1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990 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 7.23 (Berkeley) 06/28/90 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_systm.h" 34 #include "ip.h" 35 #include "in_pcb.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 #define VAN 46 int tcprexmtthresh = 3; 47 int tcppredack; /* XXX debugging: times hdr predict ok for acks */ 48 int tcppreddat; /* XXX # times header prediction ok for data packets */ 49 int tcppcbcachemiss; 50 struct tcpiphdr tcp_saveti; 51 struct inpcb *tcp_last_inpcb = &tcb; 52 53 struct tcpcb *tcp_newtcpcb(); 54 55 /* 56 * Insert segment ti into reassembly queue of tcp with 57 * control block tp. Return TH_FIN if reassembly now includes 58 * a segment with FIN. The macro form does the common case inline 59 * (segment is the next to be received on an established connection, 60 * and the queue is empty), avoiding linkage into and removal 61 * from the queue and repetition of various conversions. 62 * Set DELACK for segments received in order, but ack immediately 63 * when segments are out of order (so fast retransmit can work). 64 */ 65 #define TCP_REASS(tp, ti, m, so, flags) { \ 66 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 67 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 68 (tp)->t_state == TCPS_ESTABLISHED) { \ 69 tp->t_flags |= TF_DELACK; \ 70 (tp)->rcv_nxt += (ti)->ti_len; \ 71 flags = (ti)->ti_flags & TH_FIN; \ 72 tcpstat.tcps_rcvpack++;\ 73 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 74 sbappend(&(so)->so_rcv, (m)); \ 75 sorwakeup(so); \ 76 } else { \ 77 (flags) = tcp_reass((tp), (ti), (m)); \ 78 tp->t_flags |= TF_ACKNOW; \ 79 } \ 80 } 81 82 tcp_reass(tp, ti, m) 83 register struct tcpcb *tp; 84 register struct tcpiphdr *ti; 85 struct mbuf *m; 86 { 87 register struct tcpiphdr *q; 88 struct socket *so = tp->t_inpcb->inp_socket; 89 int flags; 90 91 /* 92 * Call with ti==0 after become established to 93 * force pre-ESTABLISHED data up to user socket. 94 */ 95 if (ti == 0) 96 goto present; 97 98 /* 99 * Find a segment which begins after this one does. 100 */ 101 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 102 q = (struct tcpiphdr *)q->ti_next) 103 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 104 break; 105 106 /* 107 * If there is a preceding segment, it may provide some of 108 * our data already. If so, drop the data from the incoming 109 * segment. If it provides all of our data, drop us. 110 */ 111 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 112 register int i; 113 q = (struct tcpiphdr *)q->ti_prev; 114 /* conversion to int (in i) handles seq wraparound */ 115 i = q->ti_seq + q->ti_len - ti->ti_seq; 116 if (i > 0) { 117 if (i >= ti->ti_len) { 118 tcpstat.tcps_rcvduppack++; 119 tcpstat.tcps_rcvdupbyte += ti->ti_len; 120 m_freem(m); 121 return (0); 122 } 123 m_adj(m, i); 124 ti->ti_len -= i; 125 ti->ti_seq += i; 126 } 127 q = (struct tcpiphdr *)(q->ti_next); 128 } 129 tcpstat.tcps_rcvoopack++; 130 tcpstat.tcps_rcvoobyte += ti->ti_len; 131 REASS_MBUF(ti) = m; /* XXX */ 132 133 /* 134 * While we overlap succeeding segments trim them or, 135 * if they are completely covered, dequeue them. 136 */ 137 while (q != (struct tcpiphdr *)tp) { 138 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 139 if (i <= 0) 140 break; 141 if (i < q->ti_len) { 142 q->ti_seq += i; 143 q->ti_len -= i; 144 m_adj(REASS_MBUF(q), i); 145 break; 146 } 147 q = (struct tcpiphdr *)q->ti_next; 148 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 149 remque(q->ti_prev); 150 m_freem(m); 151 } 152 153 /* 154 * Stick new segment in its place. 155 */ 156 insque(ti, q->ti_prev); 157 158 present: 159 /* 160 * Present data to user, advancing rcv_nxt through 161 * completed sequence space. 162 */ 163 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 164 return (0); 165 ti = tp->seg_next; 166 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 167 return (0); 168 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 169 return (0); 170 do { 171 tp->rcv_nxt += ti->ti_len; 172 flags = ti->ti_flags & TH_FIN; 173 remque(ti); 174 m = REASS_MBUF(ti); 175 ti = (struct tcpiphdr *)ti->ti_next; 176 if (so->so_state & SS_CANTRCVMORE) 177 m_freem(m); 178 else 179 sbappend(&so->so_rcv, m); 180 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 181 sorwakeup(so); 182 return (flags); 183 } 184 185 /* 186 * TCP input routine, follows pages 65-76 of the 187 * protocol specification dated September, 1981 very closely. 188 */ 189 tcp_input(m, iphlen) 190 register struct mbuf *m; 191 int iphlen; 192 { 193 register struct tcpiphdr *ti; 194 register struct inpcb *inp; 195 struct mbuf *om = 0; 196 int len, tlen, off; 197 register struct tcpcb *tp = 0; 198 register int tiflags; 199 struct socket *so; 200 int todrop, acked, ourfinisacked, needoutput = 0; 201 short ostate; 202 struct in_addr laddr; 203 int dropsocket = 0; 204 int iss = 0; 205 206 tcpstat.tcps_rcvtotal++; 207 /* 208 * Get IP and TCP header together in first mbuf. 209 * Note: IP leaves IP header in first mbuf. 210 */ 211 ti = mtod(m, struct tcpiphdr *); 212 if (iphlen > sizeof (struct ip)) 213 ip_stripoptions(m, (struct mbuf *)0); 214 if (m->m_len < sizeof (struct tcpiphdr)) { 215 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 216 tcpstat.tcps_rcvshort++; 217 return; 218 } 219 ti = mtod(m, struct tcpiphdr *); 220 } 221 222 /* 223 * Checksum extended TCP header and data. 224 */ 225 tlen = ((struct ip *)ti)->ip_len; 226 len = sizeof (struct ip) + tlen; 227 ti->ti_next = ti->ti_prev = 0; 228 ti->ti_x1 = 0; 229 ti->ti_len = (u_short)tlen; 230 HTONS(ti->ti_len); 231 if (ti->ti_sum = in_cksum(m, len)) { 232 tcpstat.tcps_rcvbadsum++; 233 goto drop; 234 } 235 236 /* 237 * Check that TCP offset makes sense, 238 * pull out TCP options and adjust length. XXX 239 */ 240 off = ti->ti_off << 2; 241 if (off < sizeof (struct tcphdr) || off > tlen) { 242 tcpstat.tcps_rcvbadoff++; 243 goto drop; 244 } 245 tlen -= off; 246 ti->ti_len = tlen; 247 if (off > sizeof (struct tcphdr)) { 248 if (m->m_len < sizeof(struct ip) + off) { 249 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 250 tcpstat.tcps_rcvshort++; 251 return; 252 } 253 ti = mtod(m, struct tcpiphdr *); 254 } 255 om = m_get(M_DONTWAIT, MT_DATA); 256 if (om == 0) 257 goto drop; 258 om->m_len = off - sizeof (struct tcphdr); 259 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 260 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 261 m->m_len -= om->m_len; 262 m->m_pkthdr.len -= om->m_len; 263 bcopy(op+om->m_len, op, 264 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 265 } 266 } 267 tiflags = ti->ti_flags; 268 269 /* 270 * Convert TCP protocol specific fields to host format. 271 */ 272 NTOHL(ti->ti_seq); 273 NTOHL(ti->ti_ack); 274 NTOHS(ti->ti_win); 275 NTOHS(ti->ti_urp); 276 277 /* 278 * Locate pcb for segment. 279 */ 280 findpcb: 281 inp = tcp_last_inpcb; 282 if (inp->inp_lport != ti->ti_dport || 283 inp->inp_fport != ti->ti_sport || 284 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 285 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 286 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 287 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 288 if (inp) 289 tcp_last_inpcb = inp; 290 ++tcppcbcachemiss; 291 } 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|SO_ACCEPTCONN)) { 308 if (so->so_options & SO_DEBUG) { 309 ostate = tp->t_state; 310 tcp_saveti = *ti; 311 } 312 if (so->so_options & SO_ACCEPTCONN) { 313 so = sonewconn(so, 0); 314 if (so == 0) 315 goto drop; 316 /* 317 * This is ugly, but .... 318 * 319 * Mark socket as temporary until we're 320 * committed to keeping it. The code at 321 * ``drop'' and ``dropwithreset'' check the 322 * flag dropsocket to see if the temporary 323 * socket created here should be discarded. 324 * We mark the socket as discardable until 325 * we're committed to it below in TCPS_LISTEN. 326 */ 327 dropsocket++; 328 inp = (struct inpcb *)so->so_pcb; 329 inp->inp_laddr = ti->ti_dst; 330 inp->inp_lport = ti->ti_dport; 331 #if BSD>=43 332 inp->inp_options = ip_srcroute(); 333 #endif 334 tp = intotcpcb(inp); 335 tp->t_state = TCPS_LISTEN; 336 } 337 } 338 339 /* 340 * Segment received on connection. 341 * Reset idle time and keep-alive timer. 342 */ 343 tp->t_idle = 0; 344 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 345 346 #ifndef VAN 347 /* 348 * Process options if not in LISTEN state, 349 * else do it below (after getting remote address). 350 */ 351 if (om && tp->t_state != TCPS_LISTEN) { 352 tcp_dooptions(tp, om, ti); 353 om = 0; 354 } 355 #endif VAN 356 /* 357 * Header prediction: check for the two common cases 358 * of a uni-directional data xfer. If the packet has 359 * no control flags, is in-sequence, the window didn't 360 * change and we're not retransmitting, it's a 361 * candidate. If the length is zero and the ack moved 362 * forward, we're the sender side of the xfer. Just 363 * free the data acked & wake any higher level process 364 * that was blocked waiting for space. If the length 365 * is non-zero and the ack didn't move, we're the 366 * receiver side. If we're getting packets in-order 367 * (the reassembly queue is empty), add the data to 368 * the socket buffer and note that we need a delayed ack. 369 */ 370 if (tp->t_state == TCPS_ESTABLISHED && 371 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 372 ti->ti_seq == tp->rcv_nxt && 373 ti->ti_win && ti->ti_win == tp->snd_wnd && 374 tp->snd_nxt == tp->snd_max) { 375 if (ti->ti_len == 0) { 376 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 377 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 378 tp->snd_cwnd >= tp->snd_wnd) { 379 /* 380 * this is a pure ack for outstanding data. 381 */ 382 ++tcppredack; 383 if (tp->t_rtt && SEQ_GT(ti->ti_ack,tp->t_rtseq)) 384 tcp_xmit_timer(tp); 385 acked = ti->ti_ack - tp->snd_una; 386 tcpstat.tcps_rcvackpack++; 387 tcpstat.tcps_rcvackbyte += acked; 388 sbdrop(&so->so_snd, acked); 389 tp->snd_una = ti->ti_ack; 390 m_freem(m); 391 392 /* 393 * If all outstanding data are acked, stop 394 * retransmit timer, otherwise restart timer 395 * using current (possibly backed-off) value. 396 * If process is waiting for space, 397 * wakeup/selwakeup/signal. If data 398 * are ready to send, let tcp_output 399 * decide between more output or persist. 