1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * from: @(#)tcp_input.c 7.25 (Berkeley) 6/30/90 34 * $Id: tcp_input.c,v 1.2 1993/05/18 18:20:15 cgd Exp $ 35 */ 36 37 #include "param.h" 38 #include "systm.h" 39 #include "malloc.h" 40 #include "select.h" 41 #include "mbuf.h" 42 #include "protosw.h" 43 #include "socket.h" 44 #include "socketvar.h" 45 #include "errno.h" 46 47 #include "../net/if.h" 48 #include "../net/route.h" 49 50 #include "in.h" 51 #include "in_systm.h" 52 #include "ip.h" 53 #include "in_pcb.h" 54 #include "ip_var.h" 55 #include "tcp.h" 56 #include "tcp_fsm.h" 57 #include "tcp_seq.h" 58 #include "tcp_timer.h" 59 #include "tcp_var.h" 60 #include "tcpip.h" 61 #include "tcp_debug.h" 62 63 int tcprexmtthresh = 3; 64 int tcppredack; /* XXX debugging: times hdr predict ok for acks */ 65 int tcppreddat; /* XXX # times header prediction ok for data packets */ 66 int tcppcbcachemiss; 67 struct tcpiphdr tcp_saveti; 68 struct inpcb *tcp_last_inpcb = &tcb; 69 70 struct tcpcb *tcp_newtcpcb(); 71 72 /* 73 * Insert segment ti into reassembly queue of tcp with 74 * control block tp. Return TH_FIN if reassembly now includes 75 * a segment with FIN. The macro form does the common case inline 76 * (segment is the next to be received on an established connection, 77 * and the queue is empty), avoiding linkage into and removal 78 * from the queue and repetition of various conversions. 79 * Set DELACK for segments received in order, but ack immediately 80 * when segments are out of order (so fast retransmit can work). 81 */ 82 #define TCP_REASS(tp, ti, m, so, flags) { \ 83 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 84 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 85 (tp)->t_state == TCPS_ESTABLISHED) { \ 86 tp->t_flags |= TF_DELACK; \ 87 (tp)->rcv_nxt += (ti)->ti_len; \ 88 flags = (ti)->ti_flags & TH_FIN; \ 89 tcpstat.tcps_rcvpack++;\ 90 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 91 sbappend(&(so)->so_rcv, (m)); \ 92 sorwakeup(so); \ 93 } else { \ 94 (flags) = tcp_reass((tp), (ti), (m)); \ 95 tp->t_flags |= TF_ACKNOW; \ 96 } \ 97 } 98 99 tcp_reass(tp, ti, m) 100 register struct tcpcb *tp; 101 register struct tcpiphdr *ti; 102 struct mbuf *m; 103 { 104 register struct tcpiphdr *q; 105 struct socket *so = tp->t_inpcb->inp_socket; 106 int flags; 107 108 /* 109 * Call with ti==0 after become established to 110 * force pre-ESTABLISHED data up to user socket. 111 */ 112 if (ti == 0) 113 goto present; 114 115 /* 116 * Find a segment which begins after this one does. 117 */ 118 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 119 q = (struct tcpiphdr *)q->ti_next) 120 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 121 break; 122 123 /* 124 * If there is a preceding segment, it may provide some of 125 * our data already. If so, drop the data from the incoming 126 * segment. If it provides all of our data, drop us. 127 */ 128 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 129 register int i; 130 q = (struct tcpiphdr *)q->ti_prev; 131 /* conversion to int (in i) handles seq wraparound */ 132 i = q->ti_seq + q->ti_len - ti->ti_seq; 133 if (i > 0) { 134 if (i >= ti->ti_len) { 135 tcpstat.tcps_rcvduppack++; 136 tcpstat.tcps_rcvdupbyte += ti->ti_len; 137 m_freem(m); 138 return (0); 139 } 140 m_adj(m, i); 141 ti->ti_len -= i; 142 ti->ti_seq += i; 143 } 144 q = (struct tcpiphdr *)(q->ti_next); 145 } 146 tcpstat.tcps_rcvoopack++; 147 tcpstat.tcps_rcvoobyte += ti->ti_len; 148 REASS_MBUF(ti) = m; /* XXX */ 149 150 /* 151 * While we overlap succeeding segments trim them or, 152 * if they are completely covered, dequeue them. 153 */ 154 while (q != (struct tcpiphdr *)tp) { 155 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 156 if (i <= 0) 157 break; 158 if (i < q->ti_len) { 159 q->ti_seq += i; 160 q->ti_len -= i; 161 m_adj(REASS_MBUF(q), i); 162 break; 163 } 164 q = (struct tcpiphdr *)q->ti_next; 165 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 166 remque(q->ti_prev); 167 m_freem(m); 168 } 169 170 /* 171 * Stick new segment in its place. 172 */ 173 insque(ti, q->ti_prev); 174 175 present: 176 /* 177 * Present data to user, advancing rcv_nxt through 178 * completed sequence space. 179 */ 180 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 181 return (0); 182 ti = tp->seg_next; 183 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 184 return (0); 185 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 186 return (0); 187 do { 188 tp->rcv_nxt += ti->ti_len; 189 flags = ti->ti_flags & TH_FIN; 190 remque(ti); 191 m = REASS_MBUF(ti); 192 ti = (struct tcpiphdr *)ti->ti_next; 193 if (so->so_state & SS_CANTRCVMORE) 194 m_freem(m); 195 else 196 sbappend(&so->so_rcv, m); 197 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 198 sorwakeup(so); 199 return (flags); 200 } 201 202 /* 203 * TCP input routine, follows pages 65-76 of the 204 * protocol specification dated September, 1981 very closely. 