tcp_input.c revision 1817
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 3 * The Regents of the University of California. 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 * @(#)tcp_input.c 8.5 (Berkeley) 4/10/94 34 * $Id$ 35 */ 36 37#ifndef TUBA_INCLUDE 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/malloc.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 63int tcprexmtthresh = 3; 64struct tcpiphdr tcp_saveti; 65struct inpcb *tcp_last_inpcb = &tcb; 66 67extern u_long sb_max; 68 69#endif /* TUBA_INCLUDE */ 70#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 71 72/* for modulo comparisons of timestamps */ 73#define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 74#define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 75 76 77/* 78 * Insert segment ti into reassembly queue of tcp with 79 * control block tp. Return TH_FIN if reassembly now includes 80 * a segment with FIN. The macro form does the common case inline 81 * (segment is the next to be received on an established connection, 82 * and the queue is empty), avoiding linkage into and removal 83 * from the queue and repetition of various conversions. 84 * Set DELACK for segments received in order, but ack immediately 85 * when segments are out of order (so fast retransmit can work). 86 */ 87#define TCP_REASS(tp, ti, m, so, flags) { \ 88 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 89 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 90 (tp)->t_state == TCPS_ESTABLISHED) { \ 91 tp->t_flags |= TF_DELACK; \ 92 (tp)->rcv_nxt += (ti)->ti_len; \ 93 flags = (ti)->ti_flags & TH_FIN; \ 94 tcpstat.tcps_rcvpack++;\ 95 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 96 sbappend(&(so)->so_rcv, (m)); \ 97 sorwakeup(so); \ 98 } else { \ 99 (flags) = tcp_reass((tp), (ti), (m)); \ 100 tp->t_flags |= TF_ACKNOW; \ 101 } \ 102} 103#ifndef TUBA_INCLUDE 104 105int 106tcp_reass(tp, ti, m) 107 register struct tcpcb *tp; 108 register struct tcpiphdr *ti; 109 struct mbuf *m; 110{ 111 register struct tcpiphdr *q; 112 struct socket *so = tp->t_inpcb->inp_socket; 113 int flags; 114 115 /* 116 * Call with ti==0 after become established to 117 * force pre-ESTABLISHED data up to user socket. 118 */ 119 if (ti == 0) 120 goto present; 121 122 /* 123 * Find a segment which begins after this one does. 124 */ 125 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 126 q = (struct tcpiphdr *)q->ti_next) 127 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 128 break; 129 130 /* 131 * If there is a preceding segment, it may provide some of 132 * our data already. If so, drop the data from the incoming 133 * segment. If it provides all of our data, drop us. 134 */ 135 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 136 register int i; 137 q = (struct tcpiphdr *)q->ti_prev; 138 /* conversion to int (in i) handles seq wraparound */ 139 i = q->ti_seq + q->ti_len - ti->ti_seq; 140 if (i > 0) { 141 if (i >= ti->ti_len) { 142 tcpstat.tcps_rcvduppack++; 143 tcpstat.tcps_rcvdupbyte += ti->ti_len; 144 m_freem(m); 145 return (0); 146 } 147 m_adj(m, i); 148 ti->ti_len -= i; 149 ti->ti_seq += i; 150 } 151 q = (struct tcpiphdr *)(q->ti_next); 152 } 153 tcpstat.tcps_rcvoopack++; 154 tcpstat.tcps_rcvoobyte += ti->ti_len; 155 REASS_MBUF(ti) = m; /* XXX */ 156 157 /* 158 * While we overlap succeeding segments trim them or, 159 * if they are completely covered, dequeue them. 160 */ 161 while (q != (struct tcpiphdr *)tp) { 162 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 163 if (i <= 0) 164 break; 165 if (i < q->ti_len) { 166 q->ti_seq += i; 167 q->ti_len -= i; 168 m_adj(REASS_MBUF(q), i); 169 break; 170 } 171 q = (struct tcpiphdr *)q->ti_next; 172 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 173 remque(q->ti_prev); 174 m_freem(m); 175 } 176 177 /* 178 * Stick new segment in its place. 179 */ 180 insque(ti, q->ti_prev); 181 182present: 183 /* 184 * Present data to user, advancing rcv_nxt through 185 * completed sequence space. 186 */ 187 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 188 return (0); 189 ti = tp->seg_next; 190 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 191 return (0); 192 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 193 return (0); 194 do { 195 tp->rcv_nxt += ti->ti_len; 196 flags = ti->ti_flags & TH_FIN; 197 remque(ti); 198 m = REASS_MBUF(ti); 199 ti = (struct tcpiphdr *)ti->ti_next; 200 if (so->so_state & SS_CANTRCVMORE) 201 m_freem(m); 202 else 203 sbappend(&so->so_rcv, m); 204 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 205 sorwakeup(so); 206 return (flags); 207} 208 209/* 210 * TCP input routine, follows pages 65-76 of the 211 * protocol specification dated September, 1981 very closely. 212 */ 213void 214tcp_input(m, iphlen) 215 register struct mbuf *m; 216 int iphlen; 217{ 218 register struct tcpiphdr *ti; 219 register struct inpcb *inp; 220 caddr_t optp = NULL; 221 int optlen = 0; 222 int len, tlen, off; 223 register struct tcpcb *tp = 0; 224 register int tiflags; 225 struct socket *so = 0; 226 int todrop, acked, ourfinisacked, needoutput = 0; 227 short ostate = 0; 228 struct in_addr laddr; 229 int dropsocket = 0; 230 int iss = 0; 231 u_long tiwin, ts_val, ts_ecr; 232 int ts_present = 0; 233 234 tcpstat.tcps_rcvtotal++; 235 /* 236 * Get IP and TCP header together in first mbuf. 237 * Note: IP leaves IP header in first mbuf. 238 */ 239 ti = mtod(m, struct tcpiphdr *); 240 if (iphlen > sizeof (struct ip)) 241 ip_stripoptions(m, (struct mbuf *)0); 242 if (m->m_len < sizeof (struct tcpiphdr)) { 243 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 244 tcpstat.tcps_rcvshort++; 245 return; 246 } 247 ti = mtod(m, struct tcpiphdr *); 248 } 249 250 /* 251 * Checksum extended TCP header and data. 252 */ 253 tlen = ((struct ip *)ti)->ip_len; 254 len = sizeof (struct ip) + tlen; 255 ti->ti_next = ti->ti_prev = 0; 256 ti->ti_x1 = 0; 257 ti->ti_len = (u_short)tlen; 258 HTONS(ti->ti_len); 259 if (ti->ti_sum = in_cksum(m, len)) { 260 tcpstat.tcps_rcvbadsum++; 261 goto drop; 262 } 263#endif /* TUBA_INCLUDE */ 264 265 /* 266 * Check that TCP offset makes sense, 267 * pull out TCP options and adjust length. XXX 268 */ 