1 /* 2 * Copyright (c) 1982 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 6.13 (Berkeley) 09/16/85 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 sbappend(&(so)->so_rcv, (m)); \ 55 sorwakeup(so); \ 56 } else \ 57 (flags) = tcp_reass((tp), (ti)); \ 58 } 59 60 tcp_reass(tp, ti) 61 register struct tcpcb *tp; 62 register struct tcpiphdr *ti; 63 { 64 register struct tcpiphdr *q; 65 struct socket *so = tp->t_inpcb->inp_socket; 66 struct mbuf *m; 67 int flags; 68 69 /* 70 * Call with ti==0 after become established to 71 * force pre-ESTABLISHED data up to user socket. 72 */ 73 if (ti == 0) 74 goto present; 75 76 /* 77 * Find a segment which begins after this one does. 78 */ 79 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 80 q = (struct tcpiphdr *)q->ti_next) 81 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 82 break; 83 84 /* 85 * If there is a preceding segment, it may provide some of 86 * our data already. If so, drop the data from the incoming 87 * segment. If it provides all of our data, drop us. 88 */ 89 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 90 register int i; 91 q = (struct tcpiphdr *)q->ti_prev; 92 /* conversion to int (in i) handles seq wraparound */ 93 i = q->ti_seq + q->ti_len - ti->ti_seq; 94 if (i > 0) { 95 if (i >= ti->ti_len) 96 goto drop; 97 m_adj(dtom(ti), i); 98 ti->ti_len -= i; 99 ti->ti_seq += i; 100 } 101 q = (struct tcpiphdr *)(q->ti_next); 102 } 103 104 /* 105 * While we overlap succeeding segments trim them or, 106 * if they are completely covered, dequeue them. 107 */ 108 while (q != (struct tcpiphdr *)tp) { 109 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 110 if (i <= 0) 111 break; 112 if (i < q->ti_len) { 113 q->ti_seq += i; 114 q->ti_len -= i; 115 m_adj(dtom(q), i); 116 break; 117 } 118 q = (struct tcpiphdr *)q->ti_next; 119 m = dtom(q->ti_prev); 120 remque(q->ti_prev); 121 m_freem(m); 122 } 123 124 /* 125 * Stick new segment in its place. 126 */ 127 insque(ti, q->ti_prev); 128 129 present: 130 /* 131 * Present data to user, advancing rcv_nxt through 132 * completed sequence space. 133 */ 134 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 135 return (0); 136 ti = tp->seg_next; 137 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 138 return (0); 139 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 140 return (0); 141 do { 142 tp->rcv_nxt += ti->ti_len; 143 flags = ti->ti_flags & TH_FIN; 144 remque(ti); 145 m = dtom(ti); 146 ti = (struct tcpiphdr *)ti->ti_next; 147 if (so->so_state & SS_CANTRCVMORE) 148 m_freem(m); 149 else 150 sbappend(&so->so_rcv, m); 151 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 152 sorwakeup(so); 153 return (flags); 154 drop: 155 m_freem(dtom(ti)); 156 return (0); 157 } 158 159 /* 160 * TCP input routine, follows pages 65-76 of the 161 * protocol specification dated September, 1981 very closely. 