1 /* 2 * Copyright (c) 1982, 1986, 1988 Regents of the University of California. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms are permitted 6 * provided that the above copyright notice and this paragraph are 7 * duplicated in all such forms and that any documentation, 8 * advertising materials, and other materials related to such 9 * distribution and use acknowledge that the software was developed 10 * by the University of California, Berkeley. The name of the 11 * University may not be used to endorse or promote products derived 12 * from this software without specific prior written permission. 13 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 14 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 15 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. 16 * 17 * @(#)tcp_input.c 7.21 (Berkeley) 04/08/89 18 */ 19 20 #include "param.h" 21 #include "systm.h" 22 #include "malloc.h" 23 #include "mbuf.h" 24 #include "protosw.h" 25 #include "socket.h" 26 #include "socketvar.h" 27 #include "errno.h" 28 29 #include "../net/if.h" 30 #include "../net/route.h" 31 32 #include "in.h" 33 #include "in_pcb.h" 34 #include "in_systm.h" 35 #include "ip.h" 36 #include "ip_var.h" 37 #include "tcp.h" 38 #include "tcp_fsm.h" 39 #include "tcp_seq.h" 40 #include "tcp_timer.h" 41 #include "tcp_var.h" 42 #include "tcpip.h" 43 #include "tcp_debug.h" 44 45 int tcpprintfs = 0; 46 int tcprexmtthresh = 3; 47 struct tcpiphdr tcp_saveti; 48 49 struct tcpcb *tcp_newtcpcb(); 50 51 /* 52 * Insert segment ti into reassembly queue of tcp with 53 * control block tp. Return TH_FIN if reassembly now includes 54 * a segment with FIN. The macro form does the common case inline 55 * (segment is the next to be received on an established connection, 56 * and the queue is empty), avoiding linkage into and removal 57 * from the queue and repetition of various conversions. 58 * Set DELACK for segments received in order, but ack immediately 59 * when segments are out of order (so fast retransmit can work). 60 */ 61 #define TCP_REASS(tp, ti, m, so, flags) { \ 62 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 63 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 64 (tp)->t_state == TCPS_ESTABLISHED) { \ 65 tp->t_flags |= TF_DELACK; \ 66 (tp)->rcv_nxt += (ti)->ti_len; \ 67 flags = (ti)->ti_flags & TH_FIN; \ 68 tcpstat.tcps_rcvpack++;\ 69 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 70 sbappend(&(so)->so_rcv, (m)); \ 71 sorwakeup(so); \ 72 } else { \ 73 (flags) = tcp_reass((tp), (ti)); \ 74 tp->t_flags |= TF_ACKNOW; \ 75 } \ 76 } 77 78 tcp_reass(tp, ti) 79 register struct tcpcb *tp; 80 register struct tcpiphdr *ti; 81 { 82 register struct tcpiphdr *q; 83 struct socket *so = tp->t_inpcb->inp_socket; 84 struct mbuf *m; 85 int flags; 86 87 /* 88 * Call with ti==0 after become established to 89 * force pre-ESTABLISHED data up to user socket. 90 */ 91 if (ti == 0) 92 goto present; 93 94 /* 95 * Find a segment which begins after this one does. 96 */ 97 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 98 q = (struct tcpiphdr *)q->ti_next) 99 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 100 break; 101 102 /* 103 * If there is a preceding segment, it may provide some of 104 * our data already. If so, drop the data from the incoming 105 * segment. If it provides all of our data, drop us. 106 */ 107 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 108 register int i; 109 q = (struct tcpiphdr *)q->ti_prev; 110 /* conversion to int (in i) handles seq wraparound */ 111 i = q->ti_seq + q->ti_len - ti->ti_seq; 112 if (i > 0) { 113 if (i >= ti->ti_len) { 114 tcpstat.tcps_rcvduppack++; 115 tcpstat.tcps_rcvdupbyte += ti->ti_len; 116 goto drop; 117 } 118 m_adj(dtom(ti), i); 119 ti->ti_len -= i; 120 ti->ti_seq += i; 121 } 122 q = (struct tcpiphdr *)(q->ti_next); 123 } 124 tcpstat.tcps_rcvoopack++; 125 tcpstat.tcps_rcvoobyte += ti->ti_len; 126 127 /* 128 * While we overlap succeeding segments trim them or, 129 * if they are completely covered, dequeue them. 130 */ 131 while (q != (struct tcpiphdr *)tp) { 132 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 133 if (i <= 0) 134 break; 135 if (i < q->ti_len) { 136 q->ti_seq += i; 137 q->ti_len -= i; 138 m_adj(dtom(q), i); 139 break; 140 } 141 q = (struct tcpiphdr *)q->ti_next; 142 m = dtom(q->ti_prev); 143 remque(q->ti_prev); 144 m_freem(m); 145 } 146 147 /* 148 * Stick new segment in its place. 149 */ 150 insque(ti, q->ti_prev); 151 152 present: 153 /* 154 * Present data to user, advancing rcv_nxt through 155 * completed sequence space. 