tcp_input.c revision 2112
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: tcp_input.c,v 1.5 1994/08/02 07:49:01 davidg Exp $ 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 67#endif /* TUBA_INCLUDE */ 68#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 69 70/* for modulo comparisons of timestamps */ 71#define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 72#define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 73 74 75/* 76 * Insert segment ti into reassembly queue of tcp with 77 * control block tp. Return TH_FIN if reassembly now includes 78 * a segment with FIN. The macro form does the common case inline 79 * (segment is the next to be received on an established connection, 80 * and the queue is empty), avoiding linkage into and removal 81 * from the queue and repetition of various conversions. 82 * Set DELACK for segments received in order, but ack immediately 83 * when segments are out of order (so fast retransmit can work). 84 */ 85#define TCP_REASS(tp, ti, m, so, flags) { \ 86 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 87 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 88 (tp)->t_state == TCPS_ESTABLISHED) { \ 89 tp->t_flags |= TF_DELACK; \ 90 (tp)->rcv_nxt += (ti)->ti_len; \ 91 flags = (ti)->ti_flags & TH_FIN; \ 92 tcpstat.tcps_rcvpack++;\ 93 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 94 sbappend(&(so)->so_rcv, (m)); \ 95 sorwakeup(so); \ 96 } else { \ 97 (flags) = tcp_reass((tp), (ti), (m)); \ 98 tp->t_flags |= TF_ACKNOW; \ 99 } \ 100} 101#ifndef TUBA_INCLUDE 102 103int 104tcp_reass(tp, ti, m) 105 register struct tcpcb *tp; 106 register struct tcpiphdr *ti; 107 struct mbuf *m; 108{ 109 register struct tcpiphdr *q; 110 struct socket *so = tp->t_inpcb->inp_socket; 111 int flags; 112 113 /* 114 * Call with ti==0 after become established to 115 * force pre-ESTABLISHED data up to user socket. 116 */ 117 if (ti == 0) 118 goto present; 119 120 /* 121 * Find a segment which begins after this one does. 122 */ 123 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 124 q = (struct tcpiphdr *)q->ti_next) 125 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 126 break; 127 128 /* 129 * If there is a preceding segment, it may provide some of 130 * our data already. If so, drop the data from the incoming 131 * segment. If it provides all of our data, drop us. 132 */ 133 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 134 register int i; 135 q = (struct tcpiphdr *)q->ti_prev; 136 /* conversion to int (in i) handles seq wraparound */ 137 i = q->ti_seq + q->ti_len - ti->ti_seq; 138 if (i > 0) { 139 if (i >= ti->ti_len) { 140 tcpstat.tcps_rcvduppack++; 141 tcpstat.tcps_rcvdupbyte += ti->ti_len; 142 m_freem(m); 143 return (0); 144 } 145 m_adj(m, i); 146 ti->ti_len -= i; 147 ti->ti_seq += i; 148 } 149 q = (struct tcpiphdr *)(q->ti_next); 150 } 151 tcpstat.tcps_rcvoopack++; 152 tcpstat.tcps_rcvoobyte += ti->ti_len; 153 REASS_MBUF(ti) = m; /* XXX */ 154 155 /* 156 * While we overlap succeeding segments trim them or, 157 * if they are completely covered, dequeue them. 158 */ 159 while (q != (struct tcpiphdr *)tp) { 160 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 161 if (i <= 0) 162 break; 163 if (i < q->ti_len) { 164 q->ti_seq += i; 165 q->ti_len -= i; 166 m_adj(REASS_MBUF(q), i); 167 break; 168 } 169 q = (struct tcpiphdr *)q->ti_next; 170 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 171 remque(q->ti_prev); 172 m_freem(m); 173 } 174 175 /* 176 * Stick new segment in its place. 177 */ 178 insque(ti, q->ti_prev); 179 180present: 181 /* 182 * Present data to user, advancing rcv_nxt through 183 * completed sequence space. 184 */ 185 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 186 return (0); 187 ti = tp->seg_next; 188 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 189 return (0); 190 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 191 return (0); 192 do { 193 tp->rcv_nxt += ti->ti_len; 194 flags = ti->ti_flags & TH_FIN; 195 remque(ti); 196 m = REASS_MBUF(ti); 197 ti = (struct tcpiphdr *)ti->ti_next; 198 if (so->so_state & SS_CANTRCVMORE) 199 m_freem(m); 200 else 201 sbappend(&so->so_rcv, m); 202 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 203 sorwakeup(so); 204 return (flags); 205} 206 207/* 208 * TCP input routine, follows pages 65-76 of the 209 * protocol specification dated September, 1981 very closely. 210 */ 211void 212tcp_input(m, iphlen) 213 register struct mbuf *m; 214 int iphlen; 215{ 216 register struct tcpiphdr *ti; 217 register struct inpcb *inp; 218 caddr_t optp = NULL; 219 int optlen = 0; 220 int len, tlen, off; 221 register struct tcpcb *tp = 0; 222 register int tiflags; 223 struct socket *so = 0; 224 int todrop, acked, ourfinisacked, needoutput = 0; 225 short ostate = 0; 226 struct in_addr laddr; 227 int dropsocket = 0; 228 int iss = 0; 229 u_long tiwin, ts_val, ts_ecr; 230 int ts_present = 0; 231 232 tcpstat.tcps_rcvtotal++; 233 /* 234 * Get IP and TCP header together in first mbuf. 235 * Note: IP leaves IP header in first mbuf. 236 */ 237 ti = mtod(m, struct tcpiphdr *); 238 if (iphlen > sizeof (struct ip)) 239 ip_stripoptions(m, (struct mbuf *)0); 240 if (m->m_len < sizeof (struct tcpiphdr)) { 241 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 242 tcpstat.tcps_rcvshort++; 243 return; 244 } 245 ti = mtod(m, struct tcpiphdr *); 246 } 247 248 /* 249 * Checksum extended TCP header and data. 250 */ 251 tlen = ((struct ip *)ti)->ip_len; 252 len = sizeof (struct ip) + tlen; 253 ti->ti_next = ti->ti_prev = 0; 254 ti->ti_x1 = 0; 255 ti->ti_len = (u_short)tlen; 256 HTONS(ti->ti_len); 257 if (ti->ti_sum = in_cksum(m, len)) { 258 tcpstat.tcps_rcvbadsum++; 259 goto drop; 260 } 261#endif /* TUBA_INCLUDE */ 262 263 /* 264 * Check that TCP offset makes sense, 265 * pull out TCP options and adjust length. XXX 266 */ 267 off = ti->ti_off << 2; 268 if (off < sizeof (struct tcphdr) || off > tlen) { 269 tcpstat.tcps_rcvbadoff++; 270 goto drop; 271 } 272 tlen -= off; 273 ti->ti_len = tlen; 274 if (off > sizeof (struct tcphdr)) { 275 if (m->m_len < sizeof(struct ip) + off) { 276 