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 7.8 (Berkeley) 07/01/87 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 #ifdef TCP_COMPAT_42 498 /* 499 * Don't toss RST in response to 4.2-style keepalive. 500 */ 501 if (ti->ti_seq == tp->rcv_nxt - 1 && tiflags & TH_RST) 502 goto do_rst; 503 #endif 504 tcpstat.tcps_rcvduppack++; 505 tcpstat.tcps_rcvdupbyte += ti->ti_len; 506 goto dropafterack; 507 } 508 tcpstat.tcps_rcvpartduppack++; 509 tcpstat.tcps_rcvpartdupbyte += todrop; 510 m_adj(m, todrop); 511 ti->ti_seq += todrop; 512 ti->ti_len -= todrop; 513 if (ti->ti_urp > todrop) 514 ti->ti_urp -= todrop; 515 else { 516 tiflags &= ~TH_URG; 517 ti->ti_urp = 0; 518 } 519 } 520 521 if (tp->rcv_wnd == 0) { 522 /* 523 * If window is closed can only take segments at 524 * window edge, and have to drop data and PUSH from 525 * incoming segments. 526 * 527 * If new data is received on a connection after the 528 * user processes are gone, then RST the other end. 529 */ 530 if ((so->so_state & SS_NOFDREF) && 531 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 532 tp = tcp_close(tp); 533 tcpstat.tcps_rcvafterclose++; 534 goto dropwithreset; 535 } 536 if (tp->rcv_nxt != ti->ti_seq) { 537 tcpstat.tcps_rcvpackafterwin++; 538 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 539 goto dropafterack; 540 } 541 if (ti->ti_len > 0) { 542 if (ti->ti_len == 1) 543 tcpstat.tcps_rcvwinprobe++; 544 else { 545 tcpstat.tcps_rcvpackafterwin++; 546 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 547 } 548 m_adj(m, ti->ti_len); 549 ti->ti_len = 0; 550 tiflags &= ~(TH_PUSH|TH_FIN); 551 } 552 } else { 553 /* 554 * If segment ends after window, drop trailing data 555 * (and PUSH and FIN); if nothing left, just ACK. 556 */ 557 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 558 if (todrop > 0) { 559 if (todrop >= ti->ti_len) { 560 /* 561 * If a new connection request is received 562 * while in TIME_WAIT, drop the old connection 563 * and start over if the sequence numbers 564 * are above the previous ones. 565 */ 566 if (tiflags & TH_SYN && 567 tp->t_state == TCPS_TIME_WAIT && 568 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 569 iss = tp->rcv_nxt + TCP_ISSINCR; 570 (void) tcp_close(tp); 571 goto findpcb; 572 } 573 if (todrop == 1) 574 tcpstat.tcps_rcvwinprobe++; 575 else { 576 tcpstat.tcps_rcvpackafterwin++; 577 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 578 } 579 goto dropafterack; 580 } 581 tcpstat.tcps_rcvpackafterwin++; 582 tcpstat.tcps_rcvbyteafterwin += todrop; 583 m_adj(m, -todrop); 584 ti->ti_len -= todrop; 585 tiflags &= ~(TH_PUSH|TH_FIN); 586 } 587 } 588 589 #ifdef TCP_COMPAT_42 590 do_rst: 591 #endif 592 /* 593 * If the RST bit is set examine the state: 594 * SYN_RECEIVED STATE: 595 * If passive open, return to LISTEN state. 596 * If active open, inform user that connection was refused. 597 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 598 * Inform user that connection was reset, and close tcb. 599 * CLOSING, LAST_ACK, TIME_WAIT STATES 600 * Close the tcb. 601 */ 602 if (tiflags&TH_RST) switch (tp->t_state) { 603 604 case TCPS_SYN_RECEIVED: 605 tp = tcp_drop(tp, ECONNREFUSED); 606 goto drop; 607 608 case TCPS_ESTABLISHED: 609 case TCPS_FIN_WAIT_1: 610 case TCPS_FIN_WAIT_2: 611 case TCPS_CLOSE_WAIT: 612 tp = tcp_drop(tp, ECONNRESET); 613 goto drop; 614 615 case TCPS_CLOSING: 616 case TCPS_LAST_ACK: 617 case TCPS_TIME_WAIT: 618 tp = tcp_close(tp); 619 goto drop; 620 } 621 622 /* 623 * If a SYN is in the window, then this is an 624 * error and we send an RST and drop the connection. 625 */ 626 if (tiflags & TH_SYN) { 627 tp = tcp_drop(tp, ECONNRESET); 628 goto dropwithreset; 629 } 630 631 /* 632 * If the ACK bit is off we drop the segment and return. 633 */ 634 if ((tiflags & TH_ACK) == 0) 635 goto drop; 636 637 /* 638 * Ack processing. 