1 /* 2 * Copyright (c) 1982, 1986 Regents of the University of California. 3 * All rights reserved. The Berkeley software License Agreement 4 * specifies the terms and conditions for redistribution. 5 * 6 * @(#)tcp_input.c 7.13 (Berkeley) 11/13/87 7 */ 8 9 #include "param.h" 10 #include "systm.h" 11 #include "mbuf.h" 12 #include "protosw.h" 13 #include "socket.h" 14 #include "socketvar.h" 15 #include "errno.h" 16 17 #include "../net/if.h" 18 #include "../net/route.h" 19 20 #include "in.h" 21 #include "in_pcb.h" 22 #include "in_systm.h" 23 #include "ip.h" 24 #include "ip_var.h" 25 #include "tcp.h" 26 #include "tcp_fsm.h" 27 #include "tcp_seq.h" 28 #include "tcp_timer.h" 29 #include "tcp_var.h" 30 #include "tcpip.h" 31 #include "tcp_debug.h" 32 33 int tcpprintfs = 0; 34 int tcpcksum = 1; 35 int tcprexmtthresh = 3; 36 struct tcpiphdr tcp_saveti; 37 extern tcpnodelack; 38 39 struct tcpcb *tcp_newtcpcb(); 40 41 /* 42 * Insert segment ti into reassembly queue of tcp with 43 * control block tp. Return TH_FIN if reassembly now includes 44 * a segment with FIN. The macro form does the common case inline 45 * (segment is the next to be received on an established connection, 46 * and the queue is empty), avoiding linkage into and removal 47 * from the queue and repetition of various conversions. 48 */ 49 #define TCP_REASS(tp, ti, m, so, flags) { \ 50 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 51 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 52 (tp)->t_state == TCPS_ESTABLISHED) { \ 53 (tp)->rcv_nxt += (ti)->ti_len; \ 54 flags = (ti)->ti_flags & TH_FIN; \ 55 tcpstat.tcps_rcvpack++;\ 56 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 57 sbappend(&(so)->so_rcv, (m)); \ 58 sorwakeup(so); \ 59 } else \ 60 (flags) = tcp_reass((tp), (ti)); \ 61 } 62 63 tcp_reass(tp, ti) 64 register struct tcpcb *tp; 65 register struct tcpiphdr *ti; 66 { 67 register struct tcpiphdr *q; 68 struct socket *so = tp->t_inpcb->inp_socket; 69 struct mbuf *m; 70 int flags; 71 72 /* 73 * Call with ti==0 after become established to 74 * force pre-ESTABLISHED data up to user socket. 75 */ 76 if (ti == 0) 77 goto present; 78 79 /* 80 * Find a segment which begins after this one does. 81 */ 82 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 83 q = (struct tcpiphdr *)q->ti_next) 84 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 85 break; 86 87 /* 88 * If there is a preceding segment, it may provide some of 89 * our data already. If so, drop the data from the incoming 90 * segment. If it provides all of our data, drop us. 91 */ 92 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 93 register int i; 94 q = (struct tcpiphdr *)q->ti_prev; 95 /* conversion to int (in i) handles seq wraparound */ 96 i = q->ti_seq + q->ti_len - ti->ti_seq; 97 if (i > 0) { 98 if (i >= ti->ti_len) { 99 tcpstat.tcps_rcvduppack++; 100 tcpstat.tcps_rcvdupbyte += ti->ti_len; 101 goto drop; 102 } 103 m_adj(dtom(ti), i); 104 ti->ti_len -= i; 105 ti->ti_seq += i; 106 } 107 q = (struct tcpiphdr *)(q->ti_next); 108 } 109 tcpstat.tcps_rcvoopack++; 110 tcpstat.tcps_rcvoobyte += ti->ti_len; 111 112 /* 113 * While we overlap succeeding segments trim them or, 114 * if they are completely covered, dequeue them. 115 */ 116 while (q != (struct tcpiphdr *)tp) { 117 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 118 if (i <= 0) 119 break; 120 if (i < q->ti_len) { 121 q->ti_seq += i; 122 q->ti_len -= i; 123 m_adj(dtom(q), i); 124 break; 125 } 126 q = (struct tcpiphdr *)q->ti_next; 127 m = dtom(q->ti_prev); 128 remque(q->ti_prev); 129 m_freem(m); 130 } 131 132 /* 133 * Stick new segment in its place. 134 */ 135 insque(ti, q->ti_prev); 136 137 present: 138 /* 139 * Present data to user, advancing rcv_nxt through 140 * completed sequence space. 141 */ 142 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 143 return (0); 144 ti = tp->seg_next; 145 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 146 return (0); 147 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 148 return (0); 149 do { 150 tp->rcv_nxt += ti->ti_len; 151 flags = ti->ti_flags & TH_FIN; 152 remque(ti); 153 m = dtom(ti); 154 ti = (struct tcpiphdr *)ti->ti_next; 155 if (so->so_state & SS_CANTRCVMORE) 156 m_freem(m); 157 else 158 sbappend(&so->so_rcv, m); 159 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 160 sorwakeup(so); 161 return (flags); 162 drop: 163 m_freem(dtom(ti)); 164 return (0); 165 } 166 167 /* 168 * TCP input routine, follows pages 65-76 of the 169 * protocol specification dated September, 1981 very closely. 