1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * From: @(#)tcp_input.c 8.5 (Berkeley) 4/10/94 34 * $Id: tcp_input.c,v 1.14 1995/02/16 00:55:39 wollman Exp $ 35 */ 36 37 #ifndef TUBA_INCLUDE 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/protosw.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.h> 45 #include <sys/errno.h> 46 47 #include <net/if.h> 48 #include <net/route.h> 49 50 #include <netinet/in.h> 51 #include <netinet/in_systm.h> 52 #include <netinet/ip.h> 53 #include <netinet/in_pcb.h> 54 #include <netinet/ip_var.h> 55 #include <netinet/tcp.h> 56 #include <netinet/tcp_fsm.h> 57 #include <netinet/tcp_seq.h> 58 #include <netinet/tcp_timer.h> 59 #include <netinet/tcp_var.h> 60 #include <netinet/tcpip.h> 61 #ifdef TCPDEBUG 62 #include <netinet/tcp_debug.h> 63 struct tcpiphdr tcp_saveti; 64 #endif 65 66 int tcprexmtthresh = 3; 67 struct inpcb *tcp_last_inpcb = &tcb; 68 tcp_seq tcp_iss; 69 tcp_cc tcp_ccgen; 70 struct inpcb tcb; 71 struct tcpstat tcpstat; 72 u_long tcp_now; 73 74 #endif /* TUBA_INCLUDE */ 75 76 /* 77 * Insert segment ti into reassembly queue of tcp with 78 * control block tp. Return TH_FIN if reassembly now includes 79 * a segment with FIN. The macro form does the common case inline 80 * (segment is the next to be received on an established connection, 81 * and the queue is empty), avoiding linkage into and removal 82 * from the queue and repetition of various conversions. 83 * Set DELACK for segments received in order, but ack immediately 84 * when segments are out of order (so fast retransmit can work). 85 */ 86 #define TCP_REASS(tp, ti, m, so, flags) { \ 87 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 88 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 89 (tp)->t_state == TCPS_ESTABLISHED) { \ 90 tp->t_flags |= TF_DELACK; \ 91 (tp)->rcv_nxt += (ti)->ti_len; \ 92 flags = (ti)->ti_flags & TH_FIN; \ 93 tcpstat.tcps_rcvpack++;\ 94 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 95 sbappend(&(so)->so_rcv, (m)); \ 96 sorwakeup(so); \ 97 } else { \ 98 (flags) = tcp_reass((tp), (ti), (m)); \ 99 tp->t_flags |= TF_ACKNOW; \ 100 } \ 101 } 102 #ifndef TUBA_INCLUDE 103 104 int 105 tcp_reass(tp, ti, m) 106 register struct tcpcb *tp; 107 register struct tcpiphdr *ti; 108 struct mbuf *m; 109 { 110 register struct tcpiphdr *q; 111 struct socket *so = tp->t_inpcb->inp_socket; 112 int flags; 113 114 /* 115 * Call with ti==0 after become established to 116 * force pre-ESTABLISHED data up to user socket. 117 */ 118 if (ti == 0) 119 goto present; 120 121 /* 122 * Find a segment which begins after this one does. 123 */ 124 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 125 q = (struct tcpiphdr *)q->ti_next) 126 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 127 break; 128 129 /* 130 * If there is a preceding segment, it may provide some of 131 * our data already. If so, drop the data from the incoming 132 * segment. If it provides all of our data, drop us. 133 */ 134 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 135 register int i; 136 q = (struct tcpiphdr *)q->ti_prev; 137 /* conversion to int (in i) handles seq wraparound */ 138 i = q->ti_seq + q->ti_len - ti->ti_seq; 139 if (i > 0) { 140 if (i >= ti->ti_len) { 141 tcpstat.tcps_rcvduppack++; 142 tcpstat.tcps_rcvdupbyte += ti->ti_len; 143 m_freem(m); 144 /* 145 * Try to present any queued data 146 * at the left window edge to the user. 147 * This is needed after the 3-WHS 148 * completes. 149 */ 150 goto present; /* ??? */ 151 } 152 m_adj(m, i); 153 ti->ti_len -= i; 154 ti->ti_seq += i; 155 } 156 q = (struct tcpiphdr *)(q->ti_next); 157 } 158 tcpstat.tcps_rcvoopack++; 159 tcpstat.tcps_rcvoobyte += ti->ti_len; 160 REASS_MBUF(ti) = m; /* XXX */ 161 162 /* 163 * While we overlap succeeding segments trim them or, 164 * if they are completely covered, dequeue them. 165 */ 166 while (q != (struct tcpiphdr *)tp) { 167 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 168 if (i <= 0) 169 break; 170 if (i < q->ti_len) { 171 q->ti_seq += i; 172 q->ti_len -= i; 173 m_adj(REASS_MBUF(q), i); 174 break; 175 } 176 q = (struct tcpiphdr *)q->ti_next; 177 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 178 remque(q->ti_prev); 179 m_freem(m); 180 } 181 182 /* 183 * Stick new segment in its place. 184 */ 185 insque(ti, q->ti_prev); 186 187 present: 188 /* 189 * Present data to user, advancing rcv_nxt through 190 * completed sequence space. 191 */ 192 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 193 return (0); 194 ti = tp->seg_next; 195 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 196 return (0); 197 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 198 return (0); 199 do { 200 tp->rcv_nxt += ti->ti_len; 201 flags = ti->ti_flags & TH_FIN; 202 remque(ti); 203 m = REASS_MBUF(ti); 204 ti = (struct tcpiphdr *)ti->ti_next; 205 if (so->so_state & SS_CANTRCVMORE) 206 m_freem(m); 207 else 208 sbappend(&so->so_rcv, m); 209 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 210 sorwakeup(so); 211 return (flags); 212 } 213 214 /* 215 * TCP input routine, follows pages 65-76 of the 216 * protocol specification dated September, 1981 very closely. 217 */ 218 void 219 tcp_input(m, iphlen) 220 register struct mbuf *m; 221 int iphlen; 222 { 223 register struct tcpiphdr *ti; 224 register struct inpcb *inp; 225 caddr_t optp = NULL; 226 int optlen = 0; 227 int len, tlen, off; 228 register struct tcpcb *tp = 0; 229 register int tiflags; 230 struct socket *so = 0; 231 int todrop, acked, ourfinisacked, needoutput = 0; 232 struct in_addr laddr; 233 int dropsocket = 0; 234 int iss = 0; 235 u_long tiwin; 236 struct tcpopt to; /* options in this segment */ 237 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 238 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 239 #ifdef TCPDEBUG 240 short ostate = 0; 241 #endif 242 243 bzero((char *)&to, sizeof(to)); 244 245 tcpstat.tcps_rcvtotal++; 246 /* 247 * Get IP and TCP header together in first mbuf. 248 * Note: IP leaves IP header in first mbuf. 249 */ 250 ti = mtod(m, struct tcpiphdr *); 251 if (iphlen > sizeof (struct ip)) 252 ip_stripoptions(m, (struct mbuf *)0); 253 if (m->m_len < sizeof (struct tcpiphdr)) { 254 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 255 tcpstat.tcps_rcvshort++; 256 return; 257 } 258 ti = mtod(m, struct tcpiphdr *); 259 } 260 261 /* 262 * Checksum extended TCP header and data. 263 */ 264 tlen = ((struct ip *)ti)->ip_len; 265 len = sizeof (struct ip) + tlen; 266 ti->ti_next = ti->ti_prev = 0; 267 ti->ti_x1 = 0; 268 ti->ti_len = (u_short)tlen; 269 HTONS(ti->ti_len); 270 ti->ti_sum = in_cksum(m, len); 271 if (ti->ti_sum) { 272 tcpstat.tcps_rcvbadsum++; 273 goto drop; 274 } 275 #endif /* TUBA_INCLUDE */ 276 277 /* 278 * Check that TCP offset makes sense, 279 * pull out TCP options and adjust length. XXX 280 */ 281 off = ti->ti_off << 2; 282 if (off < sizeof (struct tcphdr) || off > tlen) { 283 tcpstat.tcps_rcvbadoff++; 284 goto drop; 285 } 286 tlen -= off; 287 ti->ti_len = tlen; 288 if (off > sizeof (struct tcphdr)) { 289 if (m->m_len < sizeof(struct ip) + off) { 290 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 291 tcpstat.tcps_rcvshort++; 292 return; 293 } 294 ti = mtod(m, struct tcpiphdr *); 295 } 296 optlen = off - sizeof (struct tcphdr); 297 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 298 /* 299 * Do quick retrieval of timestamp options ("options 300 * prediction?"). If timestamp is the only option and it's 301 * formatted as recommended in RFC 1323 appendix A, we 302 * quickly get the values now and not bother calling 303 * tcp_dooptions(), etc. 304 */ 305 if ((optlen == TCPOLEN_TSTAMP_APPA || 306 (optlen > TCPOLEN_TSTAMP_APPA && 307 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 308 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 309 (ti->ti_flags & TH_SYN) == 0) { 310 to.to_flag |= TOF_TS; 311 to.to_tsval = ntohl(*(u_long *)(optp + 4)); 312 to.to_tsecr = ntohl(*(u_long *)(optp + 8)); 313 optp = NULL; /* we've parsed the options */ 314 } 315 } 316 tiflags = ti->ti_flags; 317 318 /* 319 * Convert TCP protocol specific fields to host format. 