639 */ 640 switch (tp->t_state) { 641 642 /* 643 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 644 * ESTABLISHED state and continue processing, otherwise 645 * send an RST. 646 */ 647 case TCPS_SYN_RECEIVED: 648 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 649 SEQ_GT(ti->ti_ack, tp->snd_max)) 650 goto dropwithreset; 651 tcpstat.tcps_connects++; 652 soisconnected(so); 653 tp->t_state = TCPS_ESTABLISHED; 654 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 655 (void) tcp_reass(tp, (struct tcpiphdr *)0); 656 tp->snd_wl1 = ti->ti_seq - 1; 657 /* fall into ... */ 658 659 /* 660 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 661 * ACKs. If the ack is in the range 662 * tp->snd_una < ti->ti_ack <= tp->snd_max 663 * then advance tp->snd_una to ti->ti_ack and drop 664 * data from the retransmission queue. If this ACK reflects 665 * more up to date window information we update our window information. 666 */ 667 case TCPS_ESTABLISHED: 668 case TCPS_FIN_WAIT_1: 669 case TCPS_FIN_WAIT_2: 670 case TCPS_CLOSE_WAIT: 671 case TCPS_CLOSING: 672 case TCPS_LAST_ACK: 673 case TCPS_TIME_WAIT: 674 675 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 676 if (ti->ti_len == 0) 677 tcpstat.tcps_rcvdupack++; 678 break; 679 } 680 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 681 tcpstat.tcps_rcvacktoomuch++; 682 goto dropafterack; 683 } 684 acked = ti->ti_ack - tp->snd_una; 685 tcpstat.tcps_rcvackpack++; 686 tcpstat.tcps_rcvackbyte += acked; 687 688 /* 689 * If transmit timer is running and timed sequence 690 * number was acked, update smoothed round trip time. 691 */ 692 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) { 693 tcpstat.tcps_rttupdated++; 694 if (tp->t_srtt != 0) { 695 register short delta; 696 697 /* 698 * srtt is stored as fixed point with 3 bits 699 * after the binary point (i.e., scaled by 8). 700 * The following magic is equivalent 701 * to the smoothing algorithm in rfc793 702 * with an alpha of .875 703 * (srtt = rtt/8 + srtt*7/8 in fixed point). 704 */ 705 delta = tp->t_rtt - (tp->t_srtt >> 3); 706 if ((tp->t_srtt += delta) <= 0) 707 tp->t_srtt = 1; 708 /* 709 * We accumulate a smoothed rtt variance, 710 * then set the retransmit timer to smoothed 711 * rtt + 2 times the smoothed variance. 712 * rttvar is strored as fixed point 713 * with 2 bits after the binary point 714 * (scaled by 4). The following is equivalent 715 * to rfc793 smoothing with an alpha of .75 716 * (rttvar = rttvar*3/4 + |delta| / 4). 717 * This replaces rfc793's wired-in beta. 718 */ 719 if (delta < 0) 720 delta = -delta; 721 delta -= (tp->t_rttvar >> 2); 722 if ((tp->t_rttvar += delta) <= 0) 723 tp->t_rttvar = 1; 724 } else { 725 /* 726 * No rtt measurement yet - use the 727 * unsmoothed rtt. Set the variance 728 * to half the rtt (so our first 729 * retransmit happens at 2*rtt) 730 */ 731 tp->t_srtt = tp->t_rtt << 3; 732 tp->t_rttvar = tp->t_rtt << 1; 733 } 734 tp->t_rtt = 0; 735 } 736 737 /* 738 * If all outstanding data is acked, stop retransmit 739 * timer and remember to restart (more output or persist). 740 * If there is more data to be acked, restart retransmit 741 * timer; set to smoothed rtt + 2*rttvar. 742 */ 743 if (ti->ti_ack == tp->snd_max) { 744 tp->t_timer[TCPT_REXMT] = 0; 745 needoutput = 1; 746 } else if (tp->t_timer[TCPT_PERSIST] == 0) { 747 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 748 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 749 TCPTV_MIN, TCPTV_REXMTMAX); 750 tp->t_rxtshift = 0; 751 } 752 /* 753 * When new data is acked, open the congestion window 754 * by one max sized segment. 