400 */ 401 if (tp->snd_una == tp->snd_max) 402 tp->t_timer[TCPT_REXMT] = 0; 403 else if (tp->t_timer[TCPT_PERSIST] == 0) 404 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 405 406 if (so->so_snd.sb_flags & SB_NOTIFY) 407 sowwakeup(so); 408 if (so->so_snd.sb_cc) 409 (void) tcp_output(tp); 410 return; 411 } 412 } else if (ti->ti_ack == tp->snd_una && 413 tp->seg_next == (struct tcpiphdr *)tp && 414 ti->ti_len <= sbspace(&so->so_rcv)) { 415 /* 416 * this is a pure, in-sequence data packet 417 * with nothing on the reassembly queue and 418 * we have enough buffer space to take it. 419 */ 420 ++tcppreddat; 421 tp->rcv_nxt += ti->ti_len; 422 tcpstat.tcps_rcvpack++; 423 tcpstat.tcps_rcvbyte += ti->ti_len; 424 /* 425 * Drop TCP and IP headers then add data 426 * to socket buffer 427 */ 428 m->m_data += sizeof(struct tcpiphdr); 429 m->m_len -= sizeof(struct tcpiphdr); 430 sbappend(&so->so_rcv, m); 431 sorwakeup(so); 432 tp->t_flags |= TF_DELACK; 433 return; 434 } 435 } 436 437 /* 438 * Drop TCP and IP headers; TCP options were dropped above. 439 */ 440 m->m_data += sizeof(struct tcpiphdr); 441 m->m_len -= sizeof(struct tcpiphdr); 442 443 /* 444 * Calculate amount of space in receive window, 445 * and then do TCP input processing. 446 * Receive window is amount of space in rcv queue, 447 * but not less than advertised window. 448 */ 449 { int win; 450 451 win = sbspace(&so->so_rcv); 452 if (win < 0) 453 win = 0; 454 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 455 } 456 457 switch (tp->t_state) { 458 459 /* 460 * If the state is LISTEN then ignore segment if it contains an RST. 461 * If the segment contains an ACK then it is bad and send a RST. 462 * If it does not contain a SYN then it is not interesting; drop it. 463 * Don't bother responding if the destination was a broadcast. 464 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 465 * tp->iss, and send a segment: 466 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 467 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 468 * Fill in remote peer address fields if not previously specified. 469 * Enter SYN_RECEIVED state, and process any other fields of this 470 * segment in this state. 471 */ 472 case TCPS_LISTEN: { 473 struct mbuf *am; 474 register struct sockaddr_in *sin; 475 476 if (tiflags & TH_RST) 477 goto drop; 478 if (tiflags & TH_ACK) 479 goto dropwithreset; 480 if ((tiflags & TH_SYN) == 0) 481 goto drop; 482 if (m->m_flags & M_BCAST) 483 goto drop; 484 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 485 if (am == NULL) 486 goto drop; 487 am->m_len = sizeof (struct sockaddr_in); 488 sin = mtod(am, struct sockaddr_in *); 489 sin->sin_family = AF_INET; 490 sin->sin_len = sizeof(*sin); 491 sin->sin_addr = ti->ti_src; 492 sin->sin_port = ti->ti_sport; 493 laddr = inp->inp_laddr; 494 if (inp->inp_laddr.s_addr == INADDR_ANY) 495 inp->inp_laddr = ti->ti_dst; 496 if (in_pcbconnect(inp, am)) { 497 inp->inp_laddr = laddr; 498 (void) m_free(am); 499 goto drop; 500 } 501 (void) m_free(am); 502 tp->t_template = tcp_template(tp); 503 if (tp->t_template == 0) { 504 tp = tcp_drop(tp, ENOBUFS); 505 dropsocket = 0; /* socket is already gone */ 506 goto drop; 507 } 508 if (om) { 509 tcp_dooptions(tp, om, ti); 510 om = 0; 511 } 512 if (iss) 513 tp->iss = iss; 514 else 515 tp->iss = tcp_iss; 516 tcp_iss += TCP_ISSINCR/2; 517 tp->irs = ti->ti_seq; 518 tcp_sendseqinit(tp); 519 tcp_rcvseqinit(tp); 520 tp->t_flags |= TF_ACKNOW; 521 tp->t_state = TCPS_SYN_RECEIVED; 522 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 523 dropsocket = 0; /* committed to socket */ 524 tcpstat.tcps_accepts++; 525 goto trimthenstep6; 526 } 527 528 /* 529 * If the state is SYN_SENT: 530 * if seg contains an ACK, but not for our SYN, drop the input. 531 * if seg contains a RST, then drop the connection. 532 * if seg does not contain SYN, then drop it. 533 * Otherwise this is an acceptable SYN segment 534 * initialize tp->rcv_nxt and tp->irs 535 * if seg contains ack then advance tp->snd_una 536 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 537 * arrange for segment to be acked (eventually) 538 * continue processing rest of data/controls, beginning with URG 539 */ 540 case TCPS_SYN_SENT: 541 #ifdef VAN 542 if (om) { 543 tcp_dooptions(tp, om, ti); 544 om = 0; 545 } 546 #endif VAN 547 if ((tiflags & TH_ACK) && 548 (SEQ_LEQ(ti->ti_ack, tp->iss) || 549 SEQ_GT(ti->ti_ack, tp->snd_max))) 550 goto dropwithreset; 551 if (tiflags & TH_RST) { 552 if (tiflags & TH_ACK) 553 tp = tcp_drop(tp, ECONNREFUSED); 554 goto drop; 555 } 556 if ((tiflags & TH_SYN) == 0) 557 goto drop; 558 if (tiflags & TH_ACK) { 559 tp->snd_una = ti->ti_ack; 560 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 561 tp->snd_nxt = tp->snd_una; 562 } 563 tp->t_timer[TCPT_REXMT] = 0; 564 tp->irs = ti->ti_seq; 565 tcp_rcvseqinit(tp); 566 tp->t_flags |= TF_ACKNOW; 567 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 568 tcpstat.tcps_connects++; 569 soisconnected(so); 570 tp->t_state = TCPS_ESTABLISHED; 571 (void) tcp_reass(tp, (struct tcpiphdr *)0, 572 (struct mbuf *)0); 573 /* 574 * if we didn't have to retransmit the SYN, 575 * use its rtt as our initial srtt & rtt var. 576 */ 577 if (tp->t_rtt) 578 tcp_xmit_timer(tp); 579 } else 580 tp->t_state = TCPS_SYN_RECEIVED; 581 582 trimthenstep6: 583 /* 584 * Advance ti->ti_seq to correspond to first data byte. 585 * If data, trim to stay within window, 586 * dropping FIN if necessary. 