205 */ 206 tcp_input(m, iphlen) 207 register struct mbuf *m; 208 int iphlen; 209 { 210 register struct tcpiphdr *ti; 211 register struct inpcb *inp; 212 struct mbuf *om = 0; 213 int len, tlen, off; 214 register struct tcpcb *tp = 0; 215 register int tiflags; 216 struct socket *so; 217 int todrop, acked, ourfinisacked, needoutput = 0; 218 short ostate; 219 struct in_addr laddr; 220 int dropsocket = 0; 221 int iss = 0; 222 223 tcpstat.tcps_rcvtotal++; 224 /* 225 * Get IP and TCP header together in first mbuf. 226 * Note: IP leaves IP header in first mbuf. 227 */ 228 ti = mtod(m, struct tcpiphdr *); 229 if (iphlen > sizeof (struct ip)) 230 ip_stripoptions(m, (struct mbuf *)0); 231 if (m->m_len < sizeof (struct tcpiphdr)) { 232 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 233 tcpstat.tcps_rcvshort++; 234 return; 235 } 236 ti = mtod(m, struct tcpiphdr *); 237 } 238 239 /* 240 * Checksum extended TCP header and data. 241 */ 242 tlen = ((struct ip *)ti)->ip_len; 243 len = sizeof (struct ip) + tlen; 244 ti->ti_next = ti->ti_prev = 0; 245 ti->ti_x1 = 0; 246 ti->ti_len = (u_short)tlen; 247 HTONS(ti->ti_len); 248 if (ti->ti_sum = in_cksum(m, len)) { 249 tcpstat.tcps_rcvbadsum++; 250 goto drop; 251 } 252 253 /* 254 * Check that TCP offset makes sense, 255 * pull out TCP options and adjust length. XXX 256 */ 257 off = ti->ti_off << 2; 258 if (off < sizeof (struct tcphdr) || off > tlen) { 259 tcpstat.tcps_rcvbadoff++; 260 goto drop; 261 } 262 tlen -= off; 263 ti->ti_len = tlen; 264 if (off > sizeof (struct tcphdr)) { 265 if (m->m_len < sizeof(struct ip) + off) { 266 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 267 tcpstat.tcps_rcvshort++; 268 return; 269 } 270 ti = mtod(m, struct tcpiphdr *); 271 } 272 om = m_get(M_DONTWAIT, MT_DATA); 273 if (om == 0) 274 goto drop; 275 om->m_len = off - sizeof (struct tcphdr); 276 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 277 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 278 m->m_len -= om->m_len; 279 m->m_pkthdr.len -= om->m_len; 280 bcopy(op+om->m_len, op, 281 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 282 } 283 } 284 tiflags = ti->ti_flags; 285 286 /* 287 * Convert TCP protocol specific fields to host format. 288 */ 289 NTOHL(ti->ti_seq); 290 NTOHL(ti->ti_ack); 291 NTOHS(ti->ti_win); 292 NTOHS(ti->ti_urp); 293 294 /* 295 * Locate pcb for segment. 296 */ 297 findpcb: 298 inp = tcp_last_inpcb; 299 if (inp->inp_lport != ti->ti_dport || 300 inp->inp_fport != ti->ti_sport || 301 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 302 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 303 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 304 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 305 if (inp) 306 tcp_last_inpcb = inp; 307 ++tcppcbcachemiss; 308 } 309 310 /* 311 * If the state is CLOSED (i.e., TCB does not exist) then 312 * all data in the incoming segment is discarded. 313 * If the TCB exists but is in CLOSED state, it is embryonic, 314 * but should either do a listen or a connect soon. 315 */ 316 if (inp == 0) 317 goto dropwithreset; 318 tp = intotcpcb(inp); 319 if (tp == 0) 320 goto dropwithreset; 321 if (tp->t_state == TCPS_CLOSED) 322 goto drop; 323 so = inp->inp_socket; 324 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 325 if (so->so_options & SO_DEBUG) { 326 ostate = tp->t_state; 327 tcp_saveti = *ti; 328 } 329 if (so->so_options & SO_ACCEPTCONN) { 330 so = sonewconn(so, 0); 331 if (so == 0) 332 goto drop; 333 /* 334 * This is ugly, but .... 335 * 336 * Mark socket as temporary until we're 337 * committed to keeping it. The code at 338 * ``drop'' and ``dropwithreset'' check the 339 * flag dropsocket to see if the temporary 340 * socket created here should be discarded. 341 * We mark the socket as discardable until 342 * we're committed to it below in TCPS_LISTEN. 343 */ 344 dropsocket++; 345 inp = (struct inpcb *)so->so_pcb; 346 inp->inp_laddr = ti->ti_dst; 347 inp->inp_lport = ti->ti_dport; 348 #if BSD>=43 349 inp->inp_options = ip_srcroute(); 350 #endif 351 tp = intotcpcb(inp); 352 tp->t_state = TCPS_LISTEN; 353 } 354 } 355 356 /* 357 * Segment received on connection. 358 * Reset idle time and keep-alive timer. 359 */ 360 tp->t_idle = 0; 361 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 362 363 /* 364 * Process options if not in LISTEN state, 365 * else do it below (after getting remote address). 366 */ 367 if (om && tp->t_state != TCPS_LISTEN) { 368 tcp_dooptions(tp, om, ti); 369 om = 0; 370 } 371 /* 372 * Header prediction: check for the two common cases 373 * of a uni-directional data xfer. If the packet has 374 * no control flags, is in-sequence, the window didn't 375 * change and we're not retransmitting, it's a 376 * candidate. If the length is zero and the ack moved 377 * forward, we're the sender side of the xfer. Just 378 * free the data acked & wake any higher level process 379 * that was blocked waiting for space. If the length 380 * is non-zero and the ack didn't move, we're the 381 * receiver side. If we're getting packets in-order 382 * (the reassembly queue is empty), add the data to 383 * the socket buffer and note that we need a delayed ack. 384 */ 385 if (tp->t_state == TCPS_ESTABLISHED && 386 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 387 ti->ti_seq == tp->rcv_nxt && 388 ti->ti_win && ti->ti_win == tp->snd_wnd && 389 tp->snd_nxt == tp->snd_max) { 390 if (ti->ti_len == 0) { 391 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 392 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 393 tp->snd_cwnd >= tp->snd_wnd) { 394 /* 395 * this is a pure ack for outstanding data. 396 */ 397 ++tcppredack; 398 if (tp->t_rtt && SEQ_GT(ti->ti_ack,tp->t_rtseq)) 399 tcp_xmit_timer(tp); 400 acked = ti->ti_ack - tp->snd_una; 401 tcpstat.tcps_rcvackpack++; 402 tcpstat.tcps_rcvackbyte += acked; 403 sbdrop(&so->so_snd, acked); 404 tp->snd_una = ti->ti_ack; 405 m_freem(m); 406 407 /* 408 * If all outstanding data are acked, stop 409 * retransmit timer, otherwise restart timer 410 * using current (possibly backed-off) value. 