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 7.6 (Berkeley) 06/30/87 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 struct tcpiphdr tcp_saveti; 36 extern tcpnodelack; 37 38 struct tcpcb *tcp_newtcpcb(); 39 40 /* 41 * Insert segment ti into reassembly queue of tcp with 42 * control block tp. Return TH_FIN if reassembly now includes 43 * a segment with FIN. The macro form does the common case inline 44 * (segment is the next to be received on an established connection, 45 * and the queue is empty), avoiding linkage into and removal 46 * from the queue and repetition of various conversions. 47 */ 48 #define TCP_REASS(tp, ti, m, so, flags) { \ 49 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 50 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 51 (tp)->t_state == TCPS_ESTABLISHED) { \ 52 (tp)->rcv_nxt += (ti)->ti_len; \ 53 flags = (ti)->ti_flags & TH_FIN; \ 54 tcpstat.tcps_rcvpack++;\ 55 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 56 sbappend(&(so)->so_rcv, (m)); \ 57 sorwakeup(so); \ 58 } else \ 59 (flags) = tcp_reass((tp), (ti)); \ 60 } 61 62 tcp_reass(tp, ti) 63 register struct tcpcb *tp; 64 register struct tcpiphdr *ti; 65 { 66 register struct tcpiphdr *q; 67 struct socket *so = tp->t_inpcb->inp_socket; 68 struct mbuf *m; 69 int flags; 70 71 /* 72 * Call with ti==0 after become established to 73 * force pre-ESTABLISHED data up to user socket. 74 */ 75 if (ti == 0) 76 goto present; 77 78 /* 79 * Find a segment which begins after this one does. 80 */ 81 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 82 q = (struct tcpiphdr *)q->ti_next) 83 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 84 break; 85 86 /* 87 * If there is a preceding segment, it may provide some of 88 * our data already. If so, drop the data from the incoming 89 * segment. If it provides all of our data, drop us. 90 */ 91 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 92 register int i; 93 q = (struct tcpiphdr *)q->ti_prev; 94 /* conversion to int (in i) handles seq wraparound */ 95 i = q->ti_seq + q->ti_len - ti->ti_seq; 96 if (i > 0) { 97 if (i >= ti->ti_len) { 98 tcpstat.tcps_rcvduppack++; 99 tcpstat.tcps_rcvdupbyte += ti->ti_len; 100 goto drop; 101 } 102 m_adj(dtom(ti), i); 103 ti->ti_len -= i; 104 ti->ti_seq += i; 105 } 106 q = (struct tcpiphdr *)(q->ti_next); 107 } 108 tcpstat.tcps_rcvoopack++; 109 tcpstat.tcps_rcvoobyte += ti->ti_len; 110 111 /* 112 * While we overlap succeeding segments trim them or, 113 * if they are completely covered, dequeue them. 114 */ 115 while (q != (struct tcpiphdr *)tp) { 116 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 117 if (i <= 0) 118 break; 119 if (i < q->ti_len) { 120 q->ti_seq += i; 121 q->ti_len -= i; 122 m_adj(dtom(q), i); 123 break; 124 } 125 q = (struct tcpiphdr *)q->ti_next; 126 m = dtom(q->ti_prev); 127 remque(q->ti_prev); 128 m_freem(m); 129 } 130 131 /* 132 * Stick new segment in its place. 133 */ 134 insque(ti, q->ti_prev); 135 136 present: 137 /* 138 * Present data to user, advancing rcv_nxt through 139 * completed sequence space. 140 */ 141 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 142 return (0); 143 ti = tp->seg_next; 144 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 145 return (0); 146 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 147 return (0); 148 do { 149 tp->rcv_nxt += ti->ti_len; 150 flags = ti->ti_flags & TH_FIN; 151 remque(ti); 152 m = dtom(ti); 153 ti = (struct tcpiphdr *)ti->ti_next; 154 if (so->so_state & SS_CANTRCVMORE) 155 m_freem(m); 156 else 157 sbappend(&so->so_rcv, m); 158 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 159 sorwakeup(so); 160 return (flags); 161 drop: 162 m_freem(dtom(ti)); 163 return (0); 164 } 165 166 /* 167 * TCP input routine, follows pages 65-76 of the 168 * protocol specification dated September, 1981 very closely. 