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 7.28 (Berkeley) 10/11/92 8 */ 9 10 #include <sys/param.h> 11 #include <sys/systm.h> 12 #include <sys/malloc.h> 13 #include <sys/mbuf.h> 14 #include <sys/protosw.h> 15 #include <sys/socket.h> 16 #include <sys/socketvar.h> 17 #include <sys/errno.h> 18 19 #include <net/if.h> 20 #include <net/route.h> 21 22 #include <netinet/in.h> 23 #include <netinet/in_systm.h> 24 #include <netinet/ip.h> 25 #include <netinet/in_pcb.h> 26 #include <netinet/ip_var.h> 27 #include <netinet/tcp.h> 28 #include <netinet/tcp_fsm.h> 29 #include <netinet/tcp_seq.h> 30 #include <netinet/tcp_timer.h> 31 #include <netinet/tcp_var.h> 32 #include <netinet/tcpip.h> 33 #include <netinet/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 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 481 laddr = inp->inp_laddr; 482 if (inp->inp_laddr.s_addr == INADDR_ANY) 483 inp->inp_laddr = ti->ti_dst; 484 if (in_pcbconnect(inp, am)) { 485 inp->inp_laddr = laddr; 486 (void) m_free(am); 487 goto drop; 488 } 489 (void) m_free(am); 490 tp->t_template = tcp_template(tp); 491 if (tp->t_template == 0) { 492 tp = tcp_drop(tp, ENOBUFS); 493 dropsocket = 0; /* socket is already gone */ 494 goto drop; 495 } 496 if (om) { 497 tcp_dooptions(tp, om, ti); 498 om = 0; 499 } 500 if (iss) 501 tp->iss = iss; 502 else 503 tp->iss = tcp_iss; 504 tcp_iss += TCP_ISSINCR/2; 505 tp->irs = ti->ti_seq; 506 tcp_sendseqinit(tp); 507 tcp_rcvseqinit(tp); 508 tp->t_flags |= TF_ACKNOW; 509 tp->t_state = TCPS_SYN_RECEIVED; 510 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 511 dropsocket = 0; /* committed to socket */ 512 tcpstat.tcps_accepts++; 513 goto trimthenstep6; 514 } 515 516 /* 517 * If the state is SYN_SENT: 518 * if seg contains an ACK, but not for our SYN, drop the input. 519 * if seg contains a RST, then drop the connection. 520 * if seg does not contain SYN, then drop it. 521 * Otherwise this is an acceptable SYN segment 522 * initialize tp->rcv_nxt and tp->irs 523 * if seg contains ack then advance tp->snd_una 524 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 525 * arrange for segment to be acked (eventually) 526 * continue processing rest of data/controls, beginning with URG 527 */ 528 case TCPS_SYN_SENT: 529 if ((tiflags & TH_ACK) && 530 (SEQ_LEQ(ti->ti_ack, tp->iss) || 531 SEQ_GT(ti->ti_ack, tp->snd_max))) 532 goto dropwithreset; 533 if (tiflags & TH_RST) { 534 if (tiflags & TH_ACK) 535 tp = tcp_drop(tp, ECONNREFUSED); 536 goto drop; 537 } 538 if ((tiflags & TH_SYN) == 0) 539 goto drop; 540 if (tiflags & TH_ACK) { 541 tp->snd_una = ti->ti_ack; 542 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 543 tp->snd_nxt = tp->snd_una; 544 } 545 tp->t_timer[TCPT_REXMT] = 0; 546 tp->irs = ti->ti_seq; 547 tcp_rcvseqinit(tp); 548 tp->t_flags |= TF_ACKNOW; 549 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 550 tcpstat.tcps_connects++; 551 soisconnected(so); 552 tp->t_state = TCPS_ESTABLISHED; 553 (void) tcp_reass(tp, (struct tcpiphdr *)0, 554 (struct mbuf *)0); 555 /* 556 * if we didn't have to retransmit the SYN, 557 * use its rtt as our initial srtt & rtt var. 558 */ 559 if (tp->t_rtt) 560 tcp_xmit_timer(tp); 561 } else 562 tp->t_state = TCPS_SYN_RECEIVED; 563 564 trimthenstep6: 565 /* 566 * Advance ti->ti_seq to correspond to first data byte. 567 * If data, trim to stay within window, 568 * dropping FIN if necessary. 