tcp_input.c revision 9818
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 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 * From: @(#)tcp_input.c 8.5 (Berkeley) 4/10/94 34 * $Id: tcp_input.c,v 1.27 1995/07/10 15:39:13 wollman Exp $ 35 */ 36 37#ifndef TUBA_INCLUDE 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/malloc.h> 41#include <sys/mbuf.h> 42#include <sys/protosw.h> 43#include <sys/socket.h> 44#include <sys/socketvar.h> 45#include <sys/errno.h> 46#include <sys/queue.h> 47 48#include <net/if.h> 49#include <net/route.h> 50 51#include <netinet/in.h> 52#include <netinet/in_systm.h> 53#include <netinet/ip.h> 54#include <netinet/in_pcb.h> 55#include <netinet/ip_var.h> 56#include <netinet/tcp.h> 57#include <netinet/tcp_fsm.h> 58#include <netinet/tcp_seq.h> 59#include <netinet/tcp_timer.h> 60#include <netinet/tcp_var.h> 61#include <netinet/tcpip.h> 62#ifdef TCPDEBUG 63#include <netinet/tcp_debug.h> 64struct tcpiphdr tcp_saveti; 65#endif 66 67int tcprexmtthresh = 3; 68tcp_seq tcp_iss; 69tcp_cc tcp_ccgen; 70struct tcpstat tcpstat; 71u_long tcp_now; 72struct inpcbhead tcb; 73struct inpcbinfo tcbinfo; 74 75#endif /* TUBA_INCLUDE */ 76 77/* 78 * Insert segment ti into reassembly queue of tcp with 79 * control block tp. Return TH_FIN if reassembly now includes 80 * a segment with FIN. The macro form does the common case inline 81 * (segment is the next to be received on an established connection, 82 * and the queue is empty), avoiding linkage into and removal 83 * from the queue and repetition of various conversions. 84 * Set DELACK for segments received in order, but ack immediately 85 * when segments are out of order (so fast retransmit can work). 86 */ 87#ifdef TCP_ACK_HACK 88#define TCP_REASS(tp, ti, m, so, flags) { \ 89 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 90 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 91 (tp)->t_state == TCPS_ESTABLISHED) { \ 92 if (ti->ti_flags & TH_PUSH) \ 93 tp->t_flags |= TF_ACKNOW; \ 94 else \ 95 tp->t_flags |= TF_DELACK; \ 96 (tp)->rcv_nxt += (ti)->ti_len; \ 97 flags = (ti)->ti_flags & TH_FIN; \ 98 tcpstat.tcps_rcvpack++;\ 99 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 100 sbappend(&(so)->so_rcv, (m)); \ 101 sorwakeup(so); \ 102 } else { \ 103 (flags) = tcp_reass((tp), (ti), (m)); \ 104 tp->t_flags |= TF_ACKNOW; \ 105 } \ 106} 107#else 108#define TCP_REASS(tp, ti, m, so, flags) { \ 109 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 110 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 111 (tp)->t_state == TCPS_ESTABLISHED) { \ 112 tp->t_flags |= TF_DELACK; \ 113 (tp)->rcv_nxt += (ti)->ti_len; \ 114 flags = (ti)->ti_flags & TH_FIN; \ 115 tcpstat.tcps_rcvpack++;\ 116 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 117 sbappend(&(so)->so_rcv, (m)); \ 118 sorwakeup(so); \ 119 } else { \ 120 (flags) = tcp_reass((tp), (ti), (m)); \ 121 tp->t_flags |= TF_ACKNOW; \ 122 } \ 123} 124#endif 125#ifndef TUBA_INCLUDE 126 127int 128tcp_reass(tp, ti, m) 129 register struct tcpcb *tp; 130 register struct tcpiphdr *ti; 131 struct mbuf *m; 132{ 133 register struct tcpiphdr *q; 134 struct socket *so = tp->t_inpcb->inp_socket; 135 int flags; 136 137 /* 138 * Call with ti==0 after become established to 139 * force pre-ESTABLISHED data up to user socket. 140 */ 141 if (ti == 0) 142 goto present; 143 144 /* 145 * Find a segment which begins after this one does. 146 */ 147 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 148 q = (struct tcpiphdr *)q->ti_next) 149 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 150 break; 151 152 /* 153 * If there is a preceding segment, it may provide some of 154 * our data already. If so, drop the data from the incoming 155 * segment. If it provides all of our data, drop us. 156 */ 157 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 158 register int i; 159 q = (struct tcpiphdr *)q->ti_prev; 160 /* conversion to int (in i) handles seq wraparound */ 161 i = q->ti_seq + q->ti_len - ti->ti_seq; 162 if (i > 0) { 163 if (i >= ti->ti_len) { 164 tcpstat.tcps_rcvduppack++; 165 tcpstat.tcps_rcvdupbyte += ti->ti_len; 166 m_freem(m); 167 /* 168 * Try to present any queued data 169 * at the left window edge to the user. 170 * This is needed after the 3-WHS 171 * completes. 172 */ 173 goto present; /* ??? */ 174 } 175 m_adj(m, i); 176 ti->ti_len -= i; 177 ti->ti_seq += i; 178 } 179 q = (struct tcpiphdr *)(q->ti_next); 180 } 181 tcpstat.tcps_rcvoopack++; 182 tcpstat.tcps_rcvoobyte += ti->ti_len; 183 REASS_MBUF(ti) = m; /* XXX */ 184 185 /* 186 * While we overlap succeeding segments trim them or, 187 * if they are completely covered, dequeue them. 188 */ 189 while (q != (struct tcpiphdr *)tp) { 190 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 191 if (i <= 0) 192 break; 193 if (i < q->ti_len) { 194 q->ti_seq += i; 195 q->ti_len -= i; 196 m_adj(REASS_MBUF(q), i); 197 break; 198 } 199 q = (struct tcpiphdr *)q->ti_next; 200 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 201 remque(q->ti_prev); 202 m_freem(m); 203 } 204 205 /* 206 * Stick new segment in its place. 207 */ 208 insque(ti, q->ti_prev); 209 210present: 211 /* 212 * Present data to user, advancing rcv_nxt through 213 * completed sequence space. 214 */ 215 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 216 return (0); 217 ti = tp->seg_next; 218 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 219 return (0); 220 do { 221 tp->rcv_nxt += ti->ti_len; 222 flags = ti->ti_flags & TH_FIN; 223 remque(ti); 224 m = REASS_MBUF(ti); 225 ti = (struct tcpiphdr *)ti->ti_next; 226 if (so->so_state & SS_CANTRCVMORE) 227 m_freem(m); 228 else 229 sbappend(&so->so_rcv, m); 230 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 231 sorwakeup(so); 232 return (flags); 233} 234 235/* 236 * TCP input routine, follows pages 65-76 of the 237 * protocol specification dated September, 1981 very closely. 238 */ 239void 240tcp_input(m, iphlen) 241 register struct mbuf *m; 242 int iphlen; 243{ 244 register struct tcpiphdr *ti; 245 register struct inpcb *inp; 246 caddr_t optp = NULL; 247 int optlen = 0; 248 int len, tlen, off; 249 register struct tcpcb *tp = 0; 250 register int tiflags; 251 struct socket *so = 0; 252 int todrop, acked, ourfinisacked, needoutput = 0; 253 struct in_addr laddr; 254 int dropsocket = 0; 255 int iss = 0; 256 u_long tiwin; 257 struct tcpopt to; /* options in this segment */ 258 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 259 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 260#ifdef TCPDEBUG 261 short ostate = 0; 262#endif 263 264 bzero((char *)&to, sizeof(to)); 265 266 tcpstat.tcps_rcvtotal++; 267 /* 268 * Get IP and TCP header together in first mbuf. 269 * Note: IP leaves IP header in first mbuf. 270 */ 271 ti = mtod(m, struct tcpiphdr *); 272 if (iphlen > sizeof (struct ip)) 273 ip_stripoptions(m, (struct mbuf *)0); 274 if (m->m_len < sizeof (struct tcpiphdr)) { 275 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 276 tcpstat.tcps_rcvshort++; 277 return; 278 } 279 ti = mtod(m, struct tcpiphdr *); 280 } 281 282 /* 283 * Checksum extended TCP header and data. 284 */ 285 tlen = ((struct ip *)ti)->ip_len; 286 len = sizeof (struct ip) + tlen; 287 ti->ti_next = ti->ti_prev = 0; 288 ti->ti_x1 = 0; 289 ti->ti_len = (u_short)tlen; 290 HTONS(ti->ti_len); 291 ti->ti_sum = in_cksum(m, len); 292 if (ti->ti_sum) { 293 tcpstat.tcps_rcvbadsum++; 294 goto drop; 295 } 296#endif /* TUBA_INCLUDE */ 297 298 /* 299 * Check that TCP offset makes sense, 300 * pull out TCP options and adjust length. XXX 301 */ 302 off = ti->ti_off << 2; 303 if (off < sizeof (struct tcphdr) || off > tlen) { 304 tcpstat.tcps_rcvbadoff++; 305 goto drop; 306 } 307 tlen -= off; 308 ti->ti_len = tlen; 309 if (off > sizeof (struct tcphdr)) { 310 if (m->m_len < sizeof(struct ip) + off) { 311 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 312 tcpstat.tcps_rcvshort++; 313 return; 314 } 315 ti = mtod(m, struct tcpiphdr *); 316 } 317 optlen = off - sizeof (struct tcphdr); 318 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 319 /* 320 * Do quick retrieval of timestamp options ("options 321 * prediction?"). If timestamp is the only option and it's 322 * formatted as recommended in RFC 1323 appendix A, we 323 * quickly get the values now and not bother calling 324 * tcp_dooptions(), etc. 325 */ 326 if ((optlen == TCPOLEN_TSTAMP_APPA || 327 (optlen > TCPOLEN_TSTAMP_APPA && 328 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 329 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 330 (ti->ti_flags & TH_SYN) == 0) { 331 to.to_flag |= TOF_TS; 332 to.to_tsval = ntohl(*(u_long *)(optp + 4)); 333 to.to_tsecr = ntohl(*(u_long *)(optp + 8)); 334 optp = NULL; /* we've parsed the options */ 335 } 336 } 337 tiflags = ti->ti_flags; 338 339 /* 340 * Convert TCP protocol specific fields to host format. 