1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 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 8.12 (Berkeley) 5/24/95 34 * $Id: tcp_input.c,v 1.66 1997/11/20 20:04:49 wollman Exp $ 35 */ 36 37 #include "opt_tcpdebug.h" 38 39 #ifndef TUBA_INCLUDE 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/sysctl.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/proc.h> /* for proc0 declaration */ 47 #include <sys/protosw.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/syslog.h> 51 52 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 53 54 #include <net/if.h> 55 #include <net/route.h> 56 57 #include <netinet/in.h> 58 #include <netinet/in_systm.h> 59 #include <netinet/ip.h> 60 #include <netinet/in_pcb.h> 61 #include <netinet/ip_var.h> 62 #include <netinet/tcp.h> 63 #include <netinet/tcp_fsm.h> 64 #include <netinet/tcp_seq.h> 65 #include <netinet/tcp_timer.h> 66 #include <netinet/tcp_var.h> 67 #include <netinet/tcpip.h> 68 #ifdef TCPDEBUG 69 #include <netinet/tcp_debug.h> 70 static struct tcpiphdr tcp_saveti; 71 #endif 72 73 static int tcprexmtthresh = 3; 74 tcp_seq tcp_iss; 75 tcp_cc tcp_ccgen; 76 77 struct tcpstat tcpstat; 78 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 79 CTLFLAG_RD, &tcpstat , tcpstat, ""); 80 81 static int log_in_vain = 0; 82 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 83 &log_in_vain, 0, ""); 84 85 u_long tcp_now; 86 struct inpcbhead tcb; 87 struct inpcbinfo tcbinfo; 88 89 static void tcp_dooptions __P((struct tcpcb *, 90 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 91 static void tcp_pulloutofband __P((struct socket *, 92 struct tcpiphdr *, struct mbuf *)); 93 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 94 static void tcp_xmit_timer __P((struct tcpcb *, int)); 95 96 #endif /* TUBA_INCLUDE */ 97 98 /* 99 * Insert segment ti into reassembly queue of tcp with 100 * control block tp. Return TH_FIN if reassembly now includes 101 * a segment with FIN. The macro form does the common case inline 102 * (segment is the next to be received on an established connection, 103 * and the queue is empty), avoiding linkage into and removal 104 * from the queue and repetition of various conversions. 105 * Set DELACK for segments received in order, but ack immediately 106 * when segments are out of order (so fast retransmit can work). 107 */ 108 #ifdef TCP_ACK_HACK 109 #define TCP_REASS(tp, ti, m, so, flags) { \ 110 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 111 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 112 (tp)->t_state == TCPS_ESTABLISHED) { \ 113 if (ti->ti_flags & TH_PUSH) \ 114 tp->t_flags |= TF_ACKNOW; \ 115 else \ 116 tp->t_flags |= TF_DELACK; \ 117 (tp)->rcv_nxt += (ti)->ti_len; \ 118 flags = (ti)->ti_flags & TH_FIN; \ 119 tcpstat.tcps_rcvpack++;\ 120 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 121 sbappend(&(so)->so_rcv, (m)); \ 122 sorwakeup(so); \ 123 } else { \ 124 (flags) = tcp_reass((tp), (ti), (m)); \ 125 tp->t_flags |= TF_ACKNOW; \ 126 } \ 127 } 128 #else 129 #define TCP_REASS(tp, ti, m, so, flags) { \ 130 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 131 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 132 (tp)->t_state == TCPS_ESTABLISHED) { \ 133 tp->t_flags |= TF_DELACK; \ 134 (tp)->rcv_nxt += (ti)->ti_len; \ 135 flags = (ti)->ti_flags & TH_FIN; \ 136 tcpstat.tcps_rcvpack++;\ 137 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 138 sbappend(&(so)->so_rcv, (m)); \ 139 sorwakeup(so); \ 140 } else { \ 141 (flags) = tcp_reass((tp), (ti), (m)); \ 142 tp->t_flags |= TF_ACKNOW; \ 143 } \ 144 } 145 #endif 146 #ifndef TUBA_INCLUDE 147 148 static int 149 tcp_reass(tp, ti, m) 150 register struct tcpcb *tp; 151 register struct tcpiphdr *ti; 152 struct mbuf *m; 153 { 154 register struct tcpiphdr *q; 155 struct socket *so = tp->t_inpcb->inp_socket; 156 int flags; 157 158 /* 159 * Call with ti==0 after become established to 160 * force pre-ESTABLISHED data up to user socket. 161 */ 162 if (ti == 0) 163 goto present; 164 165 /* 166 * Find a segment which begins after this one does. 167 */ 168 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 169 q = (struct tcpiphdr *)q->ti_next) 170 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 171 break; 172 173 /* 174 * If there is a preceding segment, it may provide some of 175 * our data already. If so, drop the data from the incoming 176 * segment. If it provides all of our data, drop us. 177 */ 178 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 179 register int i; 180 q = (struct tcpiphdr *)q->ti_prev; 181 /* conversion to int (in i) handles seq wraparound */ 182 i = q->ti_seq + q->ti_len - ti->ti_seq; 183 if (i > 0) { 184 if (i >= ti->ti_len) { 185 tcpstat.tcps_rcvduppack++; 186 tcpstat.tcps_rcvdupbyte += ti->ti_len; 187 m_freem(m); 188 /* 189 * Try to present any queued data 190 * at the left window edge to the user. 191 * This is needed after the 3-WHS 192 * completes. 193 */ 194 goto present; /* ??? */ 195 } 196 m_adj(m, i); 197 ti->ti_len -= i; 198 ti->ti_seq += i; 199 } 200 q = (struct tcpiphdr *)(q->ti_next); 201 } 202 tcpstat.tcps_rcvoopack++; 203 tcpstat.tcps_rcvoobyte += ti->ti_len; 204 REASS_MBUF(ti) = m; /* XXX */ 205 206 /* 207 * While we overlap succeeding segments trim them or, 208 * if they are completely covered, dequeue them. 209 */ 210 while (q != (struct tcpiphdr *)tp) { 211 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 212 if (i <= 0) 213 break; 214 if (i < q->ti_len) { 215 q->ti_seq += i; 216 q->ti_len -= i; 217 m_adj(REASS_MBUF(q), i); 218 break; 219 } 220 q = (struct tcpiphdr *)q->ti_next; 221 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 222 remque(q->ti_prev); 223 m_freem(m); 224 } 225 226 /* 227 * Stick new segment in its place. 228 */ 229 insque(ti, q->ti_prev); 230 231 present: 232 /* 233 * Present data to user, advancing rcv_nxt through 234 * completed sequence space. 235 */ 236 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 237 return (0); 238 ti = tp->seg_next; 239 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 240 return (0); 241 do { 242 tp->rcv_nxt += ti->ti_len; 243 flags = ti->ti_flags & TH_FIN; 244 remque(ti); 245 m = REASS_MBUF(ti); 246 ti = (struct tcpiphdr *)ti->ti_next; 247 if (so->so_state & SS_CANTRCVMORE) 248 m_freem(m); 249 else 250 sbappend(&so->so_rcv, m); 251 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 252 sorwakeup(so); 253 return (flags); 254 } 255 256 /* 257 * TCP input routine, follows pages 65-76 of the 258 * protocol specification dated September, 1981 very closely. 259 */ 260 void 261 tcp_input(m, iphlen) 262 register struct mbuf *m; 263 int iphlen; 264 { 265 register struct tcpiphdr *ti; 266 register struct inpcb *inp; 267 u_char *optp = NULL; 268 int optlen = 0; 269 int len, tlen, off; 270 register struct tcpcb *tp = 0; 271 register int tiflags; 272 struct socket *so = 0; 273 int todrop, acked, ourfinisacked, needoutput = 0; 274 struct in_addr laddr; 275 int dropsocket = 0; 276 int iss = 0; 277 u_long tiwin; 278 struct tcpopt to; /* options in this segment */ 279 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 280 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 281 #ifdef TCPDEBUG 282 short ostate = 0; 283 #endif 284 285 bzero((char *)&to, sizeof(to)); 286 287 tcpstat.tcps_rcvtotal++; 288 /* 289 * Get IP and TCP header together in first mbuf. 290 * Note: IP leaves IP header in first mbuf. 291 */ 292 ti = mtod(m, struct tcpiphdr *); 293 if (iphlen > sizeof (struct ip)) 294 ip_stripoptions(m, (struct mbuf *)0); 295 if (m->m_len < sizeof (struct tcpiphdr)) { 296 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 297 tcpstat.tcps_rcvshort++; 298 return; 299 } 300 ti = mtod(m, struct tcpiphdr *); 301 } 302 303 /* 304 * Checksum extended TCP header and data. 305 */ 306 tlen = ((struct ip *)ti)->ip_len; 307 len = sizeof (struct ip) + tlen; 308 ti->ti_next = ti->ti_prev = 0; 309 ti->ti_x1 = 0; 310 ti->ti_len = (u_short)tlen; 311 HTONS(ti->ti_len); 312 ti->ti_sum = in_cksum(m, len); 313 if (ti->ti_sum) { 