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 1.50 1996/09/21 06:30:06 ache 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> 69 static struct tcpiphdr tcp_saveti; 70 #endif 71 72 static int tcprexmtthresh = 3; 73 tcp_seq tcp_iss; 74 tcp_cc tcp_ccgen; 75 76 struct tcpstat tcpstat; 77 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 78 CTLFLAG_RD, &tcpstat , tcpstat, ""); 79 80 static int log_in_vain = 0; 81 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 82 &log_in_vain, 0, ""); 83 84 u_long tcp_now; 85 struct inpcbhead tcb; 86 struct inpcbinfo tcbinfo; 87 88 static void tcp_dooptions __P((struct tcpcb *, 89 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 90 static void tcp_pulloutofband __P((struct socket *, 91 struct tcpiphdr *, struct mbuf *)); 92 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 93 static 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 146 147 static int 148 tcp_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); 229 230 present: 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 */ 259 void 260 tcp_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 */ 359 findpcb: 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 char buf[4*sizeof "123"]; 382 383 strcpy(buf, inet_ntoa(ti->ti_dst)); 384 log(LOG_INFO, "Connection attempt to TCP %s:%d" 385 " from %s:%d\n", 386 buf, ntohs(ti->ti_dport), 387 inet_ntoa(ti->ti_src), ntohs(ti->ti_sport)); 388 } 389 goto dropwithreset; 390 } 391 tp = intotcpcb(inp); 392 if (tp == 0) 393 goto dropwithreset; 394 if (tp->t_state == TCPS_CLOSED) 395 goto drop; 396 397 /* Unscale the window into a 32-bit value. */ 398 if ((tiflags & TH_SYN) == 0) 399 tiwin = ti->ti_win << tp->snd_scale; 400 else 401 tiwin = ti->ti_win; 402 403 so = inp->inp_socket; 404 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 405 #ifdef TCPDEBUG 406 if (so->so_options & SO_DEBUG) { 407 ostate = tp->t_state; 408 tcp_saveti = *ti; 409 } 410 #endif 411 if (so->so_options & SO_ACCEPTCONN) { 412 register struct tcpcb *tp0 = tp; 413 struct socket *so2; 414 /* 415 * If the attempt to get onto the socket queue failed, 416 * drop the oldest queue entry and try again. 417 */ 418 so2 = sonewconn(so, 0); 419 if (!so2) { 420 tcpstat.tcps_listendrop++; 421 so2 = TAILQ_FIRST(&so->so_incomp); 422 if (so2) { 423 tcp_drop(sototcpcb(so2), ETIMEDOUT); 424 so2 = sonewconn(so, 0); 425 } 426 if (!so2) 427 goto drop; 428 } 429 so = so2; 430 /* 431 * This is ugly, but .... 432 * 433 * Mark socket as temporary until we're 434 * committed to keeping it. The code at 435 * ``drop'' and ``dropwithreset'' check the 436 * flag dropsocket to see if the temporary 437 * socket created here should be discarded. 438 * We mark the socket as discardable until 439 * we're committed to it below in TCPS_LISTEN. 440 */ 441 dropsocket++; 442 inp = (struct inpcb *)so->so_pcb; 443 inp->inp_laddr = ti->ti_dst; 444 inp->inp_lport = ti->ti_dport; 445 in_pcbrehash(inp); 446 #if BSD>=43 447 inp->inp_options = ip_srcroute(); 448 #endif 449 tp = intotcpcb(inp); 450 tp->t_state = TCPS_LISTEN; 451 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 452 453 /* Compute proper scaling value from buffer space */ 454 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 455 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 456 tp->request_r_scale++; 457 } 458 } 459 460 /* 461 * Segment received on connection. 462 * Reset idle time and keep-alive timer. 463 */ 464 tp->t_idle = 0; 465 if (TCPS_HAVEESTABLISHED(tp->t_state)) 466 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 467 468 /* 469 * Process options if not in LISTEN state, 470 * else do it below (after getting remote address). 471 */ 472 if (tp->t_state != TCPS_LISTEN) 473 tcp_dooptions(tp, optp, optlen, ti, &to); 474 475 /* 476 * Header prediction: check for the two common cases 477 * of a uni-directional data xfer. If the packet has 478 * no control flags, is in-sequence, the window didn't 479 * change and we're not retransmitting, it's a 480 * candidate. If the length is zero and the ack moved 481 * forward, we're the sender side of the xfer. Just 482 * free the data acked & wake any higher level process 483 * that was blocked waiting for space. If the length 484 * is non-zero and the ack didn't move, we're the 485 * receiver side. If we're getting packets in-order 486 * (the reassembly queue is empty), add the data to 487 * the socket buffer and note that we need a delayed ack. 488 * Make sure that the hidden state-flags are also off. 489 * Since we check for TCPS_ESTABLISHED above, it can only 490 * be TH_NEEDSYN. 491 */ 492 if (tp->t_state == TCPS_ESTABLISHED && 493 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 494 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 495 ((to.to_flag & TOF_TS) == 0 || 496 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 497 /* 498 * Using the CC option is compulsory if once started: 499 * the segment is OK if no T/TCP was negotiated or 500 * if the segment has a CC option equal to CCrecv 501 */ 502 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 503 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 504 ti->ti_seq == tp->rcv_nxt && 505 tiwin && tiwin == tp->snd_wnd && 506 tp->snd_nxt == tp->snd_max) { 507 508 /* 509 * If last ACK falls within this segment's sequence numbers, 510 * record the timestamp. 511 * NOTE that the test is modified according to the latest 512 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 513 */ 514 if ((to.to_flag & TOF_TS) != 0 && 515 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 516 tp->ts_recent_age = tcp_now; 517 tp->ts_recent = to.to_tsval; 518 } 519 520 if (ti->ti_len == 0) { 521 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 522 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 523 tp->snd_cwnd >= tp->snd_wnd && 524 tp->t_dupacks < tcprexmtthresh) { 525 /* 526 * this is a pure ack for outstanding data. 527 */ 528 ++tcpstat.tcps_predack; 529 if ((to.to_flag & TOF_TS) != 0) 530 tcp_xmit_timer(tp, 531 tcp_now - to.to_tsecr + 1); 532 else if (tp->t_rtt && 533 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 534 tcp_xmit_timer(tp, tp->t_rtt); 535 acked = ti->ti_ack - tp->snd_una; 536 tcpstat.tcps_rcvackpack++; 537 tcpstat.tcps_rcvackbyte += acked; 538 sbdrop(&so->so_snd, acked); 539 tp->snd_una = ti->ti_ack; 540 m_freem(m); 541 542 /* 543 * If all outstanding data are acked, stop 544 * retransmit timer, otherwise restart timer 545 * using current (possibly backed-off) value. 546 * If process is waiting for space, 547 * wakeup/selwakeup/signal. If data 548 * are ready to send, let tcp_output 549 * decide between more output or persist. 