170 */ 171 tcp_input(m0) 172 struct mbuf *m0; 173 { 174 register struct tcpiphdr *ti; 175 struct inpcb *inp; 176 register struct mbuf *m; 177 struct mbuf *om = 0; 178 int len, tlen, off; 179 register struct tcpcb *tp = 0; 180 register int tiflags; 181 struct socket *so; 182 int todrop, acked, ourfinisacked, needoutput = 0; 183 short ostate; 184 struct in_addr laddr; 185 int dropsocket = 0; 186 int iss = 0; 187 188 tcpstat.tcps_rcvtotal++; 189 /* 190 * Get IP and TCP header together in first mbuf. 191 * Note: IP leaves IP header in first mbuf. 192 */ 193 m = m0; 194 ti = mtod(m, struct tcpiphdr *); 195 if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2)) 196 ip_stripoptions((struct ip *)ti, (struct mbuf *)0); 197 if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) { 198 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 199 tcpstat.tcps_rcvshort++; 200 return; 201 } 202 ti = mtod(m, struct tcpiphdr *); 203 } 204 205 /* 206 * Checksum extended TCP header and data. 207 */ 208 tlen = ((struct ip *)ti)->ip_len; 209 len = sizeof (struct ip) + tlen; 210 if (tcpcksum) { 211 ti->ti_next = ti->ti_prev = 0; 212 ti->ti_x1 = 0; 213 ti->ti_len = (u_short)tlen; 214 ti->ti_len = htons((u_short)ti->ti_len); 215 if (ti->ti_sum = in_cksum(m, len)) { 216 if (tcpprintfs) 217 printf("tcp sum: src %x\n", ti->ti_src); 218 tcpstat.tcps_rcvbadsum++; 219 goto drop; 220 } 221 } 222 223 /* 224 * Check that TCP offset makes sense, 225 * pull out TCP options and adjust length. 226 */ 227 off = ti->ti_off << 2; 228 if (off < sizeof (struct tcphdr) || off > tlen) { 229 if (tcpprintfs) 230 printf("tcp off: src %x off %d\n", ti->ti_src, off); 231 tcpstat.tcps_rcvbadoff++; 232 goto drop; 233 } 234 tlen -= off; 235 ti->ti_len = tlen; 236 if (off > sizeof (struct tcphdr)) { 237 if (m->m_len < sizeof(struct ip) + off) { 238 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 239 tcpstat.tcps_rcvshort++; 240 return; 241 } 242 ti = mtod(m, struct tcpiphdr *); 243 } 244 om = m_get(M_DONTWAIT, MT_DATA); 245 if (om == 0) 246 goto drop; 247 om->m_len = off - sizeof (struct tcphdr); 248 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 249 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 250 m->m_len -= om->m_len; 251 bcopy(op+om->m_len, op, 252 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 253 } 254 } 255 tiflags = ti->ti_flags; 256 257 /* 258 * Drop TCP and IP headers; TCP options were dropped above. 259 */ 260 m->m_off += sizeof(struct tcpiphdr); 261 m->m_len -= sizeof(struct tcpiphdr); 262 263 /* 264 * Convert TCP protocol specific fields to host format. 265 */ 266 ti->ti_seq = ntohl(ti->ti_seq); 267 ti->ti_ack = ntohl(ti->ti_ack); 268 ti->ti_win = ntohs(ti->ti_win); 269 ti->ti_urp = ntohs(ti->ti_urp); 270 271 /* 272 * Locate pcb for segment. 273 */ 274 findpcb: 275 inp = in_pcblookup 276 (&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport, 277 INPLOOKUP_WILDCARD); 278 279 /* 280 * If the state is CLOSED (i.e., TCB does not exist) then 281 * all data in the incoming segment is discarded. 282 * If the TCB exists but is in CLOSED state, it is embryonic, 283 * but should either do a listen or a connect soon. 284 */ 285 if (inp == 0) 286 goto dropwithreset; 287 tp = intotcpcb(inp); 288 if (tp == 0) 289 goto dropwithreset; 290 if (tp->t_state == TCPS_CLOSED) 291 goto drop; 292 so = inp->inp_socket; 293 if (so->so_options & SO_DEBUG) { 294 ostate = tp->t_state; 295 tcp_saveti = *ti; 296 } 297 if (so->so_options & SO_ACCEPTCONN) { 298 so = sonewconn(so); 299 if (so == 0) 300 goto drop; 301 /* 302 * This is ugly, but .... 303 * 304 * Mark socket as temporary until we're 305 * committed to keeping it. The code at 306 * ``drop'' and ``dropwithreset'' check the 307 * flag dropsocket to see if the temporary 308 * socket created here should be discarded. 309 * We mark the socket as discardable until 310 * we're committed to it below in TCPS_LISTEN. 311 */ 312 dropsocket++; 313 inp = (struct inpcb *)so->so_pcb; 314 inp->inp_laddr = ti->ti_dst; 315 inp->inp_lport = ti->ti_dport; 316 inp->inp_options = ip_srcroute(); 317 tp = intotcpcb(inp); 318 tp->t_state = TCPS_LISTEN; 319 } 320 321 /* 322 * Segment received on connection. 323 * Reset idle time and keep-alive timer. 324 */ 325 tp->t_idle = 0; 326 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 327 328 /* 329 * Process options if not in LISTEN state, 330 * else do it below (after getting remote address). 