569 */ 570 ti->ti_seq++; 571 if (ti->ti_len > tp->rcv_wnd) { 572 todrop = ti->ti_len - tp->rcv_wnd; 573 m_adj(m, -todrop); 574 ti->ti_len = tp->rcv_wnd; 575 tiflags &= ~TH_FIN; 576 tcpstat.tcps_rcvpackafterwin++; 577 tcpstat.tcps_rcvbyteafterwin += todrop; 578 } 579 tp->snd_wl1 = ti->ti_seq - 1; 580 tp->rcv_up = ti->ti_seq; 581 goto step6; 582 } 583 584 /* 585 * States other than LISTEN or SYN_SENT. 586 * First check that at least some bytes of segment are within 587 * receive window. If segment begins before rcv_nxt, 588 * drop leading data (and SYN); if nothing left, just ack. 589 */ 590 todrop = tp->rcv_nxt - ti->ti_seq; 591 if (todrop > 0) { 592 if (tiflags & TH_SYN) { 593 tiflags &= ~TH_SYN; 594 ti->ti_seq++; 595 if (ti->ti_urp > 1) 596 ti->ti_urp--; 597 else 598 tiflags &= ~TH_URG; 599 todrop--; 600 } 601 if (todrop > ti->ti_len || 602 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 603 tcpstat.tcps_rcvduppack++; 604 tcpstat.tcps_rcvdupbyte += ti->ti_len; 605 /* 606 * If segment is just one to the left of the window, 607 * check two special cases: 608 * 1. Don't toss RST in response to 4.2-style keepalive. 609 * 2. If the only thing to drop is a FIN, we can drop 610 * it, but check the ACK or we will get into FIN 611 * wars if our FINs crossed (both CLOSING). 612 * In either case, send ACK to resynchronize, 613 * but keep on processing for RST or ACK. 614 */ 615 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 616 #ifdef TCP_COMPAT_42 617 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 618 #endif 619 ) { 620 todrop = ti->ti_len; 621 tiflags &= ~TH_FIN; 622 tp->t_flags |= TF_ACKNOW; 623 } else 624 goto dropafterack; 625 } else { 626 tcpstat.tcps_rcvpartduppack++; 627 tcpstat.tcps_rcvpartdupbyte += todrop; 628 } 629 m_adj(m, todrop); 630 ti->ti_seq += todrop; 631 ti->ti_len -= todrop; 632 if (ti->ti_urp > todrop) 633 ti->ti_urp -= todrop; 634 else { 635 tiflags &= ~TH_URG; 636 ti->ti_urp = 0; 637 } 638 } 639 640 /* 641 * If new data are received on a connection after the 642 * user processes are gone, then RST the other end. 643 */ 644 if ((so->so_state & SS_NOFDREF) && 645 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 646 tp = tcp_close(tp); 647 tcpstat.tcps_rcvafterclose++; 648 goto dropwithreset; 649 } 650 651 /* 652 * If segment ends after window, drop trailing data 653 * (and PUSH and FIN); if nothing left, just ACK. 654 */ 655 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 656 if (todrop > 0) { 657 tcpstat.tcps_rcvpackafterwin++; 658 if (todrop >= ti->ti_len) { 659 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 660 /* 661 * If a new connection request is received 662 * while in TIME_WAIT, drop the old connection 663 * and start over if the sequence numbers 664 * are above the previous ones. 665 */ 666 if (tiflags & TH_SYN && 667 tp->t_state == TCPS_TIME_WAIT && 668 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 669 iss = tp->rcv_nxt + TCP_ISSINCR; 670 tp = tcp_close(tp); 671 goto findpcb; 672 } 673 /* 674 * If window is closed can only take segments at 675 * window edge, and have to drop data and PUSH from 676 * incoming segments. Continue processing, but 677 * remember to ack. Otherwise, drop segment 678 * and ack. 679 */ 680 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 681 tp->t_flags |= TF_ACKNOW; 682 tcpstat.tcps_rcvwinprobe++; 683 } else 684 goto dropafterack; 685 } else 686 tcpstat.tcps_rcvbyteafterwin += todrop; 687 m_adj(m, -todrop); 688 ti->ti_len -= todrop; 689 tiflags &= ~(TH_PUSH|TH_FIN); 690 } 691 692 /* 693 * If the RST bit is set examine the state: 694 * SYN_RECEIVED STATE: 695 * If passive open, return to LISTEN state. 696 * If active open, inform user that connection was refused. 697 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 698 * Inform user that connection was reset, and close tcb. 699 * CLOSING, LAST_ACK, TIME_WAIT STATES 700 * Close the tcb. 701 */ 702 if (tiflags&TH_RST) switch (tp->t_state) { 703 704 case TCPS_SYN_RECEIVED: 705 so->so_error = ECONNREFUSED; 706 goto close; 707 708 case TCPS_ESTABLISHED: 709 case TCPS_FIN_WAIT_1: 710 case TCPS_FIN_WAIT_2: 711 case TCPS_CLOSE_WAIT: 712 so->so_error = ECONNRESET; 713 close: 714 tp->t_state = TCPS_CLOSED; 715 tcpstat.tcps_drops++; 716 tp = tcp_close(tp); 717 goto drop; 718 719 case TCPS_CLOSING: 720 case TCPS_LAST_ACK: 721 case TCPS_TIME_WAIT: 722 tp = tcp_close(tp); 723 goto drop; 724 } 725 726 /* 727 * If a SYN is in the window, then this is an 728 * error and we send an RST and drop the connection. 