); 424 so2 = sonewconn(so, 0); 425 } 426 if (!so2) 427 goto drop; 428 } 429 so = so2; 430 /* 431 * This is ugly, but .... 432 * 433 * Mark socket as temporary until we're 434 * committed to keeping it. The code at 435 * ``drop'' and ``dropwithreset'' check the 436 * flag dropsocket to see if the temporary 437 * socket created here should be discarded. 438 * We mark the socket as discardable until 439 * we're committed to it below in TCPS_LISTEN. 440 */ 441 dropsocket++; 442 inp = (struct inpcb *)so->so_pcb; 443 inp->inp_laddr = ti->ti_dst; 444 inp->inp_lport = ti->ti_dport; 445 in_pcbrehash(inp); 446#if BSD>=43 447 inp->inp_options = ip_srcroute(); 448#endif 449 tp = intotcpcb(inp); 450 tp->t_state = TCPS_LISTEN; 451 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 452 453 /* Compute proper scaling value from buffer space */ 454 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 455 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 456 tp->request_r_scale++; 457 } 458 } 459 460 /* 461 * Segment received on connection. 462 * Reset idle time and keep-alive timer. 463 */ 464 tp->t_idle = 0; 465 if (TCPS_HAVEESTABLISHED(tp->t_state)) 466 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 467 468 /* 469 * Process options if not in LISTEN state, 470 * else do it below (after getting remote address). 471 */ 472 if (tp->t_state != TCPS_LISTEN) 473 tcp_dooptions(tp, optp, optlen, ti, &to); 474 475 /* 476 * Header prediction: check for the two common cases 477 * of a uni-directional data xfer. If the packet has 478 * no control flags, is in-sequence, the window didn't 479 * change and we're not retransmitting, it's a 480 * candidate. If the length is zero and the ack moved 481 * forward, we're the sender side of the xfer. Just 482 * free the data acked & wake any higher level process 483 * that was blocked waiting for space. If the length 484 * is non-zero and the ack didn't move, we're the 485 * receiver side. If we're getting packets in-order 486 * (the reassembly queue is empty), add the data to 487 * the socket buffer and note that we need a delayed ack. 488 * Make sure that the hidden state-flags are also off. 489 * Since we check for TCPS_ESTABLISHED above, it can only 490 * be TH_NEEDSYN. 491 */ 492 if (tp->t_state == TCPS_ESTABLISHED && 493 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 494 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 495 ((to.to_flag & TOF_TS) == 0 || 496 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 497 /* 498 * Using the CC option is compulsory if once started: 499 * the segment is OK if no T/TCP was negotiated or 500 * if the segment has a CC option equal to CCrecv 501 */ 502 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 503 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 504 ti->ti_seq == tp->rcv_nxt && 505 tiwin && tiwin == tp->snd_wnd && 506 tp->snd_nxt == tp->snd_max) { 507 508 /* 509 * If last ACK falls within this segment's sequence numbers, 510 * record the timestamp. 511 * NOTE that the test is modified according to the latest 512 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 513 */ 514 if ((to.to_flag & TOF_TS) != 0 && 515 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 516 tp->ts_recent_age = tcp_now; 517 tp->ts_recent = to.to_tsval; 518 } 519 520 if (ti->ti_len == 0) { 521 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 522 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 523 tp->snd_cwnd >= tp->snd_wnd && 524 tp->t_dupacks < tcprexmtthresh) { 525 /* 526 * this is a pure ack for outstanding data. 527 */ 528 ++tcpstat.tcps_predack; 529 if ((to.to_flag & TOF_TS) != 0) 530 tcp_xmit_timer(tp, 531 tcp_now - to.to_tsecr + 1); 532 else if (tp->t_rtt && 533 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 534 tcp_xmit_timer(tp, tp->t_rtt); 535 acked = ti->ti_ack - tp->snd_una; 536 tcpstat.tcps_rcvackpack++; 537 tcpstat.tcps_rcvackbyte += acked; 538 sbdrop(&so->so_snd, acked); 539 tp->snd_una = ti->ti_ack; 540 m_freem(m); 541 542 /* 543 * If all outstanding data are acked, stop 544 * retransmit timer, otherwise restart timer 545 * using current (possibly backed-off) value. 546 * If process is waiting for space, 547 * wakeup/selwakeup/signal. If data 548 * are ready to send, let tcp_output 549 * decide between more output or persist. 550 */ 551 if (tp->snd_una == tp->snd_max) 552 tp->t_timer[TCPT_REXMT] = 0; 553 else if (tp->t_timer[TCPT_PERSIST] == 0) 554 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 555 556 if (so->so_snd.sb_flags & SB_NOTIFY) 557 sowwakeup(so); 558 if (so->so_snd.sb_cc) 559 (void) tcp_output(tp); 560 return; 561 } 562 } else if (ti->ti_ack == tp->snd_una && 563 tp->seg_next == (struct tcpiphdr *)tp && 564 ti->ti_len <= sbspace(&so->so_rcv)) { 565 /* 566 * this is a pure, in-sequence data packet 567 * with nothing on the reassembly queue and 568 * we have enough buffer space to take it. 569 */ 570 ++tcpstat.tcps_preddat; 571 tp->rcv_nxt += ti->ti_len; 572 tcpstat.tcps_rcvpack++; 573 tcpstat.tcps_rcvbyte += ti->ti_len; 574 /* 575 * Add data to socket buffer. 576 */ 577 sbappend(&so->so_rcv, m); 578 sorwakeup(so); 579#ifdef TCP_ACK_HACK 580 /* 581 * If this is a short packet, then ACK now - with Nagel 582 * congestion avoidance sender won't send more until 583 * he gets an ACK. 584 */ 585 if (tiflags & TH_PUSH) { 586 tp->t_flags |= TF_ACKNOW; 587 tcp_output(tp); 588 } else { 589 tp->t_flags |= TF_DELACK; 590 } 591#else 592 tp->t_flags |= TF_DELACK; 593#endif 594 return; 595 } 596 } 597 598 /* 599 * Calculate amount of space in receive window, 600 * and then do TCP input processing. 601 * Receive window is amount of space in rcv queue, 602 * but not less than advertised window. 603 */ 604 { int win; 605 606 win = sbspace(&so->so_rcv); 607 if (win < 0) 608 win = 0; 609 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 610 } 611 612 switch (tp->t_state) { 613 614 /* 615 * If the state is LISTEN then ignore segment if it contains an RST. 616 * If the segment contains an ACK then it is bad and send a RST. 617 * If it does not contain a SYN then it is not interesting; drop it. 618 * Don't bother responding if the destination was a broadcast. 619 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 620 * tp->iss, and send a segment: 621 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 622 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 623 * Fill in remote peer address fields if not previously specified. 624 * Enter SYN_RECEIVED state, and process any other fields of this 625 * segment in this state. 