320 */ 321 NTOHL(ti->ti_seq); 322 NTOHL(ti->ti_ack); 323 NTOHS(ti->ti_win); 324 NTOHS(ti->ti_urp); 325 326 /* 327 * Locate pcb for segment. 328 */ 329 findpcb: 330 inp = tcp_last_inpcb; 331 if (inp->inp_lport != ti->ti_dport || 332 inp->inp_fport != ti->ti_sport || 333 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 334 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 335 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 336 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 337 if (inp) 338 tcp_last_inpcb = inp; 339 ++tcpstat.tcps_pcbcachemiss; 340 } 341 342 /* 343 * If the state is CLOSED (i.e., TCB does not exist) then 344 * all data in the incoming segment is discarded. 345 * If the TCB exists but is in CLOSED state, it is embryonic, 346 * but should either do a listen or a connect soon. 347 */ 348 if (inp == 0) 349 goto dropwithreset; 350 tp = intotcpcb(inp); 351 if (tp == 0) 352 goto dropwithreset; 353 if (tp->t_state == TCPS_CLOSED) 354 goto drop; 355 356 /* Unscale the window into a 32-bit value. */ 357 if ((tiflags & TH_SYN) == 0) 358 tiwin = ti->ti_win << tp->snd_scale; 359 else 360 tiwin = ti->ti_win; 361 362 so = inp->inp_socket; 363 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 364 #ifdef TCPDEBUG 365 if (so->so_options & SO_DEBUG) { 366 ostate = tp->t_state; 367 tcp_saveti = *ti; 368 } 369 #endif 370 if (so->so_options & SO_ACCEPTCONN) { 371 register struct tcpcb *tp0 = tp; 372 so = sonewconn(so, 0); 373 if (so == 0) 374 goto drop; 375 /* 376 * This is ugly, but .... 377 * 378 * Mark socket as temporary until we're 379 * committed to keeping it. The code at 380 * ``drop'' and ``dropwithreset'' check the 381 * flag dropsocket to see if the temporary 382 * socket created here should be discarded. 383 * We mark the socket as discardable until 384 * we're committed to it below in TCPS_LISTEN. 385 */ 386 dropsocket++; 387 inp = (struct inpcb *)so->so_pcb; 388 inp->inp_laddr = ti->ti_dst; 389 inp->inp_lport = ti->ti_dport; 390 #if BSD>=43 391 inp->inp_options = ip_srcroute(); 392 #endif 393 tp = intotcpcb(inp); 394 tp->t_state = TCPS_LISTEN; 395 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 396 397 /* Compute proper scaling value from buffer space */ 398 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 399 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 400 tp->request_r_scale++; 401 } 402 } 403 404 /* 405 * Segment received on connection. 406 * Reset idle time and keep-alive timer. 407 */ 408 tp->t_idle = 0; 409 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 410 411 /* 412 * Process options if not in LISTEN state, 413 * else do it below (after getting remote address). 414 */ 415 if (optp && tp->t_state != TCPS_LISTEN) 416 tcp_dooptions(tp, optp, optlen, ti, 417 &to); 418 419 /* 420 * Header prediction: check for the two common cases 421 * of a uni-directional data xfer. If the packet has 422 * no control flags, is in-sequence, the window didn't 423 * change and we're not retransmitting, it's a 424 * candidate. If the length is zero and the ack moved 425 * forward, we're the sender side of the xfer. Just 426 * free the data acked & wake any higher level process 427 * that was blocked waiting for space. If the length 428 * is non-zero and the ack didn't move, we're the 429 * receiver side. If we're getting packets in-order 430 * (the reassembly queue is empty), add the data to 431 * the socket buffer and note that we need a delayed ack. 432 * Make sure that the hidden state-flags are also off. 433 * Since we check for TCPS_ESTABLISHED above, it can only 434 * be TH_NEEDSYN. 435 */ 436 if (tp->t_state == TCPS_ESTABLISHED && 437 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 438 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 439 ((to.to_flag & TOF_TS) == 0 || 440 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 441 /* 442 * Using the CC option is compulsory if once started: 443 * the segment is OK if no T/TCP was negotiated or 444 * if the segment has a CC option equal to CCrecv 445 */ 446 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 447 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 448 ti->ti_seq == tp->rcv_nxt && 449 tiwin && tiwin == tp->snd_wnd && 450 tp->snd_nxt == tp->snd_max) { 451 452 /* 453 * If last ACK falls within this segment's sequence numbers, 454 * record the timestamp. 455 * NOTE that the test is modified according to the latest 456 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 457 */ 458 if ((to.to_flag & TOF_TS) != 0 && 459 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 460 tp->ts_recent_age = tcp_now; 461 tp->ts_recent = to.to_tsval; 462 } 463 464 if (ti->ti_len == 0) { 465 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 466 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 467 tp->snd_cwnd >= tp->snd_wnd) { 468 /* 469 * this is a pure ack for outstanding data. 470 */ 471 ++tcpstat.tcps_predack; 472 if ((to.to_flag & TOF_TS) != 0) 473 tcp_xmit_timer(tp, 474 tcp_now - to.to_tsecr + 1); 475 else if (tp->t_rtt && 476 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 477 tcp_xmit_timer(tp, tp->t_rtt); 478 acked = ti->ti_ack - tp->snd_una; 479 tcpstat.tcps_rcvackpack++; 480 tcpstat.tcps_rcvackbyte += acked; 481 sbdrop(&so->so_snd, acked); 482 tp->snd_una = ti->ti_ack; 483 m_freem(m); 484 485 /* 486 * If all outstanding data are acked, stop 487 * retransmit timer, otherwise restart timer 488 * using current (possibly backed-off) value. 489 * If process is waiting for space, 490 * wakeup/selwakeup/signal. If data 491 * are ready to send, let tcp_output 492 * decide between more output or persist. 493 */ 494 if (tp->snd_una == tp->snd_max) 495 tp->t_timer[TCPT_REXMT] = 0; 496 else if (tp->t_timer[TCPT_PERSIST] == 0) 497 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 498 499 if (so->so_snd.sb_flags & SB_NOTIFY) 500 sowwakeup(so); 501 if (so->so_snd.sb_cc) 502 (void) tcp_output(tp); 503 return; 504 } 505 } else if (ti->ti_ack == tp->snd_una && 506 tp->seg_next == (struct tcpiphdr *)tp && 507 ti->ti_len <= sbspace(&so->so_rcv)) { 508 /* 509 * this is a pure, in-sequence data packet 510 * with nothing on the reassembly queue and 511 * we have enough buffer space to take it. 512 */ 513 ++tcpstat.tcps_preddat; 514 tp->rcv_nxt += ti->ti_len; 515 tcpstat.tcps_rcvpack++; 516 tcpstat.tcps_rcvbyte += ti->ti_len; 517 /* 518 * Drop TCP, IP headers and TCP options then add data 519 * to socket buffer. 520 */ 521 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 522 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 523 sbappend(&so->so_rcv, m); 524 sorwakeup(so); 525 tp->t_flags |= TF_DELACK; 526 return; 527 } 528 } 529 530 /* 531 * Drop TCP, IP headers and TCP options. 532 */ 533 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 534 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 535 536 /* 537 * Calculate amount of space in receive window, 538 * and then do TCP input processing. 539 * Receive window is amount of space in rcv queue, 540 * but not less than advertised window. 541 */ 542 { int win; 543 544 win = sbspace(&so->so_rcv); 545 if (win < 0) 546 win = 0; 547 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 548 } 549 550 switch (tp->t_state) { 551 552 /* 553 * If the state is LISTEN then ignore segment if it contains an RST. 554 * If the segment contains an ACK then it is bad and send a RST. 555 * If it does not contain a SYN then it is not interesting; drop it. 556 * Don't bother responding if the destination was a broadcast. 