1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_input.c 7.26 (Berkeley) 01/14/92 8 */ 9 10 #include "param.h" 11 #include "systm.h" 12 #include "malloc.h" 13 #include "mbuf.h" 14 #include "protosw.h" 15 #include "socket.h" 16 #include "socketvar.h" 17 #include "errno.h" 18 19 #include "../net/if.h" 20 #include "../net/route.h" 21 22 #include "in.h" 23 #include "in_systm.h" 24 #include "ip.h" 25 #include "in_pcb.h" 26 #include "ip_var.h" 27 #include "tcp.h" 28 #include "tcp_fsm.h" 29 #include "tcp_seq.h" 30 #include "tcp_timer.h" 31 #include "tcp_var.h" 32 #include "tcpip.h" 33 #include "tcp_debug.h" 34 35 int tcprexmtthresh = 3; 36 int tcppredack; /* XXX debugging: times hdr predict ok for acks */ 37 int tcppreddat; /* XXX # times header prediction ok for data packets */ 38 int tcppcbcachemiss; 39 struct tcpiphdr tcp_saveti; 40 struct inpcb *tcp_last_inpcb = &tcb; 41 42 struct tcpcb *tcp_newtcpcb(); 43 44 /* 45 * Insert segment ti into reassembly queue of tcp with 46 * control block tp. Return TH_FIN if reassembly now includes 47 * a segment with FIN. The macro form does the common case inline 48 * (segment is the next to be received on an established connection, 49 * and the queue is empty), avoiding linkage into and removal 50 * from the queue and repetition of various conversions. 51 * Set DELACK for segments received in order, but ack immediately 52 * when segments are out of order (so fast retransmit can work). 53 */ 54 #define TCP_REASS(tp, ti, m, so, flags) { \ 55 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 56 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 57 (tp)->t_state == TCPS_ESTABLISHED) { \ 58 tp->t_flags |= TF_DELACK; \ 59 (tp)->rcv_nxt += (ti)->ti_len; \ 60 flags = (ti)->ti_flags & TH_FIN; \ 61 tcpstat.tcps_rcvpack++;\ 62 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 63 sbappend(&(so)->so_rcv, (m)); \ 64 sorwakeup(so); \ 65 } else { \ 66 (flags) = tcp_reass((tp), (ti), (m)); \ 67 tp->t_flags |= TF_ACKNOW; \ 68 } \ 69 } 70 71 tcp_reass(tp, ti, m) 72 register struct tcpcb *tp; 73 register struct tcpiphdr *ti; 74 struct mbuf *m; 75 { 76 register struct tcpiphdr *q; 77 struct socket *so = tp->t_inpcb->inp_socket; 78 int flags; 79 80 /* 81 * Call with ti==0 after become established to 82 * force pre-ESTABLISHED data up to user socket. 83 */ 84 if (ti == 0) 85 goto present; 86 87 /* 88 * Find a segment which begins after this one does. 89 */ 90 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 91 q = (struct tcpiphdr *)q->ti_next) 92 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 93 break; 94 95 /* 96 * If there is a preceding segment, it may provide some of 97 * our data already. If so, drop the data from the incoming 98 * segment. If it provides all of our data, drop us. 99 */ 100 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 101 register int i; 102 q = (struct tcpiphdr *)q->ti_prev; 103 /* conversion to int (in i) handles seq wraparound */ 104 i = q->ti_seq + q->ti_len - ti->ti_seq; 105 if (i > 0) { 106 if (i >= ti->ti_len) { 107 tcpstat.tcps_rcvduppack++; 108 tcpstat.tcps_rcvdupbyte += ti->ti_len; 109 m_freem(m); 110 return (0); 111 } 112 m_adj(m, i); 113 ti->ti_len -= i; 114 ti->ti_seq += i; 115 } 116 q = (struct tcpiphdr *)(q->ti_next); 117 } 118 tcpstat.tcps_rcvoopack++; 119 tcpstat.tcps_rcvoobyte += ti->ti_len; 120 REASS_MBUF(ti) = m; /* XXX */ 121 122 /* 123 * While we overlap succeeding segments trim them or, 124 * if they are completely covered, dequeue them. 125 */ 126 while (q != (struct tcpiphdr *)tp) { 127 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 128 if (i <= 0) 129 break; 130 if (i < q->ti_len) { 131 q->ti_seq += i; 132 q->ti_len -= i; 133 m_adj(REASS_MBUF(q), i); 134 break; 135 } 136 q = (struct tcpiphdr *)q->ti_next; 137 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 138 remque(q->ti_prev); 139 m_freem(m); 140 } 141 142 /* 143 * Stick new segment in its place. 144 */ 145 insque(ti, q->ti_prev); 146 147 present: 148 /* 149 * Present data to user, advancing rcv_nxt through 150 * completed sequence space. 151 */ 152 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 153 return (0); 154 ti = tp->seg_next; 155 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 156 return (0); 157 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 158 return (0); 159 do { 160 tp->rcv_nxt += ti->ti_len; 161 flags = ti->ti_flags & TH_FIN; 162 remque(ti); 163 m = REASS_MBUF(ti); 164 ti = (struct tcpiphdr *)ti->ti_next; 165 if (so->so_state & SS_CANTRCVMORE) 166 m_freem(m); 167 else 168 sbappend(&so->so_rcv, m); 169 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 170 sorwakeup(so); 171 return (flags); 172 } 173 174 /* 175 * TCP input routine, follows pages 65-76 of the 176 * protocol specification dated September, 1981 very closely. 