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 1.78 1998/05/31 18:42:49 peter Exp $ 35 */ 36 37 #include "opt_ipfw.h" /* for ipfw_fwd */ 38 #include "opt_tcpdebug.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/sysctl.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/proc.h> /* for proc0 declaration */ 47 #include <sys/protosw.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/syslog.h> 51 52 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 53 54 #include <net/if.h> 55 #include <net/route.h> 56 57 #include <netinet/in.h> 58 #include <netinet/in_systm.h> 59 #include <netinet/ip.h> 60 #include <netinet/in_pcb.h> 61 #include <netinet/ip_var.h> 62 #include <netinet/tcp.h> 63 #include <netinet/tcp_fsm.h> 64 #include <netinet/tcp_seq.h> 65 #include <netinet/tcp_timer.h> 66 #include <netinet/tcp_var.h> 67 #include <netinet/tcpip.h> 68 #ifdef TCPDEBUG 69 #include <netinet/tcp_debug.h> 70 static struct tcpiphdr tcp_saveti; 71 #endif 72 73 static int tcprexmtthresh = 3; 74 tcp_seq tcp_iss; 75 tcp_cc tcp_ccgen; 76 77 struct tcpstat tcpstat; 78 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 79 CTLFLAG_RD, &tcpstat , tcpstat, ""); 80 81 static int log_in_vain = 0; 82 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 83 &log_in_vain, 0, ""); 84 85 int tcp_delack_enabled = 1; 86 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 87 &tcp_delack_enabled, 0, ""); 88 89 u_long tcp_now; 90 struct inpcbhead tcb; 91 struct inpcbinfo tcbinfo; 92 93 static void tcp_dooptions __P((struct tcpcb *, 94 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 95 static void tcp_pulloutofband __P((struct socket *, 96 struct tcpiphdr *, struct mbuf *)); 97 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 98 static void tcp_xmit_timer __P((struct tcpcb *, int)); 99 100 101 /* 102 * Insert segment ti into reassembly queue of tcp with 103 * control block tp. Return TH_FIN if reassembly now includes 104 * a segment with FIN. The macro form does the common case inline 105 * (segment is the next to be received on an established connection, 106 * and the queue is empty), avoiding linkage into and removal 107 * from the queue and repetition of various conversions. 108 * Set DELACK for segments received in order, but ack immediately 109 * when segments are out of order (so fast retransmit can work). 110 */ 111 #define TCP_REASS(tp, ti, m, so, flags) { \ 112 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 113 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 114 (tp)->t_state == TCPS_ESTABLISHED) { \ 115 if (tcp_delack_enabled) \ 116 tp->t_flags |= TF_DELACK; \ 117 else \ 118 tp->t_flags |= TF_ACKNOW; \ 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 } 130 131 static int 132 tcp_reass(tp, ti, m) 133 register struct tcpcb *tp; 134 register struct tcpiphdr *ti; 135 struct mbuf *m; 136 { 137 register struct tcpiphdr *q; 138 struct socket *so = tp->t_inpcb->inp_socket; 139 int flags; 140 141 /* 142 * Call with ti==0 after become established to 143 * force pre-ESTABLISHED data up to user socket. 144 */ 145 if (ti == 0) 146 goto present; 147 148 /* 149 * Find a segment which begins after this one does. 150 */ 151 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 152 q = (struct tcpiphdr *)q->ti_next) 153 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 154 break; 155 156 /* 157 * If there is a preceding segment, it may provide some of 158 * our data already. If so, drop the data from the incoming 159 * segment. If it provides all of our data, drop us. 160 */ 161 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 162 register int i; 163 q = (struct tcpiphdr *)q->ti_prev; 164 /* conversion to int (in i) handles seq wraparound */ 165 i = q->ti_seq + q->ti_len - ti->ti_seq; 166 if (i > 0) { 167 if (i >= ti->ti_len) { 168 tcpstat.tcps_rcvduppack++; 169 tcpstat.tcps_rcvdupbyte += ti->ti_len; 170 m_freem(m); 171 /* 172 * Try to present any queued data 173 * at the left window edge to the user. 174 * This is needed after the 3-WHS 175 * completes. 176 */ 177 goto present; /* ??? */ 178 } 179 m_adj(m, i); 180 ti->ti_len -= i; 181 ti->ti_seq += i; 182 } 183 q = (struct tcpiphdr *)(q->ti_next); 184 } 185 tcpstat.tcps_rcvoopack++; 186 tcpstat.tcps_rcvoobyte += ti->ti_len; 187 REASS_MBUF(ti) = m; /* XXX */ 188 189 /* 190 * While we overlap succeeding segments trim them or, 191 * if they are completely covered, dequeue them. 192 */ 193 while (q != (struct tcpiphdr *)tp) { 194 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 195 if (i <= 0) 196 break; 197 if (i < q->ti_len) { 198 q->ti_seq += i; 199 q->ti_len -= i; 200 m_adj(REASS_MBUF(q), i); 201 break; 202 } 203 q = (struct tcpiphdr *)q->ti_next; 204 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 205 remque(q->ti_prev); 206 m_freem(m); 207 } 208 209 /* 210 * Stick new segment in its place. 211 */ 212 insque(ti, q->ti_prev); 213 214 present: 215 /* 216 * Present data to user, advancing rcv_nxt through 217 * completed sequence space. 218 */ 219 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 220 return (0); 221 ti = tp->seg_next; 222 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 223 return (0); 224 do { 225 tp->rcv_nxt += ti->ti_len; 226 flags = ti->ti_flags & TH_FIN; 227 remque(ti); 228 m = REASS_MBUF(ti); 229 ti = (struct tcpiphdr *)ti->ti_next; 230 if (so->so_state & SS_CANTRCVMORE) 231 m_freem(m); 232 else 233 sbappend(&so->so_rcv, m); 234 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 235 sorwakeup(so); 236 return (flags); 237 } 238 239 /* 240 * TCP input routine, follows pages 65-76 of the 241 * protocol specification dated September, 1981 very closely. 242 */ 243 void 244 tcp_input(m, iphlen) 245 register struct mbuf *m; 246 int iphlen; 247 { 248 register struct tcpiphdr *ti; 249 register struct inpcb *inp; 250 u_char *optp = NULL; 251 int optlen = 0; 252 int len, tlen, off; 253 register struct tcpcb *tp = 0; 254 register int tiflags; 255 struct socket *so = 0; 256 int todrop, acked, ourfinisacked, needoutput = 0; 257 struct in_addr laddr; 258 int dropsocket = 0; 259 int iss = 0; 260 u_long tiwin; 261 struct tcpopt to; /* options in this segment */ 262 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 263 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 264 #ifdef TCPDEBUG 265 short ostate = 0; 266 #endif 267 268 bzero((char *)&to, sizeof(to)); 269 270 tcpstat.tcps_rcvtotal++; 271 /* 272 * Get IP and TCP header together in first mbuf. 273 * Note: IP leaves IP header in first mbuf. 274 */ 275 ti = mtod(m, struct tcpiphdr *); 276 if (iphlen > sizeof (struct ip)) 277 ip_stripoptions(m, (struct mbuf *)0); 278 if (m->m_len < sizeof (struct tcpiphdr)) { 279 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 280 tcpstat.tcps_rcvshort++; 281 return; 282 } 283 ti = mtod(m, struct tcpiphdr *); 284 } 285 286 /* 287 * Checksum extended TCP header and data. 288 */ 289 tlen = ((struct ip *)ti)->ip_len; 290 len = sizeof (struct ip) + tlen; 291 ti->ti_next = ti->ti_prev = 0; 292 ti->ti_x1 = 0; 293 ti->ti_len = (u_short)tlen; 294 HTONS(ti->ti_len); 295 ti->ti_sum = in_cksum(m, len); 296 if (ti->ti_sum) { 297 tcpstat.tcps_rcvbadsum++; 298 goto drop; 299 } 300 301 /* 302 * Check that TCP offset makes sense, 303 * pull out TCP options and adjust length. XXX 304 */ 305 off = ti->ti_off << 2; 306 if (off < sizeof (struct tcphdr) || off > tlen) { 307 tcpstat.tcps_rcvbadoff++; 308 goto drop; 309 } 310 tlen -= off; 311 ti->ti_len = tlen; 312 if (off > sizeof (struct tcphdr)) { 313 if (m->m_len < sizeof(struct ip) + off) { 314 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 315 tcpstat.tcps_rcvshort++; 316 return; 317 } 318 ti = mtod(m, struct tcpiphdr *); 319 } 320 optlen = off - sizeof (struct tcphdr); 321 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 322 } 323 tiflags = ti->ti_flags; 324 325 /* 326 * Convert TCP protocol specific fields to host format. 327 */ 328 NTOHL(ti->ti_seq); 329 NTOHL(ti->ti_ack); 330 NTOHS(ti->ti_win); 331 NTOHS(ti->ti_urp); 332 333 /* 334 * Drop TCP, IP headers and TCP options. 