550 */ 551 if (tp->snd_una == tp->snd_max) 552 tp->t_timer[TCPT_REXMT] = 0; 553 else if (tp->t_timer[TCPT_PERSIST] == 0) 554 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 555 556 if (so->so_snd.sb_flags & SB_NOTIFY) 557 sowwakeup(so); 558 if (so->so_snd.sb_cc) 559 (void) tcp_output(tp); 560 return; 561 } 562 } else if (ti->ti_ack == tp->snd_una && 563 tp->seg_next == (struct tcpiphdr *)tp && 564 ti->ti_len <= sbspace(&so->so_rcv)) { 565 /* 566 * this is a pure, in-sequence data packet 567 * with nothing on the reassembly queue and 568 * we have enough buffer space to take it. 569 */ 570 ++tcpstat.tcps_preddat; 571 tp->rcv_nxt += ti->ti_len; 572 tcpstat.tcps_rcvpack++; 573 tcpstat.tcps_rcvbyte += ti->ti_len; 574 /* 575 * Add data to socket buffer. 576 */ 577 sbappend(&so->so_rcv, m); 578 sorwakeup(so); 579 #ifdef TCP_ACK_HACK 580 /* 581 * If this is a short packet, then ACK now - with Nagel 582 * congestion avoidance sender won't send more until 583 * he gets an ACK. 584 */ 585 if (tiflags & TH_PUSH) { 586 tp->t_flags |= TF_ACKNOW; 587 tcp_output(tp); 588 } else { 589 tp->t_flags |= TF_DELACK; 590 } 591 #else 592 tp->t_flags |= TF_DELACK; 593 #endif 594 return; 595 } 596 } 597 598 /* 599 * Calculate amount of space in receive window, 600 * and then do TCP input processing. 601 * Receive window is amount of space in rcv queue, 602 * but not less than advertised window. 603 */ 604 { int win; 605 606 win = sbspace(&so->so_rcv); 607 if (win < 0) 608 win = 0; 609 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 610 } 611 612 switch (tp->t_state) { 613 614 /* 615 * If the state is LISTEN then ignore segment if it contains an RST. 616 * If the segment contains an ACK then it is bad and send a RST. 617 * If it does not contain a SYN then it is not interesting; drop it. 618 * Don't bother responding if the destination was a broadcast. 619 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 620 * tp->iss, and send a segment: 621 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 622 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 623 * Fill in remote peer address fields if not previously specified. 624 * Enter SYN_RECEIVED state, and process any other fields of this 625 * segment in this state. 626 */ 627 case TCPS_LISTEN: { 628 struct mbuf *am; 629 register struct sockaddr_in *sin; 630 631 if (tiflags & TH_RST) 632 goto drop; 633 if (tiflags & TH_ACK) 634 goto dropwithreset; 635 if ((tiflags & TH_SYN) == 0) 636 goto drop; 637 /* 638 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 639 * in_broadcast() should never return true on a received 640 * packet with M_BCAST not set. 641 */ 642 if (m->m_flags & (M_BCAST|M_MCAST) || 643 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 644 goto drop; 645 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 646 if (am == NULL) 647 goto drop; 648 am->m_len = sizeof (struct sockaddr_in); 649 sin = mtod(am, struct sockaddr_in *); 650 sin->sin_family = AF_INET; 651 sin->sin_len = sizeof(*sin); 652 sin->sin_addr = ti->ti_src; 653 sin->sin_port = ti->ti_sport; 654 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 655 laddr = inp->inp_laddr; 656 if (inp->inp_laddr.s_addr == INADDR_ANY) 657 inp->inp_laddr = ti->ti_dst; 658 if (in_pcbconnect(inp, am)) { 659 inp->inp_laddr = laddr; 660 (void) m_free(am); 661 goto drop; 662 } 663 (void) m_free(am); 664 tp->t_template = tcp_template(tp); 665 if (tp->t_template == 0) { 666 tp = tcp_drop(tp, ENOBUFS); 667 dropsocket = 0; /* socket is already gone */ 668 goto drop; 669 } 670 if ((taop = tcp_gettaocache(inp)) == NULL) { 671 taop = &tao_noncached; 672 bzero(taop, sizeof(*taop)); 673 } 674 tcp_dooptions(tp, optp, optlen, ti, &to); 675 if (iss) 676 tp->iss = iss; 677 else 678 tp->iss = tcp_iss; 679 tcp_iss += TCP_ISSINCR/4; 680 tp->irs = ti->ti_seq; 681 tcp_sendseqinit(tp); 682 tcp_rcvseqinit(tp); 683 /* 684 * Initialization of the tcpcb for transaction; 685 * set SND.WND = SEG.WND, 686 * initialize CCsend and CCrecv. 687 */ 688 tp->snd_wnd = tiwin; /* initial send-window */ 689 tp->cc_send = CC_INC(tcp_ccgen); 690 tp->cc_recv = to.to_cc; 691 /* 692 * Perform TAO test on incoming CC (SEG.CC) option, if any. 693 * - compare SEG.CC against cached CC from the same host, 694 * if any. 695 * - if SEG.CC > chached value, SYN must be new and is accepted 696 * immediately: save new CC in the cache, mark the socket 697 * connected, enter ESTABLISHED state, turn on flag to 698 * send a SYN in the next segment. 699 * A virtual advertised window is set in rcv_adv to 700 * initialize SWS prevention. Then enter normal segment 701 * processing: drop SYN, process data and FIN. 702 * - otherwise do a normal 3-way handshake. 703 */ 704 if ((to.to_flag & TOF_CC) != 0) { 705 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 706 taop->tao_cc = to.to_cc; 707 tp->t_state = TCPS_ESTABLISHED; 708 709 /* 710 * If there is a FIN, or if there is data and the 711 * connection is local, then delay SYN,ACK(SYN) in 712 * the hope of piggy-backing it on a response 713 * segment. Otherwise must send ACK now in case 714 * the other side is slow starting. 715 */ 716 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 717 in_localaddr(inp->inp_faddr))) 718 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 719 else 720 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 721 722 /* 723 * Limit the `virtual advertised window' to TCP_MAXWIN 724 * here. Even if we requested window scaling, it will 725 * become effective only later when our SYN is acked. 726 */ 727 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 728 tcpstat.tcps_connects++; 729 soisconnected(so); 730 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 731 dropsocket = 0; /* committed to socket */ 732 tcpstat.tcps_accepts++; 733 goto trimthenstep6; 734 } 735 /* else do standard 3-way handshake */ 736 } else { 737 /* 738 * No CC option, but maybe CC.NEW: 739 * invalidate cached value. 740 */ 741 taop->tao_cc = 0; 742 } 743 /* 744 * TAO test failed or there was no CC option, 745 * do a standard 3-way handshake. 746 */ 747 tp->t_flags |= TF_ACKNOW; 748 tp->t_state = TCPS_SYN_RECEIVED; 749 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 750 dropsocket = 0; /* committed to socket */ 751 tcpstat.tcps_accepts++; 752 goto trimthenstep6; 753 } 754 755 /* 756 * If the state is SYN_SENT: 757 * if seg contains an ACK, but not for our SYN, drop the input. 758 * if seg contains a RST, then drop the connection. 