626 */ 627 case TCPS_LISTEN: { 628 struct mbuf *am; 629 register struct sockaddr_in *sin; 630 631 if (tiflags & TH_RST) 632 goto drop; 633 if (tiflags & TH_ACK) 634 goto dropwithreset; 635 if ((tiflags & TH_SYN) == 0) 636 goto drop; 637 /* 638 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 639 * in_broadcast() should never return true on a received 640 * packet with M_BCAST not set. 641 */ 642 if (m->m_flags & (M_BCAST|M_MCAST) || 643 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 644 goto drop; 645 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 646 M_NOWAIT); 647 if (sin == NULL) 648 goto drop; 649 sin->sin_family = AF_INET; 650 sin->sin_len = sizeof(*sin); 651 sin->sin_addr = ti->ti_src; 652 sin->sin_port = ti->ti_sport; 653 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 654 laddr = inp->inp_laddr; 655 if (inp->inp_laddr.s_addr == INADDR_ANY) 656 inp->inp_laddr = ti->ti_dst; 657 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 658 inp->inp_laddr = laddr; 659 FREE(sin, M_SONAME); 660 goto drop; 661 } 662 FREE(sin, M_SONAME); 663 tp->t_template = tcp_template(tp); 664 if (tp->t_template == 0) { 665 tp = tcp_drop(tp, ENOBUFS); 666 dropsocket = 0; /* socket is already gone */ 667 goto drop; 668 } 669 if ((taop = tcp_gettaocache(inp)) == NULL) { 670 taop = &tao_noncached; 671 bzero(taop, sizeof(*taop)); 672 } 673 tcp_dooptions(tp, optp, optlen, ti, &to); 674 if (iss) 675 tp->iss = iss; 676 else 677 tp->iss = tcp_iss; 678 tcp_iss += TCP_ISSINCR/4; 679 tp->irs = ti->ti_seq; 680 tcp_sendseqinit(tp); 681 tcp_rcvseqinit(tp); 682 /* 683 * Initialization of the tcpcb for transaction; 684 * set SND.WND = SEG.WND, 685 * initialize CCsend and CCrecv. 686 */ 687 tp->snd_wnd = tiwin; /* initial send-window */ 688 tp->cc_send = CC_INC(tcp_ccgen); 689 tp->cc_recv = to.to_cc; 690 /* 691 * Perform TAO test on incoming CC (SEG.CC) option, if any. 692 * - compare SEG.CC against cached CC from the same host, 693 * if any. 694 * - if SEG.CC > chached value, SYN must be new and is accepted 695 * immediately: save new CC in the cache, mark the socket 696 * connected, enter ESTABLISHED state, turn on flag to 697 * send a SYN in the next segment. 698 * A virtual advertised window is set in rcv_adv to 699 * initialize SWS prevention. Then enter normal segment 700 * processing: drop SYN, process data and FIN. 701 * - otherwise do a normal 3-way handshake. 702 */ 703 if ((to.to_flag & TOF_CC) != 0) { 704 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 705 taop->tao_cc = to.to_cc; 706 tp->t_state = TCPS_ESTABLISHED; 707 708 /* 709 * If there is a FIN, or if there is data and the 710 * connection is local, then delay SYN,ACK(SYN) in 711 * the hope of piggy-backing it on a response 712 * segment. Otherwise must send ACK now in case 713 * the other side is slow starting. 714 */ 715 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 716 in_localaddr(inp->inp_faddr))) 717 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 718 else 719 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 720 721 /* 722 * Limit the `virtual advertised window' to TCP_MAXWIN 723 * here. Even if we requested window scaling, it will 724 * become effective only later when our SYN is acked. 725 */ 726 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 727 tcpstat.tcps_connects++; 728 soisconnected(so); 729 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 730 dropsocket = 0; /* committed to socket */ 731 tcpstat.tcps_accepts++; 732 goto trimthenstep6; 733 } 734 /* else do standard 3-way handshake */ 735 } else { 736 /* 737 * No CC option, but maybe CC.NEW: 738 * invalidate cached value. 739 */ 740 taop->tao_cc = 0; 741 } 742 /* 743 * TAO test failed or there was no CC option, 744 * do a standard 3-way handshake. 745 */ 746 tp->t_flags |= TF_ACKNOW; 747 tp->t_state = TCPS_SYN_RECEIVED; 748 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 749 dropsocket = 0; /* committed to socket */ 750 tcpstat.tcps_accepts++; 751 goto trimthenstep6; 752 } 753 754 /* 755 * If the state is SYN_SENT: 756 * if seg contains an ACK, but not for our SYN, drop the input. 757 * if seg contains a RST, then drop the connection. 758 * if seg does not contain SYN, then drop it. 759 * Otherwise this is an acceptable SYN segment 760 * initialize tp->rcv_nxt and tp->irs 761 * if seg contains ack then advance tp->snd_una 762 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 763 * arrange for segment to be acked (eventually) 764 * continue processing rest of data/controls, beginning with URG 765 */ 766 case TCPS_SYN_SENT: 767 if ((taop = tcp_gettaocache(inp)) == NULL) { 768 taop = &tao_noncached; 769 bzero(taop, sizeof(*taop)); 770 } 771 772 if ((tiflags & TH_ACK) && 773 (SEQ_LEQ(ti->ti_ack, tp->iss) || 774 SEQ_GT(ti->ti_ack, tp->snd_max))) { 775 /* 776 * If we have a cached CCsent for the remote host, 777 * hence we haven't just crashed and restarted, 778 * do not send a RST. This may be a retransmission 779 * from the other side after our earlier ACK was lost. 780 * Our new SYN, when it arrives, will serve as the 781 * needed ACK. 782 */ 783 if (taop->tao_ccsent != 0) 784 goto drop; 785 else 786 goto dropwithreset; 787 } 788 if (tiflags & TH_RST) { 789 if (tiflags & TH_ACK) 790 tp = tcp_drop(tp, ECONNREFUSED); 791 goto drop; 792 } 793 if ((tiflags & TH_SYN) == 0) 794 goto drop; 795 tp->snd_wnd = ti->ti_win; /* initial send window */ 796 tp->cc_recv = to.to_cc; /* foreign CC */ 797 798 tp->irs = ti->ti_seq; 799 tcp_rcvseqinit(tp); 800 if (tiflags & TH_ACK) { 801 /* 802 * Our SYN was acked. If segment contains CC.ECHO 803 * option, check it to make sure this segment really 804 * matches our SYN. If not, just drop it as old 805 * duplicate, but send an RST if we're still playing 806 * by the old rules. If no CC.ECHO option, make sure 807 * we don't get fooled into using T/TCP. 808 */ 809 if (to.to_flag & TOF_CCECHO) { 810 if (tp->cc_send != to.to_ccecho) 811 if (taop->tao_ccsent != 0) 812 goto drop; 813 else 814 goto dropwithreset; 815 } else 816 tp->t_flags &= ~TF_RCVD_CC; 817 tcpstat.tcps_connects++; 818 soisconnected(so); 819 /* Do window scaling on this connection? */ 820 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 821 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 822 tp->snd_scale = tp->requested_s_scale; 823 tp->rcv_scale = tp->request_r_scale; 824 } 825 /* Segment is acceptable, update cache if undefined. */ 826 if (taop->tao_ccsent == 0) 827 taop->tao_ccsent = to.to_ccecho; 828 829 tp->rcv_adv += tp->rcv_wnd; 830 tp->snd_una++; /* SYN is acked */ 831 /* 832 * If there's data, delay ACK; if there's also a FIN 833 * ACKNOW will be turned on later. 