335 */ 336 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 337 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 338 339 /* 340 * Locate pcb for segment. 341 */ 342 findpcb: 343 #ifdef IPFIREWALL_FORWARD 344 if (ip_fw_fwd_addr != NULL) { 345 /* 346 * Diverted. Pretend to be the destination. 347 * already got one like this? 348 */ 349 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport, 350 ti->ti_dst, ti->ti_dport, 0); 351 if (!inp) { 352 /* 353 * No, then it's new. Try find the ambushing socket 354 */ 355 if (!ip_fw_fwd_addr->sin_port) { 356 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, 357 ti->ti_sport, ip_fw_fwd_addr->sin_addr, 358 ti->ti_dport, 1); 359 } else { 360 inp = in_pcblookup_hash(&tcbinfo, 361 ti->ti_src, ti->ti_sport, 362 ip_fw_fwd_addr->sin_addr, 363 ntohs(ip_fw_fwd_addr->sin_port), 1); 364 } 365 } 366 ip_fw_fwd_addr = NULL; 367 } else 368 #endif /* IPFIREWALL_FORWARD */ 369 370 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport, 371 ti->ti_dst, ti->ti_dport, 1); 372 373 /* 374 * If the state is CLOSED (i.e., TCB does not exist) then 375 * all data in the incoming segment is discarded. 376 * If the TCB exists but is in CLOSED state, it is embryonic, 377 * but should either do a listen or a connect soon. 378 */ 379 if (inp == NULL) { 380 if (log_in_vain && tiflags & TH_SYN) { 381 char buf[4*sizeof "123"]; 382 383 strcpy(buf, inet_ntoa(ti->ti_dst)); 384 log(LOG_INFO, 385 "Connection attempt to TCP %s:%d from %s:%d\n", 386 buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src), 387 ntohs(ti->ti_sport)); 388 } 389 goto dropwithreset; 390 } 391 tp = intotcpcb(inp); 392 if (tp == 0) 393 goto dropwithreset; 394 if (tp->t_state == TCPS_CLOSED) 395 goto drop; 396 397 /* Unscale the window into a 32-bit value. */ 398 if ((tiflags & TH_SYN) == 0) 399 tiwin = ti->ti_win << tp->snd_scale; 400 else 401 tiwin = ti->ti_win; 402 403 so = inp->inp_socket; 404 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 405 #ifdef TCPDEBUG 406 if (so->so_options & SO_DEBUG) { 407 ostate = tp->t_state; 408 tcp_saveti = *ti; 409 } 410 #endif 411 if (so->so_options & SO_ACCEPTCONN) { 412 register struct tcpcb *tp0 = tp; 413 struct socket *so2; 414 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 415 /* 416 * Note: dropwithreset makes sure we don't 417 * send a RST in response to a RST. 418 */ 419 if (tiflags & TH_ACK) { 420 tcpstat.tcps_badsyn++; 421 goto dropwithreset; 422 } 423 goto drop; 424 } 425 so2 = sonewconn(so, 0); 426 if (so2 == 0) { 427 tcpstat.tcps_listendrop++; 428 so2 = sodropablereq(so); 429 if (so2) { 430 tcp_drop(sototcpcb(so2), ETIMEDOUT); 431 so2 = sonewconn(so, 0); 432 } 433 if (!so2) 434 goto drop; 435 } 436 so = so2; 437 /* 438 * This is ugly, but .... 439 * 440 * Mark socket as temporary until we're 441 * committed to keeping it. The code at 442 * ``drop'' and ``dropwithreset'' check the 443 * flag dropsocket to see if the temporary 444 * socket created here should be discarded. 445 * We mark the socket as discardable until 446 * we're committed to it below in TCPS_LISTEN. 447 */ 448 dropsocket++; 449 inp = (struct inpcb *)so->so_pcb; 450 inp->inp_laddr = ti->ti_dst; 451 inp->inp_lport = ti->ti_dport; 452 if (in_pcbinshash(inp) != 0) { 453 /* 454 * Undo the assignments above if we failed to put 455 * the PCB on the hash lists. 456 */ 457 inp->inp_laddr.s_addr = INADDR_ANY; 458 inp->inp_lport = 0; 459 goto drop; 460 } 461 inp->inp_options = ip_srcroute(); 462 tp = intotcpcb(inp); 463 tp->t_state = TCPS_LISTEN; 464 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 465 466 /* Compute proper scaling value from buffer space */ 467 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 468 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 469 tp->request_r_scale++; 470 } 471 } 472 473 /* 474 * Segment received on connection. 475 * Reset idle time and keep-alive timer. 476 */ 477 tp->t_idle = 0; 478 if (TCPS_HAVEESTABLISHED(tp->t_state)) 479 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 480 481 /* 482 * Process options if not in LISTEN state, 483 * else do it below (after getting remote address). 484 */ 485 if (tp->t_state != TCPS_LISTEN) 486 tcp_dooptions(tp, optp, optlen, ti, &to); 487 488 /* 489 * Header prediction: check for the two common cases 490 * of a uni-directional data xfer. If the packet has 491 * no control flags, is in-sequence, the window didn't 492 * change and we're not retransmitting, it's a 493 * candidate. If the length is zero and the ack moved 494 * forward, we're the sender side of the xfer. Just 495 * free the data acked & wake any higher level process 496 * that was blocked waiting for space. If the length 497 * is non-zero and the ack didn't move, we're the 498 * receiver side. If we're getting packets in-order 499 * (the reassembly queue is empty), add the data to 500 * the socket buffer and note that we need a delayed ack. 501 * Make sure that the hidden state-flags are also off. 502 * Since we check for TCPS_ESTABLISHED above, it can only 503 * be TH_NEEDSYN. 504 */ 505 if (tp->t_state == TCPS_ESTABLISHED && 506 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 507 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 508 ((to.to_flag & TOF_TS) == 0 || 509 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 510 /* 511 * Using the CC option is compulsory if once started: 512 * the segment is OK if no T/TCP was negotiated or 513 * if the segment has a CC option equal to CCrecv 514 */ 515 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 516 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 517 ti->ti_seq == tp->rcv_nxt && 518 tiwin && tiwin == tp->snd_wnd && 519 tp->snd_nxt == tp->snd_max) { 520 521 /* 522 * If last ACK falls within this segment's sequence numbers, 523 * record the timestamp. 524 * NOTE that the test is modified according to the latest 525 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 526 */ 527 if ((to.to_flag & TOF_TS) != 0 && 528 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 529 tp->ts_recent_age = tcp_now; 530 tp->ts_recent = to.to_tsval; 531 } 532 533 if (ti->ti_len == 0) { 534 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 535 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 536 tp->snd_cwnd >= tp->snd_wnd && 537 tp->t_dupacks < tcprexmtthresh) { 538 /* 539 * this is a pure ack for outstanding data. 540 */ 541 ++tcpstat.tcps_predack; 542 if ((to.to_flag & TOF_TS) != 0) 543 tcp_xmit_timer(tp, 544 tcp_now - to.to_tsecr + 1); 545 else if (tp->t_rtt && 546 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 547 tcp_xmit_timer(tp, tp->t_rtt); 548 acked = ti->ti_ack - tp->snd_una; 549 tcpstat.tcps_rcvackpack++; 550 tcpstat.tcps_rcvackbyte += acked; 551 sbdrop(&so->so_snd, acked); 552 tp->snd_una = ti->ti_ack; 553 m_freem(m); 554 555 /* 556 * If all outstanding data are acked, stop 557 * retransmit timer, otherwise restart timer 558 * using current (possibly backed-off) value. 559 * If process is waiting for space, 560 * wakeup/selwakeup/signal. If data 561 * are ready to send, let tcp_output 562 * decide between more output or persist. 563 */ 564 if (tp->snd_una == tp->snd_max) 565 tp->t_timer[TCPT_REXMT] = 0; 566 else if (tp->t_timer[TCPT_PERSIST] == 0) 567 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 568 569 sowwakeup(so); 570 if (so->so_snd.sb_cc) 571 (void) tcp_output(tp); 572 return; 573 } 574 } else if (ti->ti_ack == tp->snd_una && 575 tp->seg_next == (struct tcpiphdr *)tp && 576 ti->ti_len <= sbspace(&so->so_rcv)) { 577 /* 578 * this is a pure, in-sequence data packet 579 * with nothing on the reassembly queue and 580 * we have enough buffer space to take it. 581 */ 582 ++tcpstat.tcps_preddat; 583 tp->rcv_nxt += ti->ti_len; 584 tcpstat.tcps_rcvpack++; 585 tcpstat.tcps_rcvbyte += ti->ti_len; 586 /* 587 * Add data to socket buffer. 