162 */ 163 tcp_input(m0) 164 struct mbuf *m0; 165 { 166 register struct tcpiphdr *ti; 167 struct inpcb *inp; 168 register struct mbuf *m; 169 struct mbuf *om = 0; 170 int len, tlen, off; 171 register struct tcpcb *tp = 0; 172 register int tiflags; 173 struct socket *so; 174 int todrop, acked; 175 short ostate; 176 struct in_addr laddr; 177 int dropsocket = 0; 178 179 /* 180 * Get IP and TCP header together in first mbuf. 181 * Note: IP leaves IP header in first mbuf. 182 */ 183 m = m0; 184 ti = mtod(m, struct tcpiphdr *); 185 if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2)) 186 ip_stripoptions((struct ip *)ti, (struct mbuf *)0); 187 if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) { 188 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 189 tcpstat.tcps_hdrops++; 190 return; 191 } 192 ti = mtod(m, struct tcpiphdr *); 193 } 194 195 /* 196 * Checksum extended TCP header and data. 197 */ 198 tlen = ((struct ip *)ti)->ip_len; 199 len = sizeof (struct ip) + tlen; 200 if (tcpcksum) { 201 ti->ti_next = ti->ti_prev = 0; 202 ti->ti_x1 = 0; 203 ti->ti_len = (u_short)tlen; 204 ti->ti_len = htons((u_short)ti->ti_len); 205 if (ti->ti_sum = in_cksum(m, len)) { 206 if (tcpprintfs) 207 printf("tcp sum: src %x\n", ti->ti_src); 208 tcpstat.tcps_badsum++; 209 goto drop; 210 } 211 } 212 213 /* 214 * Check that TCP offset makes sense, 215 * pull out TCP options and adjust length. 216 */ 217 off = ti->ti_off << 2; 218 if (off < sizeof (struct tcphdr) || off > tlen) { 219 if (tcpprintfs) 220 printf("tcp off: src %x off %d\n", ti->ti_src, off); 221 tcpstat.tcps_badoff++; 222 goto drop; 223 } 224 tlen -= off; 225 ti->ti_len = tlen; 226 if (off > sizeof (struct tcphdr)) { 227 if (m->m_len < sizeof(struct ip) + off) { 228 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 229 tcpstat.tcps_hdrops++; 230 return; 231 } 232 ti = mtod(m, struct tcpiphdr *); 233 } 234 om = m_get(M_DONTWAIT, MT_DATA); 235 if (om == 0) 236 goto drop; 237 om->m_len = off - sizeof (struct tcphdr); 238 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 239 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 240 m->m_len -= om->m_len; 241 bcopy(op+om->m_len, op, 242 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 243 } 244 } 245 tiflags = ti->ti_flags; 246 247 /* 248 * Drop TCP and IP headers. 249 */ 250 off += sizeof (struct ip); 251 m->m_off += off; 252 m->m_len -= off; 253 254 /* 255 * Convert TCP protocol specific fields to host format. 256 */ 257 ti->ti_seq = ntohl(ti->ti_seq); 258 ti->ti_ack = ntohl(ti->ti_ack); 259 ti->ti_win = ntohs(ti->ti_win); 260 ti->ti_urp = ntohs(ti->ti_urp); 261 262 /* 263 * Locate pcb for segment. 264 */ 265 inp = in_pcblookup 266 (&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport, 267 INPLOOKUP_WILDCARD); 268 269 /* 270 * If the state is CLOSED (i.e., TCB does not exist) then 271 * all data in the incoming segment is discarded. 272 */ 273 if (inp == 0) 274 goto dropwithreset; 275 tp = intotcpcb(inp); 276 if (tp == 0) 277 goto dropwithreset; 278 so = inp->inp_socket; 279 if (so->so_options & SO_DEBUG) { 280 ostate = tp->t_state; 281 tcp_saveti = *ti; 282 } 283 if (so->so_options & SO_ACCEPTCONN) { 284 so = sonewconn(so); 285 if (so == 0) 286 goto drop; 287 /* 288 * This is ugly, but .... 289 * 290 * Mark socket as temporary until we're 291 * committed to keeping it. The code at 292 * ``drop'' and ``dropwithreset'' check the 293 * flag dropsocket to see if the temporary 294 * socket created here should be discarded. 295 * We mark the socket as discardable until 296 * we're committed to it below in TCPS_LISTEN. 297 */ 298 dropsocket++; 299 inp = (struct inpcb *)so->so_pcb; 300 inp->inp_laddr = ti->ti_dst; 301 inp->inp_lport = ti->ti_dport; 302 inp->inp_options = ip_srcroute(); 303 tp = intotcpcb(inp); 304 tp->t_state = TCPS_LISTEN; 305 } 306 307 /* 308 * Segment received on connection. 