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 277 tcpstat.tcps_rcvshort++; 278 return; 279 } 280 ti = mtod(m, struct tcpiphdr *); 281 } 282 optlen = off - sizeof (struct tcphdr); 283 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 284 /* 285 * Do quick retrieval of timestamp options ("options 286 * prediction?"). If timestamp is the only option and it's 287 * formatted as recommended in RFC 1323 appendix A, we 288 * quickly get the values now and not bother calling 289 * tcp_dooptions(), etc. 290 */ 291 if ((optlen == TCPOLEN_TSTAMP_APPA || 292 (optlen > TCPOLEN_TSTAMP_APPA && 293 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 294 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 295 (ti->ti_flags & TH_SYN) == 0) { 296 ts_present = 1; 297 ts_val = ntohl(*(u_long *)(optp + 4)); 298 ts_ecr = ntohl(*(u_long *)(optp + 8)); 299 optp = NULL; /* we've parsed the options */ 300 } 301 } 302 tiflags = ti->ti_flags; 303 304 /* 305 * Convert TCP protocol specific fields to host format. 306 */ 307 NTOHL(ti->ti_seq); 308 NTOHL(ti->ti_ack); 309 NTOHS(ti->ti_win); 310 NTOHS(ti->ti_urp); 311 312 /* 313 * Locate pcb for segment. 314 */ 315findpcb: 316 inp = tcp_last_inpcb; 317 if (inp->inp_lport != ti->ti_dport || 318 inp->inp_fport != ti->ti_sport || 319 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 320 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 321 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 322 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 323 if (inp) 324 tcp_last_inpcb = inp; 325 ++tcpstat.tcps_pcbcachemiss; 326 } 327 328 /* 329 * If the state is CLOSED (i.e., TCB does not exist) then 330 * all data in the incoming segment is discarded. 331 * If the TCB exists but is in CLOSED state, it is embryonic, 332 * but should either do a listen or a connect soon. 333 */ 334 if (inp == 0) 335 goto dropwithreset; 336 tp = intotcpcb(inp); 337 if (tp == 0) 338 goto dropwithreset; 339 if (tp->t_state == TCPS_CLOSED) 340 goto drop; 341 342 /* Unscale the window into a 32-bit value. */ 343 if ((tiflags & TH_SYN) == 0) 344 tiwin = ti->ti_win << tp->snd_scale; 345 else 346 tiwin = ti->ti_win; 347 348 so = inp->inp_socket; 349 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 350 if (so->so_options & SO_DEBUG) { 351 ostate = tp->t_state; 352 tcp_saveti = *ti; 353 } 354 if (so->so_options & SO_ACCEPTCONN) { 355 so = sonewconn(so, 0); 356 if (so == 0) 357 goto drop; 358 /* 359 * This is ugly, but .... 360 * 361 * Mark socket as temporary until we're 362 * committed to keeping it. The code at 363 * ``drop'' and ``dropwithreset'' check the 364 * flag dropsocket to see if the temporary 365 * socket created here should be discarded. 366 * We mark the socket as discardable until 367 * we're committed to it below in TCPS_LISTEN. 368 */ 369 dropsocket++; 370 inp = (struct inpcb *)so->so_pcb; 371 inp->inp_laddr = ti->ti_dst; 372 inp->inp_lport = ti->ti_dport; 373#if BSD>=43 374 inp->inp_options = ip_srcroute(); 375#endif 376 tp = intotcpcb(inp); 377 tp->t_state = TCPS_LISTEN; 378 379 /* Compute proper scaling value from buffer space 380 */ 381 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 382 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 383 tp->request_r_scale++; 384 } 385 } 386 387 /* 388 * Segment received on connection. 389 * Reset idle time and keep-alive timer. 390 */ 391 tp->t_idle = 0; 392 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 393 394 /* 395 * Process options if not in LISTEN state, 396 * else do it below (after getting remote address). 397 */ 398 if (optp && tp->t_state != TCPS_LISTEN) 399 tcp_dooptions(tp, optp, optlen, ti, 400 &ts_present, &ts_val, &ts_ecr); 401 402 /* 403 * Header prediction: check for the two common cases 404 * of a uni-directional data xfer. If the packet has 405 * no control flags, is in-sequence, the window didn't 406 * change and we're not retransmitting, it's a 407 * candidate. If the length is zero and the ack moved 408 * forward, we're the sender side of the xfer. Just 409 * free the data acked & wake any higher level process 410 * that was blocked waiting for space. If the length 411 * is non-zero and the ack didn't move, we're the 412 * receiver side. If we're getting packets in-order 413 * (the reassembly queue is empty), add the data to 414 * the socket buffer and note that we need a delayed ack. 415 */ 416 if (tp->t_state == TCPS_ESTABLISHED && 417 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 418 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 419 ti->ti_seq == tp->rcv_nxt && 420 tiwin && tiwin == tp->snd_wnd && 421 tp->snd_nxt == tp->snd_max) { 422 423 /* 424 * If last ACK falls within this segment's sequence numbers, 425 * record the timestamp. 426 */ 427 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 428 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 429 tp->ts_recent_age = tcp_now; 430 tp->ts_recent = ts_val; 431 } 432 433 if (ti->ti_len == 0) { 434 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 435 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 436 tp->snd_cwnd >= tp->snd_wnd) { 437 /* 438 * this is a pure ack for outstanding data. 439 */ 440 ++tcpstat.tcps_predack; 441 if (ts_present) 442 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 443 else if (tp->t_rtt && 444 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 445 tcp_xmit_timer(tp, tp->t_rtt); 446 acked = ti->ti_ack - tp->snd_una; 447 tcpstat.tcps_rcvackpack++; 448 tcpstat.tcps_rcvackbyte += acked; 449 sbdrop(&so->so_snd, acked); 450 tp->snd_una = ti->ti_ack; 451 m_freem(m); 452 453 /* 454 * If all outstanding data are acked, stop 455 * retransmit timer, otherwise restart timer 456 * using current (possibly backed-off) value. 457 * If process is waiting for space, 458 * wakeup/selwakeup/signal. If data 459 * are ready to send, let tcp_output 460 * decide between more output or persist. 