tcp_input.c revision 1812
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 */ 35 36#ifndef TUBA_INCLUDE 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/malloc.h> 40#include <sys/mbuf.h> 41#include <sys/protosw.h> 42#include <sys/socket.h> 43#include <sys/socketvar.h> 44#include <sys/errno.h> 45 46#include <net/if.h> 47#include <net/route.h> 48 49#include <netinet/in.h> 50#include <netinet/in_systm.h> 51#include <netinet/ip.h> 52#include <netinet/in_pcb.h> 53#include <netinet/ip_var.h> 54#include <netinet/tcp.h> 55#include <netinet/tcp_fsm.h> 56#include <netinet/tcp_seq.h> 57#include <netinet/tcp_timer.h> 58#include <netinet/tcp_var.h> 59#include <netinet/tcpip.h> 60#include <netinet/tcp_debug.h> 61 62int tcprexmtthresh = 3; 63struct tcpiphdr tcp_saveti; 64struct inpcb *tcp_last_inpcb = &tcb; 65 66extern u_long sb_max; 67 68#endif /* TUBA_INCLUDE */ 69#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 70 71/* for modulo comparisons of timestamps */ 72#define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 73#define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 74 75 76/* 77 * Insert segment ti into reassembly queue of tcp with 78 * control block tp. Return TH_FIN if reassembly now includes 79 * a segment with FIN. The macro form does the common case inline 80 * (segment is the next to be received on an established connection, 81 * and the queue is empty), avoiding linkage into and removal 82 * from the queue and repetition of various conversions. 83 * Set DELACK for segments received in order, but ack immediately 84 * when segments are out of order (so fast retransmit can work). 85 */ 86#define TCP_REASS(tp, ti, m, so, flags) { \ 87 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 88 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 89 (tp)->t_state == TCPS_ESTABLISHED) { \ 90 tp->t_flags |= TF_DELACK; \ 91 (tp)->rcv_nxt += (ti)->ti_len; \ 92 flags = (ti)->ti_flags & TH_FIN; \ 93 tcpstat.tcps_rcvpack++;\ 94 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 95 sbappend(&(so)->so_rcv, (m)); \ 96 sorwakeup(so); \ 97 } else { \ 98 (flags) = tcp_reass((tp), (ti), (m)); \ 99 tp->t_flags |= TF_ACKNOW; \ 100 } \ 101} 102#ifndef TUBA_INCLUDE 103 104int 105tcp_reass(tp, ti, m) 106 register struct tcpcb *tp; 107 register struct tcpiphdr *ti; 108 struct mbuf *m; 109{ 110 register struct tcpiphdr *q; 111 struct socket *so = tp->t_inpcb->inp_socket; 112 int flags; 113 114 /* 115 * Call with ti==0 after become established to 116 * force pre-ESTABLISHED data up to user socket. 117 */ 118 if (ti == 0) 119 goto present; 120 121 /* 122 * Find a segment which begins after this one does. 123 */ 124 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 125 q = (struct tcpiphdr *)q->ti_next) 126 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 127 break; 128 129 /* 130 * If there is a preceding segment, it may provide some of 131 * our data already. If so, drop the data from the incoming 132 * segment. If it provides all of our data, drop us. 133 */ 134 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 135 register int i; 136 q = (struct tcpiphdr *)q->ti_prev; 137 /* conversion to int (in i) handles seq wraparound */ 138 i = q->ti_seq + q->ti_len - ti->ti_seq; 139 if (i > 0) { 140 if (i >= ti->ti_len) { 141 tcpstat.tcps_rcvduppack++; 142 tcpstat.tcps_rcvdupbyte += ti->ti_len; 143 m_freem(m); 144 return (0); 145 } 146 m_adj(m, i); 147 ti->ti_len -= i; 148 ti->ti_seq += i; 149 } 150 q = (struct tcpiphdr *)(q->ti_next); 151 } 152 tcpstat.tcps_rcvoopack++; 153 tcpstat.tcps_rcvoobyte += ti->ti_len; 154 REASS_MBUF(ti) = m; /* XXX */ 155 156 /* 157 * While we overlap succeeding segments trim them or, 158 * if they are completely covered, dequeue them. 159 */ 160 while (q != (struct tcpiphdr *)tp) { 161 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 162 if (i <= 0) 163 break; 164 if (i < q->ti_len) { 165 q->ti_seq += i; 166 q->ti_len -= i; 167 m_adj(REASS_MBUF(q), i); 168 break; 169 } 170 q = (struct tcpiphdr *)q->ti_next; 171 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 172 remque(q->ti_prev); 173 m_freem(m); 174 } 175 176 /* 177 * Stick new segment in its place. 178 */ 179 insque(ti, q->ti_prev); 180 181present: 182 /* 183 * Present data to user, advancing rcv_nxt through 184 * completed sequence space. 185 */ 186 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 187 return (0); 188 ti = tp->seg_next; 189 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 190 return (0); 191 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 192 return (0); 193 do { 194 tp->rcv_nxt += ti->ti_len; 195 flags = ti->ti_flags & TH_FIN; 196 remque(ti); 197 m = REASS_MBUF(ti); 198 ti = (struct tcpiphdr *)ti->ti_next; 199 if (so->so_state & SS_CANTRCVMORE) 200 m_freem(m); 201 else 202 sbappend(&so->so_rcv, m); 203 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 204 sorwakeup(so); 205 return (flags); 206} 207 208/* 209 * TCP input routine, follows pages 65-76 of the 210 * protocol specification dated September, 1981 very closely. 211 */ 212void 213tcp_input(m, iphlen) 214 register struct mbuf *m; 215 int iphlen; 216{ 217 register struct tcpiphdr *ti; 218 register struct inpcb *inp; 219 caddr_t optp = NULL; 220 int optlen = 0; 221 int len, tlen, off; 222 register struct tcpcb *tp = 0; 223 register int tiflags; 224 struct socket *so = 0; 225 int todrop, acked, ourfinisacked, needoutput = 0; 226 short ostate = 0; 227 struct in_addr laddr; 228 int dropsocket = 0; 229 int iss = 0; 230 u_long tiwin, ts_val, ts_ecr; 231 int ts_present = 0; 232 233 tcpstat.tcps_rcvtotal++; 234 /* 235 * Get IP and TCP header together in first mbuf. 236 * Note: IP leaves IP header in first mbuf. 237 */ 238 ti = mtod(m, struct tcpiphdr *); 239 if (iphlen > sizeof (struct ip)) 240 ip_stripoptions(m, (struct mbuf *)0); 241 if (m->m_len < sizeof (struct tcpiphdr)) { 242 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 243 tcpstat.tcps_rcvshort++; 244 return; 245 } 246 ti = mtod(m, struct tcpiphdr *); 247 } 248 249 /* 250 * Checksum extended TCP header and data. 251 */ 252 tlen = ((struct ip *)ti)->ip_len; 253 len = sizeof (struct ip) + tlen; 254 ti->ti_next = ti->ti_prev = 0; 255 