1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 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_output.c 8.3 (Berkeley) 12/30/93 34 */ 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/malloc.h> 39 #include <sys/mbuf.h> 40 #include <sys/protosw.h> 41 #include <sys/socket.h> 42 #include <sys/socketvar.h> 43 #include <sys/errno.h> 44 45 #include <net/route.h> 46 47 #include <netinet/in.h> 48 #include <netinet/in_systm.h> 49 #include <netinet/ip.h> 50 #include <netinet/in_pcb.h> 51 #include <netinet/ip_var.h> 52 #include <netinet/tcp.h> 53 #define TCPOUTFLAGS 54 #include <netinet/tcp_fsm.h> 55 #include <netinet/tcp_seq.h> 56 #include <netinet/tcp_timer.h> 57 #include <netinet/tcp_var.h> 58 #include <netinet/tcpip.h> 59 #include <netinet/tcp_debug.h> 60 61 #ifdef notyet 62 extern struct mbuf *m_copypack(); 63 #endif 64 65 66 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */ 67 68 /* 69 * Tcp output routine: figure out what should be sent and send it. 70 */ 71 int 72 tcp_output(tp) 73 register struct tcpcb *tp; 74 { 75 register struct socket *so = tp->t_inpcb->inp_socket; 76 register long len, win; 77 int off, flags, error; 78 register struct mbuf *m; 79 register struct tcpiphdr *ti; 80 u_char opt[MAX_TCPOPTLEN]; 81 unsigned optlen, hdrlen; 82 int idle, sendalot; 83 84 /* 85 * Determine length of data that should be transmitted, 86 * and flags that will be used. 87 * If there is some data or critical controls (SYN, RST) 88 * to send, then transmit; otherwise, investigate further. 89 */ 90 idle = (tp->snd_max == tp->snd_una); 91 if (idle && tp->t_idle >= tp->t_rxtcur) 92 /* 93 * We have been idle for "a while" and no acks are 94 * expected to clock out any data we send -- 95 * slow start to get ack "clock" running again. 96 */ 97 tp->snd_cwnd = tp->t_maxseg; 98 again: 99 sendalot = 0; 100 off = tp->snd_nxt - tp->snd_una; 101 win = min(tp->snd_wnd, tp->snd_cwnd); 102 103 flags = tcp_outflags[tp->t_state]; 104 /* 105 * If in persist timeout with window of 0, send 1 byte. 106 * Otherwise, if window is small but nonzero 107 * and timer expired, we will send what we can 108 * and go to transmit state. 109 */ 110 if (tp->t_force) { 111 if (win == 0) { 112 /* 113 * If we still have some data to send, then 114 * clear the FIN bit. Usually this would 115 * happen below when it realizes that we 116 * aren't sending all the data. However, 117 * if we have exactly 1 byte of unset data, 118 * then it won't clear the FIN bit below, 119 * and if we are in persist state, we wind 120 * up sending the packet without recording 121 * that we sent the FIN bit. 122 * 123 * We can't just blindly clear the FIN bit, 124 * because if we don't have any more data 125 * to send then the probe will be the FIN 126 * itself. 127 */ 128 if (off < so->so_snd.sb_cc) 129 flags &= ~TH_FIN; 130 win = 1; 131 } else { 132 tp->t_timer[TCPT_PERSIST] = 0; 133 tp->t_rxtshift = 0; 134 } 135 } 136 137 len = min(so->so_snd.sb_cc, win) - off; 138 139 if (len < 0) { 140 /* 141 * If FIN has been sent but not acked, 142 * but we haven't been called to retransmit, 143 * len will be -1. Otherwise, window shrank 144 * after we sent into it. If window shrank to 0, 145 * cancel pending retransmit and pull snd_nxt 146 * back to (closed) window. We will enter persist 147 * state below. If the window didn't close completely, 148 * just wait for an ACK. 149 */ 150 len = 0; 151 if (win == 0) { 152 tp->t_timer[TCPT_REXMT] = 0; 153 tp->snd_nxt = tp->snd_una; 154 } 155 } 156 if (len > tp->t_maxseg) { 157 len = tp->t_maxseg; 158 sendalot = 1; 159 } 160 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 161 flags &= ~TH_FIN; 162 163 win = sbspace(&so->so_rcv); 164 165 /* 166 * Sender silly window avoidance. If connection is idle 167 * and can send all data, a maximum segment, 168 * at least a maximum default-size segment do it, 169 * or are forced, do it; otherwise don't bother. 170 * If peer's buffer is tiny, then send 171 * when window is at least half open. 172 * If retransmitting (possibly after persist timer forced us 173 * to send into a small window), then must resend. 