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, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * from: @(#)tcp_input.c 7.25 (Berkeley) 6/30/90 34 * $Id: tcp_input.c,v 1.8 1994/04/25 19:16:53 mycroft Exp $ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/malloc.h> 40 #include <sys/select.h> 41 #include <sys/mbuf.h> 42 #include <sys/protosw.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.h> 45 #include <sys/errno.h> 46 47 #include <net/if.h> 48 #include <net/route.h> 49 50 #include <netinet/in.h> 51 #include <netinet/in_systm.h> 52 #include <netinet/ip.h> 53 #include <netinet/in_pcb.h> 54 #include <netinet/ip_var.h> 55 #include <netinet/tcp.h> 56 #include <netinet/tcp_fsm.h> 57 #include <netinet/tcp_seq.h> 58 #include <netinet/tcp_timer.h> 59 #include <netinet/tcp_var.h> 60 #include <netinet/tcpip.h> 61 #include <netinet/tcp_debug.h> 62 63 int tcprexmtthresh = 3; 64 int tcppredack; /* XXX debugging: times hdr predict ok for acks */ 65 int tcppreddat; /* XXX # times header prediction ok for data packets */ 66 int tcppcbcachemiss; 67 struct tcpiphdr tcp_saveti; 68 struct inpcb *tcp_last_inpcb = &tcb; 69 70 /* 71 * Insert segment ti into reassembly queue of tcp with 72 * control block tp. Return TH_FIN if reassembly now includes 73 * a segment with FIN. The macro form does the common case inline 74 * (segment is the next to be received on an established connection, 75 * and the queue is empty), avoiding linkage into and removal 76 * from the queue and repetition of various conversions. 77 * Set DELACK for segments received in order, but ack immediately 78 * when segments are out of order (so fast retransmit can work). 79 */ 80 #define TCP_REASS(tp, ti, m, so, flags) { \ 81 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 82 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 83 (tp)->t_state == TCPS_ESTABLISHED) { \ 84 if ((ti)->ti_flags & TH_PUSH) \ 85 tp->t_flags |= TF_ACKNOW; \ 86 else \ 87 tp->t_flags |= TF_DELACK; \ 88 (tp)->rcv_nxt += (ti)->ti_len; \ 89 flags = (ti)->ti_flags & TH_FIN; \ 90 tcpstat.tcps_rcvpack++;\ 91 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 92 sbappend(&(so)->so_rcv, (m)); \ 93 sorwakeup(so); \ 94 } else { \ 95 (flags) = tcp_reass((tp), (ti), (m)); \ 96 tp->t_flags |= TF_ACKNOW; \ 97 } \ 98 } 99 100 int 101 tcp_reass(tp, ti, m) 102 register struct tcpcb *tp; 103 register struct tcpiphdr *ti; 104 struct mbuf *m; 105 { 106 register struct tcpiphdr *q; 107 struct socket *so = tp->t_inpcb->inp_socket; 108 int flags; 109 110 /* 111 * Call with ti==0 after become established to 112 * force pre-ESTABLISHED data up to user socket. 113 */ 114 if (ti == 0) 115 goto present; 116 117 /* 118 * Find a segment which begins after this one does. 119 */ 120 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 121 q = (struct tcpiphdr *)q->ti_next) 122 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 123 break; 124 125 /* 126 * If there is a preceding segment, it may provide some of 127 * our data already. If so, drop the data from the incoming 128 * segment. If it provides all of our data, drop us. 129 */ 130 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 131 register int i; 132 q = (struct tcpiphdr *)q->ti_prev; 133 /* conversion to int (in i) handles seq wraparound */ 134 i = q->ti_seq + q->ti_len - ti->ti_seq; 135 if (i > 0) { 136 if (i >= ti->ti_len) { 137 tcpstat.tcps_rcvduppack++; 138 tcpstat.tcps_rcvdupbyte += ti->ti_len; 139 m_freem(m); 140 return (0); 141 } 142 m_adj(m, i); 143 ti->ti_len -= i; 144 ti->ti_seq += i; 145 } 146 q = (struct tcpiphdr *)(q->ti_next); 147 } 148 tcpstat.tcps_rcvoopack++; 149 tcpstat.tcps_rcvoobyte += ti->ti_len; 150 REASS_MBUF(ti) = m; /* XXX */ 151 152 /* 153 * While we overlap succeeding segments trim them or, 154 * if they are completely covered, dequeue them. 155 */ 156 while (q != (struct tcpiphdr *)tp) { 157 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 158 if (i <= 0) 159 break; 160 if (i < q->ti_len) { 161 q->ti_seq += i; 162 q->ti_len -= i; 163 m_adj(REASS_MBUF(q), i); 164 break; 165 } 166 q = (struct tcpiphdr *)q->ti_next; 167 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 168 remque(q->ti_prev); 169 m_freem(m); 170 } 171 172 /* 173 * Stick new segment in its place. 174 */ 175 insque(ti, q->ti_prev); 176 177 present: 178 /* 179 * Present data to user, advancing rcv_nxt through 180 * completed sequence space. 181 */ 182 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 183 return (0); 184 ti = tp->seg_next; 185 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 186 return (0); 187 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 188 return (0); 189 do { 190 tp->rcv_nxt += ti->ti_len; 191 flags = ti->ti_flags & TH_FIN; 192 remque(ti); 193 m = REASS_MBUF(ti); 194 ti = (struct tcpiphdr *)ti->ti_next; 195 if (so->so_state & SS_CANTRCVMORE) 196 m_freem(m); 197 else 198 sbappend(&so->so_rcv, m); 199 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 200 sorwakeup(so); 201 return (flags); 202 } 203 204 /* 205 * TCP input routine, follows pages 65-76 of the 206 * protocol specification dated September, 1981 very closely. 