341 */ 342 NTOHL(ti->ti_seq); 343 NTOHL(ti->ti_ack); 344 NTOHS(ti->ti_win); 345 NTOHS(ti->ti_urp); 346 347 /* 348 * Drop TCP, IP headers and TCP options. 349 */ 350 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 351 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 352 353 /* 354 * Locate pcb for segment. 355 */ 356findpcb: 357 /* 358 * First look for an exact match. 359 */ 360 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 361 ti->ti_dst, ti->ti_dport); 362 /* 363 * ...and if that fails, do a wildcard search. 364 */ 365 if (inp == NULL) { 366 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 367 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 368 } 369 370 /* 371 * If the state is CLOSED (i.e., TCB does not exist) then 372 * all data in the incoming segment is discarded. 373 * If the TCB exists but is in CLOSED state, it is embryonic, 374 * but should either do a listen or a connect soon. 375 */ 376 if (inp == NULL) 377 goto dropwithreset; 378 tp = intotcpcb(inp); 379 if (tp == 0) 380 goto dropwithreset; 381 if (tp->t_state == TCPS_CLOSED) 382 goto drop; 383 384 /* Unscale the window into a 32-bit value. */ 385 if ((tiflags & TH_SYN) == 0) 386 tiwin = ti->ti_win << tp->snd_scale; 387 else 388 tiwin = ti->ti_win; 389 390 so = inp->inp_socket; 391 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 392#ifdef TCPDEBUG 393 if (so->so_options & SO_DEBUG) { 394 ostate = tp->t_state; 395 tcp_saveti = *ti; 396 } 397#endif 398 if (so->so_options & SO_ACCEPTCONN) { 399 register struct tcpcb *tp0 = tp; 400 so = sonewconn(so, 0); 401 if (so == 0) 402 goto drop; 403 /* 404 * This is ugly, but .... 405 * 406 * Mark socket as temporary until we're 407 * committed to keeping it. The code at 408 * ``drop'' and ``dropwithreset'' check the 409 * flag dropsocket to see if the temporary 410 * socket created here should be discarded. 411 * We mark the socket as discardable until 412 * we're committed to it below in TCPS_LISTEN. 413 */ 414 dropsocket++; 415 inp = (struct inpcb *)so->so_pcb; 416 inp->inp_laddr = ti->ti_dst; 417 inp->inp_lport = ti->ti_dport; 418 in_pcbrehash(inp); 419#if BSD>=43 420 inp->inp_options = ip_srcroute(); 421#endif 422 tp = intotcpcb(inp); 423 tp->t_state = TCPS_LISTEN; 424 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 425 426 /* Compute proper scaling value from buffer space */ 427 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 428 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 429 tp->request_r_scale++; 430 } 431 } 432 433 /* 434 * Segment received on connection. 435 * Reset idle time and keep-alive timer. 436 */ 437 tp->t_idle = 0; 438 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 439 440 /* 441 * Process options if not in LISTEN state, 442 * else do it below (after getting remote address). 443 */ 444 if (optp && tp->t_state != TCPS_LISTEN) 445 tcp_dooptions(tp, optp, optlen, ti, 446 &to); 447 448 /* 449 * Header prediction: check for the two common cases 450 * of a uni-directional data xfer. If the packet has 451 * no control flags, is in-sequence, the window didn't 452 * change and we're not retransmitting, it's a 453 * candidate. If the length is zero and the ack moved 454 * forward, we're the sender side of the xfer. Just 455 * free the data acked & wake any higher level process 456 * that was blocked waiting for space. If the length 457 * is non-zero and the ack didn't move, we're the 458 * receiver side. If we're getting packets in-order 459 * (the reassembly queue is empty), add the data to 460 * the socket buffer and note that we need a delayed ack. 461 * Make sure that the hidden state-flags are also off. 462 * Since we check for TCPS_ESTABLISHED above, it can only 463 * be TH_NEEDSYN. 464 */ 465 if (tp->t_state == TCPS_ESTABLISHED && 466 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 467 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 468 ((to.to_flag & TOF_TS) == 0 || 469 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 470 /* 471 * Using the CC option is compulsory if once started: 472 * the segment is OK if no T/TCP was negotiated or 473 * if the segment has a CC option equal to CCrecv 474 */ 475 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 476 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 477 ti->ti_seq == tp->rcv_nxt && 478 tiwin && tiwin == tp->snd_wnd && 479 tp->snd_nxt == tp->snd_max) { 480 481 /* 482 * If last ACK falls within this segment's sequence numbers, 483 * record the timestamp. 484 * NOTE that the test is modified according to the latest 485 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 486 */ 487 if ((to.to_flag & TOF_TS) != 0 && 488 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 489 tp->ts_recent_age = tcp_now; 490 tp->ts_recent = to.to_tsval; 491 } 492 493 if (ti->ti_len == 0) { 494 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 495 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 496 tp->snd_cwnd >= tp->snd_wnd) { 497 /* 498 * this is a pure ack for outstanding data. 499 */ 500 ++tcpstat.tcps_predack; 501 if ((to.to_flag & TOF_TS) != 0) 502 tcp_xmit_timer(tp, 503 tcp_now - to.to_tsecr + 1); 504 else if (tp->t_rtt && 505 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 506 tcp_xmit_timer(tp, tp->t_rtt); 507 acked = ti->ti_ack - tp->snd_una; 508 tcpstat.tcps_rcvackpack++; 509 tcpstat.tcps_rcvackbyte += acked; 510 sbdrop(&so->so_snd, acked); 511 tp->snd_una = ti->ti_ack; 512 m_freem(m); 513 514 /* 515 * If all outstanding data are acked, stop 516 * retransmit timer, otherwise restart timer 517 * using current (possibly backed-off) value. 518 * If process is waiting for space, 519 * wakeup/selwakeup/signal. If data 520 * are ready to send, let tcp_output 521 * decide between more output or persist. 522 */ 523 if (tp->snd_una == tp->snd_max) 524 tp->t_timer[TCPT_REXMT] = 0; 525 else if (tp->t_timer[TCPT_PERSIST] == 0) 526 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 527 528 if (so->so_snd.sb_flags & SB_NOTIFY) 529 sowwakeup(so); 530 if (so->so_snd.sb_cc) 531 (void) tcp_output(tp); 532 return; 533 } 534 } else if (ti->ti_ack == tp->snd_una && 535 tp->seg_next == (struct tcpiphdr *)tp && 536 ti->ti_len <= sbspace(&so->so_rcv)) { 537 /* 538 * this is a pure, in-sequence data packet 539 * with nothing on the reassembly queue and 540 * we have enough buffer space to take it. 541 */ 542 ++tcpstat.tcps_preddat; 543 tp->rcv_nxt += ti->ti_len; 544 tcpstat.tcps_rcvpack++; 545 tcpstat.tcps_rcvbyte += ti->ti_len; 546 /* 547 * Add data to socket buffer. 548 */ 549 sbappend(&so->so_rcv, m); 550 sorwakeup(so); 551#ifdef TCP_ACK_HACK 552 /* 553 * If this is a short packet, then ACK now - with Nagel 554 * congestion avoidance sender won't send more until 555 * he gets an ACK. 556 */ 557 if (tiflags & TH_PUSH) { 558 tp->t_flags |= TF_ACKNOW; 559 tcp_output(tp); 560 } else { 561 tp->t_flags |= TF_DELACK; 562 } 563#else 564 tp->t_flags |= TF_DELACK; 565#endif 566 return; 567 } 568 } 569 570 /* 571 * Calculate amount of space in receive window, 572 * and then do TCP input processing. 