tcp_input.c revision 30813
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 1.63 1997/10/02 02:10:40 davidg Exp $ 35 */ 36 37#include "opt_tcpdebug.h" 38 39#ifndef TUBA_INCLUDE 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/kernel.h> 43#include <sys/sysctl.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/socket.h> 49#include <sys/socketvar.h> 50#include <sys/syslog.h> 51 52#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 53 54#include <net/if.h> 55#include <net/route.h> 56 57#include <netinet/in.h> 58#include <netinet/in_systm.h> 59#include <netinet/ip.h> 60#include <netinet/in_pcb.h> 61#include <netinet/ip_var.h> 62#include <netinet/tcp.h> 63#include <netinet/tcp_fsm.h> 64#include <netinet/tcp_seq.h> 65#include <netinet/tcp_timer.h> 66#include <netinet/tcp_var.h> 67#include <netinet/tcpip.h> 68#ifdef TCPDEBUG 69#include <netinet/tcp_debug.h> 70static struct tcpiphdr tcp_saveti; 71#endif 72 73static int tcprexmtthresh = 3; 74tcp_seq tcp_iss; 75tcp_cc tcp_ccgen; 76 77struct tcpstat tcpstat; 78SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 79 CTLFLAG_RD, &tcpstat , tcpstat, ""); 80 81static int log_in_vain = 0; 82SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 83 &log_in_vain, 0, ""); 84 85u_long tcp_now; 86struct inpcbhead tcb; 87struct inpcbinfo tcbinfo; 88 89static void tcp_dooptions __P((struct tcpcb *, 90 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 91static void tcp_pulloutofband __P((struct socket *, 92 struct tcpiphdr *, struct mbuf *)); 93static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 94static void tcp_xmit_timer __P((struct tcpcb *, int)); 95 96#endif /* TUBA_INCLUDE */ 97 98/* 99 * Insert segment ti into reassembly queue of tcp with 100 * control block tp. Return TH_FIN if reassembly now includes 101 * a segment with FIN. The macro form does the common case inline 102 * (segment is the next to be received on an established connection, 103 * and the queue is empty), avoiding linkage into and removal 104 * from the queue and repetition of various conversions. 105 * Set DELACK for segments received in order, but ack immediately 106 * when segments are out of order (so fast retransmit can work). 107 */ 108#ifdef TCP_ACK_HACK 109#define TCP_REASS(tp, ti, m, so, flags) { \ 110 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 111 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 112 (tp)->t_state == TCPS_ESTABLISHED) { \ 113 if (ti->ti_flags & TH_PUSH) \ 114 tp->t_flags |= TF_ACKNOW; \ 115 else \ 116 tp->t_flags |= TF_DELACK; \ 117 (tp)->rcv_nxt += (ti)->ti_len; \ 118 flags = (ti)->ti_flags & TH_FIN; \ 119 tcpstat.tcps_rcvpack++;\ 120 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 121 sbappend(&(so)->so_rcv, (m)); \ 122 sorwakeup(so); \ 123 } else { \ 124 (flags) = tcp_reass((tp), (ti), (m)); \ 125 tp->t_flags |= TF_ACKNOW; \ 126 } \ 127} 128#else 129#define TCP_REASS(tp, ti, m, so, flags) { \ 130 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 131 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 132 (tp)->t_state == TCPS_ESTABLISHED) { \ 133 tp->t_flags |= TF_DELACK; \ 134 (tp)->rcv_nxt += (ti)->ti_len; \ 135 flags = (ti)->ti_flags & TH_FIN; \ 136 tcpstat.tcps_rcvpack++;\ 137 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 138 sbappend(&(so)->so_rcv, (m)); \ 139 sorwakeup(so); \ 140 } else { \ 141 (flags) = tcp_reass((tp), (ti), (m)); \ 142 tp->t_flags |= TF_ACKNOW; \ 143 } \ 144} 145#endif 146#ifndef TUBA_INCLUDE 147 148static int 149tcp_reass(tp, ti, m) 150 register struct tcpcb *tp; 151 register struct tcpiphdr *ti; 152 struct mbuf *m; 153{ 154 register struct tcpiphdr *q; 155 struct socket *so = tp->t_inpcb->inp_socket; 156 int flags; 157 158 /* 159 * Call with ti==0 after become established to 160 * force pre-ESTABLISHED data up to user socket. 161 */ 162 if (ti == 0) 163 goto present; 164 165 /* 166 * Find a segment which begins after this one does. 167 */ 168 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 169 q = (struct tcpiphdr *)q->ti_next) 170 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 171 break; 172 173 /* 174 * If there is a preceding segment, it may provide some of 175 * our data already. If so, drop the data from the incoming 176 * segment. If it provides all of our data, drop us. 177 */ 178 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 179 register int i; 180 q = (struct tcpiphdr *)q->ti_prev; 181 /* conversion to int (in i) handles seq wraparound */ 182 i = q->ti_seq + q->ti_len - ti->ti_seq; 183 if (i > 0) { 184 if (i >= ti->ti_len) { 185 tcpstat.tcps_rcvduppack++; 186 tcpstat.tcps_rcvdupbyte += ti->ti_len; 187 m_freem(m); 188 /* 189 * Try to present any queued data 190 * at the left window edge to the user. 191 * This is needed after the 3-WHS 192 * completes. 193 */ 194 goto present; /* ??? */ 195 } 196 m_adj(m, i); 197 ti->ti_len -= i; 198 ti->ti_seq += i; 199 } 200 q = (struct tcpiphdr *)(q->ti_next); 201 } 202 tcpstat.tcps_rcvoopack++; 203 tcpstat.tcps_rcvoobyte += ti->ti_len; 204 REASS_MBUF(ti) = m; /* XXX */ 205 206 /* 207 * While we overlap succeeding segments trim them or, 208 * if they are completely covered, dequeue them. 209 */ 210 while (q != (struct tcpiphdr *)tp) { 211 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 212 if (i <= 0) 213 break; 214 if (i < q->ti_len) { 215 q->ti_seq += i; 216 q->ti_len -= i; 217 m_adj(REASS_MBUF(q), i); 218 break; 219 } 220 q = (struct tcpiphdr *)q->ti_next; 221 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 222 remque(q->ti_prev); 223 m_freem(m); 224 } 225 226 /* 227 * Stick new segment in its place. 228 */ 229 insque(ti, q->ti_prev); 230 231present: 232 /* 233 * Present data to user, advancing rcv_nxt through 234 * completed sequence space. 235 */ 236 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 237 return (0); 238 ti = tp->seg_next; 239 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 240 return (0); 241 do { 242 tp->rcv_nxt += ti->ti_len; 243 flags = ti->ti_flags & TH_FIN; 244 remque(ti); 245 m = REASS_MBUF(ti); 246 ti = (struct tcpiphdr *)ti->ti_next; 247 if (so->so_state & SS_CANTRCVMORE) 248 m_freem(m); 249 else 250 sbappend(&so->so_rcv, m); 251 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 252 sorwakeup(so); 253 return (flags); 254} 255 256/* 257 * TCP input routine, follows pages 65-76 of the 258 * protocol specification dated September, 1981 very closely. 259 */ 260void 261tcp_input(m, iphlen) 262 register struct mbuf *m; 263 int iphlen; 264{ 265 register struct tcpiphdr *ti; 266 register struct inpcb *inp; 267 u_char *optp = NULL; 268 int optlen = 0; 269 int len, tlen, off; 270 register struct tcpcb *tp = 0; 271 register int tiflags; 272 struct socket *so = 0; 273 int todrop, acked, ourfinisacked, needoutput = 0; 274 struct in_addr laddr; 275 int dropsocket = 0; 276 int iss = 0; 277 u_long tiwin; 278 struct tcpopt to; /* options in this segment */ 279 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 280 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 281#ifdef TCPDEBUG 282 short ostate = 0; 283#endif 284 285 bzero((char *)&to, sizeof(to)); 286 287 tcpstat.tcps_rcvtotal++; 288 /* 289 * Get IP and TCP header together in first mbuf. 290 * Note: IP leaves IP header in first mbuf. 291 */ 292 ti = mtod(m, struct tcpiphdr *); 293 if (iphlen > sizeof (struct ip)) 294 ip_stripoptions(m, (struct mbuf *)0); 295 if (m->m_len < sizeof (struct tcpiphdr)) { 296 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 297 tcpstat.tcps_rcvshort++; 298 return; 299 } 300 ti = mtod(m, struct tcpiphdr *); 301 } 302 303 /* 304 * Checksum extended TCP header and data. 305 */ 306 tlen = ((struct ip *)ti)->ip_len; 307 len = sizeof (struct ip) + tlen; 308 ti->ti_next = ti->ti_prev = 0; 309 ti->ti_x1 = 0; 310 ti->ti_len = (u_short)tlen; 311 HTONS(ti->ti_len); 312 ti->ti_sum = in_cksum(m, len); 313 if (ti->ti_sum) { 314 tcpstat.tcps_rcvbadsum++; 315 goto drop; 316 } 317#endif /* TUBA_INCLUDE */ 318 319 /* 320 * Check that TCP offset makes sense, 321 * pull out TCP options and adjust length. XXX 322 */ 323 off = ti->ti_off << 2; 324 if (off < sizeof (struct tcphdr) || off > tlen) { 325 tcpstat.tcps_rcvbadoff++; 326 goto drop; 327 } 328 tlen -= off; 329 ti->ti_len = tlen; 330 if (off > sizeof (struct tcphdr)) { 331 if (m->m_len < sizeof(struct ip) + off) { 332 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 333 tcpstat.tcps_rcvshort++; 334 return; 335 } 336 ti = mtod(m, struct tcpiphdr *); 337 } 338 optlen = off - sizeof (struct tcphdr); 339 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 340 } 341 tiflags = ti->ti_flags; 342 343 /* 344 * Convert TCP protocol specific fields to host format. 