309 * Reset idle time and keep-alive timer. 310 */ 311 tp->t_idle = 0; 312 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 313 314 /* 315 * Process options if not in LISTEN state, 316 * else do it below (after getting remote address). 317 */ 318 if (om && tp->t_state != TCPS_LISTEN) { 319 tcp_dooptions(tp, om, ti); 320 om = 0; 321 } 322 323 /* 324 * Calculate amount of space in receive window, 325 * and then do TCP input processing. 326 * Receive window is amount of space in rcv queue, 327 * but not less than advertised window. 328 */ 329 tp->rcv_wnd = MAX(sbspace(&so->so_rcv), tp->rcv_adv - tp->rcv_nxt); 330 if (tp->rcv_wnd < 0) 331 tp->rcv_wnd = 0; 332 333 switch (tp->t_state) { 334 335 /* 336 * If the state is LISTEN then ignore segment if it contains an RST. 337 * If the segment contains an ACK then it is bad and send a RST. 338 * If it does not contain a SYN then it is not interesting; drop it. 339 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 340 * tp->iss, and send a segment: 341 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 342 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 343 * Fill in remote peer address fields if not previously specified. 344 * Enter SYN_RECEIVED state, and process any other fields of this 345 * segment in this state. 346 */ 347 case TCPS_LISTEN: { 348 struct mbuf *am; 349 register struct sockaddr_in *sin; 350 351 if (tiflags & TH_RST) 352 goto drop; 353 if (tiflags & TH_ACK) 354 goto dropwithreset; 355 if ((tiflags & TH_SYN) == 0) 356 goto drop; 357 am = m_get(M_DONTWAIT, MT_SONAME); 358 if (am == NULL) 359 goto drop; 360 am->m_len = sizeof (struct sockaddr_in); 361 sin = mtod(am, struct sockaddr_in *); 362 sin->sin_family = AF_INET; 363 sin->sin_addr = ti->ti_src; 364 sin->sin_port = ti->ti_sport; 365 laddr = inp->inp_laddr; 366 if (inp->inp_laddr.s_addr == INADDR_ANY) 367 inp->inp_laddr = ti->ti_dst; 368 if (in_pcbconnect(inp, am)) { 369 inp->inp_laddr = laddr; 370 (void) m_free(am); 371 goto drop; 372 } 373 (void) m_free(am); 374 tp->t_template = tcp_template(tp); 375 if (tp->t_template == 0) { 376 in_pcbdisconnect(inp); 377 dropsocket = 0; /* socket is already gone */ 378 tp = 0; 379 goto drop; 380 } 381 if (om) { 382 tcp_dooptions(tp, om, ti); 383 om = 0; 384 } 385 tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2; 386 tp->irs = ti->ti_seq; 387 tcp_sendseqinit(tp); 388 tcp_rcvseqinit(tp); 389 tp->t_state = TCPS_SYN_RECEIVED; 390 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 391 dropsocket = 0; /* committed to socket */ 392 goto trimthenstep6; 393 } 394 395 /* 396 * If the state is SYN_SENT: 397 * if seg contains an ACK, but not for our SYN, drop the input. 398 * if seg contains a RST, then drop the connection. 399 * if seg does not contain SYN, then drop it. 400 * Otherwise this is an acceptable SYN segment 401 * initialize tp->rcv_nxt and tp->irs 402 * if seg contains ack then advance tp->snd_una 403 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 404 * arrange for segment to be acked (eventually) 405 * continue processing rest of data/controls, beginning with URG 406 */ 407 case TCPS_SYN_SENT: 408 if ((tiflags & TH_ACK) && 409 (SEQ_LEQ(ti->ti_ack, tp->iss) || 410 SEQ_GT(ti->ti_ack, tp->snd_max))) 411 goto dropwithreset; 412 if (tiflags & TH_RST) { 413 if (tiflags & TH_ACK) 414 tp = tcp_drop(tp, ECONNREFUSED); 415 goto drop; 416 } 417 if ((tiflags & TH_SYN) == 0) 418 goto drop; 419 tp->snd_una = ti->ti_ack; 420 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 421 tp->snd_nxt = tp->snd_una; 422 tp->t_timer[TCPT_REXMT] = 0; 423 tp->irs = ti->ti_seq; 424 tcp_rcvseqinit(tp); 425 tp->t_flags |= TF_ACKNOW; 426 if (SEQ_GT(tp->snd_una, tp->iss)) { 427 soisconnected(so); 428 tp->t_state = TCPS_ESTABLISHED; 429 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 430 (void) tcp_reass(tp, (struct tcpiphdr *)0); 431 } else 432 tp->t_state = TCPS_SYN_RECEIVED; 433 goto trimthenstep6; 434 435 trimthenstep6: 436 /* 437 * Advance ti->ti_seq to correspond to first data byte. 438 * If data, trim to stay within window, 439 * dropping FIN if necessary. 440 */ 441 ti->ti_seq++; 442 if (ti->ti_len > tp->rcv_wnd) { 443 todrop = ti->ti_len - tp->rcv_wnd; 444 m_adj(m, -todrop); 445 ti->ti_len = tp->rcv_wnd; 446 ti->ti_flags &= ~TH_FIN; 447 } 448 tp->snd_wl1 = ti->ti_seq - 1; 449 goto step6; 450 } 451 452 /* 453 * If data is received on a connection after the 454 * user processes are gone, then RST the other end. 455 */ 456 if ((so->so_state & SS_NOFDREF) && tp->t_state > TCPS_CLOSE_WAIT && 457 ti->ti_len) { 458 tp = tcp_close(tp); 459 goto dropwithreset; 460 } 461 462 /* 463 * States other than LISTEN or SYN_SENT. 464 * First check that at least some bytes of segment are within 465 * receive window. 466 */ 467 if (tp->rcv_wnd == 0) { 468 /* 469 * If window is closed can only take segments at 470 * window edge, and have to drop data and PUSH from 471 * incoming segments. 472 */ 473 if (tp->rcv_nxt != ti->ti_seq) 474 goto dropafterack; 475 if (ti->ti_len > 0) { 476 m_adj(m, ti->ti_len); 477 ti->ti_len = 0; 478 ti->ti_flags &= ~(TH_PUSH|TH_FIN); 479 } 480 } else { 481 /* 482 * If segment begins before rcv_nxt, drop leading 483 * data (and SYN); if nothing left, just ack. 484 */ 485 todrop = tp->rcv_nxt - ti->ti_seq; 486 if (todrop > 0) { 487 if (tiflags & TH_SYN) { 488 tiflags &= ~TH_SYN; 489 ti->ti_flags &= ~TH_SYN; 490 ti->ti_seq++; 491 if (ti->ti_urp > 1) 492 ti->ti_urp--; 493 else 494 tiflags &= ~TH_URG; 495 todrop--; 496 } 497 if (todrop > ti->ti_len || 498 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) 499 goto dropafterack; 500 m_adj(m, todrop); 501 ti->ti_seq += todrop; 502 ti->ti_len -= todrop; 503 if (ti->ti_urp > todrop) 504 ti->ti_urp -= todrop; 505 else { 506 tiflags &= ~TH_URG; 507 ti->ti_flags &= ~TH_URG; 508 ti->ti_urp = 0; 509 } 510 } 511 /* 512 * If segment ends after window, drop trailing data 513 * (and PUSH and FIN); if nothing left, just ACK. 514 */ 515 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 516 if (todrop > 0) { 517 if (todrop >= ti->ti_len) 518 goto dropafterack; 519 m_adj(m, -todrop); 520 ti->ti_len -= todrop; 521 ti->ti_flags &= ~(TH_PUSH|TH_FIN); 522 } 523 } 524 525 /* 526 * If the RST bit is set examine the state: 527 * SYN_RECEIVED STATE: 528 * If passive open, return to LISTEN state. 529 * If active open, inform user that connection was refused. 