331 */ 332 if (om && tp->t_state != TCPS_LISTEN) { 333 tcp_dooptions(tp, om, ti); 334 om = 0; 335 } 336 337 /* 338 * Calculate amount of space in receive window, 339 * and then do TCP input processing. 340 * Receive window is amount of space in rcv queue, 341 * but not less than advertised window. 342 */ 343 { int win; 344 345 win = sbspace(&so->so_rcv); 346 if (win < 0) 347 win = 0; 348 tp->rcv_wnd = MAX(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 349 } 350 351 switch (tp->t_state) { 352 353 /* 354 * If the state is LISTEN then ignore segment if it contains an RST. 355 * If the segment contains an ACK then it is bad and send a RST. 356 * If it does not contain a SYN then it is not interesting; drop it. 357 * Don't bother responding if the destination was a broadcast. 358 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 359 * tp->iss, and send a segment: 360 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 361 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 362 * Fill in remote peer address fields if not previously specified. 363 * Enter SYN_RECEIVED state, and process any other fields of this 364 * segment in this state. 365 */ 366 case TCPS_LISTEN: { 367 struct mbuf *am; 368 register struct sockaddr_in *sin; 369 370 if (tiflags & TH_RST) 371 goto drop; 372 if (tiflags & TH_ACK) 373 goto dropwithreset; 374 if ((tiflags & TH_SYN) == 0) 375 goto drop; 376 if (in_broadcast(ti->ti_dst)) 377 goto drop; 378 am = m_get(M_DONTWAIT, MT_SONAME); 379 if (am == NULL) 380 goto drop; 381 am->m_len = sizeof (struct sockaddr_in); 382 sin = mtod(am, struct sockaddr_in *); 383 sin->sin_family = AF_INET; 384 sin->sin_addr = ti->ti_src; 385 sin->sin_port = ti->ti_sport; 386 laddr = inp->inp_laddr; 387 if (inp->inp_laddr.s_addr == INADDR_ANY) 388 inp->inp_laddr = ti->ti_dst; 389 if (in_pcbconnect(inp, am)) { 390 inp->inp_laddr = laddr; 391 (void) m_free(am); 392 goto drop; 393 } 394 (void) m_free(am); 395 tp->t_template = tcp_template(tp); 396 if (tp->t_template == 0) { 397 tp = tcp_drop(tp, ENOBUFS); 398 dropsocket = 0; /* socket is already gone */ 399 goto drop; 400 } 401 if (om) { 402 tcp_dooptions(tp, om, ti); 403 om = 0; 404 } 405 if (iss) 406 tp->iss = iss; 407 else 408 tp->iss = tcp_iss; 409 tcp_iss += TCP_ISSINCR/2; 410 tp->irs = ti->ti_seq; 411 tcp_sendseqinit(tp); 412 tcp_rcvseqinit(tp); 413 tp->t_flags |= TF_ACKNOW; 414 tp->t_state = TCPS_SYN_RECEIVED; 415 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 416 dropsocket = 0; /* committed to socket */ 417 tcpstat.tcps_accepts++; 418 goto trimthenstep6; 419 } 420 421 /* 422 * If the state is SYN_SENT: 423 * if seg contains an ACK, but not for our SYN, drop the input. 424 * if seg contains a RST, then drop the connection. 425 * if seg does not contain SYN, then drop it. 426 * Otherwise this is an acceptable SYN segment 427 * initialize tp->rcv_nxt and tp->irs 428 * if seg contains ack then advance tp->snd_una 429 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 430 * arrange for segment to be acked (eventually) 431 * continue processing rest of data/controls, beginning with URG 432 */ 433 case TCPS_SYN_SENT: 434 if ((tiflags & TH_ACK) && 435 (SEQ_LEQ(ti->ti_ack, tp->iss) || 436 SEQ_GT(ti->ti_ack, tp->snd_max))) 437 goto dropwithreset; 438 if (tiflags & TH_RST) { 439 if (tiflags & TH_ACK) 440 tp = tcp_drop(tp, ECONNREFUSED); 441 goto drop; 442 } 443 if ((tiflags & TH_SYN) == 0) 444 goto drop; 445 if (tiflags & TH_ACK) { 446 tp->snd_una = ti->ti_ack; 447 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 448 tp->snd_nxt = tp->snd_una; 449 } 450 tp->t_timer[TCPT_REXMT] = 0; 451 tp->irs = ti->ti_seq; 452 tcp_rcvseqinit(tp); 453 tp->t_flags |= TF_ACKNOW; 454 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 455 tcpstat.tcps_connects++; 456 soisconnected(so); 457 tp->t_state = TCPS_ESTABLISHED; 458 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 459 (void) tcp_reass(tp, (struct tcpiphdr *)0); 460 /* 461 * if we didn't have to retransmit the SYN, 462 * use its rtt as our initial srtt & rtt var. 463 */ 464 if (tp->t_rtt) { 465 tp->t_srtt = tp->t_rtt << 3; 466 tp->t_rttvar = tp->t_rtt << 1; 467 TCPT_RANGESET(tp->t_rxtcur, 468 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 469 TCPTV_MIN, TCPTV_REXMTMAX); 470 tp->t_rtt = 0; 471 } 472 } else 473 tp->t_state = TCPS_SYN_RECEIVED; 474 475 trimthenstep6: 476 /* 477 * Advance ti->ti_seq to correspond to first data byte. 478 * If data, trim to stay within window, 479 * dropping FIN if necessary. 