1 /* $NetBSD: tcp_output.c,v 1.10 1994/06/29 06:38:42 cgd Exp $ */ 2 3 /* 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)tcp_output.c 8.3 (Berkeley) 12/30/93 36 */ 37 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/route.h> 48 49 #include <netinet/in.h> 50 #include <netinet/in_systm.h> 51 #include <netinet/ip.h> 52 #include <netinet/in_pcb.h> 53 #include <netinet/ip_var.h> 54 #include <netinet/tcp.h> 55 #define TCPOUTFLAGS 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 #include <netinet/tcp_debug.h> 62 63 #ifdef notyet 64 extern struct mbuf *m_copypack(); 65 #endif 66 67 68 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */ 69 70 /* 71 * Tcp output routine: figure out what should be sent and send it. 72 */ 73 int 74 tcp_output(tp) 75 register struct tcpcb *tp; 76 { 77 register struct socket *so = tp->t_inpcb->inp_socket; 78 register long len, win; 79 int off, flags, error; 80 register struct mbuf *m; 81 register struct tcpiphdr *ti; 82 u_char opt[MAX_TCPOPTLEN]; 83 unsigned optlen, hdrlen; 84 int idle, sendalot; 85 86 /* 87 * Determine length of data that should be transmitted, 88 * and flags that will be used. 89 * If there is some data or critical controls (SYN, RST) 90 * to send, then transmit; otherwise, investigate further. 91 */ 92 idle = (tp->snd_max == tp->snd_una); 93 if (idle && tp->t_idle >= tp->t_rxtcur) 94 /* 95 * We have been idle for "a while" and no acks are 96 * expected to clock out any data we send -- 97 * slow start to get ack "clock" running again. 98 */ 99 tp->snd_cwnd = tp->t_maxseg; 100 again: 101 sendalot = 0; 102 off = tp->snd_nxt - tp->snd_una; 103 win = min(tp->snd_wnd, tp->snd_cwnd); 104 105 flags = tcp_outflags[tp->t_state]; 106 /* 107 * If in persist timeout with window of 0, send 1 byte. 108 * Otherwise, if window is small but nonzero 109 * and timer expired, we will send what we can 110 * and go to transmit state. 111 */ 112 if (tp->t_force) { 113 if (win == 0) { 114 /* 115 * If we still have some data to send, then 116 * clear the FIN bit. Usually this would 117 * happen below when it realizes that we 118 * aren't sending all the data. However, 119 * if we have exactly 1 byte of unset data, 120 * then it won't clear the FIN bit below, 121 * and if we are in persist state, we wind 122 * up sending the packet without recording 123 * that we sent the FIN bit. 124 * 125 * We can't just blindly clear the FIN bit, 126 * because if we don't have any more data 127 * to send then the probe will be the FIN 128 * itself. 129 */ 130 if (off < so->so_snd.sb_cc) 131 flags &= ~TH_FIN; 132 win = 1; 133 } else { 134 tp->t_timer[TCPT_PERSIST] = 0; 135 tp->t_rxtshift = 0; 136 } 137 } 138 139 len = min(so->so_snd.sb_cc, win) - off; 140 141 if (len < 0) { 142 /* 143 * If FIN has been sent but not acked, 144 * but we haven't been called to retransmit, 145 * len will be -1. Otherwise, window shrank 146 * after we sent into it. If window shrank to 0, 147 * cancel pending retransmit and pull snd_nxt 148 * back to (closed) window. We will enter persist 149 * state below. If the window didn't close completely, 150 * just wait for an ACK. 151 */ 152 len = 0; 153 if (win == 0) { 154 tp->t_timer[TCPT_REXMT] = 0; 155 tp->snd_nxt = tp->snd_una; 156 } 157 } 158 if (len > tp->t_maxseg) { 159 len = tp->t_maxseg; 160 sendalot = 1; 161 } 162 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 163 flags &= ~TH_FIN; 164 165 win = sbspace(&so->so_rcv); 166 167 /* 168 * Sender silly window avoidance. If connection is idle 169 * and can send all data, a maximum segment, 170 * at least a maximum default-size segment do it, 171 * or are forced, do it; otherwise don't bother. 172 * If peer's buffer is tiny, then send 173 * when window is at least half open. 174 * If retransmitting (possibly after persist timer forced us 175 * to send into a small window), then must resend. 