269 off = ti->ti_off << 2; 270 if (off < sizeof (struct tcphdr) || off > tlen) { 271 tcpstat.tcps_rcvbadoff++; 272 goto drop; 273 } 274 tlen -= off; 275 ti->ti_len = tlen; 276 if (off > sizeof (struct tcphdr)) { 277 if (m->m_len < sizeof(struct ip) + off) { 278 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 279 tcpstat.tcps_rcvshort++; 280 return; 281 } 282 ti = mtod(m, struct tcpiphdr *); 283 } 284 optlen = off - sizeof (struct tcphdr); 285 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 286 /* 287 * Do quick retrieval of timestamp options ("options 288 * prediction?"). If timestamp is the only option and it's 289 * formatted as recommended in RFC 1323 appendix A, we 290 * quickly get the values now and not bother calling 291 * tcp_dooptions(), etc. 292 */ 293 if ((optlen == TCPOLEN_TSTAMP_APPA || 294 (optlen > TCPOLEN_TSTAMP_APPA && 295 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 296 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 297 (ti->ti_flags & TH_SYN) == 0) { 298 ts_present = 1; 299 ts_val = ntohl(*(u_long *)(optp + 4)); 300 ts_ecr = ntohl(*(u_long *)(optp + 8)); 301 optp = NULL; /* we've parsed the options */ 302 } 303 } 304 tiflags = ti->ti_flags; 305 306 /* 307 * Convert TCP protocol specific fields to host format. 308 */ 309 NTOHL(ti->ti_seq); 310 NTOHL(ti->ti_ack); 311 NTOHS(ti->ti_win); 312 NTOHS(ti->ti_urp); 313 314 /* 315 * Locate pcb for segment. 316 */ 317findpcb: 318 inp = tcp_last_inpcb; 319 if (inp->inp_lport != ti->ti_dport || 320 inp->inp_fport != ti->ti_sport || 321 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 322 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 323 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 324 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 325 if (inp) 326 tcp_last_inpcb = inp; 327 ++tcpstat.tcps_pcbcachemiss; 328 } 329 330 /* 331 * If the state is CLOSED (i.e., TCB does not exist) then 332 * all data in the incoming segment is discarded. 333 * If the TCB exists but is in CLOSED state, it is embryonic, 334 * but should either do a listen or a connect soon. 335 */ 336 if (inp == 0) 337 goto dropwithreset; 338 tp = intotcpcb(inp); 339 if (tp == 0) 340 goto dropwithreset; 341 if (tp->t_state == TCPS_CLOSED) 342 goto drop; 343 344 /* Unscale the window into a 32-bit value. */ 345 if ((tiflags & TH_SYN) == 0) 346 tiwin = ti->ti_win << tp->snd_scale; 347 else 348 tiwin = ti->ti_win; 349 350 so = inp->inp_socket; 351 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 352 if (so->so_options & SO_DEBUG) { 353 ostate = tp->t_state; 354 tcp_saveti = *ti; 355 } 356 if (so->so_options & SO_ACCEPTCONN) { 357 so = sonewconn(so, 0); 358 if (so == 0) 359 goto drop; 360 /* 361 * This is ugly, but .... 362 * 363 * Mark socket as temporary until we're 364 * committed to keeping it. The code at 365 * ``drop'' and ``dropwithreset'' check the 366 * flag dropsocket to see if the temporary 367 * socket created here should be discarded. 368 * We mark the socket as discardable until 369 * we're committed to it below in TCPS_LISTEN. 370 */ 371 dropsocket++; 372 inp = (struct inpcb *)so->so_pcb; 373 inp->inp_laddr = ti->ti_dst; 374 inp->inp_lport = ti->ti_dport; 375#if BSD>=43 376 inp->inp_options = ip_srcroute(); 377#endif 378 tp = intotcpcb(inp); 379 tp->t_state = TCPS_LISTEN; 380 381 /* Compute proper scaling value from buffer space 382 */ 383 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 384 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 385 tp->request_r_scale++; 386 } 387 } 388 389 /* 390 * Segment received on connection. 391 * Reset idle time and keep-alive timer. 392 */ 393 tp->t_idle = 0; 394 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 395 396 /* 397 * Process options if not in LISTEN state, 398 * else do it below (after getting remote address). 399 */ 400 if (optp && tp->t_state != TCPS_LISTEN) 401 tcp_dooptions(tp, optp, optlen, ti, 402 &ts_present, &ts_val, &ts_ecr); 403 404 /* 405 * Header prediction: check for the two common cases 406 * of a uni-directional data xfer. If the packet has 407 * no control flags, is in-sequence, the window didn't 408 * change and we're not retransmitting, it's a 409 * candidate. If the length is zero and the ack moved 410 * forward, we're the sender side of the xfer. Just 411 * free the data acked & wake any higher level process 412 * that was blocked waiting for space. If the length 413 * is non-zero and the ack didn't move, we're the 414 * receiver side. If we're getting packets in-order 415 * (the reassembly queue is empty), add the data to 416 * the socket buffer and note that we need a delayed ack. 417 */ 418 if (tp->t_state == TCPS_ESTABLISHED && 419 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 420 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 421 ti->ti_seq == tp->rcv_nxt && 422 tiwin && tiwin == tp->snd_wnd && 423 tp->snd_nxt == tp->snd_max) { 424 425 /* 426 * If last ACK falls within this segment's sequence numbers, 427 * record the timestamp. 428 */ 429 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 430 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 431 tp->ts_recent_age = tcp_now; 432 tp->ts_recent = ts_val; 433 } 434 435 if (ti->ti_len == 0) { 436 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 437 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 438 tp->snd_cwnd >= tp->snd_wnd) { 439 /* 440 * this is a pure ack for outstanding data. 441 */ 442 ++tcpstat.tcps_predack; 443 if (ts_present) 444 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 445 else if (tp->t_rtt && 446 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 447 tcp_xmit_timer(tp, tp->t_rtt); 448 acked = ti->ti_ack - tp->snd_una; 449 tcpstat.tcps_rcvackpack++; 450 tcpstat.tcps_rcvackbyte += acked; 451 sbdrop(&so->so_snd, acked); 452 tp->snd_una = ti->ti_ack; 453 m_freem(m); 454 455 /* 456 * If all outstanding data are acked, stop 457 * retransmit timer, otherwise restart timer 458 * using current (possibly backed-off) value. 459 * If process is waiting for space, 460 * wakeup/selwakeup/signal. If data 461 * are ready to send, let tcp_output 462 * decide between more output or persist. 