345 */ 346 NTOHL(ti->ti_seq); 347 NTOHL(ti->ti_ack); 348 NTOHS(ti->ti_win); 349 NTOHS(ti->ti_urp); 350 351 /* 352 * Drop TCP, IP headers and TCP options. 353 */ 354 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 355 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 356 357 /* 358 * Locate pcb for segment. 359 */ 360findpcb: 361 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 362 ti->ti_dst, ti->ti_dport, 1); 363 364 /* 365 * If the state is CLOSED (i.e., TCB does not exist) then 366 * all data in the incoming segment is discarded. 367 * If the TCB exists but is in CLOSED state, it is embryonic, 368 * but should either do a listen or a connect soon. 369 */ 370 if (inp == NULL) { 371 if (log_in_vain && tiflags & TH_SYN) { 372 char buf[4*sizeof "123"]; 373 374 strcpy(buf, inet_ntoa(ti->ti_dst)); 375 log(LOG_INFO, "Connection attempt to TCP %s:%d" 376 " from %s:%d\n", 377 buf, ntohs(ti->ti_dport), 378 inet_ntoa(ti->ti_src), ntohs(ti->ti_sport)); 379 } 380 goto dropwithreset; 381 } 382 tp = intotcpcb(inp); 383 if (tp == 0) 384 goto dropwithreset; 385 if (tp->t_state == TCPS_CLOSED) 386 goto drop; 387 388 /* Unscale the window into a 32-bit value. */ 389 if ((tiflags & TH_SYN) == 0) 390 tiwin = ti->ti_win << tp->snd_scale; 391 else 392 tiwin = ti->ti_win; 393 394 so = inp->inp_socket; 395 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 396#ifdef TCPDEBUG 397 if (so->so_options & SO_DEBUG) { 398 ostate = tp->t_state; 399 tcp_saveti = *ti; 400 } 401#endif 402 if (so->so_options & SO_ACCEPTCONN) { 403 register struct tcpcb *tp0 = tp; 404 struct socket *so2; 405 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 406 /* 407 * Note: dropwithreset makes sure we don't 408 * send a RST in response to a RST. 409 */ 410 if (tiflags & TH_ACK) { 411 tcpstat.tcps_badsyn++; 412 goto dropwithreset; 413 } 414 goto drop; 415 } 416 so2 = sonewconn(so, 0); 417 if (so2 == 0) { 418 tcpstat.tcps_listendrop++; 419 so2 = sodropablereq(so); 420 if (so2) { 421 tcp_drop(sototcpcb(so2), ETIMEDOUT); 422 so2 = sonewconn(so, 0); 423 } 424 if (!so2) 425 goto drop; 426 } 427 so = so2; 428 /* 429 * This is ugly, but .... 430 * 431 * Mark socket as temporary until we're 432 * committed to keeping it. The code at 433 * ``drop'' and ``dropwithreset'' check the 434 * flag dropsocket to see if the temporary 435 * socket created here should be discarded. 436 * We mark the socket as discardable until 437 * we're committed to it below in TCPS_LISTEN. 438 */ 439 dropsocket++; 440 inp = (struct inpcb *)so->so_pcb; 441 inp->inp_laddr = ti->ti_dst; 442 inp->inp_lport = ti->ti_dport; 443 in_pcbrehash(inp); 444#if BSD>=43 445 inp->inp_options = ip_srcroute(); 446#endif 447 tp = intotcpcb(inp); 448 tp->t_state = TCPS_LISTEN; 449 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 450 451 /* Compute proper scaling value from buffer space */ 452 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 453 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 454 tp->request_r_scale++; 455 } 456 } 457 458 /* 459 * Segment received on connection. 460 * Reset idle time and keep-alive timer. 461 */ 462 tp->t_idle = 0; 463 if (TCPS_HAVEESTABLISHED(tp->t_state)) 464 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 465 466 /* 467 * Process options if not in LISTEN state, 468 * else do it below (after getting remote address). 469 */ 470 if (tp->t_state != TCPS_LISTEN) 471 tcp_dooptions(tp, optp, optlen, ti, &to); 472 473 /* 474 * Header prediction: check for the two common cases 475 * of a uni-directional data xfer. If the packet has 476 * no control flags, is in-sequence, the window didn't 477 * change and we're not retransmitting, it's a 478 * candidate. If the length is zero and the ack moved 479 * forward, we're the sender side of the xfer. Just 480 * free the data acked & wake any higher level process 481 * that was blocked waiting for space. If the length 482 * is non-zero and the ack didn't move, we're the 483 * receiver side. If we're getting packets in-order 484 * (the reassembly queue is empty), add the data to 485 * the socket buffer and note that we need a delayed ack. 486 * Make sure that the hidden state-flags are also off. 487 * Since we check for TCPS_ESTABLISHED above, it can only 488 * be TH_NEEDSYN. 489 */ 490 if (tp->t_state == TCPS_ESTABLISHED && 491 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 492 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 493 ((to.to_flag & TOF_TS) == 0 || 494 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 495 /* 496 * Using the CC option is compulsory if once started: 497 * the segment is OK if no T/TCP was negotiated or 498 * if the segment has a CC option equal to CCrecv 499 */ 500 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 501 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 502 ti->ti_seq == tp->rcv_nxt && 503 tiwin && tiwin == tp->snd_wnd && 504 tp->snd_nxt == tp->snd_max) { 505 506 /* 507 * If last ACK falls within this segment's sequence numbers, 508 * record the timestamp. 509 * NOTE that the test is modified according to the latest 510 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 511 */ 512 if ((to.to_flag & TOF_TS) != 0 && 513 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 514 tp->ts_recent_age = tcp_now; 515 tp->ts_recent = to.to_tsval; 516 } 517 518 if (ti->ti_len == 0) { 519 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 520 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 521 tp->snd_cwnd >= tp->snd_wnd && 522 tp->t_dupacks < tcprexmtthresh) { 523 /* 524 * this is a pure ack for outstanding data. 525 */ 526 ++tcpstat.tcps_predack; 527 if ((to.to_flag & TOF_TS) != 0) 528 tcp_xmit_timer(tp, 529 tcp_now - to.to_tsecr + 1); 530 else if (tp->t_rtt && 531 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 532 tcp_xmit_timer(tp, tp->t_rtt); 533 acked = ti->ti_ack - tp->snd_una; 534 tcpstat.tcps_rcvackpack++; 535 tcpstat.tcps_rcvackbyte += acked; 536 sbdrop(&so->so_snd, acked); 537 tp->snd_una = ti->ti_ack; 538 m_freem(m); 539 540 /* 541 * If all outstanding data are acked, stop 542 * retransmit timer, otherwise restart timer 543 * using current (possibly backed-off) value. 544 * If process is waiting for space, 545 * wakeup/selwakeup/signal. If data 546 * are ready to send, let tcp_output 547 * decide between more output or persist. 548 */ 549 if (tp->snd_una == tp->snd_max) 550 tp->t_timer[TCPT_REXMT] = 0; 551 else if (tp->t_timer[TCPT_PERSIST] == 0) 552 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 553 554 if (so->so_snd.sb_flags & SB_NOTIFY) 555 sowwakeup(so); 556 if (so->so_snd.sb_cc) 557 (void) tcp_output(tp); 558 return; 559 } 560 } else if (ti->ti_ack == tp->snd_una && 561 tp->seg_next == (struct tcpiphdr *)tp && 562 ti->ti_len <= sbspace(&so->so_rcv)) { 563 /* 564 * this is a pure, in-sequence data packet 565 * with nothing on the reassembly queue and 566 * we have enough buffer space to take it. 567 */ 568 ++tcpstat.tcps_preddat; 569 tp->rcv_nxt += ti->ti_len; 570 tcpstat.tcps_rcvpack++; 571 tcpstat.tcps_rcvbyte += ti->ti_len; 572 /* 573 * Add data to socket buffer. 