573 * Receive window is amount of space in rcv queue, 574 * but not less than advertised window. 575 */ 576 { int win; 577 578 win = sbspace(&so->so_rcv); 579 if (win < 0) 580 win = 0; 581 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 582 } 583 584 switch (tp->t_state) { 585 586 /* 587 * If the state is LISTEN then ignore segment if it contains an RST. 588 * If the segment contains an ACK then it is bad and send a RST. 589 * If it does not contain a SYN then it is not interesting; drop it. 590 * Don't bother responding if the destination was a broadcast. 591 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 592 * tp->iss, and send a segment: 593 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 594 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 595 * Fill in remote peer address fields if not previously specified. 596 * Enter SYN_RECEIVED state, and process any other fields of this 597 * segment in this state. 598 */ 599 case TCPS_LISTEN: { 600 struct mbuf *am; 601 register struct sockaddr_in *sin; 602 603 if (tiflags & TH_RST) 604 goto drop; 605 if (tiflags & TH_ACK) 606 goto dropwithreset; 607 if ((tiflags & TH_SYN) == 0) 608 goto drop; 609 /* 610 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 611 * in_broadcast() should never return true on a received 612 * packet with M_BCAST not set. 613 */ 614 if (m->m_flags & (M_BCAST|M_MCAST) || 615 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 616 goto drop; 617 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 618 if (am == NULL) 619 goto drop; 620 am->m_len = sizeof (struct sockaddr_in); 621 sin = mtod(am, struct sockaddr_in *); 622 sin->sin_family = AF_INET; 623 sin->sin_len = sizeof(*sin); 624 sin->sin_addr = ti->ti_src; 625 sin->sin_port = ti->ti_sport; 626 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 627 laddr = inp->inp_laddr; 628 if (inp->inp_laddr.s_addr == INADDR_ANY) 629 inp->inp_laddr = ti->ti_dst; 630 if (in_pcbconnect(inp, am)) { 631 inp->inp_laddr = laddr; 632 (void) m_free(am); 633 goto drop; 634 } 635 (void) m_free(am); 636 tp->t_template = tcp_template(tp); 637 if (tp->t_template == 0) { 638 tp = tcp_drop(tp, ENOBUFS); 639 dropsocket = 0; /* socket is already gone */ 640 goto drop; 641 } 642 if ((taop = tcp_gettaocache(inp)) == NULL) { 643 taop = &tao_noncached; 644 bzero(taop, sizeof(*taop)); 645 } 646 if (optp) 647 tcp_dooptions(tp, optp, optlen, ti, 648 &to); 649 if (iss) 650 tp->iss = iss; 651 else 652 tp->iss = tcp_iss; 653 tcp_iss += TCP_ISSINCR/2; 654 tp->irs = ti->ti_seq; 655 tcp_sendseqinit(tp); 656 tcp_rcvseqinit(tp); 657 /* 658 * Initialization of the tcpcb for transaction; 659 * set SND.WND = SEG.WND, 660 * initialize CCsend and CCrecv. 661 */ 662 tp->snd_wnd = tiwin; /* initial send-window */ 663 tp->cc_send = CC_INC(tcp_ccgen); 664 tp->cc_recv = to.to_cc; 665 /* 666 * Perform TAO test on incoming CC (SEG.CC) option, if any. 667 * - compare SEG.CC against cached CC from the same host, 668 * if any. 669 * - if SEG.CC > chached value, SYN must be new and is accepted 670 * immediately: save new CC in the cache, mark the socket 671 * connected, enter ESTABLISHED state, turn on flag to 672 * send a SYN in the next segment. 673 * A virtual advertised window is set in rcv_adv to 674 * initialize SWS prevention. Then enter normal segment 675 * processing: drop SYN, process data and FIN. 676 * - otherwise do a normal 3-way handshake. 677 */ 678 if ((to.to_flag & TOF_CC) != 0) { 679 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 680 taop->tao_cc = to.to_cc; 681 tp->t_state = TCPS_ESTABLISHED; 682 683 /* 684 * If there is a FIN, or if there is data and the 685 * connection is local, then delay SYN,ACK(SYN) in 686 * the hope of piggy-backing it on a response 687 * segment. Otherwise must send ACK now in case 688 * the other side is slow starting. 689 */ 690 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 691 in_localaddr(inp->inp_faddr))) 692 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 693 else 694 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 695 tp->rcv_adv += tp->rcv_wnd; 696 tcpstat.tcps_connects++; 697 soisconnected(so); 698 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 699 dropsocket = 0; /* committed to socket */ 700 tcpstat.tcps_accepts++; 701 goto trimthenstep6; 702 } 703 /* else do standard 3-way handshake */ 704 } else { 705 /* 706 * No CC option, but maybe CC.NEW: 707 * invalidate cached value. 708 */ 709 taop->tao_cc = 0; 710 } 711 /* 712 * TAO test failed or there was no CC option, 713 * do a standard 3-way handshake. 714 */ 715 tp->t_flags |= TF_ACKNOW; 716 tp->t_state = TCPS_SYN_RECEIVED; 717 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 718 dropsocket = 0; /* committed to socket */ 719 tcpstat.tcps_accepts++; 720 goto trimthenstep6; 721 } 722 723 /* 724 * If the state is SYN_SENT: 725 * if seg contains an ACK, but not for our SYN, drop the input. 726 * if seg contains a RST, then drop the connection. 727 * if seg does not contain SYN, then drop it. 728 * Otherwise this is an acceptable SYN segment 729 * initialize tp->rcv_nxt and tp->irs 730 * if seg contains ack then advance tp->snd_una 731 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 732 * arrange for segment to be acked (eventually) 733 * continue processing rest of data/controls, beginning with URG 734 */ 735 case TCPS_SYN_SENT: 736 if ((taop = tcp_gettaocache(inp)) == NULL) { 737 taop = &tao_noncached; 738 bzero(taop, sizeof(*taop)); 739 } 740 741 if ((tiflags & TH_ACK) && 742 (SEQ_LEQ(ti->ti_ack, tp->iss) || 743 SEQ_GT(ti->ti_ack, tp->snd_max))) { 744 /* 745 * If we have a cached CCsent for the remote host, 746 * hence we haven't just crashed and restarted, 747 * do not send a RST. This may be a retransmission 748 * from the other side after our earlier ACK was lost. 749 * Our new SYN, when it arrives, will serve as the 750 * needed ACK. 751 */ 752 if (taop->tao_ccsent != 0) 753 goto drop; 754 else 755 goto dropwithreset; 756 } 757 if (tiflags & TH_RST) { 758 if (tiflags & TH_ACK) 759 tp = tcp_drop(tp, ECONNREFUSED); 760 goto drop; 761 } 762 if ((tiflags & TH_SYN) == 0) 763 goto drop; 764 tp->snd_wnd = ti->ti_win; /* initial send window */ 765 tp->cc_recv = to.to_cc; /* foreign CC */ 766 767 tp->irs = ti->ti_seq; 768 tcp_rcvseqinit(tp); 769 if (tiflags & TH_ACK && SEQ_GT(ti->ti_ack, tp->iss)) { 770 tcpstat.tcps_connects++; 771 soisconnected(so); 772 /* Do window scaling on this connection? */ 773 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 774 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 775 tp->snd_scale = tp->requested_s_scale; 776 tp->rcv_scale = tp->request_r_scale; 777 } 778 /* 779 * Our SYN was acked. If segment contains CC.ECHO 780 * option, check it to make sure this segment really 781 * matches our SYN. If not, just drop it as old 782 * duplicate, but send an RST if we're still playing 783 * by the old rules. 784 */ 785 if ((to.to_flag & TOF_CCECHO) && 786 tp->cc_send != to.to_ccecho) { 787 if (taop->tao_ccsent != 0) 788 goto drop; 789 else 790 goto dropwithreset; 791 } 792 /* Segment is acceptable, update cache if undefined. */ 793 if (taop->tao_ccsent == 0) 794 taop->tao_ccsent = to.to_ccecho; 795 796 tp->rcv_adv += tp->rcv_wnd; 797 tp->snd_una++; /* SYN is acked */ 798 /* 799 * If there's data, delay ACK; if there's also a FIN 800 * ACKNOW will be turned on later. 801 */ 802 if (ti->ti_len != 0) 803 tp->t_flags |= TF_DELACK; 804 else 805 tp->t_flags |= TF_ACKNOW; 806 /* 807 * Received <SYN,ACK> in SYN_SENT[*] state. 