463 */ 464 if (tp->snd_una == tp->snd_max) 465 tp->t_timer[TCPT_REXMT] = 0; 466 else if (tp->t_timer[TCPT_PERSIST] == 0) 467 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 468 469 if (so->so_snd.sb_flags & SB_NOTIFY) 470 sowwakeup(so); 471 if (so->so_snd.sb_cc) 472 (void) tcp_output(tp); 473 return; 474 } 475 } else if (ti->ti_ack == tp->snd_una && 476 tp->seg_next == (struct tcpiphdr *)tp && 477 ti->ti_len <= sbspace(&so->so_rcv)) { 478 /* 479 * this is a pure, in-sequence data packet 480 * with nothing on the reassembly queue and 481 * we have enough buffer space to take it. 482 */ 483 ++tcpstat.tcps_preddat; 484 tp->rcv_nxt += ti->ti_len; 485 tcpstat.tcps_rcvpack++; 486 tcpstat.tcps_rcvbyte += ti->ti_len; 487 /* 488 * Drop TCP, IP headers and TCP options then add data 489 * to socket buffer. 490 */ 491 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 492 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 493 sbappend(&so->so_rcv, m); 494 sorwakeup(so); 495 /* 496 * If this is a small packet, then ACK now - with Nagel 497 * congestion avoidance sender won't send more until 498 * he gets an ACK. 499 */ 500 if ((unsigned)ti->ti_len < tp->t_maxseg) { 501 tp->t_flags |= TF_ACKNOW; 502 tcp_output(tp); 503 } else { 504 tp->t_flags |= TF_DELACK; 505 } 506 return; 507 } 508 } 509 510 /* 511 * Drop TCP, IP headers and TCP options. 512 */ 513 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 514 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 515 516 /* 517 * Calculate amount of space in receive window, 518 * and then do TCP input processing. 519 * Receive window is amount of space in rcv queue, 520 * but not less than advertised window. 521 */ 522 { int win; 523 524 win = sbspace(&so->so_rcv); 525 if (win < 0) 526 win = 0; 527 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 528 } 529 530 switch (tp->t_state) { 531 532 /* 533 * If the state is LISTEN then ignore segment if it contains an RST. 534 * If the segment contains an ACK then it is bad and send a RST. 535 * If it does not contain a SYN then it is not interesting; drop it. 536 * Don't bother responding if the destination was a broadcast. 537 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 538 * tp->iss, and send a segment: 539 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 540 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 541 * Fill in remote peer address fields if not previously specified. 542 * Enter SYN_RECEIVED state, and process any other fields of this 543 * segment in this state. 544 */ 545 case TCPS_LISTEN: { 546 struct mbuf *am; 547 register struct sockaddr_in *sin; 548 549 if (tiflags & TH_RST) 550 goto drop; 551 if (tiflags & TH_ACK) 552 goto dropwithreset; 553 if ((tiflags & TH_SYN) == 0) 554 goto drop; 555 /* 556 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 557 * in_broadcast() should never return true on a received 558 * packet with M_BCAST not set. 559 */ 560 if (m->m_flags & (M_BCAST|M_MCAST) || 561 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 562 goto drop; 563 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 564 if (am == NULL) 565 goto drop; 566 am->m_len = sizeof (struct sockaddr_in); 567 sin = mtod(am, struct sockaddr_in *); 568 sin->sin_family = AF_INET; 569 sin->sin_len = sizeof(*sin); 570 sin->sin_addr = ti->ti_src; 571 sin->sin_port = ti->ti_sport; 572 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 573 laddr = inp->inp_laddr; 574 if (inp->inp_laddr.s_addr == INADDR_ANY) 575 inp->inp_laddr = ti->ti_dst; 576 if (in_pcbconnect(inp, am)) { 577 inp->inp_laddr = laddr; 578 (void) m_free(am); 579 goto drop; 580 } 581 (void) m_free(am); 582 tp->t_template = tcp_template(tp); 583 if (tp->t_template == 0) { 584 tp = tcp_drop(tp, ENOBUFS); 585 dropsocket = 0; /* socket is already gone */ 586 goto drop; 587 } 588 if (optp) 589 tcp_dooptions(tp, optp, optlen, ti, 590 &ts_present, &ts_val, &ts_ecr); 591 if (iss) 592 tp->iss = iss; 593 else 594 tp->iss = tcp_iss; 595 tcp_iss += TCP_ISSINCR/2; 596 tp->irs = ti->ti_seq; 597 tcp_sendseqinit(tp); 598 tcp_rcvseqinit(tp); 599 tp->t_flags |= TF_ACKNOW; 600 tp->t_state = TCPS_SYN_RECEIVED; 601 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 602 dropsocket = 0; /* committed to socket */ 603 tcpstat.tcps_accepts++; 604 goto trimthenstep6; 605 } 606 607 /* 608 * If the state is SYN_SENT: 609 * if seg contains an ACK, but not for our SYN, drop the input. 610 * if seg contains a RST, then drop the connection. 611 * if seg does not contain SYN, then drop it. 612 * Otherwise this is an acceptable SYN segment 613 * initialize tp->rcv_nxt and tp->irs 614 * if seg contains ack then advance tp->snd_una 615 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 616 * arrange for segment to be acked (eventually) 617 * continue processing rest of data/controls, beginning with URG 618 */ 619 case TCPS_SYN_SENT: 620 if ((tiflags & TH_ACK) && 621 (SEQ_LEQ(ti->ti_ack, tp->iss) || 622 SEQ_GT(ti->ti_ack, tp->snd_max))) 623 goto dropwithreset; 624 if (tiflags & TH_RST) { 625 if (tiflags & TH_ACK) 626 tp = tcp_drop(tp, ECONNREFUSED); 627 goto drop; 628 } 629 if ((tiflags & TH_SYN) == 0) 630 goto drop; 631 if (tiflags & TH_ACK) { 632 tp->snd_una = ti->ti_ack; 633 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 634 tp->snd_nxt = tp->snd_una; 635 } 636 tp->t_timer[TCPT_REXMT] = 0; 637 tp->irs = ti->ti_seq; 638 tcp_rcvseqinit(tp); 639 tp->t_flags |= TF_ACKNOW; 640 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 641 tcpstat.tcps_connects++; 642 soisconnected(so); 643 tp->t_state = TCPS_ESTABLISHED; 644 /* Do window scaling on this connection? */ 645 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 646 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 647 tp->snd_scale = tp->requested_s_scale; 648 tp->rcv_scale = tp->request_r_scale; 649 } 650 (void) tcp_reass(tp, (struct tcpiphdr *)0, 651 (struct mbuf *)0); 652 /* 653 * if we didn't have to retransmit the SYN, 654 * use its rtt as our initial srtt & rtt var. 655 */ 656 if (tp->t_rtt) 657 tcp_xmit_timer(tp, tp->t_rtt); 658 } else 659 tp->t_state = TCPS_SYN_RECEIVED; 660 661trimthenstep6: 662 /* 663 * Advance ti->ti_seq to correspond to first data byte. 664 * If data, trim to stay within window, 665 * dropping FIN if necessary. 