759 * if seg does not contain SYN, then drop it. 760 * Otherwise this is an acceptable SYN segment 761 * initialize tp->rcv_nxt and tp->irs 762 * if seg contains ack then advance tp->snd_una 763 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 764 * arrange for segment to be acked (eventually) 765 * continue processing rest of data/controls, beginning with URG 766 */ 767 case TCPS_SYN_SENT: 768 if ((taop = tcp_gettaocache(inp)) == NULL) { 769 taop = &tao_noncached; 770 bzero(taop, sizeof(*taop)); 771 } 772 773 if ((tiflags & TH_ACK) && 774 (SEQ_LEQ(ti->ti_ack, tp->iss) || 775 SEQ_GT(ti->ti_ack, tp->snd_max))) { 776 /* 777 * If we have a cached CCsent for the remote host, 778 * hence we haven't just crashed and restarted, 779 * do not send a RST. This may be a retransmission 780 * from the other side after our earlier ACK was lost. 781 * Our new SYN, when it arrives, will serve as the 782 * needed ACK. 783 */ 784 if (taop->tao_ccsent != 0) 785 goto drop; 786 else 787 goto dropwithreset; 788 } 789 if (tiflags & TH_RST) { 790 if (tiflags & TH_ACK) 791 tp = tcp_drop(tp, ECONNREFUSED); 792 goto drop; 793 } 794 if ((tiflags & TH_SYN) == 0) 795 goto drop; 796 tp->snd_wnd = ti->ti_win; /* initial send window */ 797 tp->cc_recv = to.to_cc; /* foreign CC */ 798 799 tp->irs = ti->ti_seq; 800 tcp_rcvseqinit(tp); 801 if (tiflags & TH_ACK) { 802 /* 803 * Our SYN was acked. If segment contains CC.ECHO 804 * option, check it to make sure this segment really 805 * matches our SYN. If not, just drop it as old 806 * duplicate, but send an RST if we're still playing 807 * by the old rules. 808 */ 809 if ((to.to_flag & TOF_CCECHO) && 810 tp->cc_send != to.to_ccecho) { 811 if (taop->tao_ccsent != 0) 812 goto drop; 813 else 814 goto dropwithreset; 815 } 816 tcpstat.tcps_connects++; 817 soisconnected(so); 818 /* Do window scaling on this connection? */ 819 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 820 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 821 tp->snd_scale = tp->requested_s_scale; 822 tp->rcv_scale = tp->request_r_scale; 823 } 824 /* Segment is acceptable, update cache if undefined. */ 825 if (taop->tao_ccsent == 0) 826 taop->tao_ccsent = to.to_ccecho; 827 828 tp->rcv_adv += tp->rcv_wnd; 829 tp->snd_una++; /* SYN is acked */ 830 /* 831 * If there's data, delay ACK; if there's also a FIN 832 * ACKNOW will be turned on later. 833 */ 834 if (ti->ti_len != 0) 835 tp->t_flags |= TF_DELACK; 836 else 837 tp->t_flags |= TF_ACKNOW; 838 /* 839 * Received <SYN,ACK> in SYN_SENT[*] state. 840 * Transitions: 841 * SYN_SENT --> ESTABLISHED 842 * SYN_SENT* --> FIN_WAIT_1 843 */ 844 if (tp->t_flags & TF_NEEDFIN) { 845 tp->t_state = TCPS_FIN_WAIT_1; 846 tp->t_flags &= ~TF_NEEDFIN; 847 tiflags &= ~TH_SYN; 848 } else { 849 tp->t_state = TCPS_ESTABLISHED; 850 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 851 } 852 } else { 853 /* 854 * Received initial SYN in SYN-SENT[*] state => simul- 855 * taneous open. If segment contains CC option and there is 856 * a cached CC, apply TAO test; if it succeeds, connection is 857 * half-synchronized. Otherwise, do 3-way handshake: 858 * SYN-SENT -> SYN-RECEIVED 859 * SYN-SENT* -> SYN-RECEIVED* 860 * If there was no CC option, clear cached CC value. 861 */ 862 tp->t_flags |= TF_ACKNOW; 863 tp->t_timer[TCPT_REXMT] = 0; 864 if (to.to_flag & TOF_CC) { 865 if (taop->tao_cc != 0 && 866 CC_GT(to.to_cc, taop->tao_cc)) { 867 /* 868 * update cache and make transition: 869 * SYN-SENT -> ESTABLISHED* 870 * SYN-SENT* -> FIN-WAIT-1* 871 */ 872 taop->tao_cc = to.to_cc; 873 if (tp->t_flags & TF_NEEDFIN) { 874 tp->t_state = TCPS_FIN_WAIT_1; 875 tp->t_flags &= ~TF_NEEDFIN; 876 } else { 877 tp->t_state = TCPS_ESTABLISHED; 878 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 879 } 880 tp->t_flags |= TF_NEEDSYN; 881 } else 882 tp->t_state = TCPS_SYN_RECEIVED; 883 } else { 884 /* CC.NEW or no option => invalidate cache */ 885 taop->tao_cc = 0; 886 tp->t_state = TCPS_SYN_RECEIVED; 887 } 888 } 889 890 trimthenstep6: 891 /* 892 * Advance ti->ti_seq to correspond to first data byte. 893 * If data, trim to stay within window, 894 * dropping FIN if necessary. 895 */ 896 ti->ti_seq++; 897 if (ti->ti_len > tp->rcv_wnd) { 898 todrop = ti->ti_len - tp->rcv_wnd; 899 m_adj(m, -todrop); 900 ti->ti_len = tp->rcv_wnd; 901 tiflags &= ~TH_FIN; 902 tcpstat.tcps_rcvpackafterwin++; 903 tcpstat.tcps_rcvbyteafterwin += todrop; 904 } 905 tp->snd_wl1 = ti->ti_seq - 1; 906 tp->rcv_up = ti->ti_seq; 907 /* 908 * Client side of transaction: already sent SYN and data. 909 * If the remote host used T/TCP to validate the SYN, 910 * our data will be ACK'd; if so, enter normal data segment 911 * processing in the middle of step 5, ack processing. 912 * Otherwise, goto step 6. 913 */ 914 if (tiflags & TH_ACK) 915 goto process_ACK; 916 goto step6; 917 /* 918 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 919 * if segment contains a SYN and CC [not CC.NEW] option: 920 * if state == TIME_WAIT and connection duration > MSL, 921 * drop packet and send RST; 922 * 923 * if SEG.CC > CCrecv then is new SYN, and can implicitly 924 * ack the FIN (and data) in retransmission queue. 925 * Complete close and delete TCPCB. Then reprocess 926 * segment, hoping to find new TCPCB in LISTEN state; 927 * 928 * else must be old SYN; drop it. 929 * else do normal processing. 930 */ 931 case TCPS_LAST_ACK: 932 case TCPS_CLOSING: 933 case TCPS_TIME_WAIT: 934 if ((tiflags & TH_SYN) && 935 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 936 if (tp->t_state == TCPS_TIME_WAIT && 937 tp->t_duration > TCPTV_MSL) 938 goto dropwithreset; 939 if (CC_GT(to.to_cc, tp->cc_recv)) { 940 tp = tcp_close(tp); 941 goto findpcb; 942 } 943 else 944 goto drop; 945 } 946 break; /* continue normal processing */ 947 } 948 949 /* 950 * States other than LISTEN or SYN_SENT. 951 * First check timestamp, if present. 952 * Then check the connection count, if present. 953 * Then check that at least some bytes of segment are within 954 * receive window. If segment begins before rcv_nxt, 955 * drop leading data (and SYN); if nothing left, just ack. 956 * 957 * RFC 1323 PAWS: If we have a timestamp reply on this segment 958 * and it's less than ts_recent, drop it. 959 */ 960 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 961 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 962 963 /* Check to see if ts_recent is over 24 days old. */ 964 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 965 /* 966 * Invalidate ts_recent. If this segment updates 967 * ts_recent, the age will be reset later and ts_recent 968 * will get a valid value. If it does not, setting 969 * ts_recent to zero will at least satisfy the 970 * requirement that zero be placed in the timestamp 971 * echo reply when ts_recent isn't valid. The 972 * age isn't reset until we get a valid ts_recent 973 * because we don't want out-of-order segments to be 974 * dropped when ts_recent is old. 975 */ 976 tp->ts_recent = 0; 977 } else { 978 tcpstat.tcps_rcvduppack++; 979 tcpstat.tcps_rcvdupbyte += ti->ti_len; 980 tcpstat.tcps_pawsdrop++; 981 goto dropafterack; 982 } 983 } 984 985 /* 986 * T/TCP mechanism 987 * If T/TCP was negotiated and the segment doesn't have CC, 988 * or if it's CC is wrong then drop the segment. 989 * RST segments do not have to comply with this. 990 */ 991 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 992 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 993 (tiflags & TH_RST) == 0) 994 goto dropafterack; 995 996 todrop = tp->rcv_nxt - ti->ti_seq; 997 if (todrop > 0) { 998 if (tiflags & TH_SYN) { 999 tiflags &= ~TH_SYN; 1000 ti->ti_seq++; 1001 if (ti->ti_urp > 1) 1002 ti->ti_urp--; 1003 else 1004 tiflags &= ~TH_URG; 1005 todrop--; 1006 } 1007 /* 1008 * Following if statement from Stevens, vol. 2, p. 960. 