480 */ 481 ti->ti_seq++; 482 if (ti->ti_len > tp->rcv_wnd) { 483 todrop = ti->ti_len - tp->rcv_wnd; 484 m_adj(m, -todrop); 485 ti->ti_len = tp->rcv_wnd; 486 tiflags &= ~TH_FIN; 487 tcpstat.tcps_rcvpackafterwin++; 488 tcpstat.tcps_rcvbyteafterwin += todrop; 489 } 490 tp->snd_wl1 = ti->ti_seq - 1; 491 tp->rcv_up = ti->ti_seq; 492 goto step6; 493 } 494 495 /* 496 * States other than LISTEN or SYN_SENT. 497 * First check that at least some bytes of segment are within 498 * receive window. If segment begins before rcv_nxt, 499 * drop leading data (and SYN); if nothing left, just ack. 500 */ 501 todrop = tp->rcv_nxt - ti->ti_seq; 502 if (todrop > 0) { 503 if (tiflags & TH_SYN) { 504 tiflags &= ~TH_SYN; 505 ti->ti_seq++; 506 if (ti->ti_urp > 1) 507 ti->ti_urp--; 508 else 509 tiflags &= ~TH_URG; 510 todrop--; 511 } 512 if (todrop > ti->ti_len || 513 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 514 #ifdef TCP_COMPAT_42 515 /* 516 * Don't toss RST in response to 4.2-style keepalive. 517 */ 518 if (ti->ti_seq == tp->rcv_nxt - 1 && tiflags & TH_RST) 519 goto do_rst; 520 #endif 521 tcpstat.tcps_rcvduppack++; 522 tcpstat.tcps_rcvdupbyte += ti->ti_len; 523 todrop = ti->ti_len; 524 tiflags &= ~TH_FIN; 525 tp->t_flags |= TF_ACKNOW; 526 } else { 527 tcpstat.tcps_rcvpartduppack++; 528 tcpstat.tcps_rcvpartdupbyte += todrop; 529 } 530 m_adj(m, todrop); 531 ti->ti_seq += todrop; 532 ti->ti_len -= todrop; 533 if (ti->ti_urp > todrop) 534 ti->ti_urp -= todrop; 535 else { 536 tiflags &= ~TH_URG; 537 ti->ti_urp = 0; 538 } 539 } 540 541 /* 542 * If new data is received on a connection after the 543 * user processes are gone, then RST the other end. 544 */ 545 if ((so->so_state & SS_NOFDREF) && 546 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 547 tp = tcp_close(tp); 548 tcpstat.tcps_rcvafterclose++; 549 goto dropwithreset; 550 } 551 552 if (tp->rcv_wnd == 0) { 553 /* 554 * If window is closed can only take segments at 555 * window edge, and have to drop data and PUSH from 556 * incoming segments. 557 */ 558 if (tp->rcv_nxt != ti->ti_seq) { 559 tcpstat.tcps_rcvpackafterwin++; 560 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 561 goto dropafterack; 562 } 563 if (ti->ti_len > 0) { 564 if (ti->ti_len == 1) 565 tcpstat.tcps_rcvwinprobe++; 566 else { 567 tcpstat.tcps_rcvpackafterwin++; 568 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 569 } 570 m_adj(m, ti->ti_len); 571 ti->ti_len = 0; 572 tiflags &= ~(TH_PUSH|TH_FIN); 573 } 574 } else { 575 /* 576 * If segment ends after window, drop trailing data 577 * (and PUSH and FIN); if nothing left, just ACK. 578 */ 579 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 580 if (todrop > 0) { 581 if (todrop >= ti->ti_len) { 582 /* 583 * If a new connection request is received 584 * while in TIME_WAIT, drop the old connection 585 * and start over if the sequence numbers 586 * are above the previous ones. 587 */ 588 if (tiflags & TH_SYN && 589 tp->t_state == TCPS_TIME_WAIT && 590 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 591 iss = tp->rcv_nxt + TCP_ISSINCR; 592 (void) tcp_close(tp); 593 goto findpcb; 594 } 595 if (todrop == 1) 596 tcpstat.tcps_rcvwinprobe++; 597 else { 598 tcpstat.tcps_rcvpackafterwin++; 599 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 600 } 601 goto dropafterack; 602 } 603 tcpstat.tcps_rcvpackafterwin++; 604 tcpstat.tcps_rcvbyteafterwin += todrop; 605 m_adj(m, -todrop); 606 ti->ti_len -= todrop; 607 tiflags &= ~(TH_PUSH|TH_FIN); 608 } 609 } 610 611 #ifdef TCP_COMPAT_42 612 do_rst: 613 #endif 614 /* 615 * If the RST bit is set examine the state: 616 * SYN_RECEIVED STATE: 617 * If passive open, return to LISTEN state. 618 * If active open, inform user that connection was refused. 619 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 620 * Inform user that connection was reset, and close tcb. 621 * CLOSING, LAST_ACK, TIME_WAIT STATES 622 * Close the tcb. 623 */ 624 if (tiflags&TH_RST) switch (tp->t_state) { 625 626 case TCPS_SYN_RECEIVED: 627 tp = tcp_drop(tp, ECONNREFUSED); 628 goto drop; 629 630 case TCPS_ESTABLISHED: 631 case TCPS_FIN_WAIT_1: 632 case TCPS_FIN_WAIT_2: 633 case TCPS_CLOSE_WAIT: 634 tp = tcp_drop(tp, ECONNRESET); 635 goto drop; 636 637 case TCPS_CLOSING: 638 case TCPS_LAST_ACK: 639 case TCPS_TIME_WAIT: 640 tp = tcp_close(tp); 641 goto drop; 642 } 643 644 /* 645 * If a SYN is in the window, then this is an 646 * error and we send an RST and drop the connection. 647 */ 648 if (tiflags & TH_SYN) { 649 tp = tcp_drop(tp, ECONNRESET); 650 goto dropwithreset; 651 } 652 653 /* 654 * If the ACK bit is off we drop the segment and return. 