755 */ 756 tp->snd_cwnd = MIN(tp->snd_cwnd + tp->t_maxseg, 65535); 757 if (acked > so->so_snd.sb_cc) { 758 tp->snd_wnd -= so->so_snd.sb_cc; 759 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 760 ourfinisacked = 1; 761 } else { 762 sbdrop(&so->so_snd, acked); 763 tp->snd_wnd -= acked; 764 ourfinisacked = 0; 765 } 766 if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel) 767 sowwakeup(so); 768 tp->snd_una = ti->ti_ack; 769 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 770 tp->snd_nxt = tp->snd_una; 771 772 switch (tp->t_state) { 773 774 /* 775 * In FIN_WAIT_1 STATE in addition to the processing 776 * for the ESTABLISHED state if our FIN is now acknowledged 777 * then enter FIN_WAIT_2. 778 */ 779 case TCPS_FIN_WAIT_1: 780 if (ourfinisacked) { 781 /* 782 * If we can't receive any more 783 * data, then closing user can proceed. 784 * Starting the timer is contrary to the 785 * specification, but if we don't get a FIN 786 * we'll hang forever. 787 */ 788 if (so->so_state & SS_CANTRCVMORE) { 789 soisdisconnected(so); 790 tp->t_timer[TCPT_2MSL] = TCPTV_MAXIDLE; 791 } 792 tp->t_state = TCPS_FIN_WAIT_2; 793 } 794 break; 795 796 /* 797 * In CLOSING STATE in addition to the processing for 798 * the ESTABLISHED state if the ACK acknowledges our FIN 799 * then enter the TIME-WAIT state, otherwise ignore 800 * the segment. 801 */ 802 case TCPS_CLOSING: 803 if (ourfinisacked) { 804 tp->t_state = TCPS_TIME_WAIT; 805 tcp_canceltimers(tp); 806 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 807 soisdisconnected(so); 808 } 809 break; 810 811 /* 812 * In LAST_ACK, we may still be waiting for data to drain 813 * and/or to be acked, as well as for the ack of our FIN. 814 * If our FIN is now acknowledged, delete the TCB, 815 * enter the closed state and return. 816 */ 817 case TCPS_LAST_ACK: 818 if (ourfinisacked) { 819 tp = tcp_close(tp); 820 goto drop; 821 } 822 break; 823 824 /* 825 * In TIME_WAIT state the only thing that should arrive 826 * is a retransmission of the remote FIN. Acknowledge 827 * it and restart the finack timer. 828 */ 829 case TCPS_TIME_WAIT: 830 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 831 goto dropafterack; 832 } 833 } 834 835 step6: 836 /* 837 * Update window information. 838 * Don't look at window if no ACK: TAC's send garbage on first SYN. 839 */ 840 if ((tiflags & TH_ACK) && 841 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 842 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 843 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 844 /* keep track of pure window updates */ 845 if (ti->ti_len == 0 && 846 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) { 847 tcpstat.tcps_rcvwinupd++; 848 tcpstat.tcps_rcvdupack--; 849 } 850 tp->snd_wnd = ti->ti_win; 851 tp->snd_wl1 = ti->ti_seq; 852 tp->snd_wl2 = ti->ti_ack; 853 if (tp->snd_wnd > tp->max_sndwnd) 854 tp->max_sndwnd = tp->snd_wnd; 855 needoutput = 1; 856 } 857 858 /* 859 * Process segments with URG. 860 */ 861 if ((tiflags & TH_URG) && ti->ti_urp && 862 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 863 /* 864 * This is a kludge, but if we receive and accept 865 * random urgent pointers, we'll crash in 866 * soreceive. It's hard to imagine someone 867 * actually wanting to send this much urgent data. 868 */ 869 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 870 ti->ti_urp = 0; /* XXX */ 871 tiflags &= ~TH_URG; /* XXX */ 872 goto dodata; /* XXX */ 873 } 874 /* 875 * If this segment advances the known urgent pointer, 876 * then mark the data stream. This should not happen 877 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 878 * a FIN has been received from the remote side. 879 * In these states we ignore the URG. 880 * 881 * According to RFC961 (Assigned Protocols), 882 * the urgent pointer points to the last octet 883 * of urgent data. We continue, however, 884 * to consider it to indicate the first octet 885 * of data past the urgent section 886 * as the original spec states. 