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 1.31 1995/11/03 22:31:54 olah Exp $ 35 */ 36 37 #ifndef TUBA_INCLUDE 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/kernel.h> 41 #include <sys/sysctl.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/errno.h> 48 #include <sys/queue.h> 49 50 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 51 52 #include <net/if.h> 53 #include <net/route.h> 54 55 #include <netinet/in.h> 56 #include <netinet/in_systm.h> 57 #include <netinet/ip.h> 58 #include <netinet/in_pcb.h> 59 #include <netinet/ip_var.h> 60 #include <netinet/tcp.h> 61 #include <netinet/tcp_fsm.h> 62 #include <netinet/tcp_seq.h> 63 #include <netinet/tcp_timer.h> 64 #include <netinet/tcp_var.h> 65 #include <netinet/tcpip.h> 66 #ifdef TCPDEBUG 67 #include <netinet/tcp_debug.h> 68 struct tcpiphdr tcp_saveti; 69 #endif 70 71 int tcprexmtthresh = 3; 72 tcp_seq tcp_iss; 73 tcp_cc tcp_ccgen; 74 struct tcpstat tcpstat; 75 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 76 CTLFLAG_RD, &tcpstat , tcpstat, ""); 77 u_long tcp_now; 78 struct inpcbhead tcb; 79 struct inpcbinfo tcbinfo; 80 81 #endif /* TUBA_INCLUDE */ 82 83 /* 84 * Insert segment ti into reassembly queue of tcp with 85 * control block tp. Return TH_FIN if reassembly now includes 86 * a segment with FIN. The macro form does the common case inline 87 * (segment is the next to be received on an established connection, 88 * and the queue is empty), avoiding linkage into and removal 89 * from the queue and repetition of various conversions. 90 * Set DELACK for segments received in order, but ack immediately 91 * when segments are out of order (so fast retransmit can work). 92 */ 93 #ifdef TCP_ACK_HACK 94 #define TCP_REASS(tp, ti, m, so, flags) { \ 95 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 96 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 97 (tp)->t_state == TCPS_ESTABLISHED) { \ 98 if (ti->ti_flags & TH_PUSH) \ 99 tp->t_flags |= TF_ACKNOW; \ 100 else \ 101 tp->t_flags |= TF_DELACK; \ 102 (tp)->rcv_nxt += (ti)->ti_len; \ 103 flags = (ti)->ti_flags & TH_FIN; \ 104 tcpstat.tcps_rcvpack++;\ 105 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 106 sbappend(&(so)->so_rcv, (m)); \ 107 sorwakeup(so); \ 108 } else { \ 109 (flags) = tcp_reass((tp), (ti), (m)); \ 110 tp->t_flags |= TF_ACKNOW; \ 111 } \ 112 } 113 #else 114 #define TCP_REASS(tp, ti, m, so, flags) { \ 115 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 116 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 117 (tp)->t_state == TCPS_ESTABLISHED) { \ 118 tp->t_flags |= TF_DELACK; \ 119 (tp)->rcv_nxt += (ti)->ti_len; \ 120 flags = (ti)->ti_flags & TH_FIN; \ 121 tcpstat.tcps_rcvpack++;\ 122 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 123 sbappend(&(so)->so_rcv, (m)); \ 124 sorwakeup(so); \ 125 } else { \ 126 (flags) = tcp_reass((tp), (ti), (m)); \ 127 tp->t_flags |= TF_ACKNOW; \ 128 } \ 129 } 130 #endif 131 #ifndef TUBA_INCLUDE 132 133 int 134 tcp_reass(tp, ti, m) 135 register struct tcpcb *tp; 136 register struct tcpiphdr *ti; 137 struct mbuf *m; 138 { 139 register struct tcpiphdr *q; 140 struct socket *so = tp->t_inpcb->inp_socket; 141 int flags; 142 143 /* 144 * Call with ti==0 after become established to 145 * force pre-ESTABLISHED data up to user socket. 146 */ 147 if (ti == 0) 148 goto present; 149 150 /* 151 * Find a segment which begins after this one does. 152 */ 153 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 154 q = (struct tcpiphdr *)q->ti_next) 155 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 156 break; 157 158 /* 159 * If there is a preceding segment, it may provide some of 160 * our data already. If so, drop the data from the incoming 161 * segment. If it provides all of our data, drop us. 162 */ 163 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 164 register int i; 165 q = (struct tcpiphdr *)q->ti_prev; 166 /* conversion to int (in i) handles seq wraparound */ 167 i = q->ti_seq + q->ti_len - ti->ti_seq; 168 if (i > 0) { 169 if (i >= ti->ti_len) { 170 tcpstat.tcps_rcvduppack++; 171 tcpstat.tcps_rcvdupbyte += ti->ti_len; 172 m_freem(m); 173 /* 174 * Try to present any queued data 175 * at the left window edge to the user. 176 * This is needed after the 3-WHS 177 * completes. 178 */ 179 goto present; /* ??? */ 180 } 181 m_adj(m, i); 182 ti->ti_len -= i; 183 ti->ti_seq += i; 184 } 185 q = (struct tcpiphdr *)(q->ti_next); 186 } 187 tcpstat.tcps_rcvoopack++; 188 tcpstat.tcps_rcvoobyte += ti->ti_len; 189 REASS_MBUF(ti) = m; /* XXX */ 190 191 /* 192 * While we overlap succeeding segments trim them or, 193 * if they are completely covered, dequeue them. 194 */ 195 while (q != (struct tcpiphdr *)tp) { 196 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 197 if (i <= 0) 198 break; 199 if (i < q->ti_len) { 200 q->ti_seq += i; 201 q->ti_len -= i; 202 m_adj(REASS_MBUF(q), i); 203 break; 204 } 205 q = (struct tcpiphdr *)q->ti_next; 206 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 207 remque(q->ti_prev); 208 m_freem(m); 209 } 210 211 /* 212 * Stick new segment in its place. 213 */ 214 insque(ti, q->ti_prev); 215 216 present: 217 /* 218 * Present data to user, advancing rcv_nxt through 219 * completed sequence space. 220 */ 221 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 222 return (0); 223 ti = tp->seg_next; 224 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 225 return (0); 226 do { 227 tp->rcv_nxt += ti->ti_len; 228 flags = ti->ti_flags & TH_FIN; 229 remque(ti); 230 m = REASS_MBUF(ti); 231 ti = (struct tcpiphdr *)ti->ti_next; 232 if (so->so_state & SS_CANTRCVMORE) 233 m_freem(m); 234 else 235 sbappend(&so->so_rcv, m); 236 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 237 sorwakeup(so); 238 return (flags); 239 } 240 241 /* 242 * TCP input routine, follows pages 65-76 of the 243 * protocol specification dated September, 1981 very closely. 244 */ 245 void 246 tcp_input(m, iphlen) 247 register struct mbuf *m; 248 int iphlen; 249 { 250 register struct tcpiphdr *ti; 251 register struct inpcb *inp; 252 u_char *optp = NULL; 253 int optlen = 0; 254 int len, tlen, off; 255 register struct tcpcb *tp = 0; 256 register int tiflags; 257 struct socket *so = 0; 258 int todrop, acked, ourfinisacked, needoutput = 0; 259 struct in_addr laddr; 260 int dropsocket = 0; 261 int iss = 0; 262 u_long tiwin; 263 struct tcpopt to; /* options in this segment */ 264 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 265 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 266 #ifdef TCPDEBUG 267 short ostate = 0; 268 #endif 269 270 bzero((char *)&to, sizeof(to)); 271 272 tcpstat.tcps_rcvtotal++; 273 /* 274 * Get IP and TCP header together in first mbuf. 275 * Note: IP leaves IP header in first mbuf. 276 */ 277 ti = mtod(m, struct tcpiphdr *); 278 if (iphlen > sizeof (struct ip)) 279 ip_stripoptions(m, (struct mbuf *)0); 280 if (m->m_len < sizeof (struct tcpiphdr)) { 281 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 282 tcpstat.tcps_rcvshort++; 283 return; 284 } 285 ti = mtod(m, struct tcpiphdr *); 286 } 287 288 /* 289 * Checksum extended TCP header and data. 290 */ 291 tlen = ((struct ip *)ti)->ip_len; 292 len = sizeof (struct ip) + tlen; 293 ti->ti_next = ti->ti_prev = 0; 294 ti->ti_x1 = 0; 295 ti->ti_len = (u_short)tlen; 296 HTONS(ti->ti_len); 297 ti->ti_sum = in_cksum(m, len); 298 if (ti->ti_sum) { 299 tcpstat.tcps_rcvbadsum++; 300 goto drop; 301 } 302 #endif /* TUBA_INCLUDE */ 303 304 /* 305 * Check that TCP offset makes sense, 306 * pull out TCP options and adjust length. XXX 307 */ 308 off = ti->ti_off << 2; 309 if (off < sizeof (struct tcphdr) || off > tlen) { 310 tcpstat.tcps_rcvbadoff++; 311 goto drop; 312 } 313 tlen -= off; 314 ti->ti_len = tlen; 315 if (off > sizeof (struct tcphdr)) { 316 if (m->m_len < sizeof(struct ip) + off) { 317 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 318 tcpstat.tcps_rcvshort++; 319 return; 320 } 321 ti = mtod(m, struct tcpiphdr *); 322 } 323 optlen = off - sizeof (struct tcphdr); 324 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 325 /* 326 * Do quick retrieval of timestamp options ("options 327 * prediction?"). If timestamp is the only option and it's 328 * formatted as recommended in RFC 1323 appendix A, we 329 * quickly get the values now and not bother calling 330 * tcp_dooptions(), etc. 331 */ 332 if ((optlen == TCPOLEN_TSTAMP_APPA || 333 (optlen > TCPOLEN_TSTAMP_APPA && 334 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 335 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 336 (ti->ti_flags & TH_SYN) == 0) { 337 to.to_flag |= TOF_TS; 338 to.to_tsval = ntohl(*(u_long *)(optp + 4)); 339 to.to_tsecr = ntohl(*(u_long *)(optp + 8)); 340 optp = NULL; /* we've parsed the options */ 341 } 342 } 343 tiflags = ti->ti_flags; 344 345 /* 346 * Convert TCP protocol specific fields to host format. 