587 */ 588 ti->ti_seq++; 589 if (ti->ti_len > tp->rcv_wnd) { 590 todrop = ti->ti_len - tp->rcv_wnd; 591 m_adj(m, -todrop); 592 ti->ti_len = tp->rcv_wnd; 593 tiflags &= ~TH_FIN; 594 tcpstat.tcps_rcvpackafterwin++; 595 tcpstat.tcps_rcvbyteafterwin += todrop; 596 } 597 tp->snd_wl1 = ti->ti_seq - 1; 598 tp->rcv_up = ti->ti_seq; 599 goto step6; 600 #ifdef VAN 601 602 default: 603 if (om) { 604 tcp_dooptions(tp, om, ti); 605 om = 0; 606 } 607 #endif VAN 608 } 609 610 /* 611 * States other than LISTEN or SYN_SENT. 612 * First check that at least some bytes of segment are within 613 * receive window. If segment begins before rcv_nxt, 614 * drop leading data (and SYN); if nothing left, just ack. 615 */ 616 todrop = tp->rcv_nxt - ti->ti_seq; 617 if (todrop > 0) { 618 if (tiflags & TH_SYN) { 619 tiflags &= ~TH_SYN; 620 ti->ti_seq++; 621 if (ti->ti_urp > 1) 622 ti->ti_urp--; 623 else 624 tiflags &= ~TH_URG; 625 todrop--; 626 } 627 if (todrop > ti->ti_len || 628 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 629 tcpstat.tcps_rcvduppack++; 630 tcpstat.tcps_rcvdupbyte += ti->ti_len; 631 /* 632 * If segment is just one to the left of the window, 633 * check two special cases: 634 * 1. Don't toss RST in response to 4.2-style keepalive. 635 * 2. If the only thing to drop is a FIN, we can drop 636 * it, but check the ACK or we will get into FIN 637 * wars if our FINs crossed (both CLOSING). 638 * In either case, send ACK to resynchronize, 639 * but keep on processing for RST or ACK. 640 */ 641 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 642 #ifdef TCP_COMPAT_42 643 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 644 #endif 645 ) { 646 todrop = ti->ti_len; 647 tiflags &= ~TH_FIN; 648 tp->t_flags |= TF_ACKNOW; 649 } else 650 goto dropafterack; 651 } else { 652 tcpstat.tcps_rcvpartduppack++; 653 tcpstat.tcps_rcvpartdupbyte += todrop; 654 } 655 m_adj(m, todrop); 656 ti->ti_seq += todrop; 657 ti->ti_len -= todrop; 658 if (ti->ti_urp > todrop) 659 ti->ti_urp -= todrop; 660 else { 661 tiflags &= ~TH_URG; 662 ti->ti_urp = 0; 663 } 664 } 665 666 /* 667 * If new data are received on a connection after the 668 * user processes are gone, then RST the other end. 669 */ 670 if ((so->so_state & SS_NOFDREF) && 671 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 672 tp = tcp_close(tp); 673 tcpstat.tcps_rcvafterclose++; 674 goto dropwithreset; 675 } 676 677 /* 678 * If segment ends after window, drop trailing data 679 * (and PUSH and FIN); if nothing left, just ACK. 680 */ 681 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 682 if (todrop > 0) { 683 tcpstat.tcps_rcvpackafterwin++; 684 if (todrop >= ti->ti_len) { 685 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 686 /* 687 * If a new connection request is received 688 * while in TIME_WAIT, drop the old connection 689 * and start over if the sequence numbers 690 * are above the previous ones. 691 */ 692 if (tiflags & TH_SYN && 693 tp->t_state == TCPS_TIME_WAIT && 694 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 695 iss = tp->rcv_nxt + TCP_ISSINCR; 696 tp = tcp_close(tp); 697 goto findpcb; 698 } 699 /* 700 * If window is closed can only take segments at 701 * window edge, and have to drop data and PUSH from 702 * incoming segments. Continue processing, but 703 * remember to ack. Otherwise, drop segment 704 * and ack. 705 */ 706 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 707 tp->t_flags |= TF_ACKNOW; 708 tcpstat.tcps_rcvwinprobe++; 709 } else 710 goto dropafterack; 711 } else 712 tcpstat.tcps_rcvbyteafterwin += todrop; 713 m_adj(m, -todrop); 714 ti->ti_len -= todrop; 715 tiflags &= ~(TH_PUSH|TH_FIN); 716 } 717 718 /* 719 * If the RST bit is set examine the state: 720 * SYN_RECEIVED STATE: 721 * If passive open, return to LISTEN state. 722 * If active open, inform user that connection was refused. 723 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 724 * Inform user that connection was reset, and close tcb. 725 * CLOSING, LAST_ACK, TIME_WAIT STATES 726 * Close the tcb. 727 */ 728 if (tiflags&TH_RST) switch (tp->t_state) { 729 730 case TCPS_SYN_RECEIVED: 731 so->so_error = ECONNREFUSED; 732 goto close; 733 734 case TCPS_ESTABLISHED: 735 case TCPS_FIN_WAIT_1: 736 case TCPS_FIN_WAIT_2: 737 case TCPS_CLOSE_WAIT: 738 so->so_error = ECONNRESET; 739 close: 740 tp->t_state = TCPS_CLOSED; 741 tcpstat.tcps_drops++; 742 tp = tcp_close(tp); 743 goto drop; 744 745 case TCPS_CLOSING: 746 case TCPS_LAST_ACK: 747 case TCPS_TIME_WAIT: 748 tp = tcp_close(tp); 749 goto drop; 750 } 751 752 /* 753 * If a SYN is in the window, then this is an 754 * error and we send an RST and drop the connection. 755 */ 756 if (tiflags & TH_SYN) { 757 tp = tcp_drop(tp, ECONNRESET); 758 goto dropwithreset; 759 } 760 761 /* 762 * If the ACK bit is off we drop the segment and return. 763 */ 764 if ((tiflags & TH_ACK) == 0) 765 goto drop; 766 767 /* 768 * Ack processing. 769 */ 770 switch (tp->t_state) { 771 772 /* 773 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 774 * ESTABLISHED state and continue processing, otherwise 775 * send an RST. 776 */ 777 case TCPS_SYN_RECEIVED: 778 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 779 SEQ_GT(ti->ti_ack, tp->snd_max)) 780 goto dropwithreset; 781 tcpstat.tcps_connects++; 782 soisconnected(so); 783 tp->t_state = TCPS_ESTABLISHED; 784 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 785 tp->snd_wl1 = ti->ti_seq - 1; 786 /* fall into ... */ 787 788 /* 789 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 790 * ACKs. If the ack is in the range 791 * tp->snd_una < ti->ti_ack <= tp->snd_max 792 * then advance tp->snd_una to ti->ti_ack and drop 793 * data from the retransmission queue. If this ACK reflects 794 * more up to date window information we update our window information. 