666 */ 667 ti->ti_seq++; 668 if (ti->ti_len > tp->rcv_wnd) { 669 todrop = ti->ti_len - tp->rcv_wnd; 670 m_adj(m, -todrop); 671 ti->ti_len = tp->rcv_wnd; 672 tiflags &= ~TH_FIN; 673 tcpstat.tcps_rcvpackafterwin++; 674 tcpstat.tcps_rcvbyteafterwin += todrop; 675 } 676 tp->snd_wl1 = ti->ti_seq - 1; 677 tp->rcv_up = ti->ti_seq; 678 goto step6; 679 } 680 681 /* 682 * States other than LISTEN or SYN_SENT. 683 * First check timestamp, if present. 684 * Then check that at least some bytes of segment are within 685 * receive window. If segment begins before rcv_nxt, 686 * drop leading data (and SYN); if nothing left, just ack. 687 * 688 * RFC 1323 PAWS: If we have a timestamp reply on this segment 689 * and it's less than ts_recent, drop it. 690 */ 691 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 692 TSTMP_LT(ts_val, tp->ts_recent)) { 693 694 /* Check to see if ts_recent is over 24 days old. */ 695 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 696 /* 697 * Invalidate ts_recent. If this segment updates 698 * ts_recent, the age will be reset later and ts_recent 699 * will get a valid value. If it does not, setting 700 * ts_recent to zero will at least satisfy the 701 * requirement that zero be placed in the timestamp 702 * echo reply when ts_recent isn't valid. The 703 * age isn't reset until we get a valid ts_recent 704 * because we don't want out-of-order segments to be 705 * dropped when ts_recent is old. 706 */ 707 tp->ts_recent = 0; 708 } else { 709 tcpstat.tcps_rcvduppack++; 710 tcpstat.tcps_rcvdupbyte += ti->ti_len; 711 tcpstat.tcps_pawsdrop++; 712 goto dropafterack; 713 } 714 } 715 716 todrop = tp->rcv_nxt - ti->ti_seq; 717 if (todrop > 0) { 718 if (tiflags & TH_SYN) { 719 tiflags &= ~TH_SYN; 720 ti->ti_seq++; 721 if (ti->ti_urp > 1) 722 ti->ti_urp--; 723 else 724 tiflags &= ~TH_URG; 725 todrop--; 726 } 727 if (todrop >= ti->ti_len) { 728 tcpstat.tcps_rcvduppack++; 729 tcpstat.tcps_rcvdupbyte += ti->ti_len; 730 /* 731 * If segment is just one to the left of the window, 732 * check two special cases: 733 * 1. Don't toss RST in response to 4.2-style keepalive. 734 * 2. If the only thing to drop is a FIN, we can drop 735 * it, but check the ACK or we will get into FIN 736 * wars if our FINs crossed (both CLOSING). 737 * In either case, send ACK to resynchronize, 738 * but keep on processing for RST or ACK. 739 */ 740 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 741#ifdef TCP_COMPAT_42 742 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 743#endif 744 ) { 745 todrop = ti->ti_len; 746 tiflags &= ~TH_FIN; 747 tp->t_flags |= TF_ACKNOW; 748 } else { 749 /* 750 * Handle the case when a bound socket connects 751 * to itself. Allow packets with a SYN and 752 * an ACK to continue with the processing. 753 */ 754 if (todrop != 0 || (tiflags & TH_ACK) == 0) 755 goto dropafterack; 756 } 757 } else { 758 tcpstat.tcps_rcvpartduppack++; 759 tcpstat.tcps_rcvpartdupbyte += todrop; 760 } 761 m_adj(m, todrop); 762 ti->ti_seq += todrop; 763 ti->ti_len -= todrop; 764 if (ti->ti_urp > todrop) 765 ti->ti_urp -= todrop; 766 else { 767 tiflags &= ~TH_URG; 768 ti->ti_urp = 0; 769 } 770 } 771 772 /* 773 * If new data are received on a connection after the 774 * user processes are gone, then RST the other end. 775 */ 776 if ((so->so_state & SS_NOFDREF) && 777 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 778 tp = tcp_close(tp); 779 tcpstat.tcps_rcvafterclose++; 780 goto dropwithreset; 781 } 782 783 /* 784 * If segment ends after window, drop trailing data 785 * (and PUSH and FIN); if nothing left, just ACK. 786 */ 787 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 788 if (todrop > 0) { 789 tcpstat.tcps_rcvpackafterwin++; 790 if (todrop >= ti->ti_len) { 791 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 792 /* 793 * If a new connection request is received 794 * while in TIME_WAIT, drop the old connection 795 * and start over if the sequence numbers 796 * are above the previous ones. 797 */ 798 if (tiflags & TH_SYN && 799 tp->t_state == TCPS_TIME_WAIT && 800 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 801 iss = tp->rcv_nxt + TCP_ISSINCR; 802 tp = tcp_close(tp); 803 goto findpcb; 804 } 805 /* 806 * If window is closed can only take segments at 807 * window edge, and have to drop data and PUSH from 808 * incoming segments. Continue processing, but 809 * remember to ack. Otherwise, drop segment 810 * and ack. 811 */ 812 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 813 tp->t_flags |= TF_ACKNOW; 814 tcpstat.tcps_rcvwinprobe++; 815 } else 816 goto dropafterack; 817 } else 818 tcpstat.tcps_rcvbyteafterwin += todrop; 819 m_adj(m, -todrop); 820 ti->ti_len -= todrop; 821 tiflags &= ~(TH_PUSH|TH_FIN); 822 } 823 824 /* 825 * If last ACK falls within this segment's sequence numbers, 826 * record its timestamp. 827 */ 828 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 829 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 830 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 831 tp->ts_recent_age = tcp_now; 832 tp->ts_recent = ts_val; 833 } 834 835 /* 836 * If the RST bit is set examine the state: 837 * SYN_RECEIVED STATE: 838 * If passive open, return to LISTEN state. 839 * If active open, inform user that connection was refused. 840 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 841 * Inform user that connection was reset, and close tcb. 842 * CLOSING, LAST_ACK, TIME_WAIT STATES 843 * Close the tcb. 844 */ 845 if (tiflags&TH_RST) switch (tp->t_state) { 846 847 case TCPS_SYN_RECEIVED: 848 so->so_error = ECONNREFUSED; 849 goto close; 850 851 case TCPS_ESTABLISHED: 852 case TCPS_FIN_WAIT_1: 853 case TCPS_FIN_WAIT_2: 854 case TCPS_CLOSE_WAIT: 855 so->so_error = ECONNRESET; 856 close: 857 tp->t_state = TCPS_CLOSED; 858 tcpstat.tcps_drops++; 859 tp = tcp_close(tp); 860 goto drop; 861 862 case TCPS_CLOSING: 863 case TCPS_LAST_ACK: 864 case TCPS_TIME_WAIT: 865 tp = tcp_close(tp); 866 goto drop; 867 } 868 869 /* 870 * If a SYN is in the window, then this is an 871 * error and we send an RST and drop the connection. 