177 */ 178 tcp_input(m, iphlen) 179 register struct mbuf *m; 180 int iphlen; 181 { 182 register struct tcpiphdr *ti; 183 register struct inpcb *inp; 184 struct mbuf *om = 0; 185 int len, tlen, off; 186 register struct tcpcb *tp = 0; 187 register int tiflags; 188 struct socket *so; 189 int todrop, acked, ourfinisacked, needoutput = 0; 190 short ostate; 191 struct in_addr laddr; 192 int dropsocket = 0; 193 int iss = 0; 194 195 tcpstat.tcps_rcvtotal++; 196 /* 197 * Get IP and TCP header together in first mbuf. 198 * Note: IP leaves IP header in first mbuf. 199 */ 200 ti = mtod(m, struct tcpiphdr *); 201 if (iphlen > sizeof (struct ip)) 202 ip_stripoptions(m, (struct mbuf *)0); 203 if (m->m_len < sizeof (struct tcpiphdr)) { 204 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 205 tcpstat.tcps_rcvshort++; 206 return; 207 } 208 ti = mtod(m, struct tcpiphdr *); 209 } 210 211 /* 212 * Checksum extended TCP header and data. 213 */ 214 tlen = ((struct ip *)ti)->ip_len; 215 len = sizeof (struct ip) + tlen; 216 ti->ti_next = ti->ti_prev = 0; 217 ti->ti_x1 = 0; 218 ti->ti_len = (u_short)tlen; 219 HTONS(ti->ti_len); 220 if (ti->ti_sum = in_cksum(m, len)) { 221 tcpstat.tcps_rcvbadsum++; 222 goto drop; 223 } 224 225 /* 226 * Check that TCP offset makes sense, 227 * pull out TCP options and adjust length. XXX 228 */ 229 off = ti->ti_off << 2; 230 if (off < sizeof (struct tcphdr) || off > tlen) { 231 tcpstat.tcps_rcvbadoff++; 232 goto drop; 233 } 234 tlen -= off; 235 ti->ti_len = tlen; 236 if (off > sizeof (struct tcphdr)) { 237 if (m->m_len < sizeof(struct ip) + off) { 238 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 239 tcpstat.tcps_rcvshort++; 240 return; 241 } 242 ti = mtod(m, struct tcpiphdr *); 243 } 244 om = m_get(M_DONTWAIT, MT_DATA); 245 if (om == 0) 246 goto drop; 247 om->m_len = off - sizeof (struct tcphdr); 248 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 249 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 250 m->m_len -= om->m_len; 251 m->m_pkthdr.len -= om->m_len; 252 bcopy(op+om->m_len, op, 253 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 254 } 255 } 256 tiflags = ti->ti_flags; 257 258 /* 259 * Convert TCP protocol specific fields to host format. 260 */ 261 NTOHL(ti->ti_seq); 262 NTOHL(ti->ti_ack); 263 NTOHS(ti->ti_win); 264 NTOHS(ti->ti_urp); 265 266 /* 267 * Locate pcb for segment. 268 */ 269 findpcb: 270 inp = tcp_last_inpcb; 271 if (inp->inp_lport != ti->ti_dport || 272 inp->inp_fport != ti->ti_sport || 273 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 274 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 275 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 276 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 277 if (inp) 278 tcp_last_inpcb = inp; 279 ++tcppcbcachemiss; 280 } 281 282 /* 283 * If the state is CLOSED (i.e., TCB does not exist) then 284 * all data in the incoming segment is discarded. 285 * If the TCB exists but is in CLOSED state, it is embryonic, 286 * but should either do a listen or a connect soon. 287 */ 288 if (inp == 0) 289 goto dropwithreset; 290 tp = intotcpcb(inp); 291 if (tp == 0) 292 goto dropwithreset; 293 if (tp->t_state == TCPS_CLOSED) 294 goto drop; 295 so = inp->inp_socket; 296 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 297 if (so->so_options & SO_DEBUG) { 298 ostate = tp->t_state; 299 tcp_saveti = *ti; 300 } 301 if (so->so_options & SO_ACCEPTCONN) { 302 so = sonewconn(so, 0); 303 if (so == 0) 304 goto drop; 305 /* 306 * This is ugly, but .... 307 * 308 * Mark socket as temporary until we're 309 * committed to keeping it. The code at 310 * ``drop'' and ``dropwithreset'' check the 311 * flag dropsocket to see if the temporary 312 * socket created here should be discarded. 313 * We mark the socket as discardable until 314 * we're committed to it below in TCPS_LISTEN. 315 */ 316 dropsocket++; 317 inp = (struct inpcb *)so->so_pcb; 318 inp->inp_laddr = ti->ti_dst; 319 inp->inp_lport = ti->ti_dport; 320 #if BSD>=43 321 inp->inp_options = ip_srcroute(); 322 #endif 323 tp = intotcpcb(inp); 324 tp->t_state = TCPS_LISTEN; 325 } 326 } 327 328 /* 329 * Segment received on connection. 330 * Reset idle time and keep-alive timer. 331 */ 332 tp->t_idle = 0; 333 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 334 335 /* 336 * Process options if not in LISTEN state, 337 * else do it below (after getting remote address). 338 */ 339 if (om && tp->t_state != TCPS_LISTEN) { 340 tcp_dooptions(tp, om, ti); 341 om = 0; 342 } 343 /* 344 * Header prediction: check for the two common cases 345 * of a uni-directional data xfer. If the packet has 346 * no control flags, is in-sequence, the window didn't 347 * change and we're not retransmitting, it's a 348 * candidate. If the length is zero and the ack moved 349 * forward, we're the sender side of the xfer. Just 350 * free the data acked & wake any higher level process 351 * that was blocked waiting for space. If the length 352 * is non-zero and the ack didn't move, we're the 353 * receiver side. If we're getting packets in-order 354 * (the reassembly queue is empty), add the data to 355 * the socket buffer and note that we need a delayed ack. 356 */ 357 if (tp->t_state == TCPS_ESTABLISHED && 358 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 359 ti->ti_seq == tp->rcv_nxt && 360 ti->ti_win && ti->ti_win == tp->snd_wnd && 361 tp->snd_nxt == tp->snd_max) { 362 if (ti->ti_len == 0) { 363 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 364 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 365 tp->snd_cwnd >= tp->snd_wnd) { 366 /* 367 * this is a pure ack for outstanding data. 368 */ 369 ++tcppredack; 370 if (tp->t_rtt && SEQ_GT(ti->ti_ack,tp->t_rtseq)) 371 tcp_xmit_timer(tp); 372 acked = ti->ti_ack - tp->snd_una; 373 tcpstat.tcps_rcvackpack++; 374 tcpstat.tcps_rcvackbyte += acked; 375 sbdrop(&so->so_snd, acked); 376 tp->snd_una = ti->ti_ack; 377 m_freem(m); 378 379 /* 380 * If all outstanding data are acked, stop 381 * retransmit timer, otherwise restart timer 382 * using current (possibly backed-off) value. 383 * If process is waiting for space, 384 * wakeup/selwakeup/signal. If data 385 * are ready to send, let tcp_output 386 * decide between more output or persist. 