347 */ 348 NTOHL(ti->ti_seq); 349 NTOHL(ti->ti_ack); 350 NTOHS(ti->ti_win); 351 NTOHS(ti->ti_urp); 352 353 /* 354 * Drop TCP, IP headers and TCP options. 355 */ 356 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 357 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 358 359 /* 360 * Locate pcb for segment. 361 */ 362 findpcb: 363 /* 364 * First look for an exact match. 365 */ 366 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 367 ti->ti_dst, ti->ti_dport); 368 /* 369 * ...and if that fails, do a wildcard search. 370 */ 371 if (inp == NULL) { 372 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 373 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 374 } 375 376 /* 377 * If the state is CLOSED (i.e., TCB does not exist) then 378 * all data in the incoming segment is discarded. 379 * If the TCB exists but is in CLOSED state, it is embryonic, 380 * but should either do a listen or a connect soon. 381 */ 382 if (inp == NULL) 383 goto dropwithreset; 384 tp = intotcpcb(inp); 385 if (tp == 0) 386 goto dropwithreset; 387 if (tp->t_state == TCPS_CLOSED) 388 goto drop; 389 390 /* Unscale the window into a 32-bit value. */ 391 if ((tiflags & TH_SYN) == 0) 392 tiwin = ti->ti_win << tp->snd_scale; 393 else 394 tiwin = ti->ti_win; 395 396 so = inp->inp_socket; 397 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 398 #ifdef TCPDEBUG 399 if (so->so_options & SO_DEBUG) { 400 ostate = tp->t_state; 401 tcp_saveti = *ti; 402 } 403 #endif 404 if (so->so_options & SO_ACCEPTCONN) { 405 register struct tcpcb *tp0 = tp; 406 so = sonewconn(so, 0); 407 if (so == 0) 408 goto drop; 409 /* 410 * This is ugly, but .... 411 * 412 * Mark socket as temporary until we're 413 * committed to keeping it. The code at 414 * ``drop'' and ``dropwithreset'' check the 415 * flag dropsocket to see if the temporary 416 * socket created here should be discarded. 417 * We mark the socket as discardable until 418 * we're committed to it below in TCPS_LISTEN. 419 */ 420 dropsocket++; 421 inp = (struct inpcb *)so->so_pcb; 422 inp->inp_laddr = ti->ti_dst; 423 inp->inp_lport = ti->ti_dport; 424 in_pcbrehash(inp); 425 #if BSD>=43 426 inp->inp_options = ip_srcroute(); 427 #endif 428 tp = intotcpcb(inp); 429 tp->t_state = TCPS_LISTEN; 430 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 431 432 /* Compute proper scaling value from buffer space */ 433 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 434 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 435 tp->request_r_scale++; 436 } 437 } 438 439 /* 440 * Segment received on connection. 441 * Reset idle time and keep-alive timer. 442 */ 443 tp->t_idle = 0; 444 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 445 446 /* 447 * Process options if not in LISTEN state, 448 * else do it below (after getting remote address). 449 */ 450 if (optp && tp->t_state != TCPS_LISTEN) 451 tcp_dooptions(tp, optp, optlen, ti, 452 &to); 453 454 /* 455 * Header prediction: check for the two common cases 456 * of a uni-directional data xfer. If the packet has 457 * no control flags, is in-sequence, the window didn't 458 * change and we're not retransmitting, it's a 459 * candidate. If the length is zero and the ack moved 460 * forward, we're the sender side of the xfer. Just 461 * free the data acked & wake any higher level process 462 * that was blocked waiting for space. If the length 463 * is non-zero and the ack didn't move, we're the 464 * receiver side. If we're getting packets in-order 465 * (the reassembly queue is empty), add the data to 466 * the socket buffer and note that we need a delayed ack. 467 * Make sure that the hidden state-flags are also off. 468 * Since we check for TCPS_ESTABLISHED above, it can only 469 * be TH_NEEDSYN. 470 */ 471 if (tp->t_state == TCPS_ESTABLISHED && 472 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 473 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 474 ((to.to_flag & TOF_TS) == 0 || 475 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 476 /* 477 * Using the CC option is compulsory if once started: 478 * the segment is OK if no T/TCP was negotiated or 479 * if the segment has a CC option equal to CCrecv 480 */ 481 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 482 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 483 ti->ti_seq == tp->rcv_nxt && 484 tiwin && tiwin == tp->snd_wnd && 485 tp->snd_nxt == tp->snd_max) { 486 487 /* 488 * If last ACK falls within this segment's sequence numbers, 489 * record the timestamp. 490 * NOTE that the test is modified according to the latest 491 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 492 */ 493 if ((to.to_flag & TOF_TS) != 0 && 494 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 495 tp->ts_recent_age = tcp_now; 496 tp->ts_recent = to.to_tsval; 497 } 498 499 if (ti->ti_len == 0) { 500 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 501 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 502 tp->snd_cwnd >= tp->snd_wnd) { 503 /* 504 * this is a pure ack for outstanding data. 505 */ 506 ++tcpstat.tcps_predack; 507 if ((to.to_flag & TOF_TS) != 0) 508 tcp_xmit_timer(tp, 509 tcp_now - to.to_tsecr + 1); 510 else if (tp->t_rtt && 511 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 512 tcp_xmit_timer(tp, tp->t_rtt); 513 acked = ti->ti_ack - tp->snd_una; 514 tcpstat.tcps_rcvackpack++; 515 tcpstat.tcps_rcvackbyte += acked; 516 sbdrop(&so->so_snd, acked); 517 tp->snd_una = ti->ti_ack; 518 m_freem(m); 519 520 /* 521 * If all outstanding data are acked, stop 522 * retransmit timer, otherwise restart timer 523 * using current (possibly backed-off) value. 524 * If process is waiting for space, 525 * wakeup/selwakeup/signal. If data 526 * are ready to send, let tcp_output 527 * decide between more output or persist. 528 */ 529 if (tp->snd_una == tp->snd_max) 530 tp->t_timer[TCPT_REXMT] = 0; 531 else if (tp->t_timer[TCPT_PERSIST] == 0) 532 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 533 534 if (so->so_snd.sb_flags & SB_NOTIFY) 535 sowwakeup(so); 536 if (so->so_snd.sb_cc) 537 (void) tcp_output(tp); 538 return; 539 } 540 } else if (ti->ti_ack == tp->snd_una && 541 tp->seg_next == (struct tcpiphdr *)tp && 542 ti->ti_len <= sbspace(&so->so_rcv)) { 543 /* 544 * this is a pure, in-sequence data packet 545 * with nothing on the reassembly queue and 546 * we have enough buffer space to take it. 547 */ 548 ++tcpstat.tcps_preddat; 549 tp->rcv_nxt += ti->ti_len; 550 tcpstat.tcps_rcvpack++; 551 tcpstat.tcps_rcvbyte += ti->ti_len; 552 /* 553 * Add data to socket buffer. 