169 */ 170 tcp_input(m0) 171 struct mbuf *m0; 172 { 173 register struct tcpiphdr *ti; 174 struct inpcb *inp; 175 register struct mbuf *m; 176 struct mbuf *om = 0; 177 int len, tlen, off; 178 register struct tcpcb *tp = 0; 179 register int tiflags; 180 struct socket *so; 181 int todrop, acked, ourfinisacked, needoutput = 0; 182 short ostate; 183 struct in_addr laddr; 184 int dropsocket = 0; 185 int iss = 0; 186 187 tcpstat.tcps_rcvtotal++; 188 /* 189 * Get IP and TCP header together in first mbuf. 190 * Note: IP leaves IP header in first mbuf. 191 */ 192 m = m0; 193 ti = mtod(m, struct tcpiphdr *); 194 if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2)) 195 ip_stripoptions((struct ip *)ti, (struct mbuf *)0); 196 if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) { 197 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 198 tcpstat.tcps_rcvshort++; 199 return; 200 } 201 ti = mtod(m, struct tcpiphdr *); 202 } 203 204 /* 205 * Checksum extended TCP header and data. 206 */ 207 tlen = ((struct ip *)ti)->ip_len; 208 len = sizeof (struct ip) + tlen; 209 if (tcpcksum) { 210 ti->ti_next = ti->ti_prev = 0; 211 ti->ti_x1 = 0; 212 ti->ti_len = (u_short)tlen; 213 ti->ti_len = htons((u_short)ti->ti_len); 214 if (ti->ti_sum = in_cksum(m, len)) { 215 if (tcpprintfs) 216 printf("tcp sum: src %x\n", ti->ti_src); 217 tcpstat.tcps_rcvbadsum++; 218 goto drop; 219 } 220 } 221 222 /* 223 * Check that TCP offset makes sense, 224 * pull out TCP options and adjust length. 225 */ 226 off = ti->ti_off << 2; 227 if (off < sizeof (struct tcphdr) || off > tlen) { 228 if (tcpprintfs) 229 printf("tcp off: src %x off %d\n", ti->ti_src, off); 230 tcpstat.tcps_rcvbadoff++; 231 goto drop; 232 } 233 tlen -= off; 234 ti->ti_len = tlen; 235 if (off > sizeof (struct tcphdr)) { 236 if (m->m_len < sizeof(struct ip) + off) { 237 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 238 tcpstat.tcps_rcvshort++; 239 return; 240 } 241 ti = mtod(m, struct tcpiphdr *); 242 } 243 om = m_get(M_DONTWAIT, MT_DATA); 244 if (om == 0) 245 goto drop; 246 om->m_len = off - sizeof (struct tcphdr); 247 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 248 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 249 m->m_len -= om->m_len; 250 bcopy(op+om->m_len, op, 251 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 252 } 253 } 254 tiflags = ti->ti_flags; 255 256 /* 257 * Drop TCP and IP headers; TCP options were dropped above. 258 */ 259 m->m_off += sizeof(struct tcpiphdr); 260 m->m_len -= sizeof(struct tcpiphdr); 261 262 /* 263 * Convert TCP protocol specific fields to host format. 264 */ 265 ti->ti_seq = ntohl(ti->ti_seq); 266 ti->ti_ack = ntohl(ti->ti_ack); 267 ti->ti_win = ntohs(ti->ti_win); 268 ti->ti_urp = ntohs(ti->ti_urp); 269 270 /* 271 * Locate pcb for segment. 272 */ 273 findpcb: 274 inp = in_pcblookup 275 (&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport, 276 INPLOOKUP_WILDCARD); 277 278 /* 279 * If the state is CLOSED (i.e., TCB does not exist) then 280 * all data in the incoming segment is discarded. 281 */ 282 if (inp == 0) 283 goto dropwithreset; 284 tp = intotcpcb(inp); 285 if (tp == 0) 286 goto dropwithreset; 287 so = inp->inp_socket; 288 if (so->so_options & SO_DEBUG) { 289 ostate = tp->t_state; 290 tcp_saveti = *ti; 291 } 292 if (so->so_options & SO_ACCEPTCONN) { 293 so = sonewconn(so); 294 if (so == 0) 295 goto drop; 296 /* 297 * This is ugly, but .... 298 * 299 * Mark socket as temporary until we're 300 * committed to keeping it. The code at 301 * ``drop'' and ``dropwithreset'' check the 302 * flag dropsocket to see if the temporary 303 * socket created here should be discarded. 304 * We mark the socket as discardable until 305 * we're committed to it below in TCPS_LISTEN. 306 */ 307 dropsocket++; 308 inp = (struct inpcb *)so->so_pcb; 309 inp->inp_laddr = ti->ti_dst; 310 inp->inp_lport = ti->ti_dport; 311 inp->inp_options = ip_srcroute(); 312 tp = intotcpcb(inp); 313 tp->t_state = TCPS_LISTEN; 314 } 315 316 /* 317 * Segment received on connection. 318 * Reset idle time and keep-alive timer. 319 */ 320 tp->t_idle = 0; 321 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 322 323 /* 324 * Process options if not in LISTEN state, 325 * else do it below (after getting remote address). 