tcp_input.c revision 15396
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 1.42 1996/04/09 07:01:51 pst Exp $ 35 */ 36 37#ifndef TUBA_INCLUDE 38#include <sys/param.h> 39#include <sys/queue.h> 40#include <sys/systm.h> 41#include <sys/kernel.h> 42#include <sys/sysctl.h> 43#include <sys/malloc.h> 44#include <sys/mbuf.h> 45#include <sys/protosw.h> 46#include <sys/socket.h> 47#include <sys/socketvar.h> 48#include <sys/errno.h> 49#include <sys/syslog.h> 50 51#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 52 53#include <net/if.h> 54#include <net/route.h> 55 56#include <netinet/in.h> 57#include <netinet/in_systm.h> 58#include <netinet/ip.h> 59#include <netinet/in_pcb.h> 60#include <netinet/ip_var.h> 61#include <netinet/tcp.h> 62#include <netinet/tcp_fsm.h> 63#include <netinet/tcp_seq.h> 64#include <netinet/tcp_timer.h> 65#include <netinet/tcp_var.h> 66#include <netinet/tcpip.h> 67#ifdef TCPDEBUG 68#include <netinet/tcp_debug.h> 69static struct tcpiphdr tcp_saveti; 70#endif 71 72static int tcprexmtthresh = 3; 73tcp_seq tcp_iss; 74tcp_cc tcp_ccgen; 75 76struct tcpstat tcpstat; 77SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 78 CTLFLAG_RD, &tcpstat , tcpstat, ""); 79 80static int log_in_vain = 0; 81SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 82 &log_in_vain, 0, ""); 83 84u_long tcp_now; 85struct inpcbhead tcb; 86struct inpcbinfo tcbinfo; 87 88static void tcp_dooptions __P((struct tcpcb *, 89 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 90static void tcp_pulloutofband __P((struct socket *, 91 struct tcpiphdr *, struct mbuf *)); 92static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 93static void tcp_xmit_timer __P((struct tcpcb *, int)); 94 95#endif /* TUBA_INCLUDE */ 96 97/* 98 * Insert segment ti into reassembly queue of tcp with 99 * control block tp. Return TH_FIN if reassembly now includes 100 * a segment with FIN. The macro form does the common case inline 101 * (segment is the next to be received on an established connection, 102 * and the queue is empty), avoiding linkage into and removal 103 * from the queue and repetition of various conversions. 104 * Set DELACK for segments received in order, but ack immediately 105 * when segments are out of order (so fast retransmit can work). 106 */ 107#ifdef TCP_ACK_HACK 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 if (ti->ti_flags & TH_PUSH) \ 113 tp->t_flags |= TF_ACKNOW; \ 114 else \ 115 tp->t_flags |= TF_DELACK; \ 116 (tp)->rcv_nxt += (ti)->ti_len; \ 117 flags = (ti)->ti_flags & TH_FIN; \ 118 tcpstat.tcps_rcvpack++;\ 119 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 120 sbappend(&(so)->so_rcv, (m)); \ 121 sorwakeup(so); \ 122 } else { \ 123 (flags) = tcp_reass((tp), (ti), (m)); \ 124 tp->t_flags |= TF_ACKNOW; \ 125 } \ 126} 127#else 128#define TCP_REASS(tp, ti, m, so, flags) { \ 129 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 130 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 131 (tp)->t_state == TCPS_ESTABLISHED) { \ 132 tp->t_flags |= TF_DELACK; \ 133 (tp)->rcv_nxt += (ti)->ti_len; \ 134 flags = (ti)->ti_flags & TH_FIN; \ 135 tcpstat.tcps_rcvpack++;\ 136 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 137 sbappend(&(so)->so_rcv, (m)); \ 138 sorwakeup(so); \ 139 } else { \ 140 (flags) = tcp_reass((tp), (ti), (m)); \ 141 tp->t_flags |= TF_ACKNOW; \ 142 } \ 143} 144#endif 145#ifndef TUBA_INCLUDE 146 147static int 148tcp_reass(tp, ti, m) 149 register struct tcpcb *tp; 150 register struct tcpiphdr *ti; 151 struct mbuf *m; 152{ 153 register struct tcpiphdr *q; 154 struct socket *so = tp->t_inpcb->inp_socket; 155 int flags; 156 157 /* 158 * Call with ti==0 after become established to 159 * force pre-ESTABLISHED data up to user socket. 160 */ 161 if (ti == 0) 162 goto present; 163 164 /* 165 * Find a segment which begins after this one does. 166 */ 167 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 168 q = (struct tcpiphdr *)q->ti_next) 169 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 170 break; 171 172 /* 173 * If there is a preceding segment, it may provide some of 174 * our data already. If so, drop the data from the incoming 175 * segment. If it provides all of our data, drop us. 176 */ 177 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 178 register int i; 179 q = (struct tcpiphdr *)q->ti_prev; 180 /* conversion to int (in i) handles seq wraparound */ 181 i = q->ti_seq + q->ti_len - ti->ti_seq; 182 if (i > 0) { 183 if (i >= ti->ti_len) { 184 tcpstat.tcps_rcvduppack++; 185 tcpstat.tcps_rcvdupbyte += ti->ti_len; 186 m_freem(m); 187 /* 188 * Try to present any queued data 189 * at the left window edge to the user. 190 * This is needed after the 3-WHS 191 * completes. 192 */ 193 goto present; /* ??? */ 194 } 195 m_adj(m, i); 196 ti->ti_len -= i; 197 ti->ti_seq += i; 198 } 199 q = (struct tcpiphdr *)(q->ti_next); 200 } 201 tcpstat.tcps_rcvoopack++; 202 tcpstat.tcps_rcvoobyte += ti->ti_len; 203 REASS_MBUF(ti) = m; /* XXX */ 204 205 /* 206 * While we overlap succeeding segments trim them or, 207 * if they are completely covered, dequeue them. 208 */ 209 while (q != (struct tcpiphdr *)tp) { 210 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 211 if (i <= 0) 212 break; 213 if (i < q->ti_len) { 214 q->ti_seq += i; 215 q->ti_len -= i; 216 m_adj(REASS_MBUF(q), i); 217 break; 218 } 219 q = (struct tcpiphdr *)q->ti_next; 220 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 221 remque(q->ti_prev); 222 m_freem(m); 223 } 224 225 /* 226 * Stick new segment in its place. 227 */ 228 insque(ti, q->ti_prev); 229 230present: 231 /* 232 * Present data to user, advancing rcv_nxt through 233 * completed sequence space. 234 */ 235 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 236 return (0); 237 ti = tp->seg_next; 238 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 239 return (0); 240 do { 241 tp->rcv_nxt += ti->ti_len; 242 flags = ti->ti_flags & TH_FIN; 243 remque(ti); 244 m = REASS_MBUF(ti); 245 ti = (struct tcpiphdr *)ti->ti_next; 246 if (so->so_state & SS_CANTRCVMORE) 247 m_freem(m); 248 else 249 sbappend(&so->so_rcv, m); 250 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 251 sorwakeup(so); 252 return (flags); 253} 254 255/* 256 * TCP input routine, follows pages 65-76 of the 257 * protocol specification dated September, 1981 very closely. 258 */ 259void 260tcp_input(m, iphlen) 261 register struct mbuf *m; 262 int iphlen; 263{ 264 register struct tcpiphdr *ti; 265 register struct inpcb *inp; 266 u_char *optp = NULL; 267 int optlen = 0; 268 int len, tlen, off; 269 register struct tcpcb *tp = 0; 270 register int tiflags; 271 struct socket *so = 0; 272 int todrop, acked, ourfinisacked, needoutput = 0; 273 struct in_addr laddr; 274 int dropsocket = 0; 275 int iss = 0; 276 u_long tiwin; 277 struct tcpopt to; /* options in this segment */ 278 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 279 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 280#ifdef TCPDEBUG 281 short ostate = 0; 282#endif 283 284 bzero((char *)&to, sizeof(to)); 285 286 tcpstat.tcps_rcvtotal++; 287 /* 288 * Get IP and TCP header together in first mbuf. 289 * Note: IP leaves IP header in first mbuf. 290 */ 291 ti = mtod(m, struct tcpiphdr *); 292 if (iphlen > sizeof (struct ip)) 293 ip_stripoptions(m, (struct mbuf *)0); 294 if (m->m_len < sizeof (struct tcpiphdr)) { 295 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 296 tcpstat.tcps_rcvshort++; 297 return; 298 } 299 ti = mtod(m, struct tcpiphdr *); 300 } 301 302 /* 303 * Checksum extended TCP header and data. 304 */ 305 tlen = ((struct ip *)ti)->ip_len; 306 len = sizeof (struct ip) + tlen; 307 ti->ti_next = ti->ti_prev = 0; 308 ti->ti_x1 = 0; 309 ti->ti_len = (u_short)tlen; 310 HTONS(ti->ti_len); 311 ti->ti_sum = in_cksum(m, len); 312 if (ti->ti_sum) { 313 tcpstat.tcps_rcvbadsum++; 314 goto drop; 315 } 316#endif /* TUBA_INCLUDE */ 317 318 /* 319 * Check that TCP offset makes sense, 320 * pull out TCP options and adjust length. XXX 321 */ 322 off = ti->ti_off << 2; 323 if (off < sizeof (struct tcphdr) || off > tlen) { 324 tcpstat.tcps_rcvbadoff++; 325 goto drop; 326 } 327 tlen -= off; 328 ti->ti_len = tlen; 329 if (off > sizeof (struct tcphdr)) { 330 if (m->m_len < sizeof(struct ip) + off) { 331 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 332 tcpstat.tcps_rcvshort++; 333 return; 334 } 335 ti = mtod(m, struct tcpiphdr *); 336 } 337 optlen = off - sizeof (struct tcphdr); 338 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 339 } 340 tiflags = ti->ti_flags; 341 342 /* 343 * Convert TCP protocol specific fields to host format. 