411 * If process is waiting for space, 412 * wakeup/selwakeup/signal. If data 413 * are ready to send, let tcp_output 414 * decide between more output or persist. 415 */ 416 if (tp->snd_una == tp->snd_max) 417 tp->t_timer[TCPT_REXMT] = 0; 418 else if (tp->t_timer[TCPT_PERSIST] == 0) 419 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 420 421 if (so->so_snd.sb_flags & SB_NOTIFY) 422 sowwakeup(so); 423 if (so->so_snd.sb_cc) 424 (void) tcp_output(tp); 425 return; 426 } 427 } else if (ti->ti_ack == tp->snd_una && 428 tp->seg_next == (struct tcpiphdr *)tp && 429 ti->ti_len <= sbspace(&so->so_rcv)) { 430 /* 431 * this is a pure, in-sequence data packet 432 * with nothing on the reassembly queue and 433 * we have enough buffer space to take it. 434 */ 435 ++tcppreddat; 436 tp->rcv_nxt += ti->ti_len; 437 tcpstat.tcps_rcvpack++; 438 tcpstat.tcps_rcvbyte += ti->ti_len; 439 /* 440 * Drop TCP and IP headers then add data 441 * to socket buffer 442 */ 443 m->m_data += sizeof(struct tcpiphdr); 444 m->m_len -= sizeof(struct tcpiphdr); 445 sbappend(&so->so_rcv, m); 446 sorwakeup(so); 447 tp->t_flags |= TF_DELACK; 448 return; 449 } 450 } 451 452 /* 453 * Drop TCP and IP headers; TCP options were dropped above. 454 */ 455 m->m_data += sizeof(struct tcpiphdr); 456 m->m_len -= sizeof(struct tcpiphdr); 457 458 /* 459 * Calculate amount of space in receive window, 460 * and then do TCP input processing. 461 * Receive window is amount of space in rcv queue, 462 * but not less than advertised window. 463 */ 464 { int win; 465 466 win = sbspace(&so->so_rcv); 467 if (win < 0) 468 win = 0; 469 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 470 } 471 472 switch (tp->t_state) { 473 474 /* 475 * If the state is LISTEN then ignore segment if it contains an RST. 476 * If the segment contains an ACK then it is bad and send a RST. 477 * If it does not contain a SYN then it is not interesting; drop it. 478 * Don't bother responding if the destination was a broadcast. 479 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 480 * tp->iss, and send a segment: 481 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 482 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 483 * Fill in remote peer address fields if not previously specified. 484 * Enter SYN_RECEIVED state, and process any other fields of this 485 * segment in this state. 486 */ 487 case TCPS_LISTEN: { 488 struct mbuf *am; 489 register struct sockaddr_in *sin; 490 491 if (tiflags & TH_RST) 492 goto drop; 493 if (tiflags & TH_ACK) 494 goto dropwithreset; 495 if ((tiflags & TH_SYN) == 0) 496 goto drop; 497 if (m->m_flags & M_BCAST) 498 goto drop; 499 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 500 if (am == NULL) 501 goto drop; 502 am->m_len = sizeof (struct sockaddr_in); 503 sin = mtod(am, struct sockaddr_in *); 504 sin->sin_family = AF_INET; 505 sin->sin_len = sizeof(*sin); 506 sin->sin_addr = ti->ti_src; 507 sin->sin_port = ti->ti_sport; 508 laddr = inp->inp_laddr; 509 if (inp->inp_laddr.s_addr == INADDR_ANY) 510 inp->inp_laddr = ti->ti_dst; 511 if (in_pcbconnect(inp, am)) { 512 inp->inp_laddr = laddr; 513 (void) m_free(am); 514 goto drop; 515 } 516 (void) m_free(am); 517 tp->t_template = tcp_template(tp); 518 if (tp->t_template == 0) { 519 tp = tcp_drop(tp, ENOBUFS); 520 dropsocket = 0; /* socket is already gone */ 521 goto drop; 522 } 523 if (om) { 524 tcp_dooptions(tp, om, ti); 525 om = 0; 526 } 527 if (iss) 528 tp->iss = iss; 529 else 530 tp->iss = tcp_iss; 531 tcp_iss += TCP_ISSINCR/2; 532 tp->irs = ti->ti_seq; 533 tcp_sendseqinit(tp); 534 tcp_rcvseqinit(tp); 535 tp->t_flags |= TF_ACKNOW; 536 tp->t_state = TCPS_SYN_RECEIVED; 537 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 538 dropsocket = 0; /* committed to socket */ 539 tcpstat.tcps_accepts++; 540 goto trimthenstep6; 541 } 542 543 /* 544 * If the state is SYN_SENT: 545 * if seg contains an ACK, but not for our SYN, drop the input. 546 * if seg contains a RST, then drop the connection. 547 * if seg does not contain SYN, then drop it. 548 * Otherwise this is an acceptable SYN segment 549 * initialize tp->rcv_nxt and tp->irs 550 * if seg contains ack then advance tp->snd_una 551 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 552 * arrange for segment to be acked (eventually) 553 * continue processing rest of data/controls, beginning with URG 554 */ 555 case TCPS_SYN_SENT: 556 if ((tiflags & TH_ACK) && 557 (SEQ_LEQ(ti->ti_ack, tp->iss) || 558 SEQ_GT(ti->ti_ack, tp->snd_max))) 559 goto dropwithreset; 560 if (tiflags & TH_RST) { 561 if (tiflags & TH_ACK) 562 tp = tcp_drop(tp, ECONNREFUSED); 563 goto drop; 564 } 565 if ((tiflags & TH_SYN) == 0) 566 goto drop; 567 if (tiflags & TH_ACK) { 568 tp->snd_una = ti->ti_ack; 569 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 570 tp->snd_nxt = tp->snd_una; 571 } 572 tp->t_timer[TCPT_REXMT] = 0; 573 tp->irs = ti->ti_seq; 574 tcp_rcvseqinit(tp); 575 tp->t_flags |= TF_ACKNOW; 576 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 577 tcpstat.tcps_connects++; 578 soisconnected(so); 579 tp->t_state = TCPS_ESTABLISHED; 580 (void) tcp_reass(tp, (struct tcpiphdr *)0, 581 (struct mbuf *)0); 582 /* 583 * if we didn't have to retransmit the SYN, 584 * use its rtt as our initial srtt & rtt var. 585 */ 586 if (tp->t_rtt) 587 tcp_xmit_timer(tp); 588 } else 589 tp->t_state = TCPS_SYN_RECEIVED; 590 591 trimthenstep6: 592 /* 593 * Advance ti->ti_seq to correspond to first data byte. 594 * If data, trim to stay within window, 595 * dropping FIN if necessary. 