887 */ 888 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 889 tp->rcv_up = ti->ti_seq + ti->ti_urp; 890 so->so_oobmark = so->so_rcv.sb_cc + 891 (tp->rcv_up - tp->rcv_nxt) - 1; 892 if (so->so_oobmark == 0) 893 so->so_state |= SS_RCVATMARK; 894 sohasoutofband(so); 895 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 896 } 897 /* 898 * Remove out of band data so doesn't get presented to user. 899 * This can happen independent of advancing the URG pointer, 900 * but if two URG's are pending at once, some out-of-band 901 * data may creep in... ick. 902 */ 903 if (ti->ti_urp <= ti->ti_len && 904 (so->so_options & SO_OOBINLINE) == 0) 905 tcp_pulloutofband(so, ti); 906 } else 907 /* 908 * If no out of band data is expected, 909 * pull receive urgent pointer along 910 * with the receive window. 911 */ 912 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 913 tp->rcv_up = tp->rcv_nxt; 914 dodata: /* XXX */ 915 916 /* 917 * Process the segment text, merging it into the TCP sequencing queue, 918 * and arranging for acknowledgment of receipt if necessary. 919 * This process logically involves adjusting tp->rcv_wnd as data 920 * is presented to the user (this happens in tcp_usrreq.c, 921 * case PRU_RCVD). If a FIN has already been received on this 922 * connection then we just ignore the text. 923 */ 924 if ((ti->ti_len || (tiflags&TH_FIN)) && 925 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 926 TCP_REASS(tp, ti, m, so, tiflags); 927 if (tcpnodelack == 0) 928 tp->t_flags |= TF_DELACK; 929 else 930 tp->t_flags |= TF_ACKNOW; 931 /* 932 * Note the amount of data that peer has sent into 933 * our window, in order to estimate the sender's 934 * buffer size. 935 */ 936 len = so->so_rcv.sb_hiwat - (tp->rcv_nxt - tp->rcv_adv); 937 if (len > tp->max_rcvd) 938 tp->max_rcvd = len; 939 } else { 940 m_freem(m); 941 tiflags &= ~TH_FIN; 942 } 943 944 /* 945 * If FIN is received ACK the FIN and let the user know 946 * that the connection is closing. 947 */ 948 if (tiflags & TH_FIN) { 949 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 950 socantrcvmore(so); 951 tp->t_flags |= TF_ACKNOW; 952 tp->rcv_nxt++; 953 } 954 switch (tp->t_state) { 955 956 /* 957 * In SYN_RECEIVED and ESTABLISHED STATES 958 * enter the CLOSE_WAIT state. 959 */ 960 case TCPS_SYN_RECEIVED: 961 case TCPS_ESTABLISHED: 962 tp->t_state = TCPS_CLOSE_WAIT; 963 break; 964 965 /* 966 * If still in FIN_WAIT_1 STATE FIN has not been acked so 967 * enter the CLOSING state. 968 */ 969 case TCPS_FIN_WAIT_1: 970 tp->t_state = TCPS_CLOSING; 971 break; 972 973 /* 974 * In FIN_WAIT_2 state enter the TIME_WAIT state, 975 * starting the time-wait timer, turning off the other 976 * standard timers. 977 */ 978 case TCPS_FIN_WAIT_2: 979 tp->t_state = TCPS_TIME_WAIT; 980 tcp_canceltimers(tp); 981 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 982 soisdisconnected(so); 983 break; 984 985 /* 986 * In TIME_WAIT state restart the 2 MSL time_wait timer. 987 */ 988 case TCPS_TIME_WAIT: 989 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 990 break; 991 } 992 } 993 if (so->so_options & SO_DEBUG) 994 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 995 996 /* 997 * Return any desired output. 998 */ 999 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1000 (void) tcp_output(tp); 1001 return; 1002 1003 dropafterack: 1004 /* 1005 * Generate an ACK dropping incoming segment if it occupies 1006 * sequence space, where the ACK reflects our state. 1007 */ 1008 if (tiflags & TH_RST) 1009 goto drop; 1010 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 1011 tcp_trace(TA_RESPOND, ostate, tp, &tcp_saveti, 0); 1012 tp->t_flags |= TF_ACKNOW; 1013 (void) tcp_output(tp); 1014 return; 1015 1016 dropwithreset: 1017 if (om) { 1018 (void) m_free(om); 1019 om = 0; 1020 } 1021 /* 1022 * Generate a RST, dropping incoming segment. 1023 * Make ACK acceptable to originator of segment. 1024 * Don't bother to respond if destination was broadcast. 