729 */ 730 if (tiflags & TH_SYN) { 731 tp = tcp_drop(tp, ECONNRESET); 732 goto dropwithreset; 733 } 734 735 /* 736 * If the ACK bit is off we drop the segment and return. 737 */ 738 if ((tiflags & TH_ACK) == 0) 739 goto drop; 740 741 /* 742 * Ack processing. 743 */ 744 switch (tp->t_state) { 745 746 /* 747 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 748 * ESTABLISHED state and continue processing, otherwise 749 * send an RST. 750 */ 751 case TCPS_SYN_RECEIVED: 752 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 753 SEQ_GT(ti->ti_ack, tp->snd_max)) 754 goto dropwithreset; 755 tcpstat.tcps_connects++; 756 soisconnected(so); 757 tp->t_state = TCPS_ESTABLISHED; 758 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 759 tp->snd_wl1 = ti->ti_seq - 1; 760 /* fall into ... */ 761 762 /* 763 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 764 * ACKs. If the ack is in the range 765 * tp->snd_una < ti->ti_ack <= tp->snd_max 766 * then advance tp->snd_una to ti->ti_ack and drop 767 * data from the retransmission queue. If this ACK reflects 768 * more up to date window information we update our window information. 769 */ 770 case TCPS_ESTABLISHED: 771 case TCPS_FIN_WAIT_1: 772 case TCPS_FIN_WAIT_2: 773 case TCPS_CLOSE_WAIT: 774 case TCPS_CLOSING: 775 case TCPS_LAST_ACK: 776 case TCPS_TIME_WAIT: 777 778 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 779 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 780 tcpstat.tcps_rcvdupack++; 781 /* 782 * If we have outstanding data (other than 783 * a window probe), this is a completely 784 * duplicate ack (ie, window info didn't 785 * change), the ack is the biggest we've 786 * seen and we've seen exactly our rexmt 787 * threshhold of them, assume a packet 788 * has been dropped and retransmit it. 789 * Kludge snd_nxt & the congestion 790 * window so we send only this one 791 * packet. 792 * 793 * We know we're losing at the current 794 * window size so do congestion avoidance 795 * (set ssthresh to half the current window 796 * and pull our congestion window back to 797 * the new ssthresh). 798 * 799 * Dup acks mean that packets have left the 800 * network (they're now cached at the receiver) 801 * so bump cwnd by the amount in the receiver 802 * to keep a constant cwnd packets in the 803 * network. 804 */ 805 if (tp->t_timer[TCPT_REXMT] == 0 || 806 ti->ti_ack != tp->snd_una) 807 tp->t_dupacks = 0; 808 else if (++tp->t_dupacks == tcprexmtthresh) { 809 tcp_seq onxt = tp->snd_nxt; 810 u_int win = 811 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 812 tp->t_maxseg; 813 814 if (win < 2) 815 win = 2; 816 tp->snd_ssthresh = win * tp->t_maxseg; 817 tp->t_timer[TCPT_REXMT] = 0; 818 tp->t_rtt = 0; 819 tp->snd_nxt = ti->ti_ack; 820 tp->snd_cwnd = tp->t_maxseg; 821 (void) tcp_output(tp); 822 tp->snd_cwnd = tp->snd_ssthresh + 823 tp->t_maxseg * tp->t_dupacks; 824 if (SEQ_GT(onxt, tp->snd_nxt)) 825 tp->snd_nxt = onxt; 826 goto drop; 827 } else if (tp->t_dupacks > tcprexmtthresh) { 828 tp->snd_cwnd += tp->t_maxseg; 829 (void) tcp_output(tp); 830 goto drop; 831 } 832 } else 833 tp->t_dupacks = 0; 834 break; 835 } 836 /* 837 * If the congestion window was inflated to account 838 * for the other side's cached packets, retract it. 839 */ 840 if (tp->t_dupacks > tcprexmtthresh && 841 tp->snd_cwnd > tp->snd_ssthresh) 842 tp->snd_cwnd = tp->snd_ssthresh; 843 tp->t_dupacks = 0; 844 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 845 tcpstat.tcps_rcvacktoomuch++; 846 goto dropafterack; 847 } 848 acked = ti->ti_ack - tp->snd_una; 849 tcpstat.tcps_rcvackpack++; 850 tcpstat.tcps_rcvackbyte += acked; 851 852 /* 853 * If transmit timer is running and timed sequence 854 * number was acked, update smoothed round trip time. 855 * Since we now have an rtt measurement, cancel the 856 * timer backoff (cf., Phil Karn's retransmit alg.). 