174 */ 175 if (len) { 176 if (len == tp->t_maxseg) 177 goto send; 178 if ((idle || tp->t_flags & TF_NODELAY) && 179 len + off >= so->so_snd.sb_cc) 180 goto send; 181 if (tp->t_force) 182 goto send; 183 if (len >= tp->max_sndwnd / 2) 184 goto send; 185 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 186 goto send; 187 } 188 189 /* 190 * Compare available window to amount of window 191 * known to peer (as advertised window less 192 * next expected input). If the difference is at least two 193 * max size segments, or at least 50% of the maximum possible 194 * window, then want to send a window update to peer. 195 */ 196 if (win > 0) { 197 /* 198 * "adv" is the amount we can increase the window, 199 * taking into account that we are limited by 200 * TCP_MAXWIN << tp->rcv_scale. 201 */ 202 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 203 (tp->rcv_adv - tp->rcv_nxt); 204 205 if (adv >= (long) (2 * tp->t_maxseg)) 206 goto send; 207 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 208 goto send; 209 } 210 211 /* 212 * Send if we owe peer an ACK. 213 */ 214 if (tp->t_flags & TF_ACKNOW) 215 goto send; 216 if (flags & (TH_SYN|TH_RST)) 217 goto send; 218 if (SEQ_GT(tp->snd_up, tp->snd_una)) 219 goto send; 220 /* 221 * If our state indicates that FIN should be sent 222 * and we have not yet done so, or we're retransmitting the FIN, 223 * then we need to send. 224 */ 225 if (flags & TH_FIN && 226 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 227 goto send; 228 229 /* 230 * TCP window updates are not reliable, rather a polling protocol 231 * using ``persist'' packets is used to insure receipt of window 232 * updates. The three ``states'' for the output side are: 233 * idle not doing retransmits or persists 234 * persisting to move a small or zero window 235 * (re)transmitting and thereby not persisting 236 * 237 * tp->t_timer[TCPT_PERSIST] 238 * is set when we are in persist state. 239 * tp->t_force 240 * is set when we are called to send a persist packet. 241 * tp->t_timer[TCPT_REXMT] 242 * is set when we are retransmitting 243 * The output side is idle when both timers are zero. 244 * 245 * If send window is too small, there is data to transmit, and no 246 * retransmit or persist is pending, then go to persist state. 247 * If nothing happens soon, send when timer expires: 248 * if window is nonzero, transmit what we can, 249 * otherwise force out a byte. 250 */ 251 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && 252 tp->t_timer[TCPT_PERSIST] == 0) { 253 tp->t_rxtshift = 0; 254 tcp_setpersist(tp); 255 } 256 257 /* 258 * No reason to send a segment, just return. 259 */ 260 return (0); 261 262 send: 263 /* 264 * Before ESTABLISHED, force sending of initial options 265 * unless TCP set not to do any options. 266 * NOTE: we assume that the IP/TCP header plus TCP options 267 * always fit in a single mbuf, leaving room for a maximum 268 * link header, i.e. 269 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN 270 */ 271 optlen = 0; 272 hdrlen = sizeof (struct tcpiphdr); 273 if (flags & TH_SYN) { 274 tp->snd_nxt = tp->iss; 275 if ((tp->t_flags & TF_NOOPT) == 0) { 276 u_short mss; 277 278 opt[0] = TCPOPT_MAXSEG; 279 opt[1] = 4; 280 mss = htons((u_short) tcp_mss(tp, 0)); 281 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), sizeof(mss)); 282 optlen = 4; 283 284 if ((tp->t_flags & TF_REQ_SCALE) && 285 ((flags & TH_ACK) == 0 || 286 (tp->t_flags & TF_RCVD_SCALE))) { 287 *((u_long *) (opt + optlen)) = htonl( 288 TCPOPT_NOP << 24 | 289 TCPOPT_WINDOW << 16 | 290 TCPOLEN_WINDOW << 8 | 291 tp->request_r_scale); 292 optlen += 4; 293 } 294 } 295 } 296 297 /* 298 * Send a timestamp and echo-reply if this is a SYN and our side 299 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 300 * and our peer have sent timestamps in our SYN's. 301 */ 302 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 303 (flags & TH_RST) == 0 && 304 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 305 (tp->t_flags & TF_RCVD_TSTMP))) { 306 u_long *lp = (u_long *)(opt + optlen); 307 308 /* Form timestamp option as shown in appendix A of RFC 1323. */ 309 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 310 *lp++ = htonl(tcp_now); 311 *lp = htonl(tp->ts_recent); 312 optlen += TCPOLEN_TSTAMP_APPA; 313 } 314 315 hdrlen += optlen; 316 317 /* 318 * Adjust data length if insertion of options will 319 * bump the packet length beyond the t_maxseg length. 