588 */ 589 sbappend(&so->so_rcv, m); 590 sorwakeup(so); 591 if (tcp_delack_enabled) { 592 tp->t_flags |= TF_DELACK; 593 } else { 594 tp->t_flags |= TF_ACKNOW; 595 tcp_output(tp); 596 } 597 return; 598 } 599 } 600 601 /* 602 * Calculate amount of space in receive window, 603 * and then do TCP input processing. 604 * Receive window is amount of space in rcv queue, 605 * but not less than advertised window. 606 */ 607 { int win; 608 609 win = sbspace(&so->so_rcv); 610 if (win < 0) 611 win = 0; 612 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 613 } 614 615 switch (tp->t_state) { 616 617 /* 618 * If the state is LISTEN then ignore segment if it contains an RST. 619 * If the segment contains an ACK then it is bad and send a RST. 620 * If it does not contain a SYN then it is not interesting; drop it. 621 * If it is from this socket, drop it, it must be forged. 622 * Don't bother responding if the destination was a broadcast. 623 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 624 * tp->iss, and send a segment: 625 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 626 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 627 * Fill in remote peer address fields if not previously specified. 628 * Enter SYN_RECEIVED state, and process any other fields of this 629 * segment in this state. 630 */ 631 case TCPS_LISTEN: { 632 register struct sockaddr_in *sin; 633 634 if (tiflags & TH_RST) 635 goto drop; 636 if (tiflags & TH_ACK) 637 goto dropwithreset; 638 if ((tiflags & TH_SYN) == 0) 639 goto drop; 640 if ((ti->ti_dport == ti->ti_sport) && 641 (ti->ti_dst.s_addr == ti->ti_src.s_addr)) 642 goto drop; 643 /* 644 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 645 * in_broadcast() should never return true on a received 646 * packet with M_BCAST not set. 647 */ 648 if (m->m_flags & (M_BCAST|M_MCAST) || 649 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 650 goto drop; 651 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 652 M_NOWAIT); 653 if (sin == NULL) 654 goto drop; 655 sin->sin_family = AF_INET; 656 sin->sin_len = sizeof(*sin); 657 sin->sin_addr = ti->ti_src; 658 sin->sin_port = ti->ti_sport; 659 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 660 laddr = inp->inp_laddr; 661 if (inp->inp_laddr.s_addr == INADDR_ANY) 662 inp->inp_laddr = ti->ti_dst; 663 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 664 inp->inp_laddr = laddr; 665 FREE(sin, M_SONAME); 666 goto drop; 667 } 668 FREE(sin, M_SONAME); 669 tp->t_template = tcp_template(tp); 670 if (tp->t_template == 0) { 671 tp = tcp_drop(tp, ENOBUFS); 672 dropsocket = 0; /* socket is already gone */ 673 goto drop; 674 } 675 if ((taop = tcp_gettaocache(inp)) == NULL) { 676 taop = &tao_noncached; 677 bzero(taop, sizeof(*taop)); 678 } 679 tcp_dooptions(tp, optp, optlen, ti, &to); 680 if (iss) 681 tp->iss = iss; 682 else 683 tp->iss = tcp_iss; 684 tcp_iss += TCP_ISSINCR/4; 685 tp->irs = ti->ti_seq; 686 tcp_sendseqinit(tp); 687 tcp_rcvseqinit(tp); 688 /* 689 * Initialization of the tcpcb for transaction; 690 * set SND.WND = SEG.WND, 691 * initialize CCsend and CCrecv. 692 */ 693 tp->snd_wnd = tiwin; /* initial send-window */ 694 tp->cc_send = CC_INC(tcp_ccgen); 695 tp->cc_recv = to.to_cc; 696 /* 697 * Perform TAO test on incoming CC (SEG.CC) option, if any. 698 * - compare SEG.CC against cached CC from the same host, 699 * if any. 700 * - if SEG.CC > chached value, SYN must be new and is accepted 701 * immediately: save new CC in the cache, mark the socket 702 * connected, enter ESTABLISHED state, turn on flag to 703 * send a SYN in the next segment. 704 * A virtual advertised window is set in rcv_adv to 705 * initialize SWS prevention. Then enter normal segment 706 * processing: drop SYN, process data and FIN. 707 * - otherwise do a normal 3-way handshake. 708 */ 709 if ((to.to_flag & TOF_CC) != 0) { 710 if (((tp->t_flags & TF_NOPUSH) != 0) && 711 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 712 713 taop->tao_cc = to.to_cc; 714 tp->t_state = TCPS_ESTABLISHED; 715 716 /* 717 * If there is a FIN, or if there is data and the 718 * connection is local, then delay SYN,ACK(SYN) in 719 * the hope of piggy-backing it on a response 720 * segment. Otherwise must send ACK now in case 721 * the other side is slow starting. 722 */ 723 if (tcp_delack_enabled && ((tiflags & TH_FIN) || (ti->ti_len != 0 && 724 in_localaddr(inp->inp_faddr)))) 725 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 726 else 727 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 728 729 /* 730 * Limit the `virtual advertised window' to TCP_MAXWIN 731 * here. Even if we requested window scaling, it will 732 * become effective only later when our SYN is acked. 733 */ 734 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 735 tcpstat.tcps_connects++; 736 soisconnected(so); 737 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 738 dropsocket = 0; /* committed to socket */ 739 tcpstat.tcps_accepts++; 740 goto trimthenstep6; 741 } 742 /* else do standard 3-way handshake */ 743 } else { 744 /* 745 * No CC option, but maybe CC.NEW: 746 * invalidate cached value. 747 */ 748 taop->tao_cc = 0; 749 } 750 /* 751 * TAO test failed or there was no CC option, 752 * do a standard 3-way handshake. 753 */ 754 tp->t_flags |= TF_ACKNOW; 755 tp->t_state = TCPS_SYN_RECEIVED; 756 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 757 dropsocket = 0; /* committed to socket */ 758 tcpstat.tcps_accepts++; 759 goto trimthenstep6; 760 } 761 762 /* 763 * If the state is SYN_RECEIVED: 764 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 765 */ 766 case TCPS_SYN_RECEIVED: 767 if ((tiflags & TH_ACK) && 768 (SEQ_LEQ(ti->ti_ack, tp->snd_una) || 769 SEQ_GT(ti->ti_ack, tp->snd_max))) 770 goto dropwithreset; 771 break; 772 773 /* 774 * If the state is SYN_SENT: 775 * if seg contains an ACK, but not for our SYN, drop the input. 776 * if seg contains a RST, then drop the connection. 777 * if seg does not contain SYN, then drop it. 778 * Otherwise this is an acceptable SYN segment 779 * initialize tp->rcv_nxt and tp->irs 780 * if seg contains ack then advance tp->snd_una 781 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 782 * arrange for segment to be acked (eventually) 783 * continue processing rest of data/controls, beginning with URG 784 */ 785 case TCPS_SYN_SENT: 786 if ((taop = tcp_gettaocache(inp)) == NULL) { 787 taop = &tao_noncached; 788 bzero(taop, sizeof(*taop)); 789 } 790 791 if ((tiflags & TH_ACK) && 792 (SEQ_LEQ(ti->ti_ack, tp->iss) || 793 SEQ_GT(ti->ti_ack, tp->snd_max))) { 794 /* 795 * If we have a cached CCsent for the remote host, 796 * hence we haven't just crashed and restarted, 797 * do not send a RST. This may be a retransmission 798 * from the other side after our earlier ACK was lost. 799 * Our new SYN, when it arrives, will serve as the 800 * needed ACK. 801 */ 802 if (taop->tao_ccsent != 0) 803 goto drop; 804 else 805 goto dropwithreset; 806 } 807 if (tiflags & TH_RST) { 808 if (tiflags & TH_ACK) 809 tp = tcp_drop(tp, ECONNREFUSED); 810 goto drop; 811 } 812 if ((tiflags & TH_SYN) == 0) 813 goto drop; 814 tp->snd_wnd = ti->ti_win; /* initial send window */ 815 tp->cc_recv = to.to_cc; /* foreign CC */ 816 817 tp->irs = ti->ti_seq; 818 tcp_rcvseqinit(tp); 819 if (tiflags & TH_ACK) { 820 /* 821 * Our SYN was acked. If segment contains CC.ECHO 822 * option, check it to make sure this segment really 823 * matches our SYN. If not, just drop it as old 824 * duplicate, but send an RST if we're still playing 825 * by the old rules. If no CC.ECHO option, make sure 826 * we don't get fooled into using T/TCP. 827 */ 828 if (to.to_flag & TOF_CCECHO) { 829 if (tp->cc_send != to.to_ccecho) 830 if (taop->tao_ccsent != 0) 831 goto drop; 832 else 833 goto dropwithreset; 834 } else 835 tp->t_flags &= ~TF_RCVD_CC; 836 tcpstat.tcps_connects++; 837 soisconnected(so); 838 /* Do window scaling on this connection? */ 839 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 840 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 841 tp->snd_scale = tp->requested_s_scale; 842 tp->rcv_scale = tp->request_r_scale; 843 } 844 /* Segment is acceptable, update cache if undefined. */ 845 if (taop->tao_ccsent == 0) 846 taop->tao_ccsent = to.to_ccecho; 847 848 tp->rcv_adv += tp->rcv_wnd; 849 tp->snd_una++; /* SYN is acked */ 850 /* 851 * If there's data, delay ACK; if there's also a FIN 852 * ACKNOW will be turned on later. 