314 tcpstat.tcps_rcvbadsum++; 315 goto drop; 316 } 317 #endif /* TUBA_INCLUDE */ 318 319 /* 320 * Reject attempted self-connects. XXX This actually masks 321 * a bug elsewhere, since self-connect should work. 322 * However, a urrently-active DoS attack in the Internet 323 * sends a phony self-connect request which causes an infinite 324 * loop. 325 */ 326 if (ti->ti_src.s_addr == ti->ti_dst.s_addr 327 && ti->ti_sport == ti->ti_dport) { 328 tcpstat.tcps_badsyn++; 329 goto drop; 330 } 331 332 /* 333 * Check that TCP offset makes sense, 334 * pull out TCP options and adjust length. XXX 335 */ 336 off = ti->ti_off << 2; 337 if (off < sizeof (struct tcphdr) || off > tlen) { 338 tcpstat.tcps_rcvbadoff++; 339 goto drop; 340 } 341 tlen -= off; 342 ti->ti_len = tlen; 343 if (off > sizeof (struct tcphdr)) { 344 if (m->m_len < sizeof(struct ip) + off) { 345 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 346 tcpstat.tcps_rcvshort++; 347 return; 348 } 349 ti = mtod(m, struct tcpiphdr *); 350 } 351 optlen = off - sizeof (struct tcphdr); 352 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 353 } 354 tiflags = ti->ti_flags; 355 356 /* 357 * Convert TCP protocol specific fields to host format. 358 */ 359 NTOHL(ti->ti_seq); 360 NTOHL(ti->ti_ack); 361 NTOHS(ti->ti_win); 362 NTOHS(ti->ti_urp); 363 364 /* 365 * Drop TCP, IP headers and TCP options. 366 */ 367 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 368 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 369 370 /* 371 * Locate pcb for segment. 372 */ 373 findpcb: 374 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 375 ti->ti_dst, ti->ti_dport, 1); 376 377 /* 378 * If the state is CLOSED (i.e., TCB does not exist) then 379 * all data in the incoming segment is discarded. 380 * If the TCB exists but is in CLOSED state, it is embryonic, 381 * but should either do a listen or a connect soon. 382 */ 383 if (inp == NULL) { 384 if (log_in_vain && tiflags & TH_SYN) { 385 char buf[4*sizeof "123"]; 386 387 strcpy(buf, inet_ntoa(ti->ti_dst)); 388 log(LOG_INFO, 389 "Connection attempt to TCP %s:%d from %s:%d\n", 390 buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src), 391 ntohs(ti->ti_sport)); 392 } 393 goto dropwithreset; 394 } 395 tp = intotcpcb(inp); 396 if (tp == 0) 397 goto dropwithreset; 398 if (tp->t_state == TCPS_CLOSED) 399 goto drop; 400 401 /* Unscale the window into a 32-bit value. */ 402 if ((tiflags & TH_SYN) == 0) 403 tiwin = ti->ti_win << tp->snd_scale; 404 else 405 tiwin = ti->ti_win; 406 407 so = inp->inp_socket; 408 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 409 #ifdef TCPDEBUG 410 if (so->so_options & SO_DEBUG) { 411 ostate = tp->t_state; 412 tcp_saveti = *ti; 413 } 414 #endif 415 if (so->so_options & SO_ACCEPTCONN) { 416 register struct tcpcb *tp0 = tp; 417 struct socket *so2; 418 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 419 /* 420 * Note: dropwithreset makes sure we don't 421 * send a RST in response to a RST. 422 */ 423 if (tiflags & TH_ACK) { 424 tcpstat.tcps_badsyn++; 425 goto dropwithreset; 426 } 427 goto drop; 428 } 429 so2 = sonewconn(so, 0); 430 if (so2 == 0) { 431 tcpstat.tcps_listendrop++; 432 so2 = sodropablereq(so); 433 if (so2) { 434 tcp_drop(sototcpcb(so2), ETIMEDOUT); 435 so2 = sonewconn(so, 0); 436 } 437 if (!so2) 438 goto drop; 439 } 440 so = so2; 441 /* 442 * This is ugly, but .... 443 * 444 * Mark socket as temporary until we're 445 * committed to keeping it. The code at 446 * ``drop'' and ``dropwithreset'' check the 447 * flag dropsocket to see if the temporary 448 * socket created here should be discarded. 449 * We mark the socket as discardable until 450 * we're committed to it below in TCPS_LISTEN. 451 */ 452 dropsocket++; 453 inp = (struct inpcb *)so->so_pcb; 454 inp->inp_laddr = ti->ti_dst; 455 inp->inp_lport = ti->ti_dport; 456 in_pcbrehash(inp); 457 #if BSD>=43 458 inp->inp_options = ip_srcroute(); 459 #endif 460 tp = intotcpcb(inp); 461 tp->t_state = TCPS_LISTEN; 462 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 463 464 /* Compute proper scaling value from buffer space */ 465 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 466 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 467 tp->request_r_scale++; 468 } 469 } 470 471 /* 472 * Segment received on connection. 473 * Reset idle time and keep-alive timer. 474 */ 475 tp->t_idle = 0; 476 if (TCPS_HAVEESTABLISHED(tp->t_state)) 477 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 478 479 /* 480 * Process options if not in LISTEN state, 481 * else do it below (after getting remote address). 482 */ 483 if (tp->t_state != TCPS_LISTEN) 484 tcp_dooptions(tp, optp, optlen, ti, &to); 485 486 /* 487 * Header prediction: check for the two common cases 488 * of a uni-directional data xfer. If the packet has 489 * no control flags, is in-sequence, the window didn't 490 * change and we're not retransmitting, it's a 491 * candidate. If the length is zero and the ack moved 492 * forward, we're the sender side of the xfer. Just 493 * free the data acked & wake any higher level process 494 * that was blocked waiting for space. If the length 495 * is non-zero and the ack didn't move, we're the 496 * receiver side. If we're getting packets in-order 497 * (the reassembly queue is empty), add the data to 498 * the socket buffer and note that we need a delayed ack. 499 * Make sure that the hidden state-flags are also off. 500 * Since we check for TCPS_ESTABLISHED above, it can only 501 * be TH_NEEDSYN. 502 */ 503 if (tp->t_state == TCPS_ESTABLISHED && 504 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 505 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 506 ((to.to_flag & TOF_TS) == 0 || 507 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 508 /* 509 * Using the CC option is compulsory if once started: 510 * the segment is OK if no T/TCP was negotiated or 511 * if the segment has a CC option equal to CCrecv 512 */ 513 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 514 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 515 ti->ti_seq == tp->rcv_nxt && 516 tiwin && tiwin == tp->snd_wnd && 517 tp->snd_nxt == tp->snd_max) { 518 519 /* 520 * If last ACK falls within this segment's sequence numbers, 521 * record the timestamp. 522 * NOTE that the test is modified according to the latest 523 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 524 */ 525 if ((to.to_flag & TOF_TS) != 0 && 526 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 527 tp->ts_recent_age = tcp_now; 528 tp->ts_recent = to.to_tsval; 529 } 530 531 if (ti->ti_len == 0) { 532 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 533 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 534 tp->snd_cwnd >= tp->snd_wnd && 535 tp->t_dupacks < tcprexmtthresh) { 536 /* 537 * this is a pure ack for outstanding data. 538 */ 539 ++tcpstat.tcps_predack; 540 if ((to.to_flag & TOF_TS) != 0) 541 tcp_xmit_timer(tp, 542 tcp_now - to.to_tsecr + 1); 543 else if (tp->t_rtt && 544 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 545 tcp_xmit_timer(tp, tp->t_rtt); 546 acked = ti->ti_ack - tp->snd_una; 547 tcpstat.tcps_rcvackpack++; 548 tcpstat.tcps_rcvackbyte += acked; 549 sbdrop(&so->so_snd, acked); 550 tp->snd_una = ti->ti_ack; 551 m_freem(m); 552 553 /* 554 * If all outstanding data are acked, stop 555 * retransmit timer, otherwise restart timer 556 * using current (possibly backed-off) value. 557 * If process is waiting for space, 558 * wakeup/selwakeup/signal. If data 559 * are ready to send, let tcp_output 560 * decide between more output or persist. 561 */ 562 if (tp->snd_una == tp->snd_max) 563 tp->t_timer[TCPT_REXMT] = 0; 564 else if (tp->t_timer[TCPT_PERSIST] == 0) 565 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 566 567 if (so->so_snd.sb_flags & SB_NOTIFY) 568 sowwakeup(so); 569 if (so->so_snd.sb_cc) 570 (void) tcp_output(tp); 571 return; 572 } 573 } else if (ti->ti_ack == tp->snd_una && 574 tp->seg_next == (struct tcpiphdr *)tp && 575 ti->ti_len <= sbspace(&so->so_rcv)) { 576 /* 577 * this is a pure, in-sequence data packet 578 * with nothing on the reassembly queue and 579 * we have enough buffer space to take it. 580 */ 581 ++tcpstat.tcps_preddat; 582 tp->rcv_nxt += ti->ti_len; 583 tcpstat.tcps_rcvpack++; 584 tcpstat.tcps_rcvbyte += ti->ti_len; 585 /* 586 * Add data to socket buffer. 