207 */ 208 void 209 tcp_input(m, iphlen) 210 register struct mbuf *m; 211 int iphlen; 212 { 213 register struct tcpiphdr *ti; 214 register struct inpcb *inp; 215 struct mbuf *om = 0; 216 int len, tlen, off; 217 register struct tcpcb *tp = 0; 218 register int tiflags; 219 struct socket *so; 220 int todrop, acked, ourfinisacked, needoutput = 0; 221 short ostate; 222 struct in_addr laddr; 223 int dropsocket = 0; 224 int iss = 0; 225 226 tcpstat.tcps_rcvtotal++; 227 /* 228 * Get IP and TCP header together in first mbuf. 229 * Note: IP leaves IP header in first mbuf. 230 */ 231 ti = mtod(m, struct tcpiphdr *); 232 if (iphlen > sizeof (struct ip)) 233 ip_stripoptions(m, (struct mbuf *)0); 234 if (m->m_len < sizeof (struct tcpiphdr)) { 235 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 236 tcpstat.tcps_rcvshort++; 237 return; 238 } 239 ti = mtod(m, struct tcpiphdr *); 240 } 241 242 /* 243 * Checksum extended TCP header and data. 244 */ 245 tlen = ((struct ip *)ti)->ip_len; 246 len = sizeof (struct ip) + tlen; 247 ti->ti_next = ti->ti_prev = 0; 248 ti->ti_x1 = 0; 249 ti->ti_len = (u_short)tlen; 250 HTONS(ti->ti_len); 251 if (ti->ti_sum = in_cksum(m, len)) { 252 tcpstat.tcps_rcvbadsum++; 253 goto drop; 254 } 255 256 /* 257 * Check that TCP offset makes sense, 258 * pull out TCP options and adjust length. XXX 259 */ 260 off = ti->ti_off << 2; 261 if (off < sizeof (struct tcphdr) || off > tlen) { 262 tcpstat.tcps_rcvbadoff++; 263 goto drop; 264 } 265 tlen -= off; 266 ti->ti_len = tlen; 267 if (off > sizeof (struct tcphdr)) { 268 if (m->m_len < sizeof(struct ip) + off) { 269 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 270 tcpstat.tcps_rcvshort++; 271 return; 272 } 273 ti = mtod(m, struct tcpiphdr *); 274 } 275 om = m_get(M_DONTWAIT, MT_DATA); 276 if (om == 0) 277 goto drop; 278 om->m_len = off - sizeof (struct tcphdr); 279 { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 280 bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 281 m->m_len -= om->m_len; 282 m->m_pkthdr.len -= om->m_len; 283 bcopy(op+om->m_len, op, 284 (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 285 } 286 } 287 tiflags = ti->ti_flags; 288 289 /* 290 * Convert TCP protocol specific fields to host format. 291 */ 292 NTOHL(ti->ti_seq); 293 NTOHL(ti->ti_ack); 294 NTOHS(ti->ti_win); 295 NTOHS(ti->ti_urp); 296 297 /* 298 * Locate pcb for segment. 299 */ 300 findpcb: 301 inp = tcp_last_inpcb; 302 if (inp->inp_lport != ti->ti_dport || 303 inp->inp_fport != ti->ti_sport || 304 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 305 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 306 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 307 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 308 if (inp) 309 tcp_last_inpcb = inp; 310 ++tcppcbcachemiss; 311 } 312 313 /* 314 * If the state is CLOSED (i.e., TCB does not exist) then 315 * all data in the incoming segment is discarded. 316 * If the TCB exists but is in CLOSED state, it is embryonic, 317 * but should either do a listen or a connect soon. 318 */ 319 if (inp == 0) 320 goto dropwithreset; 321 tp = intotcpcb(inp); 322 if (tp == 0) 323 goto dropwithreset; 324 if (tp->t_state == TCPS_CLOSED) 325 goto drop; 326 so = inp->inp_socket; 327 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 328 if (so->so_options & SO_DEBUG) { 329 ostate = tp->t_state; 330 tcp_saveti = *ti; 331 } 332 if (so->so_options & SO_ACCEPTCONN) { 333 so = sonewconn(so, 0); 334 if (so == 0) 335 goto drop; 336 /* 337 * This is ugly, but .... 338 * 339 * Mark socket as temporary until we're 340 * committed to keeping it. The code at 341 * ``drop'' and ``dropwithreset'' check the 342 * flag dropsocket to see if the temporary 343 * socket created here should be discarded. 344 * We mark the socket as discardable until 345 * we're committed to it below in TCPS_LISTEN. 346 */ 347 dropsocket++; 348 inp = (struct inpcb *)so->so_pcb; 349 inp->inp_laddr = ti->ti_dst; 350 inp->inp_lport = ti->ti_dport; 351 #if BSD>=43 352 inp->inp_options = ip_srcroute(); 353 #endif 354 tp = intotcpcb(inp); 355 tp->t_state = TCPS_LISTEN; 356 } 357 } 358 359 /* 360 * Segment received on connection. 361 * Reset idle time and keep-alive timer. 362 */ 363 tp->t_idle = 0; 364 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 365 366 /* 367 * Process options if not in LISTEN state, 368 * else do it below (after getting remote address). 369 */ 370 if (om && tp->t_state != TCPS_LISTEN) { 371 tcp_dooptions(tp, om, ti); 372 om = 0; 373 } 374 /* 375 * Header prediction: check for the two common cases 376 * of a uni-directional data xfer. If the packet has 377 * no control flags, is in-sequence, the window didn't 378 * change and we're not retransmitting, it's a 379 * candidate. If the length is zero and the ack moved 380 * forward, we're the sender side of the xfer. Just 381 * free the data acked & wake any higher level process 382 * that was blocked waiting for space. If the length 383 * is non-zero and the ack didn't move, we're the 384 * receiver side. If we're getting packets in-order 385 * (the reassembly queue is empty), add the data to 386 * the socket buffer and note that we need a delayed ack. 387 */ 388 if (tp->t_state == TCPS_ESTABLISHED && 389 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 390 ti->ti_seq == tp->rcv_nxt && 391 ti->ti_win && ti->ti_win == tp->snd_wnd && 392 tp->snd_nxt == tp->snd_max) { 393 if (ti->ti_len == 0) { 394 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 395 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 396 tp->snd_cwnd >= tp->snd_wnd) { 397 /* 398 * this is a pure ack for outstanding data. 399 */ 400 ++tcppredack; 401 if (tp->t_rtt && SEQ_GT(ti->ti_ack,tp->t_rtseq)) 402 tcp_xmit_timer(tp); 403 acked = ti->ti_ack - tp->snd_una; 404 tcpstat.tcps_rcvackpack++; 405 tcpstat.tcps_rcvackbyte += acked; 406 sbdrop(&so->so_snd, acked); 407 tp->snd_una = ti->ti_ack; 408 m_freem(m); 409 410 /* 411 * If all outstanding data are acked, stop 412 * retransmit timer, otherwise restart timer 413 * using current (possibly backed-off) value. 