857 * Recompute the initial retransmit timer. 858 */ 859 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 860 tcp_xmit_timer(tp); 861 862 /* 863 * If all outstanding data is acked, stop retransmit 864 * timer and remember to restart (more output or persist). 865 * If there is more data to be acked, restart retransmit 866 * timer, using current (possibly backed-off) value. 867 */ 868 if (ti->ti_ack == tp->snd_max) { 869 tp->t_timer[TCPT_REXMT] = 0; 870 needoutput = 1; 871 } else if (tp->t_timer[TCPT_PERSIST] == 0) 872 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 873 /* 874 * When new data is acked, open the congestion window. 875 * If the window gives us less than ssthresh packets 876 * in flight, open exponentially (maxseg per packet). 877 * Otherwise open linearly: maxseg per window 878 * (maxseg^2 / cwnd per packet), plus a constant 879 * fraction of a packet (maxseg/8) to help larger windows 880 * open quickly enough. 881 */ 882 { 883 register u_int cw = tp->snd_cwnd; 884 register u_int incr = tp->t_maxseg; 885 886 if (cw > tp->snd_ssthresh) 887 incr = incr * incr / cw + incr / 8; 888 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN); 889 } 890 if (acked > so->so_snd.sb_cc) { 891 tp->snd_wnd -= so->so_snd.sb_cc; 892 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 893 ourfinisacked = 1; 894 } else { 895 sbdrop(&so->so_snd, acked); 896 tp->snd_wnd -= acked; 897 ourfinisacked = 0; 898 } 899 if (so->so_snd.sb_flags & SB_NOTIFY) 900 sowwakeup(so); 901 tp->snd_una = ti->ti_ack; 902 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 903 tp->snd_nxt = tp->snd_una; 904 905 switch (tp->t_state) { 906 907 /* 908 * In FIN_WAIT_1 STATE in addition to the processing 909 * for the ESTABLISHED state if our FIN is now acknowledged 910 * then enter FIN_WAIT_2. 911 */ 912 case TCPS_FIN_WAIT_1: 913 if (ourfinisacked) { 914 /* 915 * If we can't receive any more 916 * data, then closing user can proceed. 917 * Starting the timer is contrary to the 918 * specification, but if we don't get a FIN 919 * we'll hang forever. 920 */ 921 if (so->so_state & SS_CANTRCVMORE) { 922 soisdisconnected(so); 923 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 924 } 925 tp->t_state = TCPS_FIN_WAIT_2; 926 } 927 break; 928 929 /* 930 * In CLOSING STATE in addition to the processing for 931 * the ESTABLISHED state if the ACK acknowledges our FIN 932 * then enter the TIME-WAIT state, otherwise ignore 933 * the segment. 934 */ 935 case TCPS_CLOSING: 936 if (ourfinisacked) { 937 tp->t_state = TCPS_TIME_WAIT; 938 tcp_canceltimers(tp); 939 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 940 soisdisconnected(so); 941 } 942 break; 943 944 /* 945 * In LAST_ACK, we may still be waiting for data to drain 946 * and/or to be acked, as well as for the ack of our FIN. 947 * If our FIN is now acknowledged, delete the TCB, 948 * enter the closed state and return. 949 */ 950 case TCPS_LAST_ACK: 951 if (ourfinisacked) { 952 tp = tcp_close(tp); 953 goto drop; 954 } 955 break; 956 957 /* 958 * In TIME_WAIT state the only thing that should arrive 959 * is a retransmission of the remote FIN. Acknowledge 960 * it and restart the finack timer. 961 */ 962 case TCPS_TIME_WAIT: 963 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 964 goto dropafterack; 965 } 966 } 967 968 step6: 969 /* 970 * Update window information. 