326 */ 327 if (om && tp->t_state != TCPS_LISTEN) { 328 tcp_dooptions(tp, om, ti); 329 om = 0; 330 } 331 332 /* 333 * Calculate amount of space in receive window, 334 * and then do TCP input processing. 335 * Receive window is amount of space in rcv queue, 336 * but not less than advertised window. 337 */ 338 { int win; 339 340 win = sbspace(&so->so_rcv); 341 if (win < 0) 342 win = 0; 343 tp->rcv_wnd = MAX(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 344 } 345 346 switch (tp->t_state) { 347 348 /* 349 * If the state is LISTEN then ignore segment if it contains an RST. 350 * If the segment contains an ACK then it is bad and send a RST. 351 * If it does not contain a SYN then it is not interesting; drop it. 352 * Don't bother responding if the destination was a broadcast. 353 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 354 * tp->iss, and send a segment: 355 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 356 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 357 * Fill in remote peer address fields if not previously specified. 358 * Enter SYN_RECEIVED state, and process any other fields of this 359 * segment in this state. 360 */ 361 case TCPS_LISTEN: { 362 struct mbuf *am; 363 register struct sockaddr_in *sin; 364 365 if (tiflags & TH_RST) 366 goto drop; 367 if (tiflags & TH_ACK) 368 goto dropwithreset; 369 if ((tiflags & TH_SYN) == 0) 370 goto drop; 371 if (in_broadcast(ti->ti_dst)) 372 goto drop; 373 am = m_get(M_DONTWAIT, MT_SONAME); 374 if (am == NULL) 375 goto drop; 376 am->m_len = sizeof (struct sockaddr_in); 377 sin = mtod(am, struct sockaddr_in *); 378 sin->sin_family = AF_INET; 379 sin->sin_addr = ti->ti_src; 380 sin->sin_port = ti->ti_sport; 381 laddr = inp->inp_laddr; 382 if (inp->inp_laddr.s_addr == INADDR_ANY) 383 inp->inp_laddr = ti->ti_dst; 384 if (in_pcbconnect(inp, am)) { 385 inp->inp_laddr = laddr; 386 (void) m_free(am); 387 goto drop; 388 } 389 (void) m_free(am); 390 tp->t_template = tcp_template(tp); 391 if (tp->t_template == 0) { 392 tp = tcp_drop(tp, ENOBUFS); 393 dropsocket = 0; /* socket is already gone */ 394 goto drop; 395 } 396 if (om) { 397 tcp_dooptions(tp, om, ti); 398 om = 0; 399 } 400 if (iss) 401 tp->iss = iss; 402 else 403 tp->iss = tcp_iss; 404 tcp_iss += TCP_ISSINCR/2; 405 tp->irs = ti->ti_seq; 406 tcp_sendseqinit(tp); 407 tcp_rcvseqinit(tp); 408 tp->t_flags |= TF_ACKNOW; 409 tp->t_state = TCPS_SYN_RECEIVED; 410 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 411 dropsocket = 0; /* committed to socket */ 412 tcpstat.tcps_accepts++; 413 goto trimthenstep6; 414 } 415 416 /* 417 * If the state is SYN_SENT: 418 * if seg contains an ACK, but not for our SYN, drop the input. 419 * if seg contains a RST, then drop the connection. 420 * if seg does not contain SYN, then drop it. 421 * Otherwise this is an acceptable SYN segment 422 * initialize tp->rcv_nxt and tp->irs 423 * if seg contains ack then advance tp->snd_una 424 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 425 * arrange for segment to be acked (eventually) 426 * continue processing rest of data/controls, beginning with URG 427 */ 428 case TCPS_SYN_SENT: 429 if ((tiflags & TH_ACK) && 430 (SEQ_LEQ(ti->ti_ack, tp->iss) || 431 SEQ_GT(ti->ti_ack, tp->snd_max))) 432 goto dropwithreset; 433 if (tiflags & TH_RST) { 434 if (tiflags & TH_ACK) 435 tp = tcp_drop(tp, ECONNREFUSED); 436 goto drop; 437 } 438 if ((tiflags & TH_SYN) == 0) 439 goto drop; 440 if (tiflags & TH_ACK) { 441 tp->snd_una = ti->ti_ack; 442 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 443 tp->snd_nxt = tp->snd_una; 444 } 445 tp->t_timer[TCPT_REXMT] = 0; 446 tp->irs = ti->ti_seq; 447 tcp_rcvseqinit(tp); 448 tp->t_flags |= TF_ACKNOW; 449 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 450 tcpstat.tcps_connects++; 451 soisconnected(so); 452 tp->t_state = TCPS_ESTABLISHED; 453 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 454 (void) tcp_reass(tp, (struct tcpiphdr *)0); 455 } else 456 tp->t_state = TCPS_SYN_RECEIVED; 457 458 trimthenstep6: 459 /* 460 * Advance ti->ti_seq to correspond to first data byte. 461 * If data, trim to stay within window, 462 * dropping FIN if necessary. 