554 */ 555 sbappend(&so->so_rcv, m); 556 sorwakeup(so); 557 #ifdef TCP_ACK_HACK 558 /* 559 * If this is a short packet, then ACK now - with Nagel 560 * congestion avoidance sender won't send more until 561 * he gets an ACK. 562 */ 563 if (tiflags & TH_PUSH) { 564 tp->t_flags |= TF_ACKNOW; 565 tcp_output(tp); 566 } else { 567 tp->t_flags |= TF_DELACK; 568 } 569 #else 570 tp->t_flags |= TF_DELACK; 571 #endif 572 return; 573 } 574 } 575 576 /* 577 * Calculate amount of space in receive window, 578 * and then do TCP input processing. 579 * Receive window is amount of space in rcv queue, 580 * but not less than advertised window. 581 */ 582 { int win; 583 584 win = sbspace(&so->so_rcv); 585 if (win < 0) 586 win = 0; 587 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 588 } 589 590 switch (tp->t_state) { 591 592 /* 593 * If the state is LISTEN then ignore segment if it contains an RST. 594 * If the segment contains an ACK then it is bad and send a RST. 595 * If it does not contain a SYN then it is not interesting; drop it. 596 * Don't bother responding if the destination was a broadcast. 597 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 598 * tp->iss, and send a segment: 599 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 600 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 601 * Fill in remote peer address fields if not previously specified. 602 * Enter SYN_RECEIVED state, and process any other fields of this 603 * segment in this state. 604 */ 605 case TCPS_LISTEN: { 606 struct mbuf *am; 607 register struct sockaddr_in *sin; 608 609 if (tiflags & TH_RST) 610 goto drop; 611 if (tiflags & TH_ACK) 612 goto dropwithreset; 613 if ((tiflags & TH_SYN) == 0) 614 goto drop; 615 /* 616 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 617 * in_broadcast() should never return true on a received 618 * packet with M_BCAST not set. 619 */ 620 if (m->m_flags & (M_BCAST|M_MCAST) || 621 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 622 goto drop; 623 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 624 if (am == NULL) 625 goto drop; 626 am->m_len = sizeof (struct sockaddr_in); 627 sin = mtod(am, struct sockaddr_in *); 628 sin->sin_family = AF_INET; 629 sin->sin_len = sizeof(*sin); 630 sin->sin_addr = ti->ti_src; 631 sin->sin_port = ti->ti_sport; 632 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 633 laddr = inp->inp_laddr; 634 if (inp->inp_laddr.s_addr == INADDR_ANY) 635 inp->inp_laddr = ti->ti_dst; 636 if (in_pcbconnect(inp, am)) { 637 inp->inp_laddr = laddr; 638 (void) m_free(am); 639 goto drop; 640 } 641 (void) m_free(am); 642 tp->t_template = tcp_template(tp); 643 if (tp->t_template == 0) { 644 tp = tcp_drop(tp, ENOBUFS); 645 dropsocket = 0; /* socket is already gone */ 646 goto drop; 647 } 648 if ((taop = tcp_gettaocache(inp)) == NULL) { 649 taop = &tao_noncached; 650 bzero(taop, sizeof(*taop)); 651 } 652 if (optp) 653 tcp_dooptions(tp, optp, optlen, ti, 654 &to); 655 if (iss) 656 tp->iss = iss; 657 else 658 tp->iss = tcp_iss; 659 tcp_iss += TCP_ISSINCR/4; 660 tp->irs = ti->ti_seq; 661 tcp_sendseqinit(tp); 662 tcp_rcvseqinit(tp); 663 /* 664 * Initialization of the tcpcb for transaction; 665 * set SND.WND = SEG.WND, 666 * initialize CCsend and CCrecv. 667 */ 668 tp->snd_wnd = tiwin; /* initial send-window */ 669 tp->cc_send = CC_INC(tcp_ccgen); 670 tp->cc_recv = to.to_cc; 671 /* 672 * Perform TAO test on incoming CC (SEG.CC) option, if any. 673 * - compare SEG.CC against cached CC from the same host, 674 * if any. 675 * - if SEG.CC > chached value, SYN must be new and is accepted 676 * immediately: save new CC in the cache, mark the socket 677 * connected, enter ESTABLISHED state, turn on flag to 678 * send a SYN in the next segment. 679 * A virtual advertised window is set in rcv_adv to 680 * initialize SWS prevention. Then enter normal segment 681 * processing: drop SYN, process data and FIN. 682 * - otherwise do a normal 3-way handshake. 683 */ 684 if ((to.to_flag & TOF_CC) != 0) { 685 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 686 taop->tao_cc = to.to_cc; 687 tp->t_state = TCPS_ESTABLISHED; 688 689 /* 690 * If there is a FIN, or if there is data and the 691 * connection is local, then delay SYN,ACK(SYN) in 692 * the hope of piggy-backing it on a response 693 * segment. Otherwise must send ACK now in case 694 * the other side is slow starting. 695 */ 696 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 697 in_localaddr(inp->inp_faddr))) 698 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 699 else 700 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 701 tp->rcv_adv += tp->rcv_wnd; 702 tcpstat.tcps_connects++; 703 soisconnected(so); 704 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 705 dropsocket = 0; /* committed to socket */ 706 tcpstat.tcps_accepts++; 707 goto trimthenstep6; 708 } 709 /* else do standard 3-way handshake */ 710 } else { 711 /* 712 * No CC option, but maybe CC.NEW: 713 * invalidate cached value. 714 */ 715 taop->tao_cc = 0; 716 } 717 /* 718 * TAO test failed or there was no CC option, 719 * do a standard 3-way handshake. 720 */ 721 tp->t_flags |= TF_ACKNOW; 722 tp->t_state = TCPS_SYN_RECEIVED; 723 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 724 dropsocket = 0; /* committed to socket */ 725 tcpstat.tcps_accepts++; 726 goto trimthenstep6; 727 } 728 729 /* 730 * If the state is SYN_SENT: 731 * if seg contains an ACK, but not for our SYN, drop the input. 732 * if seg contains a RST, then drop the connection. 733 * if seg does not contain SYN, then drop it. 734 * Otherwise this is an acceptable SYN segment 735 * initialize tp->rcv_nxt and tp->irs 736 * if seg contains ack then advance tp->snd_una 737 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 738 * arrange for segment to be acked (eventually) 739 * continue processing rest of data/controls, beginning with URG 740 */ 741 case TCPS_SYN_SENT: 742 if ((taop = tcp_gettaocache(inp)) == NULL) { 743 taop = &tao_noncached; 744 bzero(taop, sizeof(*taop)); 745 } 746 747 if ((tiflags & TH_ACK) && 748 (SEQ_LEQ(ti->ti_ack, tp->iss) || 749 SEQ_GT(ti->ti_ack, tp->snd_max))) { 750 /* 751 * If we have a cached CCsent for the remote host, 752 * hence we haven't just crashed and restarted, 753 * do not send a RST. This may be a retransmission 754 * from the other side after our earlier ACK was lost. 755 * Our new SYN, when it arrives, will serve as the 756 * needed ACK. 757 */ 758 if (taop->tao_ccsent != 0) 759 goto drop; 760 else 761 goto dropwithreset; 762 } 763 if (tiflags & TH_RST) { 764 if (tiflags & TH_ACK) 765 tp = tcp_drop(tp, ECONNREFUSED); 766 goto drop; 767 } 768 if ((tiflags & TH_SYN) == 0) 769 goto drop; 770 tp->snd_wnd = ti->ti_win; /* initial send window */ 771 tp->cc_recv = to.to_cc; /* foreign CC */ 772 773 tp->irs = ti->ti_seq; 774 tcp_rcvseqinit(tp); 775 if (tiflags & TH_ACK) { 776 /* 777 * Our SYN was acked. If segment contains CC.ECHO 778 * option, check it to make sure this segment really 779 * matches our SYN. If not, just drop it as old 780 * duplicate, but send an RST if we're still playing 781 * by the old rules. 782 */ 783 if ((to.to_flag & TOF_CCECHO) && 784 tp->cc_send != to.to_ccecho) { 785 if (taop->tao_ccsent != 0) 786 goto drop; 787 else 788 goto dropwithreset; 789 } 790 tcpstat.tcps_connects++; 791 soisconnected(so); 792 /* Do window scaling on this connection? */ 793 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 794 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 795 tp->snd_scale = tp->requested_s_scale; 796 tp->rcv_scale = tp->request_r_scale; 797 } 798 /* Segment is acceptable, update cache if undefined. */ 799 if (taop->tao_ccsent == 0) 800 taop->tao_ccsent = to.to_ccecho; 801 802 tp->rcv_adv += tp->rcv_wnd; 803 tp->snd_una++; /* SYN is acked */ 804 /* 805 * If there's data, delay ACK; if there's also a FIN 806 * ACKNOW will be turned on later. 807 */ 808 if (ti->ti_len != 0) 809 tp->t_flags |= TF_DELACK; 810 else 811 tp->t_flags |= TF_ACKNOW; 812 /* 813 * Received <SYN,ACK> in SYN_SENT[*] state. 