574 */ 575 sbappend(&so->so_rcv, m); 576 sorwakeup(so); 577#ifdef TCP_ACK_HACK 578 /* 579 * If this is a short packet, then ACK now - with Nagel 580 * congestion avoidance sender won't send more until 581 * he gets an ACK. 582 */ 583 if (tiflags & TH_PUSH) { 584 tp->t_flags |= TF_ACKNOW; 585 tcp_output(tp); 586 } else { 587 tp->t_flags |= TF_DELACK; 588 } 589#else 590 tp->t_flags |= TF_DELACK; 591#endif 592 return; 593 } 594 } 595 596 /* 597 * Calculate amount of space in receive window, 598 * and then do TCP input processing. 599 * Receive window is amount of space in rcv queue, 600 * but not less than advertised window. 601 */ 602 { int win; 603 604 win = sbspace(&so->so_rcv); 605 if (win < 0) 606 win = 0; 607 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 608 } 609 610 switch (tp->t_state) { 611 612 /* 613 * If the state is LISTEN then ignore segment if it contains an RST. 614 * If the segment contains an ACK then it is bad and send a RST. 615 * If it does not contain a SYN then it is not interesting; drop it. 616 * Don't bother responding if the destination was a broadcast. 617 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 618 * tp->iss, and send a segment: 619 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 620 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 621 * Fill in remote peer address fields if not previously specified. 622 * Enter SYN_RECEIVED state, and process any other fields of this 623 * segment in this state. 624 */ 625 case TCPS_LISTEN: { 626 struct mbuf *am; 627 register struct sockaddr_in *sin; 628 629 if (tiflags & TH_RST) 630 goto drop; 631 if (tiflags & TH_ACK) 632 goto dropwithreset; 633 if ((tiflags & TH_SYN) == 0) 634 goto drop; 635 /* 636 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 637 * in_broadcast() should never return true on a received 638 * packet with M_BCAST not set. 639 */ 640 if (m->m_flags & (M_BCAST|M_MCAST) || 641 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 642 goto drop; 643 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 644 M_NOWAIT); 645 if (sin == NULL) 646 goto drop; 647 sin->sin_family = AF_INET; 648 sin->sin_len = sizeof(*sin); 649 sin->sin_addr = ti->ti_src; 650 sin->sin_port = ti->ti_sport; 651 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 652 laddr = inp->inp_laddr; 653 if (inp->inp_laddr.s_addr == INADDR_ANY) 654 inp->inp_laddr = ti->ti_dst; 655 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 656 inp->inp_laddr = laddr; 657 FREE(sin, M_SONAME); 658 goto drop; 659 } 660 FREE(sin, M_SONAME); 661 tp->t_template = tcp_template(tp); 662 if (tp->t_template == 0) { 663 tp = tcp_drop(tp, ENOBUFS); 664 dropsocket = 0; /* socket is already gone */ 665 goto drop; 666 } 667 if ((taop = tcp_gettaocache(inp)) == NULL) { 668 taop = &tao_noncached; 669 bzero(taop, sizeof(*taop)); 670 } 671 tcp_dooptions(tp, optp, optlen, ti, &to); 672 if (iss) 673 tp->iss = iss; 674 else 675 tp->iss = tcp_iss; 676 tcp_iss += TCP_ISSINCR/4; 677 tp->irs = ti->ti_seq; 678 tcp_sendseqinit(tp); 679 tcp_rcvseqinit(tp); 680 /* 681 * Initialization of the tcpcb for transaction; 682 * set SND.WND = SEG.WND, 683 * initialize CCsend and CCrecv. 684 */ 685 tp->snd_wnd = tiwin; /* initial send-window */ 686 tp->cc_send = CC_INC(tcp_ccgen); 687 tp->cc_recv = to.to_cc; 688 /* 689 * Perform TAO test on incoming CC (SEG.CC) option, if any. 690 * - compare SEG.CC against cached CC from the same host, 691 * if any. 692 * - if SEG.CC > chached value, SYN must be new and is accepted 693 * immediately: save new CC in the cache, mark the socket 694 * connected, enter ESTABLISHED state, turn on flag to 695 * send a SYN in the next segment. 696 * A virtual advertised window is set in rcv_adv to 697 * initialize SWS prevention. Then enter normal segment 698 * processing: drop SYN, process data and FIN. 699 * - otherwise do a normal 3-way handshake. 700 */ 701 if ((to.to_flag & TOF_CC) != 0) { 702 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 703 taop->tao_cc = to.to_cc; 704 tp->t_state = TCPS_ESTABLISHED; 705 706 /* 707 * If there is a FIN, or if there is data and the 708 * connection is local, then delay SYN,ACK(SYN) in 709 * the hope of piggy-backing it on a response 710 * segment. Otherwise must send ACK now in case 711 * the other side is slow starting. 712 */ 713 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 714 in_localaddr(inp->inp_faddr))) 715 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 716 else 717 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 718 719 /* 720 * Limit the `virtual advertised window' to TCP_MAXWIN 721 * here. Even if we requested window scaling, it will 722 * become effective only later when our SYN is acked. 723 */ 724 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 725 tcpstat.tcps_connects++; 726 soisconnected(so); 727 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 728 dropsocket = 0; /* committed to socket */ 729 tcpstat.tcps_accepts++; 730 goto trimthenstep6; 731 } 732 /* else do standard 3-way handshake */ 733 } else { 734 /* 735 * No CC option, but maybe CC.NEW: 736 * invalidate cached value. 737 */ 738 taop->tao_cc = 0; 739 } 740 /* 741 * TAO test failed or there was no CC option, 742 * do a standard 3-way handshake. 743 */ 744 tp->t_flags |= TF_ACKNOW; 745 tp->t_state = TCPS_SYN_RECEIVED; 746 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 747 dropsocket = 0; /* committed to socket */ 748 tcpstat.tcps_accepts++; 749 goto trimthenstep6; 750 } 751 752 /* 753 * If the state is SYN_SENT: 754 * if seg contains an ACK, but not for our SYN, drop the input. 755 * if seg contains a RST, then drop the connection. 756 * if seg does not contain SYN, then drop it. 757 * Otherwise this is an acceptable SYN segment 758 * initialize tp->rcv_nxt and tp->irs 759 * if seg contains ack then advance tp->snd_una 760 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 761 * arrange for segment to be acked (eventually) 762 * continue processing rest of data/controls, beginning with URG 763 */ 764 case TCPS_SYN_SENT: 765 if ((taop = tcp_gettaocache(inp)) == NULL) { 766 taop = &tao_noncached; 767 bzero(taop, sizeof(*taop)); 768 } 769 770 if ((tiflags & TH_ACK) && 771 (SEQ_LEQ(ti->ti_ack, tp->iss) || 772 SEQ_GT(ti->ti_ack, tp->snd_max))) { 773 /* 774 * If we have a cached CCsent for the remote host, 775 * hence we haven't just crashed and restarted, 776 * do not send a RST. This may be a retransmission 777 * from the other side after our earlier ACK was lost. 778 * Our new SYN, when it arrives, will serve as the 779 * needed ACK. 780 */ 781 if (taop->tao_ccsent != 0) 782 goto drop; 783 else 784 goto dropwithreset; 785 } 786 if (tiflags & TH_RST) { 787 if (tiflags & TH_ACK) 788 tp = tcp_drop(tp, ECONNREFUSED); 789 goto drop; 790 } 791 if ((tiflags & TH_SYN) == 0) 792 goto drop; 793 tp->snd_wnd = ti->ti_win; /* initial send window */ 794 tp->cc_recv = to.to_cc; /* foreign CC */ 795 796 tp->irs = ti->ti_seq; 797 tcp_rcvseqinit(tp); 798 if (tiflags & TH_ACK) { 799 /* 800 * Our SYN was acked. If segment contains CC.ECHO 801 * option, check it to make sure this segment really 802 * matches our SYN. If not, just drop it as old 803 * duplicate, but send an RST if we're still playing 804 * by the old rules. If no CC.ECHO option, make sure 805 * we don't get fooled into using T/TCP. 806 */ 807 if (to.to_flag & TOF_CCECHO) { 808 if (tp->cc_send != to.to_ccecho) 809 if (taop->tao_ccsent != 0) 810 goto drop; 811 else 812 goto dropwithreset; 813 } else 814 tp->t_flags &= ~TF_RCVD_CC; 815 tcpstat.tcps_connects++; 816 soisconnected(so); 817 /* Do window scaling on this connection? */ 818 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 819 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 820 tp->snd_scale = tp->requested_s_scale; 821 tp->rcv_scale = tp->request_r_scale; 822 } 823 /* Segment is acceptable, update cache if undefined. */ 824 if (taop->tao_ccsent == 0) 825 taop->tao_ccsent = to.to_ccecho; 826 827 tp->rcv_adv += tp->rcv_wnd; 828 tp->snd_una++; /* SYN is acked */ 829 /* 830 * If there's data, delay ACK; if there's also a FIN 831 * ACKNOW will be turned on later. 