834 */ 835 if (ti->ti_len != 0) 836 tp->t_flags |= TF_DELACK; 837 else 838 tp->t_flags |= TF_ACKNOW; 839 /* 840 * Received <SYN,ACK> in SYN_SENT[*] state. 841 * Transitions: 842 * SYN_SENT --> ESTABLISHED 843 * SYN_SENT* --> FIN_WAIT_1 844 */ 845 if (tp->t_flags & TF_NEEDFIN) { 846 tp->t_state = TCPS_FIN_WAIT_1; 847 tp->t_flags &= ~TF_NEEDFIN; 848 tiflags &= ~TH_SYN; 849 } else { 850 tp->t_state = TCPS_ESTABLISHED; 851 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 852 } 853 } else { 854 /* 855 * Received initial SYN in SYN-SENT[*] state => simul- 856 * taneous open. If segment contains CC option and there is 857 * a cached CC, apply TAO test; if it succeeds, connection is 858 * half-synchronized. Otherwise, do 3-way handshake: 859 * SYN-SENT -> SYN-RECEIVED 860 * SYN-SENT* -> SYN-RECEIVED* 861 * If there was no CC option, clear cached CC value. 862 */ 863 tp->t_flags |= TF_ACKNOW; 864 tp->t_timer[TCPT_REXMT] = 0; 865 if (to.to_flag & TOF_CC) { 866 if (taop->tao_cc != 0 && 867 CC_GT(to.to_cc, taop->tao_cc)) { 868 /* 869 * update cache and make transition: 870 * SYN-SENT -> ESTABLISHED* 871 * SYN-SENT* -> FIN-WAIT-1* 872 */ 873 taop->tao_cc = to.to_cc; 874 if (tp->t_flags & TF_NEEDFIN) { 875 tp->t_state = TCPS_FIN_WAIT_1; 876 tp->t_flags &= ~TF_NEEDFIN; 877 } else { 878 tp->t_state = TCPS_ESTABLISHED; 879 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 880 } 881 tp->t_flags |= TF_NEEDSYN; 882 } else 883 tp->t_state = TCPS_SYN_RECEIVED; 884 } else { 885 /* CC.NEW or no option => invalidate cache */ 886 taop->tao_cc = 0; 887 tp->t_state = TCPS_SYN_RECEIVED; 888 } 889 } 890 891trimthenstep6: 892 /* 893 * Advance ti->ti_seq to correspond to first data byte. 894 * If data, trim to stay within window, 895 * dropping FIN if necessary. 896 */ 897 ti->ti_seq++; 898 if (ti->ti_len > tp->rcv_wnd) { 899 todrop = ti->ti_len - tp->rcv_wnd; 900 m_adj(m, -todrop); 901 ti->ti_len = tp->rcv_wnd; 902 tiflags &= ~TH_FIN; 903 tcpstat.tcps_rcvpackafterwin++; 904 tcpstat.tcps_rcvbyteafterwin += todrop; 905 } 906 tp->snd_wl1 = ti->ti_seq - 1; 907 tp->rcv_up = ti->ti_seq; 908 /* 909 * Client side of transaction: already sent SYN and data. 910 * If the remote host used T/TCP to validate the SYN, 911 * our data will be ACK'd; if so, enter normal data segment 912 * processing in the middle of step 5, ack processing. 913 * Otherwise, goto step 6. 914 */ 915 if (tiflags & TH_ACK) 916 goto process_ACK; 917 goto step6; 918 /* 919 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 920 * if segment contains a SYN and CC [not CC.NEW] option: 921 * if state == TIME_WAIT and connection duration > MSL, 922 * drop packet and send RST; 923 * 924 * if SEG.CC > CCrecv then is new SYN, and can implicitly 925 * ack the FIN (and data) in retransmission queue. 926 * Complete close and delete TCPCB. Then reprocess 927 * segment, hoping to find new TCPCB in LISTEN state; 928 * 929 * else must be old SYN; drop it. 930 * else do normal processing. 931 */ 932 case TCPS_LAST_ACK: 933 case TCPS_CLOSING: 934 case TCPS_TIME_WAIT: 935 if ((tiflags & TH_SYN) && 936 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 937 if (tp->t_state == TCPS_TIME_WAIT && 938 tp->t_duration > TCPTV_MSL) 939 goto dropwithreset; 940 if (CC_GT(to.to_cc, tp->cc_recv)) { 941 tp = tcp_close(tp); 942 goto findpcb; 943 } 944 else 945 goto drop; 946 } 947 break; /* continue normal processing */ 948 } 949 950 /* 951 * States other than LISTEN or SYN_SENT. 952 * First check timestamp, if present. 953 * Then check the connection count, if present. 954 * Then check that at least some bytes of segment are within 955 * receive window. If segment begins before rcv_nxt, 956 * drop leading data (and SYN); if nothing left, just ack. 957 * 958 * RFC 1323 PAWS: If we have a timestamp reply on this segment 959 * and it's less than ts_recent, drop it. 960 */ 961 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 962 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 963 964 /* Check to see if ts_recent is over 24 days old. */ 965 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 966 /* 967 * Invalidate ts_recent. If this segment updates 968 * ts_recent, the age will be reset later and ts_recent 969 * will get a valid value. If it does not, setting 970 * ts_recent to zero will at least satisfy the 971 * requirement that zero be placed in the timestamp 972 * echo reply when ts_recent isn't valid. The 973 * age isn't reset until we get a valid ts_recent 974 * because we don't want out-of-order segments to be 975 * dropped when ts_recent is old. 976 */ 977 tp->ts_recent = 0; 978 } else { 979 tcpstat.tcps_rcvduppack++; 980 tcpstat.tcps_rcvdupbyte += ti->ti_len; 981 tcpstat.tcps_pawsdrop++; 982 goto dropafterack; 983 } 984 } 985 986 /* 987 * T/TCP mechanism 988 * If T/TCP was negotiated and the segment doesn't have CC, 989 * or if it's CC is wrong then drop the segment. 990 * RST segments do not have to comply with this. 991 */ 992 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 993 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 994 (tiflags & TH_RST) == 0) 995 goto dropafterack; 996 997 todrop = tp->rcv_nxt - ti->ti_seq; 998 if (todrop > 0) { 999 if (tiflags & TH_SYN) { 1000 tiflags &= ~TH_SYN; 1001 ti->ti_seq++; 1002 if (ti->ti_urp > 1) 1003 ti->ti_urp--; 1004 else 1005 tiflags &= ~TH_URG; 1006 todrop--; 1007 } 1008 /* 1009 * Following if statement from Stevens, vol. 2, p. 960. 1010 */ 1011 if (todrop > ti->ti_len 1012 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1013 /* 1014 * Any valid FIN must be to the left of the window. 1015 * At this point the FIN must be a duplicate or out 1016 * of sequence; drop it. 1017 */ 1018 tiflags &= ~TH_FIN; 1019 1020 /* 1021 * Send an ACK to resynchronize and drop any data. 1022 * But keep on processing for RST or ACK. 1023 */ 1024 tp->t_flags |= TF_ACKNOW; 1025 todrop = ti->ti_len; 1026 tcpstat.tcps_rcvduppack++; 1027 tcpstat.tcps_rcvdupbyte += todrop; 1028 } else { 1029 tcpstat.tcps_rcvpartduppack++; 1030 tcpstat.tcps_rcvpartdupbyte += todrop; 1031 } 1032 m_adj(m, todrop); 1033 ti->ti_seq += todrop; 1034 ti->ti_len -= todrop; 1035 if (ti->ti_urp > todrop) 1036 ti->ti_urp -= todrop; 1037 else { 1038 tiflags &= ~TH_URG; 1039 ti->ti_urp = 0; 1040 } 1041 } 1042 1043 /* 1044 * If new data are received on a connection after the 1045 * user processes are gone, then RST the other end. 1046 */ 1047 if ((so->so_state & SS_NOFDREF) && 1048 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1049 tp = tcp_close(tp); 1050 tcpstat.tcps_rcvafterclose++; 1051 goto dropwithreset; 1052 } 1053 1054 /* 1055 * If segment ends