872 */ 873 if (tiflags & TH_SYN) { 874 tp = tcp_drop(tp, ECONNRESET); 875 goto dropwithreset; 876 } 877 878 /* 879 * If the ACK bit is off we drop the segment and return. 880 */ 881 if ((tiflags & TH_ACK) == 0) 882 goto drop; 883 884 /* 885 * Ack processing. 886 */ 887 switch (tp->t_state) { 888 889 /* 890 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 891 * ESTABLISHED state and continue processing, otherwise 892 * send an RST. 893 */ 894 case TCPS_SYN_RECEIVED: 895 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 896 SEQ_GT(ti->ti_ack, tp->snd_max)) 897 goto dropwithreset; 898 tcpstat.tcps_connects++; 899 soisconnected(so); 900 tp->t_state = TCPS_ESTABLISHED; 901 /* Do window scaling? */ 902 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 903 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 904 tp->snd_scale = tp->requested_s_scale; 905 tp->rcv_scale = tp->request_r_scale; 906 } 907 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 908 tp->snd_wl1 = ti->ti_seq - 1; 909 /* fall into ... */ 910 911 /* 912 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 913 * ACKs. If the ack is in the range 914 * tp->snd_una < ti->ti_ack <= tp->snd_max 915 * then advance tp->snd_una to ti->ti_ack and drop 916 * data from the retransmission queue. If this ACK reflects 917 * more up to date window information we update our window information. 918 */ 919 case TCPS_ESTABLISHED: 920 case TCPS_FIN_WAIT_1: 921 case TCPS_FIN_WAIT_2: 922 case TCPS_CLOSE_WAIT: 923 case TCPS_CLOSING: 924 case TCPS_LAST_ACK: 925 case TCPS_TIME_WAIT: 926 927 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 928 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 929 tcpstat.tcps_rcvdupack++; 930 /* 931 * If we have outstanding data (other than 932 * a window probe), this is a completely 933 * duplicate ack (ie, window info didn't 934 * change), the ack is the biggest we've 935 * seen and we've seen exactly our rexmt 936 * threshhold of them, assume a packet 937 * has been dropped and retransmit it. 938 * Kludge snd_nxt & the congestion 939 * window so we send only this one 940 * packet. 941 * 942 * We know we're losing at the current 943 * window size so do congestion avoidance 944 * (set ssthresh to half the current window 945 * and pull our congestion window back to 946 * the new ssthresh). 947 * 948 * Dup acks mean that packets have left the 949 * network (they're now cached at the receiver) 950 * so bump cwnd by the amount in the receiver 951 * to keep a constant cwnd packets in the 952 * network. 953 */ 954 if (tp->t_timer[TCPT_REXMT] == 0 || 955 ti->ti_ack != tp->snd_una) 956 tp->t_dupacks = 0; 957 else if (++tp->t_dupacks == tcprexmtthresh) { 958 tcp_seq onxt = tp->snd_nxt; 959 u_int win = 960 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 961 tp->t_maxseg; 962 963 if (win < 2) 964 win = 2; 965 tp->snd_ssthresh = win * tp->t_maxseg; 966 tp->t_timer[TCPT_REXMT] = 0; 967 tp->t_rtt = 0; 968 tp->snd_nxt = ti->ti_ack; 969 tp->snd_cwnd = tp->t_maxseg; 970 (void) tcp_output(tp); 971 tp->snd_cwnd = tp->snd_ssthresh + 972 tp->t_maxseg * tp->t_dupacks; 973 if (SEQ_GT(onxt, tp->snd_nxt)) 974 tp->snd_nxt = onxt; 975 goto drop; 976 } else if (tp->t_dupacks > tcprexmtthresh) { 977 tp->snd_cwnd += tp->t_maxseg; 978 (void) tcp_output(tp); 979 goto drop; 980 } 981 } else 982 tp->t_dupacks = 0; 983 break; 984 } 985 /* 986 * If the congestion window was inflated to account 987 * for the other side's cached packets, retract it. 988 */ 989 if (tp->t_dupacks > tcprexmtthresh && 990 tp->snd_cwnd > tp->snd_ssthresh) 991 tp->snd_cwnd = tp->snd_ssthresh; 992 tp->t_dupacks = 0; 993 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 994 tcpstat.tcps_rcvacktoomuch++; 995 goto dropafterack; 996 } 997 acked = ti->ti_ack - tp->snd_una; 998 tcpstat.tcps_rcvackpack++; 999 tcpstat.tcps_rcvackbyte += acked; 1000 1001 /* 1002 * If we have a timestamp reply, update smoothed 1003 * round trip time. If no timestamp is present but 1004 * transmit timer is running and timed sequence 1005 * number was acked, update smoothed round trip time. 1006 * Since we now have an rtt measurement, cancel the 1007 * timer backoff (cf., Phil Karn's retransmit alg.). 1008 * Recompute the initial retransmit timer. 1009 */ 1010 if (ts_present) 1011 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 1012 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1013 tcp_xmit_timer(tp,tp->t_rtt); 1014 1015 /* 1016 * If all outstanding data is acked, stop retransmit 1017 * timer and remember to restart (more output or persist). 1018 * If there is more data to be acked, restart retransmit 1019 * timer, using current (possibly backed-off) value. 1020 */ 1021 if (ti->ti_ack == tp->snd_max) { 1022 tp->t_timer[TCPT_REXMT] = 0; 1023 needoutput = 1; 1024 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1025 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1026 /* 1027 * When new data is acked, open the congestion window. 1028 * If the window gives us less than ssthresh packets 1029 * in flight, open exponentially (maxseg per packet). 1030 * Otherwise open linearly: maxseg per window 1031 * (maxseg^2 / cwnd per packet), plus a constant 1032 * fraction of a packet (maxseg/8) to help larger windows 1033 * open quickly enough. 1034 */ 1035 { 1036 register u_int cw = tp->snd_cwnd; 1037 register u_int incr = tp->t_maxseg; 1038 1039 if (cw > tp->snd_ssthresh) 1040 incr = incr * incr / cw + incr / 8; 1041 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1042 } 1043 if (acked > so->so_snd.sb_cc) { 1044 tp->snd_wnd -= so->so_snd.sb_cc; 1045 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1046 ourfinisacked = 1; 1047 } else { 1048 sbdrop(&so->so_snd, acked); 1049 tp->snd_wnd -= acked; 1050 ourfinisacked = 0; 1051 } 1052 if (so->so_snd.sb_flags & SB_NOTIFY) 1053 sowwakeup(so); 1054 tp->snd_una = ti->ti_ack; 1055 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1056 tp->snd_nxt = tp->snd_una; 1057 1058 switch (tp->t_state) { 1059 1060 /* 1061 * In FIN_WAIT_1 STATE in addition to the processing 1062 * for the ESTABLISHED state if our FIN is now acknowledged 1063 * then enter FIN_WAIT_2. 