808 * Transitions: 809 * SYN_SENT --> ESTABLISHED 810 * SYN_SENT* --> FIN_WAIT_1 811 */ 812 if (tp->t_flags & TF_NEEDFIN) { 813 tp->t_state = TCPS_FIN_WAIT_1; 814 tp->t_flags &= ~TF_NEEDFIN; 815 tiflags &= ~TH_SYN; 816 } else 817 tp->t_state = TCPS_ESTABLISHED; 818 819 } else { 820 /* 821 * Received initial SYN in SYN-SENT[*] state => simul- 822 * taneous open. If segment contains CC option and there is 823 * a cached CC, apply TAO test; if it succeeds, connection is 824 * half-synchronized. Otherwise, do 3-way handshake: 825 * SYN-SENT -> SYN-RECEIVED 826 * SYN-SENT* -> SYN-RECEIVED* 827 * If there was no CC option, clear cached CC value. 828 */ 829 tp->t_flags |= TF_ACKNOW; 830 tp->t_timer[TCPT_REXMT] = 0; 831 if (to.to_flag & TOF_CC) { 832 if (taop->tao_cc != 0 && 833 CC_GT(to.to_cc, taop->tao_cc)) { 834 /* 835 * update cache and make transition: 836 * SYN-SENT -> ESTABLISHED* 837 * SYN-SENT* -> FIN-WAIT-1* 838 */ 839 taop->tao_cc = to.to_cc; 840 if (tp->t_flags & TF_NEEDFIN) { 841 tp->t_state = TCPS_FIN_WAIT_1; 842 tp->t_flags &= ~TF_NEEDFIN; 843 } else 844 tp->t_state = TCPS_ESTABLISHED; 845 tp->t_flags |= TF_NEEDSYN; 846 } else 847 tp->t_state = TCPS_SYN_RECEIVED; 848 } else { 849 /* CC.NEW or no option => invalidate cache */ 850 taop->tao_cc = 0; 851 tp->t_state = TCPS_SYN_RECEIVED; 852 } 853 } 854 855trimthenstep6: 856 /* 857 * Advance ti->ti_seq to correspond to first data byte. 858 * If data, trim to stay within window, 859 * dropping FIN if necessary. 860 */ 861 ti->ti_seq++; 862 if (ti->ti_len > tp->rcv_wnd) { 863 todrop = ti->ti_len - tp->rcv_wnd; 864 m_adj(m, -todrop); 865 ti->ti_len = tp->rcv_wnd; 866 tiflags &= ~TH_FIN; 867 tcpstat.tcps_rcvpackafterwin++; 868 tcpstat.tcps_rcvbyteafterwin += todrop; 869 } 870 tp->snd_wl1 = ti->ti_seq - 1; 871 tp->rcv_up = ti->ti_seq; 872 /* 873 * Client side of transaction: already sent SYN and data. 874 * If the remote host used T/TCP to validate the SYN, 875 * our data will be ACK'd; if so, enter normal data segment 876 * processing in the middle of step 5, ack processing. 877 * Otherwise, goto step 6. 878 */ 879 if (tiflags & TH_ACK) 880 goto process_ACK; 881 goto step6; 882 /* 883 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 884 * if segment contains a SYN and CC [not CC.NEW] option: 885 * if state == TIME_WAIT and connection duration > MSL, 886 * drop packet and send RST; 887 * 888 * if SEG.CC > CCrecv then is new SYN, and can implicitly 889 * ack the FIN (and data) in retransmission queue. 890 * Complete close and delete TCPCB. Then reprocess 891 * segment, hoping to find new TCPCB in LISTEN state; 892 * 893 * else must be old SYN; drop it. 894 * else do normal processing. 895 */ 896 case TCPS_LAST_ACK: 897 case TCPS_CLOSING: 898 case TCPS_TIME_WAIT: 899 if ((tiflags & TH_SYN) && 900 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 901 if (tp->t_state == TCPS_TIME_WAIT && 902 tp->t_duration > TCPTV_MSL) 903 goto dropwithreset; 904 if (CC_GT(to.to_cc, tp->cc_recv)) { 905 tp = tcp_close(tp); 906 goto findpcb; 907 } 908 else 909 goto drop; 910 } 911 break; /* continue normal processing */ 912 } 913 914 /* 915 * States other than LISTEN or SYN_SENT. 916 * First check timestamp, if present. 917 * Then check the connection count, if present. 918 * Then check that at least some bytes of segment are within 919 * receive window. If segment begins before rcv_nxt, 920 * drop leading data (and SYN); if nothing left, just ack. 921 * 922 * RFC 1323 PAWS: If we have a timestamp reply on this segment 923 * and it's less than ts_recent, drop it. 924 */ 925 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 926 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 927 928 /* Check to see if ts_recent is over 24 days old. */ 929 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 930 /* 931 * Invalidate ts_recent. If this segment updates 932 * ts_recent, the age will be reset later and ts_recent 933 * will get a valid value. If it does not, setting 934 * ts_recent to zero will at least satisfy the 935 * requirement that zero be placed in the timestamp 936 * echo reply when ts_recent isn't valid. The 937 * age isn't reset until we get a valid ts_recent 938 * because we don't want out-of-order segments to be 939 * dropped when ts_recent is old. 940 */ 941 tp->ts_recent = 0; 942 } else { 943 tcpstat.tcps_rcvduppack++; 944 tcpstat.tcps_rcvdupbyte += ti->ti_len; 945 tcpstat.tcps_pawsdrop++; 946 goto dropafterack; 947 } 948 } 949 950 /* 951 * T/TCP mechanism 952 * If T/TCP was negotiated and the segment doesn't have CC, 953 * or if it's CC is wrong then drop the segment. 954 * RST segments do not have to comply with this. 955 */ 956 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 957 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 958 (tiflags & TH_RST) == 0) 959 goto dropafterack; 960 961 todrop = tp->rcv_nxt - ti->ti_seq; 962 if (todrop > 0) { 963 if (tiflags & TH_SYN) { 964 tiflags &= ~TH_SYN; 965 ti->ti_seq++; 966 if (ti->ti_urp > 1) 967 ti->ti_urp--; 968 else 969 tiflags &= ~TH_URG; 970 todrop--; 971 } 972 /* 973 * Following if statement from Stevens, vol. 2, p. 960. 974 */ 975 if (todrop > ti->ti_len 976 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 977 /* 978 * Any valid FIN must be to the left of the window. 979 * At this point the FIN must be a duplicate or out 980 * of sequence; drop it. 981 */ 982 tiflags &= ~TH_FIN; 983 984 /* 985 * Send an ACK to resynchronize and drop any data. 986 * But keep on processing for RST or ACK. 987 */ 988 tp->t_flags |= TF_ACKNOW; 989 todrop = ti->ti_len; 990 tcpstat.tcps_rcvduppack++; 991 tcpstat.tcps_rcvdupbyte += todrop; 992 } else { 993 tcpstat.tcps_rcvpartduppack++; 994 tcpstat.tcps_rcvpartdupbyte += todrop; 995 } 996 m_adj(m, todrop); 997 ti->ti_seq += todrop; 998 ti->ti_len -= todrop; 999 if (ti->ti_urp > todrop) 1000 ti->ti_urp -= todrop; 1001 else { 1002 tiflags &= ~TH_URG; 1003 ti->ti_urp = 0; 1004 } 1005 } 1006 1007 /* 1008 * If new data are received on a connection after the 1009 * user processes are gone, then RST the other end. 1010 */ 1011 if ((so->so_state & SS_NOFDREF) && 1012 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1013 tp = tcp_close(tp); 1014 tcpstat.tcps_rcvafterclose++; 1015 goto dropwithreset; 1016 } 1017 1018 /* 1019 * If segment ends after window, drop trailing data 1020 * (and PUSH and FIN); if nothing left, just ACK. 1021 */ 1022 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1023 if (todrop > 0) { 1024 tcpstat.tcps_rcvpackafterwin++; 1025 if (todrop >= ti->ti_len) { 1026 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1027 /* 1028 * If a new connection request is received 1029 * while in TIME_WAIT, drop the old connection 1030 * and start over if the sequence numbers 1031 * are above the previous ones. 1032 */ 1033 if (tiflags & TH_SYN && 1034 tp->t_state == TCPS_TIME_WAIT && 1035 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1036 iss = tp->rcv_nxt + TCP_ISSINCR; 1037 tp = tcp_close(tp); 1038 goto findpcb; 1039 } 1040 /* 1041 * If window is closed can only take segments at 1042 * window edge, and have to drop data and PUSH from 1043 * incoming segments. Continue processing, but 1044 * remember to ack. Otherwise, drop segment 1045 * and ack. 1046 */ 1047 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1048 tp->t_flags |= TF_ACKNOW; 1049 tcpstat.tcps_rcvwinprobe++; 1050 } else 1051 goto dropafterack; 1052 } else 1053 tcpstat.tcps_rcvbyteafterwin += todrop; 1054 m_adj(m, -todrop); 1055 ti->ti_len -= todrop; 1056 tiflags &= ~(TH_PUSH|TH_FIN); 1057 } 1058 1059 /* 1060 * If last ACK falls within this segment's sequence numbers, 1061 * record its timestamp. 