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 1.36 1996/02/22 11:46:39 davidg Exp $ 35 */ 36 37 #ifndef TUBA_INCLUDE 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/kernel.h> 41 #include <sys/sysctl.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/errno.h> 48 #include <sys/queue.h> 49 50 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 51 52 #include <net/if.h> 53 #include <net/route.h> 54 55 #include <netinet/in.h> 56 #include <netinet/in_systm.h> 57 #include <netinet/ip.h> 58 #include <netinet/in_pcb.h> 59 #include <netinet/ip_var.h> 60 #include <netinet/tcp.h> 61 #include <netinet/tcp_fsm.h> 62 #include <netinet/tcp_seq.h> 63 #include <netinet/tcp_timer.h> 64 #include <netinet/tcp_var.h> 65 #include <netinet/tcpip.h> 66 #ifdef TCPDEBUG 67 #include <netinet/tcp_debug.h> 68 static struct tcpiphdr tcp_saveti; 69 #endif 70 71 static int tcprexmtthresh = 3; 72 tcp_seq tcp_iss; 73 tcp_cc tcp_ccgen; 74 75 struct tcpstat tcpstat; 76 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 77 CTLFLAG_RD, &tcpstat , tcpstat, ""); 78 79 u_long tcp_now; 80 struct inpcbhead tcb; 81 struct inpcbinfo tcbinfo; 82 83 static void tcp_dooptions __P((struct tcpcb *, 84 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 85 static void tcp_pulloutofband __P((struct socket *, 86 struct tcpiphdr *, struct mbuf *)); 87 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 88 static void tcp_xmit_timer __P((struct tcpcb *, int)); 89 90 #endif /* TUBA_INCLUDE */ 91 92 /* 93 * Insert segment ti into reassembly queue of tcp with 94 * control block tp. Return TH_FIN if reassembly now includes 95 * a segment with FIN. The macro form does the common case inline 96 * (segment is the next to be received on an established connection, 97 * and the queue is empty), avoiding linkage into and removal 98 * from the queue and repetition of various conversions. 99 * Set DELACK for segments received in order, but ack immediately 100 * when segments are out of order (so fast retransmit can work). 101 */ 102 #ifdef TCP_ACK_HACK 103 #define TCP_REASS(tp, ti, m, so, flags) { \ 104 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 105 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 106 (tp)->t_state == TCPS_ESTABLISHED) { \ 107 if (ti->ti_flags & TH_PUSH) \ 108 tp->t_flags |= TF_ACKNOW; \ 109 else \ 110 tp->t_flags |= TF_DELACK; \ 111 (tp)->rcv_nxt += (ti)->ti_len; \ 112 flags = (ti)->ti_flags & TH_FIN; \ 113 tcpstat.tcps_rcvpack++;\ 114 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 115 sbappend(&(so)->so_rcv, (m)); \ 116 sorwakeup(so); \ 117 } else { \ 118 (flags) = tcp_reass((tp), (ti), (m)); \ 119 tp->t_flags |= TF_ACKNOW; \ 120 } \ 121 } 122 #else 123 #define TCP_REASS(tp, ti, m, so, flags) { \ 124 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 125 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 126 (tp)->t_state == TCPS_ESTABLISHED) { \ 127 tp->t_flags |= TF_DELACK; \ 128 (tp)->rcv_nxt += (ti)->ti_len; \ 129 flags = (ti)->ti_flags & TH_FIN; \ 130 tcpstat.tcps_rcvpack++;\ 131 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 132 sbappend(&(so)->so_rcv, (m)); \ 133 sorwakeup(so); \ 134 } else { \ 135 (flags) = tcp_reass((tp), (ti), (m)); \ 136 tp->t_flags |= TF_ACKNOW; \ 137 } \ 138 } 139 #endif 140 #ifndef TUBA_INCLUDE 141 142 static int 143 tcp_reass(tp, ti, m) 144 register struct tcpcb *tp; 145 register struct tcpiphdr *ti; 146 struct mbuf *m; 147 { 148 register struct tcpiphdr *q; 149 struct socket *so = tp->t_inpcb->inp_socket; 150 int flags; 151 152 /* 153 * Call with ti==0 after become established to 154 * force pre-ESTABLISHED data up to user socket. 155 */ 156 if (ti == 0) 157 goto present; 158 159 /* 160 * Find a segment which begins after this one does. 161 */ 162 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 163 q = (struct tcpiphdr *)q->ti_next) 164 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 165 break; 166 167 /* 168 * If there is a preceding segment, it may provide some of 169 * our data already. If so, drop the data from the incoming 170 * segment. If it provides all of our data, drop us. 171 */ 172 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 173 register int i; 174 q = (struct tcpiphdr *)q->ti_prev; 175 /* conversion to int (in i) handles seq wraparound */ 176 i = q->ti_seq + q->ti_len - ti->ti_seq; 177 if (i > 0) { 178 if (i >= ti->ti_len) { 179 tcpstat.tcps_rcvduppack++; 180 tcpstat.tcps_rcvdupbyte += ti->ti_len; 181 m_freem(m); 182 /* 183 * Try to present any queued data 184 * at the left window edge to the user. 185 * This is needed after the 3-WHS 186 * completes. 187 */ 188 goto present; /* ??? */ 189 } 190 m_adj(m, i); 191 ti->ti_len -= i; 192 ti->ti_seq += i; 193 } 194 q = (struct tcpiphdr *)(q->ti_next); 195 } 196 tcpstat.tcps_rcvoopack++; 197 tcpstat.tcps_rcvoobyte += ti->ti_len; 198 REASS_MBUF(ti) = m; /* XXX */ 199 200 /* 201 * While we overlap succeeding segments trim them or, 202 * if they are completely covered, dequeue them. 203 */ 204 while (q != (struct tcpiphdr *)tp) { 205 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 206 if (i <= 0) 207 break; 208 if (i < q->ti_len) { 209 q->ti_seq += i; 210 q->ti_len -= i; 211 m_adj(REASS_MBUF(q), i); 212 break; 213 } 214 q = (struct tcpiphdr *)q->ti_next; 215 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 216 remque(q->ti_prev); 217 m_freem(m); 218 } 219 220 /* 221 * Stick new segment in its place. 222 */ 223 insque(ti, q->ti_prev); 224 225 present: 226 /* 227 * Present data to user, advancing rcv_nxt through 228 * completed sequence space. 229 */ 230 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 231 return (0); 232 ti = tp->seg_next; 233 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 234 return (0); 235 do { 236 tp->rcv_nxt += ti->ti_len; 237 flags = ti->ti_flags & TH_FIN; 238 remque(ti); 239 m = REASS_MBUF(ti); 240 ti = (struct tcpiphdr *)ti->ti_next; 241 if (so->so_state & SS_CANTRCVMORE) 242 m_freem(m); 243 else 244 sbappend(&so->so_rcv, m); 245 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 246 sorwakeup(so); 247 return (flags); 248 } 249 250 /* 251 * TCP input routine, follows pages 65-76 of the 252 * protocol specification dated September, 1981 very closely. 253 */ 254 void 255 tcp_input(m, iphlen) 256 register struct mbuf *m; 257 int iphlen; 258 { 259 register struct tcpiphdr *ti; 260 register struct inpcb *inp; 261 u_char *optp = NULL; 262 int optlen = 0; 263 int len, tlen, off; 264 register struct tcpcb *tp = 0; 265 register int tiflags; 266 struct socket *so = 0; 267 int todrop, acked, ourfinisacked, needoutput = 0; 268 struct in_addr laddr; 269 int dropsocket = 0; 270 int iss = 0; 271 u_long tiwin; 272 struct tcpopt to; /* options in this segment */ 273 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 274 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 275 #ifdef TCPDEBUG 276 short ostate = 0; 277 #endif 278 279 bzero((char *)&to, sizeof(to)); 280 281 tcpstat.tcps_rcvtotal++; 282 /* 283 * Get IP and TCP header together in first mbuf. 284 * Note: IP leaves IP header in first mbuf. 285 */ 286 ti = mtod(m, struct tcpiphdr *); 287 if (iphlen > sizeof (struct ip)) 288 ip_stripoptions(m, (struct mbuf *)0); 289 if (m->m_len < sizeof (struct tcpiphdr)) { 290 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 291 tcpstat.tcps_rcvshort++; 292 return; 293 } 294 ti = mtod(m, struct tcpiphdr *); 295 } 296 297 /* 298 * Checksum extended TCP header and data. 299 */ 300 tlen = ((struct ip *)ti)->ip_len; 301 len = sizeof (struct ip) + tlen; 302 ti->ti_next = ti->ti_prev = 0; 303 ti->ti_x1 = 0; 304 ti->ti_len = (u_short)tlen; 305 HTONS(ti->ti_len); 306 ti->ti_sum = in_cksum(m, len); 307 if (ti->ti_sum) { 308 tcpstat.tcps_rcvbadsum++; 309 goto drop; 310 } 311 #endif /* TUBA_INCLUDE */ 312 313 /* 314 * Check that TCP offset makes sense, 315 * pull out TCP options and adjust length. XXX 316 */ 317 off = ti->ti_off << 2; 318 if (off < sizeof (struct tcphdr) || off > tlen) { 319 tcpstat.tcps_rcvbadoff++; 320 goto drop; 321 } 322 tlen -= off; 323 ti->ti_len = tlen; 324 if (off > sizeof (struct tcphdr)) { 325 if (m->m_len < sizeof(struct ip) + off) { 326 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 327 tcpstat.tcps_rcvshort++; 328 return; 329 } 330 ti = mtod(m, struct tcpiphdr *); 331 } 332 optlen = off - sizeof (struct tcphdr); 333 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 334 } 335 tiflags = ti->ti_flags; 336 337 /* 338 * Convert TCP protocol specific fields to host format. 