557 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 558 * tp->iss, and send a segment: 559 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 560 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 561 * Fill in remote peer address fields if not previously specified. 562 * Enter SYN_RECEIVED state, and process any other fields of this 563 * segment in this state. 564 */ 565 case TCPS_LISTEN: { 566 struct mbuf *am; 567 register struct sockaddr_in *sin; 568 569 if (tiflags & TH_RST) 570 goto drop; 571 if (tiflags & TH_ACK) 572 goto dropwithreset; 573 if ((tiflags & TH_SYN) == 0) 574 goto drop; 575 /* 576 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 577 * in_broadcast() should never return true on a received 578 * packet with M_BCAST not set. 579 */ 580 if (m->m_flags & (M_BCAST|M_MCAST) || 581 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 582 goto drop; 583 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 584 if (am == NULL) 585 goto drop; 586 am->m_len = sizeof (struct sockaddr_in); 587 sin = mtod(am, struct sockaddr_in *); 588 sin->sin_family = AF_INET; 589 sin->sin_len = sizeof(*sin); 590 sin->sin_addr = ti->ti_src; 591 sin->sin_port = ti->ti_sport; 592 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 593 laddr = inp->inp_laddr; 594 if (inp->inp_laddr.s_addr == INADDR_ANY) 595 inp->inp_laddr = ti->ti_dst; 596 if (in_pcbconnect(inp, am)) { 597 inp->inp_laddr = laddr; 598 (void) m_free(am); 599 goto drop; 600 } 601 (void) m_free(am); 602 tp->t_template = tcp_template(tp); 603 if (tp->t_template == 0) { 604 tp = tcp_drop(tp, ENOBUFS); 605 dropsocket = 0; /* socket is already gone */ 606 goto drop; 607 } 608 if ((taop = tcp_gettaocache(inp)) == NULL) { 609 taop = &tao_noncached; 610 bzero(taop, sizeof(*taop)); 611 } 612 if (optp) 613 tcp_dooptions(tp, optp, optlen, ti, 614 &to); 615 if (iss) 616 tp->iss = iss; 617 else 618 tp->iss = tcp_iss; 619 tcp_iss += TCP_ISSINCR/2; 620 tp->irs = ti->ti_seq; 621 tcp_sendseqinit(tp); 622 tcp_rcvseqinit(tp); 623 /* 624 * Initialization of the tcpcb for transaction; 625 * set SND.WND = SEG.WND, 626 * initialize CCsend and CCrecv. 627 */ 628 tp->snd_wnd = tiwin; /* initial send-window */ 629 tp->cc_send = CC_INC(tcp_ccgen); 630 tp->cc_recv = to.to_cc; 631 /* 632 * Perform TAO test on incoming CC (SEG.CC) option, if any. 633 * - compare SEG.CC against cached CC from the same host, 634 * if any. 635 * - if SEG.CC > chached value, SYN must be new and is accepted 636 * immediately: save new CC in the cache, mark the socket 637 * connected, enter ESTABLISHED state, turn on flag to 638 * send a SYN in the next segment. 639 * A virtual advertised window is set in rcv_adv to 640 * initialize SWS prevention. Then enter normal segment 641 * processing: drop SYN, process data and FIN. 642 * - otherwise do a normal 3-way handshake. 643 */ 644 if ((to.to_flag & TOF_CC) != 0) { 645 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 646 taop->tao_cc = to.to_cc; 647 tp->t_state = TCPS_ESTABLISHED; 648 649 /* 650 * If there is a FIN, or if there is data and the 651 * connection is local, then delay SYN,ACK(SYN) in 652 * the hope of piggy-backing it on a response 653 * segment. Otherwise must send ACK now in case 654 * the other side is slow starting. 655 */ 656 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 657 in_localaddr(inp->inp_faddr))) 658 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 659 else 660 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 661 tp->rcv_adv += tp->rcv_wnd; 662 tcpstat.tcps_connects++; 663 soisconnected(so); 664 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 665 dropsocket = 0; /* committed to socket */ 666 tcpstat.tcps_accepts++; 667 goto trimthenstep6; 668 } 669 /* else do standard 3-way handshake */ 670 } else { 671 /* 672 * No CC option, but maybe CC.NEW: 673 * invalidate cached value. 674 */ 675 taop->tao_cc = 0; 676 } 677 /* 678 * TAO test failed or there was no CC option, 679 * do a standard 3-way handshake. 680 */ 681 tp->t_flags |= TF_ACKNOW; 682 tp->t_state = TCPS_SYN_RECEIVED; 683 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 684 dropsocket = 0; /* committed to socket */ 685 tcpstat.tcps_accepts++; 686 goto trimthenstep6; 687 } 688 689 /* 690 * If the state is SYN_SENT: 691 * if seg contains an ACK, but not for our SYN, drop the input. 692 * if seg contains a RST, then drop the connection. 693 * if seg does not contain SYN, then drop it. 694 * Otherwise this is an acceptable SYN segment 695 * initialize tp->rcv_nxt and tp->irs 696 * if seg contains ack then advance tp->snd_una 697 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 698 * arrange for segment to be acked (eventually) 699 * continue processing rest of data/controls, beginning with URG 700 */ 701 case TCPS_SYN_SENT: 702 if ((taop = tcp_gettaocache(inp)) == NULL) { 703 taop = &tao_noncached; 704 bzero(taop, sizeof(*taop)); 705 } 706 707 if ((tiflags & TH_ACK) && 708 (SEQ_LEQ(ti->ti_ack, tp->iss) || 709 SEQ_GT(ti->ti_ack, tp->snd_max))) { 710 /* 711 * If we have a cached CCsent for the remote host, 712 * hence we haven't just crashed and restarted, 713 * do not send a RST. This may be a retransmission 714 * from the other side after our earlier ACK was lost. 715 * Our new SYN, when it arrives, will serve as the 716 * needed ACK. 717 */ 718 if (taop->tao_ccsent != 0) 719 goto drop; 720 else 721 goto dropwithreset; 722 } 723 if (tiflags & TH_RST) { 724 if (tiflags & TH_ACK) 725 tp = tcp_drop(tp, ECONNREFUSED); 726 goto drop; 727 } 728 if ((tiflags & TH_SYN) == 0) 729 goto drop; 730 tp->snd_wnd = ti->ti_win; /* initial send window */ 731 tp->cc_recv = to.to_cc; /* foreign CC */ 732 733 tp->irs = ti->ti_seq; 734 tcp_rcvseqinit(tp); 735 if (tiflags & TH_ACK && SEQ_GT(ti->ti_ack, tp->iss)) { 736 tcpstat.tcps_connects++; 737 soisconnected(so); 738 /* Do window scaling on this connection? */ 739 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 740 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 741 tp->snd_scale = tp->requested_s_scale; 742 tp->rcv_scale = tp->request_r_scale; 743 } 744 /* 745 * Our SYN was acked. If segment contains CC.ECHO 746 * option, check it to make sure this segment really 747 * matches our SYN. If not, just drop it as old 748 * duplicate, but send an RST if we're still playing 749 * by the old rules. 750 */ 751 if ((to.to_flag & TOF_CCECHO) && 752 tp->cc_send != to.to_ccecho) { 753 if (taop->tao_ccsent != 0) 754 goto drop; 755 else 756 goto dropwithreset; 757 } 758 /* Segment is acceptable, update cache if undefined. */ 759 if (taop->tao_ccsent == 0) 760 taop->tao_ccsent = to.to_ccecho; 761 762 tp->rcv_adv += tp->rcv_wnd; 763 tp->snd_una++; /* SYN is acked */ 764 /* 765 * If there's data, delay ACK; if there's also a FIN 766 * ACKNOW will be turned on later. 767 */ 768 if (ti->ti_len != 0) 769 tp->t_flags |= TF_DELACK; 770 else 771 tp->t_flags |= TF_ACKNOW; 772 /* 773 * Received <SYN,ACK> in SYN_SENT[*] state. 774 * Transitions: 775 * SYN_SENT --> ESTABLISHED 776 * SYN_SENT* --> FIN_WAIT_1 777 */ 778 if (tp->t_flags & TF_NEEDFIN) { 779 tp->t_state = TCPS_FIN_WAIT_1; 780 tp->t_flags &= ~TF_NEEDFIN; 781 tiflags &= ~TH_SYN; 782 } else 783 tp->t_state = TCPS_ESTABLISHED; 784 785 } else { 786 /* 787 * Received initial SYN in SYN-SENT[*] state => simul- 788 * taneous open. If segment contains CC option and there is 789 * a cached CC, apply TAO test; if it succeeds, connection is 790 * half-synchronized. Otherwise, do 3-way handshake: 791 * SYN-SENT -> SYN-RECEIVED 792 * SYN-SENT* -> SYN-RECEIVED* 793 * If there was no CC option, clear cached CC value. 