530 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 531 * Inform user that connection was reset, and close tcb. 532 * CLOSING, LAST_ACK, TIME_WAIT STATES 533 * Close the tcb. 534 */ 535 if (tiflags&TH_RST) switch (tp->t_state) { 536 537 case TCPS_SYN_RECEIVED: 538 tp = tcp_drop(tp, ECONNREFUSED); 539 goto drop; 540 541 case TCPS_ESTABLISHED: 542 case TCPS_FIN_WAIT_1: 543 case TCPS_FIN_WAIT_2: 544 case TCPS_CLOSE_WAIT: 545 tp = tcp_drop(tp, ECONNRESET); 546 goto drop; 547 548 case TCPS_CLOSING: 549 case TCPS_LAST_ACK: 550 case TCPS_TIME_WAIT: 551 tp = tcp_close(tp); 552 goto drop; 553 } 554 555 /* 556 * If a SYN is in the window, then this is an 557 * error and we send an RST and drop the connection. 558 */ 559 if (tiflags & TH_SYN) { 560 tp = tcp_drop(tp, ECONNRESET); 561 goto dropwithreset; 562 } 563 564 /* 565 * If the ACK bit is off we drop the segment and return. 566 */ 567 if ((tiflags & TH_ACK) == 0) 568 goto drop; 569 570 /* 571 * Ack processing. 572 */ 573 switch (tp->t_state) { 574 575 /* 576 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 577 * ESTABLISHED state and continue processing, othewise 578 * send an RST. 579 */ 580 case TCPS_SYN_RECEIVED: 581 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 582 SEQ_GT(ti->ti_ack, tp->snd_max)) 583 goto dropwithreset; 584 tp->snd_una++; /* SYN acked */ 585 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 586 tp->snd_nxt = tp->snd_una; 587 tp->t_timer[TCPT_REXMT] = 0; 588 soisconnected(so); 589 tp->t_state = TCPS_ESTABLISHED; 590 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 591 (void) tcp_reass(tp, (struct tcpiphdr *)0); 592 tp->snd_wl1 = ti->ti_seq - 1; 593 /* fall into ... */ 594 595 /* 596 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 597 * ACKs. If the ack is in the range 598 * tp->snd_una < ti->ti_ack <= tp->snd_max 599 * then advance tp->snd_una to ti->ti_ack and drop 600 * data from the retransmission queue. If this ACK reflects 601 * more up to date window information we update our window information. 602 */ 603 case TCPS_ESTABLISHED: 604 case TCPS_FIN_WAIT_1: 605 case TCPS_FIN_WAIT_2: 606 case TCPS_CLOSE_WAIT: 607 case TCPS_CLOSING: 608 case TCPS_LAST_ACK: 609 case TCPS_TIME_WAIT: 610 #define ourfinisacked (acked > 0) 611 612 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) 613 break; 614 if (SEQ_GT(ti->ti_ack, tp->snd_max)) 615 goto dropafterack; 616 acked = ti->ti_ack - tp->snd_una; 617 618 /* 619 * If transmit timer is running and timed sequence 620 * number was acked, update smoothed round trip time. 621 */ 622 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) { 623 if (tp->t_srtt == 0) 624 tp->t_srtt = tp->t_rtt; 625 else 626 tp->t_srtt = 627 tcp_alpha * tp->t_srtt + 628 (1 - tcp_alpha) * tp->t_rtt; 629 tp->t_rtt = 0; 630 } 631 632 if (ti->ti_ack == tp->snd_max) 633 tp->t_timer[TCPT_REXMT] = 0; 634 else { 635 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 636 tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX); 637 tp->t_rxtshift = 0; 638 } 639 /* 640 * When new data is acked, open the congestion window a bit. 641 */ 642 if (acked > 0) 643 tp->snd_cwnd = MIN(11 * tp->snd_cwnd / 10, 65535); 644 if (acked > so->so_snd.sb_cc) { 645 tp->snd_wnd -= so->so_snd.sb_cc; 646 sbdrop(&so->so_snd, so->so_snd.sb_cc); 647 } else { 648 sbdrop(&so->so_snd, acked); 649 tp->snd_wnd -= acked; 650 acked = 0; 651 } 652 if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel) 653 sowwakeup(so); 654 tp->snd_una = ti->ti_ack; 655 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 656 tp->snd_nxt = tp->snd_una; 657 658 switch (tp->t_state) { 659 660 /* 661 * In FIN_WAIT_1 STATE in addition to the processing 662 * for the ESTABLISHED state if our FIN is now acknowledged 663 * then enter FIN_WAIT_2. 