176 */ 177 if (len) { 178 if (len == tp->t_maxseg) 179 goto send; 180 if ((idle || tp->t_flags & TF_NODELAY) && 181 len + off >= so->so_snd.sb_cc) 182 goto send; 183 if (tp->t_force) 184 goto send; 185 if (len >= tp->max_sndwnd / 2) 186 goto send; 187 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 188 goto send; 189 } 190 191 /* 192 * Compare available window to amount of window 193 * known to peer (as advertised window less 194 * next expected input). If the difference is at least two 195 * max size segments, or at least 50% of the maximum possible 196 * window, then want to send a window update to peer. 197 */ 198 if (win > 0) { 199 /* 200 * "adv" is the amount we can increase the window, 201 * taking into account that we are limited by 202 * TCP_MAXWIN << tp->rcv_scale. 203 */ 204 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 205 (tp->rcv_adv - tp->rcv_nxt); 206 207 if (adv >= (long) (2 * tp->t_maxseg)) 208 goto send; 209 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 210 goto send; 211 } 212 213 /* 214 * Send if we owe peer an ACK. 215 */ 216 if (tp->t_flags & TF_ACKNOW) 217 goto send; 218 if (flags & (TH_SYN|TH_RST)) 219 goto send; 220 if (SEQ_GT(tp->snd_up, tp->snd_una)) 221 goto send; 222 /* 223 * If our state indicates that FIN should be sent 224 * and we have not yet done so, or we're retransmitting the FIN, 225 * then we need to send. 226 */ 227 if (flags & TH_FIN && 228 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 229 goto send; 230 231 /* 232 * TCP window updates are not reliable, rather a polling protocol 233 * using ``persist'' packets is used to insure receipt of window 234 * updates. The three ``states'' for the output side are: 235 * idle not doing retransmits or persists 236 * persisting to move a small or zero window 237 * (re)transmitting and thereby not persisting 238 * 239 * tp->t_timer[TCPT_PERSIST] 240 * is set when we are in persist state. 241 * tp->t_force 242 * is set when we are called to send a persist packet. 243 * tp->t_timer[TCPT_REXMT] 244 * is set when we are retransmitting 245 * The output side is idle when both timers are zero. 246 * 247 * If send window is too small, there is data to transmit, and no 248 * retransmit or persist is pending, then go to persist state. 249 * If nothing happens soon, send when timer expires: 250 * if window is nonzero, transmit what we can, 251 * otherwise force out a byte. 252 */ 253 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && 254 tp->t_timer[TCPT_PERSIST] == 0) { 255 tp->t_rxtshift = 0; 256 tcp_setpersist(tp); 257 } 258 259 /* 260 * No reason to send a segment, just return. 261 */ 262 return (0); 263 264 send: 265 /* 266 * Before ESTABLISHED, force sending of initial options 267 * unless TCP set not to do any options. 268 * NOTE: we assume that the IP/TCP header plus TCP options 269 * always fit in a single mbuf, leaving room for a maximum 270 * link header, i.e. 271 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN 272 */ 273 optlen = 0; 274 hdrlen = sizeof (struct tcpiphdr); 275 if (flags & TH_SYN) { 276 tp->snd_nxt = tp->iss; 277 if ((tp->t_flags & TF_NOOPT) == 0) { 278 u_short mss; 279 280 opt[0] = TCPOPT_MAXSEG; 281 opt[1] = 4; 282 mss = htons((u_short) tcp_mss(tp, 0)); 283 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), sizeof(mss)); 284 optlen = 4; 285 286 if ((tp->t_flags & TF_REQ_SCALE) && 287 ((flags & TH_ACK) == 0 || 288 (tp->t_flags & TF_RCVD_SCALE))) { 289 *((u_long *) (opt + optlen)) = htonl( 290 TCPOPT_NOP << 24 | 291 TCPOPT_WINDOW << 16 | 292 TCPOLEN_WINDOW << 8 | 293 tp->request_r_scale); 294 optlen += 4; 295 } 296 } 297 } 298 299 /* 300 * Send a timestamp and echo-reply if this is a SYN and our side 301 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 302 * and our peer have sent timestamps in our SYN's. 303 */ 304 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 305 (flags & TH_RST) == 0 && 306 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 307 (tp->t_flags & TF_RCVD_TSTMP))) { 308 u_long *lp = (u_long *)(opt + optlen); 309 310 /* Form timestamp option as shown in appendix A of RFC 1323. */ 311 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 312 *lp++ = htonl(tcp_now); 313 *lp = htonl(tp->ts_recent); 314 optlen += TCPOLEN_TSTAMP_APPA; 315 } 316 317 hdrlen += optlen; 318 319 /* 320 * Adjust data length if insertion of options will 321 * bump the packet length beyond the t_maxseg length. 