156 */ 157 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 158 return (0); 159 ti = tp->seg_next; 160 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 161 return (0); 162 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 163 return (0); 164 do { 165 tp->rcv_nxt += ti->ti_len; 166 flags = ti->ti_flags & TH_FIN; 167 remque(ti); 168 m = dtom(ti); 169 ti = (struct tcpiphdr *)ti->ti_next; 170 if (so->so_state & SS_CANTRCVMORE) 171 m_freem(m); 172 else 173 sbappend(&so->so_rcv, m); 174 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 175 sorwakeup(so); 176 return (flags); 177 drop: 178 m_freem(dtom(ti)); 179 return (0); 180 } 181 182 /* 183 * TCP input routine, follows pages 65-76 of the 184 * protocol specification dated September, 1981 very closely. 185 */ 186 tcp_input(m, iphlen) 187 register struct mbuf *m; 188 int iphlen; 189 { 190 register struct tcpiphdr *ti; 191 struct inpcb *inp; 192 struct mbuf *om = 0; 193 int len, tlen, off; 194 register struct tcpcb *tp = 0; 195 register int tiflags; 196 struct socket *so; 197 int todrop, acked, ourfinisacked, needoutput = 0; 198 short ostate; 199 struct in_addr laddr; 200 int dropsocket = 0; 201 int iss = 0; 202 203 tcpstat.tcps_rcvtotal++; 204 /* 205 * Get IP and TCP header together in first mbuf. 206 * Note: IP leaves IP header in first mbuf. 207 */ 208 ti = mtod(m, struct tcpiphdr *); 209 if (iphlen > sizeof (struct ip)) 210 ip_stripoptions(m, (struct mbuf *)0); 211 if (m->m_flags & M_EXT || m->m_len < sizeof (struct tcpiphdr)) { 212 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 213 tcpstat.tcps_rcvshort++; 214 return; 215 } 216 ti = mtod(m, struct tcpiphdr *); 217 } 218 219 /* 220 * Checksum extended TCP header and data. 221 */ 222 tlen = ((struct ip *)ti)->ip_len; 223 len = sizeof (struct ip) + tlen; 224 ti->ti_next = ti->ti_prev = 0; 225 ti->ti_x1 = 0; 226 ti->ti_len = (u_short)tlen; 227 ti->ti_len = htons((u_short)ti->ti_len); 228 if (ti->ti_sum = in_cksum(m, len)) { 229 if (tcpprintfs) 230 printf("tcp sum: src %x\n", ti->ti_src); 231 tcpstat.tcps_rcvbadsum++; 232 goto drop; 233 } 234 235 /* 236 * Check that TCP offset makes sense, 237 * pull out TCP options and adjust length. 238 */ 239 off = ti->ti_off << 2; 240 if (off < sizeof (struct tcphdr) || off > tlen) { 241 if (tcpprintfs) 242 printf("tcp off: src %x off %d\n", ti->ti_src, off); 243 tcpstat.tcps_rcvbadoff++; 244 goto drop; 245 } 246 tlen -= off; 247 ti->ti_len = tlen; 248 if (off > sizeof (struct tcphdr)) { 249 if (m->m_len < sizeof(struct ip) + off) { 250 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 251 tcpstat.tcps_rcvshort++; 252 return; 253 } 254 ti = mtod(m, struct tcpiphdr *); 255 } 256 om = m_get(M_DONTWAIT, MT_DATA); 257 if (om == 0) 258 goto drop; 259 om->m_len = off - sizeof (struct tcphdr); 260 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 261 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 262 m->m_len -= om->m_len; 263 m->m_pkthdr.len -= om->m_len; 264 bcopy(op+om->m_len, op, 265 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 266 } 267 } 268 tiflags = ti->ti_flags; 269 270 /* 271 * Drop TCP and IP headers; TCP options were dropped above. 272 */ 273 m->m_data += sizeof(struct tcpiphdr); 274 m->m_len -= sizeof(struct tcpiphdr); 275 m->m_pkthdr.len -= sizeof(struct tcpiphdr); 276 277 /* 278 * Convert TCP protocol specific fields to host format. 279 */ 280 ti->ti_seq = ntohl(ti->ti_seq); 281 ti->ti_ack = ntohl(ti->ti_ack); 282 ti->ti_win = ntohs(ti->ti_win); 283 ti->ti_urp = ntohs(ti->ti_urp); 284 285 /* 286 * Locate pcb for segment. 287 */ 288 findpcb: 289 inp = in_pcblookup 290 (&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport, 291 INPLOOKUP_WILDCARD); 292 293 /* 294 * If the state is CLOSED (i.e., TCB does not exist) then 295 * all data in the incoming segment is discarded. 296 * If the TCB exists but is in CLOSED state, it is embryonic, 297 * but should either do a listen or a connect soon. 298 */ 299 if (inp == 0) 300 goto dropwithreset; 301 tp = intotcpcb(inp); 302 if (tp == 0) 303 goto dropwithreset; 304 if (tp->t_state == TCPS_CLOSED) 305 goto drop; 306 so = inp->inp_socket; 307 if (so->so_options & SO_DEBUG) { 308 ostate = tp->t_state; 309 tcp_saveti = *ti; 310 } 311 if (so->so_options & SO_ACCEPTCONN) { 312 so = sonewconn(so); 313 if (so == 0) 314 goto drop; 315 /* 316 * This is ugly, but .... 317 * 318 * Mark socket as temporary until we're 319 * committed to keeping it. The code at 320 * ``drop'' and ``dropwithreset'' check the 321 * flag dropsocket to see if the temporary 322 * socket created here should be discarded. 323 * We mark the socket as discardable until 324 * we're committed to it below in TCPS_LISTEN. 