655 */ 656 if ((tiflags & TH_ACK) == 0) 657 goto drop; 658 659 /* 660 * Ack processing. 661 */ 662 switch (tp->t_state) { 663 664 /* 665 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 666 * ESTABLISHED state and continue processing, otherwise 667 * send an RST. 668 */ 669 case TCPS_SYN_RECEIVED: 670 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 671 SEQ_GT(ti->ti_ack, tp->snd_max)) 672 goto dropwithreset; 673 tcpstat.tcps_connects++; 674 soisconnected(so); 675 tp->t_state = TCPS_ESTABLISHED; 676 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 677 (void) tcp_reass(tp, (struct tcpiphdr *)0); 678 tp->snd_wl1 = ti->ti_seq - 1; 679 /* fall into ... */ 680 681 /* 682 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 683 * ACKs. If the ack is in the range 684 * tp->snd_una < ti->ti_ack <= tp->snd_max 685 * then advance tp->snd_una to ti->ti_ack and drop 686 * data from the retransmission queue. If this ACK reflects 687 * more up to date window information we update our window information. 688 */ 689 case TCPS_ESTABLISHED: 690 case TCPS_FIN_WAIT_1: 691 case TCPS_FIN_WAIT_2: 692 case TCPS_CLOSE_WAIT: 693 case TCPS_CLOSING: 694 case TCPS_LAST_ACK: 695 case TCPS_TIME_WAIT: 696 697 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 698 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 699 tcpstat.tcps_rcvdupack++; 700 /* 701 * If we have outstanding data (not a 702 * window probe), this is a completely 703 * duplicate ack (ie, window info didn't 704 * change), the ack is the biggest we've 705 * seen and we've seen exactly our rexmt 706 * threshhold of them, assume a packet 707 * has been dropped and retransmit it. 708 * Kludge snd_nxt & the congestion 709 * window so we send only this one 710 * packet. If this packet fills the 711 * only hole in the receiver's seq. 712 * space, the next real ack will fully 713 * open our window. This means we 714 * have to do the usual slow-start to 715 * not overwhelm an intermediate gateway 716 * with a burst of packets. Leave 717 * here with the congestion window set 718 * to allow 2 packets on the next real 719 * ack and the exp-to-linear thresh 720 * set for half the current window 721 * size (since we know we're losing at 722 * the current window size). 723 */ 724 if (tp->t_timer[TCPT_REXMT] == 0 || 725 ti->ti_ack != tp->snd_una) 726 tp->t_dupacks = 0; 727 else if (++tp->t_dupacks == tcprexmtthresh) { 728 tcp_seq onxt = tp->snd_nxt; 729 u_int win = 730 MIN(tp->snd_wnd, tp->snd_cwnd) / 2 / 731 tp->t_maxseg; 732 733 if (win < 2) 734 win = 2; 735 tp->snd_ssthresh = win * tp->t_maxseg; 736 737 tp->t_timer[TCPT_REXMT] = 0; 738 tp->t_rtt = 0; 739 tp->snd_nxt = ti->ti_ack; 740 tp->snd_cwnd = tp->t_maxseg; 741 (void) tcp_output(tp); 742 743 if (SEQ_GT(onxt, tp->snd_nxt)) 744 tp->snd_nxt = onxt; 745 goto drop; 746 } 747 } else 748 tp->t_dupacks = 0; 749 break; 750 } 751 tp->t_dupacks = 0; 752 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 753 tcpstat.tcps_rcvacktoomuch++; 754 goto dropafterack; 755 } 756 acked = ti->ti_ack - tp->snd_una; 757 tcpstat.tcps_rcvackpack++; 758 tcpstat.tcps_rcvackbyte += acked; 759 760 /* 761 * If transmit timer is running and timed sequence 762 * number was acked, update smoothed round trip time. 763 * Since we now have an rtt measurement, cancel the 764 * timer backoff (cf., Phil Karn's retransmit alg.). 765 * Recompute the initial retransmit timer. 766 */ 767 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) { 768 tcpstat.tcps_rttupdated++; 769 if (tp->t_srtt != 0) { 770 register short delta; 771 772 /* 773 * srtt is stored as fixed point with 3 bits 774 * after the binary point (i.e., scaled by 8). 775 * The following magic is equivalent 776 * to the smoothing algorithm in rfc793 777 * with an alpha of .875 778 * (srtt = rtt/8 + srtt*7/8 in fixed point). 779 * Adjust t_rtt to origin 0. 780 */ 781 tp->t_rtt--; 782 delta = tp->t_rtt - (tp->t_srtt >> 3); 783 if ((tp->t_srtt += delta) <= 0) 784 tp->t_srtt = 1; 785 /* 786 * We accumulate a smoothed rtt variance 787 * (actually, a smoothed mean difference), 788 * then set the retransmit timer to smoothed 789 * rtt + 2 times the smoothed variance. 790 * rttvar is stored as fixed point 791 * with 2 bits after the binary point 792 * (scaled by 4). The following is equivalent 793 * to rfc793 smoothing with an alpha of .75 794 * (rttvar = rttvar*3/4 + |delta| / 4). 