853 */ 854 if (tcp_delack_enabled && ti->ti_len != 0) 855 tp->t_flags |= TF_DELACK; 856 else 857 tp->t_flags |= TF_ACKNOW; 858 /* 859 * Received <SYN,ACK> in SYN_SENT[*] state. 860 * Transitions: 861 * SYN_SENT --> ESTABLISHED 862 * SYN_SENT* --> FIN_WAIT_1 863 */ 864 if (tp->t_flags & TF_NEEDFIN) { 865 tp->t_state = TCPS_FIN_WAIT_1; 866 tp->t_flags &= ~TF_NEEDFIN; 867 tiflags &= ~TH_SYN; 868 } else { 869 tp->t_state = TCPS_ESTABLISHED; 870 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 871 } 872 } else { 873 /* 874 * Received initial SYN in SYN-SENT[*] state => simul- 875 * taneous open. If segment contains CC option and there is 876 * a cached CC, apply TAO test; if it succeeds, connection is 877 * half-synchronized. Otherwise, do 3-way handshake: 878 * SYN-SENT -> SYN-RECEIVED 879 * SYN-SENT* -> SYN-RECEIVED* 880 * If there was no CC option, clear cached CC value. 881 */ 882 tp->t_flags |= TF_ACKNOW; 883 tp->t_timer[TCPT_REXMT] = 0; 884 if (to.to_flag & TOF_CC) { 885 if (taop->tao_cc != 0 && 886 CC_GT(to.to_cc, taop->tao_cc)) { 887 /* 888 * update cache and make transition: 889 * SYN-SENT -> ESTABLISHED* 890 * SYN-SENT* -> FIN-WAIT-1* 891 */ 892 taop->tao_cc = to.to_cc; 893 if (tp->t_flags & TF_NEEDFIN) { 894 tp->t_state = TCPS_FIN_WAIT_1; 895 tp->t_flags &= ~TF_NEEDFIN; 896 } else { 897 tp->t_state = TCPS_ESTABLISHED; 898 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 899 } 900 tp->t_flags |= TF_NEEDSYN; 901 } else 902 tp->t_state = TCPS_SYN_RECEIVED; 903 } else { 904 /* CC.NEW or no option => invalidate cache */ 905 taop->tao_cc = 0; 906 tp->t_state = TCPS_SYN_RECEIVED; 907 } 908 } 909 910 trimthenstep6: 911 /* 912 * Advance ti->ti_seq to correspond to first data byte. 913 * If data, trim to stay within window, 914 * dropping FIN if necessary. 915 */ 916 ti->ti_seq++; 917 if (ti->ti_len > tp->rcv_wnd) { 918 todrop = ti->ti_len - tp->rcv_wnd; 919 m_adj(m, -todrop); 920 ti->ti_len = tp->rcv_wnd; 921 tiflags &= ~TH_FIN; 922 tcpstat.tcps_rcvpackafterwin++; 923 tcpstat.tcps_rcvbyteafterwin += todrop; 924 } 925 tp->snd_wl1 = ti->ti_seq - 1; 926 tp->rcv_up = ti->ti_seq; 927 /* 928 * Client side of transaction: already sent SYN and data. 929 * If the remote host used T/TCP to validate the SYN, 930 * our data will be ACK'd; if so, enter normal data segment 931 * processing in the middle of step 5, ack processing. 932 * Otherwise, goto step 6. 933 */ 934 if (tiflags & TH_ACK) 935 goto process_ACK; 936 goto step6; 937 /* 938 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 939 * if segment contains a SYN and CC [not CC.NEW] option: 940 * if state == TIME_WAIT and connection duration > MSL, 941 * drop packet and send RST; 942 * 943 * if SEG.CC > CCrecv then is new SYN, and can implicitly 944 * ack the FIN (and data) in retransmission queue. 945 * Complete close and delete TCPCB. Then reprocess 946 * segment, hoping to find new TCPCB in LISTEN state; 947 * 948 * else must be old SYN; drop it. 949 * else do normal processing. 950 */ 951 case TCPS_LAST_ACK: 952 case TCPS_CLOSING: 953 case TCPS_TIME_WAIT: 954 if ((tiflags & TH_SYN) && 955 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 956 if (tp->t_state == TCPS_TIME_WAIT && 957 tp->t_duration > TCPTV_MSL) 958 goto dropwithreset; 959 if (CC_GT(to.to_cc, tp->cc_recv)) { 960 tp = tcp_close(tp); 961 goto findpcb; 962 } 963 else 964 goto drop; 965 } 966 break; /* continue normal processing */ 967 } 968 969 /* 970 * States other than LISTEN or SYN_SENT. 971 * First check timestamp, if present. 972 * Then check the connection count, if present. 973 * Then check that at least some bytes of segment are within 974 * receive window. If segment begins before rcv_nxt, 975 * drop leading data (and SYN); if nothing left, just ack. 976 * 977 * RFC 1323 PAWS: If we have a timestamp reply on this segment 978 * and it's less than ts_recent, drop it. 979 */ 980 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 981 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 982 983 /* Check to see if ts_recent is over 24 days old. */ 984 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 985 /* 986 * Invalidate ts_recent. If this segment updates 987 * ts_recent, the age will be reset later and ts_recent 988 * will get a valid value. If it does not, setting 989 * ts_recent to zero will at least satisfy the 990 * requirement that zero be placed in the timestamp 991 * echo reply when ts_recent isn't valid. The 992 * age isn't reset until we get a valid ts_recent 993 * because we don't want out-of-order segments to be 994 * dropped when ts_recent is old. 995 */ 996 tp->ts_recent = 0; 997 } else { 998 tcpstat.tcps_rcvduppack++; 999 tcpstat.tcps_rcvdupbyte += ti->ti_len; 1000 tcpstat.tcps_pawsdrop++; 1001 goto dropafterack; 1002 } 1003 } 1004 1005 /* 1006 * T/TCP mechanism 1007 * If T/TCP was negotiated and the segment doesn't have CC, 1008 * or if its CC is wrong then drop the segment. 1009 * RST segments do not have to comply with this. 1010 */ 1011 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1012 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 1013 (tiflags & TH_RST) == 0) 1014 goto dropafterack; 1015 1016 todrop = tp->rcv_nxt - ti->ti_seq; 1017 if (todrop > 0) { 1018 if (tiflags & TH_SYN) { 1019 tiflags &= ~TH_SYN; 1020 ti->ti_seq++; 1021 if (ti->ti_urp > 1) 1022 ti->ti_urp--; 1023 else 1024 tiflags &= ~TH_URG; 1025 todrop--; 1026 } 1027 /* 1028 * Following if statement from Stevens, vol. 2, p. 960. 1029 */ 1030 if (todrop > ti->ti_len 1031 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1032 /* 1033 * Any valid FIN must be to the left of the window. 1034 * At this point the FIN must be a duplicate or out 1035 * of sequence; drop it. 1036 */ 1037 tiflags &= ~TH_FIN; 1038 1039 /* 1040 * Send an ACK to resynchronize and drop any data. 1041 * But keep on processing for RST or ACK. 1042 */ 1043 tp->t_flags |= TF_ACKNOW; 1044 todrop = ti->ti_len; 1045 tcpstat.tcps_rcvduppack++; 1046 tcpstat.tcps_rcvdupbyte += todrop; 1047 } else { 1048 tcpstat.tcps_rcvpartduppack++; 1049 tcpstat.tcps_rcvpartdupbyte += todrop; 1050 } 1051 m_adj(m, todrop); 1052 ti->ti_seq += todrop; 1053 ti->ti_len -= todrop; 1054 if (ti->ti_urp > todrop) 1055 ti->ti_urp -= todrop; 1056 else { 1057 tiflags &= ~TH_URG; 1058 ti->ti_urp = 0; 1059 } 1060 } 1061 1062 /* 1063 * If new data are received on a connection after the 1064 * user processes are gone, then RST the other end. 1065 */ 1066 if ((so->so_state & SS_NOFDREF) && 1067 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1068 tp = tcp_close(tp); 1069 tcpstat.tcps_rcvafterclose++; 1070 goto dropwithreset; 1071 } 1072 1073 /* 1074 * If segment ends after window, drop trailing data 1075 * (and PUSH and FIN); if nothing left, just ACK. 1076 */ 1077 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1078 if (todrop > 0) { 1079 tcpstat.tcps_rcvpackafterwin++; 1080 if (todrop >= ti->ti_len) { 1081 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1082 /* 1083 * If a new connection request is received 1084 * while in TIME_WAIT, drop the old connection 1085 * and start over if the sequence numbers 1086 * are above the previous ones. 1087 */ 1088 if (tiflags & TH_SYN && 1089 tp->t_state == TCPS_TIME_WAIT && 1090 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1091 iss = tp->rcv_nxt + TCP_ISSINCR; 1092 tp = tcp_close(tp); 1093 goto findpcb; 1094 } 1095 /* 1096 * If window is closed can only take segments at 1097 * window edge, and have to drop data and PUSH from 1098 * incoming segments. Continue processing, but 1099 * remember to ack. Otherwise, drop segment 1100 * and ack. 1101 */ 1102 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1103 tp->t_flags |= TF_ACKNOW; 1104 tcpstat.tcps_rcvwinprobe++; 1105 } else 1106 goto dropafterack; 1107 } else 1108 tcpstat.tcps_rcvbyteafterwin += todrop; 1109 m_adj(m, -todrop); 1110 ti->ti_len -= todrop; 1111 tiflags &= ~(TH_PUSH|TH_FIN); 1112 } 1113 1114 /* 1115 * If last ACK falls within this segment's sequence numbers, 1116 * record its timestamp. 