339 */ 340 NTOHL(ti->ti_seq); 341 NTOHL(ti->ti_ack); 342 NTOHS(ti->ti_win); 343 NTOHS(ti->ti_urp); 344 345 /* 346 * Drop TCP, IP headers and TCP options. 347 */ 348 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 349 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 350 351 /* 352 * Locate pcb for segment. 353 */ 354 findpcb: 355 /* 356 * First look for an exact match. 357 */ 358 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 359 ti->ti_dst, ti->ti_dport); 360 /* 361 * ...and if that fails, do a wildcard search. 362 */ 363 if (inp == NULL) { 364 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 365 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 366 } 367 368 /* 369 * If the state is CLOSED (i.e., TCB does not exist) then 370 * all data in the incoming segment is discarded. 371 * If the TCB exists but is in CLOSED state, it is embryonic, 372 * but should either do a listen or a connect soon. 373 */ 374 if (inp == NULL) 375 goto dropwithreset; 376 tp = intotcpcb(inp); 377 if (tp == 0) 378 goto dropwithreset; 379 if (tp->t_state == TCPS_CLOSED) 380 goto drop; 381 382 /* Unscale the window into a 32-bit value. */ 383 if ((tiflags & TH_SYN) == 0) 384 tiwin = ti->ti_win << tp->snd_scale; 385 else 386 tiwin = ti->ti_win; 387 388 so = inp->inp_socket; 389 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 390 #ifdef TCPDEBUG 391 if (so->so_options & SO_DEBUG) { 392 ostate = tp->t_state; 393 tcp_saveti = *ti; 394 } 395 #endif 396 if (so->so_options & SO_ACCEPTCONN) { 397 register struct tcpcb *tp0 = tp; 398 so = sonewconn(so, 0); 399 if (so == 0) { 400 tcpstat.tcps_listendrop++; 401 goto drop; 402 } 403 /* 404 * This is ugly, but .... 405 * 406 * Mark socket as temporary until we're 407 * committed to keeping it. The code at 408 * ``drop'' and ``dropwithreset'' check the 409 * flag dropsocket to see if the temporary 410 * socket created here should be discarded. 411 * We mark the socket as discardable until 412 * we're committed to it below in TCPS_LISTEN. 413 */ 414 dropsocket++; 415 inp = (struct inpcb *)so->so_pcb; 416 inp->inp_laddr = ti->ti_dst; 417 inp->inp_lport = ti->ti_dport; 418 in_pcbrehash(inp); 419 #if BSD>=43 420 inp->inp_options = ip_srcroute(); 421 #endif 422 tp = intotcpcb(inp); 423 tp->t_state = TCPS_LISTEN; 424 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 425 426 /* Compute proper scaling value from buffer space */ 427 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 428 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 429 tp->request_r_scale++; 430 } 431 } 432 433 /* 434 * Segment received on connection. 435 * Reset idle time and keep-alive timer. 436 */ 437 tp->t_idle = 0; 438 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 439 440 /* 441 * Process options if not in LISTEN state, 442 * else do it below (after getting remote address). 443 */ 444 if (tp->t_state != TCPS_LISTEN) 445 tcp_dooptions(tp, optp, optlen, ti, &to); 446 447 /* 448 * Header prediction: check for the two common cases 449 * of a uni-directional data xfer. If the packet has 450 * no control flags, is in-sequence, the window didn't 451 * change and we're not retransmitting, it's a 452 * candidate. If the length is zero and the ack moved 453 * forward, we're the sender side of the xfer. Just 454 * free the data acked & wake any higher level process 455 * that was blocked waiting for space. If the length 456 * is non-zero and the ack didn't move, we're the 457 * receiver side. If we're getting packets in-order 458 * (the reassembly queue is empty), add the data to 459 * the socket buffer and note that we need a delayed ack. 460 * Make sure that the hidden state-flags are also off. 461 * Since we check for TCPS_ESTABLISHED above, it can only 462 * be TH_NEEDSYN. 463 */ 464 if (tp->t_state == TCPS_ESTABLISHED && 465 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 466 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 467 ((to.to_flag & TOF_TS) == 0 || 468 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 469 /* 470 * Using the CC option is compulsory if once started: 471 * the segment is OK if no T/TCP was negotiated or 472 * if the segment has a CC option equal to CCrecv 473 */ 474 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 475 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 476 ti->ti_seq == tp->rcv_nxt && 477 tiwin && tiwin == tp->snd_wnd && 478 tp->snd_nxt == tp->snd_max) { 479 480 /* 481 * If last ACK falls within this segment's sequence numbers, 482 * record the timestamp. 483 * NOTE that the test is modified according to the latest 484 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 485 */ 486 if ((to.to_flag & TOF_TS) != 0 && 487 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 488 tp->ts_recent_age = tcp_now; 489 tp->ts_recent = to.to_tsval; 490 } 491 492 if (ti->ti_len == 0) { 493 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 494 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 495 tp->snd_cwnd >= tp->snd_wnd) { 496 /* 497 * this is a pure ack for outstanding data. 498 */ 499 ++tcpstat.tcps_predack; 500 if ((to.to_flag & TOF_TS) != 0) 501 tcp_xmit_timer(tp, 502 tcp_now - to.to_tsecr + 1); 503 else if (tp->t_rtt && 504 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 505 tcp_xmit_timer(tp, tp->t_rtt); 506 acked = ti->ti_ack - tp->snd_una; 507 tcpstat.tcps_rcvackpack++; 508 tcpstat.tcps_rcvackbyte += acked; 509 sbdrop(&so->so_snd, acked); 510 tp->snd_una = ti->ti_ack; 511 m_freem(m); 512 513 /* 514 * If all outstanding data are acked, stop 515 * retransmit timer, otherwise restart timer 516 * using current (possibly backed-off) value. 517 * If process is waiting for space, 518 * wakeup/selwakeup/signal. If data 519 * are ready to send, let tcp_output 520 * decide between more output or persist. 521 */ 522 if (tp->snd_una == tp->snd_max) 523 tp->t_timer[TCPT_REXMT] = 0; 524 else if (tp->t_timer[TCPT_PERSIST] == 0) 525 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 526 527 if (so->so_snd.sb_flags & SB_NOTIFY) 528 sowwakeup(so); 529 if (so->so_snd.sb_cc) 530 (void) tcp_output(tp); 531 return; 532 } 533 } else if (ti->ti_ack == tp->snd_una && 534 tp->seg_next == (struct tcpiphdr *)tp && 535 ti->ti_len <= sbspace(&so->so_rcv)) { 536 /* 537 * this is a pure, in-sequence data packet 538 * with nothing on the reassembly queue and 539 * we have enough buffer space to take it. 540 */ 541 ++tcpstat.tcps_preddat; 542 tp->rcv_nxt += ti->ti_len; 543 tcpstat.tcps_rcvpack++; 544 tcpstat.tcps_rcvbyte += ti->ti_len; 545 /* 546 * Add data to socket buffer. 547 */ 548 sbappend(&so->so_rcv, m); 549 sorwakeup(so); 550 #ifdef TCP_ACK_HACK 551 /* 552 * If this is a short packet, then ACK now - with Nagel 553 * congestion avoidance sender won't send more until 554 * he gets an ACK. 555 */ 556 if (tiflags & TH_PUSH) { 557 tp->t_flags |= TF_ACKNOW; 558 tcp_output(tp); 559 } else { 560 tp->t_flags |= TF_DELACK; 561 } 562 #else 563 tp->t_flags |= TF_DELACK; 564 #endif 565 return; 566 } 567 } 568 569 /* 570 * Calculate amount of space in receive window, 571 * and then do TCP input processing. 