461 */ 462 if (tp->snd_una == tp->snd_max) 463 tp->t_timer[TCPT_REXMT] = 0; 464 else if (tp->t_timer[TCPT_PERSIST] == 0) 465 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 466 467 if (so->so_snd.sb_flags & SB_NOTIFY) 468 sowwakeup(so); 469 if (so->so_snd.sb_cc) 470 (void) tcp_output(tp); 471 return; 472 } 473 } else if (ti->ti_ack == tp->snd_una && 474 tp->seg_next == (struct tcpiphdr *)tp && 475 ti->ti_len <= sbspace(&so->so_rcv)) { 476 /* 477 * this is a pure, in-sequence data packet 478 * with nothing on the reassembly queue and 479 * we have enough buffer space to take it. 480 */ 481 ++tcpstat.tcps_preddat; 482 tp->rcv_nxt += ti->ti_len; 483 tcpstat.tcps_rcvpack++; 484 tcpstat.tcps_rcvbyte += ti->ti_len; 485 /* 486 * Drop TCP, IP headers and TCP options then add data 487 * to socket buffer. 488 */ 489 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 490 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 491 sbappend(&so->so_rcv, m); 492 sorwakeup(so); 493 /* 494 * If this is a small packet, then ACK now - with Nagel 495 * congestion avoidance sender won't send more until 496 * he gets an ACK. 497 */ 498 if ((unsigned)ti->ti_len < tp->t_maxseg) { 499 tp->t_flags |= TF_ACKNOW; 500 tcp_output(tp); 501 } else { 502 tp->t_flags |= TF_DELACK; 503 } 504 return; 505 } 506 } 507 508 /* 509 * Drop TCP, IP headers and TCP options. 510 */ 511 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 512 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 513 514 /* 515 * Calculate amount of space in receive window, 516 * and then do TCP input processing. 517 * Receive window is amount of space in rcv queue, 518 * but not less than advertised window. 519 */ 520 { int win; 521 522 win = sbspace(&so->so_rcv); 523 if (win < 0) 524 win = 0; 525 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 526 } 527 528 switch (tp->t_state) { 529 530 /* 531 * If the state is LISTEN then ignore segment if it contains an RST. 532 * If the segment contains an ACK then it is bad and send a RST. 533 * If it does not contain a SYN then it is not interesting; drop it. 534 * Don't bother responding if the destination was a broadcast. 535 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 536 * tp->iss, and send a segment: 537 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 538 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 539 * Fill in remote peer address fields if not previously specified. 540 * Enter SYN_RECEIVED state, and process any other fields of this 541 * segment in this state. 542 */ 543 case TCPS_LISTEN: { 544 struct mbuf *am; 545 register struct sockaddr_in *sin; 546 547 if (tiflags & TH_RST) 548 goto drop; 549 if (tiflags & TH_ACK) 550 goto dropwithreset; 551 if ((tiflags & TH_SYN) == 0) 552 goto drop; 553 /* 554 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 555 * in_broadcast() should never return true on a received 556 * packet with M_BCAST not set. 557 */ 558 if (m->m_flags & (M_BCAST|M_MCAST) || 559 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 560 goto drop; 561 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 562 if (am == NULL) 563 goto drop; 564 am->m_len = sizeof (struct sockaddr_in); 565 sin = mtod(am, struct sockaddr_in *); 566 sin->sin_family = AF_INET; 567 sin->sin_len = sizeof(*sin); 568 sin->sin_addr = ti->ti_src; 569 sin->sin_port = ti->ti_sport; 570 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 571 laddr = inp->inp_laddr; 572 if (inp->inp_laddr.s_addr == INADDR_ANY) 573 inp->inp_laddr = ti->ti_dst; 574 if (in_pcbconnect(inp, am)) { 575 inp->inp_laddr = laddr; 576 (void) m_free(am); 577 goto drop; 578 } 579 (void) m_free(am); 580 tp->t_template = tcp_template(tp); 581 if (tp->t_template == 0) { 582 tp = tcp_drop(tp, ENOBUFS); 583 dropsocket = 0; /* socket is already gone */ 584 goto drop; 585 } 586 if (optp) 587 tcp_dooptions(tp, optp, optlen, ti, 588 &ts_present, &ts_val, &ts_ecr); 589 if (iss) 590 tp->iss = iss; 591 else 592 tp->iss = tcp_iss; 593 tcp_iss += TCP_ISSINCR/2; 594 tp->irs = ti->ti_seq; 595 tcp_sendseqinit(tp); 596 tcp_rcvseqinit(tp); 597 tp->t_flags |= TF_ACKNOW; 598 tp->t_state = TCPS_SYN_RECEIVED; 599 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 600 dropsocket = 0; /* committed to socket */ 601 tcpstat.tcps_accepts++; 602 goto trimthenstep6; 603 } 604 605 /* 606 * If the state is SYN_SENT: 607 * if seg contains an ACK, but not for our SYN, drop the input. 608 * if seg contains a RST, then drop the connection. 609 * if seg does not contain SYN, then drop it. 610 * Otherwise this is an acceptable SYN segment 611 * initialize tp->rcv_nxt and tp->irs 612 * if seg contains ack then advance tp->snd_una 613 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 614 * arrange for segment to be acked (eventually) 615 * continue processing rest of data/controls, beginning with URG 616 */ 617 case TCPS_SYN_SENT: 618 if ((tiflags & TH_ACK) && 619 (SEQ_LEQ(ti->ti_ack, tp->iss) || 620 SEQ_GT(ti->ti_ack, tp->snd_max))) 621 goto dropwithreset; 622 if (tiflags & TH_RST) { 623 if (tiflags & TH_ACK) 624 tp = tcp_drop(tp, ECONNREFUSED); 625 goto drop; 626 } 627 if ((tiflags & TH_SYN) == 0) 628 goto drop; 629 if (tiflags & TH_ACK) { 630 tp->snd_una = ti->ti_ack; 631 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 632 tp->snd_nxt = tp->snd_una; 633 } 634 tp->t_timer[TCPT_REXMT] = 0; 635 tp->irs = ti->ti_seq; 636 tcp_rcvseqinit(tp); 637 tp->t_flags |= TF_ACKNOW; 638 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 639 tcpstat.tcps_connects++; 640 soisconnected(so); 641 tp->t_state = TCPS_ESTABLISHED; 642 /* Do window scaling on this connection? */ 643 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 644 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 645 tp->snd_scale = tp->requested_s_scale; 646 tp->rcv_scale = tp->request_r_scale; 647 } 648 (void) tcp_reass(tp, (struct tcpiphdr *)0, 649 (struct mbuf *)0); 650 /* 651 * if we didn't have to retransmit the SYN, 652 * use its rtt as our initial srtt & rtt var. 653 */ 654 if (tp->t_rtt) 655 tcp_xmit_timer(tp, tp->t_rtt); 656 } else 657 tp->t_state = TCPS_SYN_RECEIVED; 658 659trimthenstep6: 660 /* 661 * Advance ti->ti_seq to correspond to first data byte. 662 * If data, trim to stay within window, 663 * dropping FIN if necessary. 