1009 */ 1010 if (todrop > ti->ti_len 1011 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1012 /* 1013 * Any valid FIN must be to the left of the window. 1014 * At this point the FIN must be a duplicate or out 1015 * of sequence; drop it. 1016 */ 1017 tiflags &= ~TH_FIN; 1018 1019 /* 1020 * Send an ACK to resynchronize and drop any data. 1021 * But keep on processing for RST or ACK. 1022 */ 1023 tp->t_flags |= TF_ACKNOW; 1024 todrop = ti->ti_len; 1025 tcpstat.tcps_rcvduppack++; 1026 tcpstat.tcps_rcvdupbyte += todrop; 1027 } else { 1028 tcpstat.tcps_rcvpartduppack++; 1029 tcpstat.tcps_rcvpartdupbyte += todrop; 1030 } 1031 m_adj(m, todrop); 1032 ti->ti_seq += todrop; 1033 ti->ti_len -= todrop; 1034 if (ti->ti_urp > todrop) 1035 ti->ti_urp -= todrop; 1036 else { 1037 tiflags &= ~TH_URG; 1038 ti->ti_urp = 0; 1039 } 1040 } 1041 1042 /* 1043 * If new data are received on a connection after the 1044 * user processes are gone, then RST the other end. 1045 */ 1046 if ((so->so_state & SS_NOFDREF) && 1047 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1048 tp = tcp_close(tp); 1049 tcpstat.tcps_rcvafterclose++; 1050 goto dropwithreset; 1051 } 1052 1053 /* 1054 * If segment ends after window, drop trailing data 1055 * (and PUSH and FIN); if nothing left, just ACK. 1056 */ 1057 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1058 if (todrop > 0) { 1059 tcpstat.tcps_rcvpackafterwin++; 1060 if (todrop >= ti->ti_len) { 1061 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1062 /* 1063 * If a new connection request is received 1064 * while in TIME_WAIT, drop the old connection 1065 * and start over if the sequence numbers 1066 * are above the previous ones. 1067 */ 1068 if (tiflags & TH_SYN && 1069 tp->t_state == TCPS_TIME_WAIT && 1070 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1071 iss = tp->rcv_nxt + TCP_ISSINCR; 1072 tp = tcp_close(tp); 1073 goto findpcb; 1074 } 1075 /* 1076 * If window is closed can only take segments at 1077 * window edge, and have to drop data and PUSH from 1078 * incoming segments. Continue processing, but 1079 * remember to ack. Otherwise, drop segment 1080 * and ack. 1081 */ 1082 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1083 tp->t_flags |= TF_ACKNOW; 1084 tcpstat.tcps_rcvwinprobe++; 1085 } else 1086 goto dropafterack; 1087 } else 1088 tcpstat.tcps_rcvbyteafterwin += todrop; 1089 m_adj(m, -todrop); 1090 ti->ti_len -= todrop; 1091 tiflags &= ~(TH_PUSH|TH_FIN); 1092 } 1093 1094 /* 1095 * If last ACK falls within this segment's sequence numbers, 1096 * record its timestamp. 1097 * NOTE that the test is modified according to the latest 1098 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1099 */ 1100 if ((to.to_flag & TOF_TS) != 0 && 1101 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1102 tp->ts_recent_age = tcp_now; 1103 tp->ts_recent = to.to_tsval; 1104 } 1105 1106 /* 1107 * If the RST bit is set examine the state: 1108 * SYN_RECEIVED STATE: 1109 * If passive open, return to LISTEN state. 1110 * If active open, inform user that connection was refused. 1111 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1112 * Inform user that connection was reset, and close tcb. 1113 * CLOSING, LAST_ACK, TIME_WAIT STATES 1114 * Close the tcb. 1115 */ 1116 if (tiflags&TH_RST) switch (tp->t_state) { 1117 1118 case TCPS_SYN_RECEIVED: 1119 so->so_error = ECONNREFUSED; 1120 goto close; 1121 1122 case TCPS_ESTABLISHED: 1123 case TCPS_FIN_WAIT_1: 1124 case TCPS_FIN_WAIT_2: 1125 case TCPS_CLOSE_WAIT: 1126 so->so_error = ECONNRESET; 1127 close: 1128 tp->t_state = TCPS_CLOSED; 1129 tcpstat.tcps_drops++; 1130 tp = tcp_close(tp); 1131 goto drop; 1132 1133 case TCPS_CLOSING: 1134 case TCPS_LAST_ACK: 1135 case TCPS_TIME_WAIT: 1136 tp = tcp_close(tp); 1137 goto drop; 1138 } 1139 1140 /* 1141 * If a SYN is in the window, then this is an 1142 * error and we send an RST and drop the connection. 1143 */ 1144 if (tiflags & TH_SYN) { 1145 tp = tcp_drop(tp, ECONNRESET); 1146 goto dropwithreset; 1147 } 1148 1149 /* 1150 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1151 * flag is on (half-synchronized state), then queue data for 1152 * later processing; else drop segment and return. 1153 */ 1154 if ((tiflags & TH_ACK) == 0) { 1155 if (tp->t_state == TCPS_SYN_RECEIVED || 1156 (tp->t_flags & TF_NEEDSYN)) 1157 goto step6; 1158 else 1159 goto drop; 1160 } 1161 1162 /* 1163 * Ack processing. 1164 */ 1165 switch (tp->t_state) { 1166 1167 /* 1168 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1169 * ESTABLISHED state and continue processing, otherwise 1170 * send an RST. 1171 */ 1172 case TCPS_SYN_RECEIVED: 1173 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1174 SEQ_GT(ti->ti_ack, tp->snd_max)) 1175 goto dropwithreset; 1176 1177 tcpstat.tcps_connects++; 1178 soisconnected(so); 1179 /* Do window scaling? */ 1180 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1181 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1182 tp->snd_scale = tp->requested_s_scale; 1183 tp->rcv_scale = tp->request_r_scale; 1184 } 1185 /* 1186 * Upon successful completion of 3-way handshake, 1187 * update cache.CC if it was undefined, pass any queued 1188 * data to the user, and advance state appropriately. 1189 */ 1190 if ((taop = tcp_gettaocache(inp)) != NULL && 1191 taop->tao_cc == 0) 1192 taop->tao_cc = tp->cc_recv; 1193 1194 /* 1195 * Make transitions: 1196 * SYN-RECEIVED -> ESTABLISHED 1197 * SYN-RECEIVED* -> FIN-WAIT-1 1198 */ 1199 if (tp->t_flags & TF_NEEDFIN) { 1200 tp->t_state = TCPS_FIN_WAIT_1; 1201 tp->t_flags &= ~TF_NEEDFIN; 1202 } else { 1203 tp->t_state = TCPS_ESTABLISHED; 1204 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1205 } 1206 /* 1207 * If segment contains data or ACK, will call tcp_reass() 1208 * later; if not, do so now to pass queued data to user. 1209 */ 1210 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1211 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1212 (struct mbuf *)0); 1213 tp->snd_wl1 = ti->ti_seq - 1; 1214 /* fall into ... */ 1215 1216 /* 1217 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1218 * ACKs. If the ack is in the range 1219 * tp->snd_una < ti->ti_ack <= tp->snd_max 1220 * then advance tp->snd_una to ti->ti_ack and drop 1221 * data from the retransmission queue. If this ACK reflects 1222 * more up to date window information we update our window information. 