344 */ 345 NTOHL(ti->ti_seq); 346 NTOHL(ti->ti_ack); 347 NTOHS(ti->ti_win); 348 NTOHS(ti->ti_urp); 349 350 /* 351 * Drop TCP, IP headers and TCP options. 352 */ 353 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 354 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 355 356 /* 357 * Locate pcb for segment. 358 */ 359findpcb: 360 /* 361 * First look for an exact match. 362 */ 363 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 364 ti->ti_dst, ti->ti_dport); 365 /* 366 * ...and if that fails, do a wildcard search. 367 */ 368 if (inp == NULL) { 369 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 370 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 371 } 372 373 /* 374 * If the state is CLOSED (i.e., TCB does not exist) then 375 * all data in the incoming segment is discarded. 376 * If the TCB exists but is in CLOSED state, it is embryonic, 377 * but should either do a listen or a connect soon. 378 */ 379 if (inp == NULL) { 380 if (log_in_vain && tiflags & TH_SYN) 381 log(LOG_INFO, "Connection attempt to TCP %s:%d" 382 " from %s:%d\n", 383 inet_ntoa(ti->ti_dst), ntohs(ti->ti_dport), 384 inet_ntoa(ti->ti_src), ntohs(ti->ti_sport)); 385 goto dropwithreset; 386 } 387 tp = intotcpcb(inp); 388 if (tp == 0) 389 goto dropwithreset; 390 if (tp->t_state == TCPS_CLOSED) 391 goto drop; 392 393 /* Unscale the window into a 32-bit value. */ 394 if ((tiflags & TH_SYN) == 0) 395 tiwin = ti->ti_win << tp->snd_scale; 396 else 397 tiwin = ti->ti_win; 398 399 so = inp->inp_socket; 400 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 401#ifdef TCPDEBUG 402 if (so->so_options & SO_DEBUG) { 403 ostate = tp->t_state; 404 tcp_saveti = *ti; 405 } 406#endif 407 if (so->so_options & SO_ACCEPTCONN) { 408 register struct tcpcb *tp0 = tp; 409 so = sonewconn(so, 0); 410 if (so == 0) { 411 tcpstat.tcps_listendrop++; 412 goto drop; 413 } 414 /* 415 * This is ugly, but .... 416 * 417 * Mark socket as temporary until we're 418 * committed to keeping it. The code at 419 * ``drop'' and ``dropwithreset'' check the 420 * flag dropsocket to see if the temporary 421 * socket created here should be discarded. 422 * We mark the socket as discardable until 423 * we're committed to it below in TCPS_LISTEN. 424 */ 425 dropsocket++; 426 inp = (struct inpcb *)so->so_pcb; 427 inp->inp_laddr = ti->ti_dst; 428 inp->inp_lport = ti->ti_dport; 429 in_pcbrehash(inp); 430#if BSD>=43 431 inp->inp_options = ip_srcroute(); 432#endif 433 tp = intotcpcb(inp); 434 tp->t_state = TCPS_LISTEN; 435 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 436 437 /* Compute proper scaling value from buffer space */ 438 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 439 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 440 tp->request_r_scale++; 441 } 442 } 443 444 /* 445 * Segment received on connection. 446 * Reset idle time and keep-alive timer. 447 */ 448 tp->t_idle = 0; 449 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 450 451 /* 452 * Process options if not in LISTEN state, 453 * else do it below (after getting remote address). 454 */ 455 if (tp->t_state != TCPS_LISTEN) 456 tcp_dooptions(tp, optp, optlen, ti, &to); 457 458 /* 459 * Header prediction: check for the two common cases 460 * of a uni-directional data xfer. If the packet has 461 * no control flags, is in-sequence, the window didn't 462 * change and we're not retransmitting, it's a 463 * candidate. If the length is zero and the ack moved 464 * forward, we're the sender side of the xfer. Just 465 * free the data acked & wake any higher level process 466 * that was blocked waiting for space. If the length 467 * is non-zero and the ack didn't move, we're the 468 * receiver side. If we're getting packets in-order 469 * (the reassembly queue is empty), add the data to 470 * the socket buffer and note that we need a delayed ack. 471 * Make sure that the hidden state-flags are also off. 472 * Since we check for TCPS_ESTABLISHED above, it can only 473 * be TH_NEEDSYN. 474 */ 475 if (tp->t_state == TCPS_ESTABLISHED && 476 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 477 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 478 ((to.to_flag & TOF_TS) == 0 || 479 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 480 /* 481 * Using the CC option is compulsory if once started: 482 * the segment is OK if no T/TCP was negotiated or 483 * if the segment has a CC option equal to CCrecv 484 */ 485 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 486 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 487 ti->ti_seq == tp->rcv_nxt && 488 tiwin && tiwin == tp->snd_wnd && 489 tp->snd_nxt == tp->snd_max) { 490 491 /* 492 * If last ACK falls within this segment's sequence numbers, 493 * record the timestamp. 494 * NOTE that the test is modified according to the latest 495 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 496 */ 497 if ((to.to_flag & TOF_TS) != 0 && 498 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 499 tp->ts_recent_age = tcp_now; 500 tp->ts_recent = to.to_tsval; 501 } 502 503 if (ti->ti_len == 0) { 504 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 505 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 506 tp->snd_cwnd >= tp->snd_wnd && 507 tp->t_dupacks < tcprexmtthresh) { 508 /* 509 * this is a pure ack for outstanding data. 510 */ 511 ++tcpstat.tcps_predack; 512 if ((to.to_flag & TOF_TS) != 0) 513 tcp_xmit_timer(tp, 514 tcp_now - to.to_tsecr + 1); 515 else if (tp->t_rtt && 516 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 517 tcp_xmit_timer(tp, tp->t_rtt); 518 acked = ti->ti_ack - tp->snd_una; 519 tcpstat.tcps_rcvackpack++; 520 tcpstat.tcps_rcvackbyte += acked; 521 sbdrop(&so->so_snd, acked); 522 tp->snd_una = ti->ti_ack; 523 m_freem(m); 524 525 /* 526 * If all outstanding data are acked, stop 527 * retransmit timer, otherwise restart timer 528 * using current (possibly backed-off) value. 529 * If process is waiting for space, 530 * wakeup/selwakeup/signal. If data 531 * are ready to send, let tcp_output 532 * decide between more output or persist. 533 */ 534 if (tp->snd_una == tp->snd_max) 535 tp->t_timer[TCPT_REXMT] = 0; 536 else if (tp->t_timer[TCPT_PERSIST] == 0) 537 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 538 539 if (so->so_snd.sb_flags & SB_NOTIFY) 540 sowwakeup(so); 541 if (so->so_snd.sb_cc) 542 (void) tcp_output(tp); 543 return; 544 } 545 } else if (ti->ti_ack == tp->snd_una && 546 tp->seg_next == (struct tcpiphdr *)tp && 547 ti->ti_len <= sbspace(&so->so_rcv)) { 548 /* 549 * this is a pure, in-sequence data packet 550 * with nothing on the reassembly queue and 551 * we have enough buffer space to take it. 552 */ 553 ++tcpstat.tcps_preddat; 554 tp->rcv_nxt += ti->ti_len; 555 tcpstat.tcps_rcvpack++; 556 tcpstat.tcps_rcvbyte += ti->ti_len; 557 /* 558 * Add data to socket buffer. 559 */ 560 sbappend(&so->so_rcv, m); 561 sorwakeup(so); 562#ifdef TCP_ACK_HACK 563 /* 564 * If this is a short packet, then ACK now - with Nagel 565 * congestion avoidance sender won't send more until 566 * he gets an ACK. 567 */ 568 if (tiflags & TH_PUSH) { 569 tp->t_flags |= TF_ACKNOW; 570 tcp_output(tp); 571 } else { 572 tp->t_flags |= TF_DELACK; 573 } 574#else 575 tp->t_flags |= TF_DELACK; 576#endif 577 return; 578 } 579 } 580 581 /* 582 * Calculate amount of space in receive window, 583 * and then do TCP input processing. 