325 */ 326 dropsocket++; 327 inp = (struct inpcb *)so->so_pcb; 328 inp->inp_laddr = ti->ti_dst; 329 inp->inp_lport = ti->ti_dport; 330 inp->inp_options = ip_srcroute(); 331 tp = intotcpcb(inp); 332 tp->t_state = TCPS_LISTEN; 333 } 334 335 /* 336 * Segment received on connection. 337 * Reset idle time and keep-alive timer. 338 */ 339 tp->t_idle = 0; 340 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 341 342 /* 343 * Process options if not in LISTEN state, 344 * else do it below (after getting remote address). 345 */ 346 if (om && tp->t_state != TCPS_LISTEN) { 347 tcp_dooptions(tp, om, ti); 348 om = 0; 349 } 350 351 /* 352 * Calculate amount of space in receive window, 353 * and then do TCP input processing. 354 * Receive window is amount of space in rcv queue, 355 * but not less than advertised window. 356 */ 357 { int win; 358 359 win = sbspace(&so->so_rcv); 360 if (win < 0) 361 win = 0; 362 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 363 } 364 365 switch (tp->t_state) { 366 367 /* 368 * If the state is LISTEN then ignore segment if it contains an RST. 369 * If the segment contains an ACK then it is bad and send a RST. 370 * If it does not contain a SYN then it is not interesting; drop it. 371 * Don't bother responding if the destination was a broadcast. 372 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 373 * tp->iss, and send a segment: 374 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 375 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 376 * Fill in remote peer address fields if not previously specified. 377 * Enter SYN_RECEIVED state, and process any other fields of this 378 * segment in this state. 379 */ 380 case TCPS_LISTEN: { 381 struct mbuf *am; 382 register struct sockaddr_in *sin; 383 384 if (tiflags & TH_RST) 385 goto drop; 386 if (tiflags & TH_ACK) 387 goto dropwithreset; 388 if ((tiflags & TH_SYN) == 0) 389 goto drop; 390 if (m->m_flags & M_BCAST) 391 goto drop; 392 am = m_get(M_DONTWAIT, MT_SONAME); 393 if (am == NULL) 394 goto drop; 395 am->m_len = sizeof (struct sockaddr_in); 396 sin = mtod(am, struct sockaddr_in *); 397 sin->sin_family = AF_INET; 398 sin->sin_len = sizeof(*sin); 399 sin->sin_addr = ti->ti_src; 400 sin->sin_port = ti->ti_sport; 401 laddr = inp->inp_laddr; 402 if (inp->inp_laddr.s_addr == INADDR_ANY) 403 inp->inp_laddr = ti->ti_dst; 404 if (in_pcbconnect(inp, am)) { 405 inp->inp_laddr = laddr; 406 (void) m_free(am); 407 goto drop; 408 } 409 (void) m_free(am); 410 tp->t_template = tcp_template(tp); 411 if (tp->t_template == 0) { 412 tp = tcp_drop(tp, ENOBUFS); 413 dropsocket = 0; /* socket is already gone */ 414 goto drop; 415 } 416 if (om) { 417 tcp_dooptions(tp, om, ti); 418 om = 0; 419 } 420 if (iss) 421 tp->iss = iss; 422 else 423 tp->iss = tcp_iss; 424 tcp_iss += TCP_ISSINCR/2; 425 tp->irs = ti->ti_seq; 426 tcp_sendseqinit(tp); 427 tcp_rcvseqinit(tp); 428 tp->t_flags |= TF_ACKNOW; 429 tp->t_state = TCPS_SYN_RECEIVED; 430 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 431 dropsocket = 0; /* committed to socket */ 432 tcpstat.tcps_accepts++; 433 goto trimthenstep6; 434 } 435 436 /* 437 * If the state is SYN_SENT: 438 * if seg contains an ACK, but not for our SYN, drop the input. 439 * if seg contains a RST, then drop the connection. 440 * if seg does not contain SYN, then drop it. 441 * Otherwise this is an acceptable SYN segment 442 * initialize tp->rcv_nxt and tp->irs 443 * if seg contains ack then advance tp->snd_una 444 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 445 * arrange for segment to be acked (eventually) 446 * continue processing rest of data/controls, beginning with URG 447 */ 448 case TCPS_SYN_SENT: 449 if ((tiflags & TH_ACK) && 450 (SEQ_LEQ(ti->ti_ack, tp->iss) || 451 SEQ_GT(ti->ti_ack, tp->snd_max))) 452 goto dropwithreset; 453 if (tiflags & TH_RST) { 454 if (tiflags & TH_ACK) 455 tp = tcp_drop(tp, ECONNREFUSED); 456 goto drop; 457 } 458 if ((tiflags & TH_SYN) == 0) 459 goto drop; 460 if (tiflags & TH_ACK) { 461 tp->snd_una = ti->ti_ack; 462 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 463 tp->snd_nxt = tp->snd_una; 464 } 465 tp->t_timer[TCPT_REXMT] = 0; 466 tp->irs = ti->ti_seq; 467 tcp_rcvseqinit(tp); 468 tp->t_flags |= TF_ACKNOW; 469 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 470 tcpstat.tcps_connects++; 471 soisconnected(so); 472 tp->t_state = TCPS_ESTABLISHED; 473 tp->t_maxseg = min(tp->t_maxseg, tcp_mss(tp)); 474 (void) tcp_reass(tp, (struct tcpiphdr *)0); 475 /* 476 * if we didn't have to retransmit the SYN, 477 * use its rtt as our initial srtt & rtt var. 478 */ 479 if (tp->t_rtt) { 480 tp->t_srtt = tp->t_rtt << 3; 481 tp->t_rttvar = tp->t_rtt << 1; 482 TCPT_RANGESET(tp->t_rxtcur, 483 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 484 TCPTV_MIN, TCPTV_REXMTMAX); 485 tp->t_rtt = 0; 486 } 487 } else 488 tp->t_state = TCPS_SYN_RECEIVED; 489 490 trimthenstep6: 491 /* 492 * Advance ti->ti_seq to correspond to first data byte. 493 * If data, trim to stay within window, 494 * dropping FIN if necessary. 