414 * If process is waiting for space, 415 * wakeup/selwakeup/signal. If data 416 * are ready to send, let tcp_output 417 * decide between more output or persist. 418 */ 419 if (tp->snd_una == tp->snd_max) 420 tp->t_timer[TCPT_REXMT] = 0; 421 else if (tp->t_timer[TCPT_PERSIST] == 0) 422 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 423 424 if (so->so_snd.sb_flags & SB_NOTIFY) 425 sowwakeup(so); 426 if (so->so_snd.sb_cc) 427 (void) tcp_output(tp); 428 return; 429 } 430 } else if (ti->ti_ack == tp->snd_una && 431 tp->seg_next == (struct tcpiphdr *)tp && 432 ti->ti_len <= sbspace(&so->so_rcv)) { 433 /* 434 * this is a pure, in-sequence data packet 435 * with nothing on the reassembly queue and 436 * we have enough buffer space to take it. 437 */ 438 ++tcppreddat; 439 tp->rcv_nxt += ti->ti_len; 440 tcpstat.tcps_rcvpack++; 441 tcpstat.tcps_rcvbyte += ti->ti_len; 442 /* 443 * Drop TCP and IP headers then add data 444 * to socket buffer 445 */ 446 m->m_data += sizeof(struct tcpiphdr); 447 m->m_len -= sizeof(struct tcpiphdr); 448 sbappend(&so->so_rcv, m); 449 sorwakeup(so); 450 if (ti->ti_flags & TH_PUSH) 451 tp->t_flags |= TF_ACKNOW; 452 else 453 tp->t_flags |= TF_DELACK; 454 return; 455 } 456 } 457 458 /* 459 * Drop TCP and IP headers; TCP options were dropped above. 460 */ 461 m->m_data += sizeof(struct tcpiphdr); 462 m->m_len -= sizeof(struct tcpiphdr); 463 464 /* 465 * Calculate amount of space in receive window, 466 * and then do TCP input processing. 467 * Receive window is amount of space in rcv queue, 468 * but not less than advertised window. 469 */ 470 { int win; 471 472 win = sbspace(&so->so_rcv); 473 if (win < 0) 474 win = 0; 475 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 476 } 477 478 switch (tp->t_state) { 479 480 /* 481 * If the state is LISTEN then ignore segment if it contains an RST. 482 * If the segment contains an ACK then it is bad and send a RST. 483 * If it does not contain a SYN then it is not interesting; drop it. 484 * Don't bother responding if the destination was a broadcast. 485 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 486 * tp->iss, and send a segment: 487 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 488 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 489 * Fill in remote peer address fields if not previously specified. 490 * Enter SYN_RECEIVED state, and process any other fields of this 491 * segment in this state. 492 */ 493 case TCPS_LISTEN: { 494 struct mbuf *am; 495 register struct sockaddr_in *sin; 496 497 if (tiflags & TH_RST) 498 goto drop; 499 if (tiflags & TH_ACK) 500 goto dropwithreset; 501 if ((tiflags & TH_SYN) == 0) 502 goto drop; 503 if (m->m_flags & M_BCAST) 504 goto drop; 505 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 506 if (am == NULL) 507 goto drop; 508 am->m_len = sizeof (struct sockaddr_in); 509 sin = mtod(am, struct sockaddr_in *); 510 sin->sin_family = AF_INET; 511 sin->sin_len = sizeof(*sin); 512 sin->sin_addr = ti->ti_src; 513 sin->sin_port = ti->ti_sport; 514 laddr = inp->inp_laddr; 515 if (inp->inp_laddr.s_addr == INADDR_ANY) 516 inp->inp_laddr = ti->ti_dst; 517 if (in_pcbconnect(inp, am)) { 518 inp->inp_laddr = laddr; 519 (void) m_free(am); 520 goto drop; 521 } 522 (void) m_free(am); 523 tp->t_template = tcp_template(tp); 524 if (tp->t_template == 0) { 525 tp = tcp_drop(tp, ENOBUFS); 526 dropsocket = 0; /* socket is already gone */ 527 goto drop; 528 } 529 if (om) { 530 tcp_dooptions(tp, om, ti); 531 om = 0; 532 } 533 if (iss) 534 tp->iss = iss; 535 else 536 tp->iss = tcp_iss; 537 tcp_iss += TCP_ISSINCR/2; 538 tp->irs = ti->ti_seq; 539 tcp_sendseqinit(tp); 540 tcp_rcvseqinit(tp); 541 tp->t_flags |= TF_ACKNOW; 542 tp->t_state = TCPS_SYN_RECEIVED; 543 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 544 dropsocket = 0; /* committed to socket */ 545 tcpstat.tcps_accepts++; 546 goto trimthenstep6; 547 } 548 549 /* 550 * If the state is SYN_SENT: 551 * if seg contains an ACK, but not for our SYN, drop the input. 552 * if seg contains a RST, then drop the connection. 553 * if seg does not contain SYN, then drop it. 554 * Otherwise this is an acceptable SYN segment 555 * initialize tp->rcv_nxt and tp->irs 556 * if seg contains ack then advance tp->snd_una 557 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 558 * arrange for segment to be acked (eventually) 559 * continue processing rest of data/controls, beginning with URG 560 */ 561 case TCPS_SYN_SENT: 562 if ((tiflags & TH_ACK) && 563 (SEQ_LEQ(ti->ti_ack, tp->iss) || 564 SEQ_GT(ti->ti_ack, tp->snd_max))) 565 goto dropwithreset; 566 if (tiflags & TH_RST) { 567 if (tiflags & TH_ACK) 568 tp = tcp_drop(tp, ECONNREFUSED); 569 goto drop; 570 } 571 if ((tiflags & TH_SYN) == 0) 572 goto drop; 573 if (tiflags & TH_ACK) { 574 tp->snd_una = ti->ti_ack; 575 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 576 tp->snd_nxt = tp->snd_una; 577 } 578 tp->t_timer[TCPT_REXMT] = 0; 579 tp->irs = ti->ti_seq; 580 tcp_rcvseqinit(tp); 581 tp->t_flags |= TF_ACKNOW; 582 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 583 tcpstat.tcps_connects++; 584 soisconnected(so); 585 tp->t_state = TCPS_ESTABLISHED; 586 (void) tcp_reass(tp, (struct tcpiphdr *)0, 587 (struct mbuf *)0); 588 /* 589 * if we didn't have to retransmit the SYN, 590 * use its rtt as our initial srtt & rtt var. 591 */ 592 if (tp->t_rtt) 593 tcp_xmit_timer(tp); 594 } else 595 tp->t_state = TCPS_SYN_RECEIVED; 596 597 trimthenstep6: 598 /* 599 * Advance ti->ti_seq to correspond to first data byte. 600 * If data, trim to stay within window, 601 * dropping FIN if necessary. 