387 */ 388 if (tp->snd_una == tp->snd_max) 389 tp->t_timer[TCPT_REXMT] = 0; 390 else if (tp->t_timer[TCPT_PERSIST] == 0) 391 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 392 393 if (so->so_snd.sb_flags & SB_NOTIFY) 394 sowwakeup(so); 395 if (so->so_snd.sb_cc) 396 (void) tcp_output(tp); 397 return; 398 } 399 } else if (ti->ti_ack == tp->snd_una && 400 tp->seg_next == (struct tcpiphdr *)tp && 401 ti->ti_len <= sbspace(&so->so_rcv)) { 402 /* 403 * this is a pure, in-sequence data packet 404 * with nothing on the reassembly queue and 405 * we have enough buffer space to take it. 406 */ 407 ++tcppreddat; 408 tp->rcv_nxt += ti->ti_len; 409 tcpstat.tcps_rcvpack++; 410 tcpstat.tcps_rcvbyte += ti->ti_len; 411 /* 412 * Drop TCP and IP headers then add data 413 * to socket buffer 414 */ 415 m->m_data += sizeof(struct tcpiphdr); 416 m->m_len -= sizeof(struct tcpiphdr); 417 sbappend(&so->so_rcv, m); 418 sorwakeup(so); 419 tp->t_flags |= TF_DELACK; 420 return; 421 } 422 } 423 424 /* 425 * Drop TCP and IP headers; TCP options were dropped above. 426 */ 427 m->m_data += sizeof(struct tcpiphdr); 428 m->m_len -= sizeof(struct tcpiphdr); 429 430 /* 431 * Calculate amount of space in receive window, 432 * and then do TCP input processing. 433 * Receive window is amount of space in rcv queue, 434 * but not less than advertised window. 435 */ 436 { int win; 437 438 win = sbspace(&so->so_rcv); 439 if (win < 0) 440 win = 0; 441 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 442 } 443 444 switch (tp->t_state) { 445 446 /* 447 * If the state is LISTEN then ignore segment if it contains an RST. 448 * If the segment contains an ACK then it is bad and send a RST. 449 * If it does not contain a SYN then it is not interesting; drop it. 450 * Don't bother responding if the destination was a broadcast. 451 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 452 * tp->iss, and send a segment: 453 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 454 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 455 * Fill in remote peer address fields if not previously specified. 456 * Enter SYN_RECEIVED state, and process any other fields of this 457 * segment in this state. 458 */ 459 case TCPS_LISTEN: { 460 struct mbuf *am; 461 register struct sockaddr_in *sin; 462 463 if (tiflags & TH_RST) 464 goto drop; 465 if (tiflags & TH_ACK) 466 goto dropwithreset; 467 if ((tiflags & TH_SYN) == 0) 468 goto drop; 469 if (m->m_flags & M_BCAST) 470 goto drop; 471 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 472 if (am == NULL) 473 goto drop; 474 am->m_len = sizeof (struct sockaddr_in); 475 sin = mtod(am, struct sockaddr_in *); 476 sin->sin_family = AF_INET; 477 sin->sin_len = sizeof(*sin); 478 sin->sin_addr = ti->ti_src; 479 sin->sin_port = ti->ti_sport; 480 laddr = inp->inp_laddr; 481 if (inp->inp_laddr.s_addr == INADDR_ANY) 482 inp->inp_laddr = ti->ti_dst; 483 if (in_pcbconnect(inp, am)) { 484 inp->inp_laddr = laddr; 485 (void) m_free(am); 486 goto drop; 487 } 488 (void) m_free(am); 489 tp->t_template = tcp_template(tp); 490 if (tp->t_template == 0) { 491 tp = tcp_drop(tp, ENOBUFS); 492 dropsocket = 0; /* socket is already gone */ 493 goto drop; 494 } 495 if (om) { 496 tcp_dooptions(tp, om, ti); 497 om = 0; 498 } 499 if (iss) 500 tp->iss = iss; 501 else 502 tp->iss = tcp_iss; 503 tcp_iss += TCP_ISSINCR/2; 504 tp->irs = ti->ti_seq; 505 tcp_sendseqinit(tp); 506 tcp_rcvseqinit(tp); 507 tp->t_flags |= TF_ACKNOW; 508 tp->t_state = TCPS_SYN_RECEIVED; 509 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 510 dropsocket = 0; /* committed to socket */ 511 tcpstat.tcps_accepts++; 512 goto trimthenstep6; 513 } 514 515 /* 516 * If the state is SYN_SENT: 517 * if seg contains an ACK, but not for our SYN, drop the input. 518 * if seg contains a RST, then drop the connection. 519 * if seg does not contain SYN, then drop it. 520 * Otherwise this is an acceptable SYN segment 521 * initialize tp->rcv_nxt and tp->irs 522 * if seg contains ack then advance tp->snd_una 523 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 524 * arrange for segment to be acked (eventually) 525 * continue processing rest of data/controls, beginning with URG 526 */ 527 case TCPS_SYN_SENT: 528 if ((tiflags & TH_ACK) && 529 (SEQ_LEQ(ti->ti_ack, tp->iss) || 530 SEQ_GT(ti->ti_ack, tp->snd_max))) 531 goto dropwithreset; 532 if (tiflags & TH_RST) { 533 if (tiflags & TH_ACK) 534 tp = tcp_drop(tp, ECONNREFUSED); 535 goto drop; 536 } 537 if ((tiflags & TH_SYN) == 0) 538 goto drop; 539 if (tiflags & TH_ACK) { 540 tp->snd_una = ti->ti_ack; 541 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 542 tp->snd_nxt = tp->snd_una; 543 } 544 tp->t_timer[TCPT_REXMT] = 0; 545 tp->irs = ti->ti_seq; 546 tcp_rcvseqinit(tp); 547 tp->t_flags |= TF_ACKNOW; 548 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 549 tcpstat.tcps_connects++; 550 soisconnected(so); 551 tp->t_state = TCPS_ESTABLISHED; 552 (void) tcp_reass(tp, (struct tcpiphdr *)0, 553 (struct mbuf *)0); 554 /* 555 * if we didn't have to retransmit the SYN, 556 * use its rtt as our initial srtt & rtt var. 557 */ 558 if (tp->t_rtt) 559 tcp_xmit_timer(tp); 560 } else 561 tp->t_state = TCPS_SYN_RECEIVED; 562 563 trimthenstep6: 564 /* 565 * Advance ti->ti_seq to correspond to first data byte. 566 * If data, trim to stay within window, 567 * dropping FIN if necessary. 