795 */ 796 case TCPS_ESTABLISHED: 797 case TCPS_FIN_WAIT_1: 798 case TCPS_FIN_WAIT_2: 799 case TCPS_CLOSE_WAIT: 800 case TCPS_CLOSING: 801 case TCPS_LAST_ACK: 802 case TCPS_TIME_WAIT: 803 804 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 805 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 806 tcpstat.tcps_rcvdupack++; 807 /* 808 * If we have outstanding data (other than 809 * a window probe), this is a completely 810 * duplicate ack (ie, window info didn't 811 * change), the ack is the biggest we've 812 * seen and we've seen exactly our rexmt 813 * threshhold of them, assume a packet 814 * has been dropped and retransmit it. 815 * Kludge snd_nxt & the congestion 816 * window so we send only this one 817 * packet. 818 * 819 * We know we're losing at the current 820 * window size so do congestion avoidance 821 * (set ssthresh to half the current window 822 * and pull our congestion window back to 823 * the new ssthresh). 824 * 825 * Dup acks mean that packets have left the 826 * network (they're now cached at the receiver) 827 * so bump cwnd by the amount in the receiver 828 * to keep a constant cwnd packets in the 829 * network. 830 */ 831 if (tp->t_timer[TCPT_REXMT] == 0 || 832 ti->ti_ack != tp->snd_una) 833 tp->t_dupacks = 0; 834 else if (++tp->t_dupacks == tcprexmtthresh) { 835 tcp_seq onxt = tp->snd_nxt; 836 u_int win = 837 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 838 tp->t_maxseg; 839 840 if (win < 2) 841 win = 2; 842 tp->snd_ssthresh = win * tp->t_maxseg; 843 tp->t_timer[TCPT_REXMT] = 0; 844 tp->t_rtt = 0; 845 tp->snd_nxt = ti->ti_ack; 846 tp->snd_cwnd = tp->t_maxseg; 847 (void) tcp_output(tp); 848 tp->snd_cwnd = tp->snd_ssthresh + 849 tp->t_maxseg * tp->t_dupacks; 850 if (SEQ_GT(onxt, tp->snd_nxt)) 851 tp->snd_nxt = onxt; 852 goto drop; 853 } else if (tp->t_dupacks > tcprexmtthresh) { 854 tp->snd_cwnd += tp->t_maxseg; 855 (void) tcp_output(tp); 856 goto drop; 857 } 858 } else 859 tp->t_dupacks = 0; 860 break; 861 } 862 /* 863 * If the congestion window was inflated to account 864 * for the other side's cached packets, retract it. 865 */ 866 if (tp->t_dupacks > tcprexmtthresh && 867 tp->snd_cwnd > tp->snd_ssthresh) 868 tp->snd_cwnd = tp->snd_ssthresh; 869 tp->t_dupacks = 0; 870 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 871 tcpstat.tcps_rcvacktoomuch++; 872 goto dropafterack; 873 } 874 acked = ti->ti_ack - tp->snd_una; 875 tcpstat.tcps_rcvackpack++; 876 tcpstat.tcps_rcvackbyte += acked; 877 878 /* 879 * If transmit timer is running and timed sequence 880 * number was acked, update smoothed round trip time. 881 * Since we now have an rtt measurement, cancel the 882 * timer backoff (cf., Phil Karn's retransmit alg.). 883 * Recompute the initial retransmit timer. 884 */ 885 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 886 tcp_xmit_timer(tp); 887 888 /* 889 * If all outstanding data is acked, stop retransmit 890 * timer and remember to restart (more output or persist). 891 * If there is more data to be acked, restart retransmit 892 * timer, using current (possibly backed-off) value. 893 */ 894 if (ti->ti_ack == tp->snd_max) { 895 tp->t_timer[TCPT_REXMT] = 0; 896 needoutput = 1; 897 } else if (tp->t_timer[TCPT_PERSIST] == 0) 898 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 899 /* 900 * When new data is acked, open the congestion window. 901 * If the window gives us less than ssthresh packets 902 * in flight, open exponentially (maxseg per packet). 903 * Otherwise open linearly: maxseg per window 904 * (maxseg^2 / cwnd per packet), plus a constant 905 * fraction of a packet (maxseg/8) to help larger windows 906 * open quickly enough. 907 */ 908 { 909 register u_int cw = tp->snd_cwnd; 910 register u_int incr = tp->t_maxseg; 911 912 if (cw > tp->snd_ssthresh) 913 incr = incr * incr / cw + incr / 8; 914 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN); 915 } 916 if (acked > so->so_snd.sb_cc) { 917 tp->snd_wnd -= so->so_snd.sb_cc; 918 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 919 ourfinisacked = 1; 920 } else { 921 sbdrop(&so->so_snd, acked); 922 tp->snd_wnd -= acked; 923 ourfinisacked = 0; 924 } 925 if (so->so_snd.sb_flags & SB_NOTIFY) 926 sowwakeup(so); 927 tp->snd_una = ti->ti_ack; 928 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 929 tp->snd_nxt = tp->snd_una; 930 931 switch (tp->t_state) { 932 933 /* 934 * In FIN_WAIT_1 STATE in addition to the processing 935 * for the ESTABLISHED state if our FIN is now acknowledged 936 * then enter FIN_WAIT_2. 937 */ 938 case TCPS_FIN_WAIT_1: 939 if (ourfinisacked) { 940 /* 941 * If we can't receive any more 942 * data, then closing user can proceed. 943 * Starting the timer is contrary to the 944 * specification, but if we don't get a FIN 945 * we'll hang forever. 946 */ 947 if (so->so_state & SS_CANTRCVMORE) { 948 soisdisconnected(so); 949 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 950 } 951 tp->t_state = TCPS_FIN_WAIT_2; 952 } 953 break; 954 955 /* 956 * In CLOSING STATE in addition to the processing for 957 * the ESTABLISHED state if the ACK acknowledges our FIN 958 * then enter the TIME-WAIT state, otherwise ignore 959 * the segment. 960 */ 961 case TCPS_CLOSING: 962 if (ourfinisacked) { 963 tp->t_state = TCPS_TIME_WAIT; 964 tcp_canceltimers(tp); 965 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 966 soisdisconnected(so); 967 } 968 break; 969 970 /* 971 * In LAST_ACK, we may still be waiting for data to drain 972 * and/or to be acked, as well as for the ack of our FIN. 973 * If our FIN is now acknowledged, delete the TCB, 974 * enter the closed state and return. 