596 */ 597 ti->ti_seq++; 598 if (ti->ti_len > tp->rcv_wnd) { 599 todrop = ti->ti_len - tp->rcv_wnd; 600 m_adj(m, -todrop); 601 ti->ti_len = tp->rcv_wnd; 602 tiflags &= ~TH_FIN; 603 tcpstat.tcps_rcvpackafterwin++; 604 tcpstat.tcps_rcvbyteafterwin += todrop; 605 } 606 tp->snd_wl1 = ti->ti_seq - 1; 607 tp->rcv_up = ti->ti_seq; 608 goto step6; 609 } 610 611 /* 612 * States other than LISTEN or SYN_SENT. 613 * First check that at least some bytes of segment are within 614 * receive window. If segment begins before rcv_nxt, 615 * drop leading data (and SYN); if nothing left, just ack. 616 */ 617 todrop = tp->rcv_nxt - ti->ti_seq; 618 if (todrop > 0) { 619 if (tiflags & TH_SYN) { 620 tiflags &= ~TH_SYN; 621 ti->ti_seq++; 622 if (ti->ti_urp > 1) 623 ti->ti_urp--; 624 else 625 tiflags &= ~TH_URG; 626 todrop--; 627 } 628 if (todrop > ti->ti_len || 629 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 630 tcpstat.tcps_rcvduppack++; 631 tcpstat.tcps_rcvdupbyte += ti->ti_len; 632 /* 633 * If segment is just one to the left of the window, 634 * check two special cases: 635 * 1. Don't toss RST in response to 4.2-style keepalive. 636 * 2. If the only thing to drop is a FIN, we can drop 637 * it, but check the ACK or we will get into FIN 638 * wars if our FINs crossed (both CLOSING). 639 * In either case, send ACK to resynchronize, 640 * but keep on processing for RST or ACK. 641 */ 642 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 643 #ifdef TCP_COMPAT_42 644 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 645 #endif 646 ) { 647 todrop = ti->ti_len; 648 tiflags &= ~TH_FIN; 649 tp->t_flags |= TF_ACKNOW; 650 } else 651 goto dropafterack; 652 } else { 653 tcpstat.tcps_rcvpartduppack++; 654 tcpstat.tcps_rcvpartdupbyte += todrop; 655 } 656 m_adj(m, todrop); 657 ti->ti_seq += todrop; 658 ti->ti_len -= todrop; 659 if (ti->ti_urp > todrop) 660 ti->ti_urp -= todrop; 661 else { 662 tiflags &= ~TH_URG; 663 ti->ti_urp = 0; 664 } 665 } 666 667 /* 668 * If new data are received on a connection after the 669 * user processes are gone, then RST the other end. 670 */ 671 if ((so->so_state & SS_NOFDREF) && 672 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 673 tp = tcp_close(tp); 674 tcpstat.tcps_rcvafterclose++; 675 goto dropwithreset; 676 } 677 678 /* 679 * If segment ends after window, drop trailing data 680 * (and PUSH and FIN); if nothing left, just ACK. 681 */ 682 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 683 if (todrop > 0) { 684 tcpstat.tcps_rcvpackafterwin++; 685 if (todrop >= ti->ti_len) { 686 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 687 /* 688 * If a new connection request is received 689 * while in TIME_WAIT, drop the old connection 690 * and start over if the sequence numbers 691 * are above the previous ones. 692 */ 693 if (tiflags & TH_SYN && 694 tp->t_state == TCPS_TIME_WAIT && 695 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 696 iss = tp->rcv_nxt + TCP_ISSINCR; 697 tp = tcp_close(tp); 698 goto findpcb; 699 } 700 /* 701 * If window is closed can only take segments at 702 * window edge, and have to drop data and PUSH from 703 * incoming segments. Continue processing, but 704 * remember to ack. Otherwise, drop segment 705 * and ack. 706 */ 707 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 708 tp->t_flags |= TF_ACKNOW; 709 tcpstat.tcps_rcvwinprobe++; 710 } else 711 goto dropafterack; 712 } else 713 tcpstat.tcps_rcvbyteafterwin += todrop; 714 m_adj(m, -todrop); 715 ti->ti_len -= todrop; 716 tiflags &= ~(TH_PUSH|TH_FIN); 717 } 718 719 /* 720 * If the RST bit is set examine the state: 721 * SYN_RECEIVED STATE: 722 * If passive open, return to LISTEN state. 723 * If active open, inform user that connection was refused. 724 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 725 * Inform user that connection was reset, and close tcb. 726 * CLOSING, LAST_ACK, TIME_WAIT STATES 727 * Close the tcb. 728 */ 729 if (tiflags&TH_RST) switch (tp->t_state) { 730 731 case TCPS_SYN_RECEIVED: 732 so->so_error = ECONNREFUSED; 733 goto close; 734 735 case TCPS_ESTABLISHED: 736 case TCPS_FIN_WAIT_1: 737 case TCPS_FIN_WAIT_2: 738 case TCPS_CLOSE_WAIT: 739 so->so_error = ECONNRESET; 740 close: 741 tp->t_state = TCPS_CLOSED; 742 tcpstat.tcps_drops++; 743 tp = tcp_close(tp); 744 goto drop; 745 746 case TCPS_CLOSING: 747 case TCPS_LAST_ACK: 748 case TCPS_TIME_WAIT: 749 tp = tcp_close(tp); 750 goto drop; 751 } 752 753 /* 754 * If a SYN is in the window, then this is an 755 * error and we send an RST and drop the connection. 756 */ 757 if (tiflags & TH_SYN) { 758 tp = tcp_drop(tp, ECONNRESET); 759 goto dropwithreset; 760 } 761 762 /* 763 * If the ACK bit is off we drop the segment and return. 764 */ 765 if ((tiflags & TH_ACK) == 0) 766 goto drop; 767 768 /* 769 * Ack processing. 770 */ 771 switch (tp->t_state) { 772 773 /* 774 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 775 * ESTABLISHED state and continue processing, otherwise 776 * send an RST. 777 */ 778 case TCPS_SYN_RECEIVED: 779 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 780 SEQ_GT(ti->ti_ack, tp->snd_max)) 781 goto dropwithreset; 782 tcpstat.tcps_connects++; 783 soisconnected(so); 784 tp->t_state = TCPS_ESTABLISHED; 785 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 786 tp->snd_wl1 = ti->ti_seq - 1; 787 /* fall into ... */ 788 789 /* 790 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 791 * ACKs. If the ack is in the range 792 * tp->snd_una < ti->ti_ack <= tp->snd_max 793 * then advance tp->snd_una to ti->ti_ack and drop 794 * data from the retransmission queue. If this ACK reflects 795 * more up to date window information we update our window information. 