832 */ 833 if (ti->ti_len != 0) 834 tp->t_flags |= TF_DELACK; 835 else 836 tp->t_flags |= TF_ACKNOW; 837 /* 838 * Received <SYN,ACK> in SYN_SENT[*] state. 839 * Transitions: 840 * SYN_SENT --> ESTABLISHED 841 * SYN_SENT* --> FIN_WAIT_1 842 */ 843 if (tp->t_flags & TF_NEEDFIN) { 844 tp->t_state = TCPS_FIN_WAIT_1; 845 tp->t_flags &= ~TF_NEEDFIN; 846 tiflags &= ~TH_SYN; 847 } else { 848 tp->t_state = TCPS_ESTABLISHED; 849 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 850 } 851 } else { 852 /* 853 * Received initial SYN in SYN-SENT[*] state => simul- 854 * taneous open. If segment contains CC option and there is 855 * a cached CC, apply TAO test; if it succeeds, connection is 856 * half-synchronized. Otherwise, do 3-way handshake: 857 * SYN-SENT -> SYN-RECEIVED 858 * SYN-SENT* -> SYN-RECEIVED* 859 * If there was no CC option, clear cached CC value. 860 */ 861 tp->t_flags |= TF_ACKNOW; 862 tp->t_timer[TCPT_REXMT] = 0; 863 if (to.to_flag & TOF_CC) { 864 if (taop->tao_cc != 0 && 865 CC_GT(to.to_cc, taop->tao_cc)) { 866 /* 867 * update cache and make transition: 868 * SYN-SENT -> ESTABLISHED* 869 * SYN-SENT* -> FIN-WAIT-1* 870 */ 871 taop->tao_cc = to.to_cc; 872 if (tp->t_flags & TF_NEEDFIN) { 873 tp->t_state = TCPS_FIN_WAIT_1; 874 tp->t_flags &= ~TF_NEEDFIN; 875 } else { 876 tp->t_state = TCPS_ESTABLISHED; 877 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 878 } 879 tp->t_flags |= TF_NEEDSYN; 880 } else 881 tp->t_state = TCPS_SYN_RECEIVED; 882 } else { 883 /* CC.NEW or no option => invalidate cache */ 884 taop->tao_cc = 0; 885 tp->t_state = TCPS_SYN_RECEIVED; 886 } 887 } 888 889trimthenstep6: 890 /* 891 * Advance ti->ti_seq to correspond to first data byte. 892 * If data, trim to stay within window, 893 * dropping FIN if necessary. 894 */ 895 ti->ti_seq++; 896 if (ti->ti_len > tp->rcv_wnd) { 897 todrop = ti->ti_len - tp->rcv_wnd; 898 m_adj(m, -todrop); 899 ti->ti_len = tp->rcv_wnd; 900 tiflags &= ~TH_FIN; 901 tcpstat.tcps_rcvpackafterwin++; 902 tcpstat.tcps_rcvbyteafterwin += todrop; 903 } 904 tp->snd_wl1 = ti->ti_seq - 1; 905 tp->rcv_up = ti->ti_seq; 906 /* 907 * Client side of transaction: already sent SYN and data. 908 * If the remote host used T/TCP to validate the SYN, 909 * our data will be ACK'd; if so, enter normal data segment 910 * processing in the middle of step 5, ack processing. 911 * Otherwise, goto step 6. 912 */ 913 if (tiflags & TH_ACK) 914 goto process_ACK; 915 goto step6; 916 /* 917 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 918 * if segment contains a SYN and CC [not CC.NEW] option: 919 * if state == TIME_WAIT and connection duration > MSL, 920 * drop packet and send RST; 921 * 922 * if SEG.CC > CCrecv then is new SYN, and can implicitly 923 * ack the FIN (and data) in retransmission queue. 924 * Complete close and delete TCPCB. Then reprocess 925 * segment, hoping to find new TCPCB in LISTEN state; 926 * 927 * else must be old SYN; drop it. 928 * else do normal processing. 929 */ 930 case TCPS_LAST_ACK: 931 case TCPS_CLOSING: 932 case TCPS_TIME_WAIT: 933 if ((tiflags & TH_SYN) && 934 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 935 if (tp->t_state == TCPS_TIME_WAIT && 936 tp->t_duration > TCPTV_MSL) 937 goto dropwithreset; 938 if (CC_GT(to.to_cc, tp->cc_recv)) { 939 tp = tcp_close(tp); 940 goto findpcb; 941 } 942 else 943 goto drop; 944 } 945 break; /* continue normal processing */ 946 } 947 948 /* 949 * States other than LISTEN or SYN_SENT. 950 * First check timestamp, if present. 951 * Then check the connection count, if present. 952 * Then check that at least some bytes of segment are within 953 * receive window. If segment begins before rcv_nxt, 954 * drop leading data (and SYN); if nothing left, just ack. 955 * 956 * RFC 1323 PAWS: If we have a timestamp reply on this segment 957 * and it's less than ts_recent, drop it. 958 */ 959 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 960 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 961 962 /* Check to see if ts_recent is over 24 days old. */ 963 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 964 /* 965 * Invalidate ts_recent. If this segment updates 966 * ts_recent, the age will be reset later and ts_recent 967 * will get a valid value. If it does not, setting 968 * ts_recent to zero will at least satisfy the 969 * requirement that zero be placed in the timestamp 970 * echo reply when ts_recent isn't valid. The 971 * age isn't reset until we get a valid ts_recent 972 * because we don't want out-of-order segments to be 973 * dropped when ts_recent is old. 974 */ 975 tp->ts_recent = 0; 976 } else { 977 tcpstat.tcps_rcvduppack++; 978 tcpstat.tcps_rcvdupbyte += ti->ti_len; 979 tcpstat.tcps_pawsdrop++; 980 goto dropafterack; 981 } 982 } 983 984 /* 985 * T/TCP mechanism 986 * If T/TCP was negotiated and the segment doesn't have CC, 987 * or if it's CC is wrong then drop the segment. 988 * RST segments do not have to comply with this. 989 */ 990 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 991 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 992 (tiflags & TH_RST) == 0) 993 goto dropafterack; 994 995 todrop = tp->rcv_nxt - ti->ti_seq; 996 if (todrop > 0) { 997 if (tiflags & TH_SYN) { 998 tiflags &= ~TH_SYN; 999 ti->ti_seq++; 1000 if (ti->ti_urp > 1) 1001 ti->ti_urp--; 1002 else 1003 tiflags &= ~TH_URG; 1004 todrop--; 1005 } 1006 /* 1007 * Following if statement from Stevens, vol. 2, p. 960. 1008 */ 1009 if (todrop > ti->ti_len 1010 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1011 /* 1012 * Any valid FIN must be to the left of the window. 1013 * At this point the FIN must be a duplicate or out 1014 * of sequence; drop it. 1015 */ 1016 tiflags &= ~TH_FIN; 1017 1018 /* 1019 * Send an ACK to resynchronize and drop any data. 1020 * But keep on processing for RST or ACK. 1021 */ 1022 tp->t_flags |= TF_ACKNOW; 1023 todrop = ti->ti_len; 1024 tcpstat.tcps_rcvduppack++; 1025 tcpstat.tcps_rcvdupbyte += todrop; 1026 } else { 1027 tcpstat.tcps_rcvpartduppack++; 1028 tcpstat.tcps_rcvpartdupbyte += todrop; 1029 } 1030 m_adj(m, todrop); 1031 ti->ti_seq += todrop; 1032 ti->ti_len -= todrop; 1033 if (ti->ti_urp > todrop) 1034 ti->ti_urp -= todrop; 1035 else { 1036 tiflags &= ~TH_URG; 1037 ti->ti_urp = 0; 1038 } 1039 } 1040 1041 /* 1042 * If new data are received on a connection after the 1043 * user processes are gone, then RST the other end. 1044 */ 1045 if ((so->so_state & SS_NOFDREF) && 1046 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1047 tp = tcp_close(tp); 1048 tcpstat.tcps_rcvafterclose++; 1049 goto dropwithreset; 1050 } 1051 1052 /* 1053 * If segment ends after window, drop trailing data 1054 * (and PUSH and FIN); if nothing left, just ACK. 1055 */ 1056 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1057 if (todrop > 0) { 1058 tcpstat.tcps_rcvpackafterwin++; 1059 if (todrop >= ti->ti_len) { 1060 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1061 /* 1062 * If a new connection request is received 1063 * while in TIME_WAIT, drop the old connection 1064 * and start over if the sequence numbers 1065 * are above the previous ones. 1066 */ 1067 if (tiflags & TH_SYN && 1068 tp->t_state == TCPS_TIME_WAIT && 1069 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1070 iss = tp->rcv_nxt + TCP_ISSINCR; 1071 tp = tcp_close(tp); 1072 goto findpcb; 1073 } 1074 /* 1075 * If window is closed can only take segments at 1076 * window edge, and have to drop data and PUSH from 1077 * incoming segments. Continue processing, but 1078 * remember to ack. Otherwise, drop segment 1079 * and ack. 1080 */ 1081 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1082 tp->t_flags |= TF_ACKNOW; 1083 tcpstat.tcps_rcvwinprobe++; 1084 } else 1085 goto dropafterack; 1086 } else 1087 tcpstat.tcps_rcvbyteafterwin += todrop; 1088 m_adj(m, -todrop); 1089 ti->ti_len -= todrop; 1090 tiflags &= ~(TH_PUSH|TH_FIN); 1091 } 1092 1093 /* 1094 * If last ACK falls within this segment's sequence numbers, 1095 * record its timestamp. 