971 * Don't look at window if no ACK: TAC's send garbage on first SYN. 972 */ 973 if ((tiflags & TH_ACK) && 974 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 975 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 976 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 977 /* keep track of pure window updates */ 978 if (ti->ti_len == 0 && 979 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 980 tcpstat.tcps_rcvwinupd++; 981 tp->snd_wnd = ti->ti_win; 982 tp->snd_wl1 = ti->ti_seq; 983 tp->snd_wl2 = ti->ti_ack; 984 if (tp->snd_wnd > tp->max_sndwnd) 985 tp->max_sndwnd = tp->snd_wnd; 986 needoutput = 1; 987 } 988 989 /* 990 * Process segments with URG. 991 */ 992 if ((tiflags & TH_URG) && ti->ti_urp && 993 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 994 /* 995 * This is a kludge, but if we receive and accept 996 * random urgent pointers, we'll crash in 997 * soreceive. It's hard to imagine someone 998 * actually wanting to send this much urgent data. 999 */ 1000 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 1001 ti->ti_urp = 0; /* XXX */ 1002 tiflags &= ~TH_URG; /* XXX */ 1003 goto dodata; /* XXX */ 1004 } 1005 /* 1006 * If this segment advances the known urgent pointer, 1007 * then mark the data stream. This should not happen 1008 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1009 * a FIN has been received from the remote side. 1010 * In these states we ignore the URG. 1011 * 1012 * According to RFC961 (Assigned Protocols), 1013 * the urgent pointer points to the last octet 1014 * of urgent data. We continue, however, 1015 * to consider it to indicate the first octet 1016 * of data past the urgent section as the original 1017 * spec states (in one of two places). 1018 */ 1019 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1020 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1021 so->so_oobmark = so->so_rcv.sb_cc + 1022 (tp->rcv_up - tp->rcv_nxt) - 1; 1023 if (so->so_oobmark == 0) 1024 so->so_state |= SS_RCVATMARK; 1025 sohasoutofband(so); 1026 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1027 } 1028 /* 1029 * Remove out of band data so doesn't get presented to user. 1030 * This can happen independent of advancing the URG pointer, 1031 * but if two URG's are pending at once, some out-of-band 1032 * data may creep in... ick. 1033 */ 1034 if (ti->ti_urp <= ti->ti_len 1035 #ifdef SO_OOBINLINE 1036 && (so->so_options & SO_OOBINLINE) == 0 1037 #endif 1038 ) 1039 tcp_pulloutofband(so, ti, m); 1040 } else 1041 /* 1042 * If no out of band data is expected, 1043 * pull receive urgent pointer along 1044 * with the receive window. 1045 */ 1046 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1047 tp->rcv_up = tp->rcv_nxt; 1048 dodata: /* XXX */ 1049 1050 /* 1051 * Process the segment text, merging it into the TCP sequencing queue, 1052 * and arranging for acknowledgment of receipt if necessary. 1053 * This process logically involves adjusting tp->rcv_wnd as data 1054 * is presented to the user (this happens in tcp_usrreq.c, 1055 * case PRU_RCVD). If a FIN has already been received on this 1056 * connection then we just ignore the text. 1057 */ 1058 if ((ti->ti_len || (tiflags&TH_FIN)) && 1059 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1060 TCP_REASS(tp, ti, m, so, tiflags); 1061 /* 1062 * Note the amount of data that peer has sent into 1063 * our window, in order to estimate the sender's 1064 * buffer size. 1065 */ 1066 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1067 } else { 1068 m_freem(m); 1069 tiflags &= ~TH_FIN; 1070 } 1071 1072 /* 1073 * If FIN is received ACK the FIN and let the user know 1074 * that the connection is closing. 1075 */ 1076 if (tiflags & TH_FIN) { 1077 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1078 socantrcvmore(so); 1079 tp->t_flags |= TF_ACKNOW; 1080 tp->rcv_nxt++; 1081 } 1082 switch (tp->t_state) { 1083 1084 /* 1085 * In SYN_RECEIVED and ESTABLISHED STATES 1086 * enter the CLOSE_WAIT state. 1087 */ 1088 case TCPS_SYN_RECEIVED: 1089 case TCPS_ESTABLISHED: 1090 tp->t_state = TCPS_CLOSE_WAIT; 1091 break; 1092 1093 /* 1094 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1095 * enter the CLOSING state. 