587 */ 588 sbappend(&so->so_rcv, m); 589 sorwakeup(so); 590 #ifdef TCP_ACK_HACK 591 /* 592 * If this is a short packet, then ACK now - with Nagel 593 * congestion avoidance sender won't send more until 594 * he gets an ACK. 595 */ 596 if (tiflags & TH_PUSH) { 597 tp->t_flags |= TF_ACKNOW; 598 tcp_output(tp); 599 } else { 600 tp->t_flags |= TF_DELACK; 601 } 602 #else 603 tp->t_flags |= TF_DELACK; 604 #endif 605 return; 606 } 607 } 608 609 /* 610 * Calculate amount of space in receive window, 611 * and then do TCP input processing. 612 * Receive window is amount of space in rcv queue, 613 * but not less than advertised window. 614 */ 615 { int win; 616 617 win = sbspace(&so->so_rcv); 618 if (win < 0) 619 win = 0; 620 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 621 } 622 623 switch (tp->t_state) { 624 625 /* 626 * If the state is LISTEN then ignore segment if it contains an RST. 627 * If the segment contains an ACK then it is bad and send a RST. 628 * If it does not contain a SYN then it is not interesting; drop it. 629 * Don't bother responding if the destination was a broadcast. 630 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 631 * tp->iss, and send a segment: 632 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 633 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 634 * Fill in remote peer address fields if not previously specified. 635 * Enter SYN_RECEIVED state, and process any other fields of this 636 * segment in this state. 637 */ 638 case TCPS_LISTEN: { 639 register struct sockaddr_in *sin; 640 641 if (tiflags & TH_RST) 642 goto drop; 643 if (tiflags & TH_ACK) 644 goto dropwithreset; 645 if ((tiflags & TH_SYN) == 0) 646 goto drop; 647 /* 648 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 649 * in_broadcast() should never return true on a received 650 * packet with M_BCAST not set. 651 */ 652 if (m->m_flags & (M_BCAST|M_MCAST) || 653 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 654 goto drop; 655 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 656 M_NOWAIT); 657 if (sin == NULL) 658 goto drop; 659 sin->sin_family = AF_INET; 660 sin->sin_len = sizeof(*sin); 661 sin->sin_addr = ti->ti_src; 662 sin->sin_port = ti->ti_sport; 663 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 664 laddr = inp->inp_laddr; 665 if (inp->inp_laddr.s_addr == INADDR_ANY) 666 inp->inp_laddr = ti->ti_dst; 667 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 668 inp->inp_laddr = laddr; 669 FREE(sin, M_SONAME); 670 goto drop; 671 } 672 FREE(sin, M_SONAME); 673 tp->t_template = tcp_template(tp); 674 if (tp->t_template == 0) { 675 tp = tcp_drop(tp, ENOBUFS); 676 dropsocket = 0; /* socket is already gone */ 677 goto drop; 678 } 679 if ((taop = tcp_gettaocache(inp)) == NULL) { 680 taop = &tao_noncached; 681 bzero(taop, sizeof(*taop)); 682 } 683 tcp_dooptions(tp, optp, optlen, ti, &to); 684 if (iss) 685 tp->iss = iss; 686 else 687 tp->iss = tcp_iss; 688 tcp_iss += TCP_ISSINCR/4; 689 tp->irs = ti->ti_seq; 690 tcp_sendseqinit(tp); 691 tcp_rcvseqinit(tp); 692 /* 693 * Initialization of the tcpcb for transaction; 694 * set SND.WND = SEG.WND, 695 * initialize CCsend and CCrecv. 696 */ 697 tp->snd_wnd = tiwin; /* initial send-window */ 698 tp->cc_send = CC_INC(tcp_ccgen); 699 tp->cc_recv = to.to_cc; 700 /* 701 * Perform TAO test on incoming CC (SEG.CC) option, if any. 702 * - compare SEG.CC against cached CC from the same host, 703 * if any. 704 * - if SEG.CC > chached value, SYN must be new and is accepted 705 * immediately: save new CC in the cache, mark the socket 706 * connected, enter ESTABLISHED state, turn on flag to 707 * send a SYN in the next segment. 708 * A virtual advertised window is set in rcv_adv to 709 * initialize SWS prevention. Then enter normal segment 710 * processing: drop SYN, process data and FIN. 711 * - otherwise do a normal 3-way handshake. 712 */ 713 if ((to.to_flag & TOF_CC) != 0) { 714 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 715 taop->tao_cc = to.to_cc; 716 tp->t_state = TCPS_ESTABLISHED; 717 718 /* 719 * If there is a FIN, or if there is data and the 720 * connection is local, then delay SYN,ACK(SYN) in 721 * the hope of piggy-backing it on a response 722 * segment. Otherwise must send ACK now in case 723 * the other side is slow starting. 724 */ 725 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 726 in_localaddr(inp->inp_faddr))) 727 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 728 else 729 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 730 731 /* 732 * Limit the `virtual advertised window' to TCP_MAXWIN 733 * here. Even if we requested window scaling, it will 734 * become effective only later when our SYN is acked. 735 */ 736 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 737 tcpstat.tcps_connects++; 738 soisconnected(so); 739 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 740 dropsocket = 0; /* committed to socket */ 741 tcpstat.tcps_accepts++; 742 goto trimthenstep6; 743 } 744 /* else do standard 3-way handshake */ 745 } else { 746 /* 747 * No CC option, but maybe CC.NEW: 748 * invalidate cached value. 749 */ 750 taop->tao_cc = 0; 751 } 752 /* 753 * TAO test failed or there was no CC option, 754 * do a standard 3-way handshake. 755 */ 756 tp->t_flags |= TF_ACKNOW; 757 tp->t_state = TCPS_SYN_RECEIVED; 758 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 759 dropsocket = 0; /* committed to socket */ 760 tcpstat.tcps_accepts++; 761 goto trimthenstep6; 762 } 763 764 /* 765 * If the state is SYN_SENT: 766 * if seg contains an ACK, but not for our SYN, drop the input. 767 * if seg contains a RST, then drop the connection. 768 * if seg does not contain SYN, then drop it. 769 * Otherwise this is an acceptable SYN segment 770 * initialize tp->rcv_nxt and tp->irs 771 * if seg contains ack then advance tp->snd_una 772 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 773 * arrange for segment to be acked (eventually) 774 * continue processing rest of data/controls, beginning with URG 775 */ 776 case TCPS_SYN_SENT: 777 if ((taop = tcp_gettaocache(inp)) == NULL) { 778 taop = &tao_noncached; 779 bzero(taop, sizeof(*taop)); 780 } 781 782 if ((tiflags & TH_ACK) && 783 (SEQ_LEQ(ti->ti_ack, tp->iss) || 784 SEQ_GT(ti->ti_ack, tp->snd_max))) { 785 /* 786 * If we have a cached CCsent for the remote host, 787 * hence we haven't just crashed and restarted, 788 * do not send a RST. This may be a retransmission 789 * from the other side after our earlier ACK was lost. 790 * Our new SYN, when it arrives, will serve as the 791 * needed ACK. 792 */ 793 if (taop->tao_ccsent != 0) 794 goto drop; 795 else 796 goto dropwithreset; 797 } 798 if (tiflags & TH_RST) { 799 if (tiflags & TH_ACK) 800 tp = tcp_drop(tp, ECONNREFUSED); 801 goto drop; 802 } 803 if ((tiflags & TH_SYN) == 0) 804 goto drop; 805 tp->snd_wnd = ti->ti_win; /* initial send window */ 806 tp->cc_recv = to.to_cc; /* foreign CC */ 807 808 tp->irs = ti->ti_seq; 809 tcp_rcvseqinit(tp); 810 if (tiflags & TH_ACK) { 811 /* 812 * Our SYN was acked. If segment contains CC.ECHO 813 * option, check it to make sure this segment really 814 * matches our SYN. If not, just drop it as old 815 * duplicate, but send an RST if we're still playing 816 * by the old rules. If no CC.ECHO option, make sure 817 * we don't get fooled into using T/TCP. 818 */ 819 if (to.to_flag & TOF_CCECHO) { 820 if (tp->cc_send != to.to_ccecho) 821 if (taop->tao_ccsent != 0) 822 goto drop; 823 else 824 goto dropwithreset; 825 } else 826 tp->t_flags &= ~TF_RCVD_CC; 827 tcpstat.tcps_connects++; 828 soisconnected(so); 829 /* Do window scaling on this connection? */ 830 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 831 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 832 tp->snd_scale = tp->requested_s_scale; 833 tp->rcv_scale = tp->request_r_scale; 834 } 835 /* Segment is acceptable, update cache if undefined. */ 836 if (taop->tao_ccsent == 0) 837 taop->tao_ccsent = to.to_ccecho; 838 839 tp->rcv_adv += tp->rcv_wnd; 840 tp->snd_una++; /* SYN is acked */ 841 /* 842 * If there's data, delay ACK; if there's also a FIN 843 * ACKNOW will be turned on later. 