1 /* 2 * Copyright (c) 1982, 1986 Regents of the University of California. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms are permitted 6 * provided that the above copyright notice and this paragraph are 7 * duplicated in all such forms and that any documentation, 8 * advertising materials, and other materials related to such 9 * distribution and use acknowledge that the software was developed 10 * by the University of California, Berkeley. The name of the 11 * University may not be used to endorse or promote products derived 12 * from this software without specific prior written permission. 13 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 14 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 15 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. 16 * 17 * @(#)tcp_output.c 7.17 (Berkeley) 06/29/88 18 */ 19 20 #include "param.h" 21 #include "systm.h" 22 #include "mbuf.h" 23 #include "protosw.h" 24 #include "socket.h" 25 #include "socketvar.h" 26 #include "errno.h" 27 28 #include "../net/route.h" 29 30 #include "in.h" 31 #include "in_pcb.h" 32 #include "in_systm.h" 33 #include "ip.h" 34 #include "ip_var.h" 35 #include "tcp.h" 36 #define TCPOUTFLAGS 37 #include "tcp_fsm.h" 38 #include "tcp_seq.h" 39 #include "tcp_timer.h" 40 #include "tcp_var.h" 41 #include "tcpip.h" 42 #include "tcp_debug.h" 43 44 /* 45 * Initial options. 46 */ 47 u_char tcp_initopt[4] = { TCPOPT_MAXSEG, 4, 0x0, 0x0, }; 48 49 /* 50 * Tcp output routine: figure out what should be sent and send it. 51 */ 52 tcp_output(tp) 53 register struct tcpcb *tp; 54 { 55 register struct socket *so = tp->t_inpcb->inp_socket; 56 register long len, win; 57 struct mbuf *m0; 58 int off, flags, error; 59 register struct mbuf *m; 60 register struct tcpiphdr *ti; 61 u_char *opt; 62 unsigned optlen = 0; 63 int idle, sendalot; 64 65 /* 66 * Determine length of data that should be transmitted, 67 * and flags that will be used. 68 * If there is some data or critical controls (SYN, RST) 69 * to send, then transmit; otherwise, investigate further. 70 */ 71 idle = (tp->snd_max == tp->snd_una); 72 again: 73 sendalot = 0; 74 off = tp->snd_nxt - tp->snd_una; 75 win = MIN(tp->snd_wnd, tp->snd_cwnd); 76 77 /* 78 * If in persist timeout with window of 0, send 1 byte. 79 * Otherwise, if window is small but nonzero 80 * and timer expired, we will send what we can 81 * and go to transmit state. 82 */ 83 if (tp->t_force) { 84 if (win == 0) 85 win = 1; 86 else { 87 tp->t_timer[TCPT_PERSIST] = 0; 88 tp->t_rxtshift = 0; 89 } 90 } 91 92 len = MIN(so->so_snd.sb_cc, win) - off; 93 flags = tcp_outflags[tp->t_state]; 94 95 if (len < 0) { 96 /* 97 * If FIN has been sent but not acked, 98 * but we haven't been called to retransmit, 99 * len will be -1. Otherwise, window shrank 100 * after we sent into it. If window shrank to 0, 101 * cancel pending retransmit and pull snd_nxt 102 * back to (closed) window. We will enter persist 103 * state below. If the window didn't close completely, 104 * just wait for an ACK. 105 */ 106 len = 0; 107 if (win == 0) { 108 tp->t_timer[TCPT_REXMT] = 0; 109 tp->snd_nxt = tp->snd_una; 110 } 111 } 112 if (len > tp->t_maxseg) { 113 len = tp->t_maxseg; 114 sendalot = 1; 115 } 116 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 117 flags &= ~TH_FIN; 118 win = sbspace(&so->so_rcv); 119 120 121 /* 122 * If our state indicates that FIN should be sent 123 * and we have not yet done so, or we're retransmitting the FIN, 124 * then we need to send. 125 */ 126 if (flags & TH_FIN && 127 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 128 goto send; 129 /* 130 * Send if we owe peer an ACK. 131 */ 132 if (tp->t_flags & TF_ACKNOW) 133 goto send; 134 if (flags & (TH_SYN|TH_RST)) 135 goto send; 136 if (SEQ_GT(tp->snd_up, tp->snd_una)) 137 goto send; 138 139 /* 140 * Sender silly window avoidance. If connection is idle 141 * and can send all data, a maximum segment, 142 * at least a maximum default-size segment do it, 143 * or are forced, do it; otherwise don't bother. 