568 */ 569 ti->ti_seq++; 570 if (ti->ti_len > tp->rcv_wnd) { 571 todrop = ti->ti_len - tp->rcv_wnd; 572 m_adj(m, -todrop); 573 ti->ti_len = tp->rcv_wnd; 574 tiflags &= ~TH_FIN; 575 tcpstat.tcps_rcvpackafterwin++; 576 tcpstat.tcps_rcvbyteafterwin += todrop; 577 } 578 tp->snd_wl1 = ti->ti_seq - 1; 579 tp->rcv_up = ti->ti_seq; 580 goto step6; 581 } 582 583 /* 584 * States other than LISTEN or SYN_SENT. 585 * First check that at least some bytes of segment are within 586 * receive window. If segment begins before rcv_nxt, 587 * drop leading data (and SYN); if nothing left, just ack. 588 */ 589 todrop = tp->rcv_nxt - ti->ti_seq; 590 if (todrop > 0) { 591 if (tiflags & TH_SYN) { 592 tiflags &= ~TH_SYN; 593 ti->ti_seq++; 594 if (ti->ti_urp > 1) 595 ti->ti_urp--; 596 else 597 tiflags &= ~TH_URG; 598 todrop--; 599 } 600 if (todrop > ti->ti_len || 601 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 602 tcpstat.tcps_rcvduppack++; 603 tcpstat.tcps_rcvdupbyte += ti->ti_len; 604 /* 605 * If segment is just one to the left of the window, 606 * check two special cases: 607 * 1. Don't toss RST in response to 4.2-style keepalive. 608 * 2. If the only thing to drop is a FIN, we can drop 609 * it, but check the ACK or we will get into FIN 610 * wars if our FINs crossed (both CLOSING). 611 * In either case, send ACK to resynchronize, 612 * but keep on processing for RST or ACK. 613 */ 614 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 615 #ifdef TCP_COMPAT_42 616 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 617 #endif 618 ) { 619 todrop = ti->ti_len; 620 tiflags &= ~TH_FIN; 621 tp->t_flags |= TF_ACKNOW; 622 } else 623 goto dropafterack; 624 } else { 625 tcpstat.tcps_rcvpartduppack++; 626 tcpstat.tcps_rcvpartdupbyte += todrop; 627 } 628 m_adj(m, todrop); 629 ti->ti_seq += todrop; 630 ti->ti_len -= todrop; 631 if (ti->ti_urp > todrop) 632 ti->ti_urp -= todrop; 633 else { 634 tiflags &= ~TH_URG; 635 ti->ti_urp = 0; 636 } 637 } 638 639 /* 640 * If new data are received on a connection after the 641 * user processes are gone, then RST the other end. 642 */ 643 if ((so->so_state & SS_NOFDREF) && 644 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 645 tp = tcp_close(tp); 646 tcpstat.tcps_rcvafterclose++; 647 goto dropwithreset; 648 } 649 650 /* 651 * If segment ends after window, drop trailing data 652 * (and PUSH and FIN); if nothing left, just ACK. 653 */ 654 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 655 if (todrop > 0) { 656 tcpstat.tcps_rcvpackafterwin++; 657 if (todrop >= ti->ti_len) { 658 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 659 /* 660 * If a new connection request is received 661 * while in TIME_WAIT, drop the old connection 662 * and start over if the sequence numbers 663 * are above the previous ones. 664 */ 665 if (tiflags & TH_SYN && 666 tp->t_state == TCPS_TIME_WAIT && 667 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 668 iss = tp->rcv_nxt + TCP_ISSINCR; 669 tp = tcp_close(tp); 670 goto findpcb; 671 } 672 /* 673 * If window is closed can only take segments at 674 * window edge, and have to drop data and PUSH from 675 * incoming segments. Continue processing, but 676 * remember to ack. Otherwise, drop segment 677 * and ack. 678 */ 679 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 680 tp->t_flags |= TF_ACKNOW; 681 tcpstat.tcps_rcvwinprobe++; 682 } else 683 goto dropafterack; 684 } else 685 tcpstat.tcps_rcvbyteafterwin += todrop; 686 m_adj(m, -todrop); 687 ti->ti_len -= todrop; 688 tiflags &= ~(TH_PUSH|TH_FIN); 689 } 690 691 /* 692 * If the RST bit is set examine the state: 693 * SYN_RECEIVED STATE: 694 * If passive open, return to LISTEN state. 695 * If active open, inform user that connection was refused. 696 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 697 * Inform user that connection was reset, and close tcb. 698 * CLOSING, LAST_ACK, TIME_WAIT STATES 699 * Close the tcb. 700 */ 701 if (tiflags&TH_RST) switch (tp->t_state) { 702 703 case TCPS_SYN_RECEIVED: 704 so->so_error = ECONNREFUSED; 705 goto close; 706 707 case TCPS_ESTABLISHED: 708 case TCPS_FIN_WAIT_1: 709 case TCPS_FIN_WAIT_2: 710 case TCPS_CLOSE_WAIT: 711 so->so_error = ECONNRESET; 712 close: 713 tp->t_state = TCPS_CLOSED; 714 tcpstat.tcps_drops++; 715 tp = tcp_close(tp); 716 goto drop; 717 718 case TCPS_CLOSING: 719 case TCPS_LAST_ACK: 720 case TCPS_TIME_WAIT: 721 tp = tcp_close(tp); 722 goto drop; 723 } 724 725 /* 726 * If a SYN is in the window, then this is an 727 * error and we send an RST and drop the connection. 728 */ 729 if (tiflags & TH_SYN) { 730 tp = tcp_drop(tp, ECONNRESET); 731 goto dropwithreset; 732 } 733 734 /* 735 * If the ACK bit is off we drop the segment and return. 736 */ 737 if ((tiflags & TH_ACK) == 0) 738 goto drop; 739 740 /* 741 * Ack processing. 742 */ 743 switch (tp->t_state) { 744 745 /* 746 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 747 * ESTABLISHED state and continue processing, otherwise 748 * send an RST. 749 */ 750 case TCPS_SYN_RECEIVED: 751 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 752 SEQ_GT(ti->ti_ack, tp->snd_max)) 753 goto dropwithreset; 754 tcpstat.tcps_connects++; 755 soisconnected(so); 756 tp->t_state = TCPS_ESTABLISHED; 757 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 758 tp->snd_wl1 = ti->ti_seq - 1; 759 /* fall into ... */ 760 761 /* 762 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 763 * ACKs. If the ack is in the range 764 * tp->snd_una < ti->ti_ack <= tp->snd_max 765 * then advance tp->snd_una to ti->ti_ack and drop 766 * data from the retransmission queue. If this ACK reflects 767 * more up to date window information we update our window information. 