975 */ 976 case TCPS_LAST_ACK: 977 if (ourfinisacked) { 978 tp = tcp_close(tp); 979 goto drop; 980 } 981 break; 982 983 /* 984 * In TIME_WAIT state the only thing that should arrive 985 * is a retransmission of the remote FIN. Acknowledge 986 * it and restart the finack timer. 987 */ 988 case TCPS_TIME_WAIT: 989 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 990 goto dropafterack; 991 } 992 } 993 994 step6: 995 /* 996 * Update window information. 997 * Don't look at window if no ACK: TAC's send garbage on first SYN. 998 */ 999 if ((tiflags & TH_ACK) && 1000 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1001 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1002 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 1003 /* keep track of pure window updates */ 1004 if (ti->ti_len == 0 && 1005 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 1006 tcpstat.tcps_rcvwinupd++; 1007 tp->snd_wnd = ti->ti_win; 1008 tp->snd_wl1 = ti->ti_seq; 1009 tp->snd_wl2 = ti->ti_ack; 1010 if (tp->snd_wnd > tp->max_sndwnd) 1011 tp->max_sndwnd = tp->snd_wnd; 1012 needoutput = 1; 1013 } 1014 1015 /* 1016 * Process segments with URG. 1017 */ 1018 if ((tiflags & TH_URG) && ti->ti_urp && 1019 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1020 /* 1021 * This is a kludge, but if we receive and accept 1022 * random urgent pointers, we'll crash in 1023 * soreceive. It's hard to imagine someone 1024 * actually wanting to send this much urgent data. 1025 */ 1026 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 1027 ti->ti_urp = 0; /* XXX */ 1028 tiflags &= ~TH_URG; /* XXX */ 1029 goto dodata; /* XXX */ 1030 } 1031 /* 1032 * If this segment advances the known urgent pointer, 1033 * then mark the data stream. This should not happen 1034 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1035 * a FIN has been received from the remote side. 1036 * In these states we ignore the URG. 1037 * 1038 * According to RFC961 (Assigned Protocols), 1039 * the urgent pointer points to the last octet 1040 * of urgent data. We continue, however, 1041 * to consider it to indicate the first octet 1042 * of data past the urgent section as the original 1043 * spec states (in one of two places). 1044 */ 1045 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1046 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1047 so->so_oobmark = so->so_rcv.sb_cc + 1048 (tp->rcv_up - tp->rcv_nxt) - 1; 1049 if (so->so_oobmark == 0) 1050 so->so_state |= SS_RCVATMARK; 1051 sohasoutofband(so); 1052 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1053 } 1054 /* 1055 * Remove out of band data so doesn't get presented to user. 1056 * This can happen independent of advancing the URG pointer, 1057 * but if two URG's are pending at once, some out-of-band 1058 * data may creep in... ick. 1059 */ 1060 if (ti->ti_urp <= ti->ti_len 1061 #ifdef SO_OOBINLINE 1062 && (so->so_options & SO_OOBINLINE) == 0 1063 #endif 1064 ) 1065 tcp_pulloutofband(so, ti, m); 1066 } else 1067 /* 1068 * If no out of band data is expected, 1069 * pull receive urgent pointer along 1070 * with the receive window. 1071 */ 1072 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1073 tp->rcv_up = tp->rcv_nxt; 1074 dodata: /* XXX */ 1075 1076 /* 1077 * Process the segment text, merging it into the TCP sequencing queue, 1078 * and arranging for acknowledgment of receipt if necessary. 1079 * This process logically involves adjusting tp->rcv_wnd as data 1080 * is presented to the user (this happens in tcp_usrreq.c, 1081 * case PRU_RCVD). If a FIN has already been received on this 1082 * connection then we just ignore the text. 1083 */ 1084 if ((ti->ti_len || (tiflags&TH_FIN)) && 1085 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1086 TCP_REASS(tp, ti, m, so, tiflags); 1087 /* 1088 * Note the amount of data that peer has sent into 1089 * our window, in order to estimate the sender's 1090 * buffer size. 1091 */ 1092 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1093 } else { 1094 m_freem(m); 1095 tiflags &= ~TH_FIN; 1096 } 1097 1098 /* 1099 * If FIN is received ACK the FIN and let the user know 1100 * that the connection is closing. 1101 */ 1102 if (tiflags & TH_FIN) { 1103 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1104 socantrcvmore(so); 1105 tp->t_flags |= TF_ACKNOW; 1106 tp->rcv_nxt++; 1107 } 1108 switch (tp->t_state) { 1109 1110 /* 1111 * In SYN_RECEIVED and ESTABLISHED STATES 1112 * enter the CLOSE_WAIT state. 1113 */ 1114 case TCPS_SYN_RECEIVED: 1115 case TCPS_ESTABLISHED: 1116 tp->t_state = TCPS_CLOSE_WAIT; 1117 break; 1118 1119 /* 1120 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1121 * enter the CLOSING state. 1122 */ 1123 case TCPS_FIN_WAIT_1: 1124 tp->t_state = TCPS_CLOSING; 1125 break; 1126 1127 /* 1128 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1129 * starting the time-wait timer, turning off the other 1130 * standard timers. 1131 */ 1132 case TCPS_FIN_WAIT_2: 1133 tp->t_state = TCPS_TIME_WAIT; 1134 tcp_canceltimers(tp); 1135 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1136 soisdisconnected(so); 1137 break; 1138 1139 /* 1140 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1141 */ 1142 case TCPS_TIME_WAIT: 1143 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1144 break; 1145 } 1146 } 1147 if (so->so_options & SO_DEBUG) 1148 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1149 1150 /* 1151 * Return any desired output. 