572 * Receive window is amount of space in rcv queue, 573 * but not less than advertised window. 574 */ 575 { int win; 576 577 win = sbspace(&so->so_rcv); 578 if (win < 0) 579 win = 0; 580 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 581 } 582 583 switch (tp->t_state) { 584 585 /* 586 * If the state is LISTEN then ignore segment if it contains an RST. 587 * If the segment contains an ACK then it is bad and send a RST. 588 * If it does not contain a SYN then it is not interesting; drop it. 589 * Don't bother responding if the destination was a broadcast. 590 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 591 * tp->iss, and send a segment: 592 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 593 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 594 * Fill in remote peer address fields if not previously specified. 595 * Enter SYN_RECEIVED state, and process any other fields of this 596 * segment in this state. 597 */ 598 case TCPS_LISTEN: { 599 struct mbuf *am; 600 register struct sockaddr_in *sin; 601 602 if (tiflags & TH_RST) 603 goto drop; 604 if (tiflags & TH_ACK) 605 goto dropwithreset; 606 if ((tiflags & TH_SYN) == 0) 607 goto drop; 608 /* 609 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 610 * in_broadcast() should never return true on a received 611 * packet with M_BCAST not set. 612 */ 613 if (m->m_flags & (M_BCAST|M_MCAST) || 614 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 615 goto drop; 616 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 617 if (am == NULL) 618 goto drop; 619 am->m_len = sizeof (struct sockaddr_in); 620 sin = mtod(am, struct sockaddr_in *); 621 sin->sin_family = AF_INET; 622 sin->sin_len = sizeof(*sin); 623 sin->sin_addr = ti->ti_src; 624 sin->sin_port = ti->ti_sport; 625 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 626 laddr = inp->inp_laddr; 627 if (inp->inp_laddr.s_addr == INADDR_ANY) 628 inp->inp_laddr = ti->ti_dst; 629 if (in_pcbconnect(inp, am)) { 630 inp->inp_laddr = laddr; 631 (void) m_free(am); 632 goto drop; 633 } 634 (void) m_free(am); 635 tp->t_template = tcp_template(tp); 636 if (tp->t_template == 0) { 637 tp = tcp_drop(tp, ENOBUFS); 638 dropsocket = 0; /* socket is already gone */ 639 goto drop; 640 } 641 if ((taop = tcp_gettaocache(inp)) == NULL) { 642 taop = &tao_noncached; 643 bzero(taop, sizeof(*taop)); 644 } 645 tcp_dooptions(tp, optp, optlen, ti, &to); 646 if (iss) 647 tp->iss = iss; 648 else 649 tp->iss = tcp_iss; 650 tcp_iss += TCP_ISSINCR/4; 651 tp->irs = ti->ti_seq; 652 tcp_sendseqinit(tp); 653 tcp_rcvseqinit(tp); 654 /* 655 * Initialization of the tcpcb for transaction; 656 * set SND.WND = SEG.WND, 657 * initialize CCsend and CCrecv. 658 */ 659 tp->snd_wnd = tiwin; /* initial send-window */ 660 tp->cc_send = CC_INC(tcp_ccgen); 661 tp->cc_recv = to.to_cc; 662 /* 663 * Perform TAO test on incoming CC (SEG.CC) option, if any. 664 * - compare SEG.CC against cached CC from the same host, 665 * if any. 666 * - if SEG.CC > chached value, SYN must be new and is accepted 667 * immediately: save new CC in the cache, mark the socket 668 * connected, enter ESTABLISHED state, turn on flag to 669 * send a SYN in the next segment. 670 * A virtual advertised window is set in rcv_adv to 671 * initialize SWS prevention. Then enter normal segment 672 * processing: drop SYN, process data and FIN. 673 * - otherwise do a normal 3-way handshake. 674 */ 675 if ((to.to_flag & TOF_CC) != 0) { 676 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 677 taop->tao_cc = to.to_cc; 678 tp->t_state = TCPS_ESTABLISHED; 679 680 /* 681 * If there is a FIN, or if there is data and the 682 * connection is local, then delay SYN,ACK(SYN) in 683 * the hope of piggy-backing it on a response 684 * segment. Otherwise must send ACK now in case 685 * the other side is slow starting. 686 */ 687 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 688 in_localaddr(inp->inp_faddr))) 689 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 690 else 691 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 692 693 /* 694 * Limit the `virtual advertised window' to TCP_MAXWIN 695 * here. Even if we requested window scaling, it will 696 * become effective only later when our SYN is acked. 697 */ 698 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 699 tcpstat.tcps_connects++; 700 soisconnected(so); 701 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 702 dropsocket = 0; /* committed to socket */ 703 tcpstat.tcps_accepts++; 704 goto trimthenstep6; 705 } 706 /* else do standard 3-way handshake */ 707 } else { 708 /* 709 * No CC option, but maybe CC.NEW: 710 * invalidate cached value. 711 */ 712 taop->tao_cc = 0; 713 } 714 /* 715 * TAO test failed or there was no CC option, 716 * do a standard 3-way handshake. 717 */ 718 tp->t_flags |= TF_ACKNOW; 719 tp->t_state = TCPS_SYN_RECEIVED; 720 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 721 dropsocket = 0; /* committed to socket */ 722 tcpstat.tcps_accepts++; 723 goto trimthenstep6; 724 } 725 726 /* 727 * If the state is SYN_SENT: 728 * if seg contains an ACK, but not for our SYN, drop the input. 729 * if seg contains a RST, then drop the connection. 730 * if seg does not contain SYN, then drop it. 731 * Otherwise this is an acceptable SYN segment 732 * initialize tp->rcv_nxt and tp->irs 733 * if seg contains ack then advance tp->snd_una 734 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 735 * arrange for segment to be acked (eventually) 736 * continue processing rest of data/controls, beginning with URG 737 */ 738 case TCPS_SYN_SENT: 739 if ((taop = tcp_gettaocache(inp)) == NULL) { 740 taop = &tao_noncached; 741 bzero(taop, sizeof(*taop)); 742 } 743 744 if ((tiflags & TH_ACK) && 745 (SEQ_LEQ(ti->ti_ack, tp->iss) || 746 SEQ_GT(ti->ti_ack, tp->snd_max))) { 747 /* 748 * If we have a cached CCsent for the remote host, 749 * hence we haven't just crashed and restarted, 750 * do not send a RST. This may be a retransmission 751 * from the other side after our earlier ACK was lost. 752 * Our new SYN, when it arrives, will serve as the 753 * needed ACK. 754 */ 755 if (taop->tao_ccsent != 0) 756 goto drop; 757 else 758 goto dropwithreset; 759 } 760 if (tiflags & TH_RST) { 761 if (tiflags & TH_ACK) 762 tp = tcp_drop(tp, ECONNREFUSED); 763 goto drop; 764 } 765 if ((tiflags & TH_SYN) == 0) 766 goto drop; 767 tp->snd_wnd = ti->ti_win; /* initial send window */ 768 tp->cc_recv = to.to_cc; /* foreign CC */ 769 770 tp->irs = ti->ti_seq; 771 tcp_rcvseqinit(tp); 772 if (tiflags & TH_ACK) { 773 /* 774 * Our SYN was acked. If segment contains CC.ECHO 775 * option, check it to make sure this segment really 776 * matches our SYN. If not, just drop it as old 777 * duplicate, but send an RST if we're still playing 778 * by the old rules. 779 */ 780 if ((to.to_flag & TOF_CCECHO) && 781 tp->cc_send != to.to_ccecho) { 782 if (taop->tao_ccsent != 0) 783 goto drop; 784 else 785 goto dropwithreset; 786 } 787 tcpstat.tcps_connects++; 788 soisconnected(so); 789 /* Do window scaling on this connection? */ 790 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 791 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 792 tp->snd_scale = tp->requested_s_scale; 793 tp->rcv_scale = tp->request_r_scale; 794 } 795 /* Segment is acceptable, update cache if undefined. */ 796 if (taop->tao_ccsent == 0) 797 taop->tao_ccsent = to.to_ccecho; 798 799 tp->rcv_adv += tp->rcv_wnd; 800 tp->snd_una++; /* SYN is acked */ 801 /* 802 * If there's data, delay ACK; if there's also a FIN 803 * ACKNOW will be turned on later. 804 */ 805 if (ti->ti_len != 0) 806 tp->t_flags |= TF_DELACK; 807 else 808 tp->t_flags |= TF_ACKNOW; 809 /* 810 * Received <SYN,ACK> in SYN_SENT[*] state. 