1223 */ 1224 case TCPS_ESTABLISHED: 1225 case TCPS_FIN_WAIT_1: 1226 case TCPS_FIN_WAIT_2: 1227 case TCPS_CLOSE_WAIT: 1228 case TCPS_CLOSING: 1229 case TCPS_LAST_ACK: 1230 case TCPS_TIME_WAIT: 1231 1232 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1233 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1234 tcpstat.tcps_rcvdupack++; 1235 /* 1236 * If we have outstanding data (other than 1237 * a window probe), this is a completely 1238 * duplicate ack (ie, window info didn't 1239 * change), the ack is the biggest we've 1240 * seen and we've seen exactly our rexmt 1241 * threshhold of them, assume a packet 1242 * has been dropped and retransmit it. 1243 * Kludge snd_nxt & the congestion 1244 * window so we send only this one 1245 * packet. 1246 * 1247 * We know we're losing at the current 1248 * window size so do congestion avoidance 1249 * (set ssthresh to half the current window 1250 * and pull our congestion window back to 1251 * the new ssthresh). 1252 * 1253 * Dup acks mean that packets have left the 1254 * network (they're now cached at the receiver) 1255 * so bump cwnd by the amount in the receiver 1256 * to keep a constant cwnd packets in the 1257 * network. 1258 */ 1259 if (tp->t_timer[TCPT_REXMT] == 0 || 1260 ti->ti_ack != tp->snd_una) 1261 tp->t_dupacks = 0; 1262 else if (++tp->t_dupacks == tcprexmtthresh) { 1263 tcp_seq onxt = tp->snd_nxt; 1264 u_int win = 1265 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1266 tp->t_maxseg; 1267 1268 if (win < 2) 1269 win = 2; 1270 tp->snd_ssthresh = win * tp->t_maxseg; 1271 tp->t_timer[TCPT_REXMT] = 0; 1272 tp->t_rtt = 0; 1273 tp->snd_nxt = ti->ti_ack; 1274 tp->snd_cwnd = tp->t_maxseg; 1275 (void) tcp_output(tp); 1276 tp->snd_cwnd = tp->snd_ssthresh + 1277 tp->t_maxseg * tp->t_dupacks; 1278 if (SEQ_GT(onxt, tp->snd_nxt)) 1279 tp->snd_nxt = onxt; 1280 goto drop; 1281 } else if (tp->t_dupacks > tcprexmtthresh) { 1282 tp->snd_cwnd += tp->t_maxseg; 1283 (void) tcp_output(tp); 1284 goto drop; 1285 } 1286 } else 1287 tp->t_dupacks = 0; 1288 break; 1289 } 1290 /* 1291 * If the congestion window was inflated to account 1292 * for the other side's cached packets, retract it. 1293 */ 1294 if (tp->t_dupacks >= tcprexmtthresh && 1295 tp->snd_cwnd > tp->snd_ssthresh) 1296 tp->snd_cwnd = tp->snd_ssthresh; 1297 tp->t_dupacks = 0; 1298 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1299 tcpstat.tcps_rcvacktoomuch++; 1300 goto dropafterack; 1301 } 1302 /* 1303 * If we reach this point, ACK is not a duplicate, 1304 * i.e., it ACKs something we sent. 1305 */ 1306 if (tp->t_flags & TF_NEEDSYN) { 1307 /* 1308 * T/TCP: Connection was half-synchronized, and our 1309 * SYN has been ACK'd (so connection is now fully 1310 * synchronized). Go to non-starred state, 1311 * increment snd_una for ACK of SYN, and check if 1312 * we can do window scaling. 1313 */ 1314 tp->t_flags &= ~TF_NEEDSYN; 1315 tp->snd_una++; 1316 /* Do window scaling? */ 1317 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1318 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1319 tp->snd_scale = tp->requested_s_scale; 1320 tp->rcv_scale = tp->request_r_scale; 1321 } 1322 } 1323 1324 process_ACK: 1325 acked = ti->ti_ack - tp->snd_una; 1326 tcpstat.tcps_rcvackpack++; 1327 tcpstat.tcps_rcvackbyte += acked; 1328 1329 /* 1330 * If we have a timestamp reply, update smoothed 1331 * round trip time. If no timestamp is present but 1332 * transmit timer is running and timed sequence 1333 * number was acked, update smoothed round trip time. 1334 * Since we now have an rtt measurement, cancel the 1335 * timer backoff (cf., Phil Karn's retransmit alg.). 1336 * Recompute the initial retransmit timer. 1337 */ 1338 if (to.to_flag & TOF_TS) 1339 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1340 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1341 tcp_xmit_timer(tp,tp->t_rtt); 1342 1343 /* 1344 * If all outstanding data is acked, stop retransmit 1345 * timer and remember to restart (more output or persist). 1346 * If there is more data to be acked, restart retransmit 1347 * timer, using current (possibly backed-off) value. 1348 */ 1349 if (ti->ti_ack == tp->snd_max) { 1350 tp->t_timer[TCPT_REXMT] = 0; 1351 needoutput = 1; 1352 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1353 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1354 1355 /* 1356 * If no data (only SYN) was ACK'd, 1357 * skip rest of ACK processing. 1358 */ 1359 if (acked == 0) 1360 goto step6; 1361 1362 /* 1363 * When new data is acked, open the congestion window. 1364 * If the window gives us less than ssthresh packets 1365 * in flight, open exponentially (maxseg per packet). 1366 * Otherwise open linearly: maxseg per window 1367 * (maxseg^2 / cwnd per packet). 1368 */ 1369 { 1370 register u_int cw = tp->snd_cwnd; 1371 register u_int incr = tp->t_maxseg; 1372 1373 if (cw > tp->snd_ssthresh) 1374 incr = incr * incr / cw; 1375 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1376 } 1377 if (acked > so->so_snd.sb_cc) { 1378 tp->snd_wnd -= so->so_snd.sb_cc; 1379 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1380 ourfinisacked = 1; 1381 } else { 1382 sbdrop(&so->so_snd, acked); 1383 tp->snd_wnd -= acked; 1384 ourfinisacked = 0; 1385 } 1386 if (so->so_snd.sb_flags & SB_NOTIFY) 1387 sowwakeup(so); 1388 tp->snd_una = ti->ti_ack; 1389 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1390 tp->snd_nxt = tp->snd_una; 1391 1392 switch (tp->t_state) { 1393 1394 /* 1395 * In FIN_WAIT_1 STATE in addition to the processing 1396 * for the ESTABLISHED state if our FIN is now acknowledged 1397 * then enter FIN_WAIT_2. 1398 */ 1399 case TCPS_FIN_WAIT_1: 1400 if (ourfinisacked) { 1401 /* 1402 * If we can't receive any more 1403 * data, then closing user can proceed. 1404 * Starting the timer is contrary to the 1405 * specification, but if we don't get a FIN 1406 * we'll hang forever. 1407 */ 1408 if (so->so_state & SS_CANTRCVMORE) { 1409 soisdisconnected(so); 1410 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1411 } 1412 tp->t_state = TCPS_FIN_WAIT_2; 1413 } 1414 break; 1415 1416 /* 1417 * In CLOSING STATE in addition to the processing for 1418 * the ESTABLISHED state if the ACK acknowledges our FIN 1419 * then enter the TIME-WAIT state, otherwise ignore 1420 * the segment. 1421 */ 1422 case TCPS_CLOSING: 1423 if (ourfinisacked) { 1424 tp->t_state = TCPS_TIME_WAIT; 1425 tcp_canceltimers(tp); 1426 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1427 if (tp->cc_recv != 0 && 1428 tp->t_duration < TCPTV_MSL) 1429 tp->t_timer[TCPT_2MSL] = 1430 tp->t_rxtcur * TCPTV_TWTRUNC; 1431 else 1432 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1433 soisdisconnected(so); 1434 } 1435 break; 1436 1437 /* 1438 * In LAST_ACK, we may still be waiting for data to drain 1439 * and/or to be acked, as well as for the ack of our FIN. 1440 * If our FIN is now acknowledged, delete the TCB, 1441 * enter the closed state and return. 1442 */ 1443 case TCPS_LAST_ACK: 1444 if (ourfinisacked) { 1445 tp = tcp_close(tp); 1446 goto drop; 1447 } 1448 break; 1449 1450 /* 1451 * In TIME_WAIT state the only thing that should arrive 1452 * is a retransmission of the remote FIN. Acknowledge 1453 * it and restart the finack timer. 