1117 * NOTE that the test is modified according to the latest 1118 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1119 */ 1120 if ((to.to_flag & TOF_TS) != 0 && 1121 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1122 tp->ts_recent_age = tcp_now; 1123 tp->ts_recent = to.to_tsval; 1124 } 1125 1126 /* 1127 * If the RST bit is set examine the state: 1128 * SYN_RECEIVED STATE: 1129 * If passive open, return to LISTEN state. 1130 * If active open, inform user that connection was refused. 1131 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1132 * Inform user that connection was reset, and close tcb. 1133 * CLOSING, LAST_ACK, TIME_WAIT STATES 1134 * Close the tcb. 1135 */ 1136 if (tiflags&TH_RST) switch (tp->t_state) { 1137 1138 case TCPS_SYN_RECEIVED: 1139 so->so_error = ECONNREFUSED; 1140 goto close; 1141 1142 case TCPS_ESTABLISHED: 1143 case TCPS_FIN_WAIT_1: 1144 case TCPS_FIN_WAIT_2: 1145 case TCPS_CLOSE_WAIT: 1146 so->so_error = ECONNRESET; 1147 close: 1148 tp->t_state = TCPS_CLOSED; 1149 tcpstat.tcps_drops++; 1150 tp = tcp_close(tp); 1151 goto drop; 1152 1153 case TCPS_CLOSING: 1154 case TCPS_LAST_ACK: 1155 case TCPS_TIME_WAIT: 1156 tp = tcp_close(tp); 1157 goto drop; 1158 } 1159 1160 /* 1161 * If a SYN is in the window, then this is an 1162 * error and we send an RST and drop the connection. 1163 */ 1164 if (tiflags & TH_SYN) { 1165 tp = tcp_drop(tp, ECONNRESET); 1166 goto dropwithreset; 1167 } 1168 1169 /* 1170 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1171 * flag is on (half-synchronized state), then queue data for 1172 * later processing; else drop segment and return. 1173 */ 1174 if ((tiflags & TH_ACK) == 0) { 1175 if (tp->t_state == TCPS_SYN_RECEIVED || 1176 (tp->t_flags & TF_NEEDSYN)) 1177 goto step6; 1178 else 1179 goto drop; 1180 } 1181 1182 /* 1183 * Ack processing. 1184 */ 1185 switch (tp->t_state) { 1186 1187 /* 1188 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1189 * ESTABLISHED state and continue processing. 1190 * The ACK was checked above. 1191 */ 1192 case TCPS_SYN_RECEIVED: 1193 1194 tcpstat.tcps_connects++; 1195 soisconnected(so); 1196 /* Do window scaling? */ 1197 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1198 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1199 tp->snd_scale = tp->requested_s_scale; 1200 tp->rcv_scale = tp->request_r_scale; 1201 } 1202 /* 1203 * Upon successful completion of 3-way handshake, 1204 * update cache.CC if it was undefined, pass any queued 1205 * data to the user, and advance state appropriately. 1206 */ 1207 if ((taop = tcp_gettaocache(inp)) != NULL && 1208 taop->tao_cc == 0) 1209 taop->tao_cc = tp->cc_recv; 1210 1211 /* 1212 * Make transitions: 1213 * SYN-RECEIVED -> ESTABLISHED 1214 * SYN-RECEIVED* -> FIN-WAIT-1 1215 */ 1216 if (tp->t_flags & TF_NEEDFIN) { 1217 tp->t_state = TCPS_FIN_WAIT_1; 1218 tp->t_flags &= ~TF_NEEDFIN; 1219 } else { 1220 tp->t_state = TCPS_ESTABLISHED; 1221 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1222 } 1223 /* 1224 * If segment contains data or ACK, will call tcp_reass() 1225 * later; if not, do so now to pass queued data to user. 1226 */ 1227 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1228 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1229 (struct mbuf *)0); 1230 tp->snd_wl1 = ti->ti_seq - 1; 1231 /* fall into ... */ 1232 1233 /* 1234 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1235 * ACKs. If the ack is in the range 1236 * tp->snd_una < ti->ti_ack <= tp->snd_max 1237 * then advance tp->snd_una to ti->ti_ack and drop 1238 * data from the retransmission queue. If this ACK reflects 1239 * more up to date window information we update our window information. 1240 */ 1241 case TCPS_ESTABLISHED: 1242 case TCPS_FIN_WAIT_1: 1243 case TCPS_FIN_WAIT_2: 1244 case TCPS_CLOSE_WAIT: 1245 case TCPS_CLOSING: 1246 case TCPS_LAST_ACK: 1247 case TCPS_TIME_WAIT: 1248 1249 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1250 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1251 tcpstat.tcps_rcvdupack++; 1252 /* 1253 * If we have outstanding data (other than 1254 * a window probe), this is a completely 1255 * duplicate ack (ie, window info didn't 1256 * change), the ack is the biggest we've 1257 * seen and we've seen exactly our rexmt 1258 * threshhold of them, assume a packet 1259 * has been dropped and retransmit it. 1260 * Kludge snd_nxt & the congestion 1261 * window so we send only this one 1262 * packet. 1263 * 1264 * We know we're losing at the current 1265 * window size so do congestion avoidance 1266 * (set ssthresh to half the current window 1267 * and pull our congestion window back to 1268 * the new ssthresh). 1269 * 1270 * Dup acks mean that packets have left the 1271 * network (they're now cached at the receiver) 1272 * so bump cwnd by the amount in the receiver 1273 * to keep a constant cwnd packets in the 1274 * network. 1275 */ 1276 if (tp->t_timer[TCPT_REXMT] == 0 || 1277 ti->ti_ack != tp->snd_una) 1278 tp->t_dupacks = 0; 1279 else if (++tp->t_dupacks == tcprexmtthresh) { 1280 tcp_seq onxt = tp->snd_nxt; 1281 u_int win = 1282 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1283 tp->t_maxseg; 1284 1285 if (win < 2) 1286 win = 2; 1287 tp->snd_ssthresh = win * tp->t_maxseg; 1288 tp->t_timer[TCPT_REXMT] = 0; 1289 tp->t_rtt = 0; 1290 tp->snd_nxt = ti->ti_ack; 1291 tp->snd_cwnd = tp->t_maxseg; 1292 (void) tcp_output(tp); 1293 tp->snd_cwnd = tp->snd_ssthresh + 1294 tp->t_maxseg * tp->t_dupacks; 1295 if (SEQ_GT(onxt, tp->snd_nxt)) 1296 tp->snd_nxt = onxt; 1297 goto drop; 1298 } else if (tp->t_dupacks > tcprexmtthresh) { 1299 tp->snd_cwnd += tp->t_maxseg; 1300 (void) tcp_output(tp); 1301 goto drop; 1302 } 1303 } else 1304 tp->t_dupacks = 0; 1305 break; 1306 } 1307 /* 1308 * If the congestion window was inflated to account 1309 * for the other side's cached packets, retract it. 1310 */ 1311 if (tp->t_dupacks >= tcprexmtthresh && 1312 tp->snd_cwnd > tp->snd_ssthresh) 1313 tp->snd_cwnd = tp->snd_ssthresh; 1314 tp->t_dupacks = 0; 1315 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1316 tcpstat.tcps_rcvacktoomuch++; 1317 goto dropafterack; 1318 } 1319 /* 1320 * If we reach this point, ACK is not a duplicate, 1321 * i.e., it ACKs something we sent. 1322 */ 1323 if (tp->t_flags & TF_NEEDSYN) { 1324 /* 1325 * T/TCP: Connection was half-synchronized, and our 1326 * SYN has been ACK'd (so connection is now fully 1327 * synchronized). Go to non-starred state, 1328 * increment snd_una for ACK of SYN, and check if 1329 * we can do window scaling. 1330 */ 1331 tp->t_flags &= ~TF_NEEDSYN; 1332 tp->snd_una++; 1333 /* Do window scaling? */ 1334 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1335 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1336 tp->snd_scale = tp->requested_s_scale; 1337 tp->rcv_scale = tp->request_r_scale; 1338 } 1339 } 1340 1341 process_ACK: 1342 acked = ti->ti_ack - tp->snd_una; 1343 tcpstat.tcps_rcvackpack++; 1344 tcpstat.tcps_rcvackbyte += acked; 1345 1346 /* 1347 * If we have a timestamp reply, update smoothed 1348 * round trip time. If no timestamp is present but 1349 * transmit timer is running and timed sequence 1350 * number was acked, update smoothed round trip time. 1351 * Since we now have an rtt measurement, cancel the 1352 * timer backoff (cf., Phil Karn's retransmit alg.). 1353 * Recompute the initial retransmit timer. 1354 */ 1355 if (to.to_flag & TOF_TS) 1356 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1357 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1358 tcp_xmit_timer(tp,tp->t_rtt); 1359 1360 /* 1361 * If all outstanding data is acked, stop retransmit 1362 * timer and remember to restart (more output or persist). 