811 * Transitions: 812 * SYN_SENT --> ESTABLISHED 813 * SYN_SENT* --> FIN_WAIT_1 814 */ 815 if (tp->t_flags & TF_NEEDFIN) { 816 tp->t_state = TCPS_FIN_WAIT_1; 817 tp->t_flags &= ~TF_NEEDFIN; 818 tiflags &= ~TH_SYN; 819 } else 820 tp->t_state = TCPS_ESTABLISHED; 821 822 } else { 823 /* 824 * Received initial SYN in SYN-SENT[*] state => simul- 825 * taneous open. If segment contains CC option and there is 826 * a cached CC, apply TAO test; if it succeeds, connection is 827 * half-synchronized. Otherwise, do 3-way handshake: 828 * SYN-SENT -> SYN-RECEIVED 829 * SYN-SENT* -> SYN-RECEIVED* 830 * If there was no CC option, clear cached CC value. 831 */ 832 tp->t_flags |= TF_ACKNOW; 833 tp->t_timer[TCPT_REXMT] = 0; 834 if (to.to_flag & TOF_CC) { 835 if (taop->tao_cc != 0 && 836 CC_GT(to.to_cc, taop->tao_cc)) { 837 /* 838 * update cache and make transition: 839 * SYN-SENT -> ESTABLISHED* 840 * SYN-SENT* -> FIN-WAIT-1* 841 */ 842 taop->tao_cc = to.to_cc; 843 if (tp->t_flags & TF_NEEDFIN) { 844 tp->t_state = TCPS_FIN_WAIT_1; 845 tp->t_flags &= ~TF_NEEDFIN; 846 } else 847 tp->t_state = TCPS_ESTABLISHED; 848 tp->t_flags |= TF_NEEDSYN; 849 } else 850 tp->t_state = TCPS_SYN_RECEIVED; 851 } else { 852 /* CC.NEW or no option => invalidate cache */ 853 taop->tao_cc = 0; 854 tp->t_state = TCPS_SYN_RECEIVED; 855 } 856 } 857 858 trimthenstep6: 859 /* 860 * Advance ti->ti_seq to correspond to first data byte. 861 * If data, trim to stay within window, 862 * dropping FIN if necessary. 863 */ 864 ti->ti_seq++; 865 if (ti->ti_len > tp->rcv_wnd) { 866 todrop = ti->ti_len - tp->rcv_wnd; 867 m_adj(m, -todrop); 868 ti->ti_len = tp->rcv_wnd; 869 tiflags &= ~TH_FIN; 870 tcpstat.tcps_rcvpackafterwin++; 871 tcpstat.tcps_rcvbyteafterwin += todrop; 872 } 873 tp->snd_wl1 = ti->ti_seq - 1; 874 tp->rcv_up = ti->ti_seq; 875 /* 876 * Client side of transaction: already sent SYN and data. 877 * If the remote host used T/TCP to validate the SYN, 878 * our data will be ACK'd; if so, enter normal data segment 879 * processing in the middle of step 5, ack processing. 880 * Otherwise, goto step 6. 881 */ 882 if (tiflags & TH_ACK) 883 goto process_ACK; 884 goto step6; 885 /* 886 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 887 * if segment contains a SYN and CC [not CC.NEW] option: 888 * if state == TIME_WAIT and connection duration > MSL, 889 * drop packet and send RST; 890 * 891 * if SEG.CC > CCrecv then is new SYN, and can implicitly 892 * ack the FIN (and data) in retransmission queue. 893 * Complete close and delete TCPCB. Then reprocess 894 * segment, hoping to find new TCPCB in LISTEN state; 895 * 896 * else must be old SYN; drop it. 897 * else do normal processing. 898 */ 899 case TCPS_LAST_ACK: 900 case TCPS_CLOSING: 901 case TCPS_TIME_WAIT: 902 if ((tiflags & TH_SYN) && 903 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 904 if (tp->t_state == TCPS_TIME_WAIT && 905 tp->t_duration > TCPTV_MSL) 906 goto dropwithreset; 907 if (CC_GT(to.to_cc, tp->cc_recv)) { 908 tp = tcp_close(tp); 909 goto findpcb; 910 } 911 else 912 goto drop; 913 } 914 break; /* continue normal processing */ 915 } 916 917 /* 918 * States other than LISTEN or SYN_SENT. 919 * First check timestamp, if present. 920 * Then check the connection count, if present. 921 * Then check that at least some bytes of segment are within 922 * receive window. If segment begins before rcv_nxt, 923 * drop leading data (and SYN); if nothing left, just ack. 924 * 925 * RFC 1323 PAWS: If we have a timestamp reply on this segment 926 * and it's less than ts_recent, drop it. 927 */ 928 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 929 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 930 931 /* Check to see if ts_recent is over 24 days old. */ 932 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 933 /* 934 * Invalidate ts_recent. If this segment updates 935 * ts_recent, the age will be reset later and ts_recent 936 * will get a valid value. If it does not, setting 937 * ts_recent to zero will at least satisfy the 938 * requirement that zero be placed in the timestamp 939 * echo reply when ts_recent isn't valid. The 940 * age isn't reset until we get a valid ts_recent 941 * because we don't want out-of-order segments to be 942 * dropped when ts_recent is old. 943 */ 944 tp->ts_recent = 0; 945 } else { 946 tcpstat.tcps_rcvduppack++; 947 tcpstat.tcps_rcvdupbyte += ti->ti_len; 948 tcpstat.tcps_pawsdrop++; 949 goto dropafterack; 950 } 951 } 952 953 /* 954 * T/TCP mechanism 955 * If T/TCP was negotiated and the segment doesn't have CC, 956 * or if it's CC is wrong then drop the segment. 957 * RST segments do not have to comply with this. 958 */ 959 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 960 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 961 (tiflags & TH_RST) == 0) 962 goto dropafterack; 963 964 todrop = tp->rcv_nxt - ti->ti_seq; 965 if (todrop > 0) { 966 if (tiflags & TH_SYN) { 967 tiflags &= ~TH_SYN; 968 ti->ti_seq++; 969 if (ti->ti_urp > 1) 970 ti->ti_urp--; 971 else 972 tiflags &= ~TH_URG; 973 todrop--; 974 } 975 /* 976 * Following if statement from Stevens, vol. 2, p. 960. 977 */ 978 if (todrop > ti->ti_len 979 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 980 /* 981 * Any valid FIN must be to the left of the window. 982 * At this point the FIN must be a duplicate or out 983 * of sequence; drop it. 984 */ 985 tiflags &= ~TH_FIN; 986 987 /* 988 * Send an ACK to resynchronize and drop any data. 989 * But keep on processing for RST or ACK. 990 */ 991 tp->t_flags |= TF_ACKNOW; 992 todrop = ti->ti_len; 993 tcpstat.tcps_rcvduppack++; 994 tcpstat.tcps_rcvdupbyte += todrop; 995 } else { 996 tcpstat.tcps_rcvpartduppack++; 997 tcpstat.tcps_rcvpartdupbyte += todrop; 998 } 999 m_adj(m, todrop); 1000 ti->ti_seq += todrop; 1001 ti->ti_len -= todrop; 1002 if (ti->ti_urp > todrop) 1003 ti->ti_urp -= todrop; 1004 else { 1005 tiflags &= ~TH_URG; 1006 ti->ti_urp = 0; 1007 } 1008 } 1009 1010 /* 1011 * If new data are received on a connection after the 1012 * user processes are gone, then RST the other end. 1013 */ 1014 if ((so->so_state & SS_NOFDREF) && 1015 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1016 tp = tcp_close(tp); 1017 tcpstat.tcps_rcvafterclose++; 1018 goto dropwithreset; 1019 } 1020 1021 /* 1022 * If segment ends after window, drop trailing data 1023 * (and PUSH and FIN); if nothing left, just ACK. 1024 */ 1025 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1026 if (todrop > 0) { 1027 tcpstat.tcps_rcvpackafterwin++; 1028 if (todrop >= ti->ti_len) { 1029 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1030 /* 1031 * If a new connection request is received 1032 * while in TIME_WAIT, drop the old connection 1033 * and start over if the sequence numbers 1034 * are above the previous ones. 1035 */ 1036 if (tiflags & TH_SYN && 1037 tp->t_state == TCPS_TIME_WAIT && 1038 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1039 iss = tp->rcv_nxt + TCP_ISSINCR; 1040 tp = tcp_close(tp); 1041 goto findpcb; 1042 } 1043 /* 1044 * If window is closed can only take segments at 1045 * window edge, and have to drop data and PUSH from 1046 * incoming segments. Continue processing, but 1047 * remember to ack. Otherwise, drop segment 1048 * and ack. 1049 */ 1050 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1051 tp->t_flags |= TF_ACKNOW; 1052 tcpstat.tcps_rcvwinprobe++; 1053 } else 1054 goto dropafterack; 1055 } else 1056 tcpstat.tcps_rcvbyteafterwin += todrop; 1057 m_adj(m, -todrop); 1058 ti->ti_len -= todrop; 1059 tiflags &= ~(TH_PUSH|TH_FIN); 1060 } 1061 1062 /* 1063 * If last ACK falls within this segment's sequence numbers, 1064 * record its timestamp. 1065 * NOTE that the test is modified according to the latest 1066 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1067 */ 1068 if ((to.to_flag & TOF_TS) != 0 && 1069 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1070 tp->ts_recent_age = tcp_now; 1071 tp->ts_recent = to.to_tsval; 1072 } 1073 1074 /* 1075 * If the RST bit is set examine the state: 1076 * SYN_RECEIVED STATE: 1077 * If passive open, return to LISTEN state. 