ti->ti_x1 = 0; 256 ti->ti_len = (u_short)tlen; 257 HTONS(ti->ti_len); 258 if (ti->ti_sum = in_cksum(m, len)) { 259 tcpstat.tcps_rcvbadsum++; 260 goto drop; 261 } 262#endif /* TUBA_INCLUDE */ 263 264 /* 265 * Check that TCP offset makes sense, 266 * pull out TCP options and adjust length. XXX 267 */ 268 off = ti->ti_off << 2; 269 if (off < sizeof (struct tcphdr) || off > tlen) { 270 tcpstat.tcps_rcvbadoff++; 271 goto drop; 272 } 273 tlen -= off; 274 ti->ti_len = tlen; 275 if (off > sizeof (struct tcphdr)) { 276 if (m->m_len < sizeof(struct ip) + off) { 277 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 278 tcpstat.tcps_rcvshort++; 279 return; 280 } 281 ti = mtod(m, struct tcpiphdr *); 282 } 283 optlen = off - sizeof (struct tcphdr); 284 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 285 /* 286 * Do quick retrieval of timestamp options ("options 287 * prediction?"). If timestamp is the only option and it's 288 * formatted as recommended in RFC 1323 appendix A, we 289 * quickly get the values now and not bother calling 290 * tcp_dooptions(), etc. 291 */ 292 if ((optlen == TCPOLEN_TSTAMP_APPA || 293 (optlen > TCPOLEN_TSTAMP_APPA && 294 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 295 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 296 (ti->ti_flags & TH_SYN) == 0) { 297 ts_present = 1; 298 ts_val = ntohl(*(u_long *)(optp + 4)); 299 ts_ecr = ntohl(*(u_long *)(optp + 8)); 300 optp = NULL; /* we've parsed the options */ 301 } 302 } 303 tiflags = ti->ti_flags; 304 305 /* 306 * Convert TCP protocol specific fields to host format. 307 */ 308 NTOHL(ti->ti_seq); 309 NTOHL(ti->ti_ack); 310 NTOHS(ti->ti_win); 311 NTOHS(ti->ti_urp); 312 313 /* 314 * Locate pcb for segment. 315 */ 316findpcb: 317 inp = tcp_last_inpcb; 318 if (inp->inp_lport != ti->ti_dport || 319 inp->inp_fport != ti->ti_sport || 320 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 321 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 322 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 323 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 324 if (inp) 325 tcp_last_inpcb = inp; 326 ++tcpstat.tcps_pcbcachemiss; 327 } 328 329 /* 330 * If the state is CLOSED (i.e., TCB does not exist) then 331 * all data in the incoming segment is discarded. 332 * If the TCB exists but is in CLOSED state, it is embryonic, 333 * but should either do a listen or a connect soon. 334 */ 335 if (inp == 0) 336 goto dropwithreset; 337 tp = intotcpcb(inp); 338 if (tp == 0) 339 goto dropwithreset; 340 if (tp->t_state == TCPS_CLOSED) 341 goto drop; 342 343 /* Unscale the window into a 32-bit value. */ 344 if ((tiflags & TH_SYN) == 0) 345 tiwin = ti->ti_win << tp->snd_scale; 346 else 347 tiwin = ti->ti_win; 348 349 so = inp->inp_socket; 350 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 351 if (so->so_options & SO_DEBUG) { 352 ostate = tp->t_state; 353 tcp_saveti = *ti; 354 } 355 if (so->so_options & SO_ACCEPTCONN) { 356 so = sonewconn(so, 0); 357 if (so == 0) 358 goto drop; 359 /* 360 * This is ugly, but .... 361 * 362 * Mark socket as temporary until we're 363 * committed to keeping it. The code at 364 * ``drop'' and ``dropwithreset'' check the 365 * flag dropsocket to see if the temporary 366 * socket created here should be discarded. 367 * We mark the socket as discardable until 368 * we're committed to it below in TCPS_LISTEN. 369 */ 370 dropsocket++; 371 inp = (struct inpcb *)so->so_pcb; 372 inp->inp_laddr = ti->ti_dst; 373 inp->inp_lport = ti->ti_dport; 374#if BSD>=43 375 inp->inp_options = ip_srcroute(); 376#endif 377 tp = intotcpcb(inp); 378 tp->t_state = TCPS_LISTEN; 379 380 /* Compute proper scaling value from buffer space 381 */ 382 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 383 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 384 tp->request_r_scale++; 385 } 386 } 387 388 /* 389 * Segment received on connection. 390 * Reset idle time and keep-alive timer. 391 */ 392 tp->t_idle = 0; 393 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 394 395 /* 396 * Process options if not in LISTEN state, 397 * else do it below (after getting remote address). 398 */ 399 if (optp && tp->t_state != TCPS_LISTEN) 400 tcp_dooptions(tp, optp, optlen, ti, 401 &ts_present, &ts_val, &ts_ecr); 402 403 /* 404 * Header prediction: check for the two common cases 405 * of a uni-directional data xfer. If the packet has 406 * no control flags, is in-sequence, the window didn't 407 * change and we're not retransmitting, it's a 408 * candidate. If the length is zero and the ack moved 409 * forward, we're the sender side of the xfer. Just 410 * free the data acked & wake any higher level process 411 * that was blocked waiting for space. If the length 412 * is non-zero and the ack didn't move, we're the 413 * receiver side. If we're getting packets in-order 414 * (the reassembly queue is empty), add the data to 415 * the socket buffer and note that we need a delayed ack. 416 */ 417 if (tp->t_state == TCPS_ESTABLISHED && 418 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 419 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 420 ti->ti_seq == tp->rcv_nxt && 421 tiwin && tiwin == tp->snd_wnd && 422 tp->snd_nxt == tp->snd_max) { 423 424 /* 425 * If last ACK falls within this segment's sequence numbers, 426 * record the timestamp. 427 */ 428 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 429 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 430 tp->ts_recent_age = tcp_now; 431 tp->ts_recent = ts_val; 432 } 433 434 if (ti->ti_len == 0) { 435 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 436 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 437 tp->snd_cwnd >= tp->snd_wnd) { 438 /* 439 * this is a pure ack for outstanding data. 440 */ 441 ++tcpstat.tcps_predack; 442 if (ts_present) 443 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 444 else if (tp->t_rtt && 445 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 446 tcp_xmit_timer(tp, tp->t_rtt); 447 acked = ti->ti_ack - tp->snd_una; 448 tcpstat.tcps_rcvackpack++; 449 tcpstat.tcps_rcvackbyte += acked; 450 sbdrop(&so->so_snd, acked); 451 tp->snd_una = ti->ti_ack; 452 m_freem(m); 453 454 /* 455 * If all outstanding data are acked, stop 456 * retransmit timer, otherwise restart timer 457 * using current (possibly backed-off) value. 