463 */ 464 ti->ti_seq++; 465 if (ti->ti_len > tp->rcv_wnd) { 466 todrop = ti->ti_len - tp->rcv_wnd; 467 m_adj(m, -todrop); 468 ti->ti_len = tp->rcv_wnd; 469 tiflags &= ~TH_FIN; 470 tcpstat.tcps_rcvpackafterwin++; 471 tcpstat.tcps_rcvbyteafterwin += todrop; 472 } 473 tp->snd_wl1 = ti->ti_seq - 1; 474 tp->rcv_up = ti->ti_seq; 475 goto step6; 476 } 477 478 /* 479 * States other than LISTEN or SYN_SENT. 480 * First check that at least some bytes of segment are within 481 * receive window. If segment begins before rcv_nxt, 482 * drop leading data (and SYN); if nothing left, just ack. 483 */ 484 todrop = tp->rcv_nxt - ti->ti_seq; 485 if (todrop > 0) { 486 if (tiflags & TH_SYN) { 487 tiflags &= ~TH_SYN; 488 ti->ti_seq++; 489 if (ti->ti_urp > 1) 490 ti->ti_urp--; 491 else 492 tiflags &= ~TH_URG; 493 todrop--; 494 } 495 if (todrop > ti->ti_len || 496 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 497 tcpstat.tcps_rcvduppack++; 498 tcpstat.tcps_rcvdupbyte += ti->ti_len; 499 goto dropafterack; 500 } 501 tcpstat.tcps_rcvpartduppack++; 502 tcpstat.tcps_rcvpartdupbyte += todrop; 503 m_adj(m, todrop); 504 ti->ti_seq += todrop; 505 ti->ti_len -= todrop; 506 if (ti->ti_urp > todrop) 507 ti->ti_urp -= todrop; 508 else { 509 tiflags &= ~TH_URG; 510 ti->ti_urp = 0; 511 } 512 } 513 514 if (tp->rcv_wnd == 0) { 515 /* 516 * If window is closed can only take segments at 517 * window edge, and have to drop data and PUSH from 518 * incoming segments. 519 * 520 * If new data is received on a connection after the 521 * user processes are gone, then RST the other end. 522 */ 523 if ((so->so_state & SS_NOFDREF) && 524 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 525 tp = tcp_close(tp); 526 tcpstat.tcps_rcvafterclose++; 527 goto dropwithreset; 528 } 529 if (tp->rcv_nxt != ti->ti_seq) { 530 tcpstat.tcps_rcvpackafterwin++; 531 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 532 goto dropafterack; 533 } 534 if (ti->ti_len > 0) { 535 if (ti->ti_len == 1) 536 tcpstat.tcps_rcvwinprobe++; 537 else { 538 tcpstat.tcps_rcvpackafterwin++; 539 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 540 } 541 m_adj(m, ti->ti_len); 542 ti->ti_len = 0; 543 tiflags &= ~(TH_PUSH|TH_FIN); 544 } 545 } else { 546 /* 547 * If segment ends after window, drop trailing data 548 * (and PUSH and FIN); if nothing left, just ACK. 549 */ 550 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 551 if (todrop > 0) { 552 if (todrop >= ti->ti_len) { 553 /* 554 * If a new connection request is received 555 * while in TIME_WAIT, drop the old connection 556 * and start over if the sequence numbers 557 * are above the previous ones. 558 */ 559 if (tiflags & TH_SYN && 560 tp->t_state == TCPS_TIME_WAIT && 561 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 562 iss = tp->rcv_nxt + TCP_ISSINCR; 563 (void) tcp_close(tp); 564 goto findpcb; 565 } 566 if (todrop == 1) 567 tcpstat.tcps_rcvwinprobe++; 568 else { 569 tcpstat.tcps_rcvpackafterwin++; 570 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 571 } 572 goto dropafterack; 573 } 574 tcpstat.tcps_rcvpackafterwin++; 575 tcpstat.tcps_rcvbyteafterwin += todrop; 576 m_adj(m, -todrop); 577 ti->ti_len -= todrop; 578 tiflags &= ~(TH_PUSH|TH_FIN); 579 } 580 } 581 582 /* 583 * If the RST bit is set examine the state: 584 * SYN_RECEIVED STATE: 585 * If passive open, return to LISTEN state. 586 * If active open, inform user that connection was refused. 587 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 588 * Inform user that connection was reset, and close tcb. 589 * CLOSING, LAST_ACK, TIME_WAIT STATES 590 * Close the tcb. 591 */ 592 if (tiflags&TH_RST) switch (tp->t_state) { 593 594 case TCPS_SYN_RECEIVED: 595 tp = tcp_drop(tp, ECONNREFUSED); 596 goto drop; 597 598 case TCPS_ESTABLISHED: 599 case TCPS_FIN_WAIT_1: 600 case TCPS_FIN_WAIT_2: 601 case TCPS_CLOSE_WAIT: 602 tp = tcp_drop(tp, ECONNRESET); 603 goto drop; 604 605 case TCPS_CLOSING: 606 case TCPS_LAST_ACK: 607 case TCPS_TIME_WAIT: 608 tp = tcp_close(tp); 609 goto drop; 610 } 611 612 /* 613 * If a SYN is in the window, then this is an 614 * error and we send an RST and drop the connection. 