after window, drop trailing data 1056 * (and PUSH and FIN); if nothing left, just ACK. 1057 */ 1058 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1059 if (todrop > 0) { 1060 tcpstat.tcps_rcvpackafterwin++; 1061 if (todrop >= ti->ti_len) { 1062 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1063 /* 1064 * If a new connection request is received 1065 * while in TIME_WAIT, drop the old connection 1066 * and start over if the sequence numbers 1067 * are above the previous ones. 1068 */ 1069 if (tiflags & TH_SYN && 1070 tp->t_state == TCPS_TIME_WAIT && 1071 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1072 iss = tp->rcv_nxt + TCP_ISSINCR; 1073 tp = tcp_close(tp); 1074 goto findpcb; 1075 } 1076 /* 1077 * If window is closed can only take segments at 1078 * window edge, and have to drop data and PUSH from 1079 * incoming segments. Continue processing, but 1080 * remember to ack. Otherwise, drop segment 1081 * and ack. 1082 */ 1083 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1084 tp->t_flags |= TF_ACKNOW; 1085 tcpstat.tcps_rcvwinprobe++; 1086 } else 1087 goto dropafterack; 1088 } else 1089 tcpstat.tcps_rcvbyteafterwin += todrop; 1090 m_adj(m, -todrop); 1091 ti->ti_len -= todrop; 1092 tiflags &= ~(TH_PUSH|TH_FIN); 1093 } 1094 1095 /* 1096 * If last ACK falls within this segment's sequence numbers, 1097 * record its timestamp. 1098 * NOTE that the test is modified according to the latest 1099 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1100 */ 1101 if ((to.to_flag & TOF_TS) != 0 && 1102 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1103 tp->ts_recent_age = tcp_now; 1104 tp->ts_recent = to.to_tsval; 1105 } 1106 1107 /* 1108 * If the RST bit is set examine the state: 1109 * SYN_RECEIVED STATE: 1110 * If passive open, return to LISTEN state. 1111 * If active open, inform user that connection was refused. 1112 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1113 * Inform user that connection was reset, and close tcb. 1114 * CLOSING, LAST_ACK, TIME_WAIT STATES 1115 * Close the tcb. 1116 */ 1117 if (tiflags&TH_RST) switch (tp->t_state) { 1118 1119 case TCPS_SYN_RECEIVED: 1120 so->so_error = ECONNREFUSED; 1121 goto close; 1122 1123 case TCPS_ESTABLISHED: 1124 case TCPS_FIN_WAIT_1: 1125 case TCPS_FIN_WAIT_2: 1126 case TCPS_CLOSE_WAIT: 1127 so->so_error = ECONNRESET; 1128 close: 1129 tp->t_state = TCPS_CLOSED; 1130 tcpstat.tcps_drops++; 1131 tp = tcp_close(tp); 1132 goto drop; 1133 1134 case TCPS_CLOSING: 1135 case TCPS_LAST_ACK: 1136 case TCPS_TIME_WAIT: 1137 tp = tcp_close(tp); 1138 goto drop; 1139 } 1140 1141 /* 1142 * If a SYN is in the window, then this is an 1143 * error and we send an RST and drop the connection. 1144 */ 1145 if (tiflags & TH_SYN) { 1146 tp = tcp_drop(tp, ECONNRESET); 1147 goto dropwithreset; 1148 } 1149 1150 /* 1151 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1152 * flag is on (half-synchronized state), then queue data for 1153 * later processing; else drop segment and return. 1154 */ 1155 if ((tiflags & TH_ACK) == 0) { 1156 if (tp->t_state == TCPS_SYN_RECEIVED || 1157 (tp->t_flags & TF_NEEDSYN)) 1158 goto step6; 1159 else 1160 goto drop; 1161 } 1162 1163 /* 1164 * Ack processing. 1165 */ 1166 switch (tp->t_state) { 1167 1168 /* 1169 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1170 * ESTABLISHED state and continue processing, otherwise 1171 * send an RST. 1172 */ 1173 case TCPS_SYN_RECEIVED: 1174 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1175 SEQ_GT(ti->ti_ack, tp->snd_max)) 1176 goto dropwithreset; 1177 1178 tcpstat.tcps_connects++; 1179 soisconnected(so); 1180 /* Do window scaling? */ 1181 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1182 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1183 tp->snd_scale = tp->requested_s_scale; 1184 tp->rcv_scale = tp->request_r_scale; 1185 } 1186 /* 1187 * Upon successful completion of 3-way handshake, 1188 * update cache.CC if it was undefined, pass any queued 1189 * data to the user, and advance state appropriately. 1190 */ 1191 if ((taop = tcp_gettaocache(inp)) != NULL && 1192 taop->tao_cc == 0) 1193 taop->tao_cc = tp->cc_recv; 1194 1195 /* 1196 * Make transitions: 1197 * SYN-RECEIVED -> ESTABLISHED 1198 * SYN-RECEIVED* -> FIN-WAIT-1 1199 */ 1200 if (tp->t_flags & TF_NEEDFIN) { 1201 tp->t_state = TCPS_FIN_WAIT_1; 1202 tp->t_flags &= ~TF_NEEDFIN; 1203 } else { 1204 tp->t_state = TCPS_ESTABLISHED; 1205 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1206 } 1207 /* 1208 * If segment contains data or ACK, will call tcp_reass() 1209 * later; if not, do so now to pass queued data to user. 1210 */ 1211 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1212 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1213 (struct mbuf *)0); 1214 tp->snd_wl1 = ti->ti_seq - 1; 1215 /* fall into ... */ 1216 1217 /* 1218 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1219 * ACKs. If the ack is in the range 1220 * tp->snd_una < ti->ti_ack <= tp->snd_max 1221 * then advance tp->snd_una to ti->ti_ack and drop 1222 * data from the retransmission queue. If this ACK reflects 1223 * more up to date window information we update our window information. 1224 */ 1225 case TCPS_ESTABLISHED: 1226 case TCPS_FIN_WAIT_1: 1227 case TCPS_FIN_WAIT_2: 1228 case TCPS_CLOSE_WAIT: 1229 case TCPS_CLOSING: 1230 case TCPS_LAST_ACK: 1231 case TCPS_TIME_WAIT: 1232 1233 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1234 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1235 tcpstat.tcps_rcvdupack++; 1236 /* 1237 * If we have outstanding data (other than 1238 * a window probe), this is a completely 1239 * duplicate ack (ie, window info didn't 1240 * change), the ack is the biggest we've 1241 * seen and we've seen exactly our rexmt 1242 * threshhold of them, assume a packet 1243 * has been dropped and retransmit it. 1244 * Kludge snd_nxt & the congestion 1245 * window so we send only this one 1246 * packet. 1247 * 1248 * We know we're losing at the current 1249 * window size so do congestion avoidance 1250 * (set ssthresh to half the current window 1251 * and pull our congestion window back to 1252 * the new ssthresh). 1253 * 1254 * Dup acks mean that packets have left the 1255 * network (they're now cached at the receiver) 1256 * so bump cwnd by the amount in the receiver 1257 * to keep a constant cwnd packets in the 1258 * network. 