458 * If process is waiting for space, 459 * wakeup/selwakeup/signal. If data 460 * are ready to send, let tcp_output 461 * decide between more output or persist. 462 */ 463 if (tp->snd_una == tp->snd_max) 464 tp->t_timer[TCPT_REXMT] = 0; 465 else if (tp->t_timer[TCPT_PERSIST] == 0) 466 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 467 468 if (so->so_snd.sb_flags & SB_NOTIFY) 469 sowwakeup(so); 470 if (so->so_snd.sb_cc) 471 (void) tcp_output(tp); 472 return; 473 } 474 } else if (ti->ti_ack == tp->snd_una && 475 tp->seg_next == (struct tcpiphdr *)tp && 476 ti->ti_len <= sbspace(&so->so_rcv)) { 477 /* 478 * this is a pure, in-sequence data packet 479 * with nothing on the reassembly queue and 480 * we have enough buffer space to take it. 481 */ 482 ++tcpstat.tcps_preddat; 483 tp->rcv_nxt += ti->ti_len; 484 tcpstat.tcps_rcvpack++; 485 tcpstat.tcps_rcvbyte += ti->ti_len; 486 /* 487 * Drop TCP, IP headers and TCP options then add data 488 * to socket buffer. 489 */ 490 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 491 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 492 sbappend(&so->so_rcv, m); 493 sorwakeup(so); 494 /* 495 * If this is a small packet, then ACK now - with Nagel 496 * congestion avoidance sender won't send more until 497 * he gets an ACK. 498 */ 499 if ((unsigned)ti->ti_len < tp->t_maxseg) { 500 tp->t_flags |= TF_ACKNOW; 501 tcp_output(tp); 502 } else { 503 tp->t_flags |= TF_DELACK; 504 } 505 return; 506 } 507 } 508 509 /* 510 * Drop TCP, IP headers and TCP options. 511 */ 512 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 513 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 514 515 /* 516 * Calculate amount of space in receive window, 517 * and then do TCP input processing. 518 * Receive window is amount of space in rcv queue, 519 * but not less than advertised window. 520 */ 521 { int win; 522 523 win = sbspace(&so->so_rcv); 524 if (win < 0) 525 win = 0; 526 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 527 } 528 529 switch (tp->t_state) { 530 531 /* 532 * If the state is LISTEN then ignore segment if it contains an RST. 533 * If the segment contains an ACK then it is bad and send a RST. 534 * If it does not contain a SYN then it is not interesting; drop it. 535 * Don't bother responding if the destination was a broadcast. 536 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 537 * tp->iss, and send a segment: 538 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 539 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 540 * Fill in remote peer address fields if not previously specified. 541 * Enter SYN_RECEIVED state, and process any other fields of this 542 * segment in this state. 543 */ 544 case TCPS_LISTEN: { 545 struct mbuf *am; 546 register struct sockaddr_in *sin; 547 548 if (tiflags & TH_RST) 549 goto drop; 550 if (tiflags & TH_ACK) 551 goto dropwithreset; 552 if ((tiflags & TH_SYN) == 0) 553 goto drop; 554 /* 555 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 556 * in_broadcast() should never return true on a received 557 * packet with M_BCAST not set. 558 */ 559 if (m->m_flags & (M_BCAST|M_MCAST) || 560 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 561 goto drop; 562 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 563 if (am == NULL) 564 goto drop; 565 am->m_len = sizeof (struct sockaddr_in); 566 sin = mtod(am, struct sockaddr_in *); 567 sin->sin_family = AF_INET; 568 sin->sin_len = sizeof(*sin); 569 sin->sin_addr = ti->ti_src; 570 sin->sin_port = ti->ti_sport; 571 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 572 laddr = inp->inp_laddr; 573 if (inp->inp_laddr.s_addr == INADDR_ANY) 574 inp->inp_laddr = ti->ti_dst; 575 if (in_pcbconnect(inp, am)) { 576 inp->inp_laddr = laddr; 577 (void) m_free(am); 578 goto drop; 579 } 580 (void) m_free(am); 581 tp->t_template = tcp_template(tp); 582 if (tp->t_template == 0) { 583 tp = tcp_drop(tp, ENOBUFS); 584 dropsocket = 0; /* socket is already gone */ 585 goto drop; 586 } 587 if (optp) 588 tcp_dooptions(tp, optp, optlen, ti, 589 &ts_present, &ts_val, &ts_ecr); 590 if (iss) 591 tp->iss = iss; 592 else 593 tp->iss = tcp_iss; 594 tcp_iss += TCP_ISSINCR/2; 595 tp->irs = ti->ti_seq; 596 tcp_sendseqinit(tp); 597 tcp_rcvseqinit(tp); 598 tp->t_flags |= TF_ACKNOW; 599 tp->t_state = TCPS_SYN_RECEIVED; 600 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 601 dropsocket = 0; /* committed to socket */ 602 tcpstat.tcps_accepts++; 603 goto trimthenstep6; 604 } 605 606 /* 607 * If the state is SYN_SENT: 608 * if seg contains an ACK, but not for our SYN, drop the input. 609 * if seg contains a RST, then drop the connection. 610 * if seg does not contain SYN, then drop it. 611 * Otherwise this is an acceptable SYN segment 612 * initialize tp->rcv_nxt and tp->irs 613 * if seg contains ack then advance tp->snd_una 614 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 615 * arrange for segment to be acked (eventually) 616 * continue processing rest of data/controls, beginning with URG 617 */ 618 case TCPS_SYN_SENT: 619 if ((tiflags & TH_ACK) && 620 (SEQ_LEQ(ti->ti_ack, tp->iss) || 621 SEQ_GT(ti->ti_ack, tp->snd_max))) 622 goto dropwithreset; 623 if (tiflags & TH_RST) { 624 if (tiflags & TH_ACK) 625 tp = tcp_drop(tp, ECONNREFUSED); 626 goto drop; 627 } 628 if ((tiflags & TH_SYN) == 0) 629 goto drop; 630 if (tiflags & TH_ACK) { 631 tp->snd_una = ti->ti_ack; 632 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 633 tp->snd_nxt = tp->snd_una; 634 } 635 tp->t_timer[TCPT_REXMT] = 0; 636 tp->irs = ti->ti_seq; 637 tcp_rcvseqinit(tp); 638 tp->t_flags |= TF_ACKNOW; 639 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 640 tcpstat.tcps_connects++; 641 soisconnected(so); 642 tp->t_state = TCPS_ESTABLISHED; 643 /* Do window scaling on this connection? */ 644 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 645 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 646 tp->snd_scale = tp->requested_s_scale; 647 tp->rcv_scale = tp->request_r_scale; 648 } 649 (void) tcp_reass(tp, (struct tcpiphdr *)0, 650 (struct mbuf *)0); 651 /* 652 * if we didn't have to retransmit the SYN, 653 * use its rtt as our initial srtt & rtt var. 654 */ 655 if (tp->t_rtt) 656 tcp_xmit_timer(tp, tp->t_rtt); 657 } else 658 tp->t_state = TCPS_SYN_RECEIVED; 659 660trimthenstep6: 661 /* 662 * Advance ti->ti_seq to correspond to first data byte. 663 * If data, trim to stay within window, 664 * dropping FIN if necessary. 