795 * This replaces rfc793's wired-in beta. 796 */ 797 if (delta < 0) 798 delta = -delta; 799 delta -= (tp->t_rttvar >> 2); 800 if ((tp->t_rttvar += delta) <= 0) 801 tp->t_rttvar = 1; 802 } else { 803 /* 804 * No rtt measurement yet - use the 805 * unsmoothed rtt. Set the variance 806 * to half the rtt (so our first 807 * retransmit happens at 2*rtt) 808 */ 809 tp->t_srtt = tp->t_rtt << 3; 810 tp->t_rttvar = tp->t_rtt << 1; 811 } 812 tp->t_rtt = 0; 813 tp->t_rxtshift = 0; 814 TCPT_RANGESET(tp->t_rxtcur, 815 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 816 TCPTV_MIN, TCPTV_REXMTMAX); 817 } 818 819 /* 820 * If all outstanding data is acked, stop retransmit 821 * timer and remember to restart (more output or persist). 822 * If there is more data to be acked, restart retransmit 823 * timer, using current (possibly backed-off) value. 824 */ 825 if (ti->ti_ack == tp->snd_max) { 826 tp->t_timer[TCPT_REXMT] = 0; 827 needoutput = 1; 828 } else if (tp->t_timer[TCPT_PERSIST] == 0) 829 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 830 /* 831 * When new data is acked, open the congestion window. 832 * If the window gives us less than ssthresh packets 833 * in flight, open exponentially (maxseg per packet). 834 * Otherwise open linearly (maxseg per window, 835 * or maxseg^2 / cwnd per packet). 836 */ 837 { 838 u_int incr = tp->t_maxseg; 839 840 if (tp->snd_cwnd > tp->snd_ssthresh) 841 incr = MAX(incr * incr / tp->snd_cwnd, 1); 842 843 tp->snd_cwnd = MIN(tp->snd_cwnd + incr, 65535); /* XXX */ 844 } 845 if (acked > so->so_snd.sb_cc) { 846 tp->snd_wnd -= so->so_snd.sb_cc; 847 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 848 ourfinisacked = 1; 849 } else { 850 sbdrop(&so->so_snd, acked); 851 tp->snd_wnd -= acked; 852 ourfinisacked = 0; 853 } 854 if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel) 855 sowwakeup(so); 856 tp->snd_una = ti->ti_ack; 857 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 858 tp->snd_nxt = tp->snd_una; 859 860 switch (tp->t_state) { 861 862 /* 863 * In FIN_WAIT_1 STATE in addition to the processing 864 * for the ESTABLISHED state if our FIN is now acknowledged 865 * then enter FIN_WAIT_2. 866 */ 867 case TCPS_FIN_WAIT_1: 868 if (ourfinisacked) { 869 /* 870 * If we can't receive any more 871 * data, then closing user can proceed. 872 * Starting the timer is contrary to the 873 * specification, but if we don't get a FIN 874 * we'll hang forever. 875 */ 876 if (so->so_state & SS_CANTRCVMORE) { 877 soisdisconnected(so); 878 tp->t_timer[TCPT_2MSL] = TCPTV_MAXIDLE; 879 } 880 tp->t_state = TCPS_FIN_WAIT_2; 881 } 882 break; 883 884 /* 885 * In CLOSING STATE in addition to the processing for 886 * the ESTABLISHED state if the ACK acknowledges our FIN 887 * then enter the TIME-WAIT state, otherwise ignore 888 * the segment. 889 */ 890 case TCPS_CLOSING: 891 if (ourfinisacked) { 892 tp->t_state = TCPS_TIME_WAIT; 893 tcp_canceltimers(tp); 894 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 895 soisdisconnected(so); 896 } 897 break; 898 899 /* 900 * In LAST_ACK, we may still be waiting for data to drain 901 * and/or to be acked, as well as for the ack of our FIN. 902 * If our FIN is now acknowledged, delete the TCB, 903 * enter the closed state and return. 904 */ 905 case TCPS_LAST_ACK: 906 if (ourfinisacked) { 907 tp = tcp_close(tp); 908 goto drop; 909 } 910 break; 911 912 /* 913 * In TIME_WAIT state the only thing that should arrive 914 * is a retransmission of the remote FIN. Acknowledge 915 * it and restart the finack timer. 916 */ 917 case TCPS_TIME_WAIT: 918 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 919 goto dropafterack; 920 } 921 } 922 923 step6: 924 /* 925 * Update window information. 926 * Don't look at window if no ACK: TAC's send garbage on first SYN. 927 */ 928 if ((tiflags & TH_ACK) && 929 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 930 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 931 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 932 /* keep track of pure window updates */ 933 if (ti->ti_len == 0 && 934 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 935 tcpstat.tcps_rcvwinupd++; 936 tp->snd_wnd = ti->ti_win; 937 tp->snd_wl1 = ti->ti_seq; 938 tp->snd_wl2 = ti->ti_ack; 939 if (tp->snd_wnd > tp->max_sndwnd) 940 tp->max_sndwnd = tp->snd_wnd; 941 needoutput = 1; 942 } 943 944 /* 945 * Process segments with URG. 946 */ 947 if ((tiflags & TH_URG) && ti->ti_urp && 948 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 949 /* 950 * This is a kludge, but if we receive and accept 951 * random urgent pointers, we'll crash in 952 * soreceive. It's hard to imagine someone 953 * actually