602 */ 603 ti->ti_seq++; 604 if (ti->ti_len > tp->rcv_wnd) { 605 todrop = ti->ti_len - tp->rcv_wnd; 606 m_adj(m, -todrop); 607 ti->ti_len = tp->rcv_wnd; 608 tiflags &= ~TH_FIN; 609 tcpstat.tcps_rcvpackafterwin++; 610 tcpstat.tcps_rcvbyteafterwin += todrop; 611 } 612 tp->snd_wl1 = ti->ti_seq - 1; 613 tp->rcv_up = ti->ti_seq; 614 goto step6; 615 } 616 617 /* 618 * States other than LISTEN or SYN_SENT. 619 * First check that at least some bytes of segment are within 620 * receive window. If segment begins before rcv_nxt, 621 * drop leading data (and SYN); if nothing left, just ack. 622 */ 623 todrop = tp->rcv_nxt - ti->ti_seq; 624 if (todrop > 0) { 625 if (tiflags & TH_SYN) { 626 tiflags &= ~TH_SYN; 627 ti->ti_seq++; 628 if (ti->ti_urp > 1) 629 ti->ti_urp--; 630 else 631 tiflags &= ~TH_URG; 632 todrop--; 633 } 634 if (todrop > ti->ti_len || 635 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 636 /* 637 * Any valid FIN must be to the left of the 638 * window. At this point, FIN must be a 639 * duplicate or out-of-sequence, so drop it. 640 */ 641 tiflags &= ~TH_FIN; 642 /* 643 * Send ACK to resynchronize, and drop any data, 644 * but keep on processing for RST or ACK. 645 */ 646 tp->t_flags |= TF_ACKNOW; 647 tcpstat.tcps_rcvdupbyte += todrop = ti->ti_len; 648 tcpstat.tcps_rcvduppack++; 649 } else { 650 tcpstat.tcps_rcvpartduppack++; 651 tcpstat.tcps_rcvpartdupbyte += todrop; 652 } 653 m_adj(m, todrop); 654 ti->ti_seq += todrop; 655 ti->ti_len -= todrop; 656 if (ti->ti_urp > todrop) 657 ti->ti_urp -= todrop; 658 else { 659 tiflags &= ~TH_URG; 660 ti->ti_urp = 0; 661 } 662 } 663 664 /* 665 * If new data are received on a connection after the 666 * user processes are gone, then RST the other end. 667 */ 668 if ((so->so_state & SS_NOFDREF) && 669 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 670 tp = tcp_close(tp); 671 tcpstat.tcps_rcvafterclose++; 672 goto dropwithreset; 673 } 674 675 /* 676 * If segment ends after window, drop trailing data 677 * (and PUSH and FIN); if nothing left, just ACK. 678 */ 679 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 680 if (todrop > 0) { 681 tcpstat.tcps_rcvpackafterwin++; 682 if (todrop >= ti->ti_len) { 683 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 684 /* 685 * If a new connection request is received 686 * while in TIME_WAIT, drop the old connection 687 * and start over if the sequence numbers 688 * are above the previous ones. 689 */ 690 if (tiflags & TH_SYN && 691 tp->t_state == TCPS_TIME_WAIT && 692 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 693 iss = tp->rcv_nxt + TCP_ISSINCR; 694 tp = tcp_close(tp); 695 goto findpcb; 696 } 697 /* 698 * If window is closed can only take segments at 699 * window edge, and have to drop data and PUSH from 700 * incoming segments. Continue processing, but 701 * remember to ack. Otherwise, drop segment 702 * and ack. 703 */ 704 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 705 tp->t_flags |= TF_ACKNOW; 706 tcpstat.tcps_rcvwinprobe++; 707 } else 708 goto dropafterack; 709 } else 710 tcpstat.tcps_rcvbyteafterwin += todrop; 711 m_adj(m, -todrop); 712 ti->ti_len -= todrop; 713 tiflags &= ~(TH_PUSH|TH_FIN); 714 } 715 716 /* 717 * If the RST bit is set examine the state: 718 * SYN_RECEIVED STATE: 719 * If passive open, return to LISTEN state. 720 * If active open, inform user that connection was refused. 721 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 722 * Inform user that connection was reset, and close tcb. 723 * CLOSING, LAST_ACK, TIME_WAIT STATES 724 * Close the tcb. 725 */ 726 if (tiflags&TH_RST) switch (tp->t_state) { 727 728 case TCPS_SYN_RECEIVED: 729 so->so_error = ECONNREFUSED; 730 goto close; 731 732 case TCPS_ESTABLISHED: 733 case TCPS_FIN_WAIT_1: 734 case TCPS_FIN_WAIT_2: 735 case TCPS_CLOSE_WAIT: 736 so->so_error = ECONNRESET; 737 close: 738 tp->t_state = TCPS_CLOSED; 739 tcpstat.tcps_drops++; 740 tp = tcp_close(tp); 741 goto drop; 742 743 case TCPS_CLOSING: 744 case TCPS_LAST_ACK: 745 case TCPS_TIME_WAIT: 746 tp = tcp_close(tp); 747 goto drop; 748 } 749 750 /* 751 * If a SYN is in the window, then this is an 752 * error and we send an RST and drop the connection. 753 */ 754 if (tiflags & TH_SYN) { 755 tp = tcp_drop(tp, ECONNRESET); 756 goto dropwithreset; 757 } 758 759 /* 760 * If the ACK bit is off we drop the segment and return. 761 */ 762 if ((tiflags & TH_ACK) == 0) 763 goto drop; 764 765 /* 766 * Ack processing. 