844 */ 845 if (ti->ti_len != 0) 846 tp->t_flags |= TF_DELACK; 847 else 848 tp->t_flags |= TF_ACKNOW; 849 /* 850 * Received <SYN,ACK> in SYN_SENT[*] state. 851 * Transitions: 852 * SYN_SENT --> ESTABLISHED 853 * SYN_SENT* --> FIN_WAIT_1 854 */ 855 if (tp->t_flags & TF_NEEDFIN) { 856 tp->t_state = TCPS_FIN_WAIT_1; 857 tp->t_flags &= ~TF_NEEDFIN; 858 tiflags &= ~TH_SYN; 859 } else { 860 tp->t_state = TCPS_ESTABLISHED; 861 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 862 } 863 } else { 864 /* 865 * Received initial SYN in SYN-SENT[*] state => simul- 866 * taneous open. If segment contains CC option and there is 867 * a cached CC, apply TAO test; if it succeeds, connection is 868 * half-synchronized. Otherwise, do 3-way handshake: 869 * SYN-SENT -> SYN-RECEIVED 870 * SYN-SENT* -> SYN-RECEIVED* 871 * If there was no CC option, clear cached CC value. 872 */ 873 tp->t_flags |= TF_ACKNOW; 874 tp->t_timer[TCPT_REXMT] = 0; 875 if (to.to_flag & TOF_CC) { 876 if (taop->tao_cc != 0 && 877 CC_GT(to.to_cc, taop->tao_cc)) { 878 /* 879 * update cache and make transition: 880 * SYN-SENT -> ESTABLISHED* 881 * SYN-SENT* -> FIN-WAIT-1* 882 */ 883 taop->tao_cc = to.to_cc; 884 if (tp->t_flags & TF_NEEDFIN) { 885 tp->t_state = TCPS_FIN_WAIT_1; 886 tp->t_flags &= ~TF_NEEDFIN; 887 } else { 888 tp->t_state = TCPS_ESTABLISHED; 889 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 890 } 891 tp->t_flags |= TF_NEEDSYN; 892 } else 893 tp->t_state = TCPS_SYN_RECEIVED; 894 } else { 895 /* CC.NEW or no option => invalidate cache */ 896 taop->tao_cc = 0; 897 tp->t_state = TCPS_SYN_RECEIVED; 898 } 899 } 900 901 trimthenstep6: 902 /* 903 * Advance ti->ti_seq to correspond to first data byte. 904 * If data, trim to stay within window, 905 * dropping FIN if necessary. 906 */ 907 ti->ti_seq++; 908 if (ti->ti_len > tp->rcv_wnd) { 909 todrop = ti->ti_len - tp->rcv_wnd; 910 m_adj(m, -todrop); 911 ti->ti_len = tp->rcv_wnd; 912 tiflags &= ~TH_FIN; 913 tcpstat.tcps_rcvpackafterwin++; 914 tcpstat.tcps_rcvbyteafterwin += todrop; 915 } 916 tp->snd_wl1 = ti->ti_seq - 1; 917 tp->rcv_up = ti->ti_seq; 918 /* 919 * Client side of transaction: already sent SYN and data. 920 * If the remote host used T/TCP to validate the SYN, 921 * our data will be ACK'd; if so, enter normal data segment 922 * processing in the middle of step 5, ack processing. 923 * Otherwise, goto step 6. 924 */ 925 if (tiflags & TH_ACK) 926 goto process_ACK; 927 goto step6; 928 /* 929 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 930 * if segment contains a SYN and CC [not CC.NEW] option: 931 * if state == TIME_WAIT and connection duration > MSL, 932 * drop packet and send RST; 933 * 934 * if SEG.CC > CCrecv then is new SYN, and can implicitly 935 * ack the FIN (and data) in retransmission queue. 936 * Complete close and delete TCPCB. Then reprocess 937 * segment, hoping to find new TCPCB in LISTEN state; 938 * 939 * else must be old SYN; drop it. 940 * else do normal processing. 941 */ 942 case TCPS_LAST_ACK: 943 case TCPS_CLOSING: 944 case TCPS_TIME_WAIT: 945 if ((tiflags & TH_SYN) && 946 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 947 if (tp->t_state == TCPS_TIME_WAIT && 948 tp->t_duration > TCPTV_MSL) 949 goto dropwithreset; 950 if (CC_GT(to.to_cc, tp->cc_recv)) { 951 tp = tcp_close(tp); 952 goto findpcb; 953 } 954 else 955 goto drop; 956 } 957 break; /* continue normal processing */ 958 } 959 960 /* 961 * States other than LISTEN or SYN_SENT. 962 * First check timestamp, if present. 963 * Then check the connection count, if present. 964 * Then check that at least some bytes of segment are within 965 * receive window. If segment begins before rcv_nxt, 966 * drop leading data (and SYN); if nothing left, just ack. 967 * 968 * RFC 1323 PAWS: If we have a timestamp reply on this segment 969 * and it's less than ts_recent, drop it. 970 */ 971 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 972 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 973 974 /* Check to see if ts_recent is over 24 days old. */ 975 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 976 /* 977 * Invalidate ts_recent. If this segment updates 978 * ts_recent, the age will be reset later and ts_recent 979 * will get a valid value. If it does not, setting 980 * ts_recent to zero will at least satisfy the 981 * requirement that zero be placed in the timestamp 982 * echo reply when ts_recent isn't valid. The 983 * age isn't reset until we get a valid ts_recent 984 * because we don't want out-of-order segments to be 985 * dropped when ts_recent is old. 986 */ 987 tp->ts_recent = 0; 988 } else { 989 tcpstat.tcps_rcvduppack++; 990 tcpstat.tcps_rcvdupbyte += ti->ti_len; 991 tcpstat.tcps_pawsdrop++; 992 goto dropafterack; 993 } 994 } 995 996 /* 997 * T/TCP mechanism 998 * If T/TCP was negotiated and the segment doesn't have CC, 999 * or if it's CC is wrong then drop the segment. 1000 * RST segments do not have to comply with this. 1001 */ 1002 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1003 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 1004 (tiflags & TH_RST) == 0) 1005 goto dropafterack; 1006 1007 todrop = tp->rcv_nxt - ti->ti_seq; 1008 if (todrop > 0) { 1009 if (tiflags & TH_SYN) { 1010 tiflags &= ~TH_SYN; 1011 ti->ti_seq++; 1012 if (ti->ti_urp > 1) 1013 ti->ti_urp--; 1014 else 1015 tiflags &= ~TH_URG; 1016 todrop--; 1017 } 1018 /* 1019 * Following if statement from Stevens, vol. 2, p. 960. 1020 */ 1021 if (todrop > ti->ti_len 1022 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1023 /* 1024 * Any valid FIN must be to the left of the window. 1025 * At this point the FIN must be a duplicate or out 1026 * of sequence; drop it. 1027 */ 1028 tiflags &= ~TH_FIN; 1029 1030 /* 1031 * Send an ACK to resynchronize and drop any data. 1032 * But keep on processing for RST or ACK. 1033 */ 1034 tp->t_flags |= TF_ACKNOW; 1035 todrop = ti->ti_len; 1036 tcpstat.tcps_rcvduppack++; 1037 tcpstat.tcps_rcvdupbyte += todrop; 1038 } else { 1039 tcpstat.tcps_rcvpartduppack++; 1040 tcpstat.tcps_rcvpartdupbyte += todrop; 1041 } 1042 m_adj(m, todrop); 1043 ti->ti_seq += todrop; 1044 ti->ti_len -= todrop; 1045 if (ti->ti_urp > todrop) 1046 ti->ti_urp -= todrop; 1047 else { 1048 tiflags &= ~TH_URG; 1049 ti->ti_urp = 0; 1050 } 1051 } 1052 1053 /* 1054 * If new data are received on a connection after the 1055 * user processes are gone, then RST the other end. 1056 */ 1057 if ((so->so_state & SS_NOFDREF) && 1058 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1059 tp = tcp_close(tp); 1060 tcpstat.tcps_rcvafterclose++; 1061 goto dropwithreset; 1062 } 1063 1064 /* 1065 * If segment ends after window, drop trailing data 1066 * (and PUSH and FIN); if nothing left, just ACK. 1067 */ 1068 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1069 if (todrop > 0) { 1070 tcpstat.tcps_rcvpackafterwin++; 1071 if (todrop >= ti->ti_len) { 1072 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1073 /* 1074 * If a new connection request is received 1075 * while in TIME_WAIT, drop the old connection 1076 * and start over if the sequence numbers 1077 * are above the previous ones. 1078 */ 1079 if (tiflags & TH_SYN && 1080 tp->t_state == TCPS_TIME_WAIT && 1081 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1082 iss = tp->rcv_nxt + TCP_ISSINCR; 1083 tp = tcp_close(tp); 1084 goto findpcb; 1085 } 1086 /* 1087 * If window is closed can only take segments at 1088 * window edge, and have to drop data and PUSH from 1089 * incoming segments. Continue processing, but 1090 * remember to ack. Otherwise, drop segment 1091 * and ack. 1092 */ 1093 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1094 tp->t_flags |= TF_ACKNOW; 1095 tcpstat.tcps_rcvwinprobe++; 1096 } else 1097 goto dropafterack; 1098 } else 1099 tcpstat.tcps_rcvbyteafterwin += todrop; 1100 m_adj(m, -todrop); 1101 ti->ti_len -= todrop; 1102 tiflags &= ~(TH_PUSH|TH_FIN); 1103 } 1104 1105 /* 1106 * If last ACK falls within this segment's sequence numbers, 1107 * record its timestamp. 