814 * Transitions: 815 * SYN_SENT --> ESTABLISHED 816 * SYN_SENT* --> FIN_WAIT_1 817 */ 818 if (tp->t_flags & TF_NEEDFIN) { 819 tp->t_state = TCPS_FIN_WAIT_1; 820 tp->t_flags &= ~TF_NEEDFIN; 821 tiflags &= ~TH_SYN; 822 } else 823 tp->t_state = TCPS_ESTABLISHED; 824 825 } else { 826 /* 827 * Received initial SYN in SYN-SENT[*] state => simul- 828 * taneous open. If segment contains CC option and there is 829 * a cached CC, apply TAO test; if it succeeds, connection is 830 * half-synchronized. Otherwise, do 3-way handshake: 831 * SYN-SENT -> SYN-RECEIVED 832 * SYN-SENT* -> SYN-RECEIVED* 833 * If there was no CC option, clear cached CC value. 834 */ 835 tp->t_flags |= TF_ACKNOW; 836 tp->t_timer[TCPT_REXMT] = 0; 837 if (to.to_flag & TOF_CC) { 838 if (taop->tao_cc != 0 && 839 CC_GT(to.to_cc, taop->tao_cc)) { 840 /* 841 * update cache and make transition: 842 * SYN-SENT -> ESTABLISHED* 843 * SYN-SENT* -> FIN-WAIT-1* 844 */ 845 taop->tao_cc = to.to_cc; 846 if (tp->t_flags & TF_NEEDFIN) { 847 tp->t_state = TCPS_FIN_WAIT_1; 848 tp->t_flags &= ~TF_NEEDFIN; 849 } else 850 tp->t_state = TCPS_ESTABLISHED; 851 tp->t_flags |= TF_NEEDSYN; 852 } else 853 tp->t_state = TCPS_SYN_RECEIVED; 854 } else { 855 /* CC.NEW or no option => invalidate cache */ 856 taop->tao_cc = 0; 857 tp->t_state = TCPS_SYN_RECEIVED; 858 } 859 } 860 861 trimthenstep6: 862 /* 863 * Advance ti->ti_seq to correspond to first data byte. 864 * If data, trim to stay within window, 865 * dropping FIN if necessary. 866 */ 867 ti->ti_seq++; 868 if (ti->ti_len > tp->rcv_wnd) { 869 todrop = ti->ti_len - tp->rcv_wnd; 870 m_adj(m, -todrop); 871 ti->ti_len = tp->rcv_wnd; 872 tiflags &= ~TH_FIN; 873 tcpstat.tcps_rcvpackafterwin++; 874 tcpstat.tcps_rcvbyteafterwin += todrop; 875 } 876 tp->snd_wl1 = ti->ti_seq - 1; 877 tp->rcv_up = ti->ti_seq; 878 /* 879 * Client side of transaction: already sent SYN and data. 880 * If the remote host used T/TCP to validate the SYN, 881 * our data will be ACK'd; if so, enter normal data segment 882 * processing in the middle of step 5, ack processing. 883 * Otherwise, goto step 6. 884 */ 885 if (tiflags & TH_ACK) 886 goto process_ACK; 887 goto step6; 888 /* 889 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 890 * if segment contains a SYN and CC [not CC.NEW] option: 891 * if state == TIME_WAIT and connection duration > MSL, 892 * drop packet and send RST; 893 * 894 * if SEG.CC > CCrecv then is new SYN, and can implicitly 895 * ack the FIN (and data) in retransmission queue. 896 * Complete close and delete TCPCB. Then reprocess 897 * segment, hoping to find new TCPCB in LISTEN state; 898 * 899 * else must be old SYN; drop it. 900 * else do normal processing. 901 */ 902 case TCPS_LAST_ACK: 903 case TCPS_CLOSING: 904 case TCPS_TIME_WAIT: 905 if ((tiflags & TH_SYN) && 906 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 907 if (tp->t_state == TCPS_TIME_WAIT && 908 tp->t_duration > TCPTV_MSL) 909 goto dropwithreset; 910 if (CC_GT(to.to_cc, tp->cc_recv)) { 911 tp = tcp_close(tp); 912 goto findpcb; 913 } 914 else 915 goto drop; 916 } 917 break; /* continue normal processing */ 918 } 919 920 /* 921 * States other than LISTEN or SYN_SENT. 922 * First check timestamp, if present. 923 * Then check the connection count, if present. 924 * Then check that at least some bytes of segment are within 925 * receive window. If segment begins before rcv_nxt, 926 * drop leading data (and SYN); if nothing left, just ack. 927 * 928 * RFC 1323 PAWS: If we have a timestamp reply on this segment 929 * and it's less than ts_recent, drop it. 930 */ 931 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 932 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 933 934 /* Check to see if ts_recent is over 24 days old. */ 935 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 936 /* 937 * Invalidate ts_recent. If this segment updates 938 * ts_recent, the age will be reset later and ts_recent 939 * will get a valid value. If it does not, setting 940 * ts_recent to zero will at least satisfy the 941 * requirement that zero be placed in the timestamp 942 * echo reply when ts_recent isn't valid. The 943 * age isn't reset until we get a valid ts_recent 944 * because we don't want out-of-order segments to be 945 * dropped when ts_recent is old. 946 */ 947 tp->ts_recent = 0; 948 } else { 949 tcpstat.tcps_rcvduppack++; 950 tcpstat.tcps_rcvdupbyte += ti->ti_len; 951 tcpstat.tcps_pawsdrop++; 952 goto dropafterack; 953 } 954 } 955 956 /* 957 * T/TCP mechanism 958 * If T/TCP was negotiated and the segment doesn't have CC, 959 * or if it's CC is wrong then drop the segment. 960 * RST segments do not have to comply with this. 961 */ 962 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 963 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 964 (tiflags & TH_RST) == 0) 965 goto dropafterack; 966 967 todrop = tp->rcv_nxt - ti->ti_seq; 968 if (todrop > 0) { 969 if (tiflags & TH_SYN) { 970 tiflags &= ~TH_SYN; 971 ti->ti_seq++; 972 if (ti->ti_urp > 1) 973 ti->ti_urp--; 974 else 975 tiflags &= ~TH_URG; 976 todrop--; 977 } 978 /* 979 * Following if statement from Stevens, vol. 2, p. 960. 980 */ 981 if (todrop > ti->ti_len 982 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 983 /* 984 * Any valid FIN must be to the left of the window. 985 * At this point the FIN must be a duplicate or out 986 * of sequence; drop it. 987 */ 988 tiflags &= ~TH_FIN; 989 990 /* 991 * Send an ACK to resynchronize and drop any data. 992 * But keep on processing for RST or ACK. 993 */ 994 tp->t_flags |= TF_ACKNOW; 995 todrop = ti->ti_len; 996 tcpstat.tcps_rcvduppack++; 997 tcpstat.tcps_rcvdupbyte += todrop; 998 } else { 999 tcpstat.tcps_rcvpartduppack++; 1000 tcpstat.tcps_rcvpartdupbyte += todrop; 1001 } 1002 m_adj(m, todrop); 1003 ti->ti_seq += todrop; 1004 ti->ti_len -= todrop; 1005 if (ti->ti_urp > todrop) 1006 ti->ti_urp -= todrop; 1007 else { 1008 tiflags &= ~TH_URG; 1009 ti->ti_urp = 0; 1010 } 1011 } 1012 1013 /* 1014 * If new data are received on a connection after the 1015 * user processes are gone, then RST the other end. 1016 */ 1017 if ((so->so_state & SS_NOFDREF) && 1018 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1019 tp = tcp_close(tp); 1020 tcpstat.tcps_rcvafterclose++; 1021 goto dropwithreset; 1022 } 1023 1024 /* 1025 * If segment ends after window, drop trailing data 1026 * (and PUSH and FIN); if nothing left, just ACK. 1027 */ 1028 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1029 if (todrop > 0) { 1030 tcpstat.tcps_rcvpackafterwin++; 1031 if (todrop >= ti->ti_len) { 1032 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1033 /* 1034 * If a new connection request is received 1035 * while in TIME_WAIT, drop the old connection 1036 * and start over if the sequence numbers 1037 * are above the previous ones. 1038 */ 1039 if (tiflags & TH_SYN && 1040 tp->t_state == TCPS_TIME_WAIT && 1041 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1042 iss = tp->rcv_nxt + TCP_ISSINCR; 1043 tp = tcp_close(tp); 1044 goto findpcb; 1045 } 1046 /* 1047 * If window is closed can only take segments at 1048 * window edge, and have to drop data and PUSH from 1049 * incoming segments. Continue processing, but 1050 * remember to ack. Otherwise, drop segment 1051 * and ack. 1052 */ 1053 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1054 tp->t_flags |= TF_ACKNOW; 1055 tcpstat.tcps_rcvwinprobe++; 1056 } else 1057 goto dropafterack; 1058 } else 1059 tcpstat.tcps_rcvbyteafterwin += todrop; 1060 m_adj(m, -todrop); 1061 ti->ti_len -= todrop; 1062 tiflags &= ~(TH_PUSH|TH_FIN); 1063 } 1064 1065 /* 1066 * If last ACK falls within this segment's sequence numbers, 1067 * record its timestamp. 