322 */ 323 if (len > tp->t_maxseg - optlen) { 324 len = tp->t_maxseg - optlen; 325 sendalot = 1; 326 } 327 328 329 #ifdef DIAGNOSTIC 330 if (max_linkhdr + hdrlen > MHLEN) 331 panic("tcphdr too big"); 332 #endif 333 334 /* 335 * Grab a header mbuf, attaching a copy of data to 336 * be transmitted, and initialize the header from 337 * the template for sends on this connection. 338 */ 339 if (len) { 340 if (tp->t_force && len == 1) 341 tcpstat.tcps_sndprobe++; 342 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 343 tcpstat.tcps_sndrexmitpack++; 344 tcpstat.tcps_sndrexmitbyte += len; 345 } else { 346 tcpstat.tcps_sndpack++; 347 tcpstat.tcps_sndbyte += len; 348 } 349 #ifdef notyet 350 if ((m = m_copypack(so->so_snd.sb_mb, off, 351 (int)len, max_linkhdr + hdrlen)) == 0) { 352 error = ENOBUFS; 353 goto out; 354 } 355 /* 356 * m_copypack left space for our hdr; use it. 357 */ 358 m->m_len += hdrlen; 359 m->m_data -= hdrlen; 360 #else 361 MGETHDR(m, M_DONTWAIT, MT_HEADER); 362 if (m == NULL) { 363 error = ENOBUFS; 364 goto out; 365 } 366 m->m_data += max_linkhdr; 367 m->m_len = hdrlen; 368 if (len <= MHLEN - hdrlen - max_linkhdr) { 369 m_copydata(so->so_snd.sb_mb, off, (int) len, 370 mtod(m, caddr_t) + hdrlen); 371 m->m_len += len; 372 } else { 373 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 374 if (m->m_next == 0) 375 len = 0; 376 } 377 #endif 378 /* 379 * If we're sending everything we've got, set PUSH. 380 * (This will keep happy those implementations which only 381 * give data to the user when a buffer fills or 382 * a PUSH comes in.) 383 */ 384 if (off + len == so->so_snd.sb_cc) 385 flags |= TH_PUSH; 386 } else { 387 if (tp->t_flags & TF_ACKNOW) 388 tcpstat.tcps_sndacks++; 389 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 390 tcpstat.tcps_sndctrl++; 391 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 392 tcpstat.tcps_sndurg++; 393 else 394 tcpstat.tcps_sndwinup++; 395 396 MGETHDR(m, M_DONTWAIT, MT_HEADER); 397 if (m == NULL) { 398 error = ENOBUFS; 399 goto out; 400 } 401 m->m_data += max_linkhdr; 402 m->m_len = hdrlen; 403 } 404 m->m_pkthdr.rcvif = (struct ifnet *)0; 405 ti = mtod(m, struct tcpiphdr *); 406 if (tp->t_template == 0) 407 panic("tcp_output"); 408 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 409 410 /* 411 * Fill in fields, remembering maximum advertised 412 * window for use in delaying messages about window sizes. 413 * If resending a FIN, be sure not to use a new sequence number. 414 */ 415 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 416 tp->snd_nxt == tp->snd_max) 417 tp->snd_nxt--; 418 /* 419 * If we are doing retransmissions, then snd_nxt will 420 * not reflect the first unsent octet. For ACK only 421 * packets, we do not want the sequence number of the 422 * retransmitted packet, we want the sequence number 423 * of the next unsent octet. So, if there is no data 424 * (and no SYN or FIN), use snd_max instead of snd_nxt 425 * when filling in ti_seq. But if we are in persist 426 * state, snd_max might reflect one byte beyond the 427 * right edge of the window, so use snd_nxt in that 428 * case, since we know we aren't doing a retransmission. 429 * (retransmit and persist are mutually exclusive...) 430 */ 431 if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST]) 432 ti->ti_seq = htonl(tp->snd_nxt); 433 else 434 ti->ti_seq = htonl(tp->snd_max); 435 ti->ti_ack = htonl(tp->rcv_nxt); 436 if (optlen) { 437 bcopy((caddr_t)opt, (caddr_t)(ti + 1), optlen); 438 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 439 } 440 ti->ti_flags = flags; 441 /* 442 * Calculate receive window. Don't shrink window, 443 * but avoid silly window syndrome. 444 */ 445 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) 446 win = 0; 447 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 448 win = (long)TCP_MAXWIN << tp->rcv_scale; 449 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 450 win = (long)(tp->rcv_adv - tp->rcv_nxt); 451 ti->ti_win = htons((u_short) (win>>tp->rcv_scale)); 452 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 453 ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 454 ti->ti_flags |= TH_URG; 455 } else 456 /* 457 * If no urgent pointer to send, then we pull 458 * the urgent pointer to the left edge of the send window 459 * so that it doesn't drift into the send window on sequence 460 * number wraparound. 