wanting to send this much urgent data. 954 */ 955 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 956 ti->ti_urp = 0; /* XXX */ 957 tiflags &= ~TH_URG; /* XXX */ 958 goto dodata; /* XXX */ 959 } 960 /* 961 * If this segment advances the known urgent pointer, 962 * then mark the data stream. This should not happen 963 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 964 * a FIN has been received from the remote side. 965 * In these states we ignore the URG. 966 * 967 * According to RFC961 (Assigned Protocols), 968 * the urgent pointer points to the last octet 969 * of urgent data. We continue, however, 970 * to consider it to indicate the first octet 971 * of data past the urgent section 972 * as the original spec states. 973 */ 974 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 975 tp->rcv_up = ti->ti_seq + ti->ti_urp; 976 so->so_oobmark = so->so_rcv.sb_cc + 977 (tp->rcv_up - tp->rcv_nxt) - 1; 978 if (so->so_oobmark == 0) 979 so->so_state |= SS_RCVATMARK; 980 sohasoutofband(so); 981 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 982 } 983 /* 984 * Remove out of band data so doesn't get presented to user. 985 * This can happen independent of advancing the URG pointer, 986 * but if two URG's are pending at once, some out-of-band 987 * data may creep in... ick. 988 */ 989 if (ti->ti_urp <= ti->ti_len && 990 (so->so_options & SO_OOBINLINE) == 0) 991 tcp_pulloutofband(so, ti); 992 } else 993 /* 994 * If no out of band data is expected, 995 * pull receive urgent pointer along 996 * with the receive window. 997 */ 998 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 999 tp->rcv_up = tp->rcv_nxt; 1000 dodata: /* XXX */ 1001 1002 /* 1003 * Process the segment text, merging it into the TCP sequencing queue, 1004 * and arranging for acknowledgment of receipt if necessary. 1005 * This process logically involves adjusting tp->rcv_wnd as data 1006 * is presented to the user (this happens in tcp_usrreq.c, 1007 * case PRU_RCVD). If a FIN has already been received on this 1008 * connection then we just ignore the text. 1009 */ 1010 if ((ti->ti_len || (tiflags&TH_FIN)) && 1011 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1012 TCP_REASS(tp, ti, m, so, tiflags); 1013 if (tcpnodelack == 0) 1014 tp->t_flags |= TF_DELACK; 1015 else 1016 tp->t_flags |= TF_ACKNOW; 1017 /* 1018 * Note the amount of data that peer has sent into 1019 * our window, in order to estimate the sender's 1020 * buffer size. 1021 */ 1022 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1023 if (len > tp->max_rcvd) 1024 tp->max_rcvd = len; 1025 } else { 1026 m_freem(m); 1027 tiflags &= ~TH_FIN; 1028 } 1029 1030 /* 1031 * If FIN is received ACK the FIN and let the user know 1032 * that the connection is closing. 1033 */ 1034 if (tiflags & TH_FIN) { 1035 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1036 socantrcvmore(so); 1037 tp->t_flags |= TF_ACKNOW; 1038 tp->rcv_nxt++; 1039 } 1040 switch (tp->t_state) { 1041 1042 /* 1043 * In SYN_RECEIVED and ESTABLISHED STATES 1044 * enter the CLOSE_WAIT state. 1045 */ 1046 case TCPS_SYN_RECEIVED: 1047 case TCPS_ESTABLISHED: 1048 tp->t_state = TCPS_CLOSE_WAIT; 1049 break; 1050 1051 /* 1052 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1053 * enter the CLOSING state. 1054 */ 1055 case TCPS_FIN_WAIT_1: 1056 tp->t_state = TCPS_CLOSING; 1057 break; 1058 1059 /* 1060 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1061 * starting the time-wait timer, turning off the other 1062 * standard timers. 1063 */ 1064 case TCPS_FIN_WAIT_2: 1065 tp->t_state = TCPS_TIME_WAIT; 1066 tcp_canceltimers(tp); 1067 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1068 soisdisconnected(so); 1069 break; 1070 1071 /* 1072 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1073 */ 1074 case TCPS_TIME_WAIT: 1075 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1076 break; 1077 } 1078 } 1079 if (so->so_options & SO_DEBUG) 1080 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1081 1082 /* 1083 * Return any desired output. 1084 */ 1085 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1086 (void) tcp_output(tp); 1087 return; 1088 1089 dropafterack: 1090 /* 1091 * Generate an ACK dropping incoming segment if it occupies 1092 * sequence space, where the ACK reflects our state. 1093 */ 1094 if (tiflags & TH_RST) 1095 goto drop; 1096 m_freem(m); 1097 tp->t_flags |= TF_ACKNOW; 1098 (void) tcp_output(tp); 1099 return; 1100 1101 dropwithreset: 1102 if (om) { 1103 (void) m_free(om); 1104 om = 0; 1105 } 1106 /* 1107 * Generate a RST, dropping incoming segment. 