1068 * NOTE that the test is modified according to the latest 1069 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1070 */ 1071 if ((to.to_flag & TOF_TS) != 0 && 1072 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1073 tp->ts_recent_age = tcp_now; 1074 tp->ts_recent = to.to_tsval; 1075 } 1076 1077 /* 1078 * If the RST bit is set examine the state: 1079 * SYN_RECEIVED STATE: 1080 * If passive open, return to LISTEN state. 1081 * If active open, inform user that connection was refused. 1082 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1083 * Inform user that connection was reset, and close tcb. 1084 * CLOSING, LAST_ACK, TIME_WAIT STATES 1085 * Close the tcb. 1086 */ 1087 if (tiflags&TH_RST) switch (tp->t_state) { 1088 1089 case TCPS_SYN_RECEIVED: 1090 so->so_error = ECONNREFUSED; 1091 goto close; 1092 1093 case TCPS_ESTABLISHED: 1094 case TCPS_FIN_WAIT_1: 1095 case TCPS_FIN_WAIT_2: 1096 case TCPS_CLOSE_WAIT: 1097 so->so_error = ECONNRESET; 1098 close: 1099 tp->t_state = TCPS_CLOSED; 1100 tcpstat.tcps_drops++; 1101 tp = tcp_close(tp); 1102 goto drop; 1103 1104 case TCPS_CLOSING: 1105 case TCPS_LAST_ACK: 1106 case TCPS_TIME_WAIT: 1107 tp = tcp_close(tp); 1108 goto drop; 1109 } 1110 1111 /* 1112 * If a SYN is in the window, then this is an 1113 * error and we send an RST and drop the connection. 1114 */ 1115 if (tiflags & TH_SYN) { 1116 tp = tcp_drop(tp, ECONNRESET); 1117 goto dropwithreset; 1118 } 1119 1120 /* 1121 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1122 * flag is on (half-synchronized state), then queue data for 1123 * later processing; else drop segment and return. 1124 */ 1125 if ((tiflags & TH_ACK) == 0) { 1126 if (tp->t_state == TCPS_SYN_RECEIVED || 1127 (tp->t_flags & TF_NEEDSYN)) 1128 goto step6; 1129 else 1130 goto drop; 1131 } 1132 1133 /* 1134 * Ack processing. 1135 */ 1136 switch (tp->t_state) { 1137 1138 /* 1139 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1140 * ESTABLISHED state and continue processing, otherwise 1141 * send an RST. 1142 */ 1143 case TCPS_SYN_RECEIVED: 1144 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1145 SEQ_GT(ti->ti_ack, tp->snd_max)) 1146 goto dropwithreset; 1147 1148 tcpstat.tcps_connects++; 1149 soisconnected(so); 1150 /* Do window scaling? */ 1151 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1152 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1153 tp->snd_scale = tp->requested_s_scale; 1154 tp->rcv_scale = tp->request_r_scale; 1155 } 1156 /* 1157 * Upon successful completion of 3-way handshake, 1158 * update cache.CC if it was undefined, pass any queued 1159 * data to the user, and advance state appropriately. 1160 */ 1161 if ((taop = tcp_gettaocache(inp)) != NULL && 1162 taop->tao_cc == 0) 1163 taop->tao_cc = tp->cc_recv; 1164 1165 /* 1166 * Make transitions: 1167 * SYN-RECEIVED -> ESTABLISHED 1168 * SYN-RECEIVED* -> FIN-WAIT-1 1169 */ 1170 if (tp->t_flags & TF_NEEDFIN) { 1171 tp->t_state = TCPS_FIN_WAIT_1; 1172 tp->t_flags &= ~TF_NEEDFIN; 1173 } else 1174 tp->t_state = TCPS_ESTABLISHED; 1175 /* 1176 * If segment contains data or ACK, will call tcp_reass() 1177 * later; if not, do so now to pass queued data to user. 1178 */ 1179 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1180 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1181 (struct mbuf *)0); 1182 tp->snd_wl1 = ti->ti_seq - 1; 1183 /* fall into ... */ 1184 1185 /* 1186 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1187 * ACKs. If the ack is in the range 1188 * tp->snd_una < ti->ti_ack <= tp->snd_max 1189 * then advance tp->snd_una to ti->ti_ack and drop 1190 * data from the retransmission queue. If this ACK reflects 1191 * more up to date window information we update our window information. 1192 */ 1193 case TCPS_ESTABLISHED: 1194 case TCPS_FIN_WAIT_1: 1195 case TCPS_FIN_WAIT_2: 1196 case TCPS_CLOSE_WAIT: 1197 case TCPS_CLOSING: 1198 case TCPS_LAST_ACK: 1199 case TCPS_TIME_WAIT: 1200 1201 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1202 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1203 tcpstat.tcps_rcvdupack++; 1204 /* 1205 * If we have outstanding data (other than 1206 * a window probe), this is a completely 1207 * duplicate ack (ie, window info didn't 1208 * change), the ack is the biggest we've 1209 * seen and we've seen exactly our rexmt 1210 * threshhold of them, assume a packet 1211 * has been dropped and retransmit it. 1212 * Kludge snd_nxt & the congestion 1213 * window so we send only this one 1214 * packet. 1215 * 1216 * We know we're losing at the current 1217 * window size so do congestion avoidance 1218 * (set ssthresh to half the current window 1219 * and pull our congestion window back to 1220 * the new ssthresh). 1221 * 1222 * Dup acks mean that packets have left the 1223 * network (they're now cached at the receiver) 1224 * so bump cwnd by the amount in the receiver 1225 * to keep a constant cwnd packets in the 1226 * network. 1227 */ 1228 if (tp->t_timer[TCPT_REXMT] == 0 || 1229 ti->ti_ack != tp->snd_una) 1230 tp->t_dupacks = 0; 1231 else if (++tp->t_dupacks == tcprexmtthresh) { 1232 tcp_seq onxt = tp->snd_nxt; 1233 u_int win = 1234 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1235 tp->t_maxseg; 1236 1237 if (win < 2) 1238 win = 2; 1239 tp->snd_ssthresh = win * tp->t_maxseg; 1240 tp->t_timer[TCPT_REXMT] = 0; 1241 tp->t_rtt = 0; 1242 tp->snd_nxt = ti->ti_ack; 1243 tp->snd_cwnd = tp->t_maxseg; 1244 (void) tcp_output(tp); 1245 tp->snd_cwnd = tp->snd_ssthresh + 1246 tp->t_maxseg * tp->t_dupacks; 1247 if (SEQ_GT(onxt, tp->snd_nxt)) 1248 tp->snd_nxt = onxt; 1249 goto drop; 1250 } else if (tp->t_dupacks > tcprexmtthresh) { 1251 tp->snd_cwnd += tp->t_maxseg; 1252 (void) tcp_output(tp); 1253 goto drop; 1254 } 1255 } else 1256 tp->t_dupacks = 0; 1257 break; 1258 } 1259 /* 1260 * If the congestion window was inflated to account 1261 * for the other side's cached packets, retract it. 1262 */ 1263 if (tp->t_dupacks > tcprexmtthresh && 1264 tp->snd_cwnd > tp->snd_ssthresh) 1265 tp->snd_cwnd = tp->snd_ssthresh; 1266 tp->t_dupacks = 0; 1267 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1268 tcpstat.tcps_rcvacktoomuch++; 1269 goto dropafterack; 1270 } 1271 /* 1272 * If we reach this point, ACK is not a duplicate, 1273 * i.e., it ACKs something we sent. 1274 */ 1275 if (tp->t_flags & TF_NEEDSYN) { 1276 /* 1277 * T/TCP: Connection was half-synchronized, and our 1278 * SYN has been ACK'd (so connection is now fully 1279 * synchronized). Go to non-starred state and 1280 * increment snd_una for ACK of SYN. 1281 */ 1282 tp->t_flags &= ~TF_NEEDSYN; 1283 tp->snd_una++; 1284 } 1285 1286 process_ACK: 1287 acked = ti->ti_ack - tp->snd_una; 1288 tcpstat.tcps_rcvackpack++; 1289 tcpstat.tcps_rcvackbyte += acked; 1290 1291 /* 1292 * If we have a timestamp reply, update smoothed 1293 * round trip time. If no timestamp is present but 1294 * transmit timer is running and timed sequence 1295 * number was acked, update smoothed round trip time. 1296 * Since we now have an rtt measurement, cancel the 1297 * timer backoff (cf., Phil Karn's retransmit alg.). 1298 * Recompute the initial retransmit timer. 1299 */ 1300 if (to.to_flag & TOF_TS) 1301 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1302 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1303 tcp_xmit_timer(tp,tp->t_rtt); 1304 1305 /* 1306 * If all outstanding data is acked, stop retransmit 1307 * timer and remember to restart (more output or persist). 1308 * If there is more data to be acked, restart retransmit 1309 * timer, using current (possibly backed-off) value. 1310 */ 1311 if (ti->ti_ack == tp->snd_max) { 1312 tp->t_timer[TCPT_REXMT] = 0; 1313 needoutput = 1; 1314 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1315 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1316 1317 /* 1318 * If no data (only SYN) was ACK'd, 1319 * skip rest of ACK processing. 