767 */ 768 switch (tp->t_state) { 769 770 /* 771 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 772 * ESTABLISHED state and continue processing, otherwise 773 * send an RST. 774 */ 775 case TCPS_SYN_RECEIVED: 776 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 777 SEQ_GT(ti->ti_ack, tp->snd_max)) 778 goto dropwithreset; 779 tcpstat.tcps_connects++; 780 soisconnected(so); 781 tp->t_state = TCPS_ESTABLISHED; 782 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 783 tp->snd_wl1 = ti->ti_seq - 1; 784 /* fall into ... */ 785 786 /* 787 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 788 * ACKs. If the ack is in the range 789 * tp->snd_una < ti->ti_ack <= tp->snd_max 790 * then advance tp->snd_una to ti->ti_ack and drop 791 * data from the retransmission queue. If this ACK reflects 792 * more up to date window information we update our window information. 793 */ 794 case TCPS_ESTABLISHED: 795 case TCPS_FIN_WAIT_1: 796 case TCPS_FIN_WAIT_2: 797 case TCPS_CLOSE_WAIT: 798 case TCPS_CLOSING: 799 case TCPS_LAST_ACK: 800 case TCPS_TIME_WAIT: 801 802 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 803 if (ti->ti_len == 0 && ti->ti_win == tp->snd_wnd) { 804 tcpstat.tcps_rcvdupack++; 805 /* 806 * If we have outstanding data (other than 807 * a window probe), this is a completely 808 * duplicate ack (ie, window info didn't 809 * change), the ack is the biggest we've 810 * seen and we've seen exactly our rexmt 811 * threshhold of them, assume a packet 812 * has been dropped and retransmit it. 813 * Kludge snd_nxt & the congestion 814 * window so we send only this one 815 * packet. 816 * 817 * We know we're losing at the current 818 * window size so do congestion avoidance 819 * (set ssthresh to half the current window 820 * and pull our congestion window back to 821 * the new ssthresh). 822 * 823 * Dup acks mean that packets have left the 824 * network (they're now cached at the receiver) 825 * so bump cwnd by the amount in the receiver 826 * to keep a constant cwnd packets in the 827 * network. 828 */ 829 if (tp->t_timer[TCPT_REXMT] == 0 || 830 ti->ti_ack != tp->snd_una) 831 tp->t_dupacks = 0; 832 else if (++tp->t_dupacks == tcprexmtthresh) { 833 tcp_seq onxt = tp->snd_nxt; 834 u_int win = 835 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 836 tp->t_maxseg; 837 838 if (win < 2) 839 win = 2; 840 tp->snd_ssthresh = win * tp->t_maxseg; 841 tp->t_timer[TCPT_REXMT] = 0; 842 tp->t_rtt = 0; 843 tp->snd_nxt = ti->ti_ack; 844 tp->snd_cwnd = tp->t_maxseg; 845 (void) tcp_output(tp); 846 tp->snd_cwnd = tp->snd_ssthresh + 847 tp->t_maxseg * tp->t_dupacks; 848 if (SEQ_GT(onxt, tp->snd_nxt)) 849 tp->snd_nxt = onxt; 850 goto drop; 851 } else if (tp->t_dupacks > tcprexmtthresh) { 852 tp->snd_cwnd += tp->t_maxseg; 853 (void) tcp_output(tp); 854 goto drop; 855 } 856 } else 857 tp->t_dupacks = 0; 858 break; 859 } 860 /* 861 * If the congestion window was inflated to account 862 * for the other side's cached packets, retract it. 863 */ 864 if (tp->t_dupacks > tcprexmtthresh && 865 tp->snd_cwnd > tp->snd_ssthresh) 866 tp->snd_cwnd = tp->snd_ssthresh; 867 tp->t_dupacks = 0; 868 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 869 tcpstat.tcps_rcvacktoomuch++; 870 goto dropafterack; 871 } 872 acked = ti->ti_ack - tp->snd_una; 873 tcpstat.tcps_rcvackpack++; 874 tcpstat.tcps_rcvackbyte += acked; 875 876 /* 877 * If transmit timer is running and timed sequence 878 * number was acked, update smoothed round trip time. 879 * Since we now have an rtt measurement, cancel the 880 * timer backoff (cf., Phil Karn's retransmit alg.). 881 * Recompute the initial retransmit timer. 882 */ 883 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 884 tcp_xmit_timer(tp); 885 886 /* 887 * If all outstanding data is acked, stop retransmit 888 * timer and remember to restart (more output or persist). 889 * If there is more data to be acked, restart retransmit 890 * timer, using current (possibly backed-off) value. 891 */ 892 if (ti->ti_ack == tp->snd_max) { 893 tp->t_timer[TCPT_REXMT] = 0; 894 needoutput = 1; 895 } else if (tp->t_timer[TCPT_PERSIST] == 0) 896 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 897 /* 898 * When new data is acked, open the congestion window. 899 * If the window gives us less than ssthresh packets 900 * in flight, open exponentially (maxseg per packet). 901 * Otherwise open linearly: maxseg per window 902 * (maxseg^2 / cwnd per packet), plus a constant 903 * fraction of a packet (maxseg/8) to help larger windows 904 * open quickly enough. 905 */ 906 { 907 register u_int cw = tp->snd_cwnd; 908 register u_int incr = tp->t_maxseg; 909 910 if (cw > tp->snd_ssthresh) 911 incr = incr * incr / cw + incr / 8; 912 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN); 913 } 914 if (acked > so->so_snd.sb_cc) { 915 tp->snd_wnd -= so->so_snd.sb_cc; 916 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 917 ourfinisacked = 1; 918 } else { 919 sbdrop(&so->so_snd, acked); 920 tp->snd_wnd -= acked; 921 ourfinisacked = 0; 922 } 923 if (so->so_snd.sb_flags & SB_NOTIFY) 924 sowwakeup(so); 925 tp->snd_una = ti->ti_ack; 926 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 927 tp->snd_nxt = tp->snd_una; 928 929 switch (tp->t_state) { 930 931 /* 932 * In FIN_WAIT_1 STATE in addition to the processing 933 * for the ESTABLISHED state if our FIN is now acknowledged 934 * then enter FIN_WAIT_2. 935 */ 936 case TCPS_FIN_WAIT_1: 937 if (ourfinisacked) { 938 /* 939 * If we can't receive any more 940 * data, then closing user can proceed. 941 * Starting the timer is contrary to the 942 * specification, but if we don't get a FIN 943 * we'll hang forever. 