1259 */ 1260 if (tp->t_timer[TCPT_REXMT] == 0 || 1261 ti->ti_ack != tp->snd_una) 1262 tp->t_dupacks = 0; 1263 else if (++tp->t_dupacks == tcprexmtthresh) { 1264 tcp_seq onxt = tp->snd_nxt; 1265 u_int win = 1266 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1267 tp->t_maxseg; 1268 1269 if (win < 2) 1270 win = 2; 1271 tp->snd_ssthresh = win * tp->t_maxseg; 1272 tp->t_timer[TCPT_REXMT] = 0; 1273 tp->t_rtt = 0; 1274 tp->snd_nxt = ti->ti_ack; 1275 tp->snd_cwnd = tp->t_maxseg; 1276 (void) tcp_output(tp); 1277 tp->snd_cwnd = tp->snd_ssthresh + 1278 tp->t_maxseg * tp->t_dupacks; 1279 if (SEQ_GT(onxt, tp->snd_nxt)) 1280 tp->snd_nxt = onxt; 1281 goto drop; 1282 } else if (tp->t_dupacks > tcprexmtthresh) { 1283 tp->snd_cwnd += tp->t_maxseg; 1284 (void) tcp_output(tp); 1285 goto drop; 1286 } 1287 } else 1288 tp->t_dupacks = 0; 1289 break; 1290 } 1291 /* 1292 * If the congestion window was inflated to account 1293 * for the other side's cached packets, retract it. 1294 */ 1295 if (tp->t_dupacks >= tcprexmtthresh && 1296 tp->snd_cwnd > tp->snd_ssthresh) 1297 tp->snd_cwnd = tp->snd_ssthresh; 1298 tp->t_dupacks = 0; 1299 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1300 tcpstat.tcps_rcvacktoomuch++; 1301 goto dropafterack; 1302 } 1303 /* 1304 * If we reach this point, ACK is not a duplicate, 1305 * i.e., it ACKs something we sent. 1306 */ 1307 if (tp->t_flags & TF_NEEDSYN) { 1308 /* 1309 * T/TCP: Connection was half-synchronized, and our 1310 * SYN has been ACK'd (so connection is now fully 1311 * synchronized). Go to non-starred state, 1312 * increment snd_una for ACK of SYN, and check if 1313 * we can do window scaling. 1314 */ 1315 tp->t_flags &= ~TF_NEEDSYN; 1316 tp->snd_una++; 1317 /* Do window scaling? */ 1318 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1319 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1320 tp->snd_scale = tp->requested_s_scale; 1321 tp->rcv_scale = tp->request_r_scale; 1322 } 1323 } 1324 1325process_ACK: 1326 acked = ti->ti_ack - tp->snd_una; 1327 tcpstat.tcps_rcvackpack++; 1328 tcpstat.tcps_rcvackbyte += acked; 1329 1330 /* 1331 * If we have a timestamp reply, update smoothed 1332 * round trip time. If no timestamp is present but 1333 * transmit timer is running and timed sequence 1334 * number was acked, update smoothed round trip time. 1335 * Since we now have an rtt measurement, cancel the 1336 * timer backoff (cf., Phil Karn's retransmit alg.). 1337 * Recompute the initial retransmit timer. 1338 */ 1339 if (to.to_flag & TOF_TS) 1340 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1341 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1342 tcp_xmit_timer(tp,tp->t_rtt); 1343 1344 /* 1345 * If all outstanding data is acked, stop retransmit 1346 * timer and remember to restart (more output or persist). 1347 * If there is more data to be acked, restart retransmit 1348 * timer, using current (possibly backed-off) value. 1349 */ 1350 if (ti->ti_ack == tp->snd_max) { 1351 tp->t_timer[TCPT_REXMT] = 0; 1352 needoutput = 1; 1353 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1354 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1355 1356 /* 1357 * If no data (only SYN) was ACK'd, 1358 * skip rest of ACK processing. 1359 */ 1360 if (acked == 0) 1361 goto step6; 1362 1363 /* 1364 * When new data is acked, open the congestion window. 1365 * If the window gives us less than ssthresh packets 1366 * in flight, open exponentially (maxseg per packet). 1367 * Otherwise open linearly: maxseg per window 1368 * (maxseg^2 / cwnd per packet). 1369 */ 1370 { 1371 register u_int cw = tp->snd_cwnd; 1372 register u_int incr = tp->t_maxseg; 1373 1374 if (cw > tp->snd_ssthresh) 1375 incr = incr * incr / cw; 1376 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1377 } 1378 if (acked > so->so_snd.sb_cc) { 1379 tp->snd_wnd -= so->so_snd.sb_cc; 1380 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1381 ourfinisacked = 1; 1382 } else { 1383 sbdrop(&so->so_snd, acked); 1384 tp->snd_wnd -= acked; 1385 ourfinisacked = 0; 1386 } 1387 if (so->so_snd.sb_flags & SB_NOTIFY) 1388 sowwakeup(so); 1389 tp->snd_una = ti->ti_ack; 1390 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1391 tp->snd_nxt = tp->snd_una; 1392 1393 switch (tp->t_state) { 1394 1395 /* 1396 * In FIN_WAIT_1 STATE in addition to the processing 1397 * for the ESTABLISHED state if our FIN is now acknowledged 1398 * then enter FIN_WAIT_2. 1399 */ 1400 case TCPS_FIN_WAIT_1: 1401 if (ourfinisacked) { 1402 /* 1403 * If we can't receive any more 1404 * data, then closing user can proceed. 1405 * Starting the timer is contrary to the 1406 * specification, but if we don't get a FIN 1407 * we'll hang forever. 1408 */ 1409 if (so->so_state & SS_CANTRCVMORE) { 1410 soisdisconnected(so); 1411 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1412 } 1413 tp->t_state = TCPS_FIN_WAIT_2; 1414 } 1415 break; 1416 1417 /* 1418 * In CLOSING STATE in addition to the processing for 1419 * the ESTABLISHED state if the ACK acknowledges our FIN 1420 * then enter the TIME-WAIT state, otherwise ignore 1421 * the segment. 1422 */ 1423 case TCPS_CLOSING: 1424 if (ourfinisacked) { 1425 tp->t_state = TCPS_TIME_WAIT; 1426 tcp_canceltimers(tp); 1427 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1428 if (tp->cc_recv != 0 && 1429 tp->t_duration < TCPTV_MSL) 1430 tp->t_timer[TCPT_2MSL] = 1431 tp->t_rxtcur * TCPTV_TWTRUNC; 1432 else 1433 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1434 soisdisconnected(so); 1435 } 1436 break; 1437 1438 /* 1439 * In LAST_ACK, we may still be waiting for data to drain 1440 * and/or to be acked, as well as for the ack of our FIN. 1441 * If our FIN is now acknowledged, delete the TCB, 1442 * enter the closed state and return. 1443 */ 1444 case TCPS_LAST_ACK: 1445 if (ourfinisacked) { 1446 tp = tcp_close(tp); 1447 goto drop; 1448 } 1449 break; 1450 1451 /* 1452 * In TIME_WAIT state the only thing that should arrive 1453 * is a retransmission of the remote FIN. Acknowledge 1454 * it and restart the finack timer. 1455 */ 1456 case TCPS_TIME_WAIT: 1457 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1458 goto dropafterack; 1459 } 1460 } 1461 1462step6: 1463 /* 1464 * Update window information. 