1320 */ 1321 if (acked == 0) 1322 goto step6; 1323 1324 /* 1325 * When new data is acked, open the congestion window. 1326 * If the window gives us less than ssthresh packets 1327 * in flight, open exponentially (maxseg per packet). 1328 * Otherwise open linearly: maxseg per window 1329 * (maxseg^2 / cwnd per packet). 1330 */ 1331 { 1332 register u_int cw = tp->snd_cwnd; 1333 register u_int incr = tp->t_maxseg; 1334 1335 if (cw > tp->snd_ssthresh) 1336 incr = incr * incr / cw; 1337 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1338 } 1339 if (acked > so->so_snd.sb_cc) { 1340 tp->snd_wnd -= so->so_snd.sb_cc; 1341 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1342 ourfinisacked = 1; 1343 } else { 1344 sbdrop(&so->so_snd, acked); 1345 tp->snd_wnd -= acked; 1346 ourfinisacked = 0; 1347 } 1348 if (so->so_snd.sb_flags & SB_NOTIFY) 1349 sowwakeup(so); 1350 tp->snd_una = ti->ti_ack; 1351 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1352 tp->snd_nxt = tp->snd_una; 1353 1354 switch (tp->t_state) { 1355 1356 /* 1357 * In FIN_WAIT_1 STATE in addition to the processing 1358 * for the ESTABLISHED state if our FIN is now acknowledged 1359 * then enter FIN_WAIT_2. 1360 */ 1361 case TCPS_FIN_WAIT_1: 1362 if (ourfinisacked) { 1363 /* 1364 * If we can't receive any more 1365 * data, then closing user can proceed. 1366 * Starting the timer is contrary to the 1367 * specification, but if we don't get a FIN 1368 * we'll hang forever. 1369 */ 1370 if (so->so_state & SS_CANTRCVMORE) { 1371 soisdisconnected(so); 1372 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1373 } 1374 tp->t_state = TCPS_FIN_WAIT_2; 1375 } 1376 break; 1377 1378 /* 1379 * In CLOSING STATE in addition to the processing for 1380 * the ESTABLISHED state if the ACK acknowledges our FIN 1381 * then enter the TIME-WAIT state, otherwise ignore 1382 * the segment. 1383 */ 1384 case TCPS_CLOSING: 1385 if (ourfinisacked) { 1386 tp->t_state = TCPS_TIME_WAIT; 1387 tcp_canceltimers(tp); 1388 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1389 if (tp->cc_recv != 0 && 1390 tp->t_duration < TCPTV_MSL) 1391 tp->t_timer[TCPT_2MSL] = 1392 tp->t_rxtcur * TCPTV_TWTRUNC; 1393 else 1394 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1395 soisdisconnected(so); 1396 } 1397 break; 1398 1399 /* 1400 * In LAST_ACK, we may still be waiting for data to drain 1401 * and/or to be acked, as well as for the ack of our FIN. 1402 * If our FIN is now acknowledged, delete the TCB, 1403 * enter the closed state and return. 1404 */ 1405 case TCPS_LAST_ACK: 1406 if (ourfinisacked) { 1407 tp = tcp_close(tp); 1408 goto drop; 1409 } 1410 break; 1411 1412 /* 1413 * In TIME_WAIT state the only thing that should arrive 1414 * is a retransmission of the remote FIN. Acknowledge 1415 * it and restart the finack timer. 1416 */ 1417 case TCPS_TIME_WAIT: 1418 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1419 goto dropafterack; 1420 } 1421 } 1422 1423 step6: 1424 /* 1425 * Update window information. 1426 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1427 */ 1428 if ((tiflags & TH_ACK) && 1429 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1430 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1431 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1432 /* keep track of pure window updates */ 1433 if (ti->ti_len == 0 && 1434 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1435 tcpstat.tcps_rcvwinupd++; 1436 tp->snd_wnd = tiwin; 1437 tp->snd_wl1 = ti->ti_seq; 1438 tp->snd_wl2 = ti->ti_ack; 1439 if (tp->snd_wnd > tp->max_sndwnd) 1440 tp->max_sndwnd = tp->snd_wnd; 1441 needoutput = 1; 1442 } 1443 1444 /* 1445 * Process segments with URG. 1446 */ 1447 if ((tiflags & TH_URG) && ti->ti_urp && 1448 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1449 /* 1450 * This is a kludge, but if we receive and accept 1451 * random urgent pointers, we'll crash in 1452 * soreceive. It's hard to imagine someone 1453 * actually wanting to send this much urgent data. 1454 */ 1455 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1456 ti->ti_urp = 0; /* XXX */ 1457 tiflags &= ~TH_URG; /* XXX */ 1458 goto dodata; /* XXX */ 1459 } 1460 /* 1461 * If this segment advances the known urgent pointer, 1462 * then mark the data stream. This should not happen 1463 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1464 * a FIN has been received from the remote side. 1465 * In these states we ignore the URG. 1466 * 1467 * According to RFC961 (Assigned Protocols), 1468 * the urgent pointer points to the last octet 1469 * of urgent data. We continue, however, 1470 * to consider it to indicate the first octet 1471 * of data past the urgent section as the original 1472 * spec states (in one of two places). 1473 */ 1474 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1475 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1476 so->so_oobmark = so->so_rcv.sb_cc + 1477 (tp->rcv_up - tp->rcv_nxt) - 1; 1478 if (so->so_oobmark == 0) 1479 so->so_state |= SS_RCVATMARK; 1480 sohasoutofband(so); 1481 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1482 } 1483 /* 1484 * Remove out of band data so doesn't get presented to user. 1485 * This can happen independent of advancing the URG pointer, 1486 * but if two URG's are pending at once, some out-of-band 1487 * data may creep in... ick. 1488 */ 1489 if (ti->ti_urp <= (u_long)ti->ti_len 1490 #ifdef SO_OOBINLINE 1491 && (so->so_options & SO_OOBINLINE) == 0 1492 #endif 1493 ) 1494 tcp_pulloutofband(so, ti, m); 1495 } else 1496 /* 1497 * If no out of band data is expected, 1498 * pull receive urgent pointer along 1499 * with the receive window. 1500 */ 1501 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1502 tp->rcv_up = tp->rcv_nxt; 1503 dodata: /* XXX */ 1504 1505 /* 1506 * Process the segment text, merging it into the TCP sequencing queue, 1507 * and arranging for acknowledgment of receipt if necessary. 1508 * This process logically involves adjusting tp->rcv_wnd as data 1509 * is presented to the user (this happens in tcp_usrreq.c, 1510 * case PRU_RCVD). If a FIN has already been received on this 1511 * connection then we just ignore the text. 1512 */ 1513 if ((ti->ti_len || (tiflags&TH_FIN)) && 1514 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1515 TCP_REASS(tp, ti, m, so, tiflags); 1516 /* 1517 * Note the amount of data that peer has sent into 1518 * our window, in order to estimate the sender's 1519 * buffer size. 1520 */ 1521 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1522 } else { 1523 m_freem(m); 1524 tiflags &= ~TH_FIN; 1525 } 1526 1527 /* 1528 * If FIN is received ACK the FIN and let the user know 1529 * that the connection is closing. 1530 */ 1531 if (tiflags & TH_FIN) { 1532 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1533 socantrcvmore(so); 1534 /* 1535 * If connection is half-synchronized 1536 * (ie NEEDSYN flag on) then delay ACK, 1537 * so it may be piggybacked when SYN is sent. 1538 * Otherwise, since we received a FIN then no 1539 * more input can be expected, send ACK now. 1540 */ 1541 if (tp->t_flags & TF_NEEDSYN) 1542 tp->t_flags |= TF_DELACK; 1543 else 1544 tp->t_flags |= TF_ACKNOW; 1545 tp->rcv_nxt++; 1546 } 1547 switch (tp->t_state) { 1548 1549 /* 1550 * In SYN_RECEIVED and ESTABLISHED STATES 1551 * enter the CLOSE_WAIT state. 1552 */ 1553 case TCPS_SYN_RECEIVED: 1554 case TCPS_ESTABLISHED: 1555 tp->t_state = TCPS_CLOSE_WAIT; 1556 break; 1557 1558 /* 1559 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1560 * enter the CLOSING state. 1561 */ 1562 case TCPS_FIN_WAIT_1: 1563 tp->t_state = TCPS_CLOSING; 1564 break; 1565 1566 /* 1567 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1568 * starting the time-wait timer, turning off the other 1569 * standard timers. 