944 */ 945 if (so->so_state & SS_CANTRCVMORE) { 946 soisdisconnected(so); 947 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 948 } 949 tp->t_state = TCPS_FIN_WAIT_2; 950 } 951 break; 952 953 /* 954 * In CLOSING STATE in addition to the processing for 955 * the ESTABLISHED state if the ACK acknowledges our FIN 956 * then enter the TIME-WAIT state, otherwise ignore 957 * the segment. 958 */ 959 case TCPS_CLOSING: 960 if (ourfinisacked) { 961 tp->t_state = TCPS_TIME_WAIT; 962 tcp_canceltimers(tp); 963 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 964 soisdisconnected(so); 965 } 966 break; 967 968 /* 969 * In LAST_ACK, we may still be waiting for data to drain 970 * and/or to be acked, as well as for the ack of our FIN. 971 * If our FIN is now acknowledged, delete the TCB, 972 * enter the closed state and return. 973 */ 974 case TCPS_LAST_ACK: 975 if (ourfinisacked) { 976 tp = tcp_close(tp); 977 goto drop; 978 } 979 break; 980 981 /* 982 * In TIME_WAIT state the only thing that should arrive 983 * is a retransmission of the remote FIN. Acknowledge 984 * it and restart the finack timer. 985 */ 986 case TCPS_TIME_WAIT: 987 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 988 goto dropafterack; 989 } 990 } 991 992 step6: 993 /* 994 * Update window information. 995 * Don't look at window if no ACK: TAC's send garbage on first SYN. 996 */ 997 if ((tiflags & TH_ACK) && 998 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 999 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1000 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) { 1001 /* keep track of pure window updates */ 1002 if (ti->ti_len == 0 && 1003 tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd) 1004 tcpstat.tcps_rcvwinupd++; 1005 tp->snd_wnd = ti->ti_win; 1006 tp->snd_wl1 = ti->ti_seq; 1007 tp->snd_wl2 = ti->ti_ack; 1008 if (tp->snd_wnd > tp->max_sndwnd) 1009 tp->max_sndwnd = tp->snd_wnd; 1010 needoutput = 1; 1011 } 1012 1013 /* 1014 * Process segments with URG. 1015 */ 1016 if ((tiflags & TH_URG) && ti->ti_urp && 1017 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1018 /* 1019 * This is a kludge, but if we receive and accept 1020 * random urgent pointers, we'll crash in 1021 * soreceive. It's hard to imagine someone 1022 * actually wanting to send this much urgent data. 1023 */ 1024 if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) { 1025 ti->ti_urp = 0; /* XXX */ 1026 tiflags &= ~TH_URG; /* XXX */ 1027 goto dodata; /* XXX */ 1028 } 1029 /* 1030 * If this segment advances the known urgent pointer, 1031 * then mark the data stream. This should not happen 1032 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1033 * a FIN has been received from the remote side. 1034 * In these states we ignore the URG. 1035 * 1036 * According to RFC961 (Assigned Protocols), 1037 * the urgent pointer points to the last octet 1038 * of urgent data. We continue, however, 1039 * to consider it to indicate the first octet 1040 * of data past the urgent section as the original 1041 * spec states (in one of two places). 1042 */ 1043 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1044 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1045 so->so_oobmark = so->so_rcv.sb_cc + 1046 (tp->rcv_up - tp->rcv_nxt) - 1; 1047 if (so->so_oobmark == 0) 1048 so->so_state |= SS_RCVATMARK; 1049 sohasoutofband(so); 1050 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1051 } 1052 /* 1053 * Remove out of band data so doesn't get presented to user. 1054 * This can happen independent of advancing the URG pointer, 1055 * but if two URG's are pending at once, some out-of-band 1056 * data may creep in... ick. 1057 */ 1058 if (ti->ti_urp <= ti->ti_len 1059 #ifdef SO_OOBINLINE 1060 && (so->so_options & SO_OOBINLINE) == 0 1061 #endif 1062 ) 1063 tcp_pulloutofband(so, ti, m); 1064 } else 1065 /* 1066 * If no out of band data is expected, 1067 * pull receive urgent pointer along 1068 * with the receive window. 1069 */ 1070 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1071 tp->rcv_up = tp->rcv_nxt; 1072 dodata: /* XXX */ 1073 1074 /* 1075 * Process the segment text, merging it into the TCP sequencing queue, 1076 * and arranging for acknowledgment of receipt if necessary. 1077 * This process logically involves adjusting tp->rcv_wnd as data 1078 * is presented to the user (this happens in tcp_usrreq.c, 1079 * case PRU_RCVD). If a FIN has already been received on this 1080 * connection then we just ignore the text. 1081 */ 1082 if ((ti->ti_len || (tiflags&TH_FIN)) && 1083 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1084 TCP_REASS(tp, ti, m, so, tiflags); 1085 /* 1086 * Note the amount of data that peer has sent into 1087 * our window, in order to estimate the sender's 1088 * buffer size. 1089 */ 1090 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1091 } else { 1092 m_freem(m); 1093 tiflags &= ~TH_FIN; 1094 } 1095 1096 /* 1097 * If FIN is received ACK the FIN and let the user know 1098 * that the connection is closing. 1099 */ 1100 if (tiflags & TH_FIN) { 1101 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1102 socantrcvmore(so); 1103 tp->t_flags |= TF_ACKNOW; 1104 tp->rcv_nxt++; 1105 } 1106 switch (tp->t_state) { 1107 1108 /* 1109 * In SYN_RECEIVED and ESTABLISHED STATES 1110 * enter the CLOSE_WAIT state. 1111 */ 1112 case TCPS_SYN_RECEIVED: 1113 case TCPS_ESTABLISHED: 1114 tp->t_state = TCPS_CLOSE_WAIT; 1115 break; 1116 1117 /* 1118 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1119 * enter the CLOSING state. 