1078 * If active open, inform user that connection was refused. 1079 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1080 * Inform user that connection was reset, and close tcb. 1081 * CLOSING, LAST_ACK, TIME_WAIT STATES 1082 * Close the tcb. 1083 */ 1084 if (tiflags&TH_RST) switch (tp->t_state) { 1085 1086 case TCPS_SYN_RECEIVED: 1087 so->so_error = ECONNREFUSED; 1088 goto close; 1089 1090 case TCPS_ESTABLISHED: 1091 case TCPS_FIN_WAIT_1: 1092 case TCPS_FIN_WAIT_2: 1093 case TCPS_CLOSE_WAIT: 1094 so->so_error = ECONNRESET; 1095 close: 1096 tp->t_state = TCPS_CLOSED; 1097 tcpstat.tcps_drops++; 1098 tp = tcp_close(tp); 1099 goto drop; 1100 1101 case TCPS_CLOSING: 1102 case TCPS_LAST_ACK: 1103 case TCPS_TIME_WAIT: 1104 tp = tcp_close(tp); 1105 goto drop; 1106 } 1107 1108 /* 1109 * If a SYN is in the window, then this is an 1110 * error and we send an RST and drop the connection. 1111 */ 1112 if (tiflags & TH_SYN) { 1113 tp = tcp_drop(tp, ECONNRESET); 1114 goto dropwithreset; 1115 } 1116 1117 /* 1118 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1119 * flag is on (half-synchronized state), then queue data for 1120 * later processing; else drop segment and return. 1121 */ 1122 if ((tiflags & TH_ACK) == 0) { 1123 if (tp->t_state == TCPS_SYN_RECEIVED || 1124 (tp->t_flags & TF_NEEDSYN)) 1125 goto step6; 1126 else 1127 goto drop; 1128 } 1129 1130 /* 1131 * Ack processing. 1132 */ 1133 switch (tp->t_state) { 1134 1135 /* 1136 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1137 * ESTABLISHED state and continue processing, otherwise 1138 * send an RST. 1139 */ 1140 case TCPS_SYN_RECEIVED: 1141 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1142 SEQ_GT(ti->ti_ack, tp->snd_max)) 1143 goto dropwithreset; 1144 1145 tcpstat.tcps_connects++; 1146 soisconnected(so); 1147 /* Do window scaling? */ 1148 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1149 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1150 tp->snd_scale = tp->requested_s_scale; 1151 tp->rcv_scale = tp->request_r_scale; 1152 } 1153 /* 1154 * Upon successful completion of 3-way handshake, 1155 * update cache.CC if it was undefined, pass any queued 1156 * data to the user, and advance state appropriately. 1157 */ 1158 if ((taop = tcp_gettaocache(inp)) != NULL && 1159 taop->tao_cc == 0) 1160 taop->tao_cc = tp->cc_recv; 1161 1162 /* 1163 * Make transitions: 1164 * SYN-RECEIVED -> ESTABLISHED 1165 * SYN-RECEIVED* -> FIN-WAIT-1 1166 */ 1167 if (tp->t_flags & TF_NEEDFIN) { 1168 tp->t_state = TCPS_FIN_WAIT_1; 1169 tp->t_flags &= ~TF_NEEDFIN; 1170 } else 1171 tp->t_state = TCPS_ESTABLISHED; 1172 /* 1173 * If segment contains data or ACK, will call tcp_reass() 1174 * later; if not, do so now to pass queued data to user. 1175 */ 1176 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1177 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1178 (struct mbuf *)0); 1179 tp->snd_wl1 = ti->ti_seq - 1; 1180 /* fall into ... */ 1181 1182 /* 1183 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1184 * ACKs. If the ack is in the range 1185 * tp->snd_una < ti->ti_ack <= tp->snd_max 1186 * then advance tp->snd_una to ti->ti_ack and drop 1187 * data from the retransmission queue. If this ACK reflects 1188 * more up to date window information we update our window information. 1189 */ 1190 case TCPS_ESTABLISHED: 1191 case TCPS_FIN_WAIT_1: 1192 case TCPS_FIN_WAIT_2: 1193 case TCPS_CLOSE_WAIT: 1194 case TCPS_CLOSING: 1195 case TCPS_LAST_ACK: 1196 case TCPS_TIME_WAIT: 1197 1198 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1199 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1200 tcpstat.tcps_rcvdupack++; 1201 /* 1202 * If we have outstanding data (other than 1203 * a window probe), this is a completely 1204 * duplicate ack (ie, window info didn't 1205 * change), the ack is the biggest we've 1206 * seen and we've seen exactly our rexmt 1207 * threshhold of them, assume a packet 1208 * has been dropped and retransmit it. 1209 * Kludge snd_nxt & the congestion 1210 * window so we send only this one 1211 * packet. 1212 * 1213 * We know we're losing at the current 1214 * window size so do congestion avoidance 1215 * (set ssthresh to half the current window 1216 * and pull our congestion window back to 1217 * the new ssthresh). 1218 * 1219 * Dup acks mean that packets have left the 1220 * network (they're now cached at the receiver) 1221 * so bump cwnd by the amount in the receiver 1222 * to keep a constant cwnd packets in the 1223 * network. 1224 */ 1225 if (tp->t_timer[TCPT_REXMT] == 0 || 1226 ti->ti_ack != tp->snd_una) 1227 tp->t_dupacks = 0; 1228 else if (++tp->t_dupacks == tcprexmtthresh) { 1229 tcp_seq onxt = tp->snd_nxt; 1230 u_int win = 1231 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1232 tp->t_maxseg; 1233 1234 if (win < 2) 1235 win = 2; 1236 tp->snd_ssthresh = win * tp->t_maxseg; 1237 tp->t_timer[TCPT_REXMT] = 0; 1238 tp->t_rtt = 0; 1239 tp->snd_nxt = ti->ti_ack; 1240 tp->snd_cwnd = tp->t_maxseg; 1241 (void) tcp_output(tp); 1242 tp->snd_cwnd = tp->snd_ssthresh + 1243 tp->t_maxseg * tp->t_dupacks; 1244 if (SEQ_GT(onxt, tp->snd_nxt)) 1245 tp->snd_nxt = onxt; 1246 goto drop; 1247 } else if (tp->t_dupacks > tcprexmtthresh) { 1248 tp->snd_cwnd += tp->t_maxseg; 1249 (void) tcp_output(tp); 1250 goto drop; 1251 } 1252 } else 1253 tp->t_dupacks = 0; 1254 break; 1255 } 1256 /* 1257 * If the congestion window was inflated to account 1258 * for the other side's cached packets, retract it. 1259 */ 1260 if (tp->t_dupacks > tcprexmtthresh && 1261 tp->snd_cwnd > tp->snd_ssthresh) 1262 tp->snd_cwnd = tp->snd_ssthresh; 1263 tp->t_dupacks = 0; 1264 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1265 tcpstat.tcps_rcvacktoomuch++; 1266 goto dropafterack; 1267 } 1268 /* 1269 * If we reach this point, ACK is not a duplicate, 1270 * i.e., it ACKs something we sent. 1271 */ 1272 if (tp->t_flags & TF_NEEDSYN) { 1273 /* 1274 * T/TCP: Connection was half-synchronized, and our 1275 * SYN has been ACK'd (so connection is now fully 1276 * synchronized). Go to non-starred state, 1277 * increment snd_una for ACK of SYN, and check if 1278 * we can do window scaling. 1279 */ 1280 tp->t_flags &= ~TF_NEEDSYN; 1281 tp->snd_una++; 1282 /* Do window scaling? */ 1283 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1284 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1285 tp->snd_scale = tp->requested_s_scale; 1286 tp->rcv_scale = tp->request_r_scale; 1287 } 1288 } 1289 1290 process_ACK: 1291 acked = ti->ti_ack - tp->snd_una; 1292 tcpstat.tcps_rcvackpack++; 1293 tcpstat.tcps_rcvackbyte += acked; 1294 1295 /* 1296 * If we have a timestamp reply, update smoothed 1297 * round trip time. If no timestamp is present but 1298 * transmit timer is running and timed sequence 1299 * number was acked, update smoothed round trip time. 1300 * Since we now have an rtt measurement, cancel the 1301 * timer backoff (cf., Phil Karn's retransmit alg.). 1302 * Recompute the initial retransmit timer. 1303 */ 1304 if (to.to_flag & TOF_TS) 1305 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1306 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1307 tcp_xmit_timer(tp,tp->t_rtt); 1308 1309 /* 1310 * If all outstanding data is acked, stop retransmit 1311 * timer and remember to restart (more output or persist). 1312 * If there is more data to be acked, restart retransmit 1313 * timer, using current (possibly backed-off) value. 1314 */ 1315 if (ti->ti_ack == tp->snd_max) { 1316 tp->t_timer[TCPT_REXMT] = 0; 1317 needoutput = 1; 1318 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1319 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1320 1321 /* 1322 * If no data (only SYN) was ACK'd, 1323 * skip rest of ACK processing. 