1064 */ 1065 case TCPS_FIN_WAIT_1: 1066 if (ourfinisacked) { 1067 /* 1068 * If we can't receive any more 1069 * data, then closing user can proceed. 1070 * Starting the timer is contrary to the 1071 * specification, but if we don't get a FIN 1072 * we'll hang forever. 1073 */ 1074 if (so->so_state & SS_CANTRCVMORE) { 1075 soisdisconnected(so); 1076 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1077 } 1078 tp->t_state = TCPS_FIN_WAIT_2; 1079 } 1080 break; 1081 1082 /* 1083 * In CLOSING STATE in addition to the processing for 1084 * the ESTABLISHED state if the ACK acknowledges our FIN 1085 * then enter the TIME-WAIT state, otherwise ignore 1086 * the segment. 1087 */ 1088 case TCPS_CLOSING: 1089 if (ourfinisacked) { 1090 tp->t_state = TCPS_TIME_WAIT; 1091 tcp_canceltimers(tp); 1092 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1093 soisdisconnected(so); 1094 } 1095 break; 1096 1097 /* 1098 * In LAST_ACK, we may still be waiting for data to drain 1099 * and/or to be acked, as well as for the ack of our FIN. 1100 * If our FIN is now acknowledged, delete the TCB, 1101 * enter the closed state and return. 1102 */ 1103 case TCPS_LAST_ACK: 1104 if (ourfinisacked) { 1105 tp = tcp_close(tp); 1106 goto drop; 1107 } 1108 break; 1109 1110 /* 1111 * In TIME_WAIT state the only thing that should arrive 1112 * is a retransmission of the remote FIN. Acknowledge 1113 * it and restart the finack timer. 1114 */ 1115 case TCPS_TIME_WAIT: 1116 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1117 goto dropafterack; 1118 } 1119 } 1120 1121step6: 1122 /* 1123 * Update window information. 1124 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1125 */ 1126 if ((tiflags & TH_ACK) && 1127 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1128 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1129 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1130 /* keep track of pure window updates */ 1131 if (ti->ti_len == 0 && 1132 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1133 tcpstat.tcps_rcvwinupd++; 1134 tp->snd_wnd = tiwin; 1135 tp->snd_wl1 = ti->ti_seq; 1136 tp->snd_wl2 = ti->ti_ack; 1137 if (tp->snd_wnd > tp->max_sndwnd) 1138 tp->max_sndwnd = tp->snd_wnd; 1139 needoutput = 1; 1140 } 1141 1142 /* 1143 * Process segments with URG. 1144 */ 1145 if ((tiflags & TH_URG) && ti->ti_urp && 1146 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1147 /* 1148 * This is a kludge, but if we receive and accept 1149 * random urgent pointers, we'll crash in 1150 * soreceive. It's hard to imagine someone 1151 * actually wanting to send this much urgent data. 1152 */ 1153 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1154 ti->ti_urp = 0; /* XXX */ 1155 tiflags &= ~TH_URG; /* XXX */ 1156 goto dodata; /* XXX */ 1157 } 1158 /* 1159 * If this segment advances the known urgent pointer, 1160 * then mark the data stream. This should not happen 1161 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1162 * a FIN has been received from the remote side. 1163 * In these states we ignore the URG. 1164 * 1165 * According to RFC961 (Assigned Protocols), 1166 * the urgent pointer points to the last octet 1167 * of urgent data. We continue, however, 1168 * to consider it to indicate the first octet 1169 * of data past the urgent section as the original 1170 * spec states (in one of two places). 1171 */ 1172 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1173 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1174 so->so_oobmark = so->so_rcv.sb_cc + 1175 (tp->rcv_up - tp->rcv_nxt) - 1; 1176 if (so->so_oobmark == 0) 1177 so->so_state |= SS_RCVATMARK; 1178 sohasoutofband(so); 1179 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1180 } 1181 /* 1182 * Remove out of band data so doesn't get presented to user. 1183 * This can happen independent of advancing the URG pointer, 1184 * but if two URG's are pending at once, some out-of-band 1185 * data may creep in... ick. 1186 */ 1187 if (ti->ti_urp <= (u_long)ti->ti_len 1188#ifdef SO_OOBINLINE 1189 && (so->so_options & SO_OOBINLINE) == 0 1190#endif 1191 ) 1192 tcp_pulloutofband(so, ti, m); 1193 } else 1194 /* 1195 * If no out of band data is expected, 1196 * pull receive urgent pointer along 1197 * with the receive window. 1198 */ 1199 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1200 tp->rcv_up = tp->rcv_nxt; 1201dodata: /* XXX */ 1202 1203 /* 1204 * Process the segment text, merging it into the TCP sequencing queue, 1205 * and arranging for acknowledgment of receipt if necessary. 1206 * This process logically involves adjusting tp->rcv_wnd as data 1207 * is presented to the user (this happens in tcp_usrreq.c, 1208 * case PRU_RCVD). If a FIN has already been received on this 1209 * connection then we just ignore the text. 1210 */ 1211 if ((ti->ti_len || (tiflags&TH_FIN)) && 1212 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1213 TCP_REASS(tp, ti, m, so, tiflags); 1214 /* 1215 * Note the amount of data that peer has sent into 1216 * our window, in order to estimate the sender's 1217 * buffer size. 1218 */ 1219 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1220 } else { 1221 m_freem(m); 1222 tiflags &= ~TH_FIN; 1223 } 1224 1225 /* 1226 * If FIN is received ACK the FIN and let the user know 1227 * that the connection is closing. 1228 */ 1229 if (tiflags & TH_FIN) { 1230 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1231 socantrcvmore(so); 1232 tp->t_flags |= TF_ACKNOW; 1233 tp->rcv_nxt++; 1234 } 1235 switch (tp->t_state) { 1236 1237 /* 1238 * In SYN_RECEIVED and ESTABLISHED STATES 1239 * enter the CLOSE_WAIT state. 1240 */ 1241 case TCPS_SYN_RECEIVED: 1242 case TCPS_ESTABLISHED: 1243 tp->t_state = TCPS_CLOSE_WAIT; 1244 break; 1245 1246 /* 1247 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1248 * enter the CLOSING state. 1249 */ 1250 case TCPS_FIN_WAIT_1: 1251 tp->t_state = TCPS_CLOSING; 1252 break; 1253 1254 /* 1255 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1256 * starting the time-wait timer, turning off the other 1257 * standard timers. 