495 */ 496 ti->ti_seq++; 497 if (ti->ti_len > tp->rcv_wnd) { 498 todrop = ti->ti_len - tp->rcv_wnd; 499 m_adj(m, -todrop); 500 ti->ti_len = tp->rcv_wnd; 501 tiflags &= ~TH_FIN; 502 tcpstat.tcps_rcvpackafterwin++; 503 tcpstat.tcps_rcvbyteafterwin += todrop; 504 } 505 tp->snd_wl1 = ti->ti_seq - 1; 506 tp->rcv_up = ti->ti_seq; 507 goto step6; 508 } 509 510 /* 511 * States other than LISTEN or SYN_SENT. 512 * First check that at least some bytes of segment are within 513 * receive window. If segment begins before rcv_nxt, 514 * drop leading data (and SYN); if nothing left, just ack. 515 */ 516 todrop = tp->rcv_nxt - ti->ti_seq; 517 if (todrop > 0) { 518 if (tiflags & TH_SYN) { 519 tiflags &= ~TH_SYN; 520 ti->ti_seq++; 521 if (ti->ti_urp > 1) 522 ti->ti_urp--; 523 else 524 tiflags &= ~TH_URG; 525 todrop--; 526 } 527 if (todrop > ti->ti_len || 528 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 529 tcpstat.tcps_rcvduppack++; 530 tcpstat.tcps_rcvdupbyte += ti->ti_len; 531 /* 532 * If segment is just one to the left of the window, 533 * check two special cases: 534 * 1. Don't toss RST in response to 4.2-style keepalive. 535 * 2. If the only thing to drop is a FIN, we can drop 536 * it, but check the ACK or we will get into FIN 537 * wars if our FINs crossed (both CLOSING). 538 * In either case, send ACK to resynchronize, 539 * but keep on processing for RST or ACK. 540 */ 541 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 542 #ifdef TCP_COMPAT_42 543 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 544 #endif 545 ) { 546 todrop = ti->ti_len; 547 tiflags &= ~TH_FIN; 548 tp->t_flags |= TF_ACKNOW; 549 } else 550 goto dropafterack; 551 } else { 552 tcpstat.tcps_rcvpartduppack++; 553 tcpstat.tcps_rcvpartdupbyte += todrop; 554 } 555 m_adj(m, todrop); 556 ti->ti_seq += todrop; 557 ti->ti_len -= todrop; 558 if (ti->ti_urp > todrop) 559 ti->ti_urp -= todrop; 560 else { 561 tiflags &= ~TH_URG; 562 ti->ti_urp = 0; 563 } 564 } 565 566 /* 567 * If new data are received on a connection after the 568 * user processes are gone, then RST the other end. 569 */ 570 if ((so->so_state & SS_NOFDREF) && 571 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 572 tp = tcp_close(tp); 573 tcpstat.tcps_rcvafterclose++; 574 goto dropwithreset; 575 } 576 577 /* 578 * If segment ends after window, drop trailing data 579 * (and PUSH and FIN); if nothing left, just ACK. 580 */ 581 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 582 if (todrop > 0) { 583 tcpstat.tcps_rcvpackafterwin++; 584 if (todrop >= ti->ti_len) { 585 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 586 /* 587 * If a new connection request is received 588 * while in TIME_WAIT, drop the old connection 589 * and start over if the sequence numbers 590 * are above the previous ones. 591 */ 592 if (tiflags & TH_SYN && 593 tp->t_state == TCPS_TIME_WAIT && 594 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 595 iss = tp->rcv_nxt + TCP_ISSINCR; 596 (void) tcp_close(tp); 597 goto findpcb; 598 } 599 /* 600 * If window is closed can only take segments at 601 * window edge, and have to drop data and PUSH from 602 * incoming segments. Continue processing, but 603 * remember to ack. Otherwise, drop segment 604 * and ack. 605 */ 606 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 607 tp->t_flags |= TF_ACKNOW; 608 tcpstat.tcps_rcvwinprobe++; 609 } else 610 goto dropafterack; 611 } else 612 tcpstat.tcps_rcvbyteafterwin += todrop; 613 m_adj(m, -todrop); 614 ti->ti_len -= todrop; 615 tiflags &= ~(TH_PUSH|TH_FIN); 616 } 617 618 /* 619 * If the RST bit is set examine the state: 620 * SYN_RECEIVED STATE: 621 * If passive open, return to LISTEN state. 622 * If active open, inform user that connection was refused. 623 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 624 * Inform user that connection was reset, and close tcb. 625 * CLOSING, LAST_ACK, TIME_WAIT STATES 626 * Close the tcb. 627 */ 628 if (tiflags&TH_RST) switch (tp->t_state) { 629 630 case TCPS_SYN_RECEIVED: 631 so->so_error = ECONNREFUSED; 632 goto close; 633 634 case TCPS_ESTABLISHED: 635 case TCPS_FIN_WAIT_1: 636 case TCPS_FIN_WAIT_2: 637 case TCPS_CLOSE_WAIT: 638 so->so_error = ECONNRESET; 639 close: 640 tp->t_state = TCPS_CLOSED; 641 tcpstat.tcps_drops++; 642 tp = tcp_close(tp); 643 goto drop; 644 645 case TCPS_CLOSING: 646 case TCPS_LAST_ACK: 647 case TCPS_TIME_WAIT: 648 tp = tcp_close(tp); 649 goto drop; 650 } 651 652 /* 653 * If a SYN is in the window, then this is an 654 * error and we send an RST and drop the connection. 