1096 * NOTE that the test is modified according to the latest 1097 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1098 */ 1099 if ((to.to_flag & TOF_TS) != 0 && 1100 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1101 tp->ts_recent_age = tcp_now; 1102 tp->ts_recent = to.to_tsval; 1103 } 1104 1105 /* 1106 * If the RST bit is set examine the state: 1107 * SYN_RECEIVED STATE: 1108 * If passive open, return to LISTEN state. 1109 * If active open, inform user that connection was refused. 1110 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1111 * Inform user that connection was reset, and close tcb. 1112 * CLOSING, LAST_ACK, TIME_WAIT STATES 1113 * Close the tcb. 1114 */ 1115 if (tiflags&TH_RST) switch (tp->t_state) { 1116 1117 case TCPS_SYN_RECEIVED: 1118 so->so_error = ECONNREFUSED; 1119 goto close; 1120 1121 case TCPS_ESTABLISHED: 1122 case TCPS_FIN_WAIT_1: 1123 case TCPS_FIN_WAIT_2: 1124 case TCPS_CLOSE_WAIT: 1125 so->so_error = ECONNRESET; 1126 close: 1127 tp->t_state = TCPS_CLOSED; 1128 tcpstat.tcps_drops++; 1129 tp = tcp_close(tp); 1130 goto drop; 1131 1132 case TCPS_CLOSING: 1133 case TCPS_LAST_ACK: 1134 case TCPS_TIME_WAIT: 1135 tp = tcp_close(tp); 1136 goto drop; 1137 } 1138 1139 /* 1140 * If a SYN is in the window, then this is an 1141 * error and we send an RST and drop the connection. 1142 */ 1143 if (tiflags & TH_SYN) { 1144 tp = tcp_drop(tp, ECONNRESET); 1145 goto dropwithreset; 1146 } 1147 1148 /* 1149 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1150 * flag is on (half-synchronized state), then queue data for 1151 * later processing; else drop segment and return. 1152 */ 1153 if ((tiflags & TH_ACK) == 0) { 1154 if (tp->t_state == TCPS_SYN_RECEIVED || 1155 (tp->t_flags & TF_NEEDSYN)) 1156 goto step6; 1157 else 1158 goto drop; 1159 } 1160 1161 /* 1162 * Ack processing. 1163 */ 1164 switch (tp->t_state) { 1165 1166 /* 1167 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1168 * ESTABLISHED state and continue processing, otherwise 1169 * send an RST. 1170 */ 1171 case TCPS_SYN_RECEIVED: 1172 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1173 SEQ_GT(ti->ti_ack, tp->snd_max)) 1174 goto dropwithreset; 1175 1176 tcpstat.tcps_connects++; 1177 soisconnected(so); 1178 /* Do window scaling? */ 1179 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1180 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1181 tp->snd_scale = tp->requested_s_scale; 1182 tp->rcv_scale = tp->request_r_scale; 1183 } 1184 /* 1185 * Upon successful completion of 3-way handshake, 1186 * update cache.CC if it was undefined, pass any queued 1187 * data to the user, and advance state appropriately. 1188 */ 1189 if ((taop = tcp_gettaocache(inp)) != NULL && 1190 taop->tao_cc == 0) 1191 taop->tao_cc = tp->cc_recv; 1192 1193 /* 1194 * Make transitions: 1195 * SYN-RECEIVED -> ESTABLISHED 1196 * SYN-RECEIVED* -> FIN-WAIT-1 1197 */ 1198 if (tp->t_flags & TF_NEEDFIN) { 1199 tp->t_state = TCPS_FIN_WAIT_1; 1200 tp->t_flags &= ~TF_NEEDFIN; 1201 } else { 1202 tp->t_state = TCPS_ESTABLISHED; 1203 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1204 } 1205 /* 1206 * If segment contains data or ACK, will call tcp_reass() 1207 * later; if not, do so now to pass queued data to user. 1208 */ 1209 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1210 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1211 (struct mbuf *)0); 1212 tp->snd_wl1 = ti->ti_seq - 1; 1213 /* fall into ... */ 1214 1215 /* 1216 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1217 * ACKs. If the ack is in the range 1218 * tp->snd_una < ti->ti_ack <= tp->snd_max 1219 * then advance tp->snd_una to ti->ti_ack and drop 1220 * data from the retransmission queue. If this ACK reflects 1221 * more up to date window information we update our window information. 1222 */ 1223 case TCPS_ESTABLISHED: 1224 case TCPS_FIN_WAIT_1: 1225 case TCPS_FIN_WAIT_2: 1226 case TCPS_CLOSE_WAIT: 1227 case TCPS_CLOSING: 1228 case TCPS_LAST_ACK: 1229 case TCPS_TIME_WAIT: 1230 1231 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1232 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1233 tcpstat.tcps_rcvdupack++; 1234 /* 1235 * If we have outstanding data (other than 1236 * a window probe), this is a completely 1237 * duplicate ack (ie, window info didn't 1238 * change), the ack is the biggest we've 1239 * seen and we've seen exactly our rexmt 1240 * threshhold of them, assume a packet 1241 * has been dropped and retransmit it. 1242 * Kludge snd_nxt & the congestion 1243 * window so we send only this one 1244 * packet. 1245 * 1246 * We know we're losing at the current 1247 * window size so do congestion avoidance 1248 * (set ssthresh to half the current window 1249 * and pull our congestion window back to 1250 * the new ssthresh). 1251 * 1252 * Dup acks mean that packets have left the 1253 * network (they're now cached at the receiver) 1254 * so bump cwnd by the amount in the receiver 1255 * to keep a constant cwnd packets in the 1256 * network. 1257 */ 1258 if (tp->t_timer[TCPT_REXMT] == 0 || 1259 ti->ti_ack != tp->snd_una) 1260 tp->t_dupacks = 0; 1261 else if (++tp->t_dupacks == tcprexmtthresh) { 1262 tcp_seq onxt = tp->snd_nxt; 1263 u_int win = 1264 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1265 tp->t_maxseg; 1266 1267 if (win < 2) 1268 win = 2; 1269 tp->snd_ssthresh = win * tp->t_maxseg; 1270 tp->t_timer[TCPT_REXMT] = 0; 1271 tp->t_rtt = 0; 1272 tp->snd_nxt = ti->ti_ack; 1273 tp->snd_cwnd = tp->t_maxseg; 1274 (void) tcp_output(tp); 1275 tp->snd_cwnd = tp->snd_ssthresh + 1276 tp->t_maxseg * tp->t_dupacks; 1277 if (SEQ_GT(onxt, tp->snd_nxt)) 1278 tp->snd_nxt = onxt; 1279 goto drop; 1280 } else if (tp->t_dupacks > tcprexmtthresh) { 1281 tp->snd_cwnd += tp->t_maxseg; 1282 (void) tcp_output(tp); 1283 goto drop; 1284 } 1285 } else 1286 tp->t_dupacks = 0; 1287 break; 1288 } 1289 /* 1290 * If the congestion window was inflated to account 1291 * for the other side's cached packets, retract it. 1292 */ 1293 if (tp->t_dupacks >= tcprexmtthresh && 1294 tp->snd_cwnd > tp->snd_ssthresh) 1295 tp->snd_cwnd = tp->snd_ssthresh; 1296 tp->t_dupacks = 0; 1297 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1298 tcpstat.tcps_rcvacktoomuch++; 1299 goto dropafterack; 1300 } 1301 /* 1302 * If we reach this point, ACK is not a duplicate, 1303 * i.e., it ACKs something we sent. 1304 */ 1305 if (tp->t_flags & TF_NEEDSYN) { 1306 /* 1307 * T/TCP: Connection was half-synchronized, and our 1308 * SYN has been ACK'd (so connection is now fully 1309 * synchronized). Go to non-starred state, 1310 * increment snd_una for ACK of SYN, and check if 1311 * we can do window scaling. 1312 */ 1313 tp->t_flags &= ~TF_NEEDSYN; 1314 tp->snd_una++; 1315 /* Do window scaling? */ 1316 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1317 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1318 tp->snd_scale = tp->requested_s_scale; 1319 tp->rcv_scale = tp->request_r_scale; 1320 } 1321 } 1322 1323process_ACK: 1324 acked = ti->ti_ack - tp->snd_una; 1325 tcpstat.tcps_rcvackpack++; 1326 tcpstat.tcps_rcvackbyte += acked; 1327 1328 /* 1329 * If we have a timestamp reply, update smoothed 1330 * round trip time. If no timestamp is present but 1331 * transmit timer is running and timed sequence 1332 * number was acked, update smoothed round trip time. 1333 * Since we now have an rtt measurement, cancel the 1334 * timer backoff (cf., Phil Karn's retransmit alg.). 1335 * Recompute the initial retransmit timer. 1336 */ 1337 if (to.to_flag & TOF_TS) 1338 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1339 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1340 tcp_xmit_timer(tp,tp->t_rtt); 1341 1342 /* 1343 * If all outstanding data is acked, stop retransmit 1344 * timer and remember to restart (more output or persist). 