1120 */ 1121 case TCPS_FIN_WAIT_1: 1122 tp->t_state = TCPS_CLOSING; 1123 break; 1124 1125 /* 1126 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1127 * starting the time-wait timer, turning off the other 1128 * standard timers. 1129 */ 1130 case TCPS_FIN_WAIT_2: 1131 tp->t_state = TCPS_TIME_WAIT; 1132 tcp_canceltimers(tp); 1133 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1134 soisdisconnected(so); 1135 break; 1136 1137 /* 1138 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1139 */ 1140 case TCPS_TIME_WAIT: 1141 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1142 break; 1143 } 1144 } 1145 if (so->so_options & SO_DEBUG) 1146 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1147 1148 /* 1149 * Return any desired output. 1150 */ 1151 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1152 (void) tcp_output(tp); 1153 return; 1154 1155 dropafterack: 1156 /* 1157 * Generate an ACK dropping incoming segment if it occupies 1158 * sequence space, where the ACK reflects our state. 1159 */ 1160 if (tiflags & TH_RST) 1161 goto drop; 1162 m_freem(m); 1163 tp->t_flags |= TF_ACKNOW; 1164 (void) tcp_output(tp); 1165 return; 1166 1167 dropwithreset: 1168 if (om) { 1169 (void) m_free(om); 1170 om = 0; 1171 } 1172 /* 1173 * Generate a RST, dropping incoming segment. 1174 * Make ACK acceptable to originator of segment. 1175 * Don't bother to respond if destination was broadcast. 1176 */ 1177 if ((tiflags & TH_RST) || m->m_flags & M_BCAST) 1178 goto drop; 1179 if (tiflags & TH_ACK) 1180 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1181 else { 1182 if (tiflags & TH_SYN) 1183 ti->ti_len++; 1184 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1185 TH_RST|TH_ACK); 1186 } 1187 /* destroy temporarily created socket */ 1188 if (dropsocket) 1189 (void) soabort(so); 1190 return; 1191 1192 drop: 1193 if (om) 1194 (void) m_free(om); 1195 /* 1196 * Drop space held by incoming segment and return. 1197 */ 1198 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1199 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1200 m_freem(m); 1201 /* destroy temporarily created socket */ 1202 if (dropsocket) 1203 (void) soabort(so); 1204 return; 1205 } 1206 1207 void 1208 tcp_dooptions(tp, om, ti) 1209 struct tcpcb *tp; 1210 struct mbuf *om; 1211 struct tcpiphdr *ti; 1212 { 1213 register u_char *cp; 1214 u_short mss; 1215 int opt, optlen, cnt; 1216 1217 cp = mtod(om, u_char *); 1218 cnt = om->m_len; 1219 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1220 opt = cp[0]; 1221 if (opt == TCPOPT_EOL) 1222 break; 1223 if (opt == TCPOPT_NOP) 1224 optlen = 1; 1225 else { 1226 optlen = cp[1]; 1227 if (optlen <= 0) 1228 break; 1229 } 1230 switch (opt) { 1231 1232 default: 1233 continue; 1234 1235 case TCPOPT_MAXSEG: 1236 if (optlen != 4) 1237 continue; 1238 if (!(ti->ti_flags & TH_SYN)) 1239 continue; 1240 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1241 NTOHS(mss); 1242 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1243 break; 1244 } 1245 } 1246 (void) m_free(om); 1247 } 1248 1249 /* 1250 * Pull out of band byte out of a segment so 1251 * it doesn't appear in the user's data queue. 1252 * It is still reflected in the segment length for 1253 * sequencing purposes. 1254 */ 1255 void 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 void 1287 tcp_xmit_timer(tp) 1288 register struct tcpcb *tp; 1289 { 1290 register short delta; 1291 1292 tcpstat.tcps_rttupdated++; 1293 if (tp->t_srtt != 0) { 1294 /* 1295 * srtt is stored as fixed point with 3 bits after the 1296 * binary point (i.e., scaled by 8). The following magic 1297 * is equivalent to the smoothing algorithm in rfc793 with 1298 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1299 * point). Adjust t_rtt to origin 0. 1300 */ 1301 delta = tp->t_rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1302 if ((tp->t_srtt += delta) <= 0) 1303 tp->t_srtt = 1; 1304 /* 1305 * We accumulate a smoothed rtt variance (actually, a 1306 * smoothed mean difference), then set the retransmit 1307 * timer to smoothed rtt + 4 times the smoothed variance. 1308 * rttvar is stored as fixed point with 2 bits after the 1309 * binary point (scaled by 4). The following is 1310 * equivalent to rfc793 smoothing with an alpha of .75 1311 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1312 * rfc793's wired-in beta. 1313 */ 1314 if (delta < 0) 1315 delta = -delta; 1316 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1317 if ((tp->t_rttvar += delta) <= 0) 1318 tp->t_rttvar = 1; 1319 } else { 1320 /* 1321 * No rtt measurement yet - use the unsmoothed rtt. 1322 * Set the variance to half the rtt (so our first 1323 * retransmit happens at 2*rtt) 1324 */ 1325 tp->t_srtt = tp->t_rtt << TCP_RTT_SHIFT; 1326 tp->t_rttvar = tp->t_rtt << (TCP_RTTVAR_SHIFT - 1); 1327 } 1328 tp->t_rtt = 0; 1329 tp->t_rxtshift = 0; 1330 1331 /* 1332 * the retransmit should happen at rtt + 4 * rttvar. 1333 * Because of the way we do the smoothing, srtt and rttvar 1334 * will each average +1/2 tick of bias. When we compute 1335 * the retransmit timer, we want 1/2 tick of rounding and 1336 * 1 extra tick because of +-1/2 tick uncertainty in the 1337 * firing of the timer. The bias will give us exactly the 1338 * 1.5 tick we need. But, because the bias is 1339 * statistical, we have to test that we don't drop below 1340 * the minimum feasible timer (which is 2 ticks). 