655 */ 656 if (tiflags & TH_SYN) { 657 tp = tcp_drop(tp, ECONNRESET); 658 goto dropwithreset; 659 } 660 661 /* 662 * If the ACK bit is off we drop the segment and return. 663 */ 664 if ((tiflags & TH_ACK) == 0) 665 goto drop; 666 667 /* 668 * Ack processing. 669 */ 670 switch (tp->t_state) { 671 672 /* 673 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 674 * ESTABLISHED state and continue processing, otherwise 675 * send an RST. 676 */ 677 case TCPS_SYN_RECEIVED: 678 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 679 SEQ_GT(ti->ti_ack, tp->snd_max)) 680 goto dropwithreset; 681 tcpstat.tcps_connects++; 682 soisconnected(so); 683 tp->t_state = TCPS_ESTABLISHED; 684 tp->t_maxseg = min(tp->t_maxseg, tcp_mss(tp)); 685 (void) tcp_reass(tp, (struct tcpiphdr *)0); 686 tp->snd_wl1 = ti->ti_seq - 1; 687 /* fall into ... */ 688 689 /* 690 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 691 * ACKs. If the ack is in the range 692 * tp->snd_una < ti->ti_ack <= tp->snd_max 693 * then advance tp->snd_una to ti->ti_ack and drop 694 * data from the retransmission queue. If this ACK reflects 695 * more up to date window information we update our window information. 696 */ 697 case TCPS_ESTABLISHED: 698 case TCPS_FIN_WAIT_1: 699 case TCPS_FIN_WAIT_2: 700 case TCPS_CLOSE_WAIT: 701 case TCPS_CLOSING: 702 case TCPS_LAST_ACK: 703 case TCPS_TIME_WAIT: 704 705 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 706 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 707 tcpstat.tcps_rcvdupack++; 708 /* 709 * If we have outstanding data (not a 710 * window probe), this is a completely 711 * duplicate ack (ie, window info didn't 712 * change), the ack is the biggest we've 713 * seen and we've seen exactly our rexmt 714 * threshhold of them, assume a packet 715 * has been dropped and retransmit it. 716 * Kludge snd_nxt & the congestion 717 * window so we send only this one 718 * packet. If this packet fills the 719 * only hole in the receiver's seq. 720 * space, the next real ack will fully 721 * open our window. This means we 722 * have to do the usual slow-start to 723 * not overwhelm an intermediate gateway 724 * with a burst of packets. Leave 725 * here with the congestion window set 726 * to allow 2 packets on the next real 727 * ack and the exp-to-linear thresh 728 * set for half the current window 729 * size (since we know we're losing at 730 * the current window size). 731 */ 732 if (tp->t_timer[TCPT_REXMT] == 0 || 733 ti->ti_ack != tp->snd_una) 734 tp->t_dupacks = 0; 735 else if (++tp->t_dupacks == tcprexmtthresh) { 736 tcp_seq onxt = tp->snd_nxt; 737 u_int win = 738 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 739 tp->t_maxseg; 740 741 if (win < 2) 742 win = 2; 743 tp->snd_ssthresh = win * tp->t_maxseg; 744 745 tp->t_timer[TCPT_REXMT] = 0; 746 tp->t_rtt = 0; 747 tp->snd_nxt = ti->ti_ack; 748 tp->snd_cwnd = tp->t_maxseg; 749 (void) tcp_output(tp); 750 751 if (SEQ_GT(onxt, tp->snd_nxt)) 752 tp->snd_nxt = onxt; 753 goto drop; 754 } 755 } else 756 tp->t_dupacks = 0; 757 break; 758 } 759 tp->t_dupacks = 0; 760 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 761 tcpstat.tcps_rcvacktoomuch++; 762 goto dropafterack; 763 } 764 acked = ti->ti_ack - tp->snd_una; 765 tcpstat.tcps_rcvackpack++; 766 tcpstat.tcps_rcvackbyte += acked; 767 768 /* 769 * If transmit timer is running and timed sequence 770 * number was acked, update smoothed round trip time. 771 * Since we now have an rtt measurement, cancel the 772 * timer backoff (cf., Phil Karn's retransmit alg.). 773 * Recompute the initial retransmit timer. 774 */ 775 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) { 776 tcpstat.tcps_rttupdated++; 777 if (tp->t_srtt != 0) { 778 register short delta; 779 780 /* 781 * srtt is stored as fixed point with 3 bits 782 * after the binary point (i.e., scaled by 8). 783 * The following magic is equivalent 784 * to the smoothing algorithm in rfc793 785 * with an alpha of .875 786 * (srtt = rtt/8 + srtt*7/8 in fixed point). 787 * Adjust t_rtt to origin 0. 788 */ 789 delta = tp->t_rtt - 1 - (tp->t_srtt >> 3); 790 if ((tp->t_srtt += delta) <= 0) 791 tp->t_srtt = 1; 792 /* 793 * We accumulate a smoothed rtt variance 794 * (actually, a smoothed mean difference), 795 * then set the retransmit timer to smoothed 796 * rtt + 2 times the smoothed variance. 