584 * Receive window is amount of space in rcv queue, 585 * but not less than advertised window. 586 */ 587 { int win; 588 589 win = sbspace(&so->so_rcv); 590 if (win < 0) 591 win = 0; 592 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 593 } 594 595 switch (tp->t_state) { 596 597 /* 598 * If the state is LISTEN then ignore segment if it contains an RST. 599 * If the segment contains an ACK then it is bad and send a RST. 600 * If it does not contain a SYN then it is not interesting; drop it. 601 * Don't bother responding if the destination was a broadcast. 602 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 603 * tp->iss, and send a segment: 604 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 605 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 606 * Fill in remote peer address fields if not previously specified. 607 * Enter SYN_RECEIVED state, and process any other fields of this 608 * segment in this state. 609 */ 610 case TCPS_LISTEN: { 611 struct mbuf *am; 612 register struct sockaddr_in *sin; 613 614 if (tiflags & TH_RST) 615 goto drop; 616 if (tiflags & TH_ACK) 617 goto dropwithreset; 618 if ((tiflags & TH_SYN) == 0) 619 goto drop; 620 /* 621 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 622 * in_broadcast() should never return true on a received 623 * packet with M_BCAST not set. 624 */ 625 if (m->m_flags & (M_BCAST|M_MCAST) || 626 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 627 goto drop; 628 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 629 if (am == NULL) 630 goto drop; 631 am->m_len = sizeof (struct sockaddr_in); 632 sin = mtod(am, struct sockaddr_in *); 633 sin->sin_family = AF_INET; 634 sin->sin_len = sizeof(*sin); 635 sin->sin_addr = ti->ti_src; 636 sin->sin_port = ti->ti_sport; 637 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 638 laddr = inp->inp_laddr; 639 if (inp->inp_laddr.s_addr == INADDR_ANY) 640 inp->inp_laddr = ti->ti_dst; 641 if (in_pcbconnect(inp, am)) { 642 inp->inp_laddr = laddr; 643 (void) m_free(am); 644 goto drop; 645 } 646 (void) m_free(am); 647 tp->t_template = tcp_template(tp); 648 if (tp->t_template == 0) { 649 tp = tcp_drop(tp, ENOBUFS); 650 dropsocket = 0; /* socket is already gone */ 651 goto drop; 652 } 653 if ((taop = tcp_gettaocache(inp)) == NULL) { 654 taop = &tao_noncached; 655 bzero(taop, sizeof(*taop)); 656 } 657 tcp_dooptions(tp, optp, optlen, ti, &to); 658 if (iss) 659 tp->iss = iss; 660 else 661 tp->iss = tcp_iss; 662 tcp_iss += TCP_ISSINCR/4; 663 tp->irs = ti->ti_seq; 664 tcp_sendseqinit(tp); 665 tcp_rcvseqinit(tp); 666 /* 667 * Initialization of the tcpcb for transaction; 668 * set SND.WND = SEG.WND, 669 * initialize CCsend and CCrecv. 670 */ 671 tp->snd_wnd = tiwin; /* initial send-window */ 672 tp->cc_send = CC_INC(tcp_ccgen); 673 tp->cc_recv = to.to_cc; 674 /* 675 * Perform TAO test on incoming CC (SEG.CC) option, if any. 676 * - compare SEG.CC against cached CC from the same host, 677 * if any. 678 * - if SEG.CC > chached value, SYN must be new and is accepted 679 * immediately: save new CC in the cache, mark the socket 680 * connected, enter ESTABLISHED state, turn on flag to 681 * send a SYN in the next segment. 682 * A virtual advertised window is set in rcv_adv to 683 * initialize SWS prevention. Then enter normal segment 684 * processing: drop SYN, process data and FIN. 685 * - otherwise do a normal 3-way handshake. 686 */ 687 if ((to.to_flag & TOF_CC) != 0) { 688 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 689 taop->tao_cc = to.to_cc; 690 tp->t_state = TCPS_ESTABLISHED; 691 692 /* 693 * If there is a FIN, or if there is data and the 694 * connection is local, then delay SYN,ACK(SYN) in 695 * the hope of piggy-backing it on a response 696 * segment. Otherwise must send ACK now in case 697 * the other side is slow starting. 698 */ 699 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 700 in_localaddr(inp->inp_faddr))) 701 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 702 else 703 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 704 705 /* 706 * Limit the `virtual advertised window' to TCP_MAXWIN 707 * here. Even if we requested window scaling, it will 708 * become effective only later when our SYN is acked. 709 */ 710 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 711 tcpstat.tcps_connects++; 712 soisconnected(so); 713 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 714 dropsocket = 0; /* committed to socket */ 715 tcpstat.tcps_accepts++; 716 goto trimthenstep6; 717 } 718 /* else do standard 3-way handshake */ 719 } else { 720 /* 721 * No CC option, but maybe CC.NEW: 722 * invalidate cached value. 723 */ 724 taop->tao_cc = 0; 725 } 726 /* 727 * TAO test failed or there was no CC option, 728 * do a standard 3-way handshake. 729 */ 730 tp->t_flags |= TF_ACKNOW; 731 tp->t_state = TCPS_SYN_RECEIVED; 732 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 733 dropsocket = 0; /* committed to socket */ 734 tcpstat.tcps_accepts++; 735 goto trimthenstep6; 736 } 737 738 /* 739 * If the state is SYN_SENT: 740 * if seg contains an ACK, but not for our SYN, drop the input. 741 * if seg contains a RST, then drop the connection. 742 * if seg does not contain SYN, then drop it. 743 * Otherwise this is an acceptable SYN segment 744 * initialize tp->rcv_nxt and tp->irs 745 * if seg contains ack then advance tp->snd_una 746 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 747 * arrange for segment to be acked (eventually) 748 * continue processing rest of data/controls, beginning with URG 749 */ 750 case TCPS_SYN_SENT: 751 if ((taop = tcp_gettaocache(inp)) == NULL) { 752 taop = &tao_noncached; 753 bzero(taop, sizeof(*taop)); 754 } 755 756 if ((tiflags & TH_ACK) && 757 (SEQ_LEQ(ti->ti_ack, tp->iss) || 758 SEQ_GT(ti->ti_ack, tp->snd_max))) { 759 /* 760 * If we have a cached CCsent for the remote host, 761 * hence we haven't just crashed and restarted, 762 * do not send a RST. This may be a retransmission 763 * from the other side after our earlier ACK was lost. 764 * Our new SYN, when it arrives, will serve as the 765 * needed ACK. 766 */ 767 if (taop->tao_ccsent != 0) 768 goto drop; 769 else 770 goto dropwithreset; 771 } 772 if (tiflags & TH_RST) { 773 if (tiflags & TH_ACK) 774 tp = tcp_drop(tp, ECONNREFUSED); 775 goto drop; 776 } 777 if ((tiflags & TH_SYN) == 0) 778 goto drop; 779 tp->snd_wnd = ti->ti_win; /* initial send window */ 780 tp->cc_recv = to.to_cc; /* foreign CC */ 781 782 tp->irs = ti->ti_seq; 783 tcp_rcvseqinit(tp); 784 if (tiflags & TH_ACK) { 785 /* 786 * Our SYN was acked. If segment contains CC.ECHO 787 * option, check it to make sure this segment really 788 * matches our SYN. If not, just drop it as old 789 * duplicate, but send an RST if we're still playing 790 * by the old rules. 791 */ 792 if ((to.to_flag & TOF_CCECHO) && 793 tp->cc_send != to.to_ccecho) { 794 if (taop->tao_ccsent != 0) 795 goto drop; 796 else 797 goto dropwithreset; 798 } 799 tcpstat.tcps_connects++; 800 soisconnected(so); 801 /* Do window scaling on this connection? */ 802 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 803 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 804 tp->snd_scale = tp->requested_s_scale; 805 tp->rcv_scale = tp->request_r_scale; 806 } 807 /* Segment is acceptable, update cache if undefined. */ 808 if (taop->tao_ccsent == 0) 809 taop->tao_ccsent = to.to_ccecho; 810 811 tp->rcv_adv += tp->rcv_wnd; 812 tp->snd_una++; /* SYN is acked */ 813 /* 814 * If there's data, delay ACK; if there's also a FIN 815 * ACKNOW will be turned on later. 816 */ 817 if (ti->ti_len != 0) 818 tp->t_flags |= TF_DELACK; 819 else 820 tp->t_flags |= TF_ACKNOW; 821 /* 822 * Received <SYN,ACK> in SYN_SENT[*] state. 