664 */ 665 ti->ti_seq++; 666 if (ti->ti_len > tp->rcv_wnd) { 667 todrop = ti->ti_len - tp->rcv_wnd; 668 m_adj(m, -todrop); 669 ti->ti_len = tp->rcv_wnd; 670 tiflags &= ~TH_FIN; 671 tcpstat.tcps_rcvpackafterwin++; 672 tcpstat.tcps_rcvbyteafterwin += todrop; 673 } 674 tp->snd_wl1 = ti->ti_seq - 1; 675 tp->rcv_up = ti->ti_seq; 676 goto step6; 677 } 678 679 /* 680 * States other than LISTEN or SYN_SENT. 681 * First check timestamp, if present. 682 * Then check that at least some bytes of segment are within 683 * receive window. If segment begins before rcv_nxt, 684 * drop leading data (and SYN); if nothing left, just ack. 685 * 686 * RFC 1323 PAWS: If we have a timestamp reply on this segment 687 * and it's less than ts_recent, drop it. 688 */ 689 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 690 TSTMP_LT(ts_val, tp->ts_recent)) { 691 692 /* Check to see if ts_recent is over 24 days old. */ 693 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 694 /* 695 * Invalidate ts_recent. If this segment updates 696 * ts_recent, the age will be reset later and ts_recent 697 * will get a valid value. If it does not, setting 698 * ts_recent to zero will at least satisfy the 699 * requirement that zero be placed in the timestamp 700 * echo reply when ts_recent isn't valid. The 701 * age isn't reset until we get a valid ts_recent 702 * because we don't want out-of-order segments to be 703 * dropped when ts_recent is old. 704 */ 705 tp->ts_recent = 0; 706 } else { 707 tcpstat.tcps_rcvduppack++; 708 tcpstat.tcps_rcvdupbyte += ti->ti_len; 709 tcpstat.tcps_pawsdrop++; 710 goto dropafterack; 711 } 712 } 713 714 todrop = tp->rcv_nxt - ti->ti_seq; 715 if (todrop > 0) { 716 if (tiflags & TH_SYN) { 717 tiflags &= ~TH_SYN; 718 ti->ti_seq++; 719 if (ti->ti_urp > 1) 720 ti->ti_urp--; 721 else 722 tiflags &= ~TH_URG; 723 todrop--; 724 } 725 if (todrop >= ti->ti_len) { 726 tcpstat.tcps_rcvduppack++; 727 tcpstat.tcps_rcvdupbyte += ti->ti_len; 728 /* 729 * If segment is just one to the left of the window, 730 * check two special cases: 731 * 1. Don't toss RST in response to 4.2-style keepalive. 732 * 2. If the only thing to drop is a FIN, we can drop 733 * it, but check the ACK or we will get into FIN 734 * wars if our FINs crossed (both CLOSING). 735 * In either case, send ACK to resynchronize, 736 * but keep on processing for RST or ACK. 737 */ 738 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 739#ifdef TCP_COMPAT_42 740 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 741#endif 742 ) { 743 todrop = ti->ti_len; 744 tiflags &= ~TH_FIN; 745 tp->t_flags |= TF_ACKNOW; 746 } else { 747 /* 748 * Handle the case when a bound socket connects 749 * to itself. Allow packets with a SYN and 750 * an ACK to continue with the processing. 751 */ 752 if (todrop != 0 || (tiflags & TH_ACK) == 0) 753 goto dropafterack; 754 } 755 } else { 756 tcpstat.tcps_rcvpartduppack++; 757 tcpstat.tcps_rcvpartdupbyte += todrop; 758 } 759 m_adj(m, todrop); 760 ti->ti_seq += todrop; 761 ti->ti_len -= todrop; 762 if (ti->ti_urp > todrop) 763 ti->ti_urp -= todrop; 764 else { 765 tiflags &= ~TH_URG; 766 ti->ti_urp = 0; 767 } 768 } 769 770 /* 771 * If new data are received on a connection after the 772 * user processes are gone, then RST the other end. 773 */ 774 if ((so->so_state & SS_NOFDREF) && 775 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 776 tp = tcp_close(tp); 777 tcpstat.tcps_rcvafterclose++; 778 goto dropwithreset; 779 } 780 781 /* 782 * If segment ends after window, drop trailing data 783 * (and PUSH and FIN); if nothing left, just ACK. 784 */ 785 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 786 if (todrop > 0) { 787 tcpstat.tcps_rcvpackafterwin++; 788 if (todrop >= ti->ti_len) { 789 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 790 /* 791 * If a new connection request is received 792 * while in TIME_WAIT, drop the old connection 793 * and start over if the sequence numbers 794 * are above the previous ones. 795 */ 796 if (tiflags & TH_SYN && 797 tp->t_state == TCPS_TIME_WAIT && 798 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 799 iss = tp->rcv_nxt + TCP_ISSINCR; 800 tp = tcp_close(tp); 801 goto findpcb; 802 } 803 /* 804 * If window is closed can only take segments at 805 * window edge, and have to drop data and PUSH from 806 * incoming segments. Continue processing, but 807 * remember to ack. Otherwise, drop segment 808 * and ack. 809 */ 810 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 811 tp->t_flags |= TF_ACKNOW; 812 tcpstat.tcps_rcvwinprobe++; 813 } else 814 goto dropafterack; 815 } else 816 tcpstat.tcps_rcvbyteafterwin += todrop; 817 m_adj(m, -todrop); 818 ti->ti_len -= todrop; 819 tiflags &= ~(TH_PUSH|TH_FIN); 820 } 821 822 /* 823 * If last ACK falls within this segment's sequence numbers, 824 * record its timestamp. 825 */ 826 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 827 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 828 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 829 tp->ts_recent_age = tcp_now; 830 tp->ts_recent = ts_val; 831 } 832 833 /* 834 * If the RST bit is set examine the state: 835 * SYN_RECEIVED STATE: 836 * If passive open, return to LISTEN state. 837 * If active open, inform user that connection was refused. 838 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 839 * Inform user that connection was reset, and close tcb. 840 * CLOSING, LAST_ACK, TIME_WAIT STATES 841 * Close the tcb. 842 */ 843 if (tiflags&TH_RST) switch (tp->t_state) { 844 845 case TCPS_SYN_RECEIVED: 846 so->so_error = ECONNREFUSED; 847 goto close; 848 849 case TCPS_ESTABLISHED: 850 case TCPS_FIN_WAIT_1: 851 case TCPS_FIN_WAIT_2: 852 case TCPS_CLOSE_WAIT: 853 so->so_error = ECONNRESET; 854 close: 855 tp->t_state = TCPS_CLOSED; 856 tcpstat.tcps_drops++; 857 tp = tcp_close(tp); 858 goto drop; 859 860 case TCPS_CLOSING: 861 case TCPS_LAST_ACK: 862 case TCPS_TIME_WAIT: 863 tp = tcp_close(tp); 864 goto drop; 865 } 866 867 /* 868 * If a SYN is in the window, then this is an 869 * error and we send an RST and drop the connection. 870 */ 871 if (tiflags & TH_SYN) { 872 tp = tcp_drop(tp, ECONNRESET); 873 goto dropwithreset; 874 } 875 876 /* 877 * If the ACK bit is off we drop the segment and return. 