1345 * If there is more data to be acked, restart retransmit 1346 * timer, using current (possibly backed-off) value. 1347 */ 1348 if (ti->ti_ack == tp->snd_max) { 1349 tp->t_timer[TCPT_REXMT] = 0; 1350 needoutput = 1; 1351 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1352 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1353 1354 /* 1355 * If no data (only SYN) was ACK'd, 1356 * skip rest of ACK processing. 1357 */ 1358 if (acked == 0) 1359 goto step6; 1360 1361 /* 1362 * When new data is acked, open the congestion window. 1363 * If the window gives us less than ssthresh packets 1364 * in flight, open exponentially (maxseg per packet). 1365 * Otherwise open linearly: maxseg per window 1366 * (maxseg^2 / cwnd per packet). 1367 */ 1368 { 1369 register u_int cw = tp->snd_cwnd; 1370 register u_int incr = tp->t_maxseg; 1371 1372 if (cw > tp->snd_ssthresh) 1373 incr = incr * incr / cw; 1374 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1375 } 1376 if (acked > so->so_snd.sb_cc) { 1377 tp->snd_wnd -= so->so_snd.sb_cc; 1378 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1379 ourfinisacked = 1; 1380 } else { 1381 sbdrop(&so->so_snd, acked); 1382 tp->snd_wnd -= acked; 1383 ourfinisacked = 0; 1384 } 1385 if (so->so_snd.sb_flags & SB_NOTIFY) 1386 sowwakeup(so); 1387 tp->snd_una = ti->ti_ack; 1388 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1389 tp->snd_nxt = tp->snd_una; 1390 1391 switch (tp->t_state) { 1392 1393 /* 1394 * In FIN_WAIT_1 STATE in addition to the processing 1395 * for the ESTABLISHED state if our FIN is now acknowledged 1396 * then enter FIN_WAIT_2. 1397 */ 1398 case TCPS_FIN_WAIT_1: 1399 if (ourfinisacked) { 1400 /* 1401 * If we can't receive any more 1402 * data, then closing user can proceed. 1403 * Starting the timer is contrary to the 1404 * specification, but if we don't get a FIN 1405 * we'll hang forever. 1406 */ 1407 if (so->so_state & SS_CANTRCVMORE) { 1408 soisdisconnected(so); 1409 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1410 } 1411 tp->t_state = TCPS_FIN_WAIT_2; 1412 } 1413 break; 1414 1415 /* 1416 * In CLOSING STATE in addition to the processing for 1417 * the ESTABLISHED state if the ACK acknowledges our FIN 1418 * then enter the TIME-WAIT state, otherwise ignore 1419 * the segment. 1420 */ 1421 case TCPS_CLOSING: 1422 if (ourfinisacked) { 1423 tp->t_state = TCPS_TIME_WAIT; 1424 tcp_canceltimers(tp); 1425 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1426 if (tp->cc_recv != 0 && 1427 tp->t_duration < TCPTV_MSL) 1428 tp->t_timer[TCPT_2MSL] = 1429 tp->t_rxtcur * TCPTV_TWTRUNC; 1430 else 1431 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1432 soisdisconnected(so); 1433 } 1434 break; 1435 1436 /* 1437 * In LAST_ACK, we may still be waiting for data to drain 1438 * and/or to be acked, as well as for the ack of our FIN. 1439 * If our FIN is now acknowledged, delete the TCB, 1440 * enter the closed state and return. 1441 */ 1442 case TCPS_LAST_ACK: 1443 if (ourfinisacked) { 1444 tp = tcp_close(tp); 1445 goto drop; 1446 } 1447 break; 1448 1449 /* 1450 * In TIME_WAIT state the only thing that should arrive 1451 * is a retransmission of the remote FIN. Acknowledge 1452 * it and restart the finack timer. 1453 */ 1454 case TCPS_TIME_WAIT: 1455 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1456 goto dropafterack; 1457 } 1458 } 1459 1460step6: 1461 /* 1462 * Update window information. 1463 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1464 */ 1465 if ((tiflags & TH_ACK) && 1466 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1467 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1468 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1469 /* keep track of pure window updates */ 1470 if (ti->ti_len == 0 && 1471 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1472 tcpstat.tcps_rcvwinupd++; 1473 tp->snd_wnd = tiwin; 1474 tp->snd_wl1 = ti->ti_seq; 1475 tp->snd_wl2 = ti->ti_ack; 1476 if (tp->snd_wnd > tp->max_sndwnd) 1477 tp->max_sndwnd = tp->snd_wnd; 1478 needoutput = 1; 1479 } 1480 1481 /* 1482 * Process segments with URG. 1483 */ 1484 if ((tiflags & TH_URG) && ti->ti_urp && 1485 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1486 /* 1487 * This is a kludge, but if we receive and accept 1488 * random urgent pointers, we'll crash in 1489 * soreceive. It's hard to imagine someone 1490 * actually wanting to send this much urgent data. 1491 */ 1492 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1493 ti->ti_urp = 0; /* XXX */ 1494 tiflags &= ~TH_URG; /* XXX */ 1495 goto dodata; /* XXX */ 1496 } 1497 /* 1498 * If this segment advances the known urgent pointer, 1499 * then mark the data stream. This should not happen 1500 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1501 * a FIN has been received from the remote side. 1502 * In these states we ignore the URG. 1503 * 1504 * According to RFC961 (Assigned Protocols), 1505 * the urgent pointer points to the last octet 1506 * of urgent data. We continue, however, 1507 * to consider it to indicate the first octet 1508 * of data past the urgent section as the original 1509 * spec states (in one of two places). 1510 */ 1511 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1512 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1513 so->so_oobmark = so->so_rcv.sb_cc + 1514 (tp->rcv_up - tp->rcv_nxt) - 1; 1515 if (so->so_oobmark == 0) 1516 so->so_state |= SS_RCVATMARK; 1517 sohasoutofband(so); 1518 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1519 } 1520 /* 1521 * Remove out of band data so doesn't get presented to user. 1522 * This can happen independent of advancing the URG pointer, 1523 * but if two URG's are pending at once, some out-of-band 1524 * data may creep in... ick. 1525 */ 1526 if (ti->ti_urp <= (u_long)ti->ti_len 1527#ifdef SO_OOBINLINE 1528 && (so->so_options & SO_OOBINLINE) == 0 1529#endif 1530 ) 1531 tcp_pulloutofband(so, ti, m); 1532 } else 1533 /* 1534 * If no out of band data is expected, 1535 * pull receive urgent pointer along 1536 * with the receive window. 1537 */ 1538 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1539 tp->rcv_up = tp->rcv_nxt; 1540dodata: /* XXX */ 1541 1542 /* 1543 * Process the segment text, merging it into the TCP sequencing queue, 1544 * and arranging for acknowledgment of receipt if necessary. 1545 * This process logically involves adjusting tp->rcv_wnd as data 1546 * is presented to the user (this happens in tcp_usrreq.c, 1547 * case PRU_RCVD). If a FIN has already been received on this 1548 * connection then we just ignore the text. 1549 */ 1550 if ((ti->ti_len || (tiflags&TH_FIN)) && 1551 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1552 TCP_REASS(tp, ti, m, so, tiflags); 1553 /* 1554 * Note the amount of data that peer has sent into 1555 * our window, in order to estimate the sender's 1556 * buffer size. 1557 */ 1558 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1559 } else { 1560 m_freem(m); 1561 tiflags &= ~TH_FIN; 1562 } 1563 1564 /* 1565 * If FIN is received ACK the FIN and let the user know 1566 * that the connection is closing. 1567 */ 1568 if (tiflags & TH_FIN) { 1569 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1570 socantrcvmore(so); 1571 /* 1572 * If connection is half-synchronized 1573 * (ie NEEDSYN flag on) then delay ACK, 1574 * so it may be piggybacked when SYN is sent. 1575 * Otherwise, since we received a FIN then no 1576 * more input can be expected, send ACK now. 1577 */ 1578 if (tp->t_flags & TF_NEEDSYN) 1579 tp->t_flags |= TF_DELACK; 1580 else 1581 tp->t_flags |= TF_ACKNOW; 1582 tp->rcv_nxt++; 1583 } 1584 switch (tp->t_state) { 1585 1586 /* 1587 * In SYN_RECEIVED and ESTABLISHED STATES 1588 * enter the CLOSE_WAIT state. 1589 */ 1590 case TCPS_SYN_RECEIVED: 1591 case TCPS_ESTABLISHED: 1592 tp->t_state = TCPS_CLOSE_WAIT; 1593 break; 1594 1595 /* 1596 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1597 * enter the CLOSING state. 