1454 */ 1455 case TCPS_TIME_WAIT: 1456 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1457 goto dropafterack; 1458 } 1459 } 1460 1461 step6: 1462 /* 1463 * Update window information. 1464 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1465 */ 1466 if ((tiflags & TH_ACK) && 1467 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1468 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1469 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1470 /* keep track of pure window updates */ 1471 if (ti->ti_len == 0 && 1472 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1473 tcpstat.tcps_rcvwinupd++; 1474 tp->snd_wnd = tiwin; 1475 tp->snd_wl1 = ti->ti_seq; 1476 tp->snd_wl2 = ti->ti_ack; 1477 if (tp->snd_wnd > tp->max_sndwnd) 1478 tp->max_sndwnd = tp->snd_wnd; 1479 needoutput = 1; 1480 } 1481 1482 /* 1483 * Process segments with URG. 1484 */ 1485 if ((tiflags & TH_URG) && ti->ti_urp && 1486 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1487 /* 1488 * This is a kludge, but if we receive and accept 1489 * random urgent pointers, we'll crash in 1490 * soreceive. It's hard to imagine someone 1491 * actually wanting to send this much urgent data. 1492 */ 1493 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1494 ti->ti_urp = 0; /* XXX */ 1495 tiflags &= ~TH_URG; /* XXX */ 1496 goto dodata; /* XXX */ 1497 } 1498 /* 1499 * If this segment advances the known urgent pointer, 1500 * then mark the data stream. This should not happen 1501 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1502 * a FIN has been received from the remote side. 1503 * In these states we ignore the URG. 1504 * 1505 * According to RFC961 (Assigned Protocols), 1506 * the urgent pointer points to the last octet 1507 * of urgent data. We continue, however, 1508 * to consider it to indicate the first octet 1509 * of data past the urgent section as the original 1510 * spec states (in one of two places). 1511 */ 1512 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1513 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1514 so->so_oobmark = so->so_rcv.sb_cc + 1515 (tp->rcv_up - tp->rcv_nxt) - 1; 1516 if (so->so_oobmark == 0) 1517 so->so_state |= SS_RCVATMARK; 1518 sohasoutofband(so); 1519 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1520 } 1521 /* 1522 * Remove out of band data so doesn't get presented to user. 1523 * This can happen independent of advancing the URG pointer, 1524 * but if two URG's are pending at once, some out-of-band 1525 * data may creep in... ick. 1526 */ 1527 if (ti->ti_urp <= (u_long)ti->ti_len 1528 #ifdef SO_OOBINLINE 1529 && (so->so_options & SO_OOBINLINE) == 0 1530 #endif 1531 ) 1532 tcp_pulloutofband(so, ti, m); 1533 } else 1534 /* 1535 * If no out of band data is expected, 1536 * pull receive urgent pointer along 1537 * with the receive window. 1538 */ 1539 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1540 tp->rcv_up = tp->rcv_nxt; 1541 dodata: /* XXX */ 1542 1543 /* 1544 * Process the segment text, merging it into the TCP sequencing queue, 1545 * and arranging for acknowledgment of receipt if necessary. 1546 * This process logically involves adjusting tp->rcv_wnd as data 1547 * is presented to the user (this happens in tcp_usrreq.c, 1548 * case PRU_RCVD). If a FIN has already been received on this 1549 * connection then we just ignore the text. 1550 */ 1551 if ((ti->ti_len || (tiflags&TH_FIN)) && 1552 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1553 TCP_REASS(tp, ti, m, so, tiflags); 1554 /* 1555 * Note the amount of data that peer has sent into 1556 * our window, in order to estimate the sender's 1557 * buffer size. 1558 */ 1559 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1560 } else { 1561 m_freem(m); 1562 tiflags &= ~TH_FIN; 1563 } 1564 1565 /* 1566 * If FIN is received ACK the FIN and let the user know 1567 * that the connection is closing. 1568 */ 1569 if (tiflags & TH_FIN) { 1570 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1571 socantrcvmore(so); 1572 /* 1573 * If connection is half-synchronized 1574 * (ie NEEDSYN flag on) then delay ACK, 1575 * so it may be piggybacked when SYN is sent. 1576 * Otherwise, since we received a FIN then no 1577 * more input can be expected, send ACK now. 1578 */ 1579 if (tp->t_flags & TF_NEEDSYN) 1580 tp->t_flags |= TF_DELACK; 1581 else 1582 tp->t_flags |= TF_ACKNOW; 1583 tp->rcv_nxt++; 1584 } 1585 switch (tp->t_state) { 1586 1587 /* 1588 * In SYN_RECEIVED and ESTABLISHED STATES 1589 * enter the CLOSE_WAIT state. 1590 */ 1591 case TCPS_SYN_RECEIVED: 1592 case TCPS_ESTABLISHED: 1593 tp->t_state = TCPS_CLOSE_WAIT; 1594 break; 1595 1596 /* 1597 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1598 * enter the CLOSING state. 1599 */ 1600 case TCPS_FIN_WAIT_1: 1601 tp->t_state = TCPS_CLOSING; 1602 break; 1603 1604 /* 1605 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1606 * starting the time-wait timer, turning off the other 1607 * standard timers. 1608 */ 1609 case TCPS_FIN_WAIT_2: 1610 tp->t_state = TCPS_TIME_WAIT; 1611 tcp_canceltimers(tp); 1612 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1613 if (tp->cc_recv != 0 && 1614 tp->t_duration < TCPTV_MSL) { 1615 tp->t_timer[TCPT_2MSL] = 1616 tp->t_rxtcur * TCPTV_TWTRUNC; 1617 /* For transaction client, force ACK now. */ 1618 tp->t_flags |= TF_ACKNOW; 1619 } 1620 else 1621 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1622 soisdisconnected(so); 1623 break; 1624 1625 /* 1626 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1627 */ 1628 case TCPS_TIME_WAIT: 1629 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1630 break; 1631 } 1632 } 1633 #ifdef TCPDEBUG 1634 if (so->so_options & SO_DEBUG) 1635 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1636 #endif 1637 1638 /* 1639 * Return any desired output. 1640 */ 1641 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1642 (void) tcp_output(tp); 1643 return; 1644 1645 dropafterack: 1646 /* 1647 * Generate an ACK dropping incoming segment if it occupies 1648 * sequence space, where the ACK reflects our state. 1649 */ 1650 if (tiflags & TH_RST) 1651 goto drop; 1652 #ifdef TCPDEBUG 1653 if (so->so_options & SO_DEBUG) 1654 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1655 #endif 1656 m_freem(m); 1657 tp->t_flags |= TF_ACKNOW; 1658 (void) tcp_output(tp); 1659 return; 1660 1661 dropwithreset: 1662 /* 1663 * Generate a RST, dropping incoming segment. 1664 * Make ACK acceptable to originator of segment. 1665 * Don't bother to respond if destination was broadcast/multicast. 1666 */ 1667 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1668 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1669 goto drop; 1670 #ifdef TCPDEBUG 1671 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1672 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1673 #endif 1674 if (tiflags & TH_ACK) 1675 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1676 else { 1677 if (tiflags & TH_SYN) 1678 ti->ti_len++; 1679 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1680 TH_RST|TH_ACK); 1681 } 1682 /* destroy temporarily created socket */ 1683 if (dropsocket) 1684 (void) soabort(so); 1685 return; 1686 1687 drop: 1688 /* 1689 * Drop space held by incoming segment and return. 