878 */ 879 if ((tiflags & TH_ACK) == 0) 880 goto drop; 881 882 /* 883 * Ack processing. 884 */ 885 switch (tp->t_state) { 886 887 /* 888 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 889 * ESTABLISHED state and continue processing, otherwise 890 * send an RST. 891 */ 892 case TCPS_SYN_RECEIVED: 893 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 894 SEQ_GT(ti->ti_ack, tp->snd_max)) 895 goto dropwithreset; 896 tcpstat.tcps_connects++; 897 soisconnected(so); 898 tp->t_state = TCPS_ESTABLISHED; 899 /* Do window scaling? */ 900 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 901 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 902 tp->snd_scale = tp->requested_s_scale; 903 tp->rcv_scale = tp->request_r_scale; 904 } 905 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 906 tp->snd_wl1 = ti->ti_seq - 1; 907 /* fall into ... */ 908 909 /* 910 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 911 * ACKs. If the ack is in the range 912 * tp->snd_una < ti->ti_ack <= tp->snd_max 913 * then advance tp->snd_una to ti->ti_ack and drop 914 * data from the retransmission queue. If this ACK reflects 915 * more up to date window information we update our window information. 916 */ 917 case TCPS_ESTABLISHED: 918 case TCPS_FIN_WAIT_1: 919 case TCPS_FIN_WAIT_2: 920 case TCPS_CLOSE_WAIT: 921 case TCPS_CLOSING: 922 case TCPS_LAST_ACK: 923 case TCPS_TIME_WAIT: 924 925 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 926 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 927 tcpstat.tcps_rcvdupack++; 928 /* 929 * If we have outstanding data (other than 930 * a window probe), this is a completely 931 * duplicate ack (ie, window info didn't 932 * change), the ack is the biggest we've 933 * seen and we've seen exactly our rexmt 934 * threshhold of them, assume a packet 935 * has been dropped and retransmit it. 936 * Kludge snd_nxt & the congestion 937 * window so we send only this one 938 * packet. 939 * 940 * We know we're losing at the current 941 * window size so do congestion avoidance 942 * (set ssthresh to half the current window 943 * and pull our congestion window back to 944 * the new ssthresh). 945 * 946 * Dup acks mean that packets have left the 947 * network (they're now cached at the receiver) 948 * so bump cwnd by the amount in the receiver 949 * to keep a constant cwnd packets in the 950 * network. 951 */ 952 if (tp->t_timer[TCPT_REXMT] == 0 || 953 ti->ti_ack != tp->snd_una) 954 tp->t_dupacks = 0; 955 else if (++tp->t_dupacks == tcprexmtthresh) { 956 tcp_seq onxt = tp->snd_nxt; 957 u_int win = 958 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 959 tp->t_maxseg; 960 961 if (win < 2) 962 win = 2; 963 tp->snd_ssthresh = win * tp->t_maxseg; 964 tp->t_timer[TCPT_REXMT] = 0; 965 tp->t_rtt = 0; 966 tp->snd_nxt = ti->ti_ack; 967 tp->snd_cwnd = tp->t_maxseg; 968 (void) tcp_output(tp); 969 tp->snd_cwnd = tp->snd_ssthresh + 970 tp->t_maxseg * tp->t_dupacks; 971 if (SEQ_GT(onxt, tp->snd_nxt)) 972 tp->snd_nxt = onxt; 973 goto drop; 974 } else if (tp->t_dupacks > tcprexmtthresh) { 975 tp->snd_cwnd += tp->t_maxseg; 976 (void) tcp_output(tp); 977 goto drop; 978 } 979 } else 980 tp->t_dupacks = 0; 981 break; 982 } 983 /* 984 * If the congestion window was inflated to account 985 * for the other side's cached packets, retract it. 986 */ 987 if (tp->t_dupacks > tcprexmtthresh && 988 tp->snd_cwnd > tp->snd_ssthresh) 989 tp->snd_cwnd = tp->snd_ssthresh; 990 tp->t_dupacks = 0; 991 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 992 tcpstat.tcps_rcvacktoomuch++; 993 goto dropafterack; 994 } 995 acked = ti->ti_ack - tp->snd_una; 996 tcpstat.tcps_rcvackpack++; 997 tcpstat.tcps_rcvackbyte += acked; 998 999 /* 1000 * If we have a timestamp reply, update smoothed 1001 * round trip time. If no timestamp is present but 1002 * transmit timer is running and timed sequence 1003 * number was acked, update smoothed round trip time. 1004 * Since we now have an rtt measurement, cancel the 1005 * timer backoff (cf., Phil Karn's retransmit alg.). 1006 * Recompute the initial retransmit timer. 1007 */ 1008 if (ts_present) 1009 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 1010 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1011 tcp_xmit_timer(tp,tp->t_rtt); 1012 1013 /* 1014 * If all outstanding data is acked, stop retransmit 1015 * timer and remember to restart (more output or persist). 1016 * If there is more data to be acked, restart retransmit 1017 * timer, using current (possibly backed-off) value. 1018 */ 1019 if (ti->ti_ack == tp->snd_max) { 1020 tp->t_timer[TCPT_REXMT] = 0; 1021 needoutput = 1; 1022 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1023 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1024 /* 1025 * When new data is acked, open the congestion window. 1026 * If the window gives us less than ssthresh packets 1027 * in flight, open exponentially (maxseg per packet). 1028 * Otherwise open linearly: maxseg per window 1029 * (maxseg^2 / cwnd per packet), plus a constant 1030 * fraction of a packet (maxseg/8) to help larger windows 1031 * open quickly enough. 1032 */ 1033 { 1034 register u_int cw = tp->snd_cwnd; 1035 register u_int incr = tp->t_maxseg; 1036 1037 if (cw > tp->snd_ssthresh) 1038 incr = incr * incr / cw + incr / 8; 1039 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1040 } 1041 if (acked > so->so_snd.sb_cc) { 1042 tp->snd_wnd -= so->so_snd.sb_cc; 1043 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1044 ourfinisacked = 1; 1045 } else { 1046 sbdrop(&so->so_snd, acked); 1047 tp->snd_wnd -= acked; 1048 ourfinisacked = 0; 1049 } 1050 if (so->so_snd.sb_flags & SB_NOTIFY) 1051 sowwakeup(so); 1052 tp->snd_una = ti->ti_ack; 1053 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1054 tp->snd_nxt = tp->snd_una; 1055 1056 switch (tp->t_state) { 1057 1058 /* 1059 * In FIN_WAIT_1 STATE in addition to the processing 1060 * for the ESTABLISHED state if our FIN is now acknowledged 1061 * then enter FIN_WAIT_2. 1062 */ 1063 case TCPS_FIN_WAIT_1: 1064 if (ourfinisacked) { 1065 /* 1066 * If we can't receive any more 1067 * data, then closing user can proceed. 