1598 */ 1599 case TCPS_FIN_WAIT_1: 1600 tp->t_state = TCPS_CLOSING; 1601 break; 1602 1603 /* 1604 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1605 * starting the time-wait timer, turning off the other 1606 * standard timers. 1607 */ 1608 case TCPS_FIN_WAIT_2: 1609 tp->t_state = TCPS_TIME_WAIT; 1610 tcp_canceltimers(tp); 1611 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1612 if (tp->cc_recv != 0 && 1613 tp->t_duration < TCPTV_MSL) { 1614 tp->t_timer[TCPT_2MSL] = 1615 tp->t_rxtcur * TCPTV_TWTRUNC; 1616 /* For transaction client, force ACK now. */ 1617 tp->t_flags |= TF_ACKNOW; 1618 } 1619 else 1620 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1621 soisdisconnected(so); 1622 break; 1623 1624 /* 1625 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1626 */ 1627 case TCPS_TIME_WAIT: 1628 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1629 break; 1630 } 1631 } 1632#ifdef TCPDEBUG 1633 if (so->so_options & SO_DEBUG) 1634 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1635#endif 1636 1637 /* 1638 * Return any desired output. 1639 */ 1640 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1641 (void) tcp_output(tp); 1642 return; 1643 1644dropafterack: 1645 /* 1646 * Generate an ACK dropping incoming segment if it occupies 1647 * sequence space, where the ACK reflects our state. 1648 */ 1649 if (tiflags & TH_RST) 1650 goto drop; 1651#ifdef TCPDEBUG 1652 if (so->so_options & SO_DEBUG) 1653 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1654#endif 1655 m_freem(m); 1656 tp->t_flags |= TF_ACKNOW; 1657 (void) tcp_output(tp); 1658 return; 1659 1660dropwithreset: 1661 /* 1662 * Generate a RST, dropping incoming segment. 1663 * Make ACK acceptable to originator of segment. 1664 * Don't bother to respond if destination was broadcast/multicast. 1665 */ 1666 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1667 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1668 goto drop; 1669#ifdef TCPDEBUG 1670 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1671 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1672#endif 1673 if (tiflags & TH_ACK) 1674 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1675 else { 1676 if (tiflags & TH_SYN) 1677 ti->ti_len++; 1678 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1679 TH_RST|TH_ACK); 1680 } 1681 /* destroy temporarily created socket */ 1682 if (dropsocket) 1683 (void) soabort(so); 1684 return; 1685 1686drop: 1687 /* 1688 * Drop space held by incoming segment and return. 1689 */ 1690#ifdef TCPDEBUG 1691 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1692 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1693#endif 1694 m_freem(m); 1695 /* destroy temporarily created socket */ 1696 if (dropsocket) 1697 (void) soabort(so); 1698 return; 1699#ifndef TUBA_INCLUDE 1700} 1701 1702static void 1703tcp_dooptions(tp, cp, cnt, ti, to) 1704 struct tcpcb *tp; 1705 u_char *cp; 1706 int cnt; 1707 struct tcpiphdr *ti; 1708 struct tcpopt *to; 1709{ 1710 u_short mss = 0; 1711 int opt, optlen; 1712 1713 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1714 opt = cp[0]; 1715 if (opt == TCPOPT_EOL) 1716 break; 1717 if (opt == TCPOPT_NOP) 1718 optlen = 1; 1719 else { 1720 optlen = cp[1]; 1721 if (optlen <= 0) 1722 break; 1723 } 1724 switch (opt) { 1725 1726 default: 1727 continue; 1728 1729 case TCPOPT_MAXSEG: 1730 if (optlen != TCPOLEN_MAXSEG) 1731 continue; 1732 if (!(ti->ti_flags & TH_SYN)) 1733 continue; 1734 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1735 NTOHS(mss); 1736 break; 1737 1738 case TCPOPT_WINDOW: 1739 if (optlen != TCPOLEN_WINDOW) 1740 continue; 1741 if (!(ti->ti_flags & TH_SYN)) 1742 continue; 1743 tp->t_flags |= TF_RCVD_SCALE; 1744 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1745 break; 1746 1747 case TCPOPT_TIMESTAMP: 1748 if (optlen != TCPOLEN_TIMESTAMP) 1749 continue; 1750 to->to_flag |= TOF_TS; 1751 bcopy((char *)cp + 2, 1752 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1753 NTOHL(to->to_tsval); 1754 bcopy((char *)cp + 6, 1755 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1756 NTOHL(to->to_tsecr); 1757 1758 /* 1759 * A timestamp received in a SYN makes 1760 * it ok to send timestamp requests and replies. 1761 */ 1762 if (ti->ti_flags & TH_SYN) { 1763 tp->t_flags |= TF_RCVD_TSTMP; 1764 tp->ts_recent = to->to_tsval; 1765 tp->ts_recent_age = tcp_now; 1766 } 1767 break; 1768 case TCPOPT_CC: 1769 if (optlen != TCPOLEN_CC) 1770 continue; 1771 to->to_flag |= TOF_CC; 1772 bcopy((char *)cp + 2, 1773 (char *)&to->to_cc, sizeof(to->to_cc)); 1774 NTOHL(to->to_cc); 1775 /* 1776 * A CC or CC.new option received in a SYN makes 1777 * it ok to send CC in subsequent segments. 1778 */ 1779 if (ti->ti_flags & TH_SYN) 1780 tp->t_flags |= TF_RCVD_CC; 1781 break; 1782 case TCPOPT_CCNEW: 1783 if (optlen != TCPOLEN_CC) 1784 continue; 1785 if (!(ti->ti_flags & TH_SYN)) 1786 continue; 1787 to->to_flag |= TOF_CCNEW; 1788 bcopy((char *)cp + 2, 1789 (char *)&to->to_cc, sizeof(to->to_cc)); 1790 NTOHL(to->to_cc); 1791 /* 1792 * A CC or CC.new option received in a SYN makes 1793 * it ok to send CC in subsequent segments. 1794 */ 1795 tp->t_flags |= TF_RCVD_CC; 1796 break; 1797 case TCPOPT_CCECHO: 1798 if (optlen != TCPOLEN_CC) 1799 continue; 1800 if (!(ti->ti_flags & TH_SYN)) 1801 continue; 1802 to->to_flag |= TOF_CCECHO; 1803 bcopy((char *)cp + 2, 1804 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1805 NTOHL(to->to_ccecho); 1806 break; 1807 } 1808 } 1809 if (ti->ti_flags & TH_SYN) 1810 tcp_mss(tp, mss); /* sets t_maxseg */ 1811} 1812 1813/* 1814 * Pull out of band byte out of a segment so 1815 * it doesn't appear in the user's data queue. 1816 * It is still reflected in the segment length for 1817 * sequencing purposes. 1818 */ 1819static void 1820tcp_pulloutofband(so, ti, m) 1821 struct socket *so; 1822 struct tcpiphdr *ti; 1823 register struct mbuf *m; 1824{ 1825 int cnt = ti->ti_urp - 1; 1826 1827 while (cnt >= 0) { 1828 if (m->m_len > cnt) { 1829 char *cp = mtod(m, caddr_t) + cnt; 1830 struct tcpcb *tp = sototcpcb(so); 1831 1832 tp->t_iobc = *cp; 1833 tp->t_oobflags |= TCPOOB_HAVEDATA; 1834 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1835 m->m_len--; 1836 return; 1837 } 1838 cnt -= m->m_len; 1839 m = m->m_next; 1840 if (m == 0) 1841 break; 1842 } 1843 panic("tcp_pulloutofband"); 1844} 1845 1846/* 1847 * Collect new round-trip time estimate 1848 * and update averages and current timeout. 1849 */ 1850static void 1851tcp_xmit_timer(tp, rtt) 1852 register struct tcpcb *tp; 1853 short rtt; 1854{ 1855 register int delta; 1856 1857 tcpstat.tcps_rttupdated++; 1858 tp->t_rttupdated++; 1859 if (tp->t_srtt != 0) { 1860 /* 1861 * srtt is stored as fixed point with 5 bits after the 1862 * binary point (i.e., scaled by 8). The following magic 1863 * is equivalent to the smoothing algorithm in rfc793 with 1864 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1865 * point). Adjust rtt to origin 0. 1866 */ 1867 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1868 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1869 1870 if ((tp->t_srtt += delta) <= 0) 1871 tp->t_srtt = 1; 1872 1873 /* 1874 * We accumulate a smoothed rtt variance (actually, a 1875 * smoothed mean difference), then set the retransmit 1876 * timer to smoothed rtt + 4 times the smoothed variance. 1877 * rttvar is stored as fixed point with 4 bits after the 1878 * binary point (scaled by 16). The following is 1879 * equivalent to rfc793 smoothing with an alpha of .75 1880 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1881 * rfc793's wired-in beta. 1882 */ 1883 if (delta < 0) 1884 delta = -delta; 1885 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1886 if ((tp->t_rttvar += delta) <= 0) 1887 tp->t_rttvar = 1; 1888 } else { 1889 /* 1890 * No rtt measurement yet - use the unsmoothed rtt. 1891 * Set the variance to half the rtt (so our first 1892 * retransmit happens at 3*rtt). 