796 */ 797 case TCPS_ESTABLISHED: 798 case TCPS_FIN_WAIT_1: 799 case TCPS_FIN_WAIT_2: 800 case TCPS_CLOSE_WAIT: 801 case TCPS_CLOSING: 802 case TCPS_LAST_ACK: 803 case TCPS_TIME_WAIT: 804 805 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 806 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 807 tcpstat.tcps_rcvdupack++; 808 /* 809 * If we have outstanding data (other than 810 * a window probe), this is a completely 811 * duplicate ack (ie, window info didn't 812 * change), the ack is the biggest we've 813 * seen and we've seen exactly our rexmt 814 * threshhold of them, assume a packet 815 * has been dropped and retransmit it. 816 * Kludge snd_nxt & the congestion 817 * window so we send only this one 818 * packet. 819 * 820 * We know we're losing at the current 821 * window size so do congestion avoidance 822 * (set ssthresh to half the current window 823 * and pull our congestion window back to 824 * the new ssthresh). 825 * 826 * Dup acks mean that packets have left the 827 * network (they're now cached at the receiver) 828 * so bump cwnd by the amount in the receiver 829 * to keep a constant cwnd packets in the 830 * network. 831 */ 832 if (tp->t_timer[TCPT_REXMT] == 0 || 833 ti->ti_ack != tp->snd_una) 834 tp->t_dupacks = 0; 835 else if (++tp->t_dupacks == tcprexmtthresh) { 836 tcp_seq onxt = tp->snd_nxt; 837 u_int win = 838 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 839 tp->t_maxseg; 840 841 if (win < 2) 842 win = 2; 843 tp->snd_ssthresh = win * tp->t_maxseg; 844 tp->t_timer[TCPT_REXMT] = 0; 845 tp->t_rtt = 0; 846 tp->snd_nxt = ti->ti_ack; 847 tp->snd_cwnd = tp->t_maxseg; 848 (void) tcp_output(tp); 849 tp->snd_cwnd = tp->snd_ssthresh + 850 tp->t_maxseg * tp->t_dupacks; 851 if (SEQ_GT(onxt, tp->snd_nxt)) 852 tp->snd_nxt = onxt; 853 goto drop; 854 } else if (tp->t_dupacks > tcprexmtthresh) { 855 tp->snd_cwnd += tp->t_maxseg; 856 (void) tcp_output(tp); 857 goto drop; 858 } 859 } else 860 tp->t_dupacks = 0; 861 break; 862 } 863 /* 864 * If the congestion window was inflated to account 865 * for the other side's cached packets, retract it. 866 */ 867 if (tp->t_dupacks > tcprexmtthresh && 868 tp->snd_cwnd > tp->snd_ssthresh) 869 tp->snd_cwnd = tp->snd_ssthresh; 870 tp->t_dupacks = 0; 871 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 872 tcpstat.tcps_rcvacktoomuch++; 873 goto dropafterack; 874 } 875 acked = ti->ti_ack - tp->snd_una; 876 tcpstat.tcps_rcvackpack++; 877 tcpstat.tcps_rcvackbyte += acked; 878 879 /* 880 * If transmit timer is running and timed sequence 881 * number was acked, update smoothed round trip time. 882 * Since we now have an rtt measurement, cancel the 883 * timer backoff (cf., Phil Karn's retransmit alg.). 884 * Recompute the initial retransmit timer. 885 */ 886 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 887 tcp_xmit_timer(tp); 888 889 /* 890 * If all outstanding data is acked, stop retransmit 891 * timer and remember to restart (more output or persist). 892 * If there is more data to be acked, restart retransmit 893 * timer, using current (possibly backed-off) value. 894 */ 895 if (ti->ti_ack == tp->snd_max) { 896 tp->t_timer[TCPT_REXMT] = 0; 897 needoutput = 1; 898 } else if (tp->t_timer[TCPT_PERSIST] == 0) 899 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 900 /* 901 * When new data is acked, open the congestion window. 902 * If the window gives us less than ssthresh packets 903 * in flight, open exponentially (maxseg per packet). 904 * Otherwise open linearly: maxseg per window 905 * (maxseg^2 / cwnd per packet), plus a constant 906 * fraction of a packet (maxseg/8) to help larger windows 907 * open quickly enough. 908 */ 909 { 910 register u_int cw = tp->snd_cwnd; 911 register u_int incr = tp->t_maxseg; 912 913 if (cw > tp->snd_ssthresh) 914 incr = incr * incr / cw + incr / 8; 915 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN); 916 } 917 if (acked > so->so_snd.sb_cc) { 918 tp->snd_wnd -= so->so_snd.sb_cc; 919 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 920 ourfinisacked = 1; 921 } else { 922 sbdrop(&so->so_snd, acked); 923 tp->snd_wnd -= acked; 924 ourfinisacked = 0; 925 } 926 if (so->so_snd.sb_flags & SB_NOTIFY) 927 sowwakeup(so); 928 tp->snd_una = ti->ti_ack; 929 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 930 tp->snd_nxt = tp->snd_una; 931 932 switch (tp->t_state) { 933 934 /* 935 * In FIN_WAIT_1 STATE in addition to the processing 936 * for the ESTABLISHED state if our FIN is now acknowledged 937 * then enter FIN_WAIT_2. 938 */ 939 case TCPS_FIN_WAIT_1: 940 if (ourfinisacked) { 941 /* 942 * If we can't receive any more 943 * data, then closing user can proceed. 944 * Starting the timer is contrary to the 945 * specification, but if we don't get a FIN 946 * we'll hang forever. 947 */ 948 if (so->so_state & SS_CANTRCVMORE) { 949 soisdisconnected(so); 950 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 951 } 952 tp->t_state = TCPS_FIN_WAIT_2; 953 } 954 break; 955 956 /* 957 * In CLOSING STATE in addition to the processing for 958 * the ESTABLISHED state if the ACK acknowledges our FIN 959 * then enter the TIME-WAIT state, otherwise ignore 960 * the segment. 