1152 */ 1153 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1154 (void) tcp_output(tp); 1155 return; 1156 1157 dropafterack: 1158 /* 1159 * Generate an ACK dropping incoming segment if it occupies 1160 * sequence space, where the ACK reflects our state. 1161 */ 1162 if (tiflags & TH_RST) 1163 goto drop; 1164 m_freem(m); 1165 tp->t_flags |= TF_ACKNOW; 1166 (void) tcp_output(tp); 1167 return; 1168 1169 dropwithreset: 1170 if (om) { 1171 (void) m_free(om); 1172 om = 0; 1173 } 1174 /* 1175 * Generate a RST, dropping incoming segment. 1176 * Make ACK acceptable to originator of segment. 1177 * Don't bother to respond if destination was broadcast. 1178 */ 1179 if ((tiflags & TH_RST) || m->m_flags & M_BCAST) 1180 goto drop; 1181 if (tiflags & TH_ACK) 1182 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1183 else { 1184 if (tiflags & TH_SYN) 1185 ti->ti_len++; 1186 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1187 TH_RST|TH_ACK); 1188 } 1189 /* destroy temporarily created socket */ 1190 if (dropsocket) 1191 (void) soabort(so); 1192 return; 1193 1194 drop: 1195 if (om) 1196 (void) m_free(om); 1197 /* 1198 * Drop space held by incoming segment and return. 1199 */ 1200 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1201 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1202 m_freem(m); 1203 /* destroy temporarily created socket */ 1204 if (dropsocket) 1205 (void) soabort(so); 1206 return; 1207 } 1208 1209 tcp_dooptions(tp, om, ti) 1210 struct tcpcb *tp; 1211 struct mbuf *om; 1212 struct tcpiphdr *ti; 1213 { 1214 register u_char *cp; 1215 u_short mss; 1216 int opt, optlen, cnt; 1217 1218 cp = mtod(om, u_char *); 1219 cnt = om->m_len; 1220 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1221 opt = cp[0]; 1222 if (opt == TCPOPT_EOL) 1223 break; 1224 if (opt == TCPOPT_NOP) 1225 optlen = 1; 1226 else { 1227 optlen = cp[1]; 1228 if (optlen <= 0) 1229 break; 1230 } 1231 switch (opt) { 1232 1233 default: 1234 continue; 1235 1236 case TCPOPT_MAXSEG: 1237 if (optlen != 4) 1238 continue; 1239 if (!(ti->ti_flags & TH_SYN)) 1240 continue; 1241 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1242 NTOHS(mss); 1243 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1244 break; 1245 } 1246 } 1247 (void) m_free(om); 1248 } 1249 1250 /* 1251 * Pull out of band byte out of a segment so 1252 * it doesn't appear in the user's data queue. 1253 * It is still reflected in the segment length for 1254 * sequencing purposes. 1255 */ 1256 tcp_pulloutofband(so, ti, m) 1257 struct socket *so; 1258 struct tcpiphdr *ti; 1259 register struct mbuf *m; 1260 { 1261 int cnt = ti->ti_urp - 1; 1262 1263 while (cnt >= 0) { 1264 if (m->m_len > cnt) { 1265 char *cp = mtod(m, caddr_t) + cnt; 1266 struct tcpcb *tp = sototcpcb(so); 1267 1268 tp->t_iobc = *cp; 1269 tp->t_oobflags |= TCPOOB_HAVEDATA; 1270 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1271 m->m_len--; 1272 return; 1273 } 1274 cnt -= m->m_len; 1275 m = m->m_next; 1276 if (m == 0) 1277 break; 1278 } 1279 panic("tcp_pulloutofband"); 1280 } 1281 1282 /* 1283 * Collect new round-trip time estimate 1284 * and update averages and current timeout. 1285 */ 1286 tcp_xmit_timer(tp) 1287 register struct tcpcb *tp; 1288 { 1289 register short delta; 1290 1291 tcpstat.tcps_rttupdated++; 1292 if (tp->t_srtt != 0) { 1293 /* 1294 * srtt is stored as fixed point with 3 bits after the 1295 * binary point (i.e., scaled by 8). The following magic 1296 * is equivalent to the smoothing algorithm in rfc793 with 1297 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1298 * point). Adjust t_rtt to origin 0. 1299 */ 1300 delta = tp->t_rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1301 if ((tp->t_srtt += delta) <= 0) 1302 tp->t_srtt = 1; 1303 /* 1304 * We accumulate a smoothed rtt variance (actually, a 1305 * smoothed mean difference), then set the retransmit 1306 * timer to smoothed rtt + 4 times the smoothed variance. 1307 * rttvar is stored as fixed point with 2 bits after the 1308 * binary point (scaled by 4). The following is 1309 * equivalent to rfc793 smoothing with an alpha of .75 1310 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1311 * rfc793's wired-in beta. 1312 */ 1313 if (delta < 0) 1314 delta = -delta; 1315 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1316 if ((tp->t_rttvar += delta) <= 0) 1317 tp->t_rttvar = 1; 1318 } else { 1319 /* 1320 * No rtt measurement yet - use the unsmoothed rtt. 1321 * Set the variance to half the rtt (so our first 1322 * retransmit happens at 2*rtt) 1323 */ 1324 tp->t_srtt = tp->t_rtt << TCP_RTT_SHIFT; 1325 tp->t_rttvar = tp->t_rtt << (TCP_RTTVAR_SHIFT - 1); 1326 } 1327 tp->t_rtt = 0; 1328 tp->t_rxtshift = 0; 1329 1330 /* 1331 * the retransmit should happen at rtt + 4 * rttvar. 1332 * Because of the way we do the smoothing, srtt and rttvar 1333 * will each average +1/2 tick of bias. When we compute 1334 * the retransmit timer, we want 1/2 tick of rounding and 1335 * 1 extra tick because of +-1/2 tick uncertainty in the 1336 * firing of the timer. The bias will give us exactly the 1337 * 1.5 tick we need. But, because the bias is 1338 * statistical, we have to test that we don't drop below 1339 * the minimum feasible timer (which is 2 ticks). 