1258 */ 1259 case TCPS_FIN_WAIT_2: 1260 tp->t_state = TCPS_TIME_WAIT; 1261 tcp_canceltimers(tp); 1262 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1263 soisdisconnected(so); 1264 break; 1265 1266 /* 1267 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1268 */ 1269 case TCPS_TIME_WAIT: 1270 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1271 break; 1272 } 1273 } 1274 if (so->so_options & SO_DEBUG) 1275 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1276 1277 /* 1278 * If this is a small packet, then ACK now - with Nagel 1279 * congestion avoidance sender won't send more until 1280 * he gets an ACK. 1281 */ 1282 if (ti->ti_len && ((unsigned)ti->ti_len < tp->t_maxseg)) 1283 tp->t_flags |= TF_ACKNOW; 1284 1285 /* 1286 * Return any desired output. 1287 */ 1288 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1289 (void) tcp_output(tp); 1290 return; 1291 1292dropafterack: 1293 /* 1294 * Generate an ACK dropping incoming segment if it occupies 1295 * sequence space, where the ACK reflects our state. 1296 */ 1297 if (tiflags & TH_RST) 1298 goto drop; 1299 m_freem(m); 1300 tp->t_flags |= TF_ACKNOW; 1301 (void) tcp_output(tp); 1302 return; 1303 1304dropwithreset: 1305 /* 1306 * Generate a RST, dropping incoming segment. 1307 * Make ACK acceptable to originator of segment. 1308 * Don't bother to respond if destination was broadcast/multicast. 1309 */ 1310 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1311 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1312 goto drop; 1313 if (tiflags & TH_ACK) 1314 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1315 else { 1316 if (tiflags & TH_SYN) 1317 ti->ti_len++; 1318 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1319 TH_RST|TH_ACK); 1320 } 1321 /* destroy temporarily created socket */ 1322 if (dropsocket) 1323 (void) soabort(so); 1324 return; 1325 1326drop: 1327 /* 1328 * Drop space held by incoming segment and return. 1329 */ 1330 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1331 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1332 m_freem(m); 1333 /* destroy temporarily created socket */ 1334 if (dropsocket) 1335 (void) soabort(so); 1336 return; 1337#ifndef TUBA_INCLUDE 1338} 1339 1340void 1341tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1342 struct tcpcb *tp; 1343 u_char *cp; 1344 int cnt; 1345 struct tcpiphdr *ti; 1346 int *ts_present; 1347 u_long *ts_val, *ts_ecr; 1348{ 1349 u_short mss; 1350 int opt, optlen; 1351 1352 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1353 opt = cp[0]; 1354 if (opt == TCPOPT_EOL) 1355 break; 1356 if (opt == TCPOPT_NOP) 1357 optlen = 1; 1358 else { 1359 optlen = cp[1]; 1360 if (optlen <= 0) 1361 break; 1362 } 1363 switch (opt) { 1364 1365 default: 1366 continue; 1367 1368 case TCPOPT_MAXSEG: 1369 if (optlen != TCPOLEN_MAXSEG) 1370 continue; 1371 if (!(ti->ti_flags & TH_SYN)) 1372 continue; 1373 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1374 NTOHS(mss); 1375 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1376 break; 1377 1378 case TCPOPT_WINDOW: 1379 if (optlen != TCPOLEN_WINDOW) 1380 continue; 1381 if (!(ti->ti_flags & TH_SYN)) 1382 continue; 1383 tp->t_flags |= TF_RCVD_SCALE; 1384 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1385 break; 1386 1387 case TCPOPT_TIMESTAMP: 1388 if (optlen != TCPOLEN_TIMESTAMP) 1389 continue; 1390 *ts_present = 1; 1391 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1392 NTOHL(*ts_val); 1393 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1394 NTOHL(*ts_ecr); 1395 1396 /* 1397 * A timestamp received in a SYN makes 1398 * it ok to send timestamp requests and replies. 1399 */ 1400 if (ti->ti_flags & TH_SYN) { 1401 tp->t_flags |= TF_RCVD_TSTMP; 1402 tp->ts_recent = *ts_val; 1403 tp->ts_recent_age = tcp_now; 1404 } 1405 break; 1406 } 1407 } 1408} 1409 1410/* 1411 * Pull out of band byte out of a segment so 1412 * it doesn't appear in the user's data queue. 1413 * It is still reflected in the segment length for 1414 * sequencing purposes. 1415 */ 1416void 1417tcp_pulloutofband(so, ti, m) 1418 struct socket *so; 1419 struct tcpiphdr *ti; 1420 register struct mbuf *m; 1421{ 1422 int cnt = ti->ti_urp - 1; 1423 1424 while (cnt >= 0) { 1425 if (m->m_len > cnt) { 1426 char *cp = mtod(m, caddr_t) + cnt; 1427 struct tcpcb *tp = sototcpcb(so); 1428 1429 tp->t_iobc = *cp; 1430 tp->t_oobflags |= TCPOOB_HAVEDATA; 1431 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1432 m->m_len--; 1433 return; 1434 } 1435 cnt -= m->m_len; 1436 m = m->m_next; 1437 if (m == 0) 1438 break; 1439 } 1440 panic("tcp_pulloutofband"); 1441} 1442 1443/* 1444 * Collect new round-trip time estimate 1445 * and update averages and current timeout. 1446 */ 1447void 1448tcp_xmit_timer(tp, rtt) 1449 register struct tcpcb *tp; 1450 short rtt; 1451{ 1452 register short delta; 1453 1454 tcpstat.tcps_rttupdated++; 1455 if (tp->t_srtt != 0) { 1456 /* 1457 * srtt is stored as fixed point with 3 bits after the 1458 * binary point (i.e., scaled by 8). The following magic 1459 * is equivalent to the smoothing algorithm in rfc793 with 1460 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1461 * point). Adjust rtt to origin 0. 1462 */ 1463 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1464 if ((tp->t_srtt += delta) <= 0) 1465 tp->t_srtt = 1; 1466 /* 1467 * We accumulate a smoothed rtt variance (actually, a 1468 * smoothed mean difference), then set the retransmit 1469 * timer to smoothed rtt + 4 times the smoothed variance. 1470 * rttvar is stored as fixed point with 2 bits after the 1471 * binary point (scaled by 4). The following is 1472 * equivalent to rfc793 smoothing with an alpha of .75 1473 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1474 * rfc793's wired-in beta. 1475 */ 1476 if (delta < 0) 1477 delta = -delta; 1478 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1479 if ((tp->t_rttvar += delta) <= 0) 1480 tp->t_rttvar = 1; 1481 } else { 1482 /* 1483 * No rtt measurement yet - use the unsmoothed rtt. 1484 * Set the variance to half the rtt (so our first 1485 * retransmit happens at 3*rtt). 1486 */ 1487 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1488 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1489 } 1490 tp->t_rtt = 0; 1491 tp->t_rxtshift = 0; 1492 1493 /* 1494 * the retransmit should happen at rtt + 4 * rttvar. 