1690 */ 1691 #ifdef TCPDEBUG 1692 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1693 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1694 #endif 1695 m_freem(m); 1696 /* destroy temporarily created socket */ 1697 if (dropsocket) 1698 (void) soabort(so); 1699 return; 1700 #ifndef TUBA_INCLUDE 1701 } 1702 1703 static void 1704 tcp_dooptions(tp, cp, cnt, ti, to) 1705 struct tcpcb *tp; 1706 u_char *cp; 1707 int cnt; 1708 struct tcpiphdr *ti; 1709 struct tcpopt *to; 1710 { 1711 u_short mss = 0; 1712 int opt, optlen; 1713 1714 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1715 opt = cp[0]; 1716 if (opt == TCPOPT_EOL) 1717 break; 1718 if (opt == TCPOPT_NOP) 1719 optlen = 1; 1720 else { 1721 optlen = cp[1]; 1722 if (optlen <= 0) 1723 break; 1724 } 1725 switch (opt) { 1726 1727 default: 1728 continue; 1729 1730 case TCPOPT_MAXSEG: 1731 if (optlen != TCPOLEN_MAXSEG) 1732 continue; 1733 if (!(ti->ti_flags & TH_SYN)) 1734 continue; 1735 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1736 NTOHS(mss); 1737 break; 1738 1739 case TCPOPT_WINDOW: 1740 if (optlen != TCPOLEN_WINDOW) 1741 continue; 1742 if (!(ti->ti_flags & TH_SYN)) 1743 continue; 1744 tp->t_flags |= TF_RCVD_SCALE; 1745 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1746 break; 1747 1748 case TCPOPT_TIMESTAMP: 1749 if (optlen != TCPOLEN_TIMESTAMP) 1750 continue; 1751 to->to_flag |= TOF_TS; 1752 bcopy((char *)cp + 2, 1753 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1754 NTOHL(to->to_tsval); 1755 bcopy((char *)cp + 6, 1756 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1757 NTOHL(to->to_tsecr); 1758 1759 /* 1760 * A timestamp received in a SYN makes 1761 * it ok to send timestamp requests and replies. 1762 */ 1763 if (ti->ti_flags & TH_SYN) { 1764 tp->t_flags |= TF_RCVD_TSTMP; 1765 tp->ts_recent = to->to_tsval; 1766 tp->ts_recent_age = tcp_now; 1767 } 1768 break; 1769 case TCPOPT_CC: 1770 if (optlen != TCPOLEN_CC) 1771 continue; 1772 to->to_flag |= TOF_CC; 1773 bcopy((char *)cp + 2, 1774 (char *)&to->to_cc, sizeof(to->to_cc)); 1775 NTOHL(to->to_cc); 1776 /* 1777 * A CC or CC.new option received in a SYN makes 1778 * it ok to send CC in subsequent segments. 1779 */ 1780 if (ti->ti_flags & TH_SYN) 1781 tp->t_flags |= TF_RCVD_CC; 1782 break; 1783 case TCPOPT_CCNEW: 1784 if (optlen != TCPOLEN_CC) 1785 continue; 1786 if (!(ti->ti_flags & TH_SYN)) 1787 continue; 1788 to->to_flag |= TOF_CCNEW; 1789 bcopy((char *)cp + 2, 1790 (char *)&to->to_cc, sizeof(to->to_cc)); 1791 NTOHL(to->to_cc); 1792 /* 1793 * A CC or CC.new option received in a SYN makes 1794 * it ok to send CC in subsequent segments. 1795 */ 1796 tp->t_flags |= TF_RCVD_CC; 1797 break; 1798 case TCPOPT_CCECHO: 1799 if (optlen != TCPOLEN_CC) 1800 continue; 1801 if (!(ti->ti_flags & TH_SYN)) 1802 continue; 1803 to->to_flag |= TOF_CCECHO; 1804 bcopy((char *)cp + 2, 1805 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1806 NTOHL(to->to_ccecho); 1807 break; 1808 } 1809 } 1810 if (ti->ti_flags & TH_SYN) 1811 tcp_mss(tp, mss); /* sets t_maxseg */ 1812 } 1813 1814 /* 1815 * Pull out of band byte out of a segment so 1816 * it doesn't appear in the user's data queue. 1817 * It is still reflected in the segment length for 1818 * sequencing purposes. 1819 */ 1820 static void 1821 tcp_pulloutofband(so, ti, m) 1822 struct socket *so; 1823 struct tcpiphdr *ti; 1824 register struct mbuf *m; 1825 { 1826 int cnt = ti->ti_urp - 1; 1827 1828 while (cnt >= 0) { 1829 if (m->m_len > cnt) { 1830 char *cp = mtod(m, caddr_t) + cnt; 1831 struct tcpcb *tp = sototcpcb(so); 1832 1833 tp->t_iobc = *cp; 1834 tp->t_oobflags |= TCPOOB_HAVEDATA; 1835 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1836 m->m_len--; 1837 return; 1838 } 1839 cnt -= m->m_len; 1840 m = m->m_next; 1841 if (m == 0) 1842 break; 1843 } 1844 panic("tcp_pulloutofband"); 1845 } 1846 1847 /* 1848 * Collect new round-trip time estimate 1849 * and update averages and current timeout. 1850 */ 1851 static void 1852 tcp_xmit_timer(tp, rtt) 1853 register struct tcpcb *tp; 1854 short rtt; 1855 { 1856 register int delta; 1857 1858 tcpstat.tcps_rttupdated++; 1859 tp->t_rttupdated++; 1860 if (tp->t_srtt != 0) { 1861 /* 1862 * srtt is stored as fixed point with 5 bits after the 1863 * binary point (i.e., scaled by 8). The following magic 1864 * is equivalent to the smoothing algorithm in rfc793 with 1865 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1866 * point). Adjust rtt to origin 0. 1867 */ 1868 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1869 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1870 1871 if ((tp->t_srtt += delta) <= 0) 1872 tp->t_srtt = 1; 1873 1874 /* 1875 * We accumulate a smoothed rtt variance (actually, a 1876 * smoothed mean difference), then set the retransmit 1877 * timer to smoothed rtt + 4 times the smoothed variance. 1878 * rttvar is stored as fixed point with 4 bits after the 1879 * binary point (scaled by 16). The following is 1880 * equivalent to rfc793 smoothing with an alpha of .75 1881 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1882 * rfc793's wired-in beta. 1883 */ 1884 if (delta < 0) 1885 delta = -delta; 1886 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1887 if ((tp->t_rttvar += delta) <= 0) 1888 tp->t_rttvar = 1; 1889 } else { 1890 /* 1891 * No rtt measurement yet - use the unsmoothed rtt. 1892 * Set the variance to half the rtt (so our first 1893 * retransmit happens at 3*rtt). 1894 */ 1895 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1896 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1897 } 1898 tp->t_rtt = 0; 1899 tp->t_rxtshift = 0; 1900 1901 /* 1902 * the retransmit should happen at rtt + 4 * rttvar. 1903 * Because of the way we do the smoothing, srtt and rttvar 1904 * will each average +1/2 tick of bias. When we compute 1905 * the retransmit timer, we want 1/2 tick of rounding and 1906 * 1 extra tick because of +-1/2 tick uncertainty in the 1907 * firing of the timer. The bias will give us exactly the 1908 * 1.5 tick we need. But, because the bias is 1909 * statistical, we have to test that we don't drop below 1910 * the minimum feasible timer (which is 2 ticks). 1911 */ 1912 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1913 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1914 1915 /* 1916 * We received an ack for a packet that wasn't retransmitted; 1917 * it is probably safe to discard any error indications we've 1918 * received recently. This isn't quite right, but close enough 1919 * for now (a route might have failed after we sent a segment, 1920 * and the return path might not be symmetrical). 