1324 */ 1325 if (acked == 0) 1326 goto step6; 1327 1328 /* 1329 * When new data is acked, open the congestion window. 1330 * If the window gives us less than ssthresh packets 1331 * in flight, open exponentially (maxseg per packet). 1332 * Otherwise open linearly: maxseg per window 1333 * (maxseg^2 / cwnd per packet). 1334 */ 1335 { 1336 register u_int cw = tp->snd_cwnd; 1337 register u_int incr = tp->t_maxseg; 1338 1339 if (cw > tp->snd_ssthresh) 1340 incr = incr * incr / cw; 1341 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1342 } 1343 if (acked > so->so_snd.sb_cc) { 1344 tp->snd_wnd -= so->so_snd.sb_cc; 1345 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1346 ourfinisacked = 1; 1347 } else { 1348 sbdrop(&so->so_snd, acked); 1349 tp->snd_wnd -= acked; 1350 ourfinisacked = 0; 1351 } 1352 if (so->so_snd.sb_flags & SB_NOTIFY) 1353 sowwakeup(so); 1354 tp->snd_una = ti->ti_ack; 1355 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1356 tp->snd_nxt = tp->snd_una; 1357 1358 switch (tp->t_state) { 1359 1360 /* 1361 * In FIN_WAIT_1 STATE in addition to the processing 1362 * for the ESTABLISHED state if our FIN is now acknowledged 1363 * then enter FIN_WAIT_2. 1364 */ 1365 case TCPS_FIN_WAIT_1: 1366 if (ourfinisacked) { 1367 /* 1368 * If we can't receive any more 1369 * data, then closing user can proceed. 1370 * Starting the timer is contrary to the 1371 * specification, but if we don't get a FIN 1372 * we'll hang forever. 1373 */ 1374 if (so->so_state & SS_CANTRCVMORE) { 1375 soisdisconnected(so); 1376 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1377 } 1378 tp->t_state = TCPS_FIN_WAIT_2; 1379 } 1380 break; 1381 1382 /* 1383 * In CLOSING STATE in addition to the processing for 1384 * the ESTABLISHED state if the ACK acknowledges our FIN 1385 * then enter the TIME-WAIT state, otherwise ignore 1386 * the segment. 1387 */ 1388 case TCPS_CLOSING: 1389 if (ourfinisacked) { 1390 tp->t_state = TCPS_TIME_WAIT; 1391 tcp_canceltimers(tp); 1392 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1393 if (tp->cc_recv != 0 && 1394 tp->t_duration < TCPTV_MSL) 1395 tp->t_timer[TCPT_2MSL] = 1396 tp->t_rxtcur * TCPTV_TWTRUNC; 1397 else 1398 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1399 soisdisconnected(so); 1400 } 1401 break; 1402 1403 /* 1404 * In LAST_ACK, we may still be waiting for data to drain 1405 * and/or to be acked, as well as for the ack of our FIN. 1406 * If our FIN is now acknowledged, delete the TCB, 1407 * enter the closed state and return. 1408 */ 1409 case TCPS_LAST_ACK: 1410 if (ourfinisacked) { 1411 tp = tcp_close(tp); 1412 goto drop; 1413 } 1414 break; 1415 1416 /* 1417 * In TIME_WAIT state the only thing that should arrive 1418 * is a retransmission of the remote FIN. Acknowledge 1419 * it and restart the finack timer. 1420 */ 1421 case TCPS_TIME_WAIT: 1422 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1423 goto dropafterack; 1424 } 1425 } 1426 1427 step6: 1428 /* 1429 * Update window information. 1430 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1431 */ 1432 if ((tiflags & TH_ACK) && 1433 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1434 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1435 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1436 /* keep track of pure window updates */ 1437 if (ti->ti_len == 0 && 1438 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1439 tcpstat.tcps_rcvwinupd++; 1440 tp->snd_wnd = tiwin; 1441 tp->snd_wl1 = ti->ti_seq; 1442 tp->snd_wl2 = ti->ti_ack; 1443 if (tp->snd_wnd > tp->max_sndwnd) 1444 tp->max_sndwnd = tp->snd_wnd; 1445 needoutput = 1; 1446 } 1447 1448 /* 1449 * Process segments with URG. 1450 */ 1451 if ((tiflags & TH_URG) && ti->ti_urp && 1452 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1453 /* 1454 * This is a kludge, but if we receive and accept 1455 * random urgent pointers, we'll crash in 1456 * soreceive. It's hard to imagine someone 1457 * actually wanting to send this much urgent data. 1458 */ 1459 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1460 ti->ti_urp = 0; /* XXX */ 1461 tiflags &= ~TH_URG; /* XXX */ 1462 goto dodata; /* XXX */ 1463 } 1464 /* 1465 * If this segment advances the known urgent pointer, 1466 * then mark the data stream. This should not happen 1467 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1468 * a FIN has been received from the remote side. 1469 * In these states we ignore the URG. 1470 * 1471 * According to RFC961 (Assigned Protocols), 1472 * the urgent pointer points to the last octet 1473 * of urgent data. We continue, however, 1474 * to consider it to indicate the first octet 1475 * of data past the urgent section as the original 1476 * spec states (in one of two places). 1477 */ 1478 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1479 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1480 so->so_oobmark = so->so_rcv.sb_cc + 1481 (tp->rcv_up - tp->rcv_nxt) - 1; 1482 if (so->so_oobmark == 0) 1483 so->so_state |= SS_RCVATMARK; 1484 sohasoutofband(so); 1485 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1486 } 1487 /* 1488 * Remove out of band data so doesn't get presented to user. 1489 * This can happen independent of advancing the URG pointer, 1490 * but if two URG's are pending at once, some out-of-band 1491 * data may creep in... ick. 1492 */ 1493 if (ti->ti_urp <= (u_long)ti->ti_len 1494 #ifdef SO_OOBINLINE 1495 && (so->so_options & SO_OOBINLINE) == 0 1496 #endif 1497 ) 1498 tcp_pulloutofband(so, ti, m); 1499 } else 1500 /* 1501 * If no out of band data is expected, 1502 * pull receive urgent pointer along 1503 * with the receive window. 1504 */ 1505 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1506 tp->rcv_up = tp->rcv_nxt; 1507 dodata: /* XXX */ 1508 1509 /* 1510 * Process the segment text, merging it into the TCP sequencing queue, 1511 * and arranging for acknowledgment of receipt if necessary. 1512 * This process logically involves adjusting tp->rcv_wnd as data 1513 * is presented to the user (this happens in tcp_usrreq.c, 1514 * case PRU_RCVD). If a FIN has already been received on this 1515 * connection then we just ignore the text. 1516 */ 1517 if ((ti->ti_len || (tiflags&TH_FIN)) && 1518 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1519 TCP_REASS(tp, ti, m, so, tiflags); 1520 /* 1521 * Note the amount of data that peer has sent into 1522 * our window, in order to estimate the sender's 1523 * buffer size. 1524 */ 1525 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1526 } else { 1527 m_freem(m); 1528 tiflags &= ~TH_FIN; 1529 } 1530 1531 /* 1532 * If FIN is received ACK the FIN and let the user know 1533 * that the connection is closing. 1534 */ 1535 if (tiflags & TH_FIN) { 1536 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1537 socantrcvmore(so); 1538 /* 1539 * If connection is half-synchronized 1540 * (ie NEEDSYN flag on) then delay ACK, 1541 * so it may be piggybacked when SYN is sent. 1542 * Otherwise, since we received a FIN then no 1543 * more input can be expected, send ACK now. 1544 */ 1545 if (tp->t_flags & TF_NEEDSYN) 1546 tp->t_flags |= TF_DELACK; 1547 else 1548 tp->t_flags |= TF_ACKNOW; 1549 tp->rcv_nxt++; 1550 } 1551 switch (tp->t_state) { 1552 1553 /* 1554 * In SYN_RECEIVED and ESTABLISHED STATES 1555 * enter the CLOSE_WAIT state. 1556 */ 1557 case TCPS_SYN_RECEIVED: 1558 case TCPS_ESTABLISHED: 1559 tp->t_state = TCPS_CLOSE_WAIT; 1560 break; 1561 1562 /* 1563 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1564 * enter the CLOSING state. 1565 */ 1566 case TCPS_FIN_WAIT_1: 1567 tp->t_state = TCPS_CLOSING; 1568 break; 1569 1570 /* 1571 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1572 * starting the time-wait timer, turning off the other 1573 * standard timers. 