665 */ 666 ti->ti_seq++; 667 if (ti->ti_len > tp->rcv_wnd) { 668 todrop = ti->ti_len - tp->rcv_wnd; 669 m_adj(m, -todrop); 670 ti->ti_len = tp->rcv_wnd; 671 tiflags &= ~TH_FIN; 672 tcpstat.tcps_rcvpackafterwin++; 673 tcpstat.tcps_rcvbyteafterwin += todrop; 674 } 675 tp->snd_wl1 = ti->ti_seq - 1; 676 tp->rcv_up = ti->ti_seq; 677 goto step6; 678 } 679 680 /* 681 * States other than LISTEN or SYN_SENT. 682 * First check timestamp, if present. 683 * Then check that at least some bytes of segment are within 684 * receive window. If segment begins before rcv_nxt, 685 * drop leading data (and SYN); if nothing left, just ack. 686 * 687 * RFC 1323 PAWS: If we have a timestamp reply on this segment 688 * and it's less than ts_recent, drop it. 689 */ 690 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 691 TSTMP_LT(ts_val, tp->ts_recent)) { 692 693 /* Check to see if ts_recent is over 24 days old. */ 694 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 695 /* 696 * Invalidate ts_recent. If this segment updates 697 * ts_recent, the age will be reset later and ts_recent 698 * will get a valid value. If it does not, setting 699 * ts_recent to zero will at least satisfy the 700 * requirement that zero be placed in the timestamp 701 * echo reply when ts_recent isn't valid. The 702 * age isn't reset until we get a valid ts_recent 703 * because we don't want out-of-order segments to be 704 * dropped when ts_recent is old. 705 */ 706 tp->ts_recent = 0; 707 } else { 708 tcpstat.tcps_rcvduppack++; 709 tcpstat.tcps_rcvdupbyte += ti->ti_len; 710 tcpstat.tcps_pawsdrop++; 711 goto dropafterack; 712 } 713 } 714 715 todrop = tp->rcv_nxt - ti->ti_seq; 716 if (todrop > 0) { 717 if (tiflags & TH_SYN) { 718 tiflags &= ~TH_SYN; 719 ti->ti_seq++; 720 if (ti->ti_urp > 1) 721 ti->ti_urp--; 722 else 723 tiflags &= ~TH_URG; 724 todrop--; 725 } 726 if (todrop >= ti->ti_len) { 727 tcpstat.tcps_rcvduppack++; 728 tcpstat.tcps_rcvdupbyte += ti->ti_len; 729 /* 730 * If segment is just one to the left of the window, 731 * check two special cases: 732 * 1. Don't toss RST in response to 4.2-style keepalive. 733 * 2. If the only thing to drop is a FIN, we can drop 734 * it, but check the ACK or we will get into FIN 735 * wars if our FINs crossed (both CLOSING). 736 * In either case, send ACK to resynchronize, 737 * but keep on processing for RST or ACK. 738 */ 739 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 740#ifdef TCP_COMPAT_42 741 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 742#endif 743 ) { 744 todrop = ti->ti_len; 745 tiflags &= ~TH_FIN; 746 tp->t_flags |= TF_ACKNOW; 747 } else { 748 /* 749 * Handle the case when a bound socket connects 750 * to itself. Allow packets with a SYN and 751 * an ACK to continue with the processing. 752 */ 753 if (todrop != 0 || (tiflags & TH_ACK) == 0) 754 goto dropafterack; 755 } 756 } else { 757 tcpstat.tcps_rcvpartduppack++; 758 tcpstat.tcps_rcvpartdupbyte += todrop; 759 } 760 m_adj(m, todrop); 761 ti->ti_seq += todrop; 762 ti->ti_len -= todrop; 763 if (ti->ti_urp > todrop) 764 ti->ti_urp -= todrop; 765 else { 766 tiflags &= ~TH_URG; 767 ti->ti_urp = 0; 768 } 769 } 770 771 /* 772 * If new data are received on a connection after the 773 * user processes are gone, then RST the other end. 774 */ 775 if ((so->so_state & SS_NOFDREF) && 776 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 777 tp = tcp_close(tp); 778 tcpstat.tcps_rcvafterclose++; 779 goto dropwithreset; 780 } 781 782 /* 783 * If segment ends after window, drop trailing data 784 * (and PUSH and FIN); if nothing left, just ACK. 785 */ 786 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 787 if (todrop > 0) { 788 tcpstat.tcps_rcvpackafterwin++; 789 if (todrop >= ti->ti_len) { 790 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 791 /* 792 * If a new connection request is received 793 * while in TIME_WAIT, drop the old connection 794 * and start over if the sequence numbers 795 * are above the previous ones. 796 */ 797 if (tiflags & TH_SYN && 798 tp->t_state == TCPS_TIME_WAIT && 799 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 800 iss = tp->rcv_nxt + TCP_ISSINCR; 801 tp = tcp_close(tp); 802 goto findpcb; 803 } 804 /* 805 * If window is closed can only take segments at 806 * window edge, and have to drop data and PUSH from 807 * incoming segments. Continue processing, but 808 * remember to ack. Otherwise, drop segment 809 * and ack. 810 */ 811 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 812 tp->t_flags |= TF_ACKNOW; 813 tcpstat.tcps_rcvwinprobe++; 814 } else 815 goto dropafterack; 816 } else 817 tcpstat.tcps_rcvbyteafterwin += todrop; 818 m_adj(m, -todrop); 819 ti->ti_len -= todrop; 820 tiflags &= ~(TH_PUSH|TH_FIN); 821 } 822 823 /* 824 * If last ACK falls within this segment's sequence numbers, 825 * record its timestamp. 826 */ 827 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 828 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 829 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 830 tp->ts_recent_age = tcp_now; 831 tp->ts_recent = ts_val; 832 } 833 834 /* 835 * If the RST bit is set examine the state: 836 * SYN_RECEIVED STATE: 837 * If passive open, return to LISTEN state. 838 * If active open, inform user that connection was refused. 839 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 840 * Inform user that connection was reset, and close tcb. 841 * CLOSING, LAST_ACK, TIME_WAIT STATES 842 * Close the tcb. 843 */ 844 if (tiflags&TH_RST) switch (tp->t_state) { 845 846 case TCPS_SYN_RECEIVED: 847 so->so_error = ECONNREFUSED; 848 goto close; 849 850 case TCPS_ESTABLISHED: 851 case TCPS_FIN_WAIT_1: 852 case TCPS_FIN_WAIT_2: 853 case TCPS_CLOSE_WAIT: 854 so->so_error = ECONNRESET; 855 close: 856 tp->t_state = TCPS_CLOSED; 857 tcpstat.tcps_drops++; 858 tp = tcp_close(tp); 859 goto drop; 860 861 case TCPS_CLOSING: 862 case TCPS_LAST_ACK: 863 case TCPS_TIME_WAIT: 864 tp = tcp_close(tp); 865 goto drop; 866 } 867 868 /* 869 * If a SYN is in the window, then this is an 870 * error and we send an RST and drop the connection. 