615 */ 616 if (tiflags & TH_SYN) { 617 tp = tcp_drop(tp, ECONNRESET); 618 goto dropwithreset; 619 } 620 621 /* 622 * If the ACK bit is off we drop the segment and return. 623 */ 624 if ((tiflags & TH_ACK) == 0) 625 goto drop; 626 627 /* 628 * Ack processing. 629 */ 630 switch (tp->t_state) { 631 632 /* 633 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 634 * ESTABLISHED state and continue processing, otherwise 635 * send an RST. 636 */ 637 case TCPS_SYN_RECEIVED: 638 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 639 SEQ_GT(ti->ti_ack, tp->snd_max)) 640 goto dropwithreset; 641 tcpstat.tcps_connects++; 642 soisconnected(so); 643 tp->t_state = TCPS_ESTABLISHED; 644 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 645 (void) tcp_reass(tp, (struct tcpiphdr *)0); 646 tp->snd_wl1 = ti->ti_seq - 1; 647 /* fall into ... */ 648 649 /* 650 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 651 * ACKs. If the ack is in the range 652 * tp->snd_una < ti->ti_ack <= tp->snd_max 653 * then advance tp->snd_una to ti->ti_ack and drop 654 * data from the retransmission queue. If this ACK reflects 655 * more up to date window information we update our window information. 656 */ 657 case TCPS_ESTABLISHED: 658 case TCPS_FIN_WAIT_1: 659 case TCPS_FIN_WAIT_2: 660 case TCPS_CLOSE_WAIT: 661 case TCPS_CLOSING: 662 case TCPS_LAST_ACK: 663 case TCPS_TIME_WAIT: 664 665 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 666 if (ti->ti_len == 0) 667 tcpstat.tcps_rcvdupack++; 668 break; 669 } 670 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 671 tcpstat.tcps_rcvacktoomuch++; 672 goto dropafterack; 673 } 674 acked = ti->ti_ack - tp->snd_una; 675 tcpstat.tcps_rcvackpack++; 676 tcpstat.tcps_rcvackbyte += acked; 677 678 /* 679 * If transmit timer is running and timed sequence 680 * number was acked, update smoothed round trip time. 681 */ 682 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) { 683 tcpstat.tcps_rttupdated++; 684 if (tp->t_srtt != 0) { 685 register short delta; 686 687 /* 688 * srtt is stored as fixed point with 3 bits 689 * after the binary point (i.e., scaled by 8). 690 * The following magic is equivalent 691 * to the smoothing algorithm in rfc793 692 * with an alpha of .875 693 * (srtt = rtt/8 + srtt*7/8 in fixed point). 694 */ 695 delta = tp->t_rtt - (tp->t_srtt >> 3); 696 if ((tp->t_srtt += delta) <= 0) 697 tp->t_srtt = 1; 698 /* 699 * We accumulate a smoothed rtt variance, 700 * then set the retransmit timer to smoothed 701 * rtt + 2 times the smoothed variance. 702 * rttvar is strored as fixed point 703 * with 2 bits after the binary point 704 * (scaled by 4). The following is equivalent 705 * to rfc793 smoothing with an alpha of .75 706 * (rttvar = rttvar*3/4 + |delta| / 4). 707 * This replaces rfc793's wired-in beta. 708 */ 709 if (delta < 0) 710 delta = -delta; 711 delta -= (tp->t_rttvar >> 2); 712 if ((tp->t_rttvar += delta) <= 0) 713 tp->t_rttvar = 1; 714 } else { 715 /* 716 * No rtt measurement yet - use the 717 * unsmoothed rtt. Set the variance 718 * to half the rtt (so our first 719 * retransmit happens at 2*rtt) 720 */ 721 tp->t_srtt = tp->t_rtt << 3; 722 tp->t_rttvar = tp->t_rtt << 1; 723 } 724 tp->t_rtt = 0; 725 } 726 727 /* 728 * If all outstanding data is acked, stop retransmit 729 * timer and remember to restart (more output or persist). 730 * If there is more data to be acked, restart retransmit 731 * timer; set to smoothed rtt + 2*rttvar. 732 */ 733 if (ti->ti_ack == tp->snd_max) { 734 tp->t_timer[TCPT_REXMT] = 0; 735 needoutput = 1; 736 } else if (tp->t_timer[TCPT_PERSIST] == 0) { 737 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 738 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 739 TCPTV_MIN, TCPTV_REXMTMAX); 740 tp->t_rxtshift = 0; 741 } 742 /* 743 * When new data is acked, open the congestion window 744 * by one max sized segment. 