1068 * Starting the timer is contrary to the 1069 * specification, but if we don't get a FIN 1070 * we'll hang forever. 1071 */ 1072 if (so->so_state & SS_CANTRCVMORE) { 1073 soisdisconnected(so); 1074 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1075 } 1076 tp->t_state = TCPS_FIN_WAIT_2; 1077 } 1078 break; 1079 1080 /* 1081 * In CLOSING STATE in addition to the processing for 1082 * the ESTABLISHED state if the ACK acknowledges our FIN 1083 * then enter the TIME-WAIT state, otherwise ignore 1084 * the segment. 1085 */ 1086 case TCPS_CLOSING: 1087 if (ourfinisacked) { 1088 tp->t_state = TCPS_TIME_WAIT; 1089 tcp_canceltimers(tp); 1090 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1091 soisdisconnected(so); 1092 } 1093 break; 1094 1095 /* 1096 * In LAST_ACK, we may still be waiting for data to drain 1097 * and/or to be acked, as well as for the ack of our FIN. 1098 * If our FIN is now acknowledged, delete the TCB, 1099 * enter the closed state and return. 1100 */ 1101 case TCPS_LAST_ACK: 1102 if (ourfinisacked) { 1103 tp = tcp_close(tp); 1104 goto drop; 1105 } 1106 break; 1107 1108 /* 1109 * In TIME_WAIT state the only thing that should arrive 1110 * is a retransmission of the remote FIN. Acknowledge 1111 * it and restart the finack timer. 1112 */ 1113 case TCPS_TIME_WAIT: 1114 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1115 goto dropafterack; 1116 } 1117 } 1118 1119step6: 1120 /* 1121 * Update window information. 1122 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1123 */ 1124 if ((tiflags & TH_ACK) && 1125 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1126 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1127 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1128 /* keep track of pure window updates */ 1129 if (ti->ti_len == 0 && 1130 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1131 tcpstat.tcps_rcvwinupd++; 1132 tp->snd_wnd = tiwin; 1133 tp->snd_wl1 = ti->ti_seq; 1134 tp->snd_wl2 = ti->ti_ack; 1135 if (tp->snd_wnd > tp->max_sndwnd) 1136 tp->max_sndwnd = tp->snd_wnd; 1137 needoutput = 1; 1138 } 1139 1140 /* 1141 * Process segments with URG. 1142 */ 1143 if ((tiflags & TH_URG) && ti->ti_urp && 1144 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1145 /* 1146 * This is a kludge, but if we receive and accept 1147 * random urgent pointers, we'll crash in 1148 * soreceive. It's hard to imagine someone 1149 * actually wanting to send this much urgent data. 1150 */ 1151 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1152 ti->ti_urp = 0; /* XXX */ 1153 tiflags &= ~TH_URG; /* XXX */ 1154 goto dodata; /* XXX */ 1155 } 1156 /* 1157 * If this segment advances the known urgent pointer, 1158 * then mark the data stream. This should not happen 1159 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1160 * a FIN has been received from the remote side. 1161 * In these states we ignore the URG. 1162 * 1163 * According to RFC961 (Assigned Protocols), 1164 * the urgent pointer points to the last octet 1165 * of urgent data. We continue, however, 1166 * to consider it to indicate the first octet 1167 * of data past the urgent section as the original 1168 * spec states (in one of two places). 1169 */ 1170 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1171 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1172 so->so_oobmark = so->so_rcv.sb_cc + 1173 (tp->rcv_up - tp->rcv_nxt) - 1; 1174 if (so->so_oobmark == 0) 1175 so->so_state |= SS_RCVATMARK; 1176 sohasoutofband(so); 1177 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1178 } 1179 /* 1180 * Remove out of band data so doesn't get presented to user. 1181 * This can happen independent of advancing the URG pointer, 1182 * but if two URG's are pending at once, some out-of-band 1183 * data may creep in... ick. 1184 */ 1185 if (ti->ti_urp <= (u_long)ti->ti_len 1186#ifdef SO_OOBINLINE 1187 && (so->so_options & SO_OOBINLINE) == 0 1188#endif 1189 ) 1190 tcp_pulloutofband(so, ti, m); 1191 } else 1192 /* 1193 * If no out of band data is expected, 1194 * pull receive urgent pointer along 1195 * with the receive window. 1196 */ 1197 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1198 tp->rcv_up = tp->rcv_nxt; 1199dodata: /* XXX */ 1200 1201 /* 1202 * Process the segment text, merging it into the TCP sequencing queue, 1203 * and arranging for acknowledgment of receipt if necessary. 1204 * This process logically involves adjusting tp->rcv_wnd as data 1205 * is presented to the user (this happens in tcp_usrreq.c, 1206 * case PRU_RCVD). If a FIN has already been received on this 1207 * connection then we just ignore the text. 1208 */ 1209 if ((ti->ti_len || (tiflags&TH_FIN)) && 1210 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1211 TCP_REASS(tp, ti, m, so, tiflags); 1212 /* 1213 * Note the amount of data that peer has sent into 1214 * our window, in order to estimate the sender's 1215 * buffer size. 1216 */ 1217 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1218 } else { 1219 m_freem(m); 1220 tiflags &= ~TH_FIN; 1221 } 1222 1223 /* 1224 * If FIN is received ACK the FIN and let the user know 1225 * that the connection is closing. 1226 */ 1227 if (tiflags & TH_FIN) { 1228 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1229 socantrcvmore(so); 1230 tp->t_flags |= TF_ACKNOW; 1231 tp->rcv_nxt++; 1232 } 1233 switch (tp->t_state) { 1234 1235 /* 1236 * In SYN_RECEIVED and ESTABLISHED STATES 1237 * enter the CLOSE_WAIT state. 1238 */ 1239 case TCPS_SYN_RECEIVED: 1240 case TCPS_ESTABLISHED: 1241 tp->t_state = TCPS_CLOSE_WAIT; 1242 break; 1243 1244 /* 1245 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1246 * enter the CLOSING state. 1247 */ 1248 case TCPS_FIN_WAIT_1: 1249 tp->t_state = TCPS_CLOSING; 1250 break; 1251 1252 /* 1253 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1254 * starting the time-wait timer, turning off the other 1255 * standard timers. 