1363 * If there is more data to be acked, restart retransmit 1364 * timer, using current (possibly backed-off) value. 1365 */ 1366 if (ti->ti_ack == tp->snd_max) { 1367 tp->t_timer[TCPT_REXMT] = 0; 1368 needoutput = 1; 1369 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1370 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1371 1372 /* 1373 * If no data (only SYN) was ACK'd, 1374 * skip rest of ACK processing. 1375 */ 1376 if (acked == 0) 1377 goto step6; 1378 1379 /* 1380 * When new data is acked, open the congestion window. 1381 * If the window gives us less than ssthresh packets 1382 * in flight, open exponentially (maxseg per packet). 1383 * Otherwise open linearly: maxseg per window 1384 * (maxseg^2 / cwnd per packet). 1385 */ 1386 { 1387 register u_int cw = tp->snd_cwnd; 1388 register u_int incr = tp->t_maxseg; 1389 1390 if (cw > tp->snd_ssthresh) 1391 incr = incr * incr / cw; 1392 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1393 } 1394 if (acked > so->so_snd.sb_cc) { 1395 tp->snd_wnd -= so->so_snd.sb_cc; 1396 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1397 ourfinisacked = 1; 1398 } else { 1399 sbdrop(&so->so_snd, acked); 1400 tp->snd_wnd -= acked; 1401 ourfinisacked = 0; 1402 } 1403 sowwakeup(so); 1404 tp->snd_una = ti->ti_ack; 1405 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1406 tp->snd_nxt = tp->snd_una; 1407 1408 switch (tp->t_state) { 1409 1410 /* 1411 * In FIN_WAIT_1 STATE in addition to the processing 1412 * for the ESTABLISHED state if our FIN is now acknowledged 1413 * then enter FIN_WAIT_2. 1414 */ 1415 case TCPS_FIN_WAIT_1: 1416 if (ourfinisacked) { 1417 /* 1418 * If we can't receive any more 1419 * data, then closing user can proceed. 1420 * Starting the timer is contrary to the 1421 * specification, but if we don't get a FIN 1422 * we'll hang forever. 1423 */ 1424 if (so->so_state & SS_CANTRCVMORE) { 1425 soisdisconnected(so); 1426 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1427 } 1428 tp->t_state = TCPS_FIN_WAIT_2; 1429 } 1430 break; 1431 1432 /* 1433 * In CLOSING STATE in addition to the processing for 1434 * the ESTABLISHED state if the ACK acknowledges our FIN 1435 * then enter the TIME-WAIT state, otherwise ignore 1436 * the segment. 1437 */ 1438 case TCPS_CLOSING: 1439 if (ourfinisacked) { 1440 tp->t_state = TCPS_TIME_WAIT; 1441 tcp_canceltimers(tp); 1442 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1443 if (tp->cc_recv != 0 && 1444 tp->t_duration < TCPTV_MSL) 1445 tp->t_timer[TCPT_2MSL] = 1446 tp->t_rxtcur * TCPTV_TWTRUNC; 1447 else 1448 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1449 soisdisconnected(so); 1450 } 1451 break; 1452 1453 /* 1454 * In LAST_ACK, we may still be waiting for data to drain 1455 * and/or to be acked, as well as for the ack of our FIN. 1456 * If our FIN is now acknowledged, delete the TCB, 1457 * enter the closed state and return. 1458 */ 1459 case TCPS_LAST_ACK: 1460 if (ourfinisacked) { 1461 tp = tcp_close(tp); 1462 goto drop; 1463 } 1464 break; 1465 1466 /* 1467 * In TIME_WAIT state the only thing that should arrive 1468 * is a retransmission of the remote FIN. Acknowledge 1469 * it and restart the finack timer. 1470 */ 1471 case TCPS_TIME_WAIT: 1472 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1473 goto dropafterack; 1474 } 1475 } 1476 1477 step6: 1478 /* 1479 * Update window information. 1480 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1481 */ 1482 if ((tiflags & TH_ACK) && 1483 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1484 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1485 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1486 /* keep track of pure window updates */ 1487 if (ti->ti_len == 0 && 1488 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1489 tcpstat.tcps_rcvwinupd++; 1490 tp->snd_wnd = tiwin; 1491 tp->snd_wl1 = ti->ti_seq; 1492 tp->snd_wl2 = ti->ti_ack; 1493 if (tp->snd_wnd > tp->max_sndwnd) 1494 tp->max_sndwnd = tp->snd_wnd; 1495 needoutput = 1; 1496 } 1497 1498 /* 1499 * Process segments with URG. 1500 */ 1501 if ((tiflags & TH_URG) && ti->ti_urp && 1502 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1503 /* 1504 * This is a kludge, but if we receive and accept 1505 * random urgent pointers, we'll crash in 1506 * soreceive. It's hard to imagine someone 1507 * actually wanting to send this much urgent data. 1508 */ 1509 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1510 ti->ti_urp = 0; /* XXX */ 1511 tiflags &= ~TH_URG; /* XXX */ 1512 goto dodata; /* XXX */ 1513 } 1514 /* 1515 * If this segment advances the known urgent pointer, 1516 * then mark the data stream. This should not happen 1517 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1518 * a FIN has been received from the remote side. 1519 * In these states we ignore the URG. 1520 * 1521 * According to RFC961 (Assigned Protocols), 1522 * the urgent pointer points to the last octet 1523 * of urgent data. We continue, however, 1524 * to consider it to indicate the first octet 1525 * of data past the urgent section as the original 1526 * spec states (in one of two places). 1527 */ 1528 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1529 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1530 so->so_oobmark = so->so_rcv.sb_cc + 1531 (tp->rcv_up - tp->rcv_nxt) - 1; 1532 if (so->so_oobmark == 0) 1533 so->so_state |= SS_RCVATMARK; 1534 sohasoutofband(so); 1535 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1536 } 1537 /* 1538 * Remove out of band data so doesn't get presented to user. 1539 * This can happen independent of advancing the URG pointer, 1540 * but if two URG's are pending at once, some out-of-band 1541 * data may creep in... ick. 1542 */ 1543 if (ti->ti_urp <= (u_long)ti->ti_len 1544 #ifdef SO_OOBINLINE 1545 && (so->so_options & SO_OOBINLINE) == 0 1546 #endif 1547 ) 1548 tcp_pulloutofband(so, ti, m); 1549 } else 1550 /* 1551 * If no out of band data is expected, 1552 * pull receive urgent pointer along 1553 * with the receive window. 1554 */ 1555 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1556 tp->rcv_up = tp->rcv_nxt; 1557 dodata: /* XXX */ 1558 1559 /* 1560 * Process the segment text, merging it into the TCP sequencing queue, 1561 * and arranging for acknowledgment of receipt if necessary. 1562 * This process logically involves adjusting tp->rcv_wnd as data 1563 * is presented to the user (this happens in tcp_usrreq.c, 1564 * case PRU_RCVD). If a FIN has already been received on this 1565 * connection then we just ignore the text. 1566 */ 1567 if ((ti->ti_len || (tiflags&TH_FIN)) && 1568 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1569 TCP_REASS(tp, ti, m, so, tiflags); 1570 /* 1571 * Note the amount of data that peer has sent into 1572 * our window, in order to estimate the sender's 1573 * buffer size. 1574 */ 1575 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1576 } else { 1577 m_freem(m); 1578 tiflags &= ~TH_FIN; 1579 } 1580 1581 /* 1582 * If FIN is received ACK the FIN and let the user know 1583 * that the connection is closing. 1584 */ 1585 if (tiflags & TH_FIN) { 1586 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1587 socantrcvmore(so); 1588 /* 1589 * If connection is half-synchronized 1590 * (ie NEEDSYN flag on) then delay ACK, 1591 * so it may be piggybacked when SYN is sent. 1592 * Otherwise, since we received a FIN then no 1593 * more input can be expected, send ACK now. 1594 */ 1595 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) 1596 tp->t_flags |= TF_DELACK; 1597 else 1598 tp->t_flags |= TF_ACKNOW; 1599 tp->rcv_nxt++; 1600 } 1601 switch (tp->t_state) { 1602 1603 /* 1604 * In SYN_RECEIVED and ESTABLISHED STATES 1605 * enter the CLOSE_WAIT state. 1606 */ 1607 case TCPS_SYN_RECEIVED: 1608 case TCPS_ESTABLISHED: 1609 tp->t_state = TCPS_CLOSE_WAIT; 1610 break; 1611 1612 /* 1613 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1614 * enter the CLOSING state. 