1893 */ 1894 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1895 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1896 } 1897 tp->t_rtt = 0; 1898 tp->t_rxtshift = 0; 1899 1900 /* 1901 * the retransmit should happen at rtt + 4 * rttvar. 1902 * Because of the way we do the smoothing, srtt and rttvar 1903 * will each average +1/2 tick of bias. When we compute 1904 * the retransmit timer, we want 1/2 tick of rounding and 1905 * 1 extra tick because of +-1/2 tick uncertainty in the 1906 * firing of the timer. The bias will give us exactly the 1907 * 1.5 tick we need. But, because the bias is 1908 * statistical, we have to test that we don't drop below 1909 * the minimum feasible timer (which is 2 ticks). 1910 */ 1911 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1912 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1913 1914 /* 1915 * We received an ack for a packet that wasn't retransmitted; 1916 * it is probably safe to discard any error indications we've 1917 * received recently. This isn't quite right, but close enough 1918 * for now (a route might have failed after we sent a segment, 1919 * and the return path might not be symmetrical). 1920 */ 1921 tp->t_softerror = 0; 1922} 1923 1924/* 1925 * Determine a reasonable value for maxseg size. 1926 * If the route is known, check route for mtu. 1927 * If none, use an mss that can be handled on the outgoing 1928 * interface without forcing IP to fragment; if bigger than 1929 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1930 * to utilize large mbufs. If no route is found, route has no mtu, 1931 * or the destination isn't local, use a default, hopefully conservative 1932 * size (usually 512 or the default IP max size, but no more than the mtu 1933 * of the interface), as we can't discover anything about intervening 1934 * gateways or networks. We also initialize the congestion/slow start 1935 * window to be a single segment if the destination isn't local. 1936 * While looking at the routing entry, we also initialize other path-dependent 1937 * parameters from pre-set or cached values in the routing entry. 1938 * 1939 * Also take into account the space needed for options that we 1940 * send regularly. Make maxseg shorter by that amount to assure 1941 * that we can send maxseg amount of data even when the options 1942 * are present. Store the upper limit of the length of options plus 1943 * data in maxopd. 1944 * 1945 * NOTE that this routine is only called when we process an incoming 1946 * segment, for outgoing segments only tcp_mssopt is called. 1947 * 1948 * In case of T/TCP, we call this routine during implicit connection 1949 * setup as well (offer = -1), to initialize maxseg from the cached 1950 * MSS of our peer. 1951 */ 1952void 1953tcp_mss(tp, offer) 1954 struct tcpcb *tp; 1955 int offer; 1956{ 1957 register struct rtentry *rt; 1958 struct ifnet *ifp; 1959 register int rtt, mss; 1960 u_long bufsize; 1961 struct inpcb *inp; 1962 struct socket *so; 1963 struct rmxp_tao *taop; 1964 int origoffer = offer; 1965 1966 inp = tp->t_inpcb; 1967 if ((rt = tcp_rtlookup(inp)) == NULL) { 1968 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1969 return; 1970 } 1971 ifp = rt->rt_ifp; 1972 so = inp->inp_socket; 1973 1974 taop = rmx_taop(rt->rt_rmx); 1975 /* 1976 * Offer == -1 means that we didn't receive SYN yet, 1977 * use cached value in that case; 1978 */ 1979 if (offer == -1) 1980 offer = taop->tao_mssopt; 1981 /* 1982 * Offer == 0 means that there was no MSS on the SYN segment, 1983 * in this case we use tcp_mssdflt. 1984 */ 1985 if (offer == 0) 1986 offer = tcp_mssdflt; 1987 else 1988 /* 1989 * Sanity check: make sure that maxopd will be large 1990 * enough to allow some data on segments even is the 1991 * all the option space is used (40bytes). Otherwise 1992 * funny things may happen in tcp_output. 1993 */ 1994 offer = max(offer, 64); 1995 taop->tao_mssopt = offer; 1996 1997 /* 1998 * While we're here, check if there's an initial rtt 1999 * or rttvar. Convert from the route-table units 2000 * to scaled multiples of the slow timeout timer. 2001 */ 2002 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2003 /* 2004 * XXX the lock bit for RTT indicates that the value 2005 * is also a minimum value; this is subject to time. 2006 */ 2007 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2008 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2009 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2010 tcpstat.tcps_usedrtt++; 2011 if (rt->rt_rmx.rmx_rttvar) { 2012 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2013 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2014 tcpstat.tcps_usedrttvar++; 2015 } else { 2016 /* default variation is +- 1 rtt */ 2017 tp->t_rttvar = 2018 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2019 } 2020 TCPT_RANGESET(tp->t_rxtcur, 2021 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2022 tp->t_rttmin, TCPTV_REXMTMAX); 2023 } 2024 /* 2025 * if there's an mtu associated with the route, use it 2026 */ 2027 if (rt->rt_rmx.rmx_mtu) 2028 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2029 else 2030 { 2031 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2032 if (!in_localaddr(inp->inp_faddr)) 2033 mss = min(mss, tcp_mssdflt); 2034 } 2035 mss = min(mss, offer); 2036 /* 2037 * maxopd stores the maximum length of data AND options 2038 * in a segment; maxseg is the amount of data in a normal 2039 * segment. We need to store this value (maxopd) apart 2040 * from maxseg, because now every segment carries options 2041 * and thus we normally have somewhat less data in segments. 2042 */ 2043 tp->t_maxopd = mss; 2044 2045 /* 2046 * In case of T/TCP, origoffer==-1 indicates, that no segments 2047 * were received yet. In this case we just guess, otherwise 2048 * we do the same as before T/TCP. 2049 */ 2050 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2051 (origoffer == -1 || 2052 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2053 mss -= TCPOLEN_TSTAMP_APPA; 2054 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2055 (origoffer == -1 || 2056 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2057 mss -= TCPOLEN_CC_APPA; 2058 2059#if (MCLBYTES & (MCLBYTES - 1)) == 0 2060 if (mss > MCLBYTES) 2061 mss &= ~(MCLBYTES-1); 2062#else 2063 if (mss > MCLBYTES) 2064 mss = mss / MCLBYTES * MCLBYTES; 2065#endif 2066 /* 2067 * If there's a pipesize, change the socket buffer 2068 * to that size. Make the socket buffers an integral 2069 * number of mss units; if the mss is larger than 2070 * the socket buffer, decrease the mss. 2071 */ 2072#ifdef RTV_SPIPE 2073 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2074#endif 2075 bufsize = so->so_snd.sb_hiwat; 2076 if (bufsize < mss) 2077 mss = bufsize; 2078 else { 2079 bufsize = roundup(bufsize, mss); 2080 if (bufsize > sb_max) 2081 bufsize = sb_max; 2082 (void)sbreserve(&so->so_snd, bufsize); 2083 } 2084 tp->t_maxseg = mss; 2085 2086#ifdef RTV_RPIPE 2087 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2088#endif 2089 bufsize = so->so_rcv.sb_hiwat; 2090 if (bufsize > mss) { 2091 bufsize = roundup(bufsize, mss); 2092 if (bufsize > sb_max) 2093 bufsize = sb_max; 2094 (void)sbreserve(&so->so_rcv, bufsize); 2095 } 2096 /* 2097 * Don't force slow-start on local network. 2098 */ 2099 if (!in_localaddr(inp->inp_faddr)) 2100 tp->snd_cwnd = mss; 2101 2102 if (rt->rt_rmx.rmx_ssthresh) { 2103 /* 2104 * There's some sort of gateway or interface 2105 * buffer limit on the path. Use this to set 2106 * the slow start threshhold, but set the 2107 * threshold to no less than 2*mss. 2108 */ 2109 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2110 tcpstat.tcps_usedssthresh++; 2111 } 2112} 2113 2114/* 2115 * Determine the MSS option to send on an outgoing SYN. 2116 */ 2117int 2118tcp_mssopt(tp) 2119 struct tcpcb *tp; 2120{ 2121 struct rtentry *rt; 2122 2123 rt = tcp_rtlookup(tp->t_inpcb); 2124 if (rt == NULL) 2125 return tcp_mssdflt; 2126 2127 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2128} 2129#endif /* TUBA_INCLUDE */ 2130