1570 */ 1571 case TCPS_FIN_WAIT_2: 1572 tp->t_state = TCPS_TIME_WAIT; 1573 tcp_canceltimers(tp); 1574 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1575 if (tp->cc_recv != 0 && 1576 tp->t_duration < TCPTV_MSL) { 1577 tp->t_timer[TCPT_2MSL] = 1578 tp->t_rxtcur * TCPTV_TWTRUNC; 1579 /* For transaction client, force ACK now. */ 1580 tp->t_flags |= TF_ACKNOW; 1581 } 1582 else 1583 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1584 soisdisconnected(so); 1585 break; 1586 1587 /* 1588 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1589 */ 1590 case TCPS_TIME_WAIT: 1591 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1592 break; 1593 } 1594 } 1595 #ifdef TCPDEBUG 1596 if (so->so_options & SO_DEBUG) 1597 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1598 #endif 1599 1600 /* 1601 * Return any desired output. 1602 */ 1603 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1604 (void) tcp_output(tp); 1605 return; 1606 1607 dropafterack: 1608 /* 1609 * Generate an ACK dropping incoming segment if it occupies 1610 * sequence space, where the ACK reflects our state. 1611 */ 1612 if (tiflags & TH_RST) 1613 goto drop; 1614 #ifdef TCPDEBUG 1615 if (so->so_options & SO_DEBUG) 1616 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1617 #endif 1618 m_freem(m); 1619 tp->t_flags |= TF_ACKNOW; 1620 (void) tcp_output(tp); 1621 return; 1622 1623 dropwithreset: 1624 /* 1625 * Generate a RST, dropping incoming segment. 1626 * Make ACK acceptable to originator of segment. 1627 * Don't bother to respond if destination was broadcast/multicast. 1628 */ 1629 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1630 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1631 goto drop; 1632 #ifdef TCPDEBUG 1633 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1634 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1635 #endif 1636 if (tiflags & TH_ACK) 1637 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1638 else { 1639 if (tiflags & TH_SYN) 1640 ti->ti_len++; 1641 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1642 TH_RST|TH_ACK); 1643 } 1644 /* destroy temporarily created socket */ 1645 if (dropsocket) 1646 (void) soabort(so); 1647 return; 1648 1649 drop: 1650 /* 1651 * Drop space held by incoming segment and return. 1652 */ 1653 #ifdef TCPDEBUG 1654 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1655 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1656 #endif 1657 m_freem(m); 1658 /* destroy temporarily created socket */ 1659 if (dropsocket) 1660 (void) soabort(so); 1661 return; 1662 #ifndef TUBA_INCLUDE 1663 } 1664 1665 void 1666 tcp_dooptions(tp, cp, cnt, ti, to) 1667 struct tcpcb *tp; 1668 u_char *cp; 1669 int cnt; 1670 struct tcpiphdr *ti; 1671 struct tcpopt *to; 1672 { 1673 u_short mss = 0; 1674 int opt, optlen; 1675 1676 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1677 opt = cp[0]; 1678 if (opt == TCPOPT_EOL) 1679 break; 1680 if (opt == TCPOPT_NOP) 1681 optlen = 1; 1682 else { 1683 optlen = cp[1]; 1684 if (optlen <= 0) 1685 break; 1686 } 1687 switch (opt) { 1688 1689 default: 1690 continue; 1691 1692 case TCPOPT_MAXSEG: 1693 if (optlen != TCPOLEN_MAXSEG) 1694 continue; 1695 if (!(ti->ti_flags & TH_SYN)) 1696 continue; 1697 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1698 NTOHS(mss); 1699 break; 1700 1701 case TCPOPT_WINDOW: 1702 if (optlen != TCPOLEN_WINDOW) 1703 continue; 1704 if (!(ti->ti_flags & TH_SYN)) 1705 continue; 1706 tp->t_flags |= TF_RCVD_SCALE; 1707 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1708 break; 1709 1710 case TCPOPT_TIMESTAMP: 1711 if (optlen != TCPOLEN_TIMESTAMP) 1712 continue; 1713 to->to_flag |= TOF_TS; 1714 bcopy((char *)cp + 2, 1715 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1716 NTOHL(to->to_tsval); 1717 bcopy((char *)cp + 6, 1718 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1719 NTOHL(to->to_tsecr); 1720 1721 /* 1722 * A timestamp received in a SYN makes 1723 * it ok to send timestamp requests and replies. 1724 */ 1725 if (ti->ti_flags & TH_SYN) { 1726 tp->t_flags |= TF_RCVD_TSTMP; 1727 tp->ts_recent = to->to_tsval; 1728 tp->ts_recent_age = tcp_now; 1729 } 1730 break; 1731 case TCPOPT_CC: 1732 if (optlen != TCPOLEN_CC) 1733 continue; 1734 to->to_flag |= TOF_CC; 1735 bcopy((char *)cp + 2, 1736 (char *)&to->to_cc, sizeof(to->to_cc)); 1737 NTOHL(to->to_cc); 1738 /* 1739 * A CC or CC.new option received in a SYN makes 1740 * it ok to send CC in subsequent segments. 1741 */ 1742 if (ti->ti_flags & TH_SYN) 1743 tp->t_flags |= TF_RCVD_CC; 1744 break; 1745 case TCPOPT_CCNEW: 1746 if (optlen != TCPOLEN_CC) 1747 continue; 1748 if (!(ti->ti_flags & TH_SYN)) 1749 continue; 1750 to->to_flag |= TOF_CCNEW; 1751 bcopy((char *)cp + 2, 1752 (char *)&to->to_cc, sizeof(to->to_cc)); 1753 NTOHL(to->to_cc); 1754 /* 1755 * A CC or CC.new option received in a SYN makes 1756 * it ok to send CC in subsequent segments. 1757 */ 1758 tp->t_flags |= TF_RCVD_CC; 1759 break; 1760 case TCPOPT_CCECHO: 1761 if (optlen != TCPOLEN_CC) 1762 continue; 1763 if (!(ti->ti_flags & TH_SYN)) 1764 continue; 1765 to->to_flag |= TOF_CCECHO; 1766 bcopy((char *)cp + 2, 1767 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1768 NTOHL(to->to_ccecho); 1769 break; 1770 } 1771 } 1772 if (ti->ti_flags & TH_SYN) 1773 tcp_mss(tp, mss); /* sets t_maxseg */ 1774 } 1775 1776 /* 1777 * Pull out of band byte out of a segment so 1778 * it doesn't appear in the user's data queue. 1779 * It is still reflected in the segment length for 1780 * sequencing purposes. 1781 */ 1782 void 1783 tcp_pulloutofband(so, ti, m) 1784 struct socket *so; 1785 struct tcpiphdr *ti; 1786 register struct mbuf *m; 1787 { 1788 int cnt = ti->ti_urp - 1; 1789 1790 while (cnt >= 0) { 1791 if (m->m_len > cnt) { 1792 char *cp = mtod(m, caddr_t) + cnt; 1793 struct tcpcb *tp = sototcpcb(so); 1794 1795 tp->t_iobc = *cp; 1796 tp->t_oobflags |= TCPOOB_HAVEDATA; 1797 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1798 m->m_len--; 1799 return; 1800 } 1801 cnt -= m->m_len; 1802 m = m->m_next; 1803 if (m == 0) 1804 break; 1805 } 1806 panic("tcp_pulloutofband"); 1807 } 1808 1809 /* 1810 * Collect new round-trip time estimate 1811 * and update averages and current timeout. 1812 */ 1813 void 1814 tcp_xmit_timer(tp, rtt) 1815 register struct tcpcb *tp; 1816 short rtt; 1817 { 1818 register short delta; 1819 1820 tcpstat.tcps_rttupdated++; 1821 tp->t_rttupdated++; 1822 if (tp->t_srtt != 0) { 1823 /* 1824 * srtt is stored as fixed point with 3 bits after the 1825 * binary point (i.e., scaled by 8). The following magic 1826 * is equivalent to the smoothing algorithm in rfc793 with 1827 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1828 * point). Adjust rtt to origin 0. 1829 */ 1830 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1831 if ((tp->t_srtt += delta) <= 0) 1832 tp->t_srtt = 1; 1833 /* 1834 * We accumulate a smoothed rtt variance (actually, a 1835 * smoothed mean difference), then set the retransmit 1836 * timer to smoothed rtt + 4 times the smoothed variance. 1837 * rttvar is stored as fixed point with 2 bits after the 1838 * binary point (scaled by 4). The following is 1839 * equivalent to rfc793 smoothing with an alpha of .75 1840 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1841 * rfc793's wired-in beta. 1842 */ 1843 if (delta < 0) 1844 delta = -delta; 1845 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1846 if ((tp->t_rttvar += delta) <= 0) 1847 tp->t_rttvar = 1; 1848 } else { 1849 /* 1850 * No rtt measurement yet - use the unsmoothed rtt. 1851 * Set the variance to half the rtt (so our first 1852 * retransmit happens at 3*rtt). 1853 */ 1854 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1855 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1856 } 1857 tp->t_rtt = 0; 1858 tp->t_rxtshift = 0; 1859 1860 /* 1861 * the retransmit should happen at rtt + 4 * rttvar. 