1574 */ 1575 case TCPS_FIN_WAIT_2: 1576 tp->t_state = TCPS_TIME_WAIT; 1577 tcp_canceltimers(tp); 1578 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1579 if (tp->cc_recv != 0 && 1580 tp->t_duration < TCPTV_MSL) { 1581 tp->t_timer[TCPT_2MSL] = 1582 tp->t_rxtcur * TCPTV_TWTRUNC; 1583 /* For transaction client, force ACK now. */ 1584 tp->t_flags |= TF_ACKNOW; 1585 } 1586 else 1587 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1588 soisdisconnected(so); 1589 break; 1590 1591 /* 1592 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1593 */ 1594 case TCPS_TIME_WAIT: 1595 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1596 break; 1597 } 1598 } 1599 #ifdef TCPDEBUG 1600 if (so->so_options & SO_DEBUG) 1601 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1602 #endif 1603 1604 /* 1605 * Return any desired output. 1606 */ 1607 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1608 (void) tcp_output(tp); 1609 return; 1610 1611 dropafterack: 1612 /* 1613 * Generate an ACK dropping incoming segment if it occupies 1614 * sequence space, where the ACK reflects our state. 1615 */ 1616 if (tiflags & TH_RST) 1617 goto drop; 1618 #ifdef TCPDEBUG 1619 if (so->so_options & SO_DEBUG) 1620 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1621 #endif 1622 m_freem(m); 1623 tp->t_flags |= TF_ACKNOW; 1624 (void) tcp_output(tp); 1625 return; 1626 1627 dropwithreset: 1628 /* 1629 * Generate a RST, dropping incoming segment. 1630 * Make ACK acceptable to originator of segment. 1631 * Don't bother to respond if destination was broadcast/multicast. 1632 */ 1633 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1634 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1635 goto drop; 1636 #ifdef TCPDEBUG 1637 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1638 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1639 #endif 1640 if (tiflags & TH_ACK) 1641 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1642 else { 1643 if (tiflags & TH_SYN) 1644 ti->ti_len++; 1645 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1646 TH_RST|TH_ACK); 1647 } 1648 /* destroy temporarily created socket */ 1649 if (dropsocket) 1650 (void) soabort(so); 1651 return; 1652 1653 drop: 1654 /* 1655 * Drop space held by incoming segment and return. 1656 */ 1657 #ifdef TCPDEBUG 1658 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1659 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1660 #endif 1661 m_freem(m); 1662 /* destroy temporarily created socket */ 1663 if (dropsocket) 1664 (void) soabort(so); 1665 return; 1666 #ifndef TUBA_INCLUDE 1667 } 1668 1669 static void 1670 tcp_dooptions(tp, cp, cnt, ti, to) 1671 struct tcpcb *tp; 1672 u_char *cp; 1673 int cnt; 1674 struct tcpiphdr *ti; 1675 struct tcpopt *to; 1676 { 1677 u_short mss = 0; 1678 int opt, optlen; 1679 1680 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1681 opt = cp[0]; 1682 if (opt == TCPOPT_EOL) 1683 break; 1684 if (opt == TCPOPT_NOP) 1685 optlen = 1; 1686 else { 1687 optlen = cp[1]; 1688 if (optlen <= 0) 1689 break; 1690 } 1691 switch (opt) { 1692 1693 default: 1694 continue; 1695 1696 case TCPOPT_MAXSEG: 1697 if (optlen != TCPOLEN_MAXSEG) 1698 continue; 1699 if (!(ti->ti_flags & TH_SYN)) 1700 continue; 1701 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1702 NTOHS(mss); 1703 break; 1704 1705 case TCPOPT_WINDOW: 1706 if (optlen != TCPOLEN_WINDOW) 1707 continue; 1708 if (!(ti->ti_flags & TH_SYN)) 1709 continue; 1710 tp->t_flags |= TF_RCVD_SCALE; 1711 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1712 break; 1713 1714 case TCPOPT_TIMESTAMP: 1715 if (optlen != TCPOLEN_TIMESTAMP) 1716 continue; 1717 to->to_flag |= TOF_TS; 1718 bcopy((char *)cp + 2, 1719 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1720 NTOHL(to->to_tsval); 1721 bcopy((char *)cp + 6, 1722 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1723 NTOHL(to->to_tsecr); 1724 1725 /* 1726 * A timestamp received in a SYN makes 1727 * it ok to send timestamp requests and replies. 1728 */ 1729 if (ti->ti_flags & TH_SYN) { 1730 tp->t_flags |= TF_RCVD_TSTMP; 1731 tp->ts_recent = to->to_tsval; 1732 tp->ts_recent_age = tcp_now; 1733 } 1734 break; 1735 case TCPOPT_CC: 1736 if (optlen != TCPOLEN_CC) 1737 continue; 1738 to->to_flag |= TOF_CC; 1739 bcopy((char *)cp + 2, 1740 (char *)&to->to_cc, sizeof(to->to_cc)); 1741 NTOHL(to->to_cc); 1742 /* 1743 * A CC or CC.new option received in a SYN makes 1744 * it ok to send CC in subsequent segments. 1745 */ 1746 if (ti->ti_flags & TH_SYN) 1747 tp->t_flags |= TF_RCVD_CC; 1748 break; 1749 case TCPOPT_CCNEW: 1750 if (optlen != TCPOLEN_CC) 1751 continue; 1752 if (!(ti->ti_flags & TH_SYN)) 1753 continue; 1754 to->to_flag |= TOF_CCNEW; 1755 bcopy((char *)cp + 2, 1756 (char *)&to->to_cc, sizeof(to->to_cc)); 1757 NTOHL(to->to_cc); 1758 /* 1759 * A CC or CC.new option received in a SYN makes 1760 * it ok to send CC in subsequent segments. 1761 */ 1762 tp->t_flags |= TF_RCVD_CC; 1763 break; 1764 case TCPOPT_CCECHO: 1765 if (optlen != TCPOLEN_CC) 1766 continue; 1767 if (!(ti->ti_flags & TH_SYN)) 1768 continue; 1769 to->to_flag |= TOF_CCECHO; 1770 bcopy((char *)cp + 2, 1771 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1772 NTOHL(to->to_ccecho); 1773 break; 1774 } 1775 } 1776 if (ti->ti_flags & TH_SYN) 1777 tcp_mss(tp, mss); /* sets t_maxseg */ 1778 } 1779 1780 /* 1781 * Pull out of band byte out of a segment so 1782 * it doesn't appear in the user's data queue. 1783 * It is still reflected in the segment length for 1784 * sequencing purposes. 1785 */ 1786 static void 1787 tcp_pulloutofband(so, ti, m) 1788 struct socket *so; 1789 struct tcpiphdr *ti; 1790 register struct mbuf *m; 1791 { 1792 int cnt = ti->ti_urp - 1; 1793 1794 while (cnt >= 0) { 1795 if (m->m_len > cnt) { 1796 char *cp = mtod(m, caddr_t) + cnt; 1797 struct tcpcb *tp = sototcpcb(so); 1798 1799 tp->t_iobc = *cp; 1800 tp->t_oobflags |= TCPOOB_HAVEDATA; 1801 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1802 m->m_len--; 1803 return; 1804 } 1805 cnt -= m->m_len; 1806 m = m->m_next; 1807 if (m == 0) 1808 break; 1809 } 1810 panic("tcp_pulloutofband"); 1811 } 1812 1813 /* 1814 * Collect new round-trip time estimate 1815 * and update averages and current timeout. 1816 */ 1817 static void 1818 tcp_xmit_timer(tp, rtt) 1819 register struct tcpcb *tp; 1820 short rtt; 1821 { 1822 register short delta; 1823 1824 tcpstat.tcps_rttupdated++; 1825 tp->t_rttupdated++; 1826 if (tp->t_srtt != 0) { 1827 /* 1828 * srtt is stored as fixed point with 3 bits after the 1829 * binary point (i.e., scaled by 8). The following magic 1830 * is equivalent to the smoothing algorithm in rfc793 with 1831 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1832 * point). Adjust rtt to origin 0. 1833 */ 1834 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1835 if ((tp->t_srtt += delta) <= 0) 1836 tp->t_srtt = 1; 1837 /* 1838 * We accumulate a smoothed rtt variance (actually, a 1839 * smoothed mean difference), then set the retransmit 1840 * timer to smoothed rtt + 4 times the smoothed variance. 1841 * rttvar is stored as fixed point with 2 bits after the 1842 * binary point (scaled by 4). The following is 1843 * equivalent to rfc793 smoothing with an alpha of .75 1844 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1845 * rfc793's wired-in beta. 1846 */ 1847 if (delta < 0) 1848 delta = -delta; 1849 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1850 if ((tp->t_rttvar += delta) <= 0) 1851 tp->t_rttvar = 1; 1852 } else { 1853 /* 1854 * No rtt measurement yet - use the unsmoothed rtt. 1855 * Set the variance to half the rtt (so our first 1856 * retransmit happens at 3*rtt). 1857 */ 1858 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1859 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1860 } 1861 tp->t_rtt = 0; 1862 tp->t_rxtshift = 0; 1863 1864 /* 1865 * the retransmit should happen at rtt + 4 * rttvar. 