1108 * NOTE that the test is modified according to the latest 1109 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1110 */ 1111 if ((to.to_flag & TOF_TS) != 0 && 1112 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1113 tp->ts_recent_age = tcp_now; 1114 tp->ts_recent = to.to_tsval; 1115 } 1116 1117 /* 1118 * If the RST bit is set examine the state: 1119 * SYN_RECEIVED STATE: 1120 * If passive open, return to LISTEN state. 1121 * If active open, inform user that connection was refused. 1122 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1123 * Inform user that connection was reset, and close tcb. 1124 * CLOSING, LAST_ACK, TIME_WAIT STATES 1125 * Close the tcb. 1126 */ 1127 if (tiflags&TH_RST) switch (tp->t_state) { 1128 1129 case TCPS_SYN_RECEIVED: 1130 so->so_error = ECONNREFUSED; 1131 goto close; 1132 1133 case TCPS_ESTABLISHED: 1134 case TCPS_FIN_WAIT_1: 1135 case TCPS_FIN_WAIT_2: 1136 case TCPS_CLOSE_WAIT: 1137 so->so_error = ECONNRESET; 1138 close: 1139 tp->t_state = TCPS_CLOSED; 1140 tcpstat.tcps_drops++; 1141 tp = tcp_close(tp); 1142 goto drop; 1143 1144 case TCPS_CLOSING: 1145 case TCPS_LAST_ACK: 1146 case TCPS_TIME_WAIT: 1147 tp = tcp_close(tp); 1148 goto drop; 1149 } 1150 1151 /* 1152 * If a SYN is in the window, then this is an 1153 * error and we send an RST and drop the connection. 1154 */ 1155 if (tiflags & TH_SYN) { 1156 tp = tcp_drop(tp, ECONNRESET); 1157 goto dropwithreset; 1158 } 1159 1160 /* 1161 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1162 * flag is on (half-synchronized state), then queue data for 1163 * later processing; else drop segment and return. 1164 */ 1165 if ((tiflags & TH_ACK) == 0) { 1166 if (tp->t_state == TCPS_SYN_RECEIVED || 1167 (tp->t_flags & TF_NEEDSYN)) 1168 goto step6; 1169 else 1170 goto drop; 1171 } 1172 1173 /* 1174 * Ack processing. 1175 */ 1176 switch (tp->t_state) { 1177 1178 /* 1179 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1180 * ESTABLISHED state and continue processing, otherwise 1181 * send an RST. 1182 */ 1183 case TCPS_SYN_RECEIVED: 1184 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1185 SEQ_GT(ti->ti_ack, tp->snd_max)) 1186 goto dropwithreset; 1187 1188 tcpstat.tcps_connects++; 1189 soisconnected(so); 1190 /* Do window scaling? */ 1191 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1192 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1193 tp->snd_scale = tp->requested_s_scale; 1194 tp->rcv_scale = tp->request_r_scale; 1195 } 1196 /* 1197 * Upon successful completion of 3-way handshake, 1198 * update cache.CC if it was undefined, pass any queued 1199 * data to the user, and advance state appropriately. 1200 */ 1201 if ((taop = tcp_gettaocache(inp)) != NULL && 1202 taop->tao_cc == 0) 1203 taop->tao_cc = tp->cc_recv; 1204 1205 /* 1206 * Make transitions: 1207 * SYN-RECEIVED -> ESTABLISHED 1208 * SYN-RECEIVED* -> FIN-WAIT-1 1209 */ 1210 if (tp->t_flags & TF_NEEDFIN) { 1211 tp->t_state = TCPS_FIN_WAIT_1; 1212 tp->t_flags &= ~TF_NEEDFIN; 1213 } else { 1214 tp->t_state = TCPS_ESTABLISHED; 1215 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1216 } 1217 /* 1218 * If segment contains data or ACK, will call tcp_reass() 1219 * later; if not, do so now to pass queued data to user. 1220 */ 1221 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1222 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1223 (struct mbuf *)0); 1224 tp->snd_wl1 = ti->ti_seq - 1; 1225 /* fall into ... */ 1226 1227 /* 1228 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1229 * ACKs. If the ack is in the range 1230 * tp->snd_una < ti->ti_ack <= tp->snd_max 1231 * then advance tp->snd_una to ti->ti_ack and drop 1232 * data from the retransmission queue. If this ACK reflects 1233 * more up to date window information we update our window information. 1234 */ 1235 case TCPS_ESTABLISHED: 1236 case TCPS_FIN_WAIT_1: 1237 case TCPS_FIN_WAIT_2: 1238 case TCPS_CLOSE_WAIT: 1239 case TCPS_CLOSING: 1240 case TCPS_LAST_ACK: 1241 case TCPS_TIME_WAIT: 1242 1243 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1244 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1245 tcpstat.tcps_rcvdupack++; 1246 /* 1247 * If we have outstanding data (other than 1248 * a window probe), this is a completely 1249 * duplicate ack (ie, window info didn't 1250 * change), the ack is the biggest we've 1251 * seen and we've seen exactly our rexmt 1252 * threshhold of them, assume a packet 1253 * has been dropped and retransmit it. 1254 * Kludge snd_nxt & the congestion 1255 * window so we send only this one 1256 * packet. 1257 * 1258 * We know we're losing at the current 1259 * window size so do congestion avoidance 1260 * (set ssthresh to half the current window 1261 * and pull our congestion window back to 1262 * the new ssthresh). 1263 * 1264 * Dup acks mean that packets have left the 1265 * network (they're now cached at the receiver) 1266 * so bump cwnd by the amount in the receiver 1267 * to keep a constant cwnd packets in the 1268 * network. 1269 */ 1270 if (tp->t_timer[TCPT_REXMT] == 0 || 1271 ti->ti_ack != tp->snd_una) 1272 tp->t_dupacks = 0; 1273 else if (++tp->t_dupacks == tcprexmtthresh) { 1274 tcp_seq onxt = tp->snd_nxt; 1275 u_int win = 1276 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1277 tp->t_maxseg; 1278 1279 if (win < 2) 1280 win = 2; 1281 tp->snd_ssthresh = win * tp->t_maxseg; 1282 tp->t_timer[TCPT_REXMT] = 0; 1283 tp->t_rtt = 0; 1284 tp->snd_nxt = ti->ti_ack; 1285 tp->snd_cwnd = tp->t_maxseg; 1286 (void) tcp_output(tp); 1287 tp->snd_cwnd = tp->snd_ssthresh + 1288 tp->t_maxseg * tp->t_dupacks; 1289 if (SEQ_GT(onxt, tp->snd_nxt)) 1290 tp->snd_nxt = onxt; 1291 goto drop; 1292 } else if (tp->t_dupacks > tcprexmtthresh) { 1293 tp->snd_cwnd += tp->t_maxseg; 1294 (void) tcp_output(tp); 1295 goto drop; 1296 } 1297 } else 1298 tp->t_dupacks = 0; 1299 break; 1300 } 1301 /* 1302 * If the congestion window was inflated to account 1303 * for the other side's cached packets, retract it. 1304 */ 1305 if (tp->t_dupacks >= tcprexmtthresh && 1306 tp->snd_cwnd > tp->snd_ssthresh) 1307 tp->snd_cwnd = tp->snd_ssthresh; 1308 tp->t_dupacks = 0; 1309 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1310 tcpstat.tcps_rcvacktoomuch++; 1311 goto dropafterack; 1312 } 1313 /* 1314 * If we reach this point, ACK is not a duplicate, 1315 * i.e., it ACKs something we sent. 1316 */ 1317 if (tp->t_flags & TF_NEEDSYN) { 1318 /* 1319 * T/TCP: Connection was half-synchronized, and our 1320 * SYN has been ACK'd (so connection is now fully 1321 * synchronized). Go to non-starred state, 1322 * increment snd_una for ACK of SYN, and check if 1323 * we can do window scaling. 1324 */ 1325 tp->t_flags &= ~TF_NEEDSYN; 1326 tp->snd_una++; 1327 /* Do window scaling? */ 1328 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1329 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1330 tp->snd_scale = tp->requested_s_scale; 1331 tp->rcv_scale = tp->request_r_scale; 1332 } 1333 } 1334 1335 process_ACK: 1336 acked = ti->ti_ack - tp->snd_una; 1337 tcpstat.tcps_rcvackpack++; 1338 tcpstat.tcps_rcvackbyte += acked; 1339 1340 /* 1341 * If we have a timestamp reply, update smoothed 1342 * round trip time. If no timestamp is present but 1343 * transmit timer is running and timed sequence 1344 * number was acked, update smoothed round trip time. 1345 * Since we now have an rtt measurement, cancel the 1346 * timer backoff (cf., Phil Karn's retransmit alg.). 1347 * Recompute the initial retransmit timer. 1348 */ 1349 if (to.to_flag & TOF_TS) 1350 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1351 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1352 tcp_xmit_timer(tp,tp->t_rtt); 1353 1354 /* 1355 * If all outstanding data is acked, stop retransmit 1356 * timer and remember to restart (more output or persist). 