1062 * NOTE that the test is modified according to the latest 1063 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1064 */ 1065 if ((to.to_flag & TOF_TS) != 0 && 1066 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1067 tp->ts_recent_age = tcp_now; 1068 tp->ts_recent = to.to_tsval; 1069 } 1070 1071 /* 1072 * If the RST bit is set examine the state: 1073 * SYN_RECEIVED STATE: 1074 * If passive open, return to LISTEN state. 1075 * If active open, inform user that connection was refused. 1076 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1077 * Inform user that connection was reset, and close tcb. 1078 * CLOSING, LAST_ACK, TIME_WAIT STATES 1079 * Close the tcb. 1080 */ 1081 if (tiflags&TH_RST) switch (tp->t_state) { 1082 1083 case TCPS_SYN_RECEIVED: 1084 so->so_error = ECONNREFUSED; 1085 goto close; 1086 1087 case TCPS_ESTABLISHED: 1088 case TCPS_FIN_WAIT_1: 1089 case TCPS_FIN_WAIT_2: 1090 case TCPS_CLOSE_WAIT: 1091 so->so_error = ECONNRESET; 1092 close: 1093 tp->t_state = TCPS_CLOSED; 1094 tcpstat.tcps_drops++; 1095 tp = tcp_close(tp); 1096 goto drop; 1097 1098 case TCPS_CLOSING: 1099 case TCPS_LAST_ACK: 1100 case TCPS_TIME_WAIT: 1101 tp = tcp_close(tp); 1102 goto drop; 1103 } 1104 1105 /* 1106 * If a SYN is in the window, then this is an 1107 * error and we send an RST and drop the connection. 1108 */ 1109 if (tiflags & TH_SYN) { 1110 tp = tcp_drop(tp, ECONNRESET); 1111 goto dropwithreset; 1112 } 1113 1114 /* 1115 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1116 * flag is on (half-synchronized state), then queue data for 1117 * later processing; else drop segment and return. 1118 */ 1119 if ((tiflags & TH_ACK) == 0) { 1120 if (tp->t_state == TCPS_SYN_RECEIVED || 1121 (tp->t_flags & TF_NEEDSYN)) 1122 goto step6; 1123 else 1124 goto drop; 1125 } 1126 1127 /* 1128 * Ack processing. 1129 */ 1130 switch (tp->t_state) { 1131 1132 /* 1133 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1134 * ESTABLISHED state and continue processing, otherwise 1135 * send an RST. 1136 */ 1137 case TCPS_SYN_RECEIVED: 1138 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1139 SEQ_GT(ti->ti_ack, tp->snd_max)) 1140 goto dropwithreset; 1141 1142 tcpstat.tcps_connects++; 1143 soisconnected(so); 1144 /* Do window scaling? */ 1145 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1146 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1147 tp->snd_scale = tp->requested_s_scale; 1148 tp->rcv_scale = tp->request_r_scale; 1149 } 1150 /* 1151 * Upon successful completion of 3-way handshake, 1152 * update cache.CC if it was undefined, pass any queued 1153 * data to the user, and advance state appropriately. 1154 */ 1155 if ((taop = tcp_gettaocache(inp)) != NULL && 1156 taop->tao_cc == 0) 1157 taop->tao_cc = tp->cc_recv; 1158 1159 /* 1160 * Make transitions: 1161 * SYN-RECEIVED -> ESTABLISHED 1162 * SYN-RECEIVED* -> FIN-WAIT-1 1163 */ 1164 if (tp->t_flags & TF_NEEDFIN) { 1165 tp->t_state = TCPS_FIN_WAIT_1; 1166 tp->t_flags &= ~TF_NEEDFIN; 1167 } else 1168 tp->t_state = TCPS_ESTABLISHED; 1169 /* 1170 * If segment contains data or ACK, will call tcp_reass() 1171 * later; if not, do so now to pass queued data to user. 1172 */ 1173 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1174 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1175 (struct mbuf *)0); 1176 tp->snd_wl1 = ti->ti_seq - 1; 1177 /* fall into ... */ 1178 1179 /* 1180 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1181 * ACKs. If the ack is in the range 1182 * tp->snd_una < ti->ti_ack <= tp->snd_max 1183 * then advance tp->snd_una to ti->ti_ack and drop 1184 * data from the retransmission queue. If this ACK reflects 1185 * more up to date window information we update our window information. 1186 */ 1187 case TCPS_ESTABLISHED: 1188 case TCPS_FIN_WAIT_1: 1189 case TCPS_FIN_WAIT_2: 1190 case TCPS_CLOSE_WAIT: 1191 case TCPS_CLOSING: 1192 case TCPS_LAST_ACK: 1193 case TCPS_TIME_WAIT: 1194 1195 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1196 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1197 tcpstat.tcps_rcvdupack++; 1198 /* 1199 * If we have outstanding data (other than 1200 * a window probe), this is a completely 1201 * duplicate ack (ie, window info didn't 1202 * change), the ack is the biggest we've 1203 * seen and we've seen exactly our rexmt 1204 * threshhold of them, assume a packet 1205 * has been dropped and retransmit it. 1206 * Kludge snd_nxt & the congestion 1207 * window so we send only this one 1208 * packet. 1209 * 1210 * We know we're losing at the current 1211 * window size so do congestion avoidance 1212 * (set ssthresh to half the current window 1213 * and pull our congestion window back to 1214 * the new ssthresh). 1215 * 1216 * Dup acks mean that packets have left the 1217 * network (they're now cached at the receiver) 1218 * so bump cwnd by the amount in the receiver 1219 * to keep a constant cwnd packets in the 1220 * network. 1221 */ 1222 if (tp->t_timer[TCPT_REXMT] == 0 || 1223 ti->ti_ack != tp->snd_una) 1224 tp->t_dupacks = 0; 1225 else if (++tp->t_dupacks == tcprexmtthresh) { 1226 tcp_seq onxt = tp->snd_nxt; 1227 u_int win = 1228 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1229 tp->t_maxseg; 1230 1231 if (win < 2) 1232 win = 2; 1233 tp->snd_ssthresh = win * tp->t_maxseg; 1234 tp->t_timer[TCPT_REXMT] = 0; 1235 tp->t_rtt = 0; 1236 tp->snd_nxt = ti->ti_ack; 1237 tp->snd_cwnd = tp->t_maxseg; 1238 (void) tcp_output(tp); 1239 tp->snd_cwnd = tp->snd_ssthresh + 1240 tp->t_maxseg * tp->t_dupacks; 1241 if (SEQ_GT(onxt, tp->snd_nxt)) 1242 tp->snd_nxt = onxt; 1243 goto drop; 1244 } else if (tp->t_dupacks > tcprexmtthresh) { 1245 tp->snd_cwnd += tp->t_maxseg; 1246 (void) tcp_output(tp); 1247 goto drop; 1248 } 1249 } else 1250 tp->t_dupacks = 0; 1251 break; 1252 } 1253 /* 1254 * If the congestion window was inflated to account 1255 * for the other side's cached packets, retract it. 1256 */ 1257 if (tp->t_dupacks > tcprexmtthresh && 1258 tp->snd_cwnd > tp->snd_ssthresh) 1259 tp->snd_cwnd = tp->snd_ssthresh; 1260 tp->t_dupacks = 0; 1261 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1262 tcpstat.tcps_rcvacktoomuch++; 1263 goto dropafterack; 1264 } 1265 /* 1266 * If we reach this point, ACK is not a duplicate, 1267 * i.e., it ACKs something we sent. 1268 */ 1269 if (tp->t_flags & TF_NEEDSYN) { 1270 /* 1271 * T/TCP: Connection was half-synchronized, and our 1272 * SYN has been ACK'd (so connection is now fully 1273 * synchronized). Go to non-starred state and 1274 * increment snd_una for ACK of SYN. 1275 */ 1276 tp->t_flags &= ~TF_NEEDSYN; 1277 tp->snd_una++; 1278 } 1279 1280process_ACK: 1281 acked = ti->ti_ack - tp->snd_una; 1282 tcpstat.tcps_rcvackpack++; 1283 tcpstat.tcps_rcvackbyte += acked; 1284 1285 /* 1286 * If we have a timestamp reply, update smoothed 1287 * round trip time. If no timestamp is present but 1288 * transmit timer is running and timed sequence 1289 * number was acked, update smoothed round trip time. 1290 * Since we now have an rtt measurement, cancel the 1291 * timer backoff (cf., Phil Karn's retransmit alg.). 1292 * Recompute the initial retransmit timer. 1293 */ 1294 if (to.to_flag & TOF_TS) 1295 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1296 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1297 tcp_xmit_timer(tp,tp->t_rtt); 1298 1299 /* 1300 * If all outstanding data is acked, stop retransmit 1301 * timer and remember to restart (more output or persist). 1302 * If there is more data to be acked, restart retransmit 1303 * timer, using current (possibly backed-off) value. 1304 */ 1305 if (ti->ti_ack == tp->snd_max) { 1306 tp->t_timer[TCPT_REXMT] = 0; 1307 needoutput = 1; 1308 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1309 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1310 1311 /* 1312 * If no data (only SYN) was ACK'd, 1313 * skip rest of ACK processing. 1314 */ 1315 if (acked == 0) 1316 goto step6; 1317 1318 /* 1319 * When new data is acked, open the congestion window. 