797 * rttvar is stored as fixed point 798 * with 2 bits after the binary point 799 * (scaled by 4). The following is equivalent 800 * to rfc793 smoothing with an alpha of .75 801 * (rttvar = rttvar*3/4 + |delta| / 4). 802 * This replaces rfc793's wired-in beta. 803 */ 804 if (delta < 0) 805 delta = -delta; 806 delta -= (tp->t_rttvar >> 2); 807 if ((tp->t_rttvar += delta) <= 0) 808 tp->t_rttvar = 1; 809 } else { 810 /* 811 * No rtt measurement yet - use the 812 * unsmoothed rtt. Set the variance 813 * to half the rtt (so our first 814 * retransmit happens at 2*rtt) 815 */ 816 tp->t_srtt = tp->t_rtt << 3; 817 tp->t_rttvar = tp->t_rtt << 1; 818 } 819 tp->t_rtt = 0; 820 tp->t_rxtshift = 0; 821 TCPT_RANGESET(tp->t_rxtcur, 822 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 823 TCPTV_MIN, TCPTV_REXMTMAX); 824 } 825 826 /* 827 * If all outstanding data is acked, stop retransmit 828 * timer and remember to restart (more output or persist). 829 * If there is more data to be acked, restart retransmit 830 * timer, using current (possibly backed-off) value. 831 */ 832 if (ti->ti_ack == tp->snd_max) { 833 tp->t_timer[TCPT_REXMT] = 0; 834 needoutput = 1; 835 } else if (tp->t_timer[TCPT_PERSIST] == 0) 836 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 837 /* 838 * When new data is acked, open the congestion window. 839 * If the window gives us less than ssthresh packets 840 * in flight, open exponentially (maxseg per packet). 841 * Otherwise open linearly (maxseg per window, 842 * or maxseg^2 / cwnd per packet). 843 */ 844 { 845 u_int incr = tp->t_maxseg; 846 847 if (tp->snd_cwnd > tp->snd_ssthresh) 848 incr = max(incr * incr / tp->snd_cwnd, 1); 849 850 tp->snd_cwnd = min(tp->snd_cwnd + incr, USHRT_MAX); /* XXX */ 851 } 852 if (acked > so->so_snd.sb_cc) { 853 tp->snd_wnd -= so->so_snd.sb_cc; 854 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 855 ourfinisacked = 1; 856 } else { 857 sbdrop(&so->so_snd, acked); 858 tp->snd_wnd -= acked; 859 ourfinisacked = 0; 860 } 861 sowwakeup(so); 862 tp->snd_una = ti->ti_ack; 863 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 864 tp->snd_nxt = tp->snd_una; 865 866 switch (tp->t_state) { 867 868 /* 869 * In FIN_WAIT_1 STATE in addition to the processing 870 * for the ESTABLISHED state if our FIN is now acknowledged 871 * then enter FIN_WAIT_2. 872 */ 873 case TCPS_FIN_WAIT_1: 874 if (ourfinisacked) { 875 /* 876 * If we can't receive any more 877 * data, then closing user can proceed. 878 * Starting the timer is contrary to the 879 * specification, but if we don't get a FIN 880 * we'll hang forever. 881 */ 882 if (so->so_state & SS_CANTRCVMORE) { 883 soisdisconnected(so); 884 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 885 } 886 tp->t_state = TCPS_FIN_WAIT_2; 887 } 888 break; 889 890 /* 891 * In CLOSING STATE in addition to the processing for 892 * the ESTABLISHED state if the ACK acknowledges our FIN 893 * then enter the TIME-WAIT state, otherwise ignore 894 * the segment. 895 */ 896 case TCPS_CLOSING: 897 if (ourfinisacked) { 898 tp->t_state = TCPS_TIME_WAIT; 899 tcp_canceltimers(tp); 900 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 901 soisdisconnected(so); 902 } 903 break; 904 905 /* 906 * In LAST_ACK, we may still be waiting for data to drain 907 * and/or to be acked, as well as for the ack of our FIN. 908 * If our FIN is now acknowledged, delete the TCB, 909 * enter the closed state and return. 910 */ 911 case TCPS_LAST_ACK: 912 if (ourfinisacked) { 913 tp = tcp_close(tp); 914 goto drop; 915 } 916 break; 917 918 /* 919 * In TIME_WAIT state the only thing that should arrive 920 * is a retransmission of the remote FIN. Acknowledge 921 * it and restart the finack timer. 922 */ 923 case TCPS_TIME_WAIT: 924 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 925 goto dropafterack; 926 } 927 } 928 929 step6: 930 /* 931 * Update window information. 932 * Don't look at window if no ACK: TAC's send garbage on first SYN. 933 */ 934 if ((tiflags & TH_ACK) && 935 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 936 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 937 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 938 /* keep track of pure window updates */ 939 if (ti->ti_len == 0 && 940 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 941 tcpstat.tcps_rcvwinupd++; 942 tp->snd_wnd = ti->ti_win; 943 tp->snd_wl1 = ti->ti_seq; 944 tp->snd_wl2 = ti->ti_ack; 945 if (tp->snd_wnd > tp->max_sndwnd) 946 tp->max_sndwnd = tp->snd_wnd; 947 needoutput = 1; 948 } 949 950 /* 951 * Process segments with URG. 952 */ 953 if ((tiflags & TH_URG) && ti->ti_urp && 954 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 955 /* 956 * This is a kludge, but if we receive and accept 957 * random urgent pointers, we'll crash in 958 * soreceive. It's hard to imagine someone 959 * actually wanting to send this much urgent data. 960 */ 961 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 962 ti->ti_urp = 0; /* XXX */ 963 tiflags &= ~TH_URG; /* XXX */ 964 goto dodata; /* XXX */ 965 } 966 /* 967 * If this segment advances the known urgent pointer, 968 * then mark the data stream. This should not happen 969 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 970 * a FIN has been received from the remote side. 971 * In these states we ignore the URG. 972 * 973 * According to RFC961 (Assigned Protocols), 974 * the urgent pointer points to the last octet 975 * of urgent data. We continue, however, 976 * to consider it to indicate the first octet 977 * of data past the urgent section 978 * as the original spec states. 979 */ 980 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 981 tp->rcv_up = ti->ti_seq + ti->ti_urp; 982 so->so_oobmark = so->so_rcv.sb_cc + 983 (tp->rcv_up - tp->rcv_nxt) - 1; 984 if (so->so_oobmark == 0) 985 so->so_state |= SS_RCVATMARK; 986 sohasoutofband(so); 987 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 988 } 989 /* 990 * Remove out of band data so doesn't get presented to user. 991 * This can happen independent of advancing the URG pointer, 992 * but if two URG's are pending at once, some out-of-band 993 * data may creep in... ick. 994 */ 995 if (ti->ti_urp <= ti->ti_len && 996 (so->so_options & SO_OOBINLINE) == 0) 997 tcp_pulloutofband(so, ti); 998 } else 999 /* 1000 * If no out of band data is expected, 1001 * pull receive urgent pointer along 1002 * with the receive window. 1003 */ 1004 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1005 tp->rcv_up = tp->rcv_nxt; 1006 dodata: /* XXX */ 1007 1008 /* 1009 * Process the segment text, merging it into the TCP sequencing queue, 1010 * and arranging for acknowledgment of receipt if necessary. 1011 * This process logically involves adjusting tp->rcv_wnd as data 1012 * is presented to the user (this happens in tcp_usrreq.c, 1013 * case PRU_RCVD). If a FIN has already been received on this 1014 * connection then we just ignore the text. 1015 */ 1016 if ((ti->ti_len || (tiflags&TH_FIN)) && 1017 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1018 TCP_REASS(tp, ti, m, so, tiflags); 1019 /* 1020 * Note the amount of data that peer has sent into 1021 * our window, in order to estimate the sender's 1022 * buffer size. 1023 */ 1024 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1025 if (len > tp->max_rcvd) 1026 tp->max_rcvd = len; 1027 } else { 1028 m_freem(m); 1029 tiflags &= ~TH_FIN; 1030 } 1031 1032 /* 1033 * If FIN is received ACK the FIN and let the user know 1034 * that the connection is closing. 1035 */ 1036 if (tiflags & TH_FIN) { 1037 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1038 socantrcvmore(so); 1039 tp->t_flags |= TF_ACKNOW; 1040 tp->rcv_nxt++; 1041 } 1042 switch (tp->t_state) { 1043 1044 /* 1045 * In SYN_RECEIVED and ESTABLISHED STATES 1046 * enter the CLOSE_WAIT state. 1047 */ 1048 case TCPS_SYN_RECEIVED: 1049 case TCPS_ESTABLISHED: 1050 tp->t_state = TCPS_CLOSE_WAIT; 1051 break; 1052 1053 /* 1054 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1055 * enter the CLOSING state. 1056 */ 1057 case TCPS_FIN_WAIT_1: 1058 tp->t_state = TCPS_CLOSING; 1059 break; 1060 1061 /* 1062 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1063 * starting the time-wait timer, turning off the other 1064 * standard timers. 1065 */ 1066 case TCPS_FIN_WAIT_2: 1067 tp->t_state = TCPS_TIME_WAIT; 1068 tcp_canceltimers(tp); 1069 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1070 soisdisconnected(so); 1071 break; 1072 1073 /* 1074 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1075 */ 1076 case TCPS_TIME_WAIT: 1077 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1078 break; 1079 } 1080 } 1081 if (so->so_options & SO_DEBUG) 1082 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1083 1084 /* 1085 * Return any desired output. 1086 */ 1087 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1088 (void) tcp_output(tp); 1089 return; 1090 1091 dropafterack: 1092 /* 1093 * Generate an ACK dropping incoming segment if it occupies 1094 * sequence space, where the ACK reflects our state. 1095 */ 1096 if (tiflags & TH_RST) 1097 goto drop; 1098 m_freem(m); 1099 tp->t_flags |= TF_ACKNOW; 1100 (void) tcp_output(tp); 1101 return; 1102 1103 dropwithreset: 1104 if (om) { 1105 (void) m_free(om); 1106 om = 0; 1107 } 1108 /* 1109 * Generate a RST, dropping incoming segment. 