823 * Transitions: 824 * SYN_SENT --> ESTABLISHED 825 * SYN_SENT* --> FIN_WAIT_1 826 */ 827 if (tp->t_flags & TF_NEEDFIN) { 828 tp->t_state = TCPS_FIN_WAIT_1; 829 tp->t_flags &= ~TF_NEEDFIN; 830 tiflags &= ~TH_SYN; 831 } else 832 tp->t_state = TCPS_ESTABLISHED; 833 834 } else { 835 /* 836 * Received initial SYN in SYN-SENT[*] state => simul- 837 * taneous open. If segment contains CC option and there is 838 * a cached CC, apply TAO test; if it succeeds, connection is 839 * half-synchronized. Otherwise, do 3-way handshake: 840 * SYN-SENT -> SYN-RECEIVED 841 * SYN-SENT* -> SYN-RECEIVED* 842 * If there was no CC option, clear cached CC value. 843 */ 844 tp->t_flags |= TF_ACKNOW; 845 tp->t_timer[TCPT_REXMT] = 0; 846 if (to.to_flag & TOF_CC) { 847 if (taop->tao_cc != 0 && 848 CC_GT(to.to_cc, taop->tao_cc)) { 849 /* 850 * update cache and make transition: 851 * SYN-SENT -> ESTABLISHED* 852 * SYN-SENT* -> FIN-WAIT-1* 853 */ 854 taop->tao_cc = to.to_cc; 855 if (tp->t_flags & TF_NEEDFIN) { 856 tp->t_state = TCPS_FIN_WAIT_1; 857 tp->t_flags &= ~TF_NEEDFIN; 858 } else 859 tp->t_state = TCPS_ESTABLISHED; 860 tp->t_flags |= TF_NEEDSYN; 861 } else 862 tp->t_state = TCPS_SYN_RECEIVED; 863 } else { 864 /* CC.NEW or no option => invalidate cache */ 865 taop->tao_cc = 0; 866 tp->t_state = TCPS_SYN_RECEIVED; 867 } 868 } 869 870trimthenstep6: 871 /* 872 * Advance ti->ti_seq to correspond to first data byte. 873 * If data, trim to stay within window, 874 * dropping FIN if necessary. 875 */ 876 ti->ti_seq++; 877 if (ti->ti_len > tp->rcv_wnd) { 878 todrop = ti->ti_len - tp->rcv_wnd; 879 m_adj(m, -todrop); 880 ti->ti_len = tp->rcv_wnd; 881 tiflags &= ~TH_FIN; 882 tcpstat.tcps_rcvpackafterwin++; 883 tcpstat.tcps_rcvbyteafterwin += todrop; 884 } 885 tp->snd_wl1 = ti->ti_seq - 1; 886 tp->rcv_up = ti->ti_seq; 887 /* 888 * Client side of transaction: already sent SYN and data. 889 * If the remote host used T/TCP to validate the SYN, 890 * our data will be ACK'd; if so, enter normal data segment 891 * processing in the middle of step 5, ack processing. 892 * Otherwise, goto step 6. 893 */ 894 if (tiflags & TH_ACK) 895 goto process_ACK; 896 goto step6; 897 /* 898 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 899 * if segment contains a SYN and CC [not CC.NEW] option: 900 * if state == TIME_WAIT and connection duration > MSL, 901 * drop packet and send RST; 902 * 903 * if SEG.CC > CCrecv then is new SYN, and can implicitly 904 * ack the FIN (and data) in retransmission queue. 905 * Complete close and delete TCPCB. Then reprocess 906 * segment, hoping to find new TCPCB in LISTEN state; 907 * 908 * else must be old SYN; drop it. 909 * else do normal processing. 910 */ 911 case TCPS_LAST_ACK: 912 case TCPS_CLOSING: 913 case TCPS_TIME_WAIT: 914 if ((tiflags & TH_SYN) && 915 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 916 if (tp->t_state == TCPS_TIME_WAIT && 917 tp->t_duration > TCPTV_MSL) 918 goto dropwithreset; 919 if (CC_GT(to.to_cc, tp->cc_recv)) { 920 tp = tcp_close(tp); 921 goto findpcb; 922 } 923 else 924 goto drop; 925 } 926 break; /* continue normal processing */ 927 } 928 929 /* 930 * States other than LISTEN or SYN_SENT. 931 * First check timestamp, if present. 932 * Then check the connection count, if present. 933 * Then check that at least some bytes of segment are within 934 * receive window. If segment begins before rcv_nxt, 935 * drop leading data (and SYN); if nothing left, just ack. 936 * 937 * RFC 1323 PAWS: If we have a timestamp reply on this segment 938 * and it's less than ts_recent, drop it. 939 */ 940 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 941 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 942 943 /* Check to see if ts_recent is over 24 days old. */ 944 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 945 /* 946 * Invalidate ts_recent. If this segment updates 947 * ts_recent, the age will be reset later and ts_recent 948 * will get a valid value. If it does not, setting 949 * ts_recent to zero will at least satisfy the 950 * requirement that zero be placed in the timestamp 951 * echo reply when ts_recent isn't valid. The 952 * age isn't reset until we get a valid ts_recent 953 * because we don't want out-of-order segments to be 954 * dropped when ts_recent is old. 955 */ 956 tp->ts_recent = 0; 957 } else { 958 tcpstat.tcps_rcvduppack++; 959 tcpstat.tcps_rcvdupbyte += ti->ti_len; 960 tcpstat.tcps_pawsdrop++; 961 goto dropafterack; 962 } 963 } 964 965 /* 966 * T/TCP mechanism 967 * If T/TCP was negotiated and the segment doesn't have CC, 968 * or if it's CC is wrong then drop the segment. 969 * RST segments do not have to comply with this. 970 */ 971 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 972 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 973 (tiflags & TH_RST) == 0) 974 goto dropafterack; 975 976 todrop = tp->rcv_nxt - ti->ti_seq; 977 if (todrop > 0) { 978 if (tiflags & TH_SYN) { 979 tiflags &= ~TH_SYN; 980 ti->ti_seq++; 981 if (ti->ti_urp > 1) 982 ti->ti_urp--; 983 else 984 tiflags &= ~TH_URG; 985 todrop--; 986 } 987 /* 988 * Following if statement from Stevens, vol. 2, p. 960. 989 */ 990 if (todrop > ti->ti_len 991 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 992 /* 993 * Any valid FIN must be to the left of the window. 994 * At this point the FIN must be a duplicate or out 995 * of sequence; drop it. 996 */ 997 tiflags &= ~TH_FIN; 998 999 /* 1000 * Send an ACK to resynchronize and drop any data. 1001 * But keep on processing for RST or ACK. 1002 */ 1003 tp->t_flags |= TF_ACKNOW; 1004 todrop = ti->ti_len; 1005 tcpstat.tcps_rcvduppack++; 1006 tcpstat.tcps_rcvdupbyte += todrop; 1007 } else { 1008 tcpstat.tcps_rcvpartduppack++; 1009 tcpstat.tcps_rcvpartdupbyte += todrop; 1010 } 1011 m_adj(m, todrop); 1012 ti->ti_seq += todrop; 1013 ti->ti_len -= todrop; 1014 if (ti->ti_urp > todrop) 1015 ti->ti_urp -= todrop; 1016 else { 1017 tiflags &= ~TH_URG; 1018 ti->ti_urp = 0; 1019 } 1020 } 1021 1022 /* 1023 * If new data are received on a connection after the 1024 * user processes are gone, then RST the other end. 1025 */ 1026 if ((so->so_state & SS_NOFDREF) && 1027 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1028 tp = tcp_close(tp); 1029 tcpstat.tcps_rcvafterclose++; 1030 goto dropwithreset; 1031 } 1032 1033 /* 1034 * If segment ends after window, drop trailing data 1035 * (and PUSH and FIN); if nothing left, just ACK. 1036 */ 1037 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1038 if (todrop > 0) { 1039 tcpstat.tcps_rcvpackafterwin++; 1040 if (todrop >= ti->ti_len) { 1041 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1042 /* 1043 * If a new connection request is received 1044 * while in TIME_WAIT, drop the old connection 1045 * and start over if the sequence numbers 1046 * are above the previous ones. 1047 */ 1048 if (tiflags & TH_SYN && 1049 tp->t_state == TCPS_TIME_WAIT && 1050 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1051 iss = tp->rcv_nxt + TCP_ISSINCR; 1052 tp = tcp_close(tp); 1053 goto findpcb; 1054 } 1055 /* 1056 * If window is closed can only take segments at 1057 * window edge, and have to drop data and PUSH from 1058 * incoming segments. Continue processing, but 1059 * remember to ack. Otherwise, drop segment 1060 * and ack. 1061 */ 1062 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1063 tp->t_flags |= TF_ACKNOW; 1064 tcpstat.tcps_rcvwinprobe++; 1065 } else 1066 goto dropafterack; 1067 } else 1068 tcpstat.tcps_rcvbyteafterwin += todrop; 1069 m_adj(m, -todrop); 1070 ti->ti_len -= todrop; 1071 tiflags &= ~(TH_PUSH|TH_FIN); 1072 } 1073 1074 /* 1075 * If last ACK falls within this segment's sequence numbers, 1076 * record its timestamp. 