1320 * If the window gives us less than ssthresh packets 1321 * in flight, open exponentially (maxseg per packet). 1322 * Otherwise open linearly: maxseg per window 1323 * (maxseg^2 / cwnd per packet). 1324 */ 1325 { 1326 register u_int cw = tp->snd_cwnd; 1327 register u_int incr = tp->t_maxseg; 1328 1329 if (cw > tp->snd_ssthresh) 1330 incr = incr * incr / cw; 1331 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1332 } 1333 if (acked > so->so_snd.sb_cc) { 1334 tp->snd_wnd -= so->so_snd.sb_cc; 1335 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1336 ourfinisacked = 1; 1337 } else { 1338 sbdrop(&so->so_snd, acked); 1339 tp->snd_wnd -= acked; 1340 ourfinisacked = 0; 1341 } 1342 if (so->so_snd.sb_flags & SB_NOTIFY) 1343 sowwakeup(so); 1344 tp->snd_una = ti->ti_ack; 1345 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1346 tp->snd_nxt = tp->snd_una; 1347 1348 switch (tp->t_state) { 1349 1350 /* 1351 * In FIN_WAIT_1 STATE in addition to the processing 1352 * for the ESTABLISHED state if our FIN is now acknowledged 1353 * then enter FIN_WAIT_2. 1354 */ 1355 case TCPS_FIN_WAIT_1: 1356 if (ourfinisacked) { 1357 /* 1358 * If we can't receive any more 1359 * data, then closing user can proceed. 1360 * Starting the timer is contrary to the 1361 * specification, but if we don't get a FIN 1362 * we'll hang forever. 1363 */ 1364 if (so->so_state & SS_CANTRCVMORE) { 1365 soisdisconnected(so); 1366 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1367 } 1368 tp->t_state = TCPS_FIN_WAIT_2; 1369 } 1370 break; 1371 1372 /* 1373 * In CLOSING STATE in addition to the processing for 1374 * the ESTABLISHED state if the ACK acknowledges our FIN 1375 * then enter the TIME-WAIT state, otherwise ignore 1376 * the segment. 1377 */ 1378 case TCPS_CLOSING: 1379 if (ourfinisacked) { 1380 tp->t_state = TCPS_TIME_WAIT; 1381 tcp_canceltimers(tp); 1382 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1383 if (tp->cc_recv != 0 && 1384 tp->t_duration < TCPTV_MSL) 1385 tp->t_timer[TCPT_2MSL] = 1386 tp->t_rxtcur * TCPTV_TWTRUNC; 1387 else 1388 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1389 soisdisconnected(so); 1390 } 1391 break; 1392 1393 /* 1394 * In LAST_ACK, we may still be waiting for data to drain 1395 * and/or to be acked, as well as for the ack of our FIN. 1396 * If our FIN is now acknowledged, delete the TCB, 1397 * enter the closed state and return. 1398 */ 1399 case TCPS_LAST_ACK: 1400 if (ourfinisacked) { 1401 tp = tcp_close(tp); 1402 goto drop; 1403 } 1404 break; 1405 1406 /* 1407 * In TIME_WAIT state the only thing that should arrive 1408 * is a retransmission of the remote FIN. Acknowledge 1409 * it and restart the finack timer. 1410 */ 1411 case TCPS_TIME_WAIT: 1412 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1413 goto dropafterack; 1414 } 1415 } 1416 1417step6: 1418 /* 1419 * Update window information. 1420 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1421 */ 1422 if ((tiflags & TH_ACK) && 1423 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1424 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1425 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1426 /* keep track of pure window updates */ 1427 if (ti->ti_len == 0 && 1428 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1429 tcpstat.tcps_rcvwinupd++; 1430 tp->snd_wnd = tiwin; 1431 tp->snd_wl1 = ti->ti_seq; 1432 tp->snd_wl2 = ti->ti_ack; 1433 if (tp->snd_wnd > tp->max_sndwnd) 1434 tp->max_sndwnd = tp->snd_wnd; 1435 needoutput = 1; 1436 } 1437 1438 /* 1439 * Process segments with URG. 1440 */ 1441 if ((tiflags & TH_URG) && ti->ti_urp && 1442 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1443 /* 1444 * This is a kludge, but if we receive and accept 1445 * random urgent pointers, we'll crash in 1446 * soreceive. It's hard to imagine someone 1447 * actually wanting to send this much urgent data. 1448 */ 1449 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1450 ti->ti_urp = 0; /* XXX */ 1451 tiflags &= ~TH_URG; /* XXX */ 1452 goto dodata; /* XXX */ 1453 } 1454 /* 1455 * If this segment advances the known urgent pointer, 1456 * then mark the data stream. This should not happen 1457 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1458 * a FIN has been received from the remote side. 1459 * In these states we ignore the URG. 1460 * 1461 * According to RFC961 (Assigned Protocols), 1462 * the urgent pointer points to the last octet 1463 * of urgent data. We continue, however, 1464 * to consider it to indicate the first octet 1465 * of data past the urgent section as the original 1466 * spec states (in one of two places). 1467 */ 1468 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1469 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1470 so->so_oobmark = so->so_rcv.sb_cc + 1471 (tp->rcv_up - tp->rcv_nxt) - 1; 1472 if (so->so_oobmark == 0) 1473 so->so_state |= SS_RCVATMARK; 1474 sohasoutofband(so); 1475 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1476 } 1477 /* 1478 * Remove out of band data so doesn't get presented to user. 1479 * This can happen independent of advancing the URG pointer, 1480 * but if two URG's are pending at once, some out-of-band 1481 * data may creep in... ick. 1482 */ 1483 if (ti->ti_urp <= (u_long)ti->ti_len 1484#ifdef SO_OOBINLINE 1485 && (so->so_options & SO_OOBINLINE) == 0 1486#endif 1487 ) 1488 tcp_pulloutofband(so, ti, m); 1489 } else 1490 /* 1491 * If no out of band data is expected, 1492 * pull receive urgent pointer along 1493 * with the receive window. 1494 */ 1495 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1496 tp->rcv_up = tp->rcv_nxt; 1497dodata: /* XXX */ 1498 1499 /* 1500 * Process the segment text, merging it into the TCP sequencing queue, 1501 * and arranging for acknowledgment of receipt if necessary. 1502 * This process logically involves adjusting tp->rcv_wnd as data 1503 * is presented to the user (this happens in tcp_usrreq.c, 1504 * case PRU_RCVD). If a FIN has already been received on this 1505 * connection then we just ignore the text. 1506 */ 1507 if ((ti->ti_len || (tiflags&TH_FIN)) && 1508 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1509 TCP_REASS(tp, ti, m, so, tiflags); 1510 /* 1511 * Note the amount of data that peer has sent into 1512 * our window, in order to estimate the sender's 1513 * buffer size. 1514 */ 1515 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1516 } else { 1517 m_freem(m); 1518 tiflags &= ~TH_FIN; 1519 } 1520 1521 /* 1522 * If FIN is received ACK the FIN and let the user know 1523 * that the connection is closing. 1524 */ 1525 if (tiflags & TH_FIN) { 1526 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1527 socantrcvmore(so); 1528 /* 1529 * If connection is half-synchronized 1530 * (ie NEEDSYN flag on) then delay ACK, 1531 * so it may be piggybacked when SYN is sent. 1532 * Otherwise, since we received a FIN then no 1533 * more input can be expected, send ACK now. 1534 */ 1535 if (tp->t_flags & TF_NEEDSYN) 1536 tp->t_flags |= TF_DELACK; 1537 else 1538 tp->t_flags |= TF_ACKNOW; 1539 tp->rcv_nxt++; 1540 } 1541 switch (tp->t_state) { 1542 1543 /* 1544 * In SYN_RECEIVED and ESTABLISHED STATES 1545 * enter the CLOSE_WAIT state. 1546 */ 1547 case TCPS_SYN_RECEIVED: 1548 case TCPS_ESTABLISHED: 1549 tp->t_state = TCPS_CLOSE_WAIT; 1550 break; 1551 1552 /* 1553 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1554 * enter the CLOSING state. 1555 */ 1556 case TCPS_FIN_WAIT_1: 1557 tp->t_state = TCPS_CLOSING; 1558 break; 1559 1560 /* 1561 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1562 * starting the time-wait timer, turning off the other 1563 * standard timers. 