745 */ 746 tp->snd_cwnd = MIN(tp->snd_cwnd + tp->t_maxseg, 65535); 747 if (acked > so->so_snd.sb_cc) { 748 tp->snd_wnd -= so->so_snd.sb_cc; 749 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 750 ourfinisacked = 1; 751 } else { 752 sbdrop(&so->so_snd, acked); 753 tp->snd_wnd -= acked; 754 ourfinisacked = 0; 755 } 756 if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel) 757 sowwakeup(so); 758 tp->snd_una = ti->ti_ack; 759 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 760 tp->snd_nxt = tp->snd_una; 761 762 switch (tp->t_state) { 763 764 /* 765 * In FIN_WAIT_1 STATE in addition to the processing 766 * for the ESTABLISHED state if our FIN is now acknowledged 767 * then enter FIN_WAIT_2. 768 */ 769 case TCPS_FIN_WAIT_1: 770 if (ourfinisacked) { 771 /* 772 * If we can't receive any more 773 * data, then closing user can proceed. 774 * Starting the timer is contrary to the 775 * specification, but if we don't get a FIN 776 * we'll hang forever. 777 */ 778 if (so->so_state & SS_CANTRCVMORE) { 779 soisdisconnected(so); 780 tp->t_timer[TCPT_2MSL] = TCPTV_MAXIDLE; 781 } 782 tp->t_state = TCPS_FIN_WAIT_2; 783 } 784 break; 785 786 /* 787 * In CLOSING STATE in addition to the processing for 788 * the ESTABLISHED state if the ACK acknowledges our FIN 789 * then enter the TIME-WAIT state, otherwise ignore 790 * the segment. 791 */ 792 case TCPS_CLOSING: 793 if (ourfinisacked) { 794 tp->t_state = TCPS_TIME_WAIT; 795 tcp_canceltimers(tp); 796 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 797 soisdisconnected(so); 798 } 799 break; 800 801 /* 802 * The only thing that can arrive in LAST_ACK state 803 * is an acknowledgment of our FIN. If our FIN is now 804 * acknowledged, delete the TCB, enter the closed state 805 * and return. 806 */ 807 case TCPS_LAST_ACK: 808 if (ourfinisacked) 809 tp = tcp_close(tp); 810 goto drop; 811 812 /* 813 * In TIME_WAIT state the only thing that should arrive 814 * is a retransmission of the remote FIN. Acknowledge 815 * it and restart the finack timer. 816 */ 817 case TCPS_TIME_WAIT: 818 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 819 goto dropafterack; 820 } 821 } 822 823 step6: 824 /* 825 * Update window information. 826 * Don't look at window if no ACK: TAC's send garbage on first SYN. 827 */ 828 if ((tiflags & TH_ACK) && 829 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 830 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 831 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 832 /* keep track of pure window updates */ 833 if (ti->ti_len == 0 && 834 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) { 835 tcpstat.tcps_rcvwinupd++; 836 tcpstat.tcps_rcvdupack--; 837 } 838 tp->snd_wnd = ti->ti_win; 839 tp->snd_wl1 = ti->ti_seq; 840 tp->snd_wl2 = ti->ti_ack; 841 if (tp->snd_wnd > tp->max_sndwnd) 842 tp->max_sndwnd = tp->snd_wnd; 843 needoutput = 1; 844 } 845 846 /* 847 * Process segments with URG. 848 */ 849 if ((tiflags & TH_URG) && ti->ti_urp && 850 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 851 /* 852 * This is a kludge, but if we receive and accept 853 * random urgent pointers, we'll crash in 854 * soreceive. It's hard to imagine someone 855 * actually wanting to send this much urgent data. 856 */ 857 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 858 ti->ti_urp = 0; /* XXX */ 859 tiflags &= ~TH_URG; /* XXX */ 860 goto dodata; /* XXX */ 861 } 862 /* 863 * If this segment advances the known urgent pointer, 864 * then mark the data stream. This should not happen 865 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 866 * a FIN has been received from the remote side. 867 * In these states we ignore the URG. 868 * 869 * According to RFC961 (Assigned Protocols), 870 * the urgent pointer points to the last octet 871 * of urgent data. We continue, however, 872 * to consider it to indicate the first octet 873 * of data past the urgent section 874 * as the original spec states. 