961 */ 962 case TCPS_CLOSING: 963 if (ourfinisacked) { 964 tp->t_state = TCPS_TIME_WAIT; 965 tcp_canceltimers(tp); 966 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 967 soisdisconnected(so); 968 } 969 break; 970 971 /* 972 * In LAST_ACK, we may still be waiting for data to drain 973 * and/or to be acked, as well as for the ack of our FIN. 974 * If our FIN is now acknowledged, delete the TCB, 975 * enter the closed state and return. 976 */ 977 case TCPS_LAST_ACK: 978 if (ourfinisacked) { 979 tp = tcp_close(tp); 980 goto drop; 981 } 982 break; 983 984 /* 985 * In TIME_WAIT state the only thing that should arrive 986 * is a retransmission of the remote FIN. Acknowledge 987 * it and restart the finack timer. 988 */ 989 case TCPS_TIME_WAIT: 990 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 991 goto dropafterack; 992 } 993 } 994 995 step6: 996 /* 997 * Update window information. 998 * Don't look at window if no ACK: TAC's send garbage on first SYN. 999 */ 1000 if ((tiflags & TH_ACK) && 1001 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1002 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1003 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 1004 /* keep track of pure window updates */ 1005 if (ti->ti_len == 0 && 1006 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 1007 tcpstat.tcps_rcvwinupd++; 1008 tp->snd_wnd = ti->ti_win; 1009 tp->snd_wl1 = ti->ti_seq; 1010 tp->snd_wl2 = ti->ti_ack; 1011 if (tp->snd_wnd > tp->max_sndwnd) 1012 tp->max_sndwnd = tp->snd_wnd; 1013 needoutput = 1; 1014 } 1015 1016 /* 1017 * Process segments with URG. 1018 */ 1019 if ((tiflags & TH_URG) && ti->ti_urp && 1020 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1021 /* 1022 * This is a kludge, but if we receive and accept 1023 * random urgent pointers, we'll crash in 1024 * soreceive. It's hard to imagine someone 1025 * actually wanting to send this much urgent data. 1026 */ 1027 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 1028 ti->ti_urp = 0; /* XXX */ 1029 tiflags &= ~TH_URG; /* XXX */ 1030 goto dodata; /* XXX */ 1031 } 1032 /* 1033 * If this segment advances the known urgent pointer, 1034 * then mark the data stream. This should not happen 1035 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1036 * a FIN has been received from the remote side. 1037 * In these states we ignore the URG. 1038 * 1039 * According to RFC961 (Assigned Protocols), 1040 * the urgent pointer points to the last octet 1041 * of urgent data. We continue, however, 1042 * to consider it to indicate the first octet 1043 * of data past the urgent section as the original 1044 * spec states (in one of two places). 1045 */ 1046 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1047 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1048 so->so_oobmark = so->so_rcv.sb_cc + 1049 (tp->rcv_up - tp->rcv_nxt) - 1; 1050 if (so->so_oobmark == 0) 1051 so->so_state |= SS_RCVATMARK; 1052 sohasoutofband(so); 1053 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1054 } 1055 /* 1056 * Remove out of band data so doesn't get presented to user. 1057 * This can happen independent of advancing the URG pointer, 1058 * but if two URG's are pending at once, some out-of-band 1059 * data may creep in... ick. 1060 */ 1061 if (ti->ti_urp <= ti->ti_len 1062 #ifdef SO_OOBINLINE 1063 && (so->so_options & SO_OOBINLINE) == 0 1064 #endif 1065 ) 1066 tcp_pulloutofband(so, ti, m); 1067 } else 1068 /* 1069 * If no out of band data is expected, 1070 * pull receive urgent pointer along 1071 * with the receive window. 1072 */ 1073 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1074 tp->rcv_up = tp->rcv_nxt; 1075 dodata: /* XXX */ 1076 1077 /* 1078 * Process the segment text, merging it into the TCP sequencing queue, 1079 * and arranging for acknowledgment of receipt if necessary. 1080 * This process logically involves adjusting tp->rcv_wnd as data 1081 * is presented to the user (this happens in tcp_usrreq.c, 1082 * case PRU_RCVD). If a FIN has already been received on this 1083 * connection then we just ignore the text. 1084 */ 1085 if ((ti->ti_len || (tiflags&TH_FIN)) && 1086 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1087 TCP_REASS(tp, ti, m, so, tiflags); 1088 /* 1089 * Note the amount of data that peer has sent into 1090 * our window, in order to estimate the sender's 1091 * buffer size. 1092 */ 1093 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1094 } else { 1095 m_freem(m); 1096 tiflags &= ~TH_FIN; 1097 } 1098 1099 /* 1100 * If FIN is received ACK the FIN and let the user know 1101 * that the connection is closing. 1102 */ 1103 if (tiflags & TH_FIN) { 1104 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1105 socantrcvmore(so); 1106 tp->t_flags |= TF_ACKNOW; 1107 tp->rcv_nxt++; 1108 } 1109 switch (tp->t_state) { 1110 1111 /* 1112 * In SYN_RECEIVED and ESTABLISHED STATES 1113 * enter the CLOSE_WAIT state. 1114 */ 1115 case TCPS_SYN_RECEIVED: 1116 case TCPS_ESTABLISHED: 1117 tp->t_state = TCPS_CLOSE_WAIT; 1118 break; 1119 1120 /* 1121 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1122 * enter the CLOSING state. 1123 */ 1124 case TCPS_FIN_WAIT_1: 1125 tp->t_state = TCPS_CLOSING; 1126 break; 1127 1128 /* 1129 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1130 * starting the time-wait timer, turning off the other 1131 * standard timers. 