1495 * Because of the way we do the smoothing, srtt and rttvar 1496 * will each average +1/2 tick of bias. When we compute 1497 * the retransmit timer, we want 1/2 tick of rounding and 1498 * 1 extra tick because of +-1/2 tick uncertainty in the 1499 * firing of the timer. The bias will give us exactly the 1500 * 1.5 tick we need. But, because the bias is 1501 * statistical, we have to test that we don't drop below 1502 * the minimum feasible timer (which is 2 ticks). 1503 */ 1504 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1505 tp->t_rttmin, TCPTV_REXMTMAX); 1506 1507 /* 1508 * We received an ack for a packet that wasn't retransmitted; 1509 * it is probably safe to discard any error indications we've 1510 * received recently. This isn't quite right, but close enough 1511 * for now (a route might have failed after we sent a segment, 1512 * and the return path might not be symmetrical). 1513 */ 1514 tp->t_softerror = 0; 1515} 1516 1517/* 1518 * Determine a reasonable value for maxseg size. 1519 * If the route is known, check route for mtu. 1520 * If none, use an mss that can be handled on the outgoing 1521 * interface without forcing IP to fragment; if bigger than 1522 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1523 * to utilize large mbufs. If no route is found, route has no mtu, 1524 * or the destination isn't local, use a default, hopefully conservative 1525 * size (usually 512 or the default IP max size, but no more than the mtu 1526 * of the interface), as we can't discover anything about intervening 1527 * gateways or networks. We also initialize the congestion/slow start 1528 * window to be a single segment if the destination isn't local. 1529 * While looking at the routing entry, we also initialize other path-dependent 1530 * parameters from pre-set or cached values in the routing entry. 1531 */ 1532int 1533tcp_mss(tp, offer) 1534 register struct tcpcb *tp; 1535 u_int offer; 1536{ 1537 struct route *ro; 1538 register struct rtentry *rt; 1539 struct ifnet *ifp; 1540 register int rtt, mss; 1541 u_long bufsize; 1542 struct inpcb *inp; 1543 struct socket *so; 1544 extern int tcp_mssdflt; 1545 1546 inp = tp->t_inpcb; 1547 ro = &inp->inp_route; 1548 1549 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1550 /* No route yet, so try to acquire one */ 1551 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1552 ro->ro_dst.sa_family = AF_INET; 1553 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1554 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1555 inp->inp_faddr; 1556 rtalloc(ro); 1557 } 1558 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1559 return (tcp_mssdflt); 1560 } 1561 ifp = rt->rt_ifp; 1562 so = inp->inp_socket; 1563 1564#ifdef RTV_MTU /* if route characteristics exist ... */ 1565 /* 1566 * While we're here, check if there's an initial rtt 1567 * or rttvar. Convert from the route-table units 1568 * to scaled multiples of the slow timeout timer. 1569 */ 1570 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1571 /* 1572 * XXX the lock bit for MTU indicates that the value 1573 * is also a minimum value; this is subject to time. 1574 */ 1575 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1576 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1577 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1578 if (rt->rt_rmx.rmx_rttvar) 1579 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1580 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1581 else 1582 /* default variation is +- 1 rtt */ 1583 tp->t_rttvar = 1584 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1585 TCPT_RANGESET(tp->t_rxtcur, 1586 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1587 tp->t_rttmin, TCPTV_REXMTMAX); 1588 } 1589 /* 1590 * if there's an mtu associated with the route, use it 1591 */ 1592 if (rt->rt_rmx.rmx_mtu) 1593 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1594 else 1595#endif /* RTV_MTU */ 1596 { 1597 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1598#if (MCLBYTES & (MCLBYTES - 1)) == 0 1599 if (mss > MCLBYTES) 1600 mss &= ~(MCLBYTES-1); 1601#else 1602 if (mss > MCLBYTES) 1603 mss = mss / MCLBYTES * MCLBYTES; 1604#endif 1605 if (!in_localaddr(inp->inp_faddr)) 1606 mss = min(mss, tcp_mssdflt); 1607 } 1608 /* 1609 * The current mss, t_maxseg, is initialized to the default value. 1610 * If we compute a smaller value, reduce the current mss. 1611 * If we compute a larger value, return it for use in sending 1612 * a max seg size option, but don't store it for use 1613 * unless we received an offer at least that large from peer. 1614 * However, do not accept offers under 32 bytes. 1615 */ 1616 if (offer) 1617 mss = min(mss, offer); 1618 mss = max(mss, 32); /* sanity */ 1619 if (mss < tp->t_maxseg || offer != 0) { 1620 /* 1621 * If there's a pipesize, change the socket buffer 1622 * to that size. Make the socket buffers an integral 1623 * number of mss units; if the mss is larger than 1624 * the socket buffer, decrease the mss. 1625 */ 1626#ifdef RTV_SPIPE 1627 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1628#endif 1629 bufsize = so->so_snd.sb_hiwat; 1630 if (bufsize < mss) 1631 mss = bufsize; 1632 else { 1633 bufsize = roundup(bufsize, mss); 1634 if (bufsize > sb_max) 1635 bufsize = sb_max; 1636 (void)sbreserve(&so->so_snd, bufsize); 1637 } 1638 tp->t_maxseg = mss; 1639 1640#ifdef RTV_RPIPE 1641 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1642#endif 1643 bufsize = so->so_rcv.sb_hiwat; 1644 if (bufsize > mss) { 1645 bufsize = roundup(bufsize, mss); 1646 if (bufsize > sb_max) 1647 bufsize = sb_max; 1648 (void)sbreserve(&so->so_rcv, bufsize); 1649 } 1650 } 1651 tp->snd_cwnd = mss; 1652 1653#ifdef RTV_SSTHRESH 1654 if (rt->rt_rmx.rmx_ssthresh) { 1655 /* 1656 * There's some sort of gateway or interface 1657 * buffer limit on the path. Use this to set 1658 * the slow start threshhold, but set the 1659 * threshold to no less than 2*mss. 1660 */ 1661 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1662 } 1663#endif /* RTV_MTU */ 1664 return (mss); 1665} 1666#endif /* TUBA_INCLUDE */ 1667