794 */ 795 tp->t_flags |= TF_ACKNOW; 796 tp->t_timer[TCPT_REXMT] = 0; 797 if (to.to_flag & TOF_CC) { 798 if (taop->tao_cc != 0 && 799 CC_GT(to.to_cc, taop->tao_cc)) { 800 /* 801 * update cache and make transition: 802 * SYN-SENT -> ESTABLISHED* 803 * SYN-SENT* -> FIN-WAIT-1* 804 */ 805 taop->tao_cc = to.to_cc; 806 if (tp->t_flags & TF_NEEDFIN) { 807 tp->t_state = TCPS_FIN_WAIT_1; 808 tp->t_flags &= ~TF_NEEDFIN; 809 } else 810 tp->t_state = TCPS_ESTABLISHED; 811 tp->t_flags |= TF_NEEDSYN; 812 } else 813 tp->t_state = TCPS_SYN_RECEIVED; 814 } else { 815 /* CC.NEW or no option => invalidate cache */ 816 taop->tao_cc = 0; 817 tp->t_state = TCPS_SYN_RECEIVED; 818 } 819 } 820 821 trimthenstep6: 822 /* 823 * Advance ti->ti_seq to correspond to first data byte. 824 * If data, trim to stay within window, 825 * dropping FIN if necessary. 826 */ 827 ti->ti_seq++; 828 if (ti->ti_len > tp->rcv_wnd) { 829 todrop = ti->ti_len - tp->rcv_wnd; 830 m_adj(m, -todrop); 831 ti->ti_len = tp->rcv_wnd; 832 tiflags &= ~TH_FIN; 833 tcpstat.tcps_rcvpackafterwin++; 834 tcpstat.tcps_rcvbyteafterwin += todrop; 835 } 836 tp->snd_wl1 = ti->ti_seq - 1; 837 tp->rcv_up = ti->ti_seq; 838 /* 839 * Client side of transaction: already sent SYN and data. 840 * If the remote host used T/TCP to validate the SYN, 841 * our data will be ACK'd; if so, enter normal data segment 842 * processing in the middle of step 5, ack processing. 843 * Otherwise, goto step 6. 844 */ 845 if (tiflags & TH_ACK) 846 goto process_ACK; 847 goto step6; 848 /* 849 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 850 * if segment contains a SYN and CC [not CC.NEW] option: 851 * if state == TIME_WAIT and connection duration > MSL, 852 * drop packet and send RST; 853 * 854 * if SEG.CC > CCrecv then is new SYN, and can implicitly 855 * ack the FIN (and data) in retransmission queue. 856 * Complete close and delete TCPCB. Then reprocess 857 * segment, hoping to find new TCPCB in LISTEN state; 858 * 859 * else must be old SYN; drop it. 860 * else do normal processing. 861 */ 862 case TCPS_LAST_ACK: 863 case TCPS_CLOSING: 864 case TCPS_TIME_WAIT: 865 if ((tiflags & TH_SYN) && 866 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 867 if (tp->t_state == TCPS_TIME_WAIT && 868 tp->t_duration > TCPTV_MSL) 869 goto dropwithreset; 870 if (CC_GT(to.to_cc, tp->cc_recv)) { 871 tp = tcp_close(tp); 872 goto findpcb; 873 } 874 else 875 goto drop; 876 } 877 break; /* continue normal processing */ 878 } 879 880 /* 881 * States other than LISTEN or SYN_SENT. 882 * First check timestamp, if present. 883 * Then check the connection count, if present. 884 * Then check that at least some bytes of segment are within 885 * receive window. If segment begins before rcv_nxt, 886 * drop leading data (and SYN); if nothing left, just ack. 887 * 888 * RFC 1323 PAWS: If we have a timestamp reply on this segment 889 * and it's less than ts_recent, drop it. 890 */ 891 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 892 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 893 894 /* Check to see if ts_recent is over 24 days old. */ 895 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 896 /* 897 * Invalidate ts_recent. If this segment updates 898 * ts_recent, the age will be reset later and ts_recent 899 * will get a valid value. If it does not, setting 900 * ts_recent to zero will at least satisfy the 901 * requirement that zero be placed in the timestamp 902 * echo reply when ts_recent isn't valid. The 903 * age isn't reset until we get a valid ts_recent 904 * because we don't want out-of-order segments to be 905 * dropped when ts_recent is old. 906 */ 907 tp->ts_recent = 0; 908 } else { 909 tcpstat.tcps_rcvduppack++; 910 tcpstat.tcps_rcvdupbyte += ti->ti_len; 911 tcpstat.tcps_pawsdrop++; 912 goto dropafterack; 913 } 914 } 915 916 /* 917 * T/TCP mechanism 918 * If T/TCP was negotiated and the segment doesn't have CC, 919 * or if it's CC is wrong then drop the segment. 920 * RST segments do not have to comply with this. 921 */ 922 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 923 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 924 (tiflags & TH_RST) == 0) 925 goto dropafterack; 926 927 todrop = tp->rcv_nxt - ti->ti_seq; 928 if (todrop > 0) { 929 if (tiflags & TH_SYN) { 930 tiflags &= ~TH_SYN; 931 ti->ti_seq++; 932 if (ti->ti_urp > 1) 933 ti->ti_urp--; 934 else 935 tiflags &= ~TH_URG; 936 todrop--; 937 } 938 /* 939 * Following if statement from Stevens, vol. 2, p. 960. 940 */ 941 if (todrop > ti->ti_len 942 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 943 /* 944 * Any valid FIN must be to the left of the window. 945 * At this point the FIN must be a duplicate or out 946 * of sequence; drop it. 947 */ 948 tiflags &= ~TH_FIN; 949 950 /* 951 * Send an ACK to resynchronize and drop any data. 952 * But keep on processing for RST or ACK. 953 */ 954 tp->t_flags |= TF_ACKNOW; 955 todrop = ti->ti_len; 956 tcpstat.tcps_rcvduppack++; 957 tcpstat.tcps_rcvdupbyte += todrop; 958 } else { 959 tcpstat.tcps_rcvpartduppack++; 960 tcpstat.tcps_rcvpartdupbyte += todrop; 961 } 962 m_adj(m, todrop); 963 ti->ti_seq += todrop; 964 ti->ti_len -= todrop; 965 if (ti->ti_urp > todrop) 966 ti->ti_urp -= todrop; 967 else { 968 tiflags &= ~TH_URG; 969 ti->ti_urp = 0; 970 } 971 } 972 973 /* 974 * If new data are received on a connection after the 975 * user processes are gone, then RST the other end. 976 */ 977 if ((so->so_state & SS_NOFDREF) && 978 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 979 tp = tcp_close(tp); 980 tcpstat.tcps_rcvafterclose++; 981 goto dropwithreset; 982 } 983 984 /* 985 * If segment ends after window, drop trailing data 986 * (and PUSH and FIN); if nothing left, just ACK. 987 */ 988 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 989 if (todrop > 0) { 990 tcpstat.tcps_rcvpackafterwin++; 991 if (todrop >= ti->ti_len) { 992 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 993 /* 994 * If a new connection request is received 995 * while in TIME_WAIT, drop the old connection 996 * and start over if the sequence numbers 997 * are above the previous ones. 998 */ 999 if (tiflags & TH_SYN && 1000 tp->t_state == TCPS_TIME_WAIT && 1001 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1002 iss = tp->rcv_nxt + TCP_ISSINCR; 1003 tp = tcp_close(tp); 1004 goto findpcb; 1005 } 1006 /* 1007 * If window is closed can only take segments at 1008 * window edge, and have to drop data and PUSH from 1009 * incoming segments. Continue processing, but 1010 * remember to ack. Otherwise, drop segment 1011 * and ack. 1012 */ 1013 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1014 tp->t_flags |= TF_ACKNOW; 1015 tcpstat.tcps_rcvwinprobe++; 1016 } else 1017 goto dropafterack; 1018 } else 1019 tcpstat.tcps_rcvbyteafterwin += todrop; 1020 m_adj(m, -todrop); 1021 ti->ti_len -= todrop; 1022 tiflags &= ~(TH_PUSH|TH_FIN); 1023 } 1024 1025 /* 1026 * If last ACK falls within this segment's sequence numbers, 1027 * record its timestamp. 1028 * NOTE that the test is modified according to the latest 1029 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1030 */ 1031 if ((to.to_flag & TOF_TS) != 0 && 1032 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1033 tp->ts_recent_age = tcp_now; 1034 tp->ts_recent = to.to_tsval; 1035 } 1036 1037 /* 1038 * If the RST bit is set examine the state: 1039 * SYN_RECEIVED STATE: 1040 * If passive open, return to LISTEN state. 1041 * If active open, inform user that connection was refused. 