1132 */ 1133 case TCPS_FIN_WAIT_2: 1134 tp->t_state = TCPS_TIME_WAIT; 1135 tcp_canceltimers(tp); 1136 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1137 soisdisconnected(so); 1138 break; 1139 1140 /* 1141 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1142 */ 1143 case TCPS_TIME_WAIT: 1144 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1145 break; 1146 } 1147 } 1148 if (so->so_options & SO_DEBUG) 1149 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1150 1151 /* 1152 * Return any desired output. 1153 */ 1154 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1155 (void) tcp_output(tp); 1156 return; 1157 1158 dropafterack: 1159 /* 1160 * Generate an ACK dropping incoming segment if it occupies 1161 * sequence space, where the ACK reflects our state. 1162 */ 1163 if (tiflags & TH_RST) 1164 goto drop; 1165 m_freem(m); 1166 tp->t_flags |= TF_ACKNOW; 1167 (void) tcp_output(tp); 1168 return; 1169 1170 dropwithreset: 1171 if (om) { 1172 (void) m_free(om); 1173 om = 0; 1174 } 1175 /* 1176 * Generate a RST, dropping incoming segment. 1177 * Make ACK acceptable to originator of segment. 1178 * Don't bother to respond if destination was broadcast. 1179 */ 1180 if ((tiflags & TH_RST) || m->m_flags & M_BCAST) 1181 goto drop; 1182 if (tiflags & TH_ACK) 1183 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1184 else { 1185 if (tiflags & TH_SYN) 1186 ti->ti_len++; 1187 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1188 TH_RST|TH_ACK); 1189 } 1190 /* destroy temporarily created socket */ 1191 if (dropsocket) 1192 (void) soabort(so); 1193 return; 1194 1195 drop: 1196 if (om) 1197 (void) m_free(om); 1198 /* 1199 * Drop space held by incoming segment and return. 1200 */ 1201 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1202 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1203 m_freem(m); 1204 /* destroy temporarily created socket */ 1205 if (dropsocket) 1206 (void) soabort(so); 1207 return; 1208 } 1209 1210 tcp_dooptions(tp, om, ti) 1211 struct tcpcb *tp; 1212 struct mbuf *om; 1213 struct tcpiphdr *ti; 1214 { 1215 register u_char *cp; 1216 u_short mss; 1217 int opt, optlen, cnt; 1218 1219 cp = mtod(om, u_char *); 1220 cnt = om->m_len; 1221 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1222 opt = cp[0]; 1223 if (opt == TCPOPT_EOL) 1224 break; 1225 if (opt == TCPOPT_NOP) 1226 optlen = 1; 1227 else { 1228 optlen = cp[1]; 1229 if (optlen <= 0) 1230 break; 1231 } 1232 switch (opt) { 1233 1234 default: 1235 continue; 1236 1237 case TCPOPT_MAXSEG: 1238 if (optlen != 4) 1239 continue; 1240 if (!(ti->ti_flags & TH_SYN)) 1241 continue; 1242 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1243 NTOHS(mss); 1244 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1245 break; 1246 } 1247 } 1248 (void) m_free(om); 1249 } 1250 1251 /* 1252 * Pull out of band byte out of a segment so 1253 * it doesn't appear in the user's data queue. 1254 * It is still reflected in the segment length for 1255 * sequencing purposes. 1256 */ 1257 tcp_pulloutofband(so, ti, m) 1258 struct socket *so; 1259 struct tcpiphdr *ti; 1260 register struct mbuf *m; 1261 { 1262 int cnt = ti->ti_urp - 1; 1263 1264 while (cnt >= 0) { 1265 if (m->m_len > cnt) { 1266 char *cp = mtod(m, caddr_t) + cnt; 1267 struct tcpcb *tp = sototcpcb(so); 1268 1269 tp->t_iobc = *cp; 1270 tp->t_oobflags |= TCPOOB_HAVEDATA; 1271 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1272 m->m_len--; 1273 return; 1274 } 1275 cnt -= m->m_len; 1276 m = m->m_next; 1277 if (m == 0) 1278 break; 1279 } 1280 panic("tcp_pulloutofband"); 1281 } 1282 1283 /* 1284 * Collect new round-trip time estimate 1285 * and update averages and current timeout. 1286 */ 1287 tcp_xmit_timer(tp) 1288 register struct tcpcb *tp; 1289 { 1290 register short delta; 1291 1292 tcpstat.tcps_rttupdated++; 1293 if (tp->t_srtt != 0) { 1294 /* 1295 * srtt is stored as fixed point with 3 bits after the 1296 * binary point (i.e., scaled by 8). The following magic 1297 * is equivalent to the smoothing algorithm in rfc793 with 1298 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1299 * point). Adjust t_rtt to origin 0. 1300 */ 1301 delta = tp->t_rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1302 if ((tp->t_srtt += delta) <= 0) 1303 tp->t_srtt = 1; 1304 /* 1305 * We accumulate a smoothed rtt variance (actually, a 1306 * smoothed mean difference), then set the retransmit 1307 * timer to smoothed rtt + 4 times the smoothed variance. 1308 * rttvar is stored as fixed point with 2 bits after the 1309 * binary point (scaled by 4). The following is 1310 * equivalent to rfc793 smoothing with an alpha of .75 1311 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1312 * rfc793's wired-in beta. 1313 */ 1314 if (delta < 0) 1315 delta = -delta; 1316 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1317 if ((tp->t_rttvar += delta) <= 0) 1318 tp->t_rttvar = 1; 1319 } else { 1320 /* 1321 * No rtt measurement yet - use the unsmoothed rtt. 1322 * Set the variance to half the rtt (so our first 1323 * retransmit happens at 2*rtt) 1324 */ 1325 tp->t_srtt = tp->t_rtt << TCP_RTT_SHIFT; 1326 tp->t_rttvar = tp->t_rtt << (TCP_RTTVAR_SHIFT - 1); 1327 } 1328 tp->t_rtt = 0; 1329 tp->t_rxtshift = 0; 1330 1331 /* 1332 * the retransmit should happen at rtt + 4 * rttvar. 1333 * Because of the way we do the smoothing, srtt and rttvar 1334 * will each average +1/2 tick of bias. When we compute 1335 * the retransmit timer, we want 1/2 tick of rounding and 1336 * 1 extra tick because of +-1/2 tick uncertainty in the 1337 * firing of the timer. The bias will give us exactly the 1338 * 1.5 tick we need. But, because the bias is 1339 * statistical, we have to test that we don't drop below 1340 * the minimum feasible timer (which is 2 ticks). 