1077 * NOTE that the test is modified according to the latest 1078 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1079 */ 1080 if ((to.to_flag & TOF_TS) != 0 && 1081 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1082 tp->ts_recent_age = tcp_now; 1083 tp->ts_recent = to.to_tsval; 1084 } 1085 1086 /* 1087 * If the RST bit is set examine the state: 1088 * SYN_RECEIVED STATE: 1089 * If passive open, return to LISTEN state. 1090 * If active open, inform user that connection was refused. 1091 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1092 * Inform user that connection was reset, and close tcb. 1093 * CLOSING, LAST_ACK, TIME_WAIT STATES 1094 * Close the tcb. 1095 */ 1096 if (tiflags&TH_RST) switch (tp->t_state) { 1097 1098 case TCPS_SYN_RECEIVED: 1099 so->so_error = ECONNREFUSED; 1100 goto close; 1101 1102 case TCPS_ESTABLISHED: 1103 case TCPS_FIN_WAIT_1: 1104 case TCPS_FIN_WAIT_2: 1105 case TCPS_CLOSE_WAIT: 1106 so->so_error = ECONNRESET; 1107 close: 1108 tp->t_state = TCPS_CLOSED; 1109 tcpstat.tcps_drops++; 1110 tp = tcp_close(tp); 1111 goto drop; 1112 1113 case TCPS_CLOSING: 1114 case TCPS_LAST_ACK: 1115 case TCPS_TIME_WAIT: 1116 tp = tcp_close(tp); 1117 goto drop; 1118 } 1119 1120 /* 1121 * If a SYN is in the window, then this is an 1122 * error and we send an RST and drop the connection. 1123 */ 1124 if (tiflags & TH_SYN) { 1125 tp = tcp_drop(tp, ECONNRESET); 1126 goto dropwithreset; 1127 } 1128 1129 /* 1130 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1131 * flag is on (half-synchronized state), then queue data for 1132 * later processing; else drop segment and return. 1133 */ 1134 if ((tiflags & TH_ACK) == 0) { 1135 if (tp->t_state == TCPS_SYN_RECEIVED || 1136 (tp->t_flags & TF_NEEDSYN)) 1137 goto step6; 1138 else 1139 goto drop; 1140 } 1141 1142 /* 1143 * Ack processing. 1144 */ 1145 switch (tp->t_state) { 1146 1147 /* 1148 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1149 * ESTABLISHED state and continue processing, otherwise 1150 * send an RST. 1151 */ 1152 case TCPS_SYN_RECEIVED: 1153 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1154 SEQ_GT(ti->ti_ack, tp->snd_max)) 1155 goto dropwithreset; 1156 1157 tcpstat.tcps_connects++; 1158 soisconnected(so); 1159 /* Do window scaling? */ 1160 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1161 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1162 tp->snd_scale = tp->requested_s_scale; 1163 tp->rcv_scale = tp->request_r_scale; 1164 } 1165 /* 1166 * Upon successful completion of 3-way handshake, 1167 * update cache.CC if it was undefined, pass any queued 1168 * data to the user, and advance state appropriately. 1169 */ 1170 if ((taop = tcp_gettaocache(inp)) != NULL && 1171 taop->tao_cc == 0) 1172 taop->tao_cc = tp->cc_recv; 1173 1174 /* 1175 * Make transitions: 1176 * SYN-RECEIVED -> ESTABLISHED 1177 * SYN-RECEIVED* -> FIN-WAIT-1 1178 */ 1179 if (tp->t_flags & TF_NEEDFIN) { 1180 tp->t_state = TCPS_FIN_WAIT_1; 1181 tp->t_flags &= ~TF_NEEDFIN; 1182 } else 1183 tp->t_state = TCPS_ESTABLISHED; 1184 /* 1185 * If segment contains data or ACK, will call tcp_reass() 1186 * later; if not, do so now to pass queued data to user. 1187 */ 1188 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1189 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1190 (struct mbuf *)0); 1191 tp->snd_wl1 = ti->ti_seq - 1; 1192 /* fall into ... */ 1193 1194 /* 1195 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1196 * ACKs. If the ack is in the range 1197 * tp->snd_una < ti->ti_ack <= tp->snd_max 1198 * then advance tp->snd_una to ti->ti_ack and drop 1199 * data from the retransmission queue. If this ACK reflects 1200 * more up to date window information we update our window information. 1201 */ 1202 case TCPS_ESTABLISHED: 1203 case TCPS_FIN_WAIT_1: 1204 case TCPS_FIN_WAIT_2: 1205 case TCPS_CLOSE_WAIT: 1206 case TCPS_CLOSING: 1207 case TCPS_LAST_ACK: 1208 case TCPS_TIME_WAIT: 1209 1210 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1211 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1212 tcpstat.tcps_rcvdupack++; 1213 /* 1214 * If we have outstanding data (other than 1215 * a window probe), this is a completely 1216 * duplicate ack (ie, window info didn't 1217 * change), the ack is the biggest we've 1218 * seen and we've seen exactly our rexmt 1219 * threshhold of them, assume a packet 1220 * has been dropped and retransmit it. 1221 * Kludge snd_nxt & the congestion 1222 * window so we send only this one 1223 * packet. 1224 * 1225 * We know we're losing at the current 1226 * window size so do congestion avoidance 1227 * (set ssthresh to half the current window 1228 * and pull our congestion window back to 1229 * the new ssthresh). 1230 * 1231 * Dup acks mean that packets have left the 1232 * network (they're now cached at the receiver) 1233 * so bump cwnd by the amount in the receiver 1234 * to keep a constant cwnd packets in the 1235 * network. 1236 */ 1237 if (tp->t_timer[TCPT_REXMT] == 0 || 1238 ti->ti_ack != tp->snd_una) 1239 tp->t_dupacks = 0; 1240 else if (++tp->t_dupacks == tcprexmtthresh) { 1241 tcp_seq onxt = tp->snd_nxt; 1242 u_int win = 1243 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1244 tp->t_maxseg; 1245 1246 if (win < 2) 1247 win = 2; 1248 tp->snd_ssthresh = win * tp->t_maxseg; 1249 tp->t_timer[TCPT_REXMT] = 0; 1250 tp->t_rtt = 0; 1251 tp->snd_nxt = ti->ti_ack; 1252 tp->snd_cwnd = tp->t_maxseg; 1253 (void) tcp_output(tp); 1254 tp->snd_cwnd = tp->snd_ssthresh + 1255 tp->t_maxseg * tp->t_dupacks; 1256 if (SEQ_GT(onxt, tp->snd_nxt)) 1257 tp->snd_nxt = onxt; 1258 goto drop; 1259 } else if (tp->t_dupacks > tcprexmtthresh) { 1260 tp->snd_cwnd += tp->t_maxseg; 1261 (void) tcp_output(tp); 1262 goto drop; 1263 } 1264 } else 1265 tp->t_dupacks = 0; 1266 break; 1267 } 1268 /* 1269 * If the congestion window was inflated to account 1270 * for the other side's cached packets, retract it. 1271 */ 1272 if (tp->t_dupacks >= tcprexmtthresh && 1273 tp->snd_cwnd > tp->snd_ssthresh) 1274 tp->snd_cwnd = tp->snd_ssthresh; 1275 tp->t_dupacks = 0; 1276 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1277 tcpstat.tcps_rcvacktoomuch++; 1278 goto dropafterack; 1279 } 1280 /* 1281 * If we reach this point, ACK is not a duplicate, 1282 * i.e., it ACKs something we sent. 1283 */ 1284 if (tp->t_flags & TF_NEEDSYN) { 1285 /* 1286 * T/TCP: Connection was half-synchronized, and our 1287 * SYN has been ACK'd (so connection is now fully 1288 * synchronized). Go to non-starred state, 1289 * increment snd_una for ACK of SYN, and check if 1290 * we can do window scaling. 1291 */ 1292 tp->t_flags &= ~TF_NEEDSYN; 1293 tp->snd_una++; 1294 /* Do window scaling? */ 1295 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1296 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1297 tp->snd_scale = tp->requested_s_scale; 1298 tp->rcv_scale = tp->request_r_scale; 1299 } 1300 } 1301 1302process_ACK: 1303 acked = ti->ti_ack - tp->snd_una; 1304 tcpstat.tcps_rcvackpack++; 1305 tcpstat.tcps_rcvackbyte += acked; 1306 1307 /* 1308 * If we have a timestamp reply, update smoothed 1309 * round trip time. If no timestamp is present but 1310 * transmit timer is running and timed sequence 1311 * number was acked, update smoothed round trip time. 1312 * Since we now have an rtt measurement, cancel the 1313 * timer backoff (cf., Phil Karn's retransmit alg.). 1314 * Recompute the initial retransmit timer. 1315 */ 1316 if (to.to_flag & TOF_TS) 1317 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1318 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1319 tcp_xmit_timer(tp,tp->t_rtt); 1320 1321 /* 1322 * If all outstanding data is acked, stop retransmit 1323 * timer and remember to restart (more output or persist). 1324 * If there is more data to be acked, restart retransmit 1325 * timer, using current (possibly backed-off) value. 