664 */ 665 case TCPS_FIN_WAIT_1: 666 if (ourfinisacked) { 667 /* 668 * If we can't receive any more 669 * data, then closing user can proceed. 670 * Starting the timer is contrary to the 671 * specification, but if we don't get a FIN 672 * we'll hang forever. 673 */ 674 if (so->so_state & SS_CANTRCVMORE) { 675 soisdisconnected(so); 676 tp->t_timer[TCPT_2MSL] = TCPTV_MAXIDLE; 677 } 678 tp->t_state = TCPS_FIN_WAIT_2; 679 } 680 break; 681 682 /* 683 * In CLOSING STATE in addition to the processing for 684 * the ESTABLISHED state if the ACK acknowledges our FIN 685 * then enter the TIME-WAIT state, otherwise ignore 686 * the segment. 687 */ 688 case TCPS_CLOSING: 689 if (ourfinisacked) { 690 tp->t_state = TCPS_TIME_WAIT; 691 tcp_canceltimers(tp); 692 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 693 soisdisconnected(so); 694 } 695 break; 696 697 /* 698 * The only thing that can arrive in LAST_ACK state 699 * is an acknowledgment of our FIN. If our FIN is now 700 * acknowledged, delete the TCB, enter the closed state 701 * and return. 702 */ 703 case TCPS_LAST_ACK: 704 if (ourfinisacked) 705 tp = tcp_close(tp); 706 goto drop; 707 708 /* 709 * In TIME_WAIT state the only thing that should arrive 710 * is a retransmission of the remote FIN. Acknowledge 711 * it and restart the finack timer. 712 */ 713 case TCPS_TIME_WAIT: 714 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 715 goto dropafterack; 716 } 717 #undef ourfinisacked 718 } 719 720 step6: 721 /* 722 * Update window information. 723 */ 724 if (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 725 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 726 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd)) { 727 tp->snd_wnd = ti->ti_win; 728 tp->snd_wl1 = ti->ti_seq; 729 tp->snd_wl2 = ti->ti_ack; 730 } 731 732 /* 733 * Process segments with URG. 734 */ 735 if ((tiflags & TH_URG) && ti->ti_urp && 736 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 737 /* 738 * This is a kludge, but if we receive accept 739 * random urgent pointers, we'll crash in 740 * soreceive. It's hard to imagine someone 741 * actually wanting to send this much urgent data. 742 */ 743 if (ti->ti_urp + (unsigned) so->so_rcv.sb_cc > 32767) { 744 ti->ti_urp = 0; /* XXX */ 745 tiflags &= ~TH_URG; /* XXX */ 746 ti->ti_flags &= ~TH_URG; /* XXX */ 747 goto badurp; /* XXX */ 748 } 749 /* 750 * If this segment advances the known urgent pointer, 751 * then mark the data stream. This should not happen 752 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 753 * a FIN has been received from the remote side. 754 * In these states we ignore the URG. 755 */ 756 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 757 tp->rcv_up = ti->ti_seq + ti->ti_urp; 758 so->so_oobmark = so->so_rcv.sb_cc + 759 (tp->rcv_up - tp->rcv_nxt) - 1; 760 if (so->so_oobmark == 0) 761 so->so_state |= SS_RCVATMARK; 762 sohasoutofband(so); 763 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 764 } 765 /* 766 * Remove out of band data so doesn't get presented to user. 767 * This can happen independent of advancing the URG pointer, 768 * but if two URG's are pending at once, some out-of-band 769 * data may creep in... ick. 770 */ 771 if (ti->ti_urp <= ti->ti_len) 772 tcp_pulloutofband(so, ti); 773 } 774 badurp: /* XXX */ 775 776 /* 777 * Process the segment text, merging it into the TCP sequencing queue, 778 * and arranging for acknowledgment of receipt if necessary. 