1341 */ 1342 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1343 tp->t_rttmin, TCPTV_REXMTMAX); 1344 1345 /* 1346 * We received an ack for a packet that wasn't retransmitted; 1347 * it is probably safe to discard any error indications we've 1348 * received recently. This isn't quite right, but close enough 1349 * for now (a route might have failed after we sent a segment, 1350 * and the return path might not be symmetrical). 1351 */ 1352 tp->t_softerror = 0; 1353 } 1354 1355 /* 1356 * Determine a reasonable value for maxseg size. 1357 * If the route is known, check route for mtu. 1358 * If none, use an mss that can be handled on the outgoing 1359 * interface without forcing IP to fragment; if bigger than 1360 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1361 * to utilize large mbufs. If no route is found, route has no mtu, 1362 * or the destination isn't local, use a default, hopefully conservative 1363 * size (usually 512 or the default IP max size, but no more than the mtu 1364 * of the interface), as we can't discover anything about intervening 1365 * gateways or networks. We also initialize the congestion/slow start 1366 * window to be a single segment if the destination isn't local. 1367 * While looking at the routing entry, we also initialize other path-dependent 1368 * parameters from pre-set or cached values in the routing entry. 1369 */ 1370 1371 tcp_mss(tp, offer) 1372 register struct tcpcb *tp; 1373 u_short offer; 1374 { 1375 struct route *ro; 1376 register struct rtentry *rt; 1377 struct ifnet *ifp; 1378 register int rtt, mss; 1379 u_long bufsize; 1380 struct inpcb *inp; 1381 struct socket *so; 1382 extern int tcp_mssdflt, tcp_rttdflt; 1383 1384 inp = tp->t_inpcb; 1385 ro = &inp->inp_route; 1386 1387 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1388 /* No route yet, so try to acquire one */ 1389 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1390 ro->ro_dst.sa_family = AF_INET; 1391 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1392 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1393 inp->inp_faddr; 1394 rtalloc(ro); 1395 } 1396 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1397 return (tcp_mssdflt); 1398 } 1399 ifp = rt->rt_ifp; 1400 so = inp->inp_socket; 1401 1402 #ifdef RTV_MTU /* if route characteristics exist ... */ 1403 /* 1404 * While we're here, check if there's an initial rtt 1405 * or rttvar. Convert from the route-table units 1406 * to scaled multiples of the slow timeout timer. 1407 */ 1408 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1409 if (rt->rt_rmx.rmx_locks & RTV_MTU) 1410 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1411 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1412 if (rt->rt_rmx.rmx_rttvar) 1413 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1414 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1415 else 1416 /* default variation is +- 1 rtt */ 1417 tp->t_rttvar = 1418 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1419 TCPT_RANGESET(tp->t_rxtcur, 1420 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1421 tp->t_rttmin, TCPTV_REXMTMAX); 1422 } 1423 /* 1424 * if there's an mtu associated with the route, use it 1425 */ 1426 if (rt->rt_rmx.rmx_mtu) 1427 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1428 else 1429 #endif /* RTV_MTU */ 1430 { 1431 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1432 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1433 if (mss > MCLBYTES) 1434 mss &= ~(MCLBYTES-1); 1435 #else 1436 if (mss > MCLBYTES) 1437 mss = mss / MCLBYTES * MCLBYTES; 1438 #endif 1439 if (!in_localaddr(inp->inp_faddr)) 1440 mss = min(mss, tcp_mssdflt); 1441 } 1442 /* 1443 * The current mss, t_maxseg, is initialized to the default value. 1444 * If we compute a smaller value, reduce the current mss. 1445 * If we compute a larger value, return it for use in sending 1446 * a max seg size option, but don't store it for use 1447 * unless we received an offer at least that large from peer. 1448 * However, do not accept offers under 32 bytes. 1449 */ 1450 if (offer) 1451 mss = min(mss, offer); 1452 mss = max(mss, 32); /* sanity */ 1453 if (mss < tp->t_maxseg || offer != 0) { 1454 /* 1455 * If there's a pipesize, change the socket buffer 1456 * to that size. Make the socket buffers an integral 1457 * number of mss units; if the mss is larger than 1458 * the socket buffer, decrease the mss. 1459 */ 1460 #ifdef RTV_SPIPE 1461 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1462 #endif 1463 bufsize = so->so_snd.sb_hiwat; 1464 if (bufsize < mss) 1465 mss = bufsize; 1466 else { 1467 bufsize = min(bufsize, SB_MAX) / mss * mss; 1468 (void) sbreserve(&so->so_snd, bufsize); 1469 } 1470 tp->t_maxseg = mss; 1471 1472 #ifdef RTV_RPIPE 1473 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1474 #endif 1475 bufsize = so->so_rcv.sb_hiwat; 1476 if (bufsize > mss) { 1477 bufsize = min(bufsize, SB_MAX) / mss * mss; 1478 (void) sbreserve(&so->so_rcv, bufsize); 1479 } 1480 } 1481 tp->snd_cwnd = mss; 1482 1483 #ifdef RTV_SSTHRESH 1484 if (rt->rt_rmx.rmx_ssthresh) { 1485 /* 1486 * There's some sort of gateway or interface 1487 * buffer limit on the path. Use this to set 1488 * the slow start threshhold, but set the 1489 * threshold to no less than 2*mss. 1490 */ 1491 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1492 } 1493 #endif /* RTV_MTU */ 1494 return (mss); 1495 } 1496