1862 * Because of the way we do the smoothing, srtt and rttvar 1863 * will each average +1/2 tick of bias. When we compute 1864 * the retransmit timer, we want 1/2 tick of rounding and 1865 * 1 extra tick because of +-1/2 tick uncertainty in the 1866 * firing of the timer. The bias will give us exactly the 1867 * 1.5 tick we need. But, because the bias is 1868 * statistical, we have to test that we don't drop below 1869 * the minimum feasible timer (which is 2 ticks). 1870 */ 1871 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1872 tp->t_rttmin, TCPTV_REXMTMAX); 1873 1874 /* 1875 * We received an ack for a packet that wasn't retransmitted; 1876 * it is probably safe to discard any error indications we've 1877 * received recently. This isn't quite right, but close enough 1878 * for now (a route might have failed after we sent a segment, 1879 * and the return path might not be symmetrical). 1880 */ 1881 tp->t_softerror = 0; 1882 } 1883 1884 /* 1885 * Determine a reasonable value for maxseg size. 1886 * If the route is known, check route for mtu. 1887 * If none, use an mss that can be handled on the outgoing 1888 * interface without forcing IP to fragment; if bigger than 1889 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1890 * to utilize large mbufs. If no route is found, route has no mtu, 1891 * or the destination isn't local, use a default, hopefully conservative 1892 * size (usually 512 or the default IP max size, but no more than the mtu 1893 * of the interface), as we can't discover anything about intervening 1894 * gateways or networks. We also initialize the congestion/slow start 1895 * window to be a single segment if the destination isn't local. 1896 * While looking at the routing entry, we also initialize other path-dependent 1897 * parameters from pre-set or cached values in the routing entry. 1898 * 1899 * Also take into account the space needed for options that we 1900 * send regularly. Make maxseg shorter by that amount to assure 1901 * that we can send maxseg amount of data even when the options 1902 * are present. Store the upper limit of the length of options plus 1903 * data in maxopd. 1904 * 1905 * NOTE that this routine is only called when we process an incoming 1906 * segment, for outgoing segments only tcp_mssopt is called. 1907 * 1908 * In case of T/TCP, we call this routine during implicit connection 1909 * setup as well (offer = -1), to initialize maxseg from the cached 1910 * MSS of our peer. 1911 */ 1912 void 1913 tcp_mss(tp, offer) 1914 struct tcpcb *tp; 1915 int offer; 1916 { 1917 register struct rtentry *rt; 1918 struct ifnet *ifp; 1919 register int rtt, mss; 1920 u_long bufsize; 1921 struct inpcb *inp; 1922 struct socket *so; 1923 struct rmxp_tao *taop; 1924 int origoffer = offer; 1925 1926 inp = tp->t_inpcb; 1927 if ((rt = tcp_rtlookup(inp)) == NULL) { 1928 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1929 return; 1930 } 1931 ifp = rt->rt_ifp; 1932 so = inp->inp_socket; 1933 1934 taop = rmx_taop(rt->rt_rmx); 1935 /* 1936 * Offer == -1 means that we didn't receive SYN yet, 1937 * use cached value in that case; 1938 */ 1939 if (offer == -1) 1940 offer = taop->tao_mssopt; 1941 /* 1942 * Offer == 0 means that there was no MSS on the SYN segment, 1943 * in this case we use tcp_mssdflt. 1944 */ 1945 if (offer == 0) 1946 offer = tcp_mssdflt; 1947 else 1948 /* 1949 * Sanity check: make sure that maxopd will be large 1950 * enough to allow some data on segments even is the 1951 * all the option space is used (40bytes). Otherwise 1952 * funny things may happen in tcp_output. 1953 */ 1954 offer = max(offer, 64); 1955 taop->tao_mssopt = offer; 1956 1957 /* 1958 * While we're here, check if there's an initial rtt 1959 * or rttvar. Convert from the route-table units 1960 * to scaled multiples of the slow timeout timer. 1961 */ 1962 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1963 /* 1964 * XXX the lock bit for RTT indicates that the value 1965 * is also a minimum value; this is subject to time. 1966 */ 1967 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1968 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1969 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1970 tcpstat.tcps_usedrtt++; 1971 if (rt->rt_rmx.rmx_rttvar) { 1972 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1973 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1974 tcpstat.tcps_usedrttvar++; 1975 } else { 1976 /* default variation is +- 1 rtt */ 1977 tp->t_rttvar = 1978 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1979 } 1980 TCPT_RANGESET(tp->t_rxtcur, 1981 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1982 tp->t_rttmin, TCPTV_REXMTMAX); 1983 } 1984 /* 1985 * if there's an mtu associated with the route, use it 1986 */ 1987 if (rt->rt_rmx.rmx_mtu) 1988 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1989 else 1990 { 1991 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1992 if (!in_localaddr(inp->inp_faddr)) 1993 mss = min(mss, tcp_mssdflt); 1994 } 1995 mss = min(mss, offer); 1996 /* 1997 * maxopd stores the maximum length of data AND options 1998 * in a segment; maxseg is the amount of data in a normal 1999 * segment. We need to store this value (maxopd) apart 2000 * from maxseg, because now every segment carries options 2001 * and thus we normally have somewhat less data in segments. 2002 */ 2003 tp->t_maxopd = mss; 2004 2005 /* 2006 * In case of T/TCP, origoffer==-1 indicates, that no segments 2007 * were received yet. In this case we just guess, otherwise 2008 * we do the same as before T/TCP. 2009 */ 2010 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2011 (origoffer == -1 || 2012 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2013 mss -= TCPOLEN_TSTAMP_APPA; 2014 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2015 (origoffer == -1 || 2016 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2017 mss -= TCPOLEN_CC_APPA; 2018 2019 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2020 if (mss > MCLBYTES) 2021 mss &= ~(MCLBYTES-1); 2022 #else 2023 if (mss > MCLBYTES) 2024 mss = mss / MCLBYTES * MCLBYTES; 2025 #endif 2026 /* 2027 * If there's a pipesize, change the socket buffer 2028 * to that size. Make the socket buffers an integral 2029 * number of mss units; if the mss is larger than 2030 * the socket buffer, decrease the mss. 2031 */ 2032 #ifdef RTV_SPIPE 2033 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2034 #endif 2035 bufsize = so->so_snd.sb_hiwat; 2036 if (bufsize < mss) 2037 mss = bufsize; 2038 else { 2039 bufsize = roundup(bufsize, mss); 2040 if (bufsize > sb_max) 2041 bufsize = sb_max; 2042 (void)sbreserve(&so->so_snd, bufsize); 2043 } 2044 tp->t_maxseg = mss; 2045 2046 #ifdef RTV_RPIPE 2047 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2048 #endif 2049 bufsize = so->so_rcv.sb_hiwat; 2050 if (bufsize > mss) { 2051 bufsize = roundup(bufsize, mss); 2052 if (bufsize > sb_max) 2053 bufsize = sb_max; 2054 (void)sbreserve(&so->so_rcv, bufsize); 2055 } 2056 /* 2057 * Don't force slow-start on local network. 2058 */ 2059 if (!in_localaddr(inp->inp_faddr)) 2060 tp->snd_cwnd = mss; 2061 2062 if (rt->rt_rmx.rmx_ssthresh) { 2063 /* 2064 * There's some sort of gateway or interface 2065 * buffer limit on the path. Use this to set 2066 * the slow start threshhold, but set the 2067 * threshold to no less than 2*mss. 2068 */ 2069 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2070 tcpstat.tcps_usedssthresh++; 2071 } 2072 } 2073 2074 /* 2075 * Determine the MSS option to send on an outgoing SYN. 2076 */ 2077 int 2078 tcp_mssopt(tp) 2079 struct tcpcb *tp; 2080 { 2081 struct rtentry *rt; 2082 2083 rt = tcp_rtlookup(tp->t_inpcb); 2084 if (rt == NULL) 2085 return tcp_mssdflt; 2086 2087 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2088 } 2089 #endif /* TUBA_INCLUDE */ 2090