320 */ 321 if (len > tp->t_maxseg - optlen) { 322 len = tp->t_maxseg - optlen; 323 sendalot = 1; 324 } 325 326 327 #ifdef DIAGNOSTIC 328 if (max_linkhdr + hdrlen > MHLEN) 329 panic("tcphdr too big"); 330 #endif 331 332 /* 333 * Grab a header mbuf, attaching a copy of data to 334 * be transmitted, and initialize the header from 335 * the template for sends on this connection. 336 */ 337 if (len) { 338 if (tp->t_force && len == 1) 339 tcpstat.tcps_sndprobe++; 340 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 341 tcpstat.tcps_sndrexmitpack++; 342 tcpstat.tcps_sndrexmitbyte += len; 343 } else { 344 tcpstat.tcps_sndpack++; 345 tcpstat.tcps_sndbyte += len; 346 } 347 #ifdef notyet 348 if ((m = m_copypack(so->so_snd.sb_mb, off, 349 (int)len, max_linkhdr + hdrlen)) == 0) { 350 error = ENOBUFS; 351 goto out; 352 } 353 /* 354 * m_copypack left space for our hdr; use it. 355 */ 356 m->m_len += hdrlen; 357 m->m_data -= hdrlen; 358 #else 359 MGETHDR(m, M_DONTWAIT, MT_HEADER); 360 if (m == NULL) { 361 error = ENOBUFS; 362 goto out; 363 } 364 m->m_data += max_linkhdr; 365 m->m_len = hdrlen; 366 if (len <= MHLEN - hdrlen - max_linkhdr) { 367 m_copydata(so->so_snd.sb_mb, off, (int) len, 368 mtod(m, caddr_t) + hdrlen); 369 m->m_len += len; 370 } else { 371 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 372 if (m->m_next == 0) 373 len = 0; 374 } 375 #endif 376 /* 377 * If we're sending everything we've got, set PUSH. 378 * (This will keep happy those implementations which only 379 * give data to the user when a buffer fills or 380 * a PUSH comes in.) 381 */ 382 if (off + len == so->so_snd.sb_cc) 383 flags |= TH_PUSH; 384 } else { 385 if (tp->t_flags & TF_ACKNOW) 386 tcpstat.tcps_sndacks++; 387 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 388 tcpstat.tcps_sndctrl++; 389 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 390 tcpstat.tcps_sndurg++; 391 else 392 tcpstat.tcps_sndwinup++; 393 394 MGETHDR(m, M_DONTWAIT, MT_HEADER); 395 if (m == NULL) { 396 error = ENOBUFS; 397 goto out; 398 } 399 m->m_data += max_linkhdr; 400 m->m_len = hdrlen; 401 } 402 m->m_pkthdr.rcvif = (struct ifnet *)0; 403 ti = mtod(m, struct tcpiphdr *); 404 if (tp->t_template == 0) 405 panic("tcp_output"); 406 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 407 408 /* 409 * Fill in fields, remembering maximum advertised 410 * window for use in delaying messages about window sizes. 411 * If resending a FIN, be sure not to use a new sequence number. 412 */ 413 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 414 tp->snd_nxt == tp->snd_max) 415 tp->snd_nxt--; 416 /* 417 * If we are doing retransmissions, then snd_nxt will 418 * not reflect the first unsent octet. For ACK only 419 * packets, we do not want the sequence number of the 420 * retransmitted packet, we want the sequence number 421 * of the next unsent octet. So, if there is no data 422 * (and no SYN or FIN), use snd_max instead of snd_nxt 423 * when filling in ti_seq. But if we are in persist 424 * state, snd_max might reflect one byte beyond the 425 * right edge of the window, so use snd_nxt in that 426 * case, since we know we aren't doing a retransmission. 427 * (retransmit and persist are mutually exclusive...) 428 */ 429 if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST]) 430 ti->ti_seq = htonl(tp->snd_nxt); 431 else 432 ti->ti_seq = htonl(tp->snd_max); 433 ti->ti_ack = htonl(tp->rcv_nxt); 434 if (optlen) { 435 bcopy((caddr_t)opt, (caddr_t)(ti + 1), optlen); 436 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 437 } 438 ti->ti_flags = flags; 439 /* 440 * Calculate receive window. Don't shrink window, 441 * but avoid silly window syndrome. 442 */ 443 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) 444 win = 0; 445 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 446 win = (long)TCP_MAXWIN << tp->rcv_scale; 447 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 448 win = (long)(tp->rcv_adv - tp->rcv_nxt); 449 ti->ti_win = htons((u_short) (win>>tp->rcv_scale)); 450 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 451 ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 452 ti->ti_flags |= TH_URG; 453 } else 454 /* 455 * If no urgent pointer to send, then we pull 456 * the urgent pointer to the left edge of the send window 457 * so that it doesn't drift into the send window on sequence 458 * number wraparound. 