1108 * Make ACK acceptable to originator of segment. 1109 * Don't bother to respond if destination was broadcast. 1110 */ 1111 if ((tiflags & TH_RST) || in_broadcast(ti->ti_dst)) 1112 goto drop; 1113 if (tiflags & TH_ACK) 1114 tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST); 1115 else { 1116 if (tiflags & TH_SYN) 1117 ti->ti_len++; 1118 tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1119 TH_RST|TH_ACK); 1120 } 1121 /* destroy temporarily created socket */ 1122 if (dropsocket) 1123 (void) soabort(so); 1124 return; 1125 1126 drop: 1127 if (om) 1128 (void) m_free(om); 1129 /* 1130 * Drop space held by incoming segment and return. 1131 */ 1132 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1133 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1134 m_freem(m); 1135 /* destroy temporarily created socket */ 1136 if (dropsocket) 1137 (void) soabort(so); 1138 return; 1139 } 1140 1141 tcp_dooptions(tp, om, ti) 1142 struct tcpcb *tp; 1143 struct mbuf *om; 1144 struct tcpiphdr *ti; 1145 { 1146 register u_char *cp; 1147 int opt, optlen, cnt; 1148 1149 cp = mtod(om, u_char *); 1150 cnt = om->m_len; 1151 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1152 opt = cp[0]; 1153 if (opt == TCPOPT_EOL) 1154 break; 1155 if (opt == TCPOPT_NOP) 1156 optlen = 1; 1157 else { 1158 optlen = cp[1]; 1159 if (optlen <= 0) 1160 break; 1161 } 1162 switch (opt) { 1163 1164 default: 1165 break; 1166 1167 case TCPOPT_MAXSEG: 1168 if (optlen != 4) 1169 continue; 1170 if (!(ti->ti_flags & TH_SYN)) 1171 continue; 1172 tp->t_maxseg = *(u_short *)(cp + 2); 1173 tp->t_maxseg = ntohs((u_short)tp->t_maxseg); 1174 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 1175 break; 1176 } 1177 } 1178 (void) m_free(om); 1179 } 1180 1181 /* 1182 * Pull out of band byte out of a segment so 1183 * it doesn't appear in the user's data queue. 1184 * It is still reflected in the segment length for 1185 * sequencing purposes. 1186 */ 1187 tcp_pulloutofband(so, ti) 1188 struct socket *so; 1189 struct tcpiphdr *ti; 1190 { 1191 register struct mbuf *m; 1192 int cnt = ti->ti_urp - 1; 1193 1194 m = dtom(ti); 1195 while (cnt >= 0) { 1196 if (m->m_len > cnt) { 1197 char *cp = mtod(m, caddr_t) + cnt; 1198 struct tcpcb *tp = sototcpcb(so); 1199 1200 tp->t_iobc = *cp; 1201 tp->t_oobflags |= TCPOOB_HAVEDATA; 1202 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1203 m->m_len--; 1204 return; 1205 } 1206 cnt -= m->m_len; 1207 m = m->m_next; 1208 if (m == 0) 1209 break; 1210 } 1211 panic("tcp_pulloutofband"); 1212 } 1213 1214 /* 1215 * Determine a reasonable value for maxseg size. 1216 * If the route is known, use one that can be handled 1217 * on the given interface without forcing IP to fragment. 1218 * If bigger than an mbuf cluster (MCLBYTES), round down to nearest size 1219 * to utilize large mbufs. 1220 * If interface pointer is unavailable, or the destination isn't local, 1221 * use a conservative size (512 or the default IP max size, but no more 1222 * than the mtu of the interface through which we route), 1223 * as we can't discover anything about intervening gateways or networks. 1224 * We also initialize the congestion/slow start window to be a single 1225 * segment if the destination isn't local; this information should 1226 * probably all be saved with the routing entry at the transport level. 1227 * 1228 * This is ugly, and doesn't belong at this level, but has to happen somehow. 1229 */ 1230 tcp_mss(tp) 1231 register struct tcpcb *tp; 1232 { 1233 struct route *ro; 1234 struct ifnet *ifp; 1235 int mss; 1236 struct inpcb *inp; 1237 1238 inp = tp->t_inpcb; 1239 ro = &inp->inp_route; 1240 if ((ro->ro_rt == (struct rtentry *)0) || 1241 (ifp = ro->ro_rt->rt_ifp) == (struct ifnet *)0) { 1242 /* No route yet, so try to acquire one */ 1243 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1244 ro->ro_dst.sa_family = AF_INET; 1245 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1246 inp->inp_faddr; 1247 rtalloc(ro); 1248 } 1249 if ((ro->ro_rt == 0) || (ifp = ro->ro_rt->rt_ifp) == 0) 1250 return (TCP_MSS); 1251 } 1252 1253 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1254 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1255 if (mss > MCLBYTES) 1256 mss &= ~(MCLBYTES-1); 1257 #else 1258 if (mss > MCLBYTES) 1259 mss = mss / MCLBYTES * MCLBYTES; 1260 #endif 1261 if (in_localaddr(inp->inp_faddr)) 1262 return (mss); 1263 1264 mss = MIN(mss, TCP_MSS); 1265 tp->snd_cwnd = mss; 1266 return (mss); 1267 } 1268