875 */ 876 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 877 tp->rcv_up = ti->ti_seq + ti->ti_urp; 878 so->so_oobmark = so->so_rcv.sb_cc + 879 (tp->rcv_up - tp->rcv_nxt) - 1; 880 if (so->so_oobmark == 0) 881 so->so_state |= SS_RCVATMARK; 882 sohasoutofband(so); 883 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 884 } 885 /* 886 * Remove out of band data so doesn't get presented to user. 887 * This can happen independent of advancing the URG pointer, 888 * but if two URG's are pending at once, some out-of-band 889 * data may creep in... ick. 890 */ 891 if (ti->ti_urp <= ti->ti_len && 892 (so->so_options & SO_OOBINLINE) == 0) 893 tcp_pulloutofband(so, ti); 894 } else 895 /* 896 * If no out of band data is expected, 897 * pull receive urgent pointer along 898 * with the receive window. 899 */ 900 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 901 tp->rcv_up = tp->rcv_nxt; 902 dodata: /* XXX */ 903 904 /* 905 * Process the segment text, merging it into the TCP sequencing queue, 906 * and arranging for acknowledgment of receipt if necessary. 907 * This process logically involves adjusting tp->rcv_wnd as data 908 * is presented to the user (this happens in tcp_usrreq.c, 909 * case PRU_RCVD). If a FIN has already been received on this 910 * connection then we just ignore the text. 911 */ 912 if ((ti->ti_len || (tiflags&TH_FIN)) && 913 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 914 TCP_REASS(tp, ti, m, so, tiflags); 915 if (tcpnodelack == 0) 916 tp->t_flags |= TF_DELACK; 917 else 918 tp->t_flags |= TF_ACKNOW; 919 /* 920 * Note the amount of data that peer has sent into 921 * our window, in order to estimate the sender's 922 * buffer size. 923 */ 924 len = so->so_rcv.sb_hiwat - (tp->rcv_nxt - tp->rcv_adv); 925 if (len > tp->max_rcvd) 926 tp->max_rcvd = len; 927 } else { 928 m_freem(m); 929 tiflags &= ~TH_FIN; 930 } 931 932 /* 933 * If FIN is received ACK the FIN and let the user know 934 * that the connection is closing. 935 */ 936 if (tiflags & TH_FIN) { 937 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 938 socantrcvmore(so); 939 tp->t_flags |= TF_ACKNOW; 940 tp->rcv_nxt++; 941 } 942 switch (tp->t_state) { 943 944 /* 945 * In SYN_RECEIVED and ESTABLISHED STATES 946 * enter the CLOSE_WAIT state. 947 */ 948 case TCPS_SYN_RECEIVED: 949 case TCPS_ESTABLISHED: 950 tp->t_state = TCPS_CLOSE_WAIT; 951 break; 952 953 /* 954 * If still in FIN_WAIT_1 STATE FIN has not been acked so 955 * enter the CLOSING state. 956 */ 957 case TCPS_FIN_WAIT_1: 958 tp->t_state = TCPS_CLOSING; 959 break; 960 961 /* 962 * In FIN_WAIT_2 state enter the TIME_WAIT state, 963 * starting the time-wait timer, turning off the other 964 * standard timers. 965 */ 966 case TCPS_FIN_WAIT_2: 967 tp->t_state = TCPS_TIME_WAIT; 968 tcp_canceltimers(tp); 969 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 970 soisdisconnected(so); 971 break; 972 973 /* 974 * In TIME_WAIT state restart the 2 MSL time_wait timer. 975 */ 976 case TCPS_TIME_WAIT: 977 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 978 break; 979 } 980 } 981 if (so->so_options & SO_DEBUG) 982 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 983 984 /* 985 * Return any desired output. 986 */ 987 if (needoutput || (tp->t_flags & TF_ACKNOW)) 988 (void) tcp_output(tp); 989 return; 990 991 dropafterack: 992 /* 993 * Generate an ACK dropping incoming segment if it occupies 994 * sequence space, where the ACK reflects our state. 995 */ 996 if (tiflags & TH_RST) 997 goto drop; 998 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 999 tcp_trace(TA_RESPOND, ostate, tp, &tcp_saveti, 0); 1000 tp->t_flags |= TF_ACKNOW; 1001 (void) tcp_output(tp); 1002 return; 1003 1004 dropwithreset: 1005 if (om) { 1006 (void) m_free(om); 1007 om = 0; 1008 } 1009 /* 1010 * Generate a RST, dropping incoming segment. 1011 * Make ACK acceptable to originator of segment. 1012 * Don't bother to respond if destination was broadcast. 