779 * This process logically involves adjusting tp->rcv_wnd as data 780 * is presented to the user (this happens in tcp_usrreq.c, 781 * case PRU_RCVD). If a FIN has already been received on this 782 * connection then we just ignore the text. 783 */ 784 if ((ti->ti_len || (tiflags&TH_FIN)) && 785 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 786 TCP_REASS(tp, ti, m, so, tiflags); 787 if (tcpnodelack == 0) 788 tp->t_flags |= TF_DELACK; 789 else 790 tp->t_flags |= TF_ACKNOW; 791 } else { 792 m_freem(m); 793 tiflags &= ~TH_FIN; 794 } 795 796 /* 797 * If FIN is received ACK the FIN and let the user know 798 * that the connection is closing. 799 */ 800 if (tiflags & TH_FIN) { 801 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 802 socantrcvmore(so); 803 tp->t_flags |= TF_ACKNOW; 804 tp->rcv_nxt++; 805 } 806 switch (tp->t_state) { 807 808 /* 809 * In SYN_RECEIVED and ESTABLISHED STATES 810 * enter the CLOSE_WAIT state. 811 */ 812 case TCPS_SYN_RECEIVED: 813 case TCPS_ESTABLISHED: 814 tp->t_state = TCPS_CLOSE_WAIT; 815 break; 816 817 /* 818 * If still in FIN_WAIT_1 STATE FIN has not been acked so 819 * enter the CLOSING state. 820 */ 821 case TCPS_FIN_WAIT_1: 822 tp->t_state = TCPS_CLOSING; 823 break; 824 825 /* 826 * In FIN_WAIT_2 state enter the TIME_WAIT state, 827 * starting the time-wait timer, turning off the other 828 * standard timers. 829 */ 830 case TCPS_FIN_WAIT_2: 831 tp->t_state = TCPS_TIME_WAIT; 832 tcp_canceltimers(tp); 833 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 834 soisdisconnected(so); 835 break; 836 837 /* 838 * In TIME_WAIT state restart the 2 MSL time_wait timer. 839 */ 840 case TCPS_TIME_WAIT: 841 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 842 break; 843 } 844 } 845 if (so->so_options & SO_DEBUG) 846 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 847 848 /* 849 * Return any desired output. 850 */ 851 (void) tcp_output(tp); 852 return; 853 854 dropafterack: 855 /* 856 * Generate an ACK dropping incoming segment if it occupies 857 * sequence space, where the ACK reflects our state. 858 */ 859 if ((tiflags&TH_RST) || 860 tlen == 0 && (tiflags&(TH_SYN|TH_FIN)) == 0) 861 goto drop; 862 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 863 tcp_trace(TA_RESPOND, ostate, tp, &tcp_saveti, 0); 864 tcp_respond(tp, ti, tp->rcv_nxt, tp->snd_nxt, TH_ACK); 865 return; 866 867 dropwithreset: 868 if (om) { 869 (void) m_free(om); 870 om = 0; 871 } 872 /* 873 * Generate a RST, dropping incoming segment. 874 * Make ACK acceptable to originator of segment. 875 */ 876 if (tiflags & TH_RST) 877 goto drop; 878 if (tiflags & TH_ACK) 879 tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST); 880 else { 881 if (tiflags & TH_SYN) 882 ti->ti_len++; 883 tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, 884 TH_RST|TH_ACK); 885 } 886 /* destroy temporarily created socket */ 887 if (dropsocket) 888 (void) soabort(so); 889 return; 890 891 drop: 892 if (om) 893 (void) m_free(om); 894 /* 895 * Drop space held by incoming segment and return. 