459 */ 460 tp->snd_up = tp->snd_una; /* drag it along */ 461 462 /* 463 * Put TCP length in extended header, and then 464 * checksum extended header and data. 465 */ 466 if (len + optlen) 467 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + 468 optlen + len)); 469 ti->ti_sum = in_cksum(m, (int)(hdrlen + len)); 470 471 /* 472 * In transmit state, time the transmission and arrange for 473 * the retransmit. In persist state, just set snd_max. 474 */ 475 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { 476 tcp_seq startseq = tp->snd_nxt; 477 478 /* 479 * Advance snd_nxt over sequence space of this segment. 480 */ 481 if (flags & (TH_SYN|TH_FIN)) { 482 if (flags & TH_SYN) 483 tp->snd_nxt++; 484 if (flags & TH_FIN) { 485 tp->snd_nxt++; 486 tp->t_flags |= TF_SENTFIN; 487 } 488 } 489 tp->snd_nxt += len; 490 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 491 tp->snd_max = tp->snd_nxt; 492 /* 493 * Time this transmission if not a retransmission and 494 * not currently timing anything. 495 */ 496 if (tp->t_rtt == 0) { 497 tp->t_rtt = 1; 498 tp->t_rtseq = startseq; 499 tcpstat.tcps_segstimed++; 500 } 501 } 502 503 /* 504 * Set retransmit timer if not currently set, 505 * and not doing an ack or a keep-alive probe. 506 * Initial value for retransmit timer is smoothed 507 * round-trip time + 2 * round-trip time variance. 508 * Initialize shift counter which is used for backoff 509 * of retransmit time. 510 */ 511 if (tp->t_timer[TCPT_REXMT] == 0 && 512 tp->snd_nxt != tp->snd_una) { 513 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 514 if (tp->t_timer[TCPT_PERSIST]) { 515 tp->t_timer[TCPT_PERSIST] = 0; 516 tp->t_rxtshift = 0; 517 } 518 } 519 } else 520 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 521 tp->snd_max = tp->snd_nxt + len; 522 523 /* 524 * Trace. 525 */ 526 if (so->so_options & SO_DEBUG) 527 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 528 529 /* 530 * Fill in IP length and desired time to live and 531 * send to IP level. There should be a better way 532 * to handle ttl and tos; we could keep them in 533 * the template, but need a way to checksum without them. 534 */ 535 m->m_pkthdr.len = hdrlen + len; 536 #ifdef TUBA 537 if (tp->t_tuba_pcb) 538 error = tuba_output(m, tp); 539 else 540 #endif 541 { 542 ((struct ip *)ti)->ip_len = m->m_pkthdr.len; 543 ((struct ip *)ti)->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; /* XXX */ 544 ((struct ip *)ti)->ip_tos = tp->t_inpcb->inp_ip.ip_tos; /* XXX */ 545 #if BSD >= 43 546 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 547 so->so_options & SO_DONTROUTE, 0); 548 #else 549 error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route, 550 so->so_options & SO_DONTROUTE); 551 #endif 552 } 553 if (error) { 554 out: 555 if (error == ENOBUFS) { 556 tcp_quench(tp->t_inpcb, 0); 557 return (0); 558 } 559 if ((error == EHOSTUNREACH || error == ENETDOWN) 560 && TCPS_HAVERCVDSYN(tp->t_state)) { 561 tp->t_softerror = error; 562 return (0); 563 } 564 return (error); 565 } 566 tcpstat.tcps_sndtotal++; 567 568 /* 569 * Data sent (as far as we can tell). 570 * If this advertises a larger window than any other segment, 571 * then remember the size of the advertised window. 572 * Any pending ACK has now been sent. 573 */ 574 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 575 tp->rcv_adv = tp->rcv_nxt + win; 576 tp->last_ack_sent = tp->rcv_nxt; 577 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 578 if (sendalot) 579 goto again; 580 return (0); 581 } 582 583 void 584 tcp_setpersist(tp) 585 register struct tcpcb *tp; 586 { 587 register t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 588 589 if (tp->t_timer[TCPT_REXMT]) 590 panic("tcp_output REXMT"); 591 /* 592 * Start/restart persistance timer. 593 */ 594 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 595 t * tcp_backoff[tp->t_rxtshift], 596 TCPTV_PERSMIN, TCPTV_PERSMAX); 597 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 598 tp->t_rxtshift++; 599 } 600