871 */ 872 if (tiflags & TH_SYN) { 873 tp = tcp_drop(tp, ECONNRESET); 874 goto dropwithreset; 875 } 876 877 /* 878 * If the ACK bit is off we drop the segment and return. 879 */ 880 if ((tiflags & TH_ACK) == 0) 881 goto drop; 882 883 /* 884 * Ack processing. 885 */ 886 switch (tp->t_state) { 887 888 /* 889 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 890 * ESTABLISHED state and continue processing, otherwise 891 * send an RST. 892 */ 893 case TCPS_SYN_RECEIVED: 894 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 895 SEQ_GT(ti->ti_ack, tp->snd_max)) 896 goto dropwithreset; 897 tcpstat.tcps_connects++; 898 soisconnected(so); 899 tp->t_state = TCPS_ESTABLISHED; 900 /* Do window scaling? */ 901 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 902 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 903 tp->snd_scale = tp->requested_s_scale; 904 tp->rcv_scale = tp->request_r_scale; 905 } 906 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 907 tp->snd_wl1 = ti->ti_seq - 1; 908 /* fall into ... */ 909 910 /* 911 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 912 * ACKs. If the ack is in the range 913 * tp->snd_una < ti->ti_ack <= tp->snd_max 914 * then advance tp->snd_una to ti->ti_ack and drop 915 * data from the retransmission queue. If this ACK reflects 916 * more up to date window information we update our window information. 917 */ 918 case TCPS_ESTABLISHED: 919 case TCPS_FIN_WAIT_1: 920 case TCPS_FIN_WAIT_2: 921 case TCPS_CLOSE_WAIT: 922 case TCPS_CLOSING: 923 case TCPS_LAST_ACK: 924 case TCPS_TIME_WAIT: 925 926 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 927 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 928 tcpstat.tcps_rcvdupack++; 929 /* 930 * If we have outstanding data (other than 931 * a window probe), this is a completely 932 * duplicate ack (ie, window info didn't 933 * change), the ack is the biggest we've 934 * seen and we've seen exactly our rexmt 935 * threshhold of them, assume a packet 936 * has been dropped and retransmit it. 937 * Kludge snd_nxt & the congestion 938 * window so we send only this one 939 * packet. 940 * 941 * We know we're losing at the current 942 * window size so do congestion avoidance 943 * (set ssthresh to half the current window 944 * and pull our congestion window back to 945 * the new ssthresh). 946 * 947 * Dup acks mean that packets have left the 948 * network (they're now cached at the receiver) 949 * so bump cwnd by the amount in the receiver 950 * to keep a constant cwnd packets in the 951 * network. 952 */ 953 if (tp->t_timer[TCPT_REXMT] == 0 || 954 ti->ti_ack != tp->snd_una) 955 tp->t_dupacks = 0; 956 else if (++tp->t_dupacks == tcprexmtthresh) { 957 tcp_seq onxt = tp->snd_nxt; 958 u_int win = 959 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 960 tp->t_maxseg; 961 962 if (win < 2) 963 win = 2; 964 tp->snd_ssthresh = win * tp->t_maxseg; 965 tp->t_timer[TCPT_REXMT] = 0; 966 tp->t_rtt = 0; 967 tp->snd_nxt = ti->ti_ack; 968 tp->snd_cwnd = tp->t_maxseg; 969 (void) tcp_output(tp); 970 tp->snd_cwnd = tp->snd_ssthresh + 971 tp->t_maxseg * tp->t_dupacks; 972 if (SEQ_GT(onxt, tp->snd_nxt)) 973 tp->snd_nxt = onxt; 974 goto drop; 975 } else if (tp->t_dupacks > tcprexmtthresh) { 976 tp->snd_cwnd += tp->t_maxseg; 977 (void) tcp_output(tp); 978 goto drop; 979 } 980 } else 981 tp->t_dupacks = 0; 982 break; 983 } 984 /* 985 * If the congestion window was inflated to account 986 * for the other side's cached packets, retract it. 987 */ 988 if (tp->t_dupacks > tcprexmtthresh && 989 tp->snd_cwnd > tp->snd_ssthresh) 990 tp->snd_cwnd = tp->snd_ssthresh; 991 tp->t_dupacks = 0; 992 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 993 tcpstat.tcps_rcvacktoomuch++; 994 goto dropafterack; 995 } 996 acked = ti->ti_ack - tp->snd_una; 997 tcpstat.tcps_rcvackpack++; 998 tcpstat.tcps_rcvackbyte += acked; 999 1000 /* 1001 * If we have a timestamp reply, update smoothed 1002 * round trip time. If no timestamp is present but 1003 * transmit timer is running and timed sequence 1004 * number was acked, update smoothed round trip time. 1005 * Since we now have an rtt measurement, cancel the 1006 * timer backoff (cf., Phil Karn's retransmit alg.). 1007 * Recompute the initial retransmit timer. 1008 */ 1009 if (ts_present) 1010 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 1011 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1012 tcp_xmit_timer(tp,tp->t_rtt); 1013 1014 /* 1015 * If all outstanding data is acked, stop retransmit 1016 * timer and remember to restart (more output or persist). 1017 * If there is more data to be acked, restart retransmit 1018 * timer, using current (possibly backed-off) value. 1019 */ 1020 if (ti->ti_ack == tp->snd_max) { 1021 tp->t_timer[TCPT_REXMT] = 0; 1022 needoutput = 1; 1023 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1024 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1025 /* 1026 * When new data is acked, open the congestion window. 1027 * If the window gives us less than ssthresh packets 1028 * in flight, open exponentially (maxseg per packet). 1029 * Otherwise open linearly: maxseg per window 1030 * (maxseg^2 / cwnd per packet), plus a constant 1031 * fraction of a packet (maxseg/8) to help larger windows 1032 * open quickly enough. 1033 */ 1034 { 1035 register u_int cw = tp->snd_cwnd; 1036 register u_int incr = tp->t_maxseg; 1037 1038 if (cw > tp->snd_ssthresh) 1039 incr = incr * incr / cw + incr / 8; 1040 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1041 } 1042 if (acked > so->so_snd.sb_cc) { 1043 tp->snd_wnd -= so->so_snd.sb_cc; 1044 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1045 ourfinisacked = 1; 1046 } else { 1047 sbdrop(&so->so_snd, acked); 1048 tp->snd_wnd -= acked; 1049 ourfinisacked = 0; 1050 } 1051 if (so->so_snd.sb_flags & SB_NOTIFY) 1052 sowwakeup(so); 1053 tp->snd_una = ti->ti_ack; 1054 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1055 tp->snd_nxt = tp->snd_una; 1056 1057 switch (tp->t_state) { 1058 1059 /* 1060 * In FIN_WAIT_1 STATE in addition to the processing 1061 * for the ESTABLISHED state if our FIN is now acknowledged 1062 * then enter FIN_WAIT_2. 