768 */ 769 case TCPS_ESTABLISHED: 770 case TCPS_FIN_WAIT_1: 771 case TCPS_FIN_WAIT_2: 772 case TCPS_CLOSE_WAIT: 773 case TCPS_CLOSING: 774 case TCPS_LAST_ACK: 775 case TCPS_TIME_WAIT: 776 777 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 778 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 779 tcpstat.tcps_rcvdupack++; 780 /* 781 * If we have outstanding data (other than 782 * a window probe), this is a completely 783 * duplicate ack (ie, window info didn't 784 * change), the ack is the biggest we've 785 * seen and we've seen exactly our rexmt 786 * threshhold of them, assume a packet 787 * has been dropped and retransmit it. 788 * Kludge snd_nxt & the congestion 789 * window so we send only this one 790 * packet. 791 * 792 * We know we're losing at the current 793 * window size so do congestion avoidance 794 * (set ssthresh to half the current window 795 * and pull our congestion window back to 796 * the new ssthresh). 797 * 798 * Dup acks mean that packets have left the 799 * network (they're now cached at the receiver) 800 * so bump cwnd by the amount in the receiver 801 * to keep a constant cwnd packets in the 802 * network. 803 */ 804 if (tp->t_timer[TCPT_REXMT] == 0 || 805 ti->ti_ack != tp->snd_una) 806 tp->t_dupacks = 0; 807 else if (++tp->t_dupacks == tcprexmtthresh) { 808 tcp_seq onxt = tp->snd_nxt; 809 u_int win = 810 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 811 tp->t_maxseg; 812 813 if (win < 2) 814 win = 2; 815 tp->snd_ssthresh = win * tp->t_maxseg; 816 tp->t_timer[TCPT_REXMT] = 0; 817 tp->t_rtt = 0; 818 tp->snd_nxt = ti->ti_ack; 819 tp->snd_cwnd = tp->t_maxseg; 820 (void) tcp_output(tp); 821 tp->snd_cwnd = tp->snd_ssthresh + 822 tp->t_maxseg * tp->t_dupacks; 823 if (SEQ_GT(onxt, tp->snd_nxt)) 824 tp->snd_nxt = onxt; 825 goto drop; 826 } else if (tp->t_dupacks > tcprexmtthresh) { 827 tp->snd_cwnd += tp->t_maxseg; 828 (void) tcp_output(tp); 829 goto drop; 830 } 831 } else 832 tp->t_dupacks = 0; 833 break; 834 } 835 /* 836 * If the congestion window was inflated to account 837 * for the other side's cached packets, retract it. 838 */ 839 if (tp->t_dupacks > tcprexmtthresh && 840 tp->snd_cwnd > tp->snd_ssthresh) 841 tp->snd_cwnd = tp->snd_ssthresh; 842 tp->t_dupacks = 0; 843 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 844 tcpstat.tcps_rcvacktoomuch++; 845 goto dropafterack; 846 } 847 acked = ti->ti_ack - tp->snd_una; 848 tcpstat.tcps_rcvackpack++; 849 tcpstat.tcps_rcvackbyte += acked; 850 851 /* 852 * If transmit timer is running and timed sequence 853 * number was acked, update smoothed round trip time. 854 * Since we now have an rtt measurement, cancel the 855 * timer backoff (cf., Phil Karn's retransmit alg.). 856 * Recompute the initial retransmit timer. 857 */ 858 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 859 tcp_xmit_timer(tp); 860 861 /* 862 * If all outstanding data is acked, stop retransmit 863 * timer and remember to restart (more output or persist). 864 * If there is more data to be acked, restart retransmit 865 * timer, using current (possibly backed-off) value. 866 */ 867 if (ti->ti_ack == tp->snd_max) { 868 tp->t_timer[TCPT_REXMT] = 0; 869 needoutput = 1; 870 } else if (tp->t_timer[TCPT_PERSIST] == 0) 871 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 872 /* 873 * When new data is acked, open the congestion window. 874 * If the window gives us less than ssthresh packets 875 * in flight, open exponentially (maxseg per packet). 876 * Otherwise open linearly: maxseg per window 877 * (maxseg^2 / cwnd per packet), plus a constant 878 * fraction of a packet (maxseg/8) to help larger windows 879 * open quickly enough. 880 */ 881 { 882 register u_int cw = tp->snd_cwnd; 883 register u_int incr = tp->t_maxseg; 884 885 if (cw > tp->snd_ssthresh) 886 incr = incr * incr / cw + incr / 8; 887 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN); 888 } 889 if (acked > so->so_snd.sb_cc) { 890 tp->snd_wnd -= so->so_snd.sb_cc; 891 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 892 ourfinisacked = 1; 893 } else { 894 sbdrop(&so->so_snd, acked); 895 tp->snd_wnd -= acked; 896 ourfinisacked = 0; 897 } 898 if (so->so_snd.sb_flags & SB_NOTIFY) 899 sowwakeup(so); 900 tp->snd_una = ti->ti_ack; 901 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 902 tp->snd_nxt = tp->snd_una; 903 904 switch (tp->t_state) { 905 906 /* 907 * In FIN_WAIT_1 STATE in addition to the processing 908 * for the ESTABLISHED state if our FIN is now acknowledged 909 * then enter FIN_WAIT_2. 910 */ 911 case TCPS_FIN_WAIT_1: 912 if (ourfinisacked) { 913 /* 914 * If we can't receive any more 915 * data, then closing user can proceed. 916 * Starting the timer is contrary to the 917 * specification, but if we don't get a FIN 918 * we'll hang forever. 919 */ 920 if (so->so_state & SS_CANTRCVMORE) { 921 soisdisconnected(so); 922 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 923 } 924 tp->t_state = TCPS_FIN_WAIT_2; 925 } 926 break; 927 928 /* 929 * In CLOSING STATE in addition to the processing for 930 * the ESTABLISHED state if the ACK acknowledges our FIN 931 * then enter the TIME-WAIT state, otherwise ignore 932 * the segment. 