1615 */ 1616 case TCPS_FIN_WAIT_1: 1617 tp->t_state = TCPS_CLOSING; 1618 break; 1619 1620 /* 1621 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1622 * starting the time-wait timer, turning off the other 1623 * standard timers. 1624 */ 1625 case TCPS_FIN_WAIT_2: 1626 tp->t_state = TCPS_TIME_WAIT; 1627 tcp_canceltimers(tp); 1628 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1629 if (tp->cc_recv != 0 && 1630 tp->t_duration < TCPTV_MSL) { 1631 tp->t_timer[TCPT_2MSL] = 1632 tp->t_rxtcur * TCPTV_TWTRUNC; 1633 /* For transaction client, force ACK now. */ 1634 tp->t_flags |= TF_ACKNOW; 1635 } 1636 else 1637 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1638 soisdisconnected(so); 1639 break; 1640 1641 /* 1642 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1643 */ 1644 case TCPS_TIME_WAIT: 1645 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1646 break; 1647 } 1648 } 1649 #ifdef TCPDEBUG 1650 if (so->so_options & SO_DEBUG) 1651 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1652 #endif 1653 1654 /* 1655 * Return any desired output. 1656 */ 1657 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1658 (void) tcp_output(tp); 1659 return; 1660 1661 dropafterack: 1662 /* 1663 * Generate an ACK dropping incoming segment if it occupies 1664 * sequence space, where the ACK reflects our state. 1665 */ 1666 if (tiflags & TH_RST) 1667 goto drop; 1668 #ifdef TCPDEBUG 1669 if (so->so_options & SO_DEBUG) 1670 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1671 #endif 1672 m_freem(m); 1673 tp->t_flags |= TF_ACKNOW; 1674 (void) tcp_output(tp); 1675 return; 1676 1677 dropwithreset: 1678 /* 1679 * Generate a RST, dropping incoming segment. 1680 * Make ACK acceptable to originator of segment. 1681 * Don't bother to respond if destination was broadcast/multicast. 1682 */ 1683 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1684 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1685 goto drop; 1686 #ifdef TCPDEBUG 1687 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1688 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1689 #endif 1690 if (tiflags & TH_ACK) 1691 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1692 else { 1693 if (tiflags & TH_SYN) 1694 ti->ti_len++; 1695 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1696 TH_RST|TH_ACK); 1697 } 1698 /* destroy temporarily created socket */ 1699 if (dropsocket) 1700 (void) soabort(so); 1701 return; 1702 1703 drop: 1704 /* 1705 * Drop space held by incoming segment and return. 1706 */ 1707 #ifdef TCPDEBUG 1708 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1709 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1710 #endif 1711 m_freem(m); 1712 /* destroy temporarily created socket */ 1713 if (dropsocket) 1714 (void) soabort(so); 1715 return; 1716 } 1717 1718 static void 1719 tcp_dooptions(tp, cp, cnt, ti, to) 1720 struct tcpcb *tp; 1721 u_char *cp; 1722 int cnt; 1723 struct tcpiphdr *ti; 1724 struct tcpopt *to; 1725 { 1726 u_short mss = 0; 1727 int opt, optlen; 1728 1729 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1730 opt = cp[0]; 1731 if (opt == TCPOPT_EOL) 1732 break; 1733 if (opt == TCPOPT_NOP) 1734 optlen = 1; 1735 else { 1736 optlen = cp[1]; 1737 if (optlen <= 0) 1738 break; 1739 } 1740 switch (opt) { 1741 1742 default: 1743 continue; 1744 1745 case TCPOPT_MAXSEG: 1746 if (optlen != TCPOLEN_MAXSEG) 1747 continue; 1748 if (!(ti->ti_flags & TH_SYN)) 1749 continue; 1750 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1751 NTOHS(mss); 1752 break; 1753 1754 case TCPOPT_WINDOW: 1755 if (optlen != TCPOLEN_WINDOW) 1756 continue; 1757 if (!(ti->ti_flags & TH_SYN)) 1758 continue; 1759 tp->t_flags |= TF_RCVD_SCALE; 1760 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1761 break; 1762 1763 case TCPOPT_TIMESTAMP: 1764 if (optlen != TCPOLEN_TIMESTAMP) 1765 continue; 1766 to->to_flag |= TOF_TS; 1767 bcopy((char *)cp + 2, 1768 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1769 NTOHL(to->to_tsval); 1770 bcopy((char *)cp + 6, 1771 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1772 NTOHL(to->to_tsecr); 1773 1774 /* 1775 * A timestamp received in a SYN makes 1776 * it ok to send timestamp requests and replies. 1777 */ 1778 if (ti->ti_flags & TH_SYN) { 1779 tp->t_flags |= TF_RCVD_TSTMP; 1780 tp->ts_recent = to->to_tsval; 1781 tp->ts_recent_age = tcp_now; 1782 } 1783 break; 1784 case TCPOPT_CC: 1785 if (optlen != TCPOLEN_CC) 1786 continue; 1787 to->to_flag |= TOF_CC; 1788 bcopy((char *)cp + 2, 1789 (char *)&to->to_cc, sizeof(to->to_cc)); 1790 NTOHL(to->to_cc); 1791 /* 1792 * A CC or CC.new option received in a SYN makes 1793 * it ok to send CC in subsequent segments. 1794 */ 1795 if (ti->ti_flags & TH_SYN) 1796 tp->t_flags |= TF_RCVD_CC; 1797 break; 1798 case TCPOPT_CCNEW: 1799 if (optlen != TCPOLEN_CC) 1800 continue; 1801 if (!(ti->ti_flags & TH_SYN)) 1802 continue; 1803 to->to_flag |= TOF_CCNEW; 1804 bcopy((char *)cp + 2, 1805 (char *)&to->to_cc, sizeof(to->to_cc)); 1806 NTOHL(to->to_cc); 1807 /* 1808 * A CC or CC.new option received in a SYN makes 1809 * it ok to send CC in subsequent segments. 1810 */ 1811 tp->t_flags |= TF_RCVD_CC; 1812 break; 1813 case TCPOPT_CCECHO: 1814 if (optlen != TCPOLEN_CC) 1815 continue; 1816 if (!(ti->ti_flags & TH_SYN)) 1817 continue; 1818 to->to_flag |= TOF_CCECHO; 1819 bcopy((char *)cp + 2, 1820 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1821 NTOHL(to->to_ccecho); 1822 break; 1823 } 1824 } 1825 if (ti->ti_flags & TH_SYN) 1826 tcp_mss(tp, mss); /* sets t_maxseg */ 1827 } 1828 1829 /* 1830 * Pull out of band byte out of a segment so 1831 * it doesn't appear in the user's data queue. 1832 * It is still reflected in the segment length for 1833 * sequencing purposes. 1834 */ 1835 static void 1836 tcp_pulloutofband(so, ti, m) 1837 struct socket *so; 1838 struct tcpiphdr *ti; 1839 register struct mbuf *m; 1840 { 1841 int cnt = ti->ti_urp - 1; 1842 1843 while (cnt >= 0) { 1844 if (m->m_len > cnt) { 1845 char *cp = mtod(m, caddr_t) + cnt; 1846 struct tcpcb *tp = sototcpcb(so); 1847 1848 tp->t_iobc = *cp; 1849 tp->t_oobflags |= TCPOOB_HAVEDATA; 1850 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1851 m->m_len--; 1852 return; 1853 } 1854 cnt -= m->m_len; 1855 m = m->m_next; 1856 if (m == 0) 1857 break; 1858 } 1859 panic("tcp_pulloutofband"); 1860 } 1861 1862 /* 1863 * Collect new round-trip time estimate 1864 * and update averages and current timeout. 1865 */ 1866 static void 1867 tcp_xmit_timer(tp, rtt) 1868 register struct tcpcb *tp; 1869 short rtt; 1870 { 1871 register int delta; 1872 1873 tcpstat.tcps_rttupdated++; 1874 tp->t_rttupdated++; 1875 if (tp->t_srtt != 0) { 1876 /* 1877 * srtt is stored as fixed point with 5 bits after the 1878 * binary point (i.e., scaled by 8). The following magic 1879 * is equivalent to the smoothing algorithm in rfc793 with 1880 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1881 * point). Adjust rtt to origin 0. 1882 */ 1883 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1884 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1885 1886 if ((tp->t_srtt += delta) <= 0) 1887 tp->t_srtt = 1; 1888 1889 /* 1890 * We accumulate a smoothed rtt variance (actually, a 1891 * smoothed mean difference), then set the retransmit 1892 * timer to smoothed rtt + 4 times the smoothed variance. 1893 * rttvar is stored as fixed point with 4 bits after the 1894 * binary point (scaled by 16). The following is 1895 * equivalent to rfc793 smoothing with an alpha of .75 1896 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1897 * rfc793's wired-in beta. 1898 */ 1899 if (delta < 0) 1900 delta = -delta; 1901 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1902 if ((tp->t_rttvar += delta) <= 0) 1903 tp->t_rttvar = 1; 1904 } else { 1905 /* 1906 * No rtt measurement yet - use the unsmoothed rtt. 1907 * Set the variance to half the rtt (so our first 1908 * retransmit happens at 3*rtt). 