1063 */ 1064 case TCPS_FIN_WAIT_1: 1065 if (ourfinisacked) { 1066 /* 1067 * If we can't receive any more 1068 * data, then closing user can proceed. 1069 * Starting the timer is contrary to the 1070 * specification, but if we don't get a FIN 1071 * we'll hang forever. 1072 */ 1073 if (so->so_state & SS_CANTRCVMORE) { 1074 soisdisconnected(so); 1075 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1076 } 1077 tp->t_state = TCPS_FIN_WAIT_2; 1078 } 1079 break; 1080 1081 /* 1082 * In CLOSING STATE in addition to the processing for 1083 * the ESTABLISHED state if the ACK acknowledges our FIN 1084 * then enter the TIME-WAIT state, otherwise ignore 1085 * the segment. 1086 */ 1087 case TCPS_CLOSING: 1088 if (ourfinisacked) { 1089 tp->t_state = TCPS_TIME_WAIT; 1090 tcp_canceltimers(tp); 1091 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1092 soisdisconnected(so); 1093 } 1094 break; 1095 1096 /* 1097 * In LAST_ACK, we may still be waiting for data to drain 1098 * and/or to be acked, as well as for the ack of our FIN. 1099 * If our FIN is now acknowledged, delete the TCB, 1100 * enter the closed state and return. 1101 */ 1102 case TCPS_LAST_ACK: 1103 if (ourfinisacked) { 1104 tp = tcp_close(tp); 1105 goto drop; 1106 } 1107 break; 1108 1109 /* 1110 * In TIME_WAIT state the only thing that should arrive 1111 * is a retransmission of the remote FIN. Acknowledge 1112 * it and restart the finack timer. 1113 */ 1114 case TCPS_TIME_WAIT: 1115 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1116 goto dropafterack; 1117 } 1118 } 1119 1120step6: 1121 /* 1122 * Update window information. 1123 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1124 */ 1125 if ((tiflags & TH_ACK) && 1126 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1127 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1128 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1129 /* keep track of pure window updates */ 1130 if (ti->ti_len == 0 && 1131 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1132 tcpstat.tcps_rcvwinupd++; 1133 tp->snd_wnd = tiwin; 1134 tp->snd_wl1 = ti->ti_seq; 1135 tp->snd_wl2 = ti->ti_ack; 1136 if (tp->snd_wnd > tp->max_sndwnd) 1137 tp->max_sndwnd = tp->snd_wnd; 1138 needoutput = 1; 1139 } 1140 1141 /* 1142 * Process segments with URG. 1143 */ 1144 if ((tiflags & TH_URG) && ti->ti_urp && 1145 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1146 /* 1147 * This is a kludge, but if we receive and accept 1148 * random urgent pointers, we'll crash in 1149 * soreceive. It's hard to imagine someone 1150 * actually wanting to send this much urgent data. 1151 */ 1152 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1153 ti->ti_urp = 0; /* XXX */ 1154 tiflags &= ~TH_URG; /* XXX */ 1155 goto dodata; /* XXX */ 1156 } 1157 /* 1158 * If this segment advances the known urgent pointer, 1159 * then mark the data stream. This should not happen 1160 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1161 * a FIN has been received from the remote side. 1162 * In these states we ignore the URG. 1163 * 1164 * According to RFC961 (Assigned Protocols), 1165 * the urgent pointer points to the last octet 1166 * of urgent data. We continue, however, 1167 * to consider it to indicate the first octet 1168 * of data past the urgent section as the original 1169 * spec states (in one of two places). 1170 */ 1171 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1172 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1173 so->so_oobmark = so->so_rcv.sb_cc + 1174 (tp->rcv_up - tp->rcv_nxt) - 1; 1175 if (so->so_oobmark == 0) 1176 so->so_state |= SS_RCVATMARK; 1177 sohasoutofband(so); 1178 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1179 } 1180 /* 1181 * Remove out of band data so doesn't get presented to user. 1182 * This can happen independent of advancing the URG pointer, 1183 * but if two URG's are pending at once, some out-of-band 1184 * data may creep in... ick. 1185 */ 1186 if (ti->ti_urp <= (u_long)ti->ti_len 1187#ifdef SO_OOBINLINE 1188 && (so->so_options & SO_OOBINLINE) == 0 1189#endif 1190 ) 1191 tcp_pulloutofband(so, ti, m); 1192 } else 1193 /* 1194 * If no out of band data is expected, 1195 * pull receive urgent pointer along 1196 * with the receive window. 1197 */ 1198 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1199 tp->rcv_up = tp->rcv_nxt; 1200dodata: /* XXX */ 1201 1202 /* 1203 * Process the segment text, merging it into the TCP sequencing queue, 1204 * and arranging for acknowledgment of receipt if necessary. 1205 * This process logically involves adjusting tp->rcv_wnd as data 1206 * is presented to the user (this happens in tcp_usrreq.c, 1207 * case PRU_RCVD). If a FIN has already been received on this 1208 * connection then we just ignore the text. 1209 */ 1210 if ((ti->ti_len || (tiflags&TH_FIN)) && 1211 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1212 TCP_REASS(tp, ti, m, so, tiflags); 1213 /* 1214 * Note the amount of data that peer has sent into 1215 * our window, in order to estimate the sender's 1216 * buffer size. 1217 */ 1218 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1219 } else { 1220 m_freem(m); 1221 tiflags &= ~TH_FIN; 1222 } 1223 1224 /* 1225 * If FIN is received ACK the FIN and let the user know 1226 * that the connection is closing. 1227 */ 1228 if (tiflags & TH_FIN) { 1229 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1230 socantrcvmore(so); 1231 tp->t_flags |= TF_ACKNOW; 1232 tp->rcv_nxt++; 1233 } 1234 switch (tp->t_state) { 1235 1236 /* 1237 * In SYN_RECEIVED and ESTABLISHED STATES 1238 * enter the CLOSE_WAIT state. 1239 */ 1240 case TCPS_SYN_RECEIVED: 1241 case TCPS_ESTABLISHED: 1242 tp->t_state = TCPS_CLOSE_WAIT; 1243 break; 1244 1245 /* 1246 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1247 * enter the CLOSING state. 1248 */ 1249 case TCPS_FIN_WAIT_1: 1250 tp->t_state = TCPS_CLOSING; 1251 break; 1252 1253 /* 1254 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1255 * starting the time-wait timer, turning off the other 1256 * standard timers. 