461 */ 462 tp->snd_up = tp->snd_una; /* drag it along */ 463 464 /* 465 * Put TCP length in extended header, and then 466 * checksum extended header and data. 467 */ 468 if (len + optlen) 469 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + 470 optlen + len)); 471 ti->ti_sum = in_cksum(m, (int)(hdrlen + len)); 472 473 /* 474 * In transmit state, time the transmission and arrange for 475 * the retransmit. In persist state, just set snd_max. 476 */ 477 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { 478 tcp_seq startseq = tp->snd_nxt; 479 480 /* 481 * Advance snd_nxt over sequence space of this segment. 482 */ 483 if (flags & (TH_SYN|TH_FIN)) { 484 if (flags & TH_SYN) 485 tp->snd_nxt++; 486 if (flags & TH_FIN) { 487 tp->snd_nxt++; 488 tp->t_flags |= TF_SENTFIN; 489 } 490 } 491 tp->snd_nxt += len; 492 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 493 tp->snd_max = tp->snd_nxt; 494 /* 495 * Time this transmission if not a retransmission and 496 * not currently timing anything. 497 */ 498 if (tp->t_rtt == 0) { 499 tp->t_rtt = 1; 500 tp->t_rtseq = startseq; 501 tcpstat.tcps_segstimed++; 502 } 503 } 504 505 /* 506 * Set retransmit timer if not currently set, 507 * and not doing an ack or a keep-alive probe. 508 * Initial value for retransmit timer is smoothed 509 * round-trip time + 2 * round-trip time variance. 510 * Initialize shift counter which is used for backoff 511 * of retransmit time. 512 */ 513 if (tp->t_timer[TCPT_REXMT] == 0 && 514 tp->snd_nxt != tp->snd_una) { 515 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 516 if (tp->t_timer[TCPT_PERSIST]) { 517 tp->t_timer[TCPT_PERSIST] = 0; 518 tp->t_rxtshift = 0; 519 } 520 } 521 } else 522 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 523 tp->snd_max = tp->snd_nxt + len; 524 525 /* 526 * Trace. 527 */ 528 if (so->so_options & SO_DEBUG) 529 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 530 531 /* 532 * Fill in IP length and desired time to live and 533 * send to IP level. There should be a better way 534 * to handle ttl and tos; we could keep them in 535 * the template, but need a way to checksum without them. 536 */ 537 m->m_pkthdr.len = hdrlen + len; 538 #ifdef TUBA 539 if (tp->t_tuba_pcb) 540 error = tuba_output(m, tp); 541 else 542 #endif 543 { 544 ((struct ip *)ti)->ip_len = m->m_pkthdr.len; 545 ((struct ip *)ti)->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; /* XXX */ 546 ((struct ip *)ti)->ip_tos = tp->t_inpcb->inp_ip.ip_tos; /* XXX */ 547 #if BSD >= 43 548 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 549 so->so_options & SO_DONTROUTE, 0); 550 #else 551 error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route, 552 so->so_options & SO_DONTROUTE); 553 #endif 554 } 555 if (error) { 556 out: 557 if (error == ENOBUFS) { 558 tcp_quench(tp->t_inpcb, 0); 559 return (0); 560 } 561 if ((error == EHOSTUNREACH || error == ENETDOWN) 562 && TCPS_HAVERCVDSYN(tp->t_state)) { 563 tp->t_softerror = error; 564 return (0); 565 } 566 return (error); 567 } 568 tcpstat.tcps_sndtotal++; 569 570 /* 571 * Data sent (as far as we can tell). 572 * If this advertises a larger window than any other segment, 573 * then remember the size of the advertised window. 574 * Any pending ACK has now been sent. 575 */ 576 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 577 tp->rcv_adv = tp->rcv_nxt + win; 578 tp->last_ack_sent = tp->rcv_nxt; 579 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 580 if (sendalot) 581 goto again; 582 return (0); 583 } 584 585 void 586 tcp_setpersist(tp) 587 register struct tcpcb *tp; 588 { 589 register t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 590 591 if (tp->t_timer[TCPT_REXMT]) 592 panic("tcp_output REXMT"); 593 /* 594 * Start/restart persistance timer. 595 */ 596 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 597 t * tcp_backoff[tp->t_rxtshift], 598 TCPTV_PERSMIN, TCPTV_PERSMAX); 599 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 600 tp->t_rxtshift++; 601 } 602