1025 */ 1026 if ((tiflags & TH_RST) || in_broadcast(ti->ti_dst)) 1027 goto drop; 1028 if (tiflags & TH_ACK) 1029 tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST); 1030 else { 1031 if (tiflags & TH_SYN) 1032 ti->ti_len++; 1033 tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1034 TH_RST|TH_ACK); 1035 } 1036 /* destroy temporarily created socket */ 1037 if (dropsocket) 1038 (void) soabort(so); 1039 return; 1040 1041 drop: 1042 if (om) 1043 (void) m_free(om); 1044 /* 1045 * Drop space held by incoming segment and return. 1046 */ 1047 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1048 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1049 m_freem(m); 1050 /* destroy temporarily created socket */ 1051 if (dropsocket) 1052 (void) soabort(so); 1053 return; 1054 } 1055 1056 tcp_dooptions(tp, om, ti) 1057 struct tcpcb *tp; 1058 struct mbuf *om; 1059 struct tcpiphdr *ti; 1060 { 1061 register u_char *cp; 1062 int opt, optlen, cnt; 1063 1064 cp = mtod(om, u_char *); 1065 cnt = om->m_len; 1066 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1067 opt = cp[0]; 1068 if (opt == TCPOPT_EOL) 1069 break; 1070 if (opt == TCPOPT_NOP) 1071 optlen = 1; 1072 else { 1073 optlen = cp[1]; 1074 if (optlen <= 0) 1075 break; 1076 } 1077 switch (opt) { 1078 1079 default: 1080 break; 1081 1082 case TCPOPT_MAXSEG: 1083 if (optlen != 4) 1084 continue; 1085 if (!(ti->ti_flags & TH_SYN)) 1086 continue; 1087 tp->t_maxseg = *(u_short *)(cp + 2); 1088 tp->t_maxseg = ntohs((u_short)tp->t_maxseg); 1089 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 1090 break; 1091 } 1092 } 1093 (void) m_free(om); 1094 } 1095 1096 /* 1097 * Pull out of band byte out of a segment so 1098 * it doesn't appear in the user's data queue. 1099 * It is still reflected in the segment length for 1100 * sequencing purposes. 1101 */ 1102 tcp_pulloutofband(so, ti) 1103 struct socket *so; 1104 struct tcpiphdr *ti; 1105 { 1106 register struct mbuf *m; 1107 int cnt = ti->ti_urp - 1; 1108 1109 m = dtom(ti); 1110 while (cnt >= 0) { 1111 if (m->m_len > cnt) { 1112 char *cp = mtod(m, caddr_t) + cnt; 1113 struct tcpcb *tp = sototcpcb(so); 1114 1115 tp->t_iobc = *cp; 1116 tp->t_oobflags |= TCPOOB_HAVEDATA; 1117 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1118 m->m_len--; 1119 return; 1120 } 1121 cnt -= m->m_len; 1122 m = m->m_next; 1123 if (m == 0) 1124 break; 1125 } 1126 panic("tcp_pulloutofband"); 1127 } 1128 1129 /* 1130 * Determine a reasonable value for maxseg size. 1131 * If the route is known, use one that can be handled 1132 * on the given interface without forcing IP to fragment. 1133 * If bigger than an mbuf cluster (MCLBYTES), round down to nearest size 1134 * to utilize large mbufs. 1135 * If interface pointer is unavailable, or the destination isn't local, 1136 * use a conservative size (512 or the default IP max size, but no more 1137 * than the mtu of the interface through which we route), 1138 * as we can't discover anything about intervening gateways or networks. 1139 * 1140 * This is ugly, and doesn't belong at this level, but has to happen somehow. 1141 */ 1142 tcp_mss(tp) 1143 register struct tcpcb *tp; 1144 { 1145 struct route *ro; 1146 struct ifnet *ifp; 1147 int mss; 1148 struct inpcb *inp; 1149 1150 inp = tp->t_inpcb; 1151 ro = &inp->inp_route; 1152 if ((ro->ro_rt == (struct rtentry *)0) || 1153 (ifp = ro->ro_rt->rt_ifp) == (struct ifnet *)0) { 1154 /* No route yet, so try to acquire one */ 1155 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1156 ro->ro_dst.sa_family = AF_INET; 1157 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1158 inp->inp_faddr; 1159 rtalloc(ro); 1160 } 1161 if ((ro->ro_rt == 0) || (ifp = ro->ro_rt->rt_ifp) == 0) 1162 return (TCP_MSS); 1163 } 1164 1165 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1166 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1167 if (mss > MCLBYTES) 1168 mss &= ~(MCLBYTES-1); 1169 #else 1170 if (mss > MCLBYTES) 1171 mss = mss / MCLBYTES * MCLBYTES; 1172 #endif 1173 if (in_localaddr(inp->inp_faddr)) 1174 return (mss); 1175 return (MIN(mss, TCP_MSS)); 1176 } 1177