1257 */ 1258 case TCPS_FIN_WAIT_2: 1259 tp->t_state = TCPS_TIME_WAIT; 1260 tcp_canceltimers(tp); 1261 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1262 soisdisconnected(so); 1263 break; 1264 1265 /* 1266 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1267 */ 1268 case TCPS_TIME_WAIT: 1269 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1270 break; 1271 } 1272 } 1273 if (so->so_options & SO_DEBUG) 1274 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1275 1276 /* 1277 * If this is a small packet, then ACK now - with Nagel 1278 * congestion avoidance sender won't send more until 1279 * he gets an ACK. 1280 */ 1281 if (ti->ti_len && ((unsigned)ti->ti_len < tp->t_maxseg)) 1282 tp->t_flags |= TF_ACKNOW; 1283 1284 /* 1285 * Return any desired output. 1286 */ 1287 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1288 (void) tcp_output(tp); 1289 return; 1290 1291dropafterack: 1292 /* 1293 * Generate an ACK dropping incoming segment if it occupies 1294 * sequence space, where the ACK reflects our state. 1295 */ 1296 if (tiflags & TH_RST) 1297 goto drop; 1298 m_freem(m); 1299 tp->t_flags |= TF_ACKNOW; 1300 (void) tcp_output(tp); 1301 return; 1302 1303dropwithreset: 1304 /* 1305 * Generate a RST, dropping incoming segment. 1306 * Make ACK acceptable to originator of segment. 1307 * Don't bother to respond if destination was broadcast/multicast. 1308 */ 1309 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1310 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1311 goto drop; 1312 if (tiflags & TH_ACK) 1313 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1314 else { 1315 if (tiflags & TH_SYN) 1316 ti->ti_len++; 1317 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1318 TH_RST|TH_ACK); 1319 } 1320 /* destroy temporarily created socket */ 1321 if (dropsocket) 1322 (void) soabort(so); 1323 return; 1324 1325drop: 1326 /* 1327 * Drop space held by incoming segment and return. 1328 */ 1329 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1330 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1331 m_freem(m); 1332 /* destroy temporarily created socket */ 1333 if (dropsocket) 1334 (void) soabort(so); 1335 return; 1336#ifndef TUBA_INCLUDE 1337} 1338 1339void 1340tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1341 struct tcpcb *tp; 1342 u_char *cp; 1343 int cnt; 1344 struct tcpiphdr *ti; 1345 int *ts_present; 1346 u_long *ts_val, *ts_ecr; 1347{ 1348 u_short mss; 1349 int opt, optlen; 1350 1351 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1352 opt = cp[0]; 1353 if (opt == TCPOPT_EOL) 1354 break; 1355 if (opt == TCPOPT_NOP) 1356 optlen = 1; 1357 else { 1358 optlen = cp[1]; 1359 if (optlen <= 0) 1360 break; 1361 } 1362 switch (opt) { 1363 1364 default: 1365 continue; 1366 1367 case TCPOPT_MAXSEG: 1368 if (optlen != TCPOLEN_MAXSEG) 1369 continue; 1370 if (!(ti->ti_flags & TH_SYN)) 1371 continue; 1372 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1373 NTOHS(mss); 1374 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1375 break; 1376 1377 case TCPOPT_WINDOW: 1378 if (optlen != TCPOLEN_WINDOW) 1379 continue; 1380 if (!(ti->ti_flags & TH_SYN)) 1381 continue; 1382 tp->t_flags |= TF_RCVD_SCALE; 1383 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1384 break; 1385 1386 case TCPOPT_TIMESTAMP: 1387 if (optlen != TCPOLEN_TIMESTAMP) 1388 continue; 1389 *ts_present = 1; 1390 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1391 NTOHL(*ts_val); 1392 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1393 NTOHL(*ts_ecr); 1394 1395 /* 1396 * A timestamp received in a SYN makes 1397 * it ok to send timestamp requests and replies. 1398 */ 1399 if (ti->ti_flags & TH_SYN) { 1400 tp->t_flags |= TF_RCVD_TSTMP; 1401 tp->ts_recent = *ts_val; 1402 tp->ts_recent_age = tcp_now; 1403 } 1404 break; 1405 } 1406 } 1407} 1408 1409/* 1410 * Pull out of band byte out of a segment so 1411 * it doesn't appear in the user's data queue. 1412 * It is still reflected in the segment length for 1413 * sequencing purposes. 1414 */ 1415void 1416tcp_pulloutofband(so, ti, m) 1417 struct socket *so; 1418 struct tcpiphdr *ti; 1419 register struct mbuf *m; 1420{ 1421 int cnt = ti->ti_urp - 1; 1422 1423 while (cnt >= 0) { 1424 if (m->m_len > cnt) { 1425 char *cp = mtod(m, caddr_t) + cnt; 1426 struct tcpcb *tp = sototcpcb(so); 1427 1428 tp->t_iobc = *cp; 1429 tp->t_oobflags |= TCPOOB_HAVEDATA; 1430 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1431 m->m_len--; 1432 return; 1433 } 1434 cnt -= m->m_len; 1435 m = m->m_next; 1436 if (m == 0) 1437 break; 1438 } 1439 panic("tcp_pulloutofband"); 1440} 1441 1442/* 1443 * Collect new round-trip time estimate 1444 * and update averages and current timeout. 1445 */ 1446void 1447tcp_xmit_timer(tp, rtt) 1448 register struct tcpcb *tp; 1449 short rtt; 1450{ 1451 register short delta; 1452 1453 tcpstat.tcps_rttupdated++; 1454 if (tp->t_srtt != 0) { 1455 /* 1456 * srtt is stored as fixed point with 3 bits after the 1457 * binary point (i.e., scaled by 8). The following magic 1458 * is equivalent to the smoothing algorithm in rfc793 with 1459 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1460 * point). Adjust rtt to origin 0. 1461 */ 1462 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1463 if ((tp->t_srtt += delta) <= 0) 1464 tp->t_srtt = 1; 1465 /* 1466 * We accumulate a smoothed rtt variance (actually, a 1467 * smoothed mean difference), then set the retransmit 1468 * timer to smoothed rtt + 4 times the smoothed variance. 1469 * rttvar is stored as fixed point with 2 bits after the 1470 * binary point (scaled by 4). The following is 1471 * equivalent to rfc793 smoothing with an alpha of .75 1472 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1473 * rfc793's wired-in beta. 1474 */ 1475 if (delta < 0) 1476 delta = -delta; 1477 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1478 if ((tp->t_rttvar += delta) <= 0) 1479 tp->t_rttvar = 1; 1480 } else { 1481 /* 1482 * No rtt measurement yet - use the unsmoothed rtt. 1483 * Set the variance to half the rtt (so our first 1484 * retransmit happens at 3*rtt). 1485 */ 1486 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1487 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1488 } 1489 tp->t_rtt = 0; 1490 tp->t_rxtshift = 0; 1491 1492 /* 1493 * the retransmit should happen at rtt + 4 * rttvar. 