1326 */ 1327 if (ti->ti_ack == tp->snd_max) { 1328 tp->t_timer[TCPT_REXMT] = 0; 1329 needoutput = 1; 1330 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1331 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1332 1333 /* 1334 * If no data (only SYN) was ACK'd, 1335 * skip rest of ACK processing. 1336 */ 1337 if (acked == 0) 1338 goto step6; 1339 1340 /* 1341 * When new data is acked, open the congestion window. 1342 * If the window gives us less than ssthresh packets 1343 * in flight, open exponentially (maxseg per packet). 1344 * Otherwise open linearly: maxseg per window 1345 * (maxseg^2 / cwnd per packet). 1346 */ 1347 { 1348 register u_int cw = tp->snd_cwnd; 1349 register u_int incr = tp->t_maxseg; 1350 1351 if (cw > tp->snd_ssthresh) 1352 incr = incr * incr / cw; 1353 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1354 } 1355 if (acked > so->so_snd.sb_cc) { 1356 tp->snd_wnd -= so->so_snd.sb_cc; 1357 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1358 ourfinisacked = 1; 1359 } else { 1360 sbdrop(&so->so_snd, acked); 1361 tp->snd_wnd -= acked; 1362 ourfinisacked = 0; 1363 } 1364 if (so->so_snd.sb_flags & SB_NOTIFY) 1365 sowwakeup(so); 1366 tp->snd_una = ti->ti_ack; 1367 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1368 tp->snd_nxt = tp->snd_una; 1369 1370 switch (tp->t_state) { 1371 1372 /* 1373 * In FIN_WAIT_1 STATE in addition to the processing 1374 * for the ESTABLISHED state if our FIN is now acknowledged 1375 * then enter FIN_WAIT_2. 1376 */ 1377 case TCPS_FIN_WAIT_1: 1378 if (ourfinisacked) { 1379 /* 1380 * If we can't receive any more 1381 * data, then closing user can proceed. 1382 * Starting the timer is contrary to the 1383 * specification, but if we don't get a FIN 1384 * we'll hang forever. 1385 */ 1386 if (so->so_state & SS_CANTRCVMORE) { 1387 soisdisconnected(so); 1388 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1389 } 1390 tp->t_state = TCPS_FIN_WAIT_2; 1391 } 1392 break; 1393 1394 /* 1395 * In CLOSING STATE in addition to the processing for 1396 * the ESTABLISHED state if the ACK acknowledges our FIN 1397 * then enter the TIME-WAIT state, otherwise ignore 1398 * the segment. 1399 */ 1400 case TCPS_CLOSING: 1401 if (ourfinisacked) { 1402 tp->t_state = TCPS_TIME_WAIT; 1403 tcp_canceltimers(tp); 1404 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1405 if (tp->cc_recv != 0 && 1406 tp->t_duration < TCPTV_MSL) 1407 tp->t_timer[TCPT_2MSL] = 1408 tp->t_rxtcur * TCPTV_TWTRUNC; 1409 else 1410 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1411 soisdisconnected(so); 1412 } 1413 break; 1414 1415 /* 1416 * In LAST_ACK, we may still be waiting for data to drain 1417 * and/or to be acked, as well as for the ack of our FIN. 1418 * If our FIN is now acknowledged, delete the TCB, 1419 * enter the closed state and return. 1420 */ 1421 case TCPS_LAST_ACK: 1422 if (ourfinisacked) { 1423 tp = tcp_close(tp); 1424 goto drop; 1425 } 1426 break; 1427 1428 /* 1429 * In TIME_WAIT state the only thing that should arrive 1430 * is a retransmission of the remote FIN. Acknowledge 1431 * it and restart the finack timer. 1432 */ 1433 case TCPS_TIME_WAIT: 1434 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1435 goto dropafterack; 1436 } 1437 } 1438 1439step6: 1440 /* 1441 * Update window information. 1442 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1443 */ 1444 if ((tiflags & TH_ACK) && 1445 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1446 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1447 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1448 /* keep track of pure window updates */ 1449 if (ti->ti_len == 0 && 1450 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1451 tcpstat.tcps_rcvwinupd++; 1452 tp->snd_wnd = tiwin; 1453 tp->snd_wl1 = ti->ti_seq; 1454 tp->snd_wl2 = ti->ti_ack; 1455 if (tp->snd_wnd > tp->max_sndwnd) 1456 tp->max_sndwnd = tp->snd_wnd; 1457 needoutput = 1; 1458 } 1459 1460 /* 1461 * Process segments with URG. 1462 */ 1463 if ((tiflags & TH_URG) && ti->ti_urp && 1464 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1465 /* 1466 * This is a kludge, but if we receive and accept 1467 * random urgent pointers, we'll crash in 1468 * soreceive. It's hard to imagine someone 1469 * actually wanting to send this much urgent data. 1470 */ 1471 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1472 ti->ti_urp = 0; /* XXX */ 1473 tiflags &= ~TH_URG; /* XXX */ 1474 goto dodata; /* XXX */ 1475 } 1476 /* 1477 * If this segment advances the known urgent pointer, 1478 * then mark the data stream. This should not happen 1479 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1480 * a FIN has been received from the remote side. 1481 * In these states we ignore the URG. 1482 * 1483 * According to RFC961 (Assigned Protocols), 1484 * the urgent pointer points to the last octet 1485 * of urgent data. We continue, however, 1486 * to consider it to indicate the first octet 1487 * of data past the urgent section as the original 1488 * spec states (in one of two places). 1489 */ 1490 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1491 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1492 so->so_oobmark = so->so_rcv.sb_cc + 1493 (tp->rcv_up - tp->rcv_nxt) - 1; 1494 if (so->so_oobmark == 0) 1495 so->so_state |= SS_RCVATMARK; 1496 sohasoutofband(so); 1497 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1498 } 1499 /* 1500 * Remove out of band data so doesn't get presented to user. 1501 * This can happen independent of advancing the URG pointer, 1502 * but if two URG's are pending at once, some out-of-band 1503 * data may creep in... ick. 1504 */ 1505 if (ti->ti_urp <= (u_long)ti->ti_len 1506#ifdef SO_OOBINLINE 1507 && (so->so_options & SO_OOBINLINE) == 0 1508#endif 1509 ) 1510 tcp_pulloutofband(so, ti, m); 1511 } else 1512 /* 1513 * If no out of band data is expected, 1514 * pull receive urgent pointer along 1515 * with the receive window. 1516 */ 1517 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1518 tp->rcv_up = tp->rcv_nxt; 1519dodata: /* XXX */ 1520 1521 /* 1522 * Process the segment text, merging it into the TCP sequencing queue, 1523 * and arranging for acknowledgment of receipt if necessary. 1524 * This process logically involves adjusting tp->rcv_wnd as data 1525 * is presented to the user (this happens in tcp_usrreq.c, 1526 * case PRU_RCVD). If a FIN has already been received on this 1527 * connection then we just ignore the text. 1528 */ 1529 if ((ti->ti_len || (tiflags&TH_FIN)) && 1530 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1531 TCP_REASS(tp, ti, m, so, tiflags); 1532 /* 1533 * Note the amount of data that peer has sent into 1534 * our window, in order to estimate the sender's 1535 * buffer size. 1536 */ 1537 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1538 } else { 1539 m_freem(m); 1540 tiflags &= ~TH_FIN; 1541 } 1542 1543 /* 1544 * If FIN is received ACK the FIN and let the user know 1545 * that the connection is closing. 1546 */ 1547 if (tiflags & TH_FIN) { 1548 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1549 socantrcvmore(so); 1550 /* 1551 * If connection is half-synchronized 1552 * (ie NEEDSYN flag on) then delay ACK, 1553 * so it may be piggybacked when SYN is sent. 1554 * Otherwise, since we received a FIN then no 1555 * more input can be expected, send ACK now. 1556 */ 1557 if (tp->t_flags & TF_NEEDSYN) 1558 tp->t_flags |= TF_DELACK; 1559 else 1560 tp->t_flags |= TF_ACKNOW; 1561 tp->rcv_nxt++; 1562 } 1563 switch (tp->t_state) { 1564 1565 /* 1566 * In SYN_RECEIVED and ESTABLISHED STATES 1567 * enter the CLOSE_WAIT state. 1568 */ 1569 case TCPS_SYN_RECEIVED: 1570 case TCPS_ESTABLISHED: 1571 tp->t_state = TCPS_CLOSE_WAIT; 1572 break; 1573 1574 /* 1575 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1576 * enter the CLOSING state. 1577 */ 1578 case TCPS_FIN_WAIT_1: 1579 tp->t_state = TCPS_CLOSING; 1580 break; 1581 1582 /* 1583 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1584 * starting the time-wait timer, turning off the other 1585 * standard timers. 