1866 * Because of the way we do the smoothing, srtt and rttvar 1867 * will each average +1/2 tick of bias. When we compute 1868 * the retransmit timer, we want 1/2 tick of rounding and 1869 * 1 extra tick because of +-1/2 tick uncertainty in the 1870 * firing of the timer. The bias will give us exactly the 1871 * 1.5 tick we need. But, because the bias is 1872 * statistical, we have to test that we don't drop below 1873 * the minimum feasible timer (which is 2 ticks). 1874 */ 1875 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1876 tp->t_rttmin, TCPTV_REXMTMAX); 1877 1878 /* 1879 * We received an ack for a packet that wasn't retransmitted; 1880 * it is probably safe to discard any error indications we've 1881 * received recently. This isn't quite right, but close enough 1882 * for now (a route might have failed after we sent a segment, 1883 * and the return path might not be symmetrical). 1884 */ 1885 tp->t_softerror = 0; 1886 } 1887 1888 /* 1889 * Determine a reasonable value for maxseg size. 1890 * If the route is known, check route for mtu. 1891 * If none, use an mss that can be handled on the outgoing 1892 * interface without forcing IP to fragment; if bigger than 1893 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1894 * to utilize large mbufs. If no route is found, route has no mtu, 1895 * or the destination isn't local, use a default, hopefully conservative 1896 * size (usually 512 or the default IP max size, but no more than the mtu 1897 * of the interface), as we can't discover anything about intervening 1898 * gateways or networks. We also initialize the congestion/slow start 1899 * window to be a single segment if the destination isn't local. 1900 * While looking at the routing entry, we also initialize other path-dependent 1901 * parameters from pre-set or cached values in the routing entry. 1902 * 1903 * Also take into account the space needed for options that we 1904 * send regularly. Make maxseg shorter by that amount to assure 1905 * that we can send maxseg amount of data even when the options 1906 * are present. Store the upper limit of the length of options plus 1907 * data in maxopd. 1908 * 1909 * NOTE that this routine is only called when we process an incoming 1910 * segment, for outgoing segments only tcp_mssopt is called. 1911 * 1912 * In case of T/TCP, we call this routine during implicit connection 1913 * setup as well (offer = -1), to initialize maxseg from the cached 1914 * MSS of our peer. 1915 */ 1916 void 1917 tcp_mss(tp, offer) 1918 struct tcpcb *tp; 1919 int offer; 1920 { 1921 register struct rtentry *rt; 1922 struct ifnet *ifp; 1923 register int rtt, mss; 1924 u_long bufsize; 1925 struct inpcb *inp; 1926 struct socket *so; 1927 struct rmxp_tao *taop; 1928 int origoffer = offer; 1929 1930 inp = tp->t_inpcb; 1931 if ((rt = tcp_rtlookup(inp)) == NULL) { 1932 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1933 return; 1934 } 1935 ifp = rt->rt_ifp; 1936 so = inp->inp_socket; 1937 1938 taop = rmx_taop(rt->rt_rmx); 1939 /* 1940 * Offer == -1 means that we didn't receive SYN yet, 1941 * use cached value in that case; 1942 */ 1943 if (offer == -1) 1944 offer = taop->tao_mssopt; 1945 /* 1946 * Offer == 0 means that there was no MSS on the SYN segment, 1947 * in this case we use tcp_mssdflt. 1948 */ 1949 if (offer == 0) 1950 offer = tcp_mssdflt; 1951 else 1952 /* 1953 * Sanity check: make sure that maxopd will be large 1954 * enough to allow some data on segments even is the 1955 * all the option space is used (40bytes). Otherwise 1956 * funny things may happen in tcp_output. 1957 */ 1958 offer = max(offer, 64); 1959 taop->tao_mssopt = offer; 1960 1961 /* 1962 * While we're here, check if there's an initial rtt 1963 * or rttvar. Convert from the route-table units 1964 * to scaled multiples of the slow timeout timer. 1965 */ 1966 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1967 /* 1968 * XXX the lock bit for RTT indicates that the value 1969 * is also a minimum value; this is subject to time. 1970 */ 1971 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1972 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1973 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1974 tcpstat.tcps_usedrtt++; 1975 if (rt->rt_rmx.rmx_rttvar) { 1976 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1977 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1978 tcpstat.tcps_usedrttvar++; 1979 } else { 1980 /* default variation is +- 1 rtt */ 1981 tp->t_rttvar = 1982 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1983 } 1984 TCPT_RANGESET(tp->t_rxtcur, 1985 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1986 tp->t_rttmin, TCPTV_REXMTMAX); 1987 } 1988 /* 1989 * if there's an mtu associated with the route, use it 1990 */ 1991 if (rt->rt_rmx.rmx_mtu) 1992 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1993 else 1994 { 1995 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1996 if (!in_localaddr(inp->inp_faddr)) 1997 mss = min(mss, tcp_mssdflt); 1998 } 1999 mss = min(mss, offer); 2000 /* 2001 * maxopd stores the maximum length of data AND options 2002 * in a segment; maxseg is the amount of data in a normal 2003 * segment. We need to store this value (maxopd) apart 2004 * from maxseg, because now every segment carries options 2005 * and thus we normally have somewhat less data in segments. 2006 */ 2007 tp->t_maxopd = mss; 2008 2009 /* 2010 * In case of T/TCP, origoffer==-1 indicates, that no segments 2011 * were received yet. In this case we just guess, otherwise 2012 * we do the same as before T/TCP. 2013 */ 2014 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2015 (origoffer == -1 || 2016 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2017 mss -= TCPOLEN_TSTAMP_APPA; 2018 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2019 (origoffer == -1 || 2020 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2021 mss -= TCPOLEN_CC_APPA; 2022 2023 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2024 if (mss > MCLBYTES) 2025 mss &= ~(MCLBYTES-1); 2026 #else 2027 if (mss > MCLBYTES) 2028 mss = mss / MCLBYTES * MCLBYTES; 2029 #endif 2030 /* 2031 * If there's a pipesize, change the socket buffer 2032 * to that size. Make the socket buffers an integral 2033 * number of mss units; if the mss is larger than 2034 * the socket buffer, decrease the mss. 2035 */ 2036 #ifdef RTV_SPIPE 2037 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2038 #endif 2039 bufsize = so->so_snd.sb_hiwat; 2040 if (bufsize < mss) 2041 mss = bufsize; 2042 else { 2043 bufsize = roundup(bufsize, mss); 2044 if (bufsize > sb_max) 2045 bufsize = sb_max; 2046 (void)sbreserve(&so->so_snd, bufsize); 2047 } 2048 tp->t_maxseg = mss; 2049 2050 #ifdef RTV_RPIPE 2051 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2052 #endif 2053 bufsize = so->so_rcv.sb_hiwat; 2054 if (bufsize > mss) { 2055 bufsize = roundup(bufsize, mss); 2056 if (bufsize > sb_max) 2057 bufsize = sb_max; 2058 (void)sbreserve(&so->so_rcv, bufsize); 2059 } 2060 /* 2061 * Don't force slow-start on local network. 2062 */ 2063 if (!in_localaddr(inp->inp_faddr)) 2064 tp->snd_cwnd = mss; 2065 2066 if (rt->rt_rmx.rmx_ssthresh) { 2067 /* 2068 * There's some sort of gateway or interface 2069 * buffer limit on the path. Use this to set 2070 * the slow start threshhold, but set the 2071 * threshold to no less than 2*mss. 2072 */ 2073 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2074 tcpstat.tcps_usedssthresh++; 2075 } 2076 } 2077 2078 /* 2079 * Determine the MSS option to send on an outgoing SYN. 2080 */ 2081 int 2082 tcp_mssopt(tp) 2083 struct tcpcb *tp; 2084 { 2085 struct rtentry *rt; 2086 2087 rt = tcp_rtlookup(tp->t_inpcb); 2088 if (rt == NULL) 2089 return tcp_mssdflt; 2090 2091 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2092 } 2093 #endif /* TUBA_INCLUDE */ 2094