1357 * If there is more data to be acked, restart retransmit 1358 * timer, using current (possibly backed-off) value. 1359 */ 1360 if (ti->ti_ack == tp->snd_max) { 1361 tp->t_timer[TCPT_REXMT] = 0; 1362 needoutput = 1; 1363 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1364 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1365 1366 /* 1367 * If no data (only SYN) was ACK'd, 1368 * skip rest of ACK processing. 1369 */ 1370 if (acked == 0) 1371 goto step6; 1372 1373 /* 1374 * When new data is acked, open the congestion window. 1375 * If the window gives us less than ssthresh packets 1376 * in flight, open exponentially (maxseg per packet). 1377 * Otherwise open linearly: maxseg per window 1378 * (maxseg^2 / cwnd per packet). 1379 */ 1380 { 1381 register u_int cw = tp->snd_cwnd; 1382 register u_int incr = tp->t_maxseg; 1383 1384 if (cw > tp->snd_ssthresh) 1385 incr = incr * incr / cw; 1386 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1387 } 1388 if (acked > so->so_snd.sb_cc) { 1389 tp->snd_wnd -= so->so_snd.sb_cc; 1390 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1391 ourfinisacked = 1; 1392 } else { 1393 sbdrop(&so->so_snd, acked); 1394 tp->snd_wnd -= acked; 1395 ourfinisacked = 0; 1396 } 1397 if (so->so_snd.sb_flags & SB_NOTIFY) 1398 sowwakeup(so); 1399 tp->snd_una = ti->ti_ack; 1400 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1401 tp->snd_nxt = tp->snd_una; 1402 1403 switch (tp->t_state) { 1404 1405 /* 1406 * In FIN_WAIT_1 STATE in addition to the processing 1407 * for the ESTABLISHED state if our FIN is now acknowledged 1408 * then enter FIN_WAIT_2. 1409 */ 1410 case TCPS_FIN_WAIT_1: 1411 if (ourfinisacked) { 1412 /* 1413 * If we can't receive any more 1414 * data, then closing user can proceed. 1415 * Starting the timer is contrary to the 1416 * specification, but if we don't get a FIN 1417 * we'll hang forever. 1418 */ 1419 if (so->so_state & SS_CANTRCVMORE) { 1420 soisdisconnected(so); 1421 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1422 } 1423 tp->t_state = TCPS_FIN_WAIT_2; 1424 } 1425 break; 1426 1427 /* 1428 * In CLOSING STATE in addition to the processing for 1429 * the ESTABLISHED state if the ACK acknowledges our FIN 1430 * then enter the TIME-WAIT state, otherwise ignore 1431 * the segment. 1432 */ 1433 case TCPS_CLOSING: 1434 if (ourfinisacked) { 1435 tp->t_state = TCPS_TIME_WAIT; 1436 tcp_canceltimers(tp); 1437 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1438 if (tp->cc_recv != 0 && 1439 tp->t_duration < TCPTV_MSL) 1440 tp->t_timer[TCPT_2MSL] = 1441 tp->t_rxtcur * TCPTV_TWTRUNC; 1442 else 1443 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1444 soisdisconnected(so); 1445 } 1446 break; 1447 1448 /* 1449 * In LAST_ACK, we may still be waiting for data to drain 1450 * and/or to be acked, as well as for the ack of our FIN. 1451 * If our FIN is now acknowledged, delete the TCB, 1452 * enter the closed state and return. 1453 */ 1454 case TCPS_LAST_ACK: 1455 if (ourfinisacked) { 1456 tp = tcp_close(tp); 1457 goto drop; 1458 } 1459 break; 1460 1461 /* 1462 * In TIME_WAIT state the only thing that should arrive 1463 * is a retransmission of the remote FIN. Acknowledge 1464 * it and restart the finack timer. 1465 */ 1466 case TCPS_TIME_WAIT: 1467 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1468 goto dropafterack; 1469 } 1470 } 1471 1472 step6: 1473 /* 1474 * Update window information. 1475 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1476 */ 1477 if ((tiflags & TH_ACK) && 1478 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1479 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1480 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1481 /* keep track of pure window updates */ 1482 if (ti->ti_len == 0 && 1483 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1484 tcpstat.tcps_rcvwinupd++; 1485 tp->snd_wnd = tiwin; 1486 tp->snd_wl1 = ti->ti_seq; 1487 tp->snd_wl2 = ti->ti_ack; 1488 if (tp->snd_wnd > tp->max_sndwnd) 1489 tp->max_sndwnd = tp->snd_wnd; 1490 needoutput = 1; 1491 } 1492 1493 /* 1494 * Process segments with URG. 1495 */ 1496 if ((tiflags & TH_URG) && ti->ti_urp && 1497 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1498 /* 1499 * This is a kludge, but if we receive and accept 1500 * random urgent pointers, we'll crash in 1501 * soreceive. It's hard to imagine someone 1502 * actually wanting to send this much urgent data. 1503 */ 1504 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1505 ti->ti_urp = 0; /* XXX */ 1506 tiflags &= ~TH_URG; /* XXX */ 1507 goto dodata; /* XXX */ 1508 } 1509 /* 1510 * If this segment advances the known urgent pointer, 1511 * then mark the data stream. This should not happen 1512 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1513 * a FIN has been received from the remote side. 1514 * In these states we ignore the URG. 1515 * 1516 * According to RFC961 (Assigned Protocols), 1517 * the urgent pointer points to the last octet 1518 * of urgent data. We continue, however, 1519 * to consider it to indicate the first octet 1520 * of data past the urgent section as the original 1521 * spec states (in one of two places). 1522 */ 1523 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1524 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1525 so->so_oobmark = so->so_rcv.sb_cc + 1526 (tp->rcv_up - tp->rcv_nxt) - 1; 1527 if (so->so_oobmark == 0) 1528 so->so_state |= SS_RCVATMARK; 1529 sohasoutofband(so); 1530 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1531 } 1532 /* 1533 * Remove out of band data so doesn't get presented to user. 1534 * This can happen independent of advancing the URG pointer, 1535 * but if two URG's are pending at once, some out-of-band 1536 * data may creep in... ick. 1537 */ 1538 if (ti->ti_urp <= (u_long)ti->ti_len 1539 #ifdef SO_OOBINLINE 1540 && (so->so_options & SO_OOBINLINE) == 0 1541 #endif 1542 ) 1543 tcp_pulloutofband(so, ti, m); 1544 } else 1545 /* 1546 * If no out of band data is expected, 1547 * pull receive urgent pointer along 1548 * with the receive window. 1549 */ 1550 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1551 tp->rcv_up = tp->rcv_nxt; 1552 dodata: /* XXX */ 1553 1554 /* 1555 * Process the segment text, merging it into the TCP sequencing queue, 1556 * and arranging for acknowledgment of receipt if necessary. 1557 * This process logically involves adjusting tp->rcv_wnd as data 1558 * is presented to the user (this happens in tcp_usrreq.c, 1559 * case PRU_RCVD). If a FIN has already been received on this 1560 * connection then we just ignore the text. 1561 */ 1562 if ((ti->ti_len || (tiflags&TH_FIN)) && 1563 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1564 TCP_REASS(tp, ti, m, so, tiflags); 1565 /* 1566 * Note the amount of data that peer has sent into 1567 * our window, in order to estimate the sender's 1568 * buffer size. 1569 */ 1570 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1571 } else { 1572 m_freem(m); 1573 tiflags &= ~TH_FIN; 1574 } 1575 1576 /* 1577 * If FIN is received ACK the FIN and let the user know 1578 * that the connection is closing. 1579 */ 1580 if (tiflags & TH_FIN) { 1581 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1582 socantrcvmore(so); 1583 /* 1584 * If connection is half-synchronized 1585 * (ie NEEDSYN flag on) then delay ACK, 1586 * so it may be piggybacked when SYN is sent. 1587 * Otherwise, since we received a FIN then no 1588 * more input can be expected, send ACK now. 1589 */ 1590 if (tp->t_flags & TF_NEEDSYN) 1591 tp->t_flags |= TF_DELACK; 1592 else 1593 tp->t_flags |= TF_ACKNOW; 1594 tp->rcv_nxt++; 1595 } 1596 switch (tp->t_state) { 1597 1598 /* 1599 * In SYN_RECEIVED and ESTABLISHED STATES 1600 * enter the CLOSE_WAIT state. 1601 */ 1602 case TCPS_SYN_RECEIVED: 1603 case TCPS_ESTABLISHED: 1604 tp->t_state = TCPS_CLOSE_WAIT; 1605 break; 1606 1607 /* 1608 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1609 * enter the CLOSING state. 