933 */ 934 case TCPS_CLOSING: 935 if (ourfinisacked) { 936 tp->t_state = TCPS_TIME_WAIT; 937 tcp_canceltimers(tp); 938 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 939 soisdisconnected(so); 940 } 941 break; 942 943 /* 944 * In LAST_ACK, we may still be waiting for data to drain 945 * and/or to be acked, as well as for the ack of our FIN. 946 * If our FIN is now acknowledged, delete the TCB, 947 * enter the closed state and return. 948 */ 949 case TCPS_LAST_ACK: 950 if (ourfinisacked) { 951 tp = tcp_close(tp); 952 goto drop; 953 } 954 break; 955 956 /* 957 * In TIME_WAIT state the only thing that should arrive 958 * is a retransmission of the remote FIN. Acknowledge 959 * it and restart the finack timer. 960 */ 961 case TCPS_TIME_WAIT: 962 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 963 goto dropafterack; 964 } 965 } 966 967 step6: 968 /* 969 * Update window information. 970 * Don't look at window if no ACK: TAC's send garbage on first SYN. 971 */ 972 if ((tiflags & TH_ACK) && 973 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 974 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 975 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 976 /* keep track of pure window updates */ 977 if (ti->ti_len == 0 && 978 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 979 tcpstat.tcps_rcvwinupd++; 980 tp->snd_wnd = ti->ti_win; 981 tp->snd_wl1 = ti->ti_seq; 982 tp->snd_wl2 = ti->ti_ack; 983 if (tp->snd_wnd > tp->max_sndwnd) 984 tp->max_sndwnd = tp->snd_wnd; 985 needoutput = 1; 986 } 987 988 /* 989 * Process segments with URG. 990 */ 991 if ((tiflags & TH_URG) && ti->ti_urp && 992 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 993 /* 994 * This is a kludge, but if we receive and accept 995 * random urgent pointers, we'll crash in 996 * soreceive. It's hard to imagine someone 997 * actually wanting to send this much urgent data. 998 */ 999 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 1000 ti->ti_urp = 0; /* XXX */ 1001 tiflags &= ~TH_URG; /* XXX */ 1002 goto dodata; /* XXX */ 1003 } 1004 /* 1005 * If this segment advances the known urgent pointer, 1006 * then mark the data stream. This should not happen 1007 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1008 * a FIN has been received from the remote side. 1009 * In these states we ignore the URG. 1010 * 1011 * According to RFC961 (Assigned Protocols), 1012 * the urgent pointer points to the last octet 1013 * of urgent data. We continue, however, 1014 * to consider it to indicate the first octet 1015 * of data past the urgent section as the original 1016 * spec states (in one of two places). 1017 */ 1018 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1019 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1020 so->so_oobmark = so->so_rcv.sb_cc + 1021 (tp->rcv_up - tp->rcv_nxt) - 1; 1022 if (so->so_oobmark == 0) 1023 so->so_state |= SS_RCVATMARK; 1024 sohasoutofband(so); 1025 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1026 } 1027 /* 1028 * Remove out of band data so doesn't get presented to user. 1029 * This can happen independent of advancing the URG pointer, 1030 * but if two URG's are pending at once, some out-of-band 1031 * data may creep in... ick. 1032 */ 1033 if (ti->ti_urp <= ti->ti_len 1034 #ifdef SO_OOBINLINE 1035 && (so->so_options & SO_OOBINLINE) == 0 1036 #endif 1037 ) 1038 tcp_pulloutofband(so, ti, m); 1039 } else 1040 /* 1041 * If no out of band data is expected, 1042 * pull receive urgent pointer along 1043 * with the receive window. 1044 */ 1045 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1046 tp->rcv_up = tp->rcv_nxt; 1047 dodata: /* XXX */ 1048 1049 /* 1050 * Process the segment text, merging it into the TCP sequencing queue, 1051 * and arranging for acknowledgment of receipt if necessary. 1052 * This process logically involves adjusting tp->rcv_wnd as data 1053 * is presented to the user (this happens in tcp_usrreq.c, 1054 * case PRU_RCVD). If a FIN has already been received on this 1055 * connection then we just ignore the text. 1056 */ 1057 if ((ti->ti_len || (tiflags&TH_FIN)) && 1058 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1059 TCP_REASS(tp, ti, m, so, tiflags); 1060 /* 1061 * Note the amount of data that peer has sent into 1062 * our window, in order to estimate the sender's 1063 * buffer size. 1064 */ 1065 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1066 } else { 1067 m_freem(m); 1068 tiflags &= ~TH_FIN; 1069 } 1070 1071 /* 1072 * If FIN is received ACK the FIN and let the user know 1073 * that the connection is closing. 1074 */ 1075 if (tiflags & TH_FIN) { 1076 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1077 socantrcvmore(so); 1078 tp->t_flags |= TF_ACKNOW; 1079 tp->rcv_nxt++; 1080 } 1081 switch (tp->t_state) { 1082 1083 /* 1084 * In SYN_RECEIVED and ESTABLISHED STATES 1085 * enter the CLOSE_WAIT state. 1086 */ 1087 case TCPS_SYN_RECEIVED: 1088 case TCPS_ESTABLISHED: 1089 tp->t_state = TCPS_CLOSE_WAIT; 1090 break; 1091 1092 /* 1093 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1094 * enter the CLOSING state. 1095 */ 1096 case TCPS_FIN_WAIT_1: 1097 tp->t_state = TCPS_CLOSING; 1098 break; 1099 1100 /* 1101 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1102 * starting the time-wait timer, turning off the other 1103 * standard timers. 1104 */ 1105 case TCPS_FIN_WAIT_2: 1106 tp->t_state = TCPS_TIME_WAIT; 1107 tcp_canceltimers(tp); 1108 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1109 soisdisconnected(so); 1110 break; 1111 1112 /* 1113 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1114 */ 1115 case TCPS_TIME_WAIT: 1116 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1117 break; 1118 } 1119 } 1120 if (so->so_options & SO_DEBUG) 1121 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1122 1123 /* 1124 * Return any desired output. 