1465 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1466 */ 1467 if ((tiflags & TH_ACK) && 1468 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1469 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1470 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1471 /* keep track of pure window updates */ 1472 if (ti->ti_len == 0 && 1473 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1474 tcpstat.tcps_rcvwinupd++; 1475 tp->snd_wnd = tiwin; 1476 tp->snd_wl1 = ti->ti_seq; 1477 tp->snd_wl2 = ti->ti_ack; 1478 if (tp->snd_wnd > tp->max_sndwnd) 1479 tp->max_sndwnd = tp->snd_wnd; 1480 needoutput = 1; 1481 } 1482 1483 /* 1484 * Process segments with URG. 1485 */ 1486 if ((tiflags & TH_URG) && ti->ti_urp && 1487 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1488 /* 1489 * This is a kludge, but if we receive and accept 1490 * random urgent pointers, we'll crash in 1491 * soreceive. It's hard to imagine someone 1492 * actually wanting to send this much urgent data. 1493 */ 1494 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1495 ti->ti_urp = 0; /* XXX */ 1496 tiflags &= ~TH_URG; /* XXX */ 1497 goto dodata; /* XXX */ 1498 } 1499 /* 1500 * If this segment advances the known urgent pointer, 1501 * then mark the data stream. This should not happen 1502 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1503 * a FIN has been received from the remote side. 1504 * In these states we ignore the URG. 1505 * 1506 * According to RFC961 (Assigned Protocols), 1507 * the urgent pointer points to the last octet 1508 * of urgent data. We continue, however, 1509 * to consider it to indicate the first octet 1510 * of data past the urgent section as the original 1511 * spec states (in one of two places). 1512 */ 1513 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1514 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1515 so->so_oobmark = so->so_rcv.sb_cc + 1516 (tp->rcv_up - tp->rcv_nxt) - 1; 1517 if (so->so_oobmark == 0) 1518 so->so_state |= SS_RCVATMARK; 1519 sohasoutofband(so); 1520 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1521 } 1522 /* 1523 * Remove out of band data so doesn't get presented to user. 1524 * This can happen independent of advancing the URG pointer, 1525 * but if two URG's are pending at once, some out-of-band 1526 * data may creep in... ick. 1527 */ 1528 if (ti->ti_urp <= (u_long)ti->ti_len 1529#ifdef SO_OOBINLINE 1530 && (so->so_options & SO_OOBINLINE) == 0 1531#endif 1532 ) 1533 tcp_pulloutofband(so, ti, m); 1534 } else 1535 /* 1536 * If no out of band data is expected, 1537 * pull receive urgent pointer along 1538 * with the receive window. 1539 */ 1540 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1541 tp->rcv_up = tp->rcv_nxt; 1542dodata: /* XXX */ 1543 1544 /* 1545 * Process the segment text, merging it into the TCP sequencing queue, 1546 * and arranging for acknowledgment of receipt if necessary. 1547 * This process logically involves adjusting tp->rcv_wnd as data 1548 * is presented to the user (this happens in tcp_usrreq.c, 1549 * case PRU_RCVD). If a FIN has already been received on this 1550 * connection then we just ignore the text. 1551 */ 1552 if ((ti->ti_len || (tiflags&TH_FIN)) && 1553 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1554 TCP_REASS(tp, ti, m, so, tiflags); 1555 /* 1556 * Note the amount of data that peer has sent into 1557 * our window, in order to estimate the sender's 1558 * buffer size. 1559 */ 1560 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1561 } else { 1562 m_freem(m); 1563 tiflags &= ~TH_FIN; 1564 } 1565 1566 /* 1567 * If FIN is received ACK the FIN and let the user know 1568 * that the connection is closing. 1569 */ 1570 if (tiflags & TH_FIN) { 1571 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1572 socantrcvmore(so); 1573 /* 1574 * If connection is half-synchronized 1575 * (ie NEEDSYN flag on) then delay ACK, 1576 * so it may be piggybacked when SYN is sent. 1577 * Otherwise, since we received a FIN then no 1578 * more input can be expected, send ACK now. 1579 */ 1580 if (tp->t_flags & TF_NEEDSYN) 1581 tp->t_flags |= TF_DELACK; 1582 else 1583 tp->t_flags |= TF_ACKNOW; 1584 tp->rcv_nxt++; 1585 } 1586 switch (tp->t_state) { 1587 1588 /* 1589 * In SYN_RECEIVED and ESTABLISHED STATES 1590 * enter the CLOSE_WAIT state. 1591 */ 1592 case TCPS_SYN_RECEIVED: 1593 case TCPS_ESTABLISHED: 1594 tp->t_state = TCPS_CLOSE_WAIT; 1595 break; 1596 1597 /* 1598 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1599 * enter the CLOSING state. 1600 */ 1601 case TCPS_FIN_WAIT_1: 1602 tp->t_state = TCPS_CLOSING; 1603 break; 1604 1605 /* 1606 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1607 * starting the time-wait timer, turning off the other 1608 * standard timers. 1609 */ 1610 case TCPS_FIN_WAIT_2: 1611 tp->t_state = TCPS_TIME_WAIT; 1612 tcp_canceltimers(tp); 1613 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1614 if (tp->cc_recv != 0 && 1615 tp->t_duration < TCPTV_MSL) { 1616 tp->t_timer[TCPT_2MSL] = 1617 tp->t_rxtcur * TCPTV_TWTRUNC; 1618 /* For transaction client, force ACK now. */ 1619 tp->t_flags |= TF_ACKNOW; 1620 } 1621 else 1622 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1623 soisdisconnected(so); 1624 break; 1625 1626 /* 1627 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1628 */ 1629 case TCPS_TIME_WAIT: 1630 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1631 break; 1632 } 1633 } 1634#ifdef TCPDEBUG 1635 if (so->so_options & SO_DEBUG) 1636 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1637#endif 1638 1639 /* 1640 * Return any desired output. 1641 */ 1642 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1643 (void) tcp_output(tp); 1644 return; 1645 1646dropafterack: 1647 /* 1648 * Generate an ACK dropping incoming segment if it occupies 1649 * sequence space, where the ACK reflects our state. 1650 */ 1651 if (tiflags & TH_RST) 1652 goto drop; 1653#ifdef TCPDEBUG 1654 if (so->so_options & SO_DEBUG) 1655 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1656#endif 1657 m_freem(m); 1658 tp->t_flags |= TF_ACKNOW; 1659 (void) tcp_output(tp); 1660 return; 1661 1662dropwithreset: 1663 /* 1664 * Generate a RST, dropping incoming segment. 1665 * Make ACK acceptable to originator of segment. 1666 * Don't bother to respond if destination was broadcast/multicast. 1667 */ 1668 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1669 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1670 goto drop; 1671#ifdef TCPDEBUG 1672 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1673 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1674#endif 1675 if (tiflags & TH_ACK) 1676 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1677 else { 1678 if (tiflags & TH_SYN) 1679 ti->ti_len++; 1680 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1681 TH_RST|TH_ACK); 1682 } 1683 /* destroy temporarily created socket */ 1684 if (dropsocket) 1685 (void) soabort(so); 1686 return; 1687 1688drop: 1689 /* 1690 * Drop space held by incoming segment and return. 