1610 */ 1611 case TCPS_FIN_WAIT_1: 1612 tp->t_state = TCPS_CLOSING; 1613 break; 1614 1615 /* 1616 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1617 * starting the time-wait timer, turning off the other 1618 * standard timers. 1619 */ 1620 case TCPS_FIN_WAIT_2: 1621 tp->t_state = TCPS_TIME_WAIT; 1622 tcp_canceltimers(tp); 1623 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1624 if (tp->cc_recv != 0 && 1625 tp->t_duration < TCPTV_MSL) { 1626 tp->t_timer[TCPT_2MSL] = 1627 tp->t_rxtcur * TCPTV_TWTRUNC; 1628 /* For transaction client, force ACK now. */ 1629 tp->t_flags |= TF_ACKNOW; 1630 } 1631 else 1632 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1633 soisdisconnected(so); 1634 break; 1635 1636 /* 1637 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1638 */ 1639 case TCPS_TIME_WAIT: 1640 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1641 break; 1642 } 1643 } 1644 #ifdef TCPDEBUG 1645 if (so->so_options & SO_DEBUG) 1646 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1647 #endif 1648 1649 /* 1650 * Return any desired output. 1651 */ 1652 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1653 (void) tcp_output(tp); 1654 return; 1655 1656 dropafterack: 1657 /* 1658 * Generate an ACK dropping incoming segment if it occupies 1659 * sequence space, where the ACK reflects our state. 1660 */ 1661 if (tiflags & TH_RST) 1662 goto drop; 1663 #ifdef TCPDEBUG 1664 if (so->so_options & SO_DEBUG) 1665 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1666 #endif 1667 m_freem(m); 1668 tp->t_flags |= TF_ACKNOW; 1669 (void) tcp_output(tp); 1670 return; 1671 1672 dropwithreset: 1673 /* 1674 * Generate a RST, dropping incoming segment. 1675 * Make ACK acceptable to originator of segment. 1676 * Don't bother to respond if destination was broadcast/multicast. 1677 */ 1678 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1679 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1680 goto drop; 1681 #ifdef TCPDEBUG 1682 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1683 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1684 #endif 1685 if (tiflags & TH_ACK) 1686 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1687 else { 1688 if (tiflags & TH_SYN) 1689 ti->ti_len++; 1690 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1691 TH_RST|TH_ACK); 1692 } 1693 /* destroy temporarily created socket */ 1694 if (dropsocket) 1695 (void) soabort(so); 1696 return; 1697 1698 drop: 1699 /* 1700 * Drop space held by incoming segment and return. 1701 */ 1702 #ifdef TCPDEBUG 1703 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1704 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1705 #endif 1706 m_freem(m); 1707 /* destroy temporarily created socket */ 1708 if (dropsocket) 1709 (void) soabort(so); 1710 return; 1711 #ifndef TUBA_INCLUDE 1712 } 1713 1714 static void 1715 tcp_dooptions(tp, cp, cnt, ti, to) 1716 struct tcpcb *tp; 1717 u_char *cp; 1718 int cnt; 1719 struct tcpiphdr *ti; 1720 struct tcpopt *to; 1721 { 1722 u_short mss = 0; 1723 int opt, optlen; 1724 1725 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1726 opt = cp[0]; 1727 if (opt == TCPOPT_EOL) 1728 break; 1729 if (opt == TCPOPT_NOP) 1730 optlen = 1; 1731 else { 1732 optlen = cp[1]; 1733 if (optlen <= 0) 1734 break; 1735 } 1736 switch (opt) { 1737 1738 default: 1739 continue; 1740 1741 case TCPOPT_MAXSEG: 1742 if (optlen != TCPOLEN_MAXSEG) 1743 continue; 1744 if (!(ti->ti_flags & TH_SYN)) 1745 continue; 1746 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1747 NTOHS(mss); 1748 break; 1749 1750 case TCPOPT_WINDOW: 1751 if (optlen != TCPOLEN_WINDOW) 1752 continue; 1753 if (!(ti->ti_flags & TH_SYN)) 1754 continue; 1755 tp->t_flags |= TF_RCVD_SCALE; 1756 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1757 break; 1758 1759 case TCPOPT_TIMESTAMP: 1760 if (optlen != TCPOLEN_TIMESTAMP) 1761 continue; 1762 to->to_flag |= TOF_TS; 1763 bcopy((char *)cp + 2, 1764 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1765 NTOHL(to->to_tsval); 1766 bcopy((char *)cp + 6, 1767 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1768 NTOHL(to->to_tsecr); 1769 1770 /* 1771 * A timestamp received in a SYN makes 1772 * it ok to send timestamp requests and replies. 1773 */ 1774 if (ti->ti_flags & TH_SYN) { 1775 tp->t_flags |= TF_RCVD_TSTMP; 1776 tp->ts_recent = to->to_tsval; 1777 tp->ts_recent_age = tcp_now; 1778 } 1779 break; 1780 case TCPOPT_CC: 1781 if (optlen != TCPOLEN_CC) 1782 continue; 1783 to->to_flag |= TOF_CC; 1784 bcopy((char *)cp + 2, 1785 (char *)&to->to_cc, sizeof(to->to_cc)); 1786 NTOHL(to->to_cc); 1787 /* 1788 * A CC or CC.new option received in a SYN makes 1789 * it ok to send CC in subsequent segments. 1790 */ 1791 if (ti->ti_flags & TH_SYN) 1792 tp->t_flags |= TF_RCVD_CC; 1793 break; 1794 case TCPOPT_CCNEW: 1795 if (optlen != TCPOLEN_CC) 1796 continue; 1797 if (!(ti->ti_flags & TH_SYN)) 1798 continue; 1799 to->to_flag |= TOF_CCNEW; 1800 bcopy((char *)cp + 2, 1801 (char *)&to->to_cc, sizeof(to->to_cc)); 1802 NTOHL(to->to_cc); 1803 /* 1804 * A CC or CC.new option received in a SYN makes 1805 * it ok to send CC in subsequent segments. 1806 */ 1807 tp->t_flags |= TF_RCVD_CC; 1808 break; 1809 case TCPOPT_CCECHO: 1810 if (optlen != TCPOLEN_CC) 1811 continue; 1812 if (!(ti->ti_flags & TH_SYN)) 1813 continue; 1814 to->to_flag |= TOF_CCECHO; 1815 bcopy((char *)cp + 2, 1816 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1817 NTOHL(to->to_ccecho); 1818 break; 1819 } 1820 } 1821 if (ti->ti_flags & TH_SYN) 1822 tcp_mss(tp, mss); /* sets t_maxseg */ 1823 } 1824 1825 /* 1826 * Pull out of band byte out of a segment so 1827 * it doesn't appear in the user's data queue. 1828 * It is still reflected in the segment length for 1829 * sequencing purposes. 1830 */ 1831 static void 1832 tcp_pulloutofband(so, ti, m) 1833 struct socket *so; 1834 struct tcpiphdr *ti; 1835 register struct mbuf *m; 1836 { 1837 int cnt = ti->ti_urp - 1; 1838 1839 while (cnt >= 0) { 1840 if (m->m_len > cnt) { 1841 char *cp = mtod(m, caddr_t) + cnt; 1842 struct tcpcb *tp = sototcpcb(so); 1843 1844 tp->t_iobc = *cp; 1845 tp->t_oobflags |= TCPOOB_HAVEDATA; 1846 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1847 m->m_len--; 1848 return; 1849 } 1850 cnt -= m->m_len; 1851 m = m->m_next; 1852 if (m == 0) 1853 break; 1854 } 1855 panic("tcp_pulloutofband"); 1856 } 1857 1858 /* 1859 * Collect new round-trip time estimate 1860 * and update averages and current timeout. 1861 */ 1862 static void 1863 tcp_xmit_timer(tp, rtt) 1864 register struct tcpcb *tp; 1865 short rtt; 1866 { 1867 register int delta; 1868 1869 tcpstat.tcps_rttupdated++; 1870 tp->t_rttupdated++; 1871 if (tp->t_srtt != 0) { 1872 /* 1873 * srtt is stored as fixed point with 5 bits after the 1874 * binary point (i.e., scaled by 8). The following magic 1875 * is equivalent to the smoothing algorithm in rfc793 with 1876 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1877 * point). Adjust rtt to origin 0. 1878 */ 1879 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1880 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1881 1882 if ((tp->t_srtt += delta) <= 0) 1883 tp->t_srtt = 1; 1884 1885 /* 1886 * We accumulate a smoothed rtt variance (actually, a 1887 * smoothed mean difference), then set the retransmit 1888 * timer to smoothed rtt + 4 times the smoothed variance. 1889 * rttvar is stored as fixed point with 4 bits after the 1890 * binary point (scaled by 16). The following is 1891 * equivalent to rfc793 smoothing with an alpha of .75 1892 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1893 * rfc793's wired-in beta. 1894 */ 1895 if (delta < 0) 1896 delta = -delta; 1897 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1898 if ((tp->t_rttvar += delta) <= 0) 1899 tp->t_rttvar = 1; 1900 } else { 1901 /* 1902 * No rtt measurement yet - use the unsmoothed rtt. 1903 * Set the variance to half the rtt (so our first 1904 * retransmit happens at 3*rtt). 