1586 */ 1587 case TCPS_FIN_WAIT_2: 1588 tp->t_state = TCPS_TIME_WAIT; 1589 tcp_canceltimers(tp); 1590 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1591 if (tp->cc_recv != 0 && 1592 tp->t_duration < TCPTV_MSL) { 1593 tp->t_timer[TCPT_2MSL] = 1594 tp->t_rxtcur * TCPTV_TWTRUNC; 1595 /* For transaction client, force ACK now. */ 1596 tp->t_flags |= TF_ACKNOW; 1597 } 1598 else 1599 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1600 soisdisconnected(so); 1601 break; 1602 1603 /* 1604 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1605 */ 1606 case TCPS_TIME_WAIT: 1607 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1608 break; 1609 } 1610 } 1611#ifdef TCPDEBUG 1612 if (so->so_options & SO_DEBUG) 1613 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1614#endif 1615 1616 /* 1617 * Return any desired output. 1618 */ 1619 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1620 (void) tcp_output(tp); 1621 return; 1622 1623dropafterack: 1624 /* 1625 * Generate an ACK dropping incoming segment if it occupies 1626 * sequence space, where the ACK reflects our state. 1627 */ 1628 if (tiflags & TH_RST) 1629 goto drop; 1630#ifdef TCPDEBUG 1631 if (so->so_options & SO_DEBUG) 1632 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1633#endif 1634 m_freem(m); 1635 tp->t_flags |= TF_ACKNOW; 1636 (void) tcp_output(tp); 1637 return; 1638 1639dropwithreset: 1640 /* 1641 * Generate a RST, dropping incoming segment. 1642 * Make ACK acceptable to originator of segment. 1643 * Don't bother to respond if destination was broadcast/multicast. 1644 */ 1645 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1646 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1647 goto drop; 1648#ifdef TCPDEBUG 1649 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1650 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1651#endif 1652 if (tiflags & TH_ACK) 1653 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1654 else { 1655 if (tiflags & TH_SYN) 1656 ti->ti_len++; 1657 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1658 TH_RST|TH_ACK); 1659 } 1660 /* destroy temporarily created socket */ 1661 if (dropsocket) 1662 (void) soabort(so); 1663 return; 1664 1665drop: 1666 /* 1667 * Drop space held by incoming segment and return. 1668 */ 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 m_freem(m); 1674 /* destroy temporarily created socket */ 1675 if (dropsocket) 1676 (void) soabort(so); 1677 return; 1678#ifndef TUBA_INCLUDE 1679} 1680 1681static void 1682tcp_dooptions(tp, cp, cnt, ti, to) 1683 struct tcpcb *tp; 1684 u_char *cp; 1685 int cnt; 1686 struct tcpiphdr *ti; 1687 struct tcpopt *to; 1688{ 1689 u_short mss = 0; 1690 int opt, optlen; 1691 1692 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1693 opt = cp[0]; 1694 if (opt == TCPOPT_EOL) 1695 break; 1696 if (opt == TCPOPT_NOP) 1697 optlen = 1; 1698 else { 1699 optlen = cp[1]; 1700 if (optlen <= 0) 1701 break; 1702 } 1703 switch (opt) { 1704 1705 default: 1706 continue; 1707 1708 case TCPOPT_MAXSEG: 1709 if (optlen != TCPOLEN_MAXSEG) 1710 continue; 1711 if (!(ti->ti_flags & TH_SYN)) 1712 continue; 1713 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1714 NTOHS(mss); 1715 break; 1716 1717 case TCPOPT_WINDOW: 1718 if (optlen != TCPOLEN_WINDOW) 1719 continue; 1720 if (!(ti->ti_flags & TH_SYN)) 1721 continue; 1722 tp->t_flags |= TF_RCVD_SCALE; 1723 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1724 break; 1725 1726 case TCPOPT_TIMESTAMP: 1727 if (optlen != TCPOLEN_TIMESTAMP) 1728 continue; 1729 to->to_flag |= TOF_TS; 1730 bcopy((char *)cp + 2, 1731 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1732 NTOHL(to->to_tsval); 1733 bcopy((char *)cp + 6, 1734 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1735 NTOHL(to->to_tsecr); 1736 1737 /* 1738 * A timestamp received in a SYN makes 1739 * it ok to send timestamp requests and replies. 1740 */ 1741 if (ti->ti_flags & TH_SYN) { 1742 tp->t_flags |= TF_RCVD_TSTMP; 1743 tp->ts_recent = to->to_tsval; 1744 tp->ts_recent_age = tcp_now; 1745 } 1746 break; 1747 case TCPOPT_CC: 1748 if (optlen != TCPOLEN_CC) 1749 continue; 1750 to->to_flag |= TOF_CC; 1751 bcopy((char *)cp + 2, 1752 (char *)&to->to_cc, sizeof(to->to_cc)); 1753 NTOHL(to->to_cc); 1754 /* 1755 * A CC or CC.new option received in a SYN makes 1756 * it ok to send CC in subsequent segments. 1757 */ 1758 if (ti->ti_flags & TH_SYN) 1759 tp->t_flags |= TF_RCVD_CC; 1760 break; 1761 case TCPOPT_CCNEW: 1762 if (optlen != TCPOLEN_CC) 1763 continue; 1764 if (!(ti->ti_flags & TH_SYN)) 1765 continue; 1766 to->to_flag |= TOF_CCNEW; 1767 bcopy((char *)cp + 2, 1768 (char *)&to->to_cc, sizeof(to->to_cc)); 1769 NTOHL(to->to_cc); 1770 /* 1771 * A CC or CC.new option received in a SYN makes 1772 * it ok to send CC in subsequent segments. 1773 */ 1774 tp->t_flags |= TF_RCVD_CC; 1775 break; 1776 case TCPOPT_CCECHO: 1777 if (optlen != TCPOLEN_CC) 1778 continue; 1779 if (!(ti->ti_flags & TH_SYN)) 1780 continue; 1781 to->to_flag |= TOF_CCECHO; 1782 bcopy((char *)cp + 2, 1783 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1784 NTOHL(to->to_ccecho); 1785 break; 1786 } 1787 } 1788 if (ti->ti_flags & TH_SYN) 1789 tcp_mss(tp, mss); /* sets t_maxseg */ 1790} 1791 1792/* 1793 * Pull out of band byte out of a segment so 1794 * it doesn't appear in the user's data queue. 1795 * It is still reflected in the segment length for 1796 * sequencing purposes. 1797 */ 1798static void 1799tcp_pulloutofband(so, ti, m) 1800 struct socket *so; 1801 struct tcpiphdr *ti; 1802 register struct mbuf *m; 1803{ 1804 int cnt = ti->ti_urp - 1; 1805 1806 while (cnt >= 0) { 1807 if (m->m_len > cnt) { 1808 char *cp = mtod(m, caddr_t) + cnt; 1809 struct tcpcb *tp = sototcpcb(so); 1810 1811 tp->t_iobc = *cp; 1812 tp->t_oobflags |= TCPOOB_HAVEDATA; 1813 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1814 m->m_len--; 1815 return; 1816 } 1817 cnt -= m->m_len; 1818 m = m->m_next; 1819 if (m == 0) 1820 break; 1821 } 1822 panic("tcp_pulloutofband"); 1823} 1824 1825/* 1826 * Collect new round-trip time estimate 1827 * and update averages and current timeout. 1828 */ 1829static void 1830tcp_xmit_timer(tp, rtt) 1831 register struct tcpcb *tp; 1832 short rtt; 1833{ 1834 register int delta; 1835 1836 tcpstat.tcps_rttupdated++; 1837 tp->t_rttupdated++; 1838 if (tp->t_srtt != 0) { 1839 /* 1840 * srtt is stored as fixed point with 5 bits after the 1841 * binary point (i.e., scaled by 8). The following magic 1842 * is equivalent to the smoothing algorithm in rfc793 with 1843 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1844 * point). Adjust rtt to origin 0. 1845 */ 1846 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1847 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1848 1849 if ((tp->t_srtt += delta) <= 0) 1850 tp->t_srtt = 1; 1851 1852 /* 1853 * We accumulate a smoothed rtt variance (actually, a 1854 * smoothed mean difference), then set the retransmit 1855 * timer to smoothed rtt + 4 times the smoothed variance. 1856 * rttvar is stored as fixed point with 4 bits after the 1857 * binary point (scaled by 16). The following is 1858 * equivalent to rfc793 smoothing with an alpha of .75 1859 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1860 * rfc793's wired-in beta. 1861 */ 1862 if (delta < 0) 1863 delta = -delta; 1864 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1865 if ((tp->t_rttvar += delta) <= 0) 1866 tp->t_rttvar = 1; 1867 } else { 1868 /* 1869 * No rtt measurement yet - use the unsmoothed rtt. 1870 * Set the variance to half the rtt (so our first 1871 * retransmit happens at 3*rtt). 1872 */ 1873 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1874 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1875 } 1876 tp->t_rtt = 0; 1877 tp->t_rxtshift = 0; 1878 1879 /* 1880 * the retransmit should happen at rtt + 4 * rttvar. 