1096 */ 1097 case TCPS_FIN_WAIT_1: 1098 tp->t_state = TCPS_CLOSING; 1099 break; 1100 1101 /* 1102 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1103 * starting the time-wait timer, turning off the other 1104 * standard timers. 1105 */ 1106 case TCPS_FIN_WAIT_2: 1107 tp->t_state = TCPS_TIME_WAIT; 1108 tcp_canceltimers(tp); 1109 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1110 soisdisconnected(so); 1111 break; 1112 1113 /* 1114 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1115 */ 1116 case TCPS_TIME_WAIT: 1117 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1118 break; 1119 } 1120 } 1121 if (so->so_options & SO_DEBUG) 1122 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1123 1124 /* 1125 * Return any desired output. 1126 */ 1127 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1128 (void) tcp_output(tp); 1129 return; 1130 1131 dropafterack: 1132 /* 1133 * Generate an ACK dropping incoming segment if it occupies 1134 * sequence space, where the ACK reflects our state. 1135 */ 1136 if (tiflags & TH_RST) 1137 goto drop; 1138 m_freem(m); 1139 tp->t_flags |= TF_ACKNOW; 1140 (void) tcp_output(tp); 1141 return; 1142 1143 dropwithreset: 1144 if (om) { 1145 (void) m_free(om); 1146 om = 0; 1147 } 1148 /* 1149 * Generate a RST, dropping incoming segment. 1150 * Make ACK acceptable to originator of segment. 1151 * Don't bother to respond if destination was broadcast. 1152 */ 1153 if ((tiflags & TH_RST) || m->m_flags & M_BCAST) 1154 goto drop; 1155 if (tiflags & TH_ACK) 1156 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1157 else { 1158 if (tiflags & TH_SYN) 1159 ti->ti_len++; 1160 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1161 TH_RST|TH_ACK); 1162 } 1163 /* destroy temporarily created socket */ 1164 if (dropsocket) 1165 (void) soabort(so); 1166 return; 1167 1168 drop: 1169 if (om) 1170 (void) m_free(om); 1171 /* 1172 * Drop space held by incoming segment and return. 1173 */ 1174 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1175 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1176 m_freem(m); 1177 /* destroy temporarily created socket */ 1178 if (dropsocket) 1179 (void) soabort(so); 1180 return; 1181 } 1182 1183 tcp_dooptions(tp, om, ti) 1184 struct tcpcb *tp; 1185 struct mbuf *om; 1186 struct tcpiphdr *ti; 1187 { 1188 register u_char *cp; 1189 u_short mss; 1190 int opt, optlen, cnt; 1191 1192 cp = mtod(om, u_char *); 1193 cnt = om->m_len; 1194 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1195 opt = cp[0]; 1196 if (opt == TCPOPT_EOL) 1197 break; 1198 if (opt == TCPOPT_NOP) 1199 optlen = 1; 1200 else { 1201 optlen = cp[1]; 1202 if (optlen <= 0) 1203 break; 1204 } 1205 switch (opt) { 1206 1207 default: 1208 continue; 1209 1210 case TCPOPT_MAXSEG: 1211 if (optlen != 4) 1212 continue; 1213 if (!(ti->ti_flags & TH_SYN)) 1214 continue; 1215 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1216 NTOHS(mss); 1217 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1218 break; 1219 } 1220 } 1221 (void) m_free(om); 1222 } 1223 1224 /* 1225 * Pull out of band byte out of a segment so 1226 * it doesn't appear in the user's data queue. 1227 * It is still reflected in the segment length for 1228 * sequencing purposes. 1229 */ 1230 tcp_pulloutofband(so, ti, m) 1231 struct socket *so; 1232 struct tcpiphdr *ti; 1233 register struct mbuf *m; 1234 { 1235 int cnt = ti->ti_urp - 1; 1236 1237 while (cnt >= 0) { 1238 if (m->m_len > cnt) { 1239 char *cp = mtod(m, caddr_t) + cnt; 1240 struct tcpcb *tp = sototcpcb(so); 1241 1242 tp->t_iobc = *cp; 1243 tp->t_oobflags |= TCPOOB_HAVEDATA; 1244 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1245 m->m_len--; 1246 return; 1247 } 1248 cnt -= m->m_len; 1249 m = m->m_next; 1250 if (m == 0) 1251 break; 1252 } 1253 panic("tcp_pulloutofband"); 1254 } 1255 1256 /* 1257 * Collect new round-trip time estimate 1258 * and update averages and current timeout. 