1905 */ 1906 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1907 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1908 } 1909 tp->t_rtt = 0; 1910 tp->t_rxtshift = 0; 1911 1912 /* 1913 * the retransmit should happen at rtt + 4 * rttvar. 1914 * Because of the way we do the smoothing, srtt and rttvar 1915 * will each average +1/2 tick of bias. When we compute 1916 * the retransmit timer, we want 1/2 tick of rounding and 1917 * 1 extra tick because of +-1/2 tick uncertainty in the 1918 * firing of the timer. The bias will give us exactly the 1919 * 1.5 tick we need. But, because the bias is 1920 * statistical, we have to test that we don't drop below 1921 * the minimum feasible timer (which is 2 ticks). 1922 */ 1923 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1924 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1925 1926 /* 1927 * We received an ack for a packet that wasn't retransmitted; 1928 * it is probably safe to discard any error indications we've 1929 * received recently. This isn't quite right, but close enough 1930 * for now (a route might have failed after we sent a segment, 1931 * and the return path might not be symmetrical). 1932 */ 1933 tp->t_softerror = 0; 1934 } 1935 1936 /* 1937 * Determine a reasonable value for maxseg size. 1938 * If the route is known, check route for mtu. 1939 * If none, use an mss that can be handled on the outgoing 1940 * interface without forcing IP to fragment; if bigger than 1941 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1942 * to utilize large mbufs. If no route is found, route has no mtu, 1943 * or the destination isn't local, use a default, hopefully conservative 1944 * size (usually 512 or the default IP max size, but no more than the mtu 1945 * of the interface), as we can't discover anything about intervening 1946 * gateways or networks. We also initialize the congestion/slow start 1947 * window to be a single segment if the destination isn't local. 1948 * While looking at the routing entry, we also initialize other path-dependent 1949 * parameters from pre-set or cached values in the routing entry. 1950 * 1951 * Also take into account the space needed for options that we 1952 * send regularly. Make maxseg shorter by that amount to assure 1953 * that we can send maxseg amount of data even when the options 1954 * are present. Store the upper limit of the length of options plus 1955 * data in maxopd. 1956 * 1957 * NOTE that this routine is only called when we process an incoming 1958 * segment, for outgoing segments only tcp_mssopt is called. 1959 * 1960 * In case of T/TCP, we call this routine during implicit connection 1961 * setup as well (offer = -1), to initialize maxseg from the cached 1962 * MSS of our peer. 1963 */ 1964 void 1965 tcp_mss(tp, offer) 1966 struct tcpcb *tp; 1967 int offer; 1968 { 1969 register struct rtentry *rt; 1970 struct ifnet *ifp; 1971 register int rtt, mss; 1972 u_long bufsize; 1973 struct inpcb *inp; 1974 struct socket *so; 1975 struct rmxp_tao *taop; 1976 int origoffer = offer; 1977 1978 inp = tp->t_inpcb; 1979 if ((rt = tcp_rtlookup(inp)) == NULL) { 1980 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1981 return; 1982 } 1983 ifp = rt->rt_ifp; 1984 so = inp->inp_socket; 1985 1986 taop = rmx_taop(rt->rt_rmx); 1987 /* 1988 * Offer == -1 means that we didn't receive SYN yet, 1989 * use cached value in that case; 1990 */ 1991 if (offer == -1) 1992 offer = taop->tao_mssopt; 1993 /* 1994 * Offer == 0 means that there was no MSS on the SYN segment, 1995 * in this case we use tcp_mssdflt. 1996 */ 1997 if (offer == 0) 1998 offer = tcp_mssdflt; 1999 else 2000 /* 2001 * Sanity check: make sure that maxopd will be large 2002 * enough to allow some data on segments even is the 2003 * all the option space is used (40bytes). Otherwise 2004 * funny things may happen in tcp_output. 2005 */ 2006 offer = max(offer, 64); 2007 taop->tao_mssopt = offer; 2008 2009 /* 2010 * While we're here, check if there's an initial rtt 2011 * or rttvar. Convert from the route-table units 2012 * to scaled multiples of the slow timeout timer. 2013 */ 2014 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2015 /* 2016 * XXX the lock bit for RTT indicates that the value 2017 * is also a minimum value; this is subject to time. 2018 */ 2019 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2020 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2021 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2022 tcpstat.tcps_usedrtt++; 2023 if (rt->rt_rmx.rmx_rttvar) { 2024 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2025 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2026 tcpstat.tcps_usedrttvar++; 2027 } else { 2028 /* default variation is +- 1 rtt */ 2029 tp->t_rttvar = 2030 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2031 } 2032 TCPT_RANGESET(tp->t_rxtcur, 2033 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2034 tp->t_rttmin, TCPTV_REXMTMAX); 2035 } 2036 /* 2037 * if there's an mtu associated with the route, use it 2038 */ 2039 if (rt->rt_rmx.rmx_mtu) 2040 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2041 else 2042 { 2043 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2044 if (!in_localaddr(inp->inp_faddr)) 2045 mss = min(mss, tcp_mssdflt); 2046 } 2047 mss = min(mss, offer); 2048 /* 2049 * maxopd stores the maximum length of data AND options 2050 * in a segment; maxseg is the amount of data in a normal 2051 * segment. We need to store this value (maxopd) apart 2052 * from maxseg, because now every segment carries options 2053 * and thus we normally have somewhat less data in segments. 2054 */ 2055 tp->t_maxopd = mss; 2056 2057 /* 2058 * In case of T/TCP, origoffer==-1 indicates, that no segments 2059 * were received yet. In this case we just guess, otherwise 2060 * we do the same as before T/TCP. 2061 */ 2062 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2063 (origoffer == -1 || 2064 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2065 mss -= TCPOLEN_TSTAMP_APPA; 2066 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2067 (origoffer == -1 || 2068 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2069 mss -= TCPOLEN_CC_APPA; 2070 2071 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2072 if (mss > MCLBYTES) 2073 mss &= ~(MCLBYTES-1); 2074 #else 2075 if (mss > MCLBYTES) 2076 mss = mss / MCLBYTES * MCLBYTES; 2077 #endif 2078 /* 2079 * If there's a pipesize, change the socket buffer 2080 * to that size. Make the socket buffers an integral 2081 * number of mss units; if the mss is larger than 2082 * the socket buffer, decrease the mss. 2083 */ 2084 #ifdef RTV_SPIPE 2085 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2086 #endif 2087 bufsize = so->so_snd.sb_hiwat; 2088 if (bufsize < mss) 2089 mss = bufsize; 2090 else { 2091 bufsize = roundup(bufsize, mss); 2092 if (bufsize > sb_max) 2093 bufsize = sb_max; 2094 (void)sbreserve(&so->so_snd, bufsize); 2095 } 2096 tp->t_maxseg = mss; 2097 2098 #ifdef RTV_RPIPE 2099 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2100 #endif 2101 bufsize = so->so_rcv.sb_hiwat; 2102 if (bufsize > mss) { 2103 bufsize = roundup(bufsize, mss); 2104 if (bufsize > sb_max) 2105 bufsize = sb_max; 2106 (void)sbreserve(&so->so_rcv, bufsize); 2107 } 2108 /* 2109 * Don't force slow-start on local network. 2110 */ 2111 if (!in_localaddr(inp->inp_faddr)) 2112 tp->snd_cwnd = mss; 2113 2114 if (rt->rt_rmx.rmx_ssthresh) { 2115 /* 2116 * There's some sort of gateway or interface 2117 * buffer limit on the path. Use this to set 2118 * the slow start threshhold, but set the 2119 * threshold to no less than 2*mss. 2120 */ 2121 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2122 tcpstat.tcps_usedssthresh++; 2123 } 2124 } 2125 2126 /* 2127 * Determine the MSS option to send on an outgoing SYN. 2128 */ 2129 int 2130 tcp_mssopt(tp) 2131 struct tcpcb *tp; 2132 { 2133 struct rtentry *rt; 2134 2135 rt = tcp_rtlookup(tp->t_inpcb); 2136 if (rt == NULL) 2137 return tcp_mssdflt; 2138 2139 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2140 } 2141 #endif /* TUBA_INCLUDE */ 2142