1494 * Because of the way we do the smoothing, srtt and rttvar 1495 * will each average +1/2 tick of bias. When we compute 1496 * the retransmit timer, we want 1/2 tick of rounding and 1497 * 1 extra tick because of +-1/2 tick uncertainty in the 1498 * firing of the timer. The bias will give us exactly the 1499 * 1.5 tick we need. But, because the bias is 1500 * statistical, we have to test that we don't drop below 1501 * the minimum feasible timer (which is 2 ticks). 1502 */ 1503 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1504 tp->t_rttmin, TCPTV_REXMTMAX); 1505 1506 /* 1507 * We received an ack for a packet that wasn't retransmitted; 1508 * it is probably safe to discard any error indications we've 1509 * received recently. This isn't quite right, but close enough 1510 * for now (a route might have failed after we sent a segment, 1511 * and the return path might not be symmetrical). 1512 */ 1513 tp->t_softerror = 0; 1514} 1515 1516/* 1517 * Determine a reasonable value for maxseg size. 1518 * If the route is known, check route for mtu. 1519 * If none, use an mss that can be handled on the outgoing 1520 * interface without forcing IP to fragment; if bigger than 1521 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1522 * to utilize large mbufs. If no route is found, route has no mtu, 1523 * or the destination isn't local, use a default, hopefully conservative 1524 * size (usually 512 or the default IP max size, but no more than the mtu 1525 * of the interface), as we can't discover anything about intervening 1526 * gateways or networks. We also initialize the congestion/slow start 1527 * window to be a single segment if the destination isn't local. 1528 * While looking at the routing entry, we also initialize other path-dependent 1529 * parameters from pre-set or cached values in the routing entry. 1530 */ 1531int 1532tcp_mss(tp, offer) 1533 register struct tcpcb *tp; 1534 u_int offer; 1535{ 1536 struct route *ro; 1537 register struct rtentry *rt; 1538 struct ifnet *ifp; 1539 register int rtt, mss; 1540 u_long bufsize; 1541 struct inpcb *inp; 1542 struct socket *so; 1543 extern int tcp_mssdflt; 1544 1545 inp = tp->t_inpcb; 1546 ro = &inp->inp_route; 1547 1548 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1549 /* No route yet, so try to acquire one */ 1550 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1551 ro->ro_dst.sa_family = AF_INET; 1552 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1553 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1554 inp->inp_faddr; 1555 rtalloc(ro); 1556 } 1557 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1558 return (tcp_mssdflt); 1559 } 1560 ifp = rt->rt_ifp; 1561 so = inp->inp_socket; 1562 1563#ifdef RTV_MTU /* if route characteristics exist ... */ 1564 /* 1565 * While we're here, check if there's an initial rtt 1566 * or rttvar. Convert from the route-table units 1567 * to scaled multiples of the slow timeout timer. 1568 */ 1569 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1570 /* 1571 * XXX the lock bit for MTU indicates that the value 1572 * is also a minimum value; this is subject to time. 1573 */ 1574 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1575 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1576 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1577 if (rt->rt_rmx.rmx_rttvar) 1578 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1579 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1580 else 1581 /* default variation is +- 1 rtt */ 1582 tp->t_rttvar = 1583 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1584 TCPT_RANGESET(tp->t_rxtcur, 1585 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1586 tp->t_rttmin, TCPTV_REXMTMAX); 1587 } 1588 /* 1589 * if there's an mtu associated with the route, use it 1590 */ 1591 if (rt->rt_rmx.rmx_mtu) 1592 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1593 else 1594#endif /* RTV_MTU */ 1595 { 1596 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1597#if (MCLBYTES & (MCLBYTES - 1)) == 0 1598 if (mss > MCLBYTES) 1599 mss &= ~(MCLBYTES-1); 1600#else 1601 if (mss > MCLBYTES) 1602 mss = mss / MCLBYTES * MCLBYTES; 1603#endif 1604 if (!in_localaddr(inp->inp_faddr)) 1605 mss = min(mss, tcp_mssdflt); 1606 } 1607 /* 1608 * The current mss, t_maxseg, is initialized to the default value. 1609 * If we compute a smaller value, reduce the current mss. 1610 * If we compute a larger value, return it for use in sending 1611 * a max seg size option, but don't store it for use 1612 * unless we received an offer at least that large from peer. 1613 * However, do not accept offers under 32 bytes. 1614 */ 1615 if (offer) 1616 mss = min(mss, offer); 1617 mss = max(mss, 32); /* sanity */ 1618 if (mss < tp->t_maxseg || offer != 0) { 1619 /* 1620 * If there's a pipesize, change the socket buffer 1621 * to that size. Make the socket buffers an integral 1622 * number of mss units; if the mss is larger than 1623 * the socket buffer, decrease the mss. 1624 */ 1625#ifdef RTV_SPIPE 1626 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1627#endif 1628 bufsize = so->so_snd.sb_hiwat; 1629 if (bufsize < mss) 1630 mss = bufsize; 1631 else { 1632 bufsize = roundup(bufsize, mss); 1633 if (bufsize > sb_max) 1634 bufsize = sb_max; 1635 (void)sbreserve(&so->so_snd, bufsize); 1636 } 1637 tp->t_maxseg = mss; 1638 1639#ifdef RTV_RPIPE 1640 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1641#endif 1642 bufsize = so->so_rcv.sb_hiwat; 1643 if (bufsize > mss) { 1644 bufsize = roundup(bufsize, mss); 1645 if (bufsize > sb_max) 1646 bufsize = sb_max; 1647 (void)sbreserve(&so->so_rcv, bufsize); 1648 } 1649 } 1650 tp->snd_cwnd = mss; 1651 1652#ifdef RTV_SSTHRESH 1653 if (rt->rt_rmx.rmx_ssthresh) { 1654 /* 1655 * There's some sort of gateway or interface 1656 * buffer limit on the path. Use this to set 1657 * the slow start threshhold, but set the 1658 * threshold to no less than 2*mss. 1659 */ 1660 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1661 } 1662#endif /* RTV_MTU */ 1663 return (mss); 1664} 1665#endif /* TUBA_INCLUDE */ 1666