1259 */ 1260 tcp_xmit_timer(tp) 1261 register struct tcpcb *tp; 1262 { 1263 register short delta; 1264 1265 tcpstat.tcps_rttupdated++; 1266 if (tp->t_srtt != 0) { 1267 /* 1268 * srtt is stored as fixed point with 3 bits after the 1269 * binary point (i.e., scaled by 8). The following magic 1270 * is equivalent to the smoothing algorithm in rfc793 with 1271 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1272 * point). Adjust t_rtt to origin 0. 1273 */ 1274 delta = tp->t_rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1275 if ((tp->t_srtt += delta) <= 0) 1276 tp->t_srtt = 1; 1277 /* 1278 * We accumulate a smoothed rtt variance (actually, a 1279 * smoothed mean difference), then set the retransmit 1280 * timer to smoothed rtt + 4 times the smoothed variance. 1281 * rttvar is stored as fixed point with 2 bits after the 1282 * binary point (scaled by 4). The following is 1283 * equivalent to rfc793 smoothing with an alpha of .75 1284 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1285 * rfc793's wired-in beta. 1286 */ 1287 if (delta < 0) 1288 delta = -delta; 1289 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1290 if ((tp->t_rttvar += delta) <= 0) 1291 tp->t_rttvar = 1; 1292 } else { 1293 /* 1294 * No rtt measurement yet - use the unsmoothed rtt. 1295 * Set the variance to half the rtt (so our first 1296 * retransmit happens at 3*rtt). 1297 */ 1298 tp->t_srtt = tp->t_rtt << TCP_RTT_SHIFT; 1299 tp->t_rttvar = tp->t_rtt << (TCP_RTTVAR_SHIFT - 1); 1300 } 1301 tp->t_rtt = 0; 1302 tp->t_rxtshift = 0; 1303 1304 /* 1305 * the retransmit should happen at rtt + 4 * rttvar. 1306 * Because of the way we do the smoothing, srtt and rttvar 1307 * will each average +1/2 tick of bias. When we compute 1308 * the retransmit timer, we want 1/2 tick of rounding and 1309 * 1 extra tick because of +-1/2 tick uncertainty in the 1310 * firing of the timer. The bias will give us exactly the 1311 * 1.5 tick we need. But, because the bias is 1312 * statistical, we have to test that we don't drop below 1313 * the minimum feasible timer (which is 2 ticks). 1314 */ 1315 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1316 tp->t_rttmin, TCPTV_REXMTMAX); 1317 1318 /* 1319 * We received an ack for a packet that wasn't retransmitted; 1320 * it is probably safe to discard any error indications we've 1321 * received recently. This isn't quite right, but close enough 1322 * for now (a route might have failed after we sent a segment, 1323 * and the return path might not be symmetrical). 1324 */ 1325 tp->t_softerror = 0; 1326 } 1327 1328 /* 1329 * Determine a reasonable value for maxseg size. 1330 * If the route is known, check route for mtu. 1331 * If none, use an mss that can be handled on the outgoing 1332 * interface without forcing IP to fragment; if bigger than 1333 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1334 * to utilize large mbufs. If no route is found, route has no mtu, 1335 * or the destination isn't local, use a default, hopefully conservative 1336 * size (usually 512 or the default IP max size, but no more than the mtu 1337 * of the interface), as we can't discover anything about intervening 1338 * gateways or networks. We also initialize the congestion/slow start 1339 * window to be a single segment if the destination isn't local. 1340 * While looking at the routing entry, we also initialize other path-dependent 1341 * parameters from pre-set or cached values in the routing entry. 