1042 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1043 * Inform user that connection was reset, and close tcb. 1044 * CLOSING, LAST_ACK, TIME_WAIT STATES 1045 * Close the tcb. 1046 */ 1047 if (tiflags&TH_RST) switch (tp->t_state) { 1048 1049 case TCPS_SYN_RECEIVED: 1050 so->so_error = ECONNREFUSED; 1051 goto close; 1052 1053 case TCPS_ESTABLISHED: 1054 case TCPS_FIN_WAIT_1: 1055 case TCPS_FIN_WAIT_2: 1056 case TCPS_CLOSE_WAIT: 1057 so->so_error = ECONNRESET; 1058 close: 1059 tp->t_state = TCPS_CLOSED; 1060 tcpstat.tcps_drops++; 1061 tp = tcp_close(tp); 1062 goto drop; 1063 1064 case TCPS_CLOSING: 1065 case TCPS_LAST_ACK: 1066 case TCPS_TIME_WAIT: 1067 tp = tcp_close(tp); 1068 goto drop; 1069 } 1070 1071 /* 1072 * If a SYN is in the window, then this is an 1073 * error and we send an RST and drop the connection. 1074 */ 1075 if (tiflags & TH_SYN) { 1076 tp = tcp_drop(tp, ECONNRESET); 1077 goto dropwithreset; 1078 } 1079 1080 /* 1081 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1082 * flag is on (half-synchronized state), then queue data for 1083 * later processing; else drop segment and return. 1084 */ 1085 if ((tiflags & TH_ACK) == 0) { 1086 if (tp->t_state == TCPS_SYN_RECEIVED || 1087 (tp->t_flags & TF_NEEDSYN)) 1088 goto step6; 1089 else 1090 goto drop; 1091 } 1092 1093 /* 1094 * Ack processing. 1095 */ 1096 switch (tp->t_state) { 1097 1098 /* 1099 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1100 * ESTABLISHED state and continue processing, otherwise 1101 * send an RST. 1102 */ 1103 case TCPS_SYN_RECEIVED: 1104 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1105 SEQ_GT(ti->ti_ack, tp->snd_max)) 1106 goto dropwithreset; 1107 1108 tcpstat.tcps_connects++; 1109 soisconnected(so); 1110 /* Do window scaling? */ 1111 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1112 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1113 tp->snd_scale = tp->requested_s_scale; 1114 tp->rcv_scale = tp->request_r_scale; 1115 } 1116 /* 1117 * Upon successful completion of 3-way handshake, 1118 * update cache.CC if it was undefined, pass any queued 1119 * data to the user, and advance state appropriately. 1120 */ 1121 if ((taop = tcp_gettaocache(inp)) != NULL && 1122 taop->tao_cc == 0) 1123 taop->tao_cc = tp->cc_recv; 1124 1125 /* 1126 * Make transitions: 1127 * SYN-RECEIVED -> ESTABLISHED 1128 * SYN-RECEIVED* -> FIN-WAIT-1 1129 */ 1130 if (tp->t_flags & TF_NEEDFIN) { 1131 tp->t_state = TCPS_FIN_WAIT_1; 1132 tp->t_flags &= ~TF_NEEDFIN; 1133 } else 1134 tp->t_state = TCPS_ESTABLISHED; 1135 /* 1136 * If segment contains data or ACK, will call tcp_reass() 1137 * later; if not, do so now to pass queued data to user. 1138 */ 1139 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1140 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1141 (struct mbuf *)0); 1142 tp->snd_wl1 = ti->ti_seq - 1; 1143 /* fall into ... */ 1144 1145 /* 1146 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1147 * ACKs. If the ack is in the range 1148 * tp->snd_una < ti->ti_ack <= tp->snd_max 1149 * then advance tp->snd_una to ti->ti_ack and drop 1150 * data from the retransmission queue. If this ACK reflects 1151 * more up to date window information we update our window information. 1152 */ 1153 case TCPS_ESTABLISHED: 1154 case TCPS_FIN_WAIT_1: 1155 case TCPS_FIN_WAIT_2: 1156 case TCPS_CLOSE_WAIT: 1157 case TCPS_CLOSING: 1158 case TCPS_LAST_ACK: 1159 case TCPS_TIME_WAIT: 1160 1161 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1162 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1163 tcpstat.tcps_rcvdupack++; 1164 /* 1165 * If we have outstanding data (other than 1166 * a window probe), this is a completely 1167 * duplicate ack (ie, window info didn't 1168 * change), the ack is the biggest we've 1169 * seen and we've seen exactly our rexmt 1170 * threshhold of them, assume a packet 1171 * has been dropped and retransmit it. 1172 * Kludge snd_nxt & the congestion 1173 * window so we send only this one 1174 * packet. 1175 * 1176 * We know we're losing at the current 1177 * window size so do congestion avoidance 1178 * (set ssthresh to half the current window 1179 * and pull our congestion window back to 1180 * the new ssthresh). 1181 * 1182 * Dup acks mean that packets have left the 1183 * network (they're now cached at the receiver) 1184 * so bump cwnd by the amount in the receiver 1185 * to keep a constant cwnd packets in the 1186 * network. 1187 */ 1188 if (tp->t_timer[TCPT_REXMT] == 0 || 1189 ti->ti_ack != tp->snd_una) 1190 tp->t_dupacks = 0; 1191 else if (++tp->t_dupacks == tcprexmtthresh) { 1192 tcp_seq onxt = tp->snd_nxt; 1193 u_int win = 1194 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1195 tp->t_maxseg; 1196 1197 if (win < 2) 1198 win = 2; 1199 tp->snd_ssthresh = win * tp->t_maxseg; 1200 tp->t_timer[TCPT_REXMT] = 0; 1201 tp->t_rtt = 0; 1202 tp->snd_nxt = ti->ti_ack; 1203 tp->snd_cwnd = tp->t_maxseg; 1204 (void) tcp_output(tp); 1205 tp->snd_cwnd = tp->snd_ssthresh + 1206 tp->t_maxseg * tp->t_dupacks; 1207 if (SEQ_GT(onxt, tp->snd_nxt)) 1208 tp->snd_nxt = onxt; 1209 goto drop; 1210 } else if (tp->t_dupacks > tcprexmtthresh) { 1211 tp->snd_cwnd += tp->t_maxseg; 1212 (void) tcp_output(tp); 1213 goto drop; 1214 } 1215 } else 1216 tp->t_dupacks = 0; 1217 break; 1218 } 1219 /* 1220 * If the congestion window was inflated to account 1221 * for the other side's cached packets, retract it. 1222 */ 1223 if (tp->t_dupacks > tcprexmtthresh && 1224 tp->snd_cwnd > tp->snd_ssthresh) 1225 tp->snd_cwnd = tp->snd_ssthresh; 1226 tp->t_dupacks = 0; 1227 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1228 tcpstat.tcps_rcvacktoomuch++; 1229 goto dropafterack; 1230 } 1231 /* 1232 * If we reach this point, ACK is not a duplicate, 1233 * i.e., it ACKs something we sent. 1234 */ 1235 if (tp->t_flags & TF_NEEDSYN) { 1236 /* 1237 * T/TCP: Connection was half-synchronized, and our 1238 * SYN has been ACK'd (so connection is now fully 1239 * synchronized). Go to non-starred state and 1240 * increment snd_una for ACK of SYN. 1241 */ 1242 tp->t_flags &= ~TF_NEEDSYN; 1243 tp->snd_una++; 1244 } 1245 1246 process_ACK: 1247 acked = ti->ti_ack - tp->snd_una; 1248 tcpstat.tcps_rcvackpack++; 1249 tcpstat.tcps_rcvackbyte += acked; 1250 1251 /* 1252 * If we have a timestamp reply, update smoothed 1253 * round trip time. If no timestamp is present but 1254 * transmit timer is running and timed sequence 1255 * number was acked, update smoothed round trip time. 1256 * Since we now have an rtt measurement, cancel the 1257 * timer backoff (cf., Phil Karn's retransmit alg.). 1258 * Recompute the initial retransmit timer. 1259 */ 1260 if (to.to_flag & TOF_TS) 1261 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1262 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1263 tcp_xmit_timer(tp,tp->t_rtt); 1264 1265 /* 1266 * If all outstanding data is acked, stop retransmit 1267 * timer and remember to restart (more output or persist). 1268 * If there is more data to be acked, restart retransmit 1269 * timer, using current (possibly backed-off) value. 1270 */ 1271 if (ti->ti_ack == tp->snd_max) { 1272 tp->t_timer[TCPT_REXMT] = 0; 1273 needoutput = 1; 1274 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1275 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1276 1277 /* 1278 * If no data (only SYN) was ACK'd, 1279 * skip rest of ACK processing. 1280 */ 1281 if (acked == 0) 1282 goto step6; 1283 1284 /* 1285 * When new data is acked, open the congestion window. 1286 * If the window gives us less than ssthresh packets 1287 * in flight, open exponentially (maxseg per packet). 1288 * Otherwise open linearly: maxseg per window 1289 * (maxseg^2 / cwnd per packet). 