1691 */ 1692#ifdef TCPDEBUG 1693 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1694 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1695#endif 1696 m_freem(m); 1697 /* destroy temporarily created socket */ 1698 if (dropsocket) 1699 (void) soabort(so); 1700 return; 1701#ifndef TUBA_INCLUDE 1702} 1703 1704static void 1705tcp_dooptions(tp, cp, cnt, ti, to) 1706 struct tcpcb *tp; 1707 u_char *cp; 1708 int cnt; 1709 struct tcpiphdr *ti; 1710 struct tcpopt *to; 1711{ 1712 u_short mss = 0; 1713 int opt, optlen; 1714 1715 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1716 opt = cp[0]; 1717 if (opt == TCPOPT_EOL) 1718 break; 1719 if (opt == TCPOPT_NOP) 1720 optlen = 1; 1721 else { 1722 optlen = cp[1]; 1723 if (optlen <= 0) 1724 break; 1725 } 1726 switch (opt) { 1727 1728 default: 1729 continue; 1730 1731 case TCPOPT_MAXSEG: 1732 if (optlen != TCPOLEN_MAXSEG) 1733 continue; 1734 if (!(ti->ti_flags & TH_SYN)) 1735 continue; 1736 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1737 NTOHS(mss); 1738 break; 1739 1740 case TCPOPT_WINDOW: 1741 if (optlen != TCPOLEN_WINDOW) 1742 continue; 1743 if (!(ti->ti_flags & TH_SYN)) 1744 continue; 1745 tp->t_flags |= TF_RCVD_SCALE; 1746 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1747 break; 1748 1749 case TCPOPT_TIMESTAMP: 1750 if (optlen != TCPOLEN_TIMESTAMP) 1751 continue; 1752 to->to_flag |= TOF_TS; 1753 bcopy((char *)cp + 2, 1754 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1755 NTOHL(to->to_tsval); 1756 bcopy((char *)cp + 6, 1757 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1758 NTOHL(to->to_tsecr); 1759 1760 /* 1761 * A timestamp received in a SYN makes 1762 * it ok to send timestamp requests and replies. 1763 */ 1764 if (ti->ti_flags & TH_SYN) { 1765 tp->t_flags |= TF_RCVD_TSTMP; 1766 tp->ts_recent = to->to_tsval; 1767 tp->ts_recent_age = tcp_now; 1768 } 1769 break; 1770 case TCPOPT_CC: 1771 if (optlen != TCPOLEN_CC) 1772 continue; 1773 to->to_flag |= TOF_CC; 1774 bcopy((char *)cp + 2, 1775 (char *)&to->to_cc, sizeof(to->to_cc)); 1776 NTOHL(to->to_cc); 1777 /* 1778 * A CC or CC.new option received in a SYN makes 1779 * it ok to send CC in subsequent segments. 1780 */ 1781 if (ti->ti_flags & TH_SYN) 1782 tp->t_flags |= TF_RCVD_CC; 1783 break; 1784 case TCPOPT_CCNEW: 1785 if (optlen != TCPOLEN_CC) 1786 continue; 1787 if (!(ti->ti_flags & TH_SYN)) 1788 continue; 1789 to->to_flag |= TOF_CCNEW; 1790 bcopy((char *)cp + 2, 1791 (char *)&to->to_cc, sizeof(to->to_cc)); 1792 NTOHL(to->to_cc); 1793 /* 1794 * A CC or CC.new option received in a SYN makes 1795 * it ok to send CC in subsequent segments. 1796 */ 1797 tp->t_flags |= TF_RCVD_CC; 1798 break; 1799 case TCPOPT_CCECHO: 1800 if (optlen != TCPOLEN_CC) 1801 continue; 1802 if (!(ti->ti_flags & TH_SYN)) 1803 continue; 1804 to->to_flag |= TOF_CCECHO; 1805 bcopy((char *)cp + 2, 1806 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1807 NTOHL(to->to_ccecho); 1808 break; 1809 } 1810 } 1811 if (ti->ti_flags & TH_SYN) 1812 tcp_mss(tp, mss); /* sets t_maxseg */ 1813} 1814 1815/* 1816 * Pull out of band byte out of a segment so 1817 * it doesn't appear in the user's data queue. 1818 * It is still reflected in the segment length for 1819 * sequencing purposes. 1820 */ 1821static void 1822tcp_pulloutofband(so, ti, m) 1823 struct socket *so; 1824 struct tcpiphdr *ti; 1825 register struct mbuf *m; 1826{ 1827 int cnt = ti->ti_urp - 1; 1828 1829 while (cnt >= 0) { 1830 if (m->m_len > cnt) { 1831 char *cp = mtod(m, caddr_t) + cnt; 1832 struct tcpcb *tp = sototcpcb(so); 1833 1834 tp->t_iobc = *cp; 1835 tp->t_oobflags |= TCPOOB_HAVEDATA; 1836 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1837 m->m_len--; 1838 return; 1839 } 1840 cnt -= m->m_len; 1841 m = m->m_next; 1842 if (m == 0) 1843 break; 1844 } 1845 panic("tcp_pulloutofband"); 1846} 1847 1848/* 1849 * Collect new round-trip time estimate 1850 * and update averages and current timeout. 1851 */ 1852static void 1853tcp_xmit_timer(tp, rtt) 1854 register struct tcpcb *tp; 1855 short rtt; 1856{ 1857 register int delta; 1858 1859 tcpstat.tcps_rttupdated++; 1860 tp->t_rttupdated++; 1861 if (tp->t_srtt != 0) { 1862 /* 1863 * srtt is stored as fixed point with 5 bits after the 1864 * binary point (i.e., scaled by 8). The following magic 1865 * is equivalent to the smoothing algorithm in rfc793 with 1866 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1867 * point). Adjust rtt to origin 0. 1868 */ 1869 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1870 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1871 1872 if ((tp->t_srtt += delta) <= 0) 1873 tp->t_srtt = 1; 1874 1875 /* 1876 * We accumulate a smoothed rtt variance (actually, a 1877 * smoothed mean difference), then set the retransmit 1878 * timer to smoothed rtt + 4 times the smoothed variance. 1879 * rttvar is stored as fixed point with 4 bits after the 1880 * binary point (scaled by 16). The following is 1881 * equivalent to rfc793 smoothing with an alpha of .75 1882 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1883 * rfc793's wired-in beta. 1884 */ 1885 if (delta < 0) 1886 delta = -delta; 1887 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1888 if ((tp->t_rttvar += delta) <= 0) 1889 tp->t_rttvar = 1; 1890 } else { 1891 /* 1892 * No rtt measurement yet - use the unsmoothed rtt. 1893 * Set the variance to half the rtt (so our first 1894 * retransmit happens at 3*rtt). 1895 */ 1896 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1897 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1898 } 1899 tp->t_rtt = 0; 1900 tp->t_rxtshift = 0; 1901 1902 /* 1903 * the retransmit should happen at rtt + 4 * rttvar. 1904 * Because of the way we do the smoothing, srtt and rttvar 1905 * will each average +1/2 tick of bias. When we compute 1906 * the retransmit timer, we want 1/2 tick of rounding and 1907 * 1 extra tick because of +-1/2 tick uncertainty in the 1908 * firing of the timer. The bias will give us exactly the 1909 * 1.5 tick we need. But, because the bias is 1910 * statistical, we have to test that we don't drop below 1911 * the minimum feasible timer (which is 2 ticks). 1912 */ 1913 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1914 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1915 1916 /* 1917 * We received an ack for a packet that wasn't retransmitted; 1918 * it is probably safe to discard any error indications we've 1919 * received recently. This isn't quite right, but close enough 1920 * for now (a route might have failed after we sent a segment, 1921 * and the return path might not be symmetrical). 