1340 */ 1341 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1342 tp->t_rttmin, TCPTV_REXMTMAX); 1343 1344 /* 1345 * We received an ack for a packet that wasn't retransmitted; 1346 * it is probably safe to discard any error indications we've 1347 * received recently. This isn't quite right, but close enough 1348 * for now (a route might have failed after we sent a segment, 1349 * and the return path might not be symmetrical). 1350 */ 1351 tp->t_softerror = 0; 1352 } 1353 1354 /* 1355 * Determine a reasonable value for maxseg size. 1356 * If the route is known, check route for mtu. 1357 * If none, use an mss that can be handled on the outgoing 1358 * interface without forcing IP to fragment; if bigger than 1359 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1360 * to utilize large mbufs. If no route is found, route has no mtu, 1361 * or the destination isn't local, use a default, hopefully conservative 1362 * size (usually 512 or the default IP max size, but no more than the mtu 1363 * of the interface), as we can't discover anything about intervening 1364 * gateways or networks. We also initialize the congestion/slow start 1365 * window to be a single segment if the destination isn't local. 1366 * While looking at the routing entry, we also initialize other path-dependent 1367 * parameters from pre-set or cached values in the routing entry. 1368 */ 1369 1370 tcp_mss(tp, offer) 1371 register struct tcpcb *tp; 1372 u_short offer; 1373 { 1374 struct route *ro; 1375 register struct rtentry *rt; 1376 struct ifnet *ifp; 1377 register int rtt, mss; 1378 u_long bufsize; 1379 struct inpcb *inp; 1380 struct socket *so; 1381 extern int tcp_mssdflt, tcp_rttdflt; 1382 1383 inp = tp->t_inpcb; 1384 ro = &inp->inp_route; 1385 1386 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1387 /* No route yet, so try to acquire one */ 1388 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1389 ro->ro_dst.sa_family = AF_INET; 1390 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1391 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1392 inp->inp_faddr; 1393 rtalloc(ro); 1394 } 1395 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1396 return (tcp_mssdflt); 1397 } 1398 ifp = rt->rt_ifp; 1399 so = inp->inp_socket; 1400 1401 #ifdef RTV_MTU /* if route characteristics exist ... */ 1402 /* 1403 * While we're here, check if there's an initial rtt 1404 * or rttvar. Convert from the route-table units 1405 * to scaled multiples of the slow timeout timer. 1406 */ 1407 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1408 if (rt->rt_rmx.rmx_locks & RTV_MTU) 1409 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1410 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1411 if (rt->rt_rmx.rmx_rttvar) 1412 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1413 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1414 else 1415 /* default variation is +- 1 rtt */ 1416 tp->t_rttvar = 1417 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1418 TCPT_RANGESET(tp->t_rxtcur, 1419 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1420 tp->t_rttmin, TCPTV_REXMTMAX); 1421 } 1422 /* 1423 * if there's an mtu associated with the route, use it 1424 */ 1425 if (rt->rt_rmx.rmx_mtu) 1426 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1427 else 1428 #endif /* RTV_MTU */ 1429 { 1430 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1431 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1432 if (mss > MCLBYTES) 1433 mss &= ~(MCLBYTES-1); 1434 #else 1435 if (mss > MCLBYTES) 1436 mss = mss / MCLBYTES * MCLBYTES; 1437 #endif 1438 if (!in_localaddr(inp->inp_faddr)) 1439 mss = min(mss, tcp_mssdflt); 1440 } 1441 /* 1442 * The current mss, t_maxseg, is initialized to the default value. 1443 * If we compute a smaller value, reduce the current mss. 1444 * If we compute a larger value, return it for use in sending 1445 * a max seg size option, but don't store it for use 1446 * unless we received an offer at least that large from peer. 1447 * However, do not accept offers under 32 bytes. 1448 */ 1449 if (offer) 1450 mss = min(mss, offer); 1451 mss = max(mss, 32); /* sanity */ 1452 if (mss < tp->t_maxseg || offer != 0) { 1453 /* 1454 * If there's a pipesize, change the socket buffer 1455 * to that size. Make the socket buffers an integral 1456 * number of mss units; if the mss is larger than 1457 * the socket buffer, decrease the mss. 1458 */ 1459 #ifdef RTV_SPIPE 1460 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1461 #endif 1462 bufsize = so->so_snd.sb_hiwat; 1463 if (bufsize < mss) 1464 mss = bufsize; 1465 else { 1466 bufsize = min(bufsize, SB_MAX) / mss * mss; 1467 (void) sbreserve(&so->so_snd, bufsize); 1468 } 1469 tp->t_maxseg = mss; 1470 1471 #ifdef RTV_RPIPE 1472 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1473 #endif 1474 bufsize = so->so_rcv.sb_hiwat; 1475 if (bufsize > mss) { 1476 bufsize = min(bufsize, SB_MAX) / mss * mss; 1477 (void) sbreserve(&so->so_rcv, bufsize); 1478 } 1479 } 1480 tp->snd_cwnd = mss; 1481 1482 #ifdef RTV_SSTHRESH 1483 if (rt->rt_rmx.rmx_ssthresh) { 1484 /* 1485 * There's some sort of gateway or interface 1486 * buffer limit on the path. Use this to set 1487 * the slow start threshhold, but set the 1488 * threshold to no less than 2*mss. 1489 */ 1490 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1491 } 1492 #endif /* RTV_MTU */ 1493 return (mss); 1494 } 1495