1564 */ 1565 case TCPS_FIN_WAIT_2: 1566 tp->t_state = TCPS_TIME_WAIT; 1567 tcp_canceltimers(tp); 1568 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1569 if (tp->cc_recv != 0 && 1570 tp->t_duration < TCPTV_MSL) { 1571 tp->t_timer[TCPT_2MSL] = 1572 tp->t_rxtcur * TCPTV_TWTRUNC; 1573 /* For transaction client, force ACK now. */ 1574 tp->t_flags |= TF_ACKNOW; 1575 } 1576 else 1577 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1578 soisdisconnected(so); 1579 break; 1580 1581 /* 1582 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1583 */ 1584 case TCPS_TIME_WAIT: 1585 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1586 break; 1587 } 1588 } 1589#ifdef TCPDEBUG 1590 if (so->so_options & SO_DEBUG) 1591 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1592#endif 1593 1594 /* 1595 * Return any desired output. 1596 */ 1597 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1598 (void) tcp_output(tp); 1599 return; 1600 1601dropafterack: 1602 /* 1603 * Generate an ACK dropping incoming segment if it occupies 1604 * sequence space, where the ACK reflects our state. 1605 */ 1606 if (tiflags & TH_RST) 1607 goto drop; 1608#ifdef TCPDEBUG 1609 if (so->so_options & SO_DEBUG) 1610 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1611#endif 1612 m_freem(m); 1613 tp->t_flags |= TF_ACKNOW; 1614 (void) tcp_output(tp); 1615 return; 1616 1617dropwithreset: 1618 /* 1619 * Generate a RST, dropping incoming segment. 1620 * Make ACK acceptable to originator of segment. 1621 * Don't bother to respond if destination was broadcast/multicast. 1622 */ 1623 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1624 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1625 goto drop; 1626#ifdef TCPDEBUG 1627 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1628 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1629#endif 1630 if (tiflags & TH_ACK) 1631 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1632 else { 1633 if (tiflags & TH_SYN) 1634 ti->ti_len++; 1635 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1636 TH_RST|TH_ACK); 1637 } 1638 /* destroy temporarily created socket */ 1639 if (dropsocket) 1640 (void) soabort(so); 1641 return; 1642 1643drop: 1644 /* 1645 * Drop space held by incoming segment and return. 1646 */ 1647#ifdef TCPDEBUG 1648 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1649 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1650#endif 1651 m_freem(m); 1652 /* destroy temporarily created socket */ 1653 if (dropsocket) 1654 (void) soabort(so); 1655 return; 1656#ifndef TUBA_INCLUDE 1657} 1658 1659void 1660tcp_dooptions(tp, cp, cnt, ti, to) 1661 struct tcpcb *tp; 1662 u_char *cp; 1663 int cnt; 1664 struct tcpiphdr *ti; 1665 struct tcpopt *to; 1666{ 1667 u_short mss = 0; 1668 int opt, optlen; 1669 1670 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1671 opt = cp[0]; 1672 if (opt == TCPOPT_EOL) 1673 break; 1674 if (opt == TCPOPT_NOP) 1675 optlen = 1; 1676 else { 1677 optlen = cp[1]; 1678 if (optlen <= 0) 1679 break; 1680 } 1681 switch (opt) { 1682 1683 default: 1684 continue; 1685 1686 case TCPOPT_MAXSEG: 1687 if (optlen != TCPOLEN_MAXSEG) 1688 continue; 1689 if (!(ti->ti_flags & TH_SYN)) 1690 continue; 1691 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1692 NTOHS(mss); 1693 break; 1694 1695 case TCPOPT_WINDOW: 1696 if (optlen != TCPOLEN_WINDOW) 1697 continue; 1698 if (!(ti->ti_flags & TH_SYN)) 1699 continue; 1700 tp->t_flags |= TF_RCVD_SCALE; 1701 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1702 break; 1703 1704 case TCPOPT_TIMESTAMP: 1705 if (optlen != TCPOLEN_TIMESTAMP) 1706 continue; 1707 to->to_flag |= TOF_TS; 1708 bcopy((char *)cp + 2, 1709 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1710 NTOHL(to->to_tsval); 1711 bcopy((char *)cp + 6, 1712 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1713 NTOHL(to->to_tsecr); 1714 1715 /* 1716 * A timestamp received in a SYN makes 1717 * it ok to send timestamp requests and replies. 1718 */ 1719 if (ti->ti_flags & TH_SYN) { 1720 tp->t_flags |= TF_RCVD_TSTMP; 1721 tp->ts_recent = to->to_tsval; 1722 tp->ts_recent_age = tcp_now; 1723 } 1724 break; 1725 case TCPOPT_CC: 1726 if (optlen != TCPOLEN_CC) 1727 continue; 1728 to->to_flag |= TOF_CC; 1729 bcopy((char *)cp + 2, 1730 (char *)&to->to_cc, sizeof(to->to_cc)); 1731 NTOHL(to->to_cc); 1732 /* 1733 * A CC or CC.new option received in a SYN makes 1734 * it ok to send CC in subsequent segments. 1735 */ 1736 if (ti->ti_flags & TH_SYN) 1737 tp->t_flags |= TF_RCVD_CC; 1738 break; 1739 case TCPOPT_CCNEW: 1740 if (optlen != TCPOLEN_CC) 1741 continue; 1742 if (!(ti->ti_flags & TH_SYN)) 1743 continue; 1744 to->to_flag |= TOF_CCNEW; 1745 bcopy((char *)cp + 2, 1746 (char *)&to->to_cc, sizeof(to->to_cc)); 1747 NTOHL(to->to_cc); 1748 /* 1749 * A CC or CC.new option received in a SYN makes 1750 * it ok to send CC in subsequent segments. 1751 */ 1752 tp->t_flags |= TF_RCVD_CC; 1753 break; 1754 case TCPOPT_CCECHO: 1755 if (optlen != TCPOLEN_CC) 1756 continue; 1757 if (!(ti->ti_flags & TH_SYN)) 1758 continue; 1759 to->to_flag |= TOF_CCECHO; 1760 bcopy((char *)cp + 2, 1761 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1762 NTOHL(to->to_ccecho); 1763 break; 1764 } 1765 } 1766 if (ti->ti_flags & TH_SYN) 1767 tcp_mss(tp, mss); /* sets t_maxseg */ 1768} 1769 1770/* 1771 * Pull out of band byte out of a segment so 1772 * it doesn't appear in the user's data queue. 1773 * It is still reflected in the segment length for 1774 * sequencing purposes. 1775 */ 1776void 1777tcp_pulloutofband(so, ti, m) 1778 struct socket *so; 1779 struct tcpiphdr *ti; 1780 register struct mbuf *m; 1781{ 1782 int cnt = ti->ti_urp - 1; 1783 1784 while (cnt >= 0) { 1785 if (m->m_len > cnt) { 1786 char *cp = mtod(m, caddr_t) + cnt; 1787 struct tcpcb *tp = sototcpcb(so); 1788 1789 tp->t_iobc = *cp; 1790 tp->t_oobflags |= TCPOOB_HAVEDATA; 1791 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1792 m->m_len--; 1793 return; 1794 } 1795 cnt -= m->m_len; 1796 m = m->m_next; 1797 if (m == 0) 1798 break; 1799 } 1800 panic("tcp_pulloutofband"); 1801} 1802 1803/* 1804 * Collect new round-trip time estimate 1805 * and update averages and current timeout. 1806 */ 1807void 1808tcp_xmit_timer(tp, rtt) 1809 register struct tcpcb *tp; 1810 short rtt; 1811{ 1812 register short delta; 1813 1814 tcpstat.tcps_rttupdated++; 1815 tp->t_rttupdated++; 1816 if (tp->t_srtt != 0) { 1817 /* 1818 * srtt is stored as fixed point with 3 bits after the 1819 * binary point (i.e., scaled by 8). The following magic 1820 * is equivalent to the smoothing algorithm in rfc793 with 1821 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1822 * point). Adjust rtt to origin 0. 1823 */ 1824 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1825 if ((tp->t_srtt += delta) <= 0) 1826 tp->t_srtt = 1; 1827 /* 1828 * We accumulate a smoothed rtt variance (actually, a 1829 * smoothed mean difference), then set the retransmit 1830 * timer to smoothed rtt + 4 times the smoothed variance. 1831 * rttvar is stored as fixed point with 2 bits after the 1832 * binary point (scaled by 4). The following is 1833 * equivalent to rfc793 smoothing with an alpha of .75 1834 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1835 * rfc793's wired-in beta. 1836 */ 1837 if (delta < 0) 1838 delta = -delta; 1839 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1840 if ((tp->t_rttvar += delta) <= 0) 1841 tp->t_rttvar = 1; 1842 } else { 1843 /* 1844 * No rtt measurement yet - use the unsmoothed rtt. 1845 * Set the variance to half the rtt (so our first 1846 * retransmit happens at 3*rtt). 1847 */ 1848 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1849 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1850 } 1851 tp->t_rtt = 0; 1852 tp->t_rxtshift = 0; 1853 1854 /* 1855 * the retransmit should happen at rtt + 4 * rttvar. 