1125 */ 1126 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1127 (void) tcp_output(tp); 1128 return; 1129 1130 dropafterack: 1131 /* 1132 * Generate an ACK dropping incoming segment if it occupies 1133 * sequence space, where the ACK reflects our state. 1134 */ 1135 if (tiflags & TH_RST) 1136 goto drop; 1137 m_freem(m); 1138 tp->t_flags |= TF_ACKNOW; 1139 (void) tcp_output(tp); 1140 return; 1141 1142 dropwithreset: 1143 if (om) { 1144 (void) m_free(om); 1145 om = 0; 1146 } 1147 /* 1148 * Generate a RST, dropping incoming segment. 1149 * Make ACK acceptable to originator of segment. 1150 * Don't bother to respond if destination was broadcast. 1151 */ 1152 if ((tiflags & TH_RST) || m->m_flags & M_BCAST) 1153 goto drop; 1154 if (tiflags & TH_ACK) 1155 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1156 else { 1157 if (tiflags & TH_SYN) 1158 ti->ti_len++; 1159 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1160 TH_RST|TH_ACK); 1161 } 1162 /* destroy temporarily created socket */ 1163 if (dropsocket) 1164 (void) soabort(so); 1165 return; 1166 1167 drop: 1168 if (om) 1169 (void) m_free(om); 1170 /* 1171 * Drop space held by incoming segment and return. 1172 */ 1173 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1174 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1175 m_freem(m); 1176 /* destroy temporarily created socket */ 1177 if (dropsocket) 1178 (void) soabort(so); 1179 return; 1180 } 1181 1182 tcp_dooptions(tp, om, ti) 1183 struct tcpcb *tp; 1184 struct mbuf *om; 1185 struct tcpiphdr *ti; 1186 { 1187 register u_char *cp; 1188 u_short mss; 1189 int opt, optlen, cnt; 1190 1191 cp = mtod(om, u_char *); 1192 cnt = om->m_len; 1193 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1194 opt = cp[0]; 1195 if (opt == TCPOPT_EOL) 1196 break; 1197 if (opt == TCPOPT_NOP) 1198 optlen = 1; 1199 else { 1200 optlen = cp[1]; 1201 if (optlen <= 0) 1202 break; 1203 } 1204 switch (opt) { 1205 1206 default: 1207 continue; 1208 1209 case TCPOPT_MAXSEG: 1210 if (optlen != 4) 1211 continue; 1212 if (!(ti->ti_flags & TH_SYN)) 1213 continue; 1214 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1215 NTOHS(mss); 1216 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1217 break; 1218 } 1219 } 1220 (void) m_free(om); 1221 } 1222 1223 /* 1224 * Pull out of band byte out of a segment so 1225 * it doesn't appear in the user's data queue. 1226 * It is still reflected in the segment length for 1227 * sequencing purposes. 1228 */ 1229 tcp_pulloutofband(so, ti, m) 1230 struct socket *so; 1231 struct tcpiphdr *ti; 1232 register struct mbuf *m; 1233 { 1234 int cnt = ti->ti_urp - 1; 1235 1236 while (cnt >= 0) { 1237 if (m->m_len > cnt) { 1238 char *cp = mtod(m, caddr_t) + cnt; 1239 struct tcpcb *tp = sototcpcb(so); 1240 1241 tp->t_iobc = *cp; 1242 tp->t_oobflags |= TCPOOB_HAVEDATA; 1243 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1244 m->m_len--; 1245 return; 1246 } 1247 cnt -= m->m_len; 1248 m = m->m_next; 1249 if (m == 0) 1250 break; 1251 } 1252 panic("tcp_pulloutofband"); 1253 } 1254 1255 /* 1256 * Collect new round-trip time estimate 1257 * and update averages and current timeout. 1258 */ 1259 tcp_xmit_timer(tp) 1260 register struct tcpcb *tp; 1261 { 1262 register short delta; 1263 1264 tcpstat.tcps_rttupdated++; 1265 if (tp->t_srtt != 0) { 1266 /* 1267 * srtt is stored as fixed point with 3 bits after the 1268 * binary point (i.e., scaled by 8). The following magic 1269 * is equivalent to the smoothing algorithm in rfc793 with 1270 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1271 * point). Adjust t_rtt to origin 0. 1272 */ 1273 delta = tp->t_rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1274 if ((tp->t_srtt += delta) <= 0) 1275 tp->t_srtt = 1; 1276 /* 1277 * We accumulate a smoothed rtt variance (actually, a 1278 * smoothed mean difference), then set the retransmit 1279 * timer to smoothed rtt + 4 times the smoothed variance. 1280 * rttvar is stored as fixed point with 2 bits after the 1281 * binary point (scaled by 4). The following is 1282 * equivalent to rfc793 smoothing with an alpha of .75 1283 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1284 * rfc793's wired-in beta. 1285 */ 1286 if (delta < 0) 1287 delta = -delta; 1288 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1289 if ((tp->t_rttvar += delta) <= 0) 1290 tp->t_rttvar = 1; 1291 } else { 1292 /* 1293 * No rtt measurement yet - use the unsmoothed rtt. 1294 * Set the variance to half the rtt (so our first 1295 * retransmit happens at 3*rtt). 1296 */ 1297 tp->t_srtt = tp->t_rtt << TCP_RTT_SHIFT; 1298 tp->t_rttvar = tp->t_rtt << (TCP_RTTVAR_SHIFT - 1); 1299 } 1300 tp->t_rtt = 0; 1301 tp->t_rxtshift = 0; 1302 1303 /* 1304 * the retransmit should happen at rtt + 4 * rttvar. 1305 * Because of the way we do the smoothing, srtt and rttvar 1306 * will each average +1/2 tick of bias. When we compute 1307 * the retransmit timer, we want 1/2 tick of rounding and 1308 * 1 extra tick because of +-1/2 tick uncertainty in the 1309 * firing of the timer. The bias will give us exactly the 1310 * 1.5 tick we need. But, because the bias is 1311 * statistical, we have to test that we don't drop below 1312 * the minimum feasible timer (which is 2 ticks). 