1921 */ 1922 tp->t_softerror = 0; 1923 } 1924 1925 /* 1926 * Determine a reasonable value for maxseg size. 1927 * If the route is known, check route for mtu. 1928 * If none, use an mss that can be handled on the outgoing 1929 * interface without forcing IP to fragment; if bigger than 1930 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1931 * to utilize large mbufs. If no route is found, route has no mtu, 1932 * or the destination isn't local, use a default, hopefully conservative 1933 * size (usually 512 or the default IP max size, but no more than the mtu 1934 * of the interface), as we can't discover anything about intervening 1935 * gateways or networks. We also initialize the congestion/slow start 1936 * window to be a single segment if the destination isn't local. 1937 * While looking at the routing entry, we also initialize other path-dependent 1938 * parameters from pre-set or cached values in the routing entry. 1939 * 1940 * Also take into account the space needed for options that we 1941 * send regularly. Make maxseg shorter by that amount to assure 1942 * that we can send maxseg amount of data even when the options 1943 * are present. Store the upper limit of the length of options plus 1944 * data in maxopd. 1945 * 1946 * NOTE that this routine is only called when we process an incoming 1947 * segment, for outgoing segments only tcp_mssopt is called. 1948 * 1949 * In case of T/TCP, we call this routine during implicit connection 1950 * setup as well (offer = -1), to initialize maxseg from the cached 1951 * MSS of our peer. 1952 */ 1953 void 1954 tcp_mss(tp, offer) 1955 struct tcpcb *tp; 1956 int offer; 1957 { 1958 register struct rtentry *rt; 1959 struct ifnet *ifp; 1960 register int rtt, mss; 1961 u_long bufsize; 1962 struct inpcb *inp; 1963 struct socket *so; 1964 struct rmxp_tao *taop; 1965 int origoffer = offer; 1966 1967 inp = tp->t_inpcb; 1968 if ((rt = tcp_rtlookup(inp)) == NULL) { 1969 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1970 return; 1971 } 1972 ifp = rt->rt_ifp; 1973 so = inp->inp_socket; 1974 1975 taop = rmx_taop(rt->rt_rmx); 1976 /* 1977 * Offer == -1 means that we didn't receive SYN yet, 1978 * use cached value in that case; 1979 */ 1980 if (offer == -1) 1981 offer = taop->tao_mssopt; 1982 /* 1983 * Offer == 0 means that there was no MSS on the SYN segment, 1984 * in this case we use tcp_mssdflt. 1985 */ 1986 if (offer == 0) 1987 offer = tcp_mssdflt; 1988 else 1989 /* 1990 * Sanity check: make sure that maxopd will be large 1991 * enough to allow some data on segments even is the 1992 * all the option space is used (40bytes). Otherwise 1993 * funny things may happen in tcp_output. 1994 */ 1995 offer = max(offer, 64); 1996 taop->tao_mssopt = offer; 1997 1998 /* 1999 * While we're here, check if there's an initial rtt 2000 * or rttvar. Convert from the route-table units 2001 * to scaled multiples of the slow timeout timer. 2002 */ 2003 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2004 /* 2005 * XXX the lock bit for RTT indicates that the value 2006 * is also a minimum value; this is subject to time. 2007 */ 2008 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2009 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2010 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2011 tcpstat.tcps_usedrtt++; 2012 if (rt->rt_rmx.rmx_rttvar) { 2013 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2014 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2015 tcpstat.tcps_usedrttvar++; 2016 } else { 2017 /* default variation is +- 1 rtt */ 2018 tp->t_rttvar = 2019 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2020 } 2021 TCPT_RANGESET(tp->t_rxtcur, 2022 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2023 tp->t_rttmin, TCPTV_REXMTMAX); 2024 } 2025 /* 2026 * if there's an mtu associated with the route, use it 2027 */ 2028 if (rt->rt_rmx.rmx_mtu) 2029 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2030 else 2031 { 2032 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2033 if (!in_localaddr(inp->inp_faddr)) 2034 mss = min(mss, tcp_mssdflt); 2035 } 2036 mss = min(mss, offer); 2037 /* 2038 * maxopd stores the maximum length of data AND options 2039 * in a segment; maxseg is the amount of data in a normal 2040 * segment. We need to store this value (maxopd) apart 2041 * from maxseg, because now every segment carries options 2042 * and thus we normally have somewhat less data in segments. 2043 */ 2044 tp->t_maxopd = mss; 2045 2046 /* 2047 * In case of T/TCP, origoffer==-1 indicates, that no segments 2048 * were received yet. In this case we just guess, otherwise 2049 * we do the same as before T/TCP. 2050 */ 2051 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2052 (origoffer == -1 || 2053 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2054 mss -= TCPOLEN_TSTAMP_APPA; 2055 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2056 (origoffer == -1 || 2057 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2058 mss -= TCPOLEN_CC_APPA; 2059 2060 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2061 if (mss > MCLBYTES) 2062 mss &= ~(MCLBYTES-1); 2063 #else 2064 if (mss > MCLBYTES) 2065 mss = mss / MCLBYTES * MCLBYTES; 2066 #endif 2067 /* 2068 * If there's a pipesize, change the socket buffer 2069 * to that size. Make the socket buffers an integral 2070 * number of mss units; if the mss is larger than 2071 * the socket buffer, decrease the mss. 2072 */ 2073 #ifdef RTV_SPIPE 2074 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2075 #endif 2076 bufsize = so->so_snd.sb_hiwat; 2077 if (bufsize < mss) 2078 mss = bufsize; 2079 else { 2080 bufsize = roundup(bufsize, mss); 2081 if (bufsize > sb_max) 2082 bufsize = sb_max; 2083 (void)sbreserve(&so->so_snd, bufsize); 2084 } 2085 tp->t_maxseg = mss; 2086 2087 #ifdef RTV_RPIPE 2088 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2089 #endif 2090 bufsize = so->so_rcv.sb_hiwat; 2091 if (bufsize > mss) { 2092 bufsize = roundup(bufsize, mss); 2093 if (bufsize > sb_max) 2094 bufsize = sb_max; 2095 (void)sbreserve(&so->so_rcv, bufsize); 2096 } 2097 /* 2098 * Don't force slow-start on local network. 2099 */ 2100 if (!in_localaddr(inp->inp_faddr)) 2101 tp->snd_cwnd = mss; 2102 2103 if (rt->rt_rmx.rmx_ssthresh) { 2104 /* 2105 * There's some sort of gateway or interface 2106 * buffer limit on the path. Use this to set 2107 * the slow start threshhold, but set the 2108 * threshold to no less than 2*mss. 2109 */ 2110 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2111 tcpstat.tcps_usedssthresh++; 2112 } 2113 } 2114 2115 /* 2116 * Determine the MSS option to send on an outgoing SYN. 2117 */ 2118 int 2119 tcp_mssopt(tp) 2120 struct tcpcb *tp; 2121 { 2122 struct rtentry *rt; 2123 2124 rt = tcp_rtlookup(tp->t_inpcb); 2125 if (rt == NULL) 2126 return tcp_mssdflt; 2127 2128 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2129 } 2130 #endif /* TUBA_INCLUDE */ 2131