1922 */ 1923 tp->t_softerror = 0; 1924} 1925 1926/* 1927 * Determine a reasonable value for maxseg size. 1928 * If the route is known, check route for mtu. 1929 * If none, use an mss that can be handled on the outgoing 1930 * interface without forcing IP to fragment; if bigger than 1931 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1932 * to utilize large mbufs. If no route is found, route has no mtu, 1933 * or the destination isn't local, use a default, hopefully conservative 1934 * size (usually 512 or the default IP max size, but no more than the mtu 1935 * of the interface), as we can't discover anything about intervening 1936 * gateways or networks. We also initialize the congestion/slow start 1937 * window to be a single segment if the destination isn't local. 1938 * While looking at the routing entry, we also initialize other path-dependent 1939 * parameters from pre-set or cached values in the routing entry. 1940 * 1941 * Also take into account the space needed for options that we 1942 * send regularly. Make maxseg shorter by that amount to assure 1943 * that we can send maxseg amount of data even when the options 1944 * are present. Store the upper limit of the length of options plus 1945 * data in maxopd. 1946 * 1947 * NOTE that this routine is only called when we process an incoming 1948 * segment, for outgoing segments only tcp_mssopt is called. 1949 * 1950 * In case of T/TCP, we call this routine during implicit connection 1951 * setup as well (offer = -1), to initialize maxseg from the cached 1952 * MSS of our peer. 1953 */ 1954void 1955tcp_mss(tp, offer) 1956 struct tcpcb *tp; 1957 int offer; 1958{ 1959 register struct rtentry *rt; 1960 struct ifnet *ifp; 1961 register int rtt, mss; 1962 u_long bufsize; 1963 struct inpcb *inp; 1964 struct socket *so; 1965 struct rmxp_tao *taop; 1966 int origoffer = offer; 1967 1968 inp = tp->t_inpcb; 1969 if ((rt = tcp_rtlookup(inp)) == NULL) { 1970 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1971 return; 1972 } 1973 ifp = rt->rt_ifp; 1974 so = inp->inp_socket; 1975 1976 taop = rmx_taop(rt->rt_rmx); 1977 /* 1978 * Offer == -1 means that we didn't receive SYN yet, 1979 * use cached value in that case; 1980 */ 1981 if (offer == -1) 1982 offer = taop->tao_mssopt; 1983 /* 1984 * Offer == 0 means that there was no MSS on the SYN segment, 1985 * in this case we use tcp_mssdflt. 1986 */ 1987 if (offer == 0) 1988 offer = tcp_mssdflt; 1989 else 1990 /* 1991 * Sanity check: make sure that maxopd will be large 1992 * enough to allow some data on segments even is the 1993 * all the option space is used (40bytes). Otherwise 1994 * funny things may happen in tcp_output. 1995 */ 1996 offer = max(offer, 64); 1997 taop->tao_mssopt = offer; 1998 1999 /* 2000 * While we're here, check if there's an initial rtt 2001 * or rttvar. Convert from the route-table units 2002 * to scaled multiples of the slow timeout timer. 2003 */ 2004 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2005 /* 2006 * XXX the lock bit for RTT indicates that the value 2007 * is also a minimum value; this is subject to time. 2008 */ 2009 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2010 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2011 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2012 tcpstat.tcps_usedrtt++; 2013 if (rt->rt_rmx.rmx_rttvar) { 2014 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2015 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2016 tcpstat.tcps_usedrttvar++; 2017 } else { 2018 /* default variation is +- 1 rtt */ 2019 tp->t_rttvar = 2020 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2021 } 2022 TCPT_RANGESET(tp->t_rxtcur, 2023 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2024 tp->t_rttmin, TCPTV_REXMTMAX); 2025 } 2026 /* 2027 * if there's an mtu associated with the route, use it 2028 */ 2029 if (rt->rt_rmx.rmx_mtu) 2030 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2031 else 2032 { 2033 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2034 if (!in_localaddr(inp->inp_faddr)) 2035 mss = min(mss, tcp_mssdflt); 2036 } 2037 mss = min(mss, offer); 2038 /* 2039 * maxopd stores the maximum length of data AND options 2040 * in a segment; maxseg is the amount of data in a normal 2041 * segment. We need to store this value (maxopd) apart 2042 * from maxseg, because now every segment carries options 2043 * and thus we normally have somewhat less data in segments. 2044 */ 2045 tp->t_maxopd = mss; 2046 2047 /* 2048 * In case of T/TCP, origoffer==-1 indicates, that no segments 2049 * were received yet. In this case we just guess, otherwise 2050 * we do the same as before T/TCP. 2051 */ 2052 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2053 (origoffer == -1 || 2054 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2055 mss -= TCPOLEN_TSTAMP_APPA; 2056 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2057 (origoffer == -1 || 2058 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2059 mss -= TCPOLEN_CC_APPA; 2060 2061#if (MCLBYTES & (MCLBYTES - 1)) == 0 2062 if (mss > MCLBYTES) 2063 mss &= ~(MCLBYTES-1); 2064#else 2065 if (mss > MCLBYTES) 2066 mss = mss / MCLBYTES * MCLBYTES; 2067#endif 2068 /* 2069 * If there's a pipesize, change the socket buffer 2070 * to that size. Make the socket buffers an integral 2071 * number of mss units; if the mss is larger than 2072 * the socket buffer, decrease the mss. 2073 */ 2074#ifdef RTV_SPIPE 2075 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2076#endif 2077 bufsize = so->so_snd.sb_hiwat; 2078 if (bufsize < mss) 2079 mss = bufsize; 2080 else { 2081 bufsize = roundup(bufsize, mss); 2082 if (bufsize > sb_max) 2083 bufsize = sb_max; 2084 (void)sbreserve(&so->so_snd, bufsize); 2085 } 2086 tp->t_maxseg = mss; 2087 2088#ifdef RTV_RPIPE 2089 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2090#endif 2091 bufsize = so->so_rcv.sb_hiwat; 2092 if (bufsize > mss) { 2093 bufsize = roundup(bufsize, mss); 2094 if (bufsize > sb_max) 2095 bufsize = sb_max; 2096 (void)sbreserve(&so->so_rcv, bufsize); 2097 } 2098 /* 2099 * Don't force slow-start on local network. 2100 */ 2101 if (!in_localaddr(inp->inp_faddr)) 2102 tp->snd_cwnd = mss; 2103 2104 if (rt->rt_rmx.rmx_ssthresh) { 2105 /* 2106 * There's some sort of gateway or interface 2107 * buffer limit on the path. Use this to set 2108 * the slow start threshhold, but set the 2109 * threshold to no less than 2*mss. 2110 */ 2111 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2112 tcpstat.tcps_usedssthresh++; 2113 } 2114} 2115 2116/* 2117 * Determine the MSS option to send on an outgoing SYN. 2118 */ 2119int 2120tcp_mssopt(tp) 2121 struct tcpcb *tp; 2122{ 2123 struct rtentry *rt; 2124 2125 rt = tcp_rtlookup(tp->t_inpcb); 2126 if (rt == NULL) 2127 return tcp_mssdflt; 2128 2129 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2130} 2131#endif /* TUBA_INCLUDE */ 2132