1313 */ 1314 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1315 tp->t_rttmin, TCPTV_REXMTMAX); 1316 1317 /* 1318 * We received an ack for a packet that wasn't retransmitted; 1319 * it is probably safe to discard any error indications we've 1320 * received recently. This isn't quite right, but close enough 1321 * for now (a route might have failed after we sent a segment, 1322 * and the return path might not be symmetrical). 1323 */ 1324 tp->t_softerror = 0; 1325 } 1326 1327 /* 1328 * Determine a reasonable value for maxseg size. 1329 * If the route is known, check route for mtu. 1330 * If none, use an mss that can be handled on the outgoing 1331 * interface without forcing IP to fragment; if bigger than 1332 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1333 * to utilize large mbufs. If no route is found, route has no mtu, 1334 * or the destination isn't local, use a default, hopefully conservative 1335 * size (usually 512 or the default IP max size, but no more than the mtu 1336 * of the interface), as we can't discover anything about intervening 1337 * gateways or networks. We also initialize the congestion/slow start 1338 * window to be a single segment if the destination isn't local. 1339 * While looking at the routing entry, we also initialize other path-dependent 1340 * parameters from pre-set or cached values in the routing entry. 1341 */ 1342 1343 tcp_mss(tp, offer) 1344 register struct tcpcb *tp; 1345 u_short offer; 1346 { 1347 struct route *ro; 1348 register struct rtentry *rt; 1349 struct ifnet *ifp; 1350 register int rtt, mss; 1351 u_long bufsize; 1352 struct inpcb *inp; 1353 struct socket *so; 1354 extern int tcp_mssdflt, tcp_rttdflt; 1355 1356 inp = tp->t_inpcb; 1357 ro = &inp->inp_route; 1358 1359 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1360 /* No route yet, so try to acquire one */ 1361 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1362 ro->ro_dst.sa_family = AF_INET; 1363 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1364 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1365 inp->inp_faddr; 1366 rtalloc(ro); 1367 } 1368 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1369 return (tcp_mssdflt); 1370 } 1371 ifp = rt->rt_ifp; 1372 so = inp->inp_socket; 1373 1374 #ifdef RTV_MTU /* if route characteristics exist ... */ 1375 /* 1376 * While we're here, check if there's an initial rtt 1377 * or rttvar. Convert from the route-table units 1378 * to scaled multiples of the slow timeout timer. 1379 */ 1380 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1381 /* 1382 * XXX the lock bit for MTU indicates that the value 1383 * is also a minimum value; this is subject to time. 1384 */ 1385 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1386 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1387 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1388 if (rt->rt_rmx.rmx_rttvar) 1389 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1390 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1391 else 1392 /* default variation is +- 1 rtt */ 1393 tp->t_rttvar = 1394 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1395 TCPT_RANGESET(tp->t_rxtcur, 1396 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1397 tp->t_rttmin, TCPTV_REXMTMAX); 1398 } 1399 /* 1400 * if there's an mtu associated with the route, use it 1401 */ 1402 if (rt->rt_rmx.rmx_mtu) 1403 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1404 else 1405 #endif /* RTV_MTU */ 1406 { 1407 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1408 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1409 if (mss > MCLBYTES) 1410 mss &= ~(MCLBYTES-1); 1411 #else 1412 if (mss > MCLBYTES) 1413 mss = mss / MCLBYTES * MCLBYTES; 1414 #endif 1415 if (!in_localaddr(inp->inp_faddr)) 1416 mss = min(mss, tcp_mssdflt); 1417 } 1418 /* 1419 * The current mss, t_maxseg, is initialized to the default value. 1420 * If we compute a smaller value, reduce the current mss. 1421 * If we compute a larger value, return it for use in sending 1422 * a max seg size option, but don't store it for use 1423 * unless we received an offer at least that large from peer. 1424 * However, do not accept offers under 32 bytes. 1425 */ 1426 if (offer) 1427 mss = min(mss, offer); 1428 mss = max(mss, 32); /* sanity */ 1429 if (mss < tp->t_maxseg || offer != 0) { 1430 /* 1431 * If there's a pipesize, change the socket buffer 1432 * to that size. Make the socket buffers an integral 1433 * number of mss units; if the mss is larger than 1434 * the socket buffer, decrease the mss. 1435 */ 1436 #ifdef RTV_SPIPE 1437 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1438 #endif 1439 bufsize = so->so_snd.sb_hiwat; 1440 if (bufsize < mss) 1441 mss = bufsize; 1442 else { 1443 bufsize = min(bufsize, SB_MAX) / mss * mss; 1444 (void) sbreserve(&so->so_snd, bufsize); 1445 } 1446 tp->t_maxseg = mss; 1447 1448 #ifdef RTV_RPIPE 1449 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1450 #endif 1451 bufsize = so->so_rcv.sb_hiwat; 1452 if (bufsize > mss) { 1453 bufsize = min(bufsize, SB_MAX) / mss * mss; 1454 (void) sbreserve(&so->so_rcv, bufsize); 1455 } 1456 } 1457 tp->snd_cwnd = mss; 1458 1459 #ifdef RTV_SSTHRESH 1460 if (rt->rt_rmx.rmx_ssthresh) { 1461 /* 1462 * There's some sort of gateway or interface 1463 * buffer limit on the path. Use this to set 1464 * the slow start threshhold, but set the 1465 * threshold to no less than 2*mss. 1466 */ 1467 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1468 } 1469 #endif /* RTV_MTU */ 1470 return (mss); 1471 } 1472