1856 * Because of the way we do the smoothing, srtt and rttvar 1857 * will each average +1/2 tick of bias. When we compute 1858 * the retransmit timer, we want 1/2 tick of rounding and 1859 * 1 extra tick because of +-1/2 tick uncertainty in the 1860 * firing of the timer. The bias will give us exactly the 1861 * 1.5 tick we need. But, because the bias is 1862 * statistical, we have to test that we don't drop below 1863 * the minimum feasible timer (which is 2 ticks). 1864 */ 1865 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1866 tp->t_rttmin, TCPTV_REXMTMAX); 1867 1868 /* 1869 * We received an ack for a packet that wasn't retransmitted; 1870 * it is probably safe to discard any error indications we've 1871 * received recently. This isn't quite right, but close enough 1872 * for now (a route might have failed after we sent a segment, 1873 * and the return path might not be symmetrical). 1874 */ 1875 tp->t_softerror = 0; 1876} 1877 1878/* 1879 * Determine a reasonable value for maxseg size. 1880 * If the route is known, check route for mtu. 1881 * If none, use an mss that can be handled on the outgoing 1882 * interface without forcing IP to fragment; if bigger than 1883 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1884 * to utilize large mbufs. If no route is found, route has no mtu, 1885 * or the destination isn't local, use a default, hopefully conservative 1886 * size (usually 512 or the default IP max size, but no more than the mtu 1887 * of the interface), as we can't discover anything about intervening 1888 * gateways or networks. We also initialize the congestion/slow start 1889 * window to be a single segment if the destination isn't local. 1890 * While looking at the routing entry, we also initialize other path-dependent 1891 * parameters from pre-set or cached values in the routing entry. 1892 * 1893 * Also take into account the space needed for options that we 1894 * send regularly. Make maxseg shorter by that amount to assure 1895 * that we can send maxseg amount of data even when the options 1896 * are present. Store the upper limit of the length of options plus 1897 * data in maxopd. 1898 * 1899 * NOTE that this routine is only called when we process an incoming 1900 * segment, for outgoing segments only tcp_mssopt is called. 1901 * 1902 * In case of T/TCP, we call this routine during implicit connection 1903 * setup as well (offer = -1), to initialize maxseg from the cached 1904 * MSS of our peer. 1905 */ 1906void 1907tcp_mss(tp, offer) 1908 struct tcpcb *tp; 1909 int offer; 1910{ 1911 register struct rtentry *rt; 1912 struct ifnet *ifp; 1913 register int rtt, mss; 1914 u_long bufsize; 1915 struct inpcb *inp; 1916 struct socket *so; 1917 struct rmxp_tao *taop; 1918 int origoffer = offer; 1919 1920 inp = tp->t_inpcb; 1921 if ((rt = tcp_rtlookup(inp)) == NULL) { 1922 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1923 return; 1924 } 1925 ifp = rt->rt_ifp; 1926 so = inp->inp_socket; 1927 1928 taop = rmx_taop(rt->rt_rmx); 1929 /* 1930 * Offer == -1 means that we didn't receive SYN yet, 1931 * use cached value in that case; 1932 */ 1933 if (offer == -1) 1934 offer = taop->tao_mssopt; 1935 /* 1936 * Offer == 0 means that there was no MSS on the SYN segment, 1937 * in this case we use tcp_mssdflt. 1938 */ 1939 if (offer == 0) 1940 offer = tcp_mssdflt; 1941 else 1942 /* 1943 * Sanity check: make sure that maxopd will be large 1944 * enough to allow some data on segments even is the 1945 * all the option space is used (40bytes). Otherwise 1946 * funny things may happen in tcp_output. 1947 */ 1948 offer = max(offer, 64); 1949 taop->tao_mssopt = offer; 1950 1951 /* 1952 * While we're here, check if there's an initial rtt 1953 * or rttvar. Convert from the route-table units 1954 * to scaled multiples of the slow timeout timer. 1955 */ 1956 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1957 /* 1958 * XXX the lock bit for RTT indicates that the value 1959 * is also a minimum value; this is subject to time. 1960 */ 1961 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1962 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1963 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1964 tcpstat.tcps_usedrtt++; 1965 if (rt->rt_rmx.rmx_rttvar) { 1966 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1967 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1968 tcpstat.tcps_usedrttvar++; 1969 } else { 1970 /* default variation is +- 1 rtt */ 1971 tp->t_rttvar = 1972 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1973 } 1974 TCPT_RANGESET(tp->t_rxtcur, 1975 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1976 tp->t_rttmin, TCPTV_REXMTMAX); 1977 } 1978 /* 1979 * if there's an mtu associated with the route, use it 1980 */ 1981 if (rt->rt_rmx.rmx_mtu) 1982 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1983 else 1984 { 1985 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1986 if (!in_localaddr(inp->inp_faddr)) 1987 mss = min(mss, tcp_mssdflt); 1988 } 1989 mss = min(mss, offer); 1990 /* 1991 * maxopd stores the maximum length of data AND options 1992 * in a segment; maxseg is the amount of data in a normal 1993 * segment. We need to store this value (maxopd) apart 1994 * from maxseg, because now every segment carries options 1995 * and thus we normally have somewhat less data in segments. 1996 */ 1997 tp->t_maxopd = mss; 1998 1999 /* 2000 * In case of T/TCP, origoffer==-1 indicates, that no segments 2001 * were received yet. In this case we just guess, otherwise 2002 * we do the same as before T/TCP. 2003 */ 2004 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2005 (origoffer == -1 || 2006 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2007 mss -= TCPOLEN_TSTAMP_APPA; 2008 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2009 (origoffer == -1 || 2010 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2011 mss -= TCPOLEN_CC_APPA; 2012 2013#if (MCLBYTES & (MCLBYTES - 1)) == 0 2014 if (mss > MCLBYTES) 2015 mss &= ~(MCLBYTES-1); 2016#else 2017 if (mss > MCLBYTES) 2018 mss = mss / MCLBYTES * MCLBYTES; 2019#endif 2020 /* 2021 * If there's a pipesize, change the socket buffer 2022 * to that size. Make the socket buffers an integral 2023 * number of mss units; if the mss is larger than 2024 * the socket buffer, decrease the mss. 2025 */ 2026#ifdef RTV_SPIPE 2027 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2028#endif 2029 bufsize = so->so_snd.sb_hiwat; 2030 if (bufsize < mss) 2031 mss = bufsize; 2032 else { 2033 bufsize = roundup(bufsize, mss); 2034 if (bufsize > sb_max) 2035 bufsize = sb_max; 2036 (void)sbreserve(&so->so_snd, bufsize); 2037 } 2038 tp->t_maxseg = mss; 2039 2040#ifdef RTV_RPIPE 2041 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2042#endif 2043 bufsize = so->so_rcv.sb_hiwat; 2044 if (bufsize > mss) { 2045 bufsize = roundup(bufsize, mss); 2046 if (bufsize > sb_max) 2047 bufsize = sb_max; 2048 (void)sbreserve(&so->so_rcv, bufsize); 2049 } 2050 /* 2051 * Don't force slow-start on local network. 2052 */ 2053 if (!in_localaddr(inp->inp_faddr)) 2054 tp->snd_cwnd = mss; 2055 2056 if (rt->rt_rmx.rmx_ssthresh) { 2057 /* 2058 * There's some sort of gateway or interface 2059 * buffer limit on the path. Use this to set 2060 * the slow start threshhold, but set the 2061 * threshold to no less than 2*mss. 2062 */ 2063 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2064 tcpstat.tcps_usedssthresh++; 2065 } 2066} 2067 2068/* 2069 * Determine the MSS option to send on an outgoing SYN. 2070 */ 2071int 2072tcp_mssopt(tp) 2073 struct tcpcb *tp; 2074{ 2075 struct rtentry *rt; 2076 2077 rt = tcp_rtlookup(tp->t_inpcb); 2078 if (rt == NULL) 2079 return tcp_mssdflt; 2080 2081 /* 2082 * if there's an mtu associated with the route, use it 2083 */ 2084 if (rt->rt_rmx.rmx_mtu) 2085 return rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2086 2087 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2088} 2089#endif /* TUBA_INCLUDE */ 2090