1290 */ 1291 { 1292 register u_int cw = tp->snd_cwnd; 1293 register u_int incr = tp->t_maxseg; 1294 1295 if (cw > tp->snd_ssthresh) 1296 incr = incr * incr / cw; 1297 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1298 } 1299 if (acked > so->so_snd.sb_cc) { 1300 tp->snd_wnd -= so->so_snd.sb_cc; 1301 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1302 ourfinisacked = 1; 1303 } else { 1304 sbdrop(&so->so_snd, acked); 1305 tp->snd_wnd -= acked; 1306 ourfinisacked = 0; 1307 } 1308 if (so->so_snd.sb_flags & SB_NOTIFY) 1309 sowwakeup(so); 1310 tp->snd_una = ti->ti_ack; 1311 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1312 tp->snd_nxt = tp->snd_una; 1313 1314 switch (tp->t_state) { 1315 1316 /* 1317 * In FIN_WAIT_1 STATE in addition to the processing 1318 * for the ESTABLISHED state if our FIN is now acknowledged 1319 * then enter FIN_WAIT_2. 1320 */ 1321 case TCPS_FIN_WAIT_1: 1322 if (ourfinisacked) { 1323 /* 1324 * If we can't receive any more 1325 * data, then closing user can proceed. 1326 * Starting the timer is contrary to the 1327 * specification, but if we don't get a FIN 1328 * we'll hang forever. 1329 */ 1330 if (so->so_state & SS_CANTRCVMORE) { 1331 soisdisconnected(so); 1332 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1333 } 1334 tp->t_state = TCPS_FIN_WAIT_2; 1335 } 1336 break; 1337 1338 /* 1339 * In CLOSING STATE in addition to the processing for 1340 * the ESTABLISHED state if the ACK acknowledges our FIN 1341 * then enter the TIME-WAIT state, otherwise ignore 1342 * the segment. 1343 */ 1344 case TCPS_CLOSING: 1345 if (ourfinisacked) { 1346 tp->t_state = TCPS_TIME_WAIT; 1347 tcp_canceltimers(tp); 1348 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1349 if (tp->cc_recv != 0 && 1350 tp->t_duration < TCPTV_MSL) 1351 tp->t_timer[TCPT_2MSL] = 1352 tp->t_rxtcur * TCPTV_TWTRUNC; 1353 else 1354 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1355 soisdisconnected(so); 1356 } 1357 break; 1358 1359 /* 1360 * In LAST_ACK, we may still be waiting for data to drain 1361 * and/or to be acked, as well as for the ack of our FIN. 1362 * If our FIN is now acknowledged, delete the TCB, 1363 * enter the closed state and return. 1364 */ 1365 case TCPS_LAST_ACK: 1366 if (ourfinisacked) { 1367 tp = tcp_close(tp); 1368 goto drop; 1369 } 1370 break; 1371 1372 /* 1373 * In TIME_WAIT state the only thing that should arrive 1374 * is a retransmission of the remote FIN. Acknowledge 1375 * it and restart the finack timer. 1376 */ 1377 case TCPS_TIME_WAIT: 1378 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1379 goto dropafterack; 1380 } 1381 } 1382 1383 step6: 1384 /* 1385 * Update window information. 1386 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1387 */ 1388 if ((tiflags & TH_ACK) && 1389 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1390 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1391 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1392 /* keep track of pure window updates */ 1393 if (ti->ti_len == 0 && 1394 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1395 tcpstat.tcps_rcvwinupd++; 1396 tp->snd_wnd = tiwin; 1397 tp->snd_wl1 = ti->ti_seq; 1398 tp->snd_wl2 = ti->ti_ack; 1399 if (tp->snd_wnd > tp->max_sndwnd) 1400 tp->max_sndwnd = tp->snd_wnd; 1401 needoutput = 1; 1402 } 1403 1404 /* 1405 * Process segments with URG. 1406 */ 1407 if ((tiflags & TH_URG) && ti->ti_urp && 1408 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1409 /* 1410 * This is a kludge, but if we receive and accept 1411 * random urgent pointers, we'll crash in 1412 * soreceive. It's hard to imagine someone 1413 * actually wanting to send this much urgent data. 1414 */ 1415 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1416 ti->ti_urp = 0; /* XXX */ 1417 tiflags &= ~TH_URG; /* XXX */ 1418 goto dodata; /* XXX */ 1419 } 1420 /* 1421 * If this segment advances the known urgent pointer, 1422 * then mark the data stream. This should not happen 1423 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1424 * a FIN has been received from the remote side. 1425 * In these states we ignore the URG. 1426 * 1427 * According to RFC961 (Assigned Protocols), 1428 * the urgent pointer points to the last octet 1429 * of urgent data. We continue, however, 1430 * to consider it to indicate the first octet 1431 * of data past the urgent section as the original 1432 * spec states (in one of two places). 1433 */ 1434 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1435 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1436 so->so_oobmark = so->so_rcv.sb_cc + 1437 (tp->rcv_up - tp->rcv_nxt) - 1; 1438 if (so->so_oobmark == 0) 1439 so->so_state |= SS_RCVATMARK; 1440 sohasoutofband(so); 1441 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1442 } 1443 /* 1444 * Remove out of band data so doesn't get presented to user. 1445 * This can happen independent of advancing the URG pointer, 1446 * but if two URG's are pending at once, some out-of-band 1447 * data may creep in... ick. 1448 */ 1449 if (ti->ti_urp <= (u_long)ti->ti_len 1450 #ifdef SO_OOBINLINE 1451 && (so->so_options & SO_OOBINLINE) == 0 1452 #endif 1453 ) 1454 tcp_pulloutofband(so, ti, m); 1455 } else 1456 /* 1457 * If no out of band data is expected, 1458 * pull receive urgent pointer along 1459 * with the receive window. 1460 */ 1461 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1462 tp->rcv_up = tp->rcv_nxt; 1463 dodata: /* XXX */ 1464 1465 /* 1466 * Process the segment text, merging it into the TCP sequencing queue, 1467 * and arranging for acknowledgment of receipt if necessary. 1468 * This process logically involves adjusting tp->rcv_wnd as data 1469 * is presented to the user (this happens in tcp_usrreq.c, 1470 * case PRU_RCVD). If a FIN has already been received on this 1471 * connection then we just ignore the text. 1472 */ 1473 if ((ti->ti_len || (tiflags&TH_FIN)) && 1474 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1475 TCP_REASS(tp, ti, m, so, tiflags); 1476 /* 1477 * Note the amount of data that peer has sent into 1478 * our window, in order to estimate the sender's 1479 * buffer size. 1480 */ 1481 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1482 } else { 1483 m_freem(m); 1484 tiflags &= ~TH_FIN; 1485 } 1486 1487 /* 1488 * If FIN is received ACK the FIN and let the user know 1489 * that the connection is closing. 1490 */ 1491 if (tiflags & TH_FIN) { 1492 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1493 socantrcvmore(so); 1494 /* 1495 * If connection is half-synchronized 1496 * (ie SEND_SYN flag on) then delay ACK, 1497 * so it may be piggybacked when SYN is sent. 1498 * Otherwise, since we received a FIN then no 1499 * more input can be expected, send ACK now. 1500 */ 1501 if (tp->t_flags & TF_NEEDSYN) 1502 tp->t_flags |= TF_DELACK; 1503 else 1504 tp->t_flags |= TF_ACKNOW; 1505 tp->rcv_nxt++; 1506 } 1507 switch (tp->t_state) { 1508 1509 /* 1510 * In SYN_RECEIVED and ESTABLISHED STATES 1511 * enter the CLOSE_WAIT state. 1512 */ 1513 case TCPS_SYN_RECEIVED: 1514 case TCPS_ESTABLISHED: 1515 tp->t_state = TCPS_CLOSE_WAIT; 1516 break; 1517 1518 /* 1519 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1520 * enter the CLOSING state. 1521 */ 1522 case TCPS_FIN_WAIT_1: 1523 tp->t_state = TCPS_CLOSING; 1524 break; 1525 1526 /* 1527 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1528 * starting the time-wait timer, turning off the other 1529 * standard timers. 1530 */ 1531 case TCPS_FIN_WAIT_2: 1532 tp->t_state = TCPS_TIME_WAIT; 1533 tcp_canceltimers(tp); 1534 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1535 if (tp->cc_recv != 0 && 1536 tp->t_duration < TCPTV_MSL) { 1537 tp->t_timer[TCPT_2MSL] = 1538 tp->t_rxtcur * TCPTV_TWTRUNC; 1539 /* For transaction client, force ACK now. */ 1540 tp->t_flags |= TF_ACKNOW; 1541 } 1542 else 1543 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1544 soisdisconnected(so); 1545 break; 1546 1547 /* 1548 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1549 */ 1550 case TCPS_TIME_WAIT: 1551 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1552 break; 1553 } 1554 } 1555 #ifdef TCPDEBUG 1556 if (so->so_options & SO_DEBUG) 1557 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1558 #endif 1559 1560 /* 1561 * Return any desired output. 