1342 */ 1343 1344 tcp_mss(tp, offer) 1345 register struct tcpcb *tp; 1346 u_short offer; 1347 { 1348 struct route *ro; 1349 register struct rtentry *rt; 1350 struct ifnet *ifp; 1351 register int rtt, mss; 1352 u_long bufsize; 1353 struct inpcb *inp; 1354 struct socket *so; 1355 extern int tcp_mssdflt, tcp_rttdflt; 1356 1357 inp = tp->t_inpcb; 1358 ro = &inp->inp_route; 1359 1360 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1361 /* No route yet, so try to acquire one */ 1362 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1363 ro->ro_dst.sa_family = AF_INET; 1364 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1365 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1366 inp->inp_faddr; 1367 rtalloc(ro); 1368 } 1369 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1370 return (tcp_mssdflt); 1371 } 1372 ifp = rt->rt_ifp; 1373 so = inp->inp_socket; 1374 1375 #ifdef RTV_MTU /* if route characteristics exist ... */ 1376 /* 1377 * While we're here, check if there's an initial rtt 1378 * or rttvar. Convert from the route-table units 1379 * to scaled multiples of the slow timeout timer. 1380 */ 1381 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1382 /* 1383 * XXX the lock bit for MTU indicates that the value 1384 * is also a minimum value; this is subject to time. 1385 */ 1386 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1387 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1388 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1389 if (rt->rt_rmx.rmx_rttvar) 1390 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1391 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1392 else 1393 /* default variation is +- 1 rtt */ 1394 tp->t_rttvar = 1395 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1396 TCPT_RANGESET(tp->t_rxtcur, 1397 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1398 tp->t_rttmin, TCPTV_REXMTMAX); 1399 } 1400 /* 1401 * if there's an mtu associated with the route, use it 1402 */ 1403 if (rt->rt_rmx.rmx_mtu) 1404 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1405 else 1406 #endif /* RTV_MTU */ 1407 { 1408 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1409 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1410 if (mss > MCLBYTES) 1411 mss &= ~(MCLBYTES-1); 1412 #else 1413 if (mss > MCLBYTES) 1414 mss = mss / MCLBYTES * MCLBYTES; 1415 #endif 1416 if (!in_localaddr(inp->inp_faddr)) 1417 mss = min(mss, tcp_mssdflt); 1418 } 1419 /* 1420 * The current mss, t_maxseg, is initialized to the default value. 1421 * If we compute a smaller value, reduce the current mss. 1422 * If we compute a larger value, return it for use in sending 1423 * a max seg size option, but don't store it for use 1424 * unless we received an offer at least that large from peer. 1425 * However, do not accept offers under 32 bytes. 1426 */ 1427 if (offer) 1428 mss = min(mss, offer); 1429 mss = max(mss, 32); /* sanity */ 1430 if (mss < tp->t_maxseg || offer != 0) { 1431 /* 1432 * If there's a pipesize, change the socket buffer 1433 * to that size. Make the socket buffers an integral 1434 * number of mss units; if the mss is larger than 1435 * the socket buffer, decrease the mss. 1436 */ 1437 #ifdef RTV_SPIPE 1438 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1439 #endif 1440 bufsize = so->so_snd.sb_hiwat; 1441 if (bufsize < mss) 1442 mss = bufsize; 1443 else { 1444 bufsize = min(bufsize, SB_MAX) / mss * mss; 1445 (void) sbreserve(&so->so_snd, bufsize); 1446 } 1447 tp->t_maxseg = mss; 1448 1449 #ifdef RTV_RPIPE 1450 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1451 #endif 1452 bufsize = so->so_rcv.sb_hiwat; 1453 if (bufsize > mss) { 1454 bufsize = min(bufsize, SB_MAX) / mss * mss; 1455 (void) sbreserve(&so->so_rcv, bufsize); 1456 } 1457 } 1458 tp->snd_cwnd = mss; 1459 1460 #ifdef RTV_SSTHRESH 1461 if (rt->rt_rmx.rmx_ssthresh) { 1462 /* 1463 * There's some sort of gateway or interface 1464 * buffer limit on the path. Use this to set 1465 * the slow start threshhold, but set the 1466 * threshold to no less than 2*mss. 1467 */ 1468 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1469 } 1470 #endif /* RTV_MTU */ 1471 return (mss); 1472 } 1473