1909 */ 1910 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1911 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1912 } 1913 tp->t_rtt = 0; 1914 tp->t_rxtshift = 0; 1915 1916 /* 1917 * the retransmit should happen at rtt + 4 * rttvar. 1918 * Because of the way we do the smoothing, srtt and rttvar 1919 * will each average +1/2 tick of bias. When we compute 1920 * the retransmit timer, we want 1/2 tick of rounding and 1921 * 1 extra tick because of +-1/2 tick uncertainty in the 1922 * firing of the timer. The bias will give us exactly the 1923 * 1.5 tick we need. But, because the bias is 1924 * statistical, we have to test that we don't drop below 1925 * the minimum feasible timer (which is 2 ticks). 1926 */ 1927 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1928 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1929 1930 /* 1931 * We received an ack for a packet that wasn't retransmitted; 1932 * it is probably safe to discard any error indications we've 1933 * received recently. This isn't quite right, but close enough 1934 * for now (a route might have failed after we sent a segment, 1935 * and the return path might not be symmetrical). 1936 */ 1937 tp->t_softerror = 0; 1938 } 1939 1940 /* 1941 * Determine a reasonable value for maxseg size. 1942 * If the route is known, check route for mtu. 1943 * If none, use an mss that can be handled on the outgoing 1944 * interface without forcing IP to fragment; if bigger than 1945 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1946 * to utilize large mbufs. If no route is found, route has no mtu, 1947 * or the destination isn't local, use a default, hopefully conservative 1948 * size (usually 512 or the default IP max size, but no more than the mtu 1949 * of the interface), as we can't discover anything about intervening 1950 * gateways or networks. We also initialize the congestion/slow start 1951 * window to be a single segment if the destination isn't local. 1952 * While looking at the routing entry, we also initialize other path-dependent 1953 * parameters from pre-set or cached values in the routing entry. 1954 * 1955 * Also take into account the space needed for options that we 1956 * send regularly. Make maxseg shorter by that amount to assure 1957 * that we can send maxseg amount of data even when the options 1958 * are present. Store the upper limit of the length of options plus 1959 * data in maxopd. 1960 * 1961 * NOTE that this routine is only called when we process an incoming 1962 * segment, for outgoing segments only tcp_mssopt is called. 1963 * 1964 * In case of T/TCP, we call this routine during implicit connection 1965 * setup as well (offer = -1), to initialize maxseg from the cached 1966 * MSS of our peer. 1967 */ 1968 void 1969 tcp_mss(tp, offer) 1970 struct tcpcb *tp; 1971 int offer; 1972 { 1973 register struct rtentry *rt; 1974 struct ifnet *ifp; 1975 register int rtt, mss; 1976 u_long bufsize; 1977 struct inpcb *inp; 1978 struct socket *so; 1979 struct rmxp_tao *taop; 1980 int origoffer = offer; 1981 1982 inp = tp->t_inpcb; 1983 if ((rt = tcp_rtlookup(inp)) == NULL) { 1984 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1985 return; 1986 } 1987 ifp = rt->rt_ifp; 1988 so = inp->inp_socket; 1989 1990 taop = rmx_taop(rt->rt_rmx); 1991 /* 1992 * Offer == -1 means that we didn't receive SYN yet, 1993 * use cached value in that case; 1994 */ 1995 if (offer == -1) 1996 offer = taop->tao_mssopt; 1997 /* 1998 * Offer == 0 means that there was no MSS on the SYN segment, 1999 * in this case we use tcp_mssdflt. 2000 */ 2001 if (offer == 0) 2002 offer = tcp_mssdflt; 2003 else 2004 /* 2005 * Sanity check: make sure that maxopd will be large 2006 * enough to allow some data on segments even is the 2007 * all the option space is used (40bytes). Otherwise 2008 * funny things may happen in tcp_output. 2009 */ 2010 offer = max(offer, 64); 2011 taop->tao_mssopt = offer; 2012 2013 /* 2014 * While we're here, check if there's an initial rtt 2015 * or rttvar. Convert from the route-table units 2016 * to scaled multiples of the slow timeout timer. 2017 */ 2018 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2019 /* 2020 * XXX the lock bit for RTT indicates that the value 2021 * is also a minimum value; this is subject to time. 2022 */ 2023 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2024 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2025 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2026 tcpstat.tcps_usedrtt++; 2027 if (rt->rt_rmx.rmx_rttvar) { 2028 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2029 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2030 tcpstat.tcps_usedrttvar++; 2031 } else { 2032 /* default variation is +- 1 rtt */ 2033 tp->t_rttvar = 2034 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2035 } 2036 TCPT_RANGESET(tp->t_rxtcur, 2037 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2038 tp->t_rttmin, TCPTV_REXMTMAX); 2039 } 2040 /* 2041 * if there's an mtu associated with the route, use it 2042 */ 2043 if (rt->rt_rmx.rmx_mtu) 2044 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2045 else 2046 { 2047 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2048 if (!in_localaddr(inp->inp_faddr)) 2049 mss = min(mss, tcp_mssdflt); 2050 } 2051 mss = min(mss, offer); 2052 /* 2053 * maxopd stores the maximum length of data AND options 2054 * in a segment; maxseg is the amount of data in a normal 2055 * segment. We need to store this value (maxopd) apart 2056 * from maxseg, because now every segment carries options 2057 * and thus we normally have somewhat less data in segments. 2058 */ 2059 tp->t_maxopd = mss; 2060 2061 /* 2062 * In case of T/TCP, origoffer==-1 indicates, that no segments 2063 * were received yet. In this case we just guess, otherwise 2064 * we do the same as before T/TCP. 2065 */ 2066 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2067 (origoffer == -1 || 2068 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2069 mss -= TCPOLEN_TSTAMP_APPA; 2070 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2071 (origoffer == -1 || 2072 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2073 mss -= TCPOLEN_CC_APPA; 2074 2075 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2076 if (mss > MCLBYTES) 2077 mss &= ~(MCLBYTES-1); 2078 #else 2079 if (mss > MCLBYTES) 2080 mss = mss / MCLBYTES * MCLBYTES; 2081 #endif 2082 /* 2083 * If there's a pipesize, change the socket buffer 2084 * to that size. Make the socket buffers an integral 2085 * number of mss units; if the mss is larger than 2086 * the socket buffer, decrease the mss. 2087 */ 2088 #ifdef RTV_SPIPE 2089 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2090 #endif 2091 bufsize = so->so_snd.sb_hiwat; 2092 if (bufsize < mss) 2093 mss = bufsize; 2094 else { 2095 bufsize = roundup(bufsize, mss); 2096 if (bufsize > sb_max) 2097 bufsize = sb_max; 2098 (void)sbreserve(&so->so_snd, bufsize); 2099 } 2100 tp->t_maxseg = mss; 2101 2102 #ifdef RTV_RPIPE 2103 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2104 #endif 2105 bufsize = so->so_rcv.sb_hiwat; 2106 if (bufsize > mss) { 2107 bufsize = roundup(bufsize, mss); 2108 if (bufsize > sb_max) 2109 bufsize = sb_max; 2110 (void)sbreserve(&so->so_rcv, bufsize); 2111 } 2112 /* 2113 * Don't force slow-start on local network. 2114 */ 2115 if (!in_localaddr(inp->inp_faddr)) 2116 tp->snd_cwnd = mss; 2117 2118 if (rt->rt_rmx.rmx_ssthresh) { 2119 /* 2120 * There's some sort of gateway or interface 2121 * buffer limit on the path. Use this to set 2122 * the slow start threshhold, but set the 2123 * threshold to no less than 2*mss. 2124 */ 2125 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2126 tcpstat.tcps_usedssthresh++; 2127 } 2128 } 2129 2130 /* 2131 * Determine the MSS option to send on an outgoing SYN. 2132 */ 2133 int 2134 tcp_mssopt(tp) 2135 struct tcpcb *tp; 2136 { 2137 struct rtentry *rt; 2138 2139 rt = tcp_rtlookup(tp->t_inpcb); 2140 if (rt == NULL) 2141 return tcp_mssdflt; 2142 2143 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2144 } 2145