1562 */ 1563 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1564 (void) tcp_output(tp); 1565 return; 1566 1567 dropafterack: 1568 /* 1569 * Generate an ACK dropping incoming segment if it occupies 1570 * sequence space, where the ACK reflects our state. 1571 */ 1572 if (tiflags & TH_RST) 1573 goto drop; 1574 #ifdef TCPDEBUG 1575 if (so->so_options & SO_DEBUG) 1576 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1577 #endif 1578 m_freem(m); 1579 tp->t_flags |= TF_ACKNOW; 1580 (void) tcp_output(tp); 1581 return; 1582 1583 dropwithreset: 1584 /* 1585 * Generate a RST, dropping incoming segment. 1586 * Make ACK acceptable to originator of segment. 1587 * Don't bother to respond if destination was broadcast/multicast. 1588 */ 1589 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1590 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1591 goto drop; 1592 #ifdef TCPDEBUG 1593 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1594 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1595 #endif 1596 if (tiflags & TH_ACK) 1597 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1598 else { 1599 if (tiflags & TH_SYN) 1600 ti->ti_len++; 1601 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1602 TH_RST|TH_ACK); 1603 } 1604 /* destroy temporarily created socket */ 1605 if (dropsocket) 1606 (void) soabort(so); 1607 return; 1608 1609 drop: 1610 /* 1611 * Drop space held by incoming segment and return. 1612 */ 1613 #ifdef TCPDEBUG 1614 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1615 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1616 #endif 1617 m_freem(m); 1618 /* destroy temporarily created socket */ 1619 if (dropsocket) 1620 (void) soabort(so); 1621 return; 1622 #ifndef TUBA_INCLUDE 1623 } 1624 1625 void 1626 tcp_dooptions(tp, cp, cnt, ti, to) 1627 struct tcpcb *tp; 1628 u_char *cp; 1629 int cnt; 1630 struct tcpiphdr *ti; 1631 struct tcpopt *to; 1632 { 1633 u_short mss = 0; 1634 int opt, optlen; 1635 1636 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1637 opt = cp[0]; 1638 if (opt == TCPOPT_EOL) 1639 break; 1640 if (opt == TCPOPT_NOP) 1641 optlen = 1; 1642 else { 1643 optlen = cp[1]; 1644 if (optlen <= 0) 1645 break; 1646 } 1647 switch (opt) { 1648 1649 default: 1650 continue; 1651 1652 case TCPOPT_MAXSEG: 1653 if (optlen != TCPOLEN_MAXSEG) 1654 continue; 1655 if (!(ti->ti_flags & TH_SYN)) 1656 continue; 1657 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1658 NTOHS(mss); 1659 break; 1660 1661 case TCPOPT_WINDOW: 1662 if (optlen != TCPOLEN_WINDOW) 1663 continue; 1664 if (!(ti->ti_flags & TH_SYN)) 1665 continue; 1666 tp->t_flags |= TF_RCVD_SCALE; 1667 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1668 break; 1669 1670 case TCPOPT_TIMESTAMP: 1671 if (optlen != TCPOLEN_TIMESTAMP) 1672 continue; 1673 to->to_flag |= TOF_TS; 1674 bcopy((char *)cp + 2, 1675 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1676 NTOHL(to->to_tsval); 1677 bcopy((char *)cp + 6, 1678 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1679 NTOHL(to->to_tsecr); 1680 1681 /* 1682 * A timestamp received in a SYN makes 1683 * it ok to send timestamp requests and replies. 1684 */ 1685 if (ti->ti_flags & TH_SYN) { 1686 tp->t_flags |= TF_RCVD_TSTMP; 1687 tp->ts_recent = to->to_tsval; 1688 tp->ts_recent_age = tcp_now; 1689 } 1690 break; 1691 case TCPOPT_CC: 1692 if (optlen != TCPOLEN_CC) 1693 continue; 1694 to->to_flag |= TCPOPT_CC; 1695 bcopy((char *)cp + 2, 1696 (char *)&to->to_cc, sizeof(to->to_cc)); 1697 NTOHL(to->to_cc); 1698 /* 1699 * A CC or CC.new option received in a SYN makes 1700 * it ok to send CC in subsequent segments. 1701 */ 1702 if (ti->ti_flags & TH_SYN) 1703 tp->t_flags |= TF_RCVD_CC; 1704 break; 1705 case TCPOPT_CCNEW: 1706 if (optlen != TCPOLEN_CC) 1707 continue; 1708 if (!(ti->ti_flags & TH_SYN)) 1709 continue; 1710 to->to_flag |= TOF_CCNEW; 1711 bcopy((char *)cp + 2, 1712 (char *)&to->to_cc, sizeof(to->to_cc)); 1713 NTOHL(to->to_cc); 1714 /* 1715 * A CC or CC.new option received in a SYN makes 1716 * it ok to send CC in subsequent segments. 1717 */ 1718 tp->t_flags |= TF_RCVD_CC; 1719 break; 1720 case TCPOPT_CCECHO: 1721 if (optlen != TCPOLEN_CC) 1722 continue; 1723 if (!(ti->ti_flags & TH_SYN)) 1724 continue; 1725 to->to_flag |= TOF_CCECHO; 1726 bcopy((char *)cp + 2, 1727 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1728 NTOHL(to->to_ccecho); 1729 break; 1730 } 1731 } 1732 if (ti->ti_flags & TH_SYN) 1733 tcp_mss(tp, mss); /* sets t_maxseg */ 1734 } 1735 1736 /* 1737 * Pull out of band byte out of a segment so 1738 * it doesn't appear in the user's data queue. 1739 * It is still reflected in the segment length for 1740 * sequencing purposes. 1741 */ 1742 void 1743 tcp_pulloutofband(so, ti, m) 1744 struct socket *so; 1745 struct tcpiphdr *ti; 1746 register struct mbuf *m; 1747 { 1748 int cnt = ti->ti_urp - 1; 1749 1750 while (cnt >= 0) { 1751 if (m->m_len > cnt) { 1752 char *cp = mtod(m, caddr_t) + cnt; 1753 struct tcpcb *tp = sototcpcb(so); 1754 1755 tp->t_iobc = *cp; 1756 tp->t_oobflags |= TCPOOB_HAVEDATA; 1757 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1758 m->m_len--; 1759 return; 1760 } 1761 cnt -= m->m_len; 1762 m = m->m_next; 1763 if (m == 0) 1764 break; 1765 } 1766 panic("tcp_pulloutofband"); 1767 } 1768 1769 /* 1770 * Collect new round-trip time estimate 1771 * and update averages and current timeout. 1772 */ 1773 void 1774 tcp_xmit_timer(tp, rtt) 1775 register struct tcpcb *tp; 1776 short rtt; 1777 { 1778 register short delta; 1779 1780 tcpstat.tcps_rttupdated++; 1781 if (tp->t_srtt != 0) { 1782 /* 1783 * srtt is stored as fixed point with 3 bits after the 1784 * binary point (i.e., scaled by 8). The following magic 1785 * is equivalent to the smoothing algorithm in rfc793 with 1786 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1787 * point). Adjust rtt to origin 0. 1788 */ 1789 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1790 if ((tp->t_srtt += delta) <= 0) 1791 tp->t_srtt = 1; 1792 /* 1793 * We accumulate a smoothed rtt variance (actually, a 1794 * smoothed mean difference), then set the retransmit 1795 * timer to smoothed rtt + 4 times the smoothed variance. 1796 * rttvar is stored as fixed point with 2 bits after the 1797 * binary point (scaled by 4). The following is 1798 * equivalent to rfc793 smoothing with an alpha of .75 1799 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1800 * rfc793's wired-in beta. 1801 */ 1802 if (delta < 0) 1803 delta = -delta; 1804 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1805 if ((tp->t_rttvar += delta) <= 0) 1806 tp->t_rttvar = 1; 1807 } else { 1808 /* 1809 * No rtt measurement yet - use the unsmoothed rtt. 1810 * Set the variance to half the rtt (so our first 1811 * retransmit happens at 3*rtt). 1812 */ 1813 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1814 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1815 } 1816 tp->t_rtt = 0; 1817 tp->t_rxtshift = 0; 1818 1819 /* 1820 * the retransmit should happen at rtt + 4 * rttvar. 1821 * Because of the way we do the smoothing, srtt and rttvar 1822 * will each average +1/2 tick of bias. When we compute 1823 * the retransmit timer, we want 1/2 tick of rounding and 1824 * 1 extra tick because of +-1/2 tick uncertainty in the 1825 * firing of the timer. The bias will give us exactly the 1826 * 1.5 tick we need. But, because the bias is 1827 * statistical, we have to test that we don't drop below 1828 * the minimum feasible timer (which is 2 ticks). 