1256 */ 1257 case TCPS_FIN_WAIT_2: 1258 tp->t_state = TCPS_TIME_WAIT; 1259 tcp_canceltimers(tp); 1260 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1261 soisdisconnected(so); 1262 break; 1263 1264 /* 1265 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1266 */ 1267 case TCPS_TIME_WAIT: 1268 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1269 break; 1270 } 1271 } 1272 if (so->so_options & SO_DEBUG) 1273 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1274 1275 /* 1276 * If this is a small packet, then ACK now - with Nagel 1277 * congestion avoidance sender won't send more until 1278 * he gets an ACK. 1279 */ 1280 if (ti->ti_len && ((unsigned)ti->ti_len < tp->t_maxseg)) 1281 tp->t_flags |= TF_ACKNOW; 1282 1283 /* 1284 * Return any desired output. 1285 */ 1286 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1287 (void) tcp_output(tp); 1288 return; 1289 1290dropafterack: 1291 /* 1292 * Generate an ACK dropping incoming segment if it occupies 1293 * sequence space, where the ACK reflects our state. 1294 */ 1295 if (tiflags & TH_RST) 1296 goto drop; 1297 m_freem(m); 1298 tp->t_flags |= TF_ACKNOW; 1299 (void) tcp_output(tp); 1300 return; 1301 1302dropwithreset: 1303 /* 1304 * Generate a RST, dropping incoming segment. 1305 * Make ACK acceptable to originator of segment. 1306 * Don't bother to respond if destination was broadcast/multicast. 1307 */ 1308 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1309 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1310 goto drop; 1311 if (tiflags & TH_ACK) 1312 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1313 else { 1314 if (tiflags & TH_SYN) 1315 ti->ti_len++; 1316 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1317 TH_RST|TH_ACK); 1318 } 1319 /* destroy temporarily created socket */ 1320 if (dropsocket) 1321 (void) soabort(so); 1322 return; 1323 1324drop: 1325 /* 1326 * Drop space held by incoming segment and return. 1327 */ 1328 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1329 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1330 m_freem(m); 1331 /* destroy temporarily created socket */ 1332 if (dropsocket) 1333 (void) soabort(so); 1334 return; 1335#ifndef TUBA_INCLUDE 1336} 1337 1338void 1339tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1340 struct tcpcb *tp; 1341 u_char *cp; 1342 int cnt; 1343 struct tcpiphdr *ti; 1344 int *ts_present; 1345 u_long *ts_val, *ts_ecr; 1346{ 1347 u_short mss; 1348 int opt, optlen; 1349 1350 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1351 opt = cp[0]; 1352 if (opt == TCPOPT_EOL) 1353 break; 1354 if (opt == TCPOPT_NOP) 1355 optlen = 1; 1356 else { 1357 optlen = cp[1]; 1358 if (optlen <= 0) 1359 break; 1360 } 1361 switch (opt) { 1362 1363 default: 1364 continue; 1365 1366 case TCPOPT_MAXSEG: 1367 if (optlen != TCPOLEN_MAXSEG) 1368 continue; 1369 if (!(ti->ti_flags & TH_SYN)) 1370 continue; 1371 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1372 NTOHS(mss); 1373 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1374 break; 1375 1376 case TCPOPT_WINDOW: 1377 if (optlen != TCPOLEN_WINDOW) 1378 continue; 1379 if (!(ti->ti_flags & TH_SYN)) 1380 continue; 1381 tp->t_flags |= TF_RCVD_SCALE; 1382 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1383 break; 1384 1385 case TCPOPT_TIMESTAMP: 1386 if (optlen != TCPOLEN_TIMESTAMP) 1387 continue; 1388 *ts_present = 1; 1389 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1390 NTOHL(*ts_val); 1391 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1392 NTOHL(*ts_ecr); 1393 1394 /* 1395 * A timestamp received in a SYN makes 1396 * it ok to send timestamp requests and replies. 1397 */ 1398 if (ti->ti_flags & TH_SYN) { 1399 tp->t_flags |= TF_RCVD_TSTMP; 1400 tp->ts_recent = *ts_val; 1401 tp->ts_recent_age = tcp_now; 1402 } 1403 break; 1404 } 1405 } 1406} 1407 1408/* 1409 * Pull out of band byte out of a segment so 1410 * it doesn't appear in the user's data queue. 1411 * It is still reflected in the segment length for 1412 * sequencing purposes. 1413 */ 1414void 1415tcp_pulloutofband(so, ti, m) 1416 struct socket *so; 1417 struct tcpiphdr *ti; 1418 register struct mbuf *m; 1419{ 1420 int cnt = ti->ti_urp - 1; 1421 1422 while (cnt >= 0) { 1423 if (m->m_len > cnt) { 1424 char *cp = mtod(m, caddr_t) + cnt; 1425 struct tcpcb *tp = sototcpcb(so); 1426 1427 tp->t_iobc = *cp; 1428 tp->t_oobflags |= TCPOOB_HAVEDATA; 1429 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1430 m->m_len--; 1431 return; 1432 } 1433 cnt -= m->m_len; 1434 m = m->m_next; 1435 if (m == 0) 1436 break; 1437 } 1438 panic("tcp_pulloutofband"); 1439} 1440 1441/* 1442 * Collect new round-trip time estimate 1443 * and update averages and current timeout. 1444 */ 1445void 1446tcp_xmit_timer(tp, rtt) 1447 register struct tcpcb *tp; 1448 short rtt; 1449{ 1450 register short delta; 1451 1452 tcpstat.tcps_rttupdated++; 1453 if (tp->t_srtt != 0) { 1454 /* 1455 * srtt is stored as fixed point with 3 bits after the 1456 * binary point (i.e., scaled by 8). The following magic 1457 * is equivalent to the smoothing algorithm in rfc793 with 1458 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1459 * point). Adjust rtt to origin 0. 1460 */ 1461 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1462 if ((tp->t_srtt += delta) <= 0) 1463 tp->t_srtt = 1; 1464 /* 1465 * We accumulate a smoothed rtt variance (actually, a 1466 * smoothed mean difference), then set the retransmit 1467 * timer to smoothed rtt + 4 times the smoothed variance. 1468 * rttvar is stored as fixed point with 2 bits after the 1469 * binary point (scaled by 4). The following is 1470 * equivalent to rfc793 smoothing with an alpha of .75 1471 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1472 * rfc793's wired-in beta. 1473 */ 1474 if (delta < 0) 1475 delta = -delta; 1476 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1477 if ((tp->t_rttvar += delta) <= 0) 1478 tp->t_rttvar = 1; 1479 } else { 1480 /* 1481 * No rtt measurement yet - use the unsmoothed rtt. 1482 * Set the variance to half the rtt (so our first 1483 * retransmit happens at 3*rtt). 1484 */ 1485 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1486 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1487 } 1488 tp->t_rtt = 0; 1489 tp->t_rxtshift = 0; 1490 1491 /* 1492 * the retransmit should happen at rtt + 4 * rttvar. 