1341 */ 1342 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1343 tp->t_rttmin, TCPTV_REXMTMAX); 1344 1345 /* 1346 * We received an ack for a packet that wasn't retransmitted; 1347 * it is probably safe to discard any error indications we've 1348 * received recently. This isn't quite right, but close enough 1349 * for now (a route might have failed after we sent a segment, 1350 * and the return path might not be symmetrical). 1351 */ 1352 tp->t_softerror = 0; 1353 } 1354 1355 /* 1356 * Determine a reasonable value for maxseg size. 1357 * If the route is known, check route for mtu. 1358 * If none, use an mss that can be handled on the outgoing 1359 * interface without forcing IP to fragment; if bigger than 1360 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1361 * to utilize large mbufs. If no route is found, route has no mtu, 1362 * or the destination isn't local, use a default, hopefully conservative 1363 * size (usually 512 or the default IP max size, but no more than the mtu 1364 * of the interface), as we can't discover anything about intervening 1365 * gateways or networks. We also initialize the congestion/slow start 1366 * window to be a single segment if the destination isn't local. 1367 * While looking at the routing entry, we also initialize other path-dependent 1368 * parameters from pre-set or cached values in the routing entry. 1369 */ 1370 int 1371 tcp_mss(tp, offer) 1372 register struct tcpcb *tp; 1373 u_short offer; 1374 { 1375 struct route *ro; 1376 register struct rtentry *rt; 1377 struct ifnet *ifp; 1378 register int rtt, mss; 1379 u_long bufsize; 1380 struct inpcb *inp; 1381 struct socket *so; 1382 extern int tcp_mssdflt, tcp_rttdflt; 1383 1384 inp = tp->t_inpcb; 1385 ro = &inp->inp_route; 1386 1387 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1388 /* No route yet, so try to acquire one */ 1389 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1390 ro->ro_dst.sa_family = AF_INET; 1391 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1392 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1393 inp->inp_faddr; 1394 rtalloc(ro); 1395 } 1396 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1397 return (tcp_mssdflt); 1398 } 1399 ifp = rt->rt_ifp; 1400 so = inp->inp_socket; 1401 1402 #ifdef RTV_MTU /* if route characteristics exist ... */ 1403 /* 1404 * While we're here, check if there's an initial rtt 1405 * or rttvar. Convert from the route-table units 1406 * to scaled multiples of the slow timeout timer. 1407 */ 1408 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1409 if (rt->rt_rmx.rmx_locks & RTV_MTU) 1410 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1411 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1412 if (rt->rt_rmx.rmx_rttvar) 1413 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1414 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1415 else 1416 /* default variation is +- 1 rtt */ 1417 tp->t_rttvar = 1418 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1419 TCPT_RANGESET(tp->t_rxtcur, 1420 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1421 tp->t_rttmin, TCPTV_REXMTMAX); 1422 } 1423 /* 1424 * if there's an mtu associated with the route, use it 1425 */ 1426 if (rt->rt_rmx.rmx_mtu) 1427 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1428 else 1429 #endif /* RTV_MTU */ 1430 { 1431 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1432 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1433 if (mss > MCLBYTES) 1434 mss &= ~(MCLBYTES-1); 1435 #else 1436 if (mss > MCLBYTES) 1437 mss = mss / MCLBYTES * MCLBYTES; 1438 #endif 1439 if (!in_localaddr(inp->inp_faddr)) 1440 mss = min(mss, tcp_mssdflt); 1441 } 1442 /* 1443 * The current mss, t_maxseg, is initialized to the default value. 1444 * If we compute a smaller value, reduce the current mss. 1445 * If we compute a larger value, return it for use in sending 1446 * a max seg size option, but don't store it for use 1447 * unless we received an offer at least that large from peer. 1448 * However, do not accept offers under 32 bytes. 1449 */ 1450 if (offer) 1451 mss = min(mss, offer); 1452 mss = max(mss, 32); /* sanity */ 1453 if (mss < tp->t_maxseg || offer != 0) { 1454 /* 1455 * If there's a pipesize, change the socket buffer 1456 * to that size. Make the socket buffers an integral 1457 * number of mss units; if the mss is larger than 1458 * the socket buffer, decrease the mss. 1459 */ 1460 #ifdef RTV_SPIPE 1461 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1462 #endif 1463 bufsize = so->so_snd.sb_hiwat; 1464 if (bufsize < mss) 1465 mss = bufsize; 1466 else { 1467 bufsize = min(bufsize, SB_MAX) / mss * mss; 1468 (void) sbreserve(&so->so_snd, bufsize); 1469 } 1470 tp->t_maxseg = mss; 1471 1472 #ifdef RTV_RPIPE 1473 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1474 #endif 1475 bufsize = so->so_rcv.sb_hiwat; 1476 if (bufsize > mss) { 1477 bufsize = min(bufsize, SB_MAX) / mss * mss; 1478 (void) sbreserve(&so->so_rcv, bufsize); 1479 } 1480 } 1481 tp->snd_cwnd = mss; 1482 1483 #ifdef RTV_SSTHRESH 1484 if (rt->rt_rmx.rmx_ssthresh) { 1485 /* 1486 * There's some sort of gateway or interface 1487 * buffer limit on the path. Use this to set 1488 * the slow start threshhold, but set the 1489 * threshold to no less than 2*mss. 1490 */ 1491 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1492 } 1493 #endif /* RTV_MTU */ 1494 return (mss); 1495 } 1496