1829 */ 1830 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1831 tp->t_rttmin, TCPTV_REXMTMAX); 1832 1833 /* 1834 * We received an ack for a packet that wasn't retransmitted; 1835 * it is probably safe to discard any error indications we've 1836 * received recently. This isn't quite right, but close enough 1837 * for now (a route might have failed after we sent a segment, 1838 * and the return path might not be symmetrical). 1839 */ 1840 tp->t_softerror = 0; 1841 } 1842 1843 /* 1844 * Determine a reasonable value for maxseg size. 1845 * If the route is known, check route for mtu. 1846 * If none, use an mss that can be handled on the outgoing 1847 * interface without forcing IP to fragment; if bigger than 1848 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1849 * to utilize large mbufs. If no route is found, route has no mtu, 1850 * or the destination isn't local, use a default, hopefully conservative 1851 * size (usually 512 or the default IP max size, but no more than the mtu 1852 * of the interface), as we can't discover anything about intervening 1853 * gateways or networks. We also initialize the congestion/slow start 1854 * window to be a single segment if the destination isn't local. 1855 * While looking at the routing entry, we also initialize other path-dependent 1856 * parameters from pre-set or cached values in the routing entry. 1857 * 1858 * Also take into account the space needed for options that we 1859 * send regularly. Make maxseg shorter by that amount to assure 1860 * that we can send maxseg amount of data even when the options 1861 * are present. Store the upper limit of the length of options plus 1862 * data in maxopd. 1863 * 1864 * NOTE that this routine is only called when we process an incoming 1865 * segment, for outgoing segments only tcp_mssopt is called. 1866 * 1867 * In case of T/TCP, we call this routine during implicit connection 1868 * setup as well (offer = -1), to initialize maxseg from the cached 1869 * MSS of our peer. 1870 */ 1871 void 1872 tcp_mss(tp, offer) 1873 struct tcpcb *tp; 1874 int offer; 1875 { 1876 register struct rtentry *rt; 1877 struct ifnet *ifp; 1878 register int rtt, mss; 1879 u_long bufsize; 1880 struct inpcb *inp; 1881 struct socket *so; 1882 struct rmxp_tao *taop; 1883 int origoffer = offer; 1884 extern int tcp_do_rfc1644; 1885 extern int tcp_mssdflt; 1886 1887 inp = tp->t_inpcb; 1888 if ((rt = tcp_rtlookup(inp)) == NULL) { 1889 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1890 return; 1891 } 1892 ifp = rt->rt_ifp; 1893 so = inp->inp_socket; 1894 1895 taop = rmx_taop(rt->rt_rmx); 1896 /* 1897 * Offer == -1 means that we didn't receive SYN yet, 1898 * use cached value in that case; 1899 */ 1900 if (offer == -1) 1901 offer = taop->tao_mssopt; 1902 /* 1903 * Offer == 0 means that there was no MSS on the SYN segment, 1904 * in this case we use tcp_mssdflt. 1905 */ 1906 if (offer == 0) 1907 offer = tcp_mssdflt; 1908 else 1909 /* 1910 * Sanity check: make sure that maxopd will be large 1911 * enough to allow some data on segments even is the 1912 * all the option space is used (40bytes). Otherwise 1913 * funny things may happen in tcp_output. 1914 */ 1915 offer = max(offer, 64); 1916 taop->tao_mssopt = offer; 1917 1918 #ifdef RTV_MTU /* if route characteristics exist ... */ 1919 /* 1920 * While we're here, check if there's an initial rtt 1921 * or rttvar. Convert from the route-table units 1922 * to scaled multiples of the slow timeout timer. 1923 */ 1924 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1925 /* 1926 * XXX the lock bit for RTT indicates that the value 1927 * is also a minimum value; this is subject to time. 1928 */ 1929 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1930 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1931 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1932 if (rt->rt_rmx.rmx_rttvar) 1933 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1934 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1935 else 1936 /* default variation is +- 1 rtt */ 1937 tp->t_rttvar = 1938 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1939 TCPT_RANGESET(tp->t_rxtcur, 1940 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1941 tp->t_rttmin, TCPTV_REXMTMAX); 1942 } 1943 /* 1944 * if there's an mtu associated with the route, use it 1945 */ 1946 if (rt->rt_rmx.rmx_mtu) 1947 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1948 else 1949 #endif /* RTV_MTU */ 1950 { 1951 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1952 if (!in_localaddr(inp->inp_faddr)) 1953 mss = min(mss, tcp_mssdflt); 1954 } 1955 mss = min(mss, offer); 1956 /* 1957 * maxopd stores the maximum length of data AND options 1958 * in a segment; maxseg is the amount of data in a normal 1959 * segment. We need to store this value (maxopd) apart 1960 * from maxseg, because now every segment carries options 1961 * and thus we normally have somewhat less data in segments. 1962 */ 1963 tp->t_maxopd = mss; 1964 1965 /* 1966 * In case of T/TCP, origoffer==-1 indicates, that no segments 1967 * were received yet. In this case we just guess, otherwise 1968 * we do the same as before T/TCP. 1969 */ 1970 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1971 (origoffer == -1 || 1972 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 1973 mss -= TCPOLEN_TSTAMP_APPA; 1974 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1975 (origoffer == -1 || 1976 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 1977 mss -= TCPOLEN_CC_APPA; 1978 1979 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1980 if (mss > MCLBYTES) 1981 mss &= ~(MCLBYTES-1); 1982 #else 1983 if (mss > MCLBYTES) 1984 mss = mss / MCLBYTES * MCLBYTES; 1985 #endif 1986 /* 1987 * If there's a pipesize, change the socket buffer 1988 * to that size. Make the socket buffers an integral 1989 * number of mss units; if the mss is larger than 1990 * the socket buffer, decrease the mss. 1991 */ 1992 #ifdef RTV_SPIPE 1993 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1994 #endif 1995 bufsize = so->so_snd.sb_hiwat; 1996 if (bufsize < mss) 1997 mss = bufsize; 1998 else { 1999 bufsize = roundup(bufsize, mss); 2000 if (bufsize > sb_max) 2001 bufsize = sb_max; 2002 (void)sbreserve(&so->so_snd, bufsize); 2003 } 2004 tp->t_maxseg = mss; 2005 2006 #ifdef RTV_RPIPE 2007 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2008 #endif 2009 bufsize = so->so_rcv.sb_hiwat; 2010 if (bufsize > mss) { 2011 bufsize = roundup(bufsize, mss); 2012 if (bufsize > sb_max) 2013 bufsize = sb_max; 2014 (void)sbreserve(&so->so_rcv, bufsize); 2015 } 2016 /* 2017 * Don't force slow-start on local network. 2018 */ 2019 if (!in_localaddr(inp->inp_faddr)) 2020 tp->snd_cwnd = mss; 2021 2022 #ifdef RTV_SSTHRESH 2023 if (rt->rt_rmx.rmx_ssthresh) { 2024 /* 2025 * There's some sort of gateway or interface 2026 * buffer limit on the path. Use this to set 2027 * the slow start threshhold, but set the 2028 * threshold to no less than 2*mss. 2029 */ 2030 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2031 } 2032 #endif 2033 } 2034 2035 /* 2036 * Determine the MSS option to send on an outgoing SYN. 2037 */ 2038 int 2039 tcp_mssopt(tp) 2040 struct tcpcb *tp; 2041 { 2042 struct rtentry *rt; 2043 extern int tcp_mssdflt; 2044 2045 rt = tcp_rtlookup(tp->t_inpcb); 2046 if (rt == NULL) 2047 return tcp_mssdflt; 2048 2049 /* 2050 * if there's an mtu associated with the route, use it 2051 */ 2052 if (rt->rt_rmx.rmx_mtu) 2053 return rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2054 2055 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2056 } 2057 #endif /* TUBA_INCLUDE */ 2058