1493 * Because of the way we do the smoothing, srtt and rttvar 1494 * will each average +1/2 tick of bias. When we compute 1495 * the retransmit timer, we want 1/2 tick of rounding and 1496 * 1 extra tick because of +-1/2 tick uncertainty in the 1497 * firing of the timer. The bias will give us exactly the 1498 * 1.5 tick we need. But, because the bias is 1499 * statistical, we have to test that we don't drop below 1500 * the minimum feasible timer (which is 2 ticks). 1501 */ 1502 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1503 tp->t_rttmin, TCPTV_REXMTMAX); 1504 1505 /* 1506 * We received an ack for a packet that wasn't retransmitted; 1507 * it is probably safe to discard any error indications we've 1508 * received recently. This isn't quite right, but close enough 1509 * for now (a route might have failed after we sent a segment, 1510 * and the return path might not be symmetrical). 1511 */ 1512 tp->t_softerror = 0; 1513} 1514 1515/* 1516 * Determine a reasonable value for maxseg size. 1517 * If the route is known, check route for mtu. 1518 * If none, use an mss that can be handled on the outgoing 1519 * interface without forcing IP to fragment; if bigger than 1520 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1521 * to utilize large mbufs. If no route is found, route has no mtu, 1522 * or the destination isn't local, use a default, hopefully conservative 1523 * size (usually 512 or the default IP max size, but no more than the mtu 1524 * of the interface), as we can't discover anything about intervening 1525 * gateways or networks. We also initialize the congestion/slow start 1526 * window to be a single segment if the destination isn't local. 1527 * While looking at the routing entry, we also initialize other path-dependent 1528 * parameters from pre-set or cached values in the routing entry. 1529 */ 1530int 1531tcp_mss(tp, offer) 1532 register struct tcpcb *tp; 1533 u_int offer; 1534{ 1535 struct route *ro; 1536 register struct rtentry *rt; 1537 struct ifnet *ifp; 1538 register int rtt, mss; 1539 u_long bufsize; 1540 struct inpcb *inp; 1541 struct socket *so; 1542 extern int tcp_mssdflt; 1543 1544 inp = tp->t_inpcb; 1545 ro = &inp->inp_route; 1546 1547 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1548 /* No route yet, so try to acquire one */ 1549 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1550 ro->ro_dst.sa_family = AF_INET; 1551 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1552 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1553 inp->inp_faddr; 1554 rtalloc(ro); 1555 } 1556 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1557 return (tcp_mssdflt); 1558 } 1559 ifp = rt->rt_ifp; 1560 so = inp->inp_socket; 1561 1562#ifdef RTV_MTU /* if route characteristics exist ... */ 1563 /* 1564 * While we're here, check if there's an initial rtt 1565 * or rttvar. Convert from the route-table units 1566 * to scaled multiples of the slow timeout timer. 1567 */ 1568 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1569 /* 1570 * XXX the lock bit for MTU indicates that the value 1571 * is also a minimum value; this is subject to time. 1572 */ 1573 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1574 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1575 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1576 if (rt->rt_rmx.rmx_rttvar) 1577 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1578 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1579 else 1580 /* default variation is +- 1 rtt */ 1581 tp->t_rttvar = 1582 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1583 TCPT_RANGESET(tp->t_rxtcur, 1584 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1585 tp->t_rttmin, TCPTV_REXMTMAX); 1586 } 1587 /* 1588 * if there's an mtu associated with the route, use it 1589 */ 1590 if (rt->rt_rmx.rmx_mtu) 1591 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1592 else 1593#endif /* RTV_MTU */ 1594 { 1595 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1596#if (MCLBYTES & (MCLBYTES - 1)) == 0 1597 if (mss > MCLBYTES) 1598 mss &= ~(MCLBYTES-1); 1599#else 1600 if (mss > MCLBYTES) 1601 mss = mss / MCLBYTES * MCLBYTES; 1602#endif 1603 if (!in_localaddr(inp->inp_faddr)) 1604 mss = min(mss, tcp_mssdflt); 1605 } 1606 /* 1607 * The current mss, t_maxseg, is initialized to the default value. 1608 * If we compute a smaller value, reduce the current mss. 1609 * If we compute a larger value, return it for use in sending 1610 * a max seg size option, but don't store it for use 1611 * unless we received an offer at least that large from peer. 1612 * However, do not accept offers under 32 bytes. 1613 */ 1614 if (offer) 1615 mss = min(mss, offer); 1616 mss = max(mss, 32); /* sanity */ 1617 if (mss < tp->t_maxseg || offer != 0) { 1618 /* 1619 * If there's a pipesize, change the socket buffer 1620 * to that size. Make the socket buffers an integral 1621 * number of mss units; if the mss is larger than 1622 * the socket buffer, decrease the mss. 1623 */ 1624#ifdef RTV_SPIPE 1625 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1626#endif 1627 bufsize = so->so_snd.sb_hiwat; 1628 if (bufsize < mss) 1629 mss = bufsize; 1630 else { 1631 bufsize = roundup(bufsize, mss); 1632 if (bufsize > sb_max) 1633 bufsize = sb_max; 1634 (void)sbreserve(&so->so_snd, bufsize); 1635 } 1636 tp->t_maxseg = mss; 1637 1638#ifdef RTV_RPIPE 1639 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1640#endif 1641 bufsize = so->so_rcv.sb_hiwat; 1642 if (bufsize > mss) { 1643 bufsize = roundup(bufsize, mss); 1644 if (bufsize > sb_max) 1645 bufsize = sb_max; 1646 (void)sbreserve(&so->so_rcv, bufsize); 1647 } 1648 } 1649 tp->snd_cwnd = mss; 1650 1651#ifdef RTV_SSTHRESH 1652 if (rt->rt_rmx.rmx_ssthresh) { 1653 /* 1654 * There's some sort of gateway or interface 1655 * buffer limit on the path. Use this to set 1656 * the slow start threshhold, but set the 1657 * threshold to no less than 2*mss. 1658 */ 1659 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1660 } 1661#endif /* RTV_MTU */ 1662 return (mss); 1663} 1664#endif /* TUBA_INCLUDE */ 1665