1110 * Make ACK acceptable to originator of segment. 1111 * Don't bother to respond if destination was broadcast. 1112 */ 1113 if ((tiflags & TH_RST) || m->m_flags & M_BCAST) 1114 goto drop; 1115 if (tiflags & TH_ACK) 1116 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1117 else { 1118 if (tiflags & TH_SYN) 1119 ti->ti_len++; 1120 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1121 TH_RST|TH_ACK); 1122 } 1123 /* destroy temporarily created socket */ 1124 if (dropsocket) 1125 (void) soabort(so); 1126 return; 1127 1128 drop: 1129 if (om) 1130 (void) m_free(om); 1131 /* 1132 * Drop space held by incoming segment and return. 1133 */ 1134 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1135 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1136 m_freem(m); 1137 /* destroy temporarily created socket */ 1138 if (dropsocket) 1139 (void) soabort(so); 1140 return; 1141 } 1142 1143 tcp_dooptions(tp, om, ti) 1144 struct tcpcb *tp; 1145 struct mbuf *om; 1146 struct tcpiphdr *ti; 1147 { 1148 register u_char *cp; 1149 int opt, optlen, cnt; 1150 1151 cp = mtod(om, u_char *); 1152 cnt = om->m_len; 1153 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1154 opt = cp[0]; 1155 if (opt == TCPOPT_EOL) 1156 break; 1157 if (opt == TCPOPT_NOP) 1158 optlen = 1; 1159 else { 1160 optlen = cp[1]; 1161 if (optlen <= 0) 1162 break; 1163 } 1164 switch (opt) { 1165 1166 default: 1167 break; 1168 1169 case TCPOPT_MAXSEG: 1170 if (optlen != 4) 1171 continue; 1172 if (!(ti->ti_flags & TH_SYN)) 1173 continue; 1174 tp->t_maxseg = *(u_short *)(cp + 2); 1175 tp->t_maxseg = ntohs((u_short)tp->t_maxseg); 1176 tp->t_maxseg = min(tp->t_maxseg, tcp_mss(tp)); 1177 break; 1178 } 1179 } 1180 (void) m_free(om); 1181 } 1182 1183 /* 1184 * Pull out of band byte out of a segment so 1185 * it doesn't appear in the user's data queue. 1186 * It is still reflected in the segment length for 1187 * sequencing purposes. 1188 */ 1189 tcp_pulloutofband(so, ti) 1190 struct socket *so; 1191 struct tcpiphdr *ti; 1192 { 1193 register struct mbuf *m; 1194 int cnt = ti->ti_urp - 1; 1195 1196 m = dtom(ti); 1197 while (cnt >= 0) { 1198 if (m->m_len > cnt) { 1199 char *cp = mtod(m, caddr_t) + cnt; 1200 struct tcpcb *tp = sototcpcb(so); 1201 1202 tp->t_iobc = *cp; 1203 tp->t_oobflags |= TCPOOB_HAVEDATA; 1204 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1205 m->m_len--; 1206 return; 1207 } 1208 cnt -= m->m_len; 1209 m = m->m_next; 1210 if (m == 0) 1211 break; 1212 } 1213 panic("tcp_pulloutofband"); 1214 } 1215 1216 /* 1217 * Determine a reasonable value for maxseg size. 1218 * If the route is known, use one that can be handled 1219 * on the given interface without forcing IP to fragment. 1220 * If bigger than an mbuf cluster (MCLBYTES), round down to nearest size 1221 * to utilize large mbufs. 1222 * If interface pointer is unavailable, or the destination isn't local, 1223 * use a conservative size (512 or the default IP max size, but no more 1224 * than the mtu of the interface through which we route), 1225 * as we can't discover anything about intervening gateways or networks. 1226 * We also initialize the congestion/slow start window to be a single 1227 * segment if the destination isn't local; this information should 1228 * probably all be saved with the routing entry at the transport level. 1229 * 1230 * This is ugly, and doesn't belong at this level, but has to happen somehow. 1231 */ 1232 tcp_mss(tp) 1233 register struct tcpcb *tp; 1234 { 1235 struct route *ro; 1236 struct ifnet *ifp; 1237 int mss; 1238 struct inpcb *inp; 1239 1240 inp = tp->t_inpcb; 1241 ro = &inp->inp_route; 1242 if ((ro->ro_rt == (struct rtentry *)0) || 1243 (ifp = ro->ro_rt->rt_ifp) == (struct ifnet *)0) { 1244 /* No route yet, so try to acquire one */ 1245 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1246 ro->ro_dst.sa_family = AF_INET; 1247 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1248 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1249 inp->inp_faddr; 1250 rtalloc(ro); 1251 } 1252 if ((ro->ro_rt == 0) || (ifp = ro->ro_rt->rt_ifp) == 0) 1253 return (TCP_MSS); 1254 } 1255 1256 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1257 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1258 if (mss > MCLBYTES) 1259 mss &= ~(MCLBYTES-1); 1260 #else 1261 if (mss > MCLBYTES) 1262 mss = mss / MCLBYTES * MCLBYTES; 1263 #endif 1264 if (in_localaddr(inp->inp_faddr)) 1265 return (mss); 1266 1267 mss = min(mss, TCP_MSS); 1268 tp->snd_cwnd = mss; 1269 return (mss); 1270 } 1271