1013 */ 1014 if ((tiflags & TH_RST) || in_broadcast(ti->ti_dst)) 1015 goto drop; 1016 if (tiflags & TH_ACK) 1017 tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST); 1018 else { 1019 if (tiflags & TH_SYN) 1020 ti->ti_len++; 1021 tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1022 TH_RST|TH_ACK); 1023 } 1024 /* destroy temporarily created socket */ 1025 if (dropsocket) 1026 (void) soabort(so); 1027 return; 1028 1029 drop: 1030 if (om) 1031 (void) m_free(om); 1032 /* 1033 * Drop space held by incoming segment and return. 1034 */ 1035 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1036 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1037 m_freem(m); 1038 /* destroy temporarily created socket */ 1039 if (dropsocket) 1040 (void) soabort(so); 1041 return; 1042 } 1043 1044 tcp_dooptions(tp, om, ti) 1045 struct tcpcb *tp; 1046 struct mbuf *om; 1047 struct tcpiphdr *ti; 1048 { 1049 register u_char *cp; 1050 int opt, optlen, cnt; 1051 1052 cp = mtod(om, u_char *); 1053 cnt = om->m_len; 1054 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1055 opt = cp[0]; 1056 if (opt == TCPOPT_EOL) 1057 break; 1058 if (opt == TCPOPT_NOP) 1059 optlen = 1; 1060 else { 1061 optlen = cp[1]; 1062 if (optlen <= 0) 1063 break; 1064 } 1065 switch (opt) { 1066 1067 default: 1068 break; 1069 1070 case TCPOPT_MAXSEG: 1071 if (optlen != 4) 1072 continue; 1073 if (!(ti->ti_flags & TH_SYN)) 1074 continue; 1075 tp->t_maxseg = *(u_short *)(cp + 2); 1076 tp->t_maxseg = ntohs((u_short)tp->t_maxseg); 1077 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 1078 break; 1079 } 1080 } 1081 (void) m_free(om); 1082 } 1083 1084 /* 1085 * Pull out of band byte out of a segment so 1086 * it doesn't appear in the user's data queue. 1087 * It is still reflected in the segment length for 1088 * sequencing purposes. 1089 */ 1090 tcp_pulloutofband(so, ti) 1091 struct socket *so; 1092 struct tcpiphdr *ti; 1093 { 1094 register struct mbuf *m; 1095 int cnt = ti->ti_urp - 1; 1096 1097 m = dtom(ti); 1098 while (cnt >= 0) { 1099 if (m->m_len > cnt) { 1100 char *cp = mtod(m, caddr_t) + cnt; 1101 struct tcpcb *tp = sototcpcb(so); 1102 1103 tp->t_iobc = *cp; 1104 tp->t_oobflags |= TCPOOB_HAVEDATA; 1105 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1106 m->m_len--; 1107 return; 1108 } 1109 cnt -= m->m_len; 1110 m = m->m_next; 1111 if (m == 0) 1112 break; 1113 } 1114 panic("tcp_pulloutofband"); 1115 } 1116 1117 /* 1118 * Determine a reasonable value for maxseg size. 1119 * If the route is known, use one that can be handled 1120 * on the given interface without forcing IP to fragment. 1121 * If bigger than an mbuf cluster (MCLBYTES), round down to nearest size 1122 * to utilize large mbufs. 1123 * If interface pointer is unavailable, or the destination isn't local, 1124 * use a conservative size (512 or the default IP max size, but no more 1125 * than the mtu of the interface through which we route), 1126 * as we can't discover anything about intervening gateways or networks. 1127 * 1128 * This is ugly, and doesn't belong at this level, but has to happen somehow. 1129 */ 1130 tcp_mss(tp) 1131 register struct tcpcb *tp; 1132 { 1133 struct route *ro; 1134 struct ifnet *ifp; 1135 int mss; 1136 struct inpcb *inp; 1137 1138 inp = tp->t_inpcb; 1139 ro = &inp->inp_route; 1140 if ((ro->ro_rt == (struct rtentry *)0) || 1141 (ifp = ro->ro_rt->rt_ifp) == (struct ifnet *)0) { 1142 /* No route yet, so try to acquire one */ 1143 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1144 ro->ro_dst.sa_family = AF_INET; 1145 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1146 inp->inp_faddr; 1147 rtalloc(ro); 1148 } 1149 if ((ro->ro_rt == 0) || (ifp = ro->ro_rt->rt_ifp) == 0) 1150 return (TCP_MSS); 1151 } 1152 1153 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1154 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1155 if (mss > MCLBYTES) 1156 mss &= ~(MCLBYTES-1); 1157 #else 1158 if (mss > MCLBYTES) 1159 mss = mss / MCLBYTES * MCLBYTES; 1160 #endif 1161 if (in_localaddr(inp->inp_faddr)) 1162 return (mss); 1163 return (MIN(mss, TCP_MSS)); 1164 } 1165