1881 * Because of the way we do the smoothing, srtt and rttvar 1882 * will each average +1/2 tick of bias. When we compute 1883 * the retransmit timer, we want 1/2 tick of rounding and 1884 * 1 extra tick because of +-1/2 tick uncertainty in the 1885 * firing of the timer. The bias will give us exactly the 1886 * 1.5 tick we need. But, because the bias is 1887 * statistical, we have to test that we don't drop below 1888 * the minimum feasible timer (which is 2 ticks). 1889 */ 1890 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1891 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1892 1893 /* 1894 * We received an ack for a packet that wasn't retransmitted; 1895 * it is probably safe to discard any error indications we've 1896 * received recently. This isn't quite right, but close enough 1897 * for now (a route might have failed after we sent a segment, 1898 * and the return path might not be symmetrical). 1899 */ 1900 tp->t_softerror = 0; 1901} 1902 1903/* 1904 * Determine a reasonable value for maxseg size. 1905 * If the route is known, check route for mtu. 1906 * If none, use an mss that can be handled on the outgoing 1907 * interface without forcing IP to fragment; if bigger than 1908 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1909 * to utilize large mbufs. If no route is found, route has no mtu, 1910 * or the destination isn't local, use a default, hopefully conservative 1911 * size (usually 512 or the default IP max size, but no more than the mtu 1912 * of the interface), as we can't discover anything about intervening 1913 * gateways or networks. We also initialize the congestion/slow start 1914 * window to be a single segment if the destination isn't local. 1915 * While looking at the routing entry, we also initialize other path-dependent 1916 * parameters from pre-set or cached values in the routing entry. 1917 * 1918 * Also take into account the space needed for options that we 1919 * send regularly. Make maxseg shorter by that amount to assure 1920 * that we can send maxseg amount of data even when the options 1921 * are present. Store the upper limit of the length of options plus 1922 * data in maxopd. 1923 * 1924 * NOTE that this routine is only called when we process an incoming 1925 * segment, for outgoing segments only tcp_mssopt is called. 1926 * 1927 * In case of T/TCP, we call this routine during implicit connection 1928 * setup as well (offer = -1), to initialize maxseg from the cached 1929 * MSS of our peer. 1930 */ 1931void 1932tcp_mss(tp, offer) 1933 struct tcpcb *tp; 1934 int offer; 1935{ 1936 register struct rtentry *rt; 1937 struct ifnet *ifp; 1938 register int rtt, mss; 1939 u_long bufsize; 1940 struct inpcb *inp; 1941 struct socket *so; 1942 struct rmxp_tao *taop; 1943 int origoffer = offer; 1944 1945 inp = tp->t_inpcb; 1946 if ((rt = tcp_rtlookup(inp)) == NULL) { 1947 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1948 return; 1949 } 1950 ifp = rt->rt_ifp; 1951 so = inp->inp_socket; 1952 1953 taop = rmx_taop(rt->rt_rmx); 1954 /* 1955 * Offer == -1 means that we didn't receive SYN yet, 1956 * use cached value in that case; 1957 */ 1958 if (offer == -1) 1959 offer = taop->tao_mssopt; 1960 /* 1961 * Offer == 0 means that there was no MSS on the SYN segment, 1962 * in this case we use tcp_mssdflt. 1963 */ 1964 if (offer == 0) 1965 offer = tcp_mssdflt; 1966 else 1967 /* 1968 * Sanity check: make sure that maxopd will be large 1969 * enough to allow some data on segments even is the 1970 * all the option space is used (40bytes). Otherwise 1971 * funny things may happen in tcp_output. 1972 */ 1973 offer = max(offer, 64); 1974 taop->tao_mssopt = offer; 1975 1976 /* 1977 * While we're here, check if there's an initial rtt 1978 * or rttvar. Convert from the route-table units 1979 * to scaled multiples of the slow timeout timer. 1980 */ 1981 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1982 /* 1983 * XXX the lock bit for RTT indicates that the value 1984 * is also a minimum value; this is subject to time. 1985 */ 1986 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1987 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1988 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1989 tcpstat.tcps_usedrtt++; 1990 if (rt->rt_rmx.rmx_rttvar) { 1991 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1992 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1993 tcpstat.tcps_usedrttvar++; 1994 } else { 1995 /* default variation is +- 1 rtt */ 1996 tp->t_rttvar = 1997 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1998 } 1999 TCPT_RANGESET(tp->t_rxtcur, 2000 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2001 tp->t_rttmin, TCPTV_REXMTMAX); 2002 } 2003 /* 2004 * if there's an mtu associated with the route, use it 2005 */ 2006 if (rt->rt_rmx.rmx_mtu) 2007 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2008 else 2009 { 2010 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2011 if (!in_localaddr(inp->inp_faddr)) 2012 mss = min(mss, tcp_mssdflt); 2013 } 2014 mss = min(mss, offer); 2015 /* 2016 * maxopd stores the maximum length of data AND options 2017 * in a segment; maxseg is the amount of data in a normal 2018 * segment. We need to store this value (maxopd) apart 2019 * from maxseg, because now every segment carries options 2020 * and thus we normally have somewhat less data in segments. 2021 */ 2022 tp->t_maxopd = mss; 2023 2024 /* 2025 * In case of T/TCP, origoffer==-1 indicates, that no segments 2026 * were received yet. In this case we just guess, otherwise 2027 * we do the same as before T/TCP. 2028 */ 2029 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2030 (origoffer == -1 || 2031 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2032 mss -= TCPOLEN_TSTAMP_APPA; 2033 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2034 (origoffer == -1 || 2035 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2036 mss -= TCPOLEN_CC_APPA; 2037 2038#if (MCLBYTES & (MCLBYTES - 1)) == 0 2039 if (mss > MCLBYTES) 2040 mss &= ~(MCLBYTES-1); 2041#else 2042 if (mss > MCLBYTES) 2043 mss = mss / MCLBYTES * MCLBYTES; 2044#endif 2045 /* 2046 * If there's a pipesize, change the socket buffer 2047 * to that size. Make the socket buffers an integral 2048 * number of mss units; if the mss is larger than 2049 * the socket buffer, decrease the mss. 2050 */ 2051#ifdef RTV_SPIPE 2052 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2053#endif 2054 bufsize = so->so_snd.sb_hiwat; 2055 if (bufsize < mss) 2056 mss = bufsize; 2057 else { 2058 bufsize = roundup(bufsize, mss); 2059 if (bufsize > sb_max) 2060 bufsize = sb_max; 2061 (void)sbreserve(&so->so_snd, bufsize); 2062 } 2063 tp->t_maxseg = mss; 2064 2065#ifdef RTV_RPIPE 2066 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2067#endif 2068 bufsize = so->so_rcv.sb_hiwat; 2069 if (bufsize > mss) { 2070 bufsize = roundup(bufsize, mss); 2071 if (bufsize > sb_max) 2072 bufsize = sb_max; 2073 (void)sbreserve(&so->so_rcv, bufsize); 2074 } 2075 /* 2076 * Don't force slow-start on local network. 2077 */ 2078 if (!in_localaddr(inp->inp_faddr)) 2079 tp->snd_cwnd = mss; 2080 2081 if (rt->rt_rmx.rmx_ssthresh) { 2082 /* 2083 * There's some sort of gateway or interface 2084 * buffer limit on the path. Use this to set 2085 * the slow start threshhold, but set the 2086 * threshold to no less than 2*mss. 2087 */ 2088 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2089 tcpstat.tcps_usedssthresh++; 2090 } 2091} 2092 2093/* 2094 * Determine the MSS option to send on an outgoing SYN. 2095 */ 2096int 2097tcp_mssopt(tp) 2098 struct tcpcb *tp; 2099{ 2100 struct rtentry *rt; 2101 2102 rt = tcp_rtlookup(tp->t_inpcb); 2103 if (rt == NULL) 2104 return tcp_mssdflt; 2105 2106 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2107} 2108#endif /* TUBA_INCLUDE */ 2109