896 */ 897 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 898 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 899 m_freem(m); 900 /* destroy temporarily created socket */ 901 if (dropsocket) 902 (void) soabort(so); 903 return; 904 } 905 906 tcp_dooptions(tp, om, ti) 907 struct tcpcb *tp; 908 struct mbuf *om; 909 struct tcpiphdr *ti; 910 { 911 register u_char *cp; 912 int opt, optlen, cnt; 913 914 cp = mtod(om, u_char *); 915 cnt = om->m_len; 916 for (; cnt > 0; cnt -= optlen, cp += optlen) { 917 opt = cp[0]; 918 if (opt == TCPOPT_EOL) 919 break; 920 if (opt == TCPOPT_NOP) 921 optlen = 1; 922 else { 923 optlen = cp[1]; 924 if (optlen <= 0) 925 break; 926 } 927 switch (opt) { 928 929 default: 930 break; 931 932 case TCPOPT_MAXSEG: 933 if (optlen != 4) 934 continue; 935 if (!(ti->ti_flags & TH_SYN)) 936 continue; 937 tp->t_maxseg = *(u_short *)(cp + 2); 938 tp->t_maxseg = ntohs((u_short)tp->t_maxseg); 939 tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp)); 940 break; 941 } 942 } 943 (void) m_free(om); 944 } 945 946 /* 947 * Pull out of band byte out of a segment so 948 * it doesn't appear in the user's data queue. 949 * It is still reflected in the segment length for 950 * sequencing purposes. 951 */ 952 tcp_pulloutofband(so, ti) 953 struct socket *so; 954 struct tcpiphdr *ti; 955 { 956 register struct mbuf *m; 957 int cnt = ti->ti_urp - 1; 958 959 m = dtom(ti); 960 while (cnt >= 0) { 961 if (m->m_len > cnt) { 962 char *cp = mtod(m, caddr_t) + cnt; 963 struct tcpcb *tp = sototcpcb(so); 964 965 tp->t_iobc = *cp; 966 tp->t_oobflags |= TCPOOB_HAVEDATA; 967 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 968 m->m_len--; 969 return; 970 } 971 cnt -= m->m_len; 972 m = m->m_next; 973 if (m == 0) 974 break; 975 } 976 panic("tcp_pulloutofband"); 977 } 978 979 /* 980 * Determine a reasonable value for maxseg size. 981 * If the route is known, use one that can be handled 982 * on the given interface without forcing IP to fragment. 983 * If bigger than a page (CLBYTES), round down to nearest pagesize 984 * to utilize pagesize mbufs. 985 * If interface pointer is unavailable, or the destination isn't local, 986 * use a conservative size (512 or the default IP max size, but no more 987 * than the mtu of the interface through which we route), 988 * as we can't discover anything about intervening gateways or networks. 989 * 990 * This is ugly, and doesn't belong at this level, but has to happen somehow. 991 */ 992 tcp_mss(tp) 993 register struct tcpcb *tp; 994 { 995 struct route *ro; 996 struct ifnet *ifp; 997 int mss; 998 struct inpcb *inp; 999 1000 inp = tp->t_inpcb; 1001 ro = &inp->inp_route; 1002 if ((ro->ro_rt == (struct rtentry *)0) || 1003 (ifp = ro->ro_rt->rt_ifp) == (struct ifnet *)0) { 1004 /* No route yet, so try to acquire one */ 1005 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1006 ro->ro_dst.sa_family = AF_INET; 1007 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1008 inp->inp_faddr; 1009 rtalloc(ro); 1010 } 1011 if ((ro->ro_rt == 0) || (ifp = ro->ro_rt->rt_ifp) == 0) 1012 return (TCP_MSS); 1013 } 1014 1015 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1016 #if (CLBYTES & (CLBYTES - 1)) == 0 1017 if (mss > CLBYTES) 1018 mss &= ~(CLBYTES-1); 1019 #else 1020 if (mss > CLBYTES) 1021 mss = mss / CLBYTES * CLBYTES; 1022 #endif 1023 if (in_localaddr(inp->inp_faddr)) 1024 return (mss); 1025 return (MIN(mss, TCP_MSS)); 1026 } 1027