144 * If peer's buffer is tiny, then send 145 * when window is at least half open. 146 * If retransmitting (possibly after persist timer forced us 147 * to send into a small window), then must resend. 148 */ 149 if (len) { 150 if (len == tp->t_maxseg) 151 goto send; 152 if ((idle || tp->t_flags & TF_NODELAY) && 153 len + off >= so->so_snd.sb_cc) 154 goto send; 155 if (tp->t_force) 156 goto send; 157 if (len >= tp->max_sndwnd / 2) 158 goto send; 159 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 160 goto send; 161 } 162 163 /* 164 * Compare available window to amount of window 165 * known to peer (as advertised window less 166 * next expected input). If the difference is at least two 167 * max size segments or at least 35% of the maximum possible 168 * window, then want to send a window update to peer. 169 */ 170 if (win > 0) { 171 int adv = win - (tp->rcv_adv - tp->rcv_nxt); 172 173 if (so->so_rcv.sb_cc == 0 && adv >= 2 * tp->t_maxseg) 174 goto send; 175 if (100 * adv / so->so_rcv.sb_hiwat >= 35) 176 goto send; 177 } 178 179 /* 180 * TCP window updates are not reliable, rather a polling protocol 181 * using ``persist'' packets is used to insure receipt of window 182 * updates. The three ``states'' for the output side are: 183 * idle not doing retransmits or persists 184 * persisting to move a small or zero window 185 * (re)transmitting and thereby not persisting 186 * 187 * tp->t_timer[TCPT_PERSIST] 188 * is set when we are in persist state. 189 * tp->t_force 190 * is set when we are called to send a persist packet. 191 * tp->t_timer[TCPT_REXMT] 192 * is set when we are retransmitting 193 * The output side is idle when both timers are zero. 194 * 195 * If send window is too small, there is data to transmit, and no 196 * retransmit or persist is pending, then go to persist state. 197 * If nothing happens soon, send when timer expires: 198 * if window is nonzero, transmit what we can, 199 * otherwise force out a byte. 200 */ 201 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && 202 tp->t_timer[TCPT_PERSIST] == 0) { 203 tp->t_rxtshift = 0; 204 tcp_setpersist(tp); 205 } 206 207 /* 208 * No reason to send a segment, just return. 209 */ 210 return (0); 211 212 send: 213 /* 214 * Grab a header mbuf, attaching a copy of data to 215 * be transmitted, and initialize the header from 216 * the template for sends on this connection. 217 */ 218 MGET(m, M_DONTWAIT, MT_HEADER); 219 if (m == NULL) 220 return (ENOBUFS); 221 #define MAXLINKHDR 32 /* belongs elsewhere */ 222 #define DATASPACE (MMAXOFF - (MMINOFF + MAXLINKHDR + sizeof (struct tcpiphdr))) 223 m->m_off = MMINOFF + MAXLINKHDR; 224 m->m_len = sizeof (struct tcpiphdr); 225 ti = mtod(m, struct tcpiphdr *); 226 if (len) { 227 if (tp->t_force && len == 1) 228 tcpstat.tcps_sndprobe++; 229 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 230 tcpstat.tcps_sndrexmitpack++; 231 tcpstat.tcps_sndrexmitbyte += len; 232 } else { 233 tcpstat.tcps_sndpack++; 234 tcpstat.tcps_sndbyte += len; 235 } 236 if (len <= DATASPACE) { 237 m_copydata(so->so_snd.sb_mb, off, (int) len, 238 mtod(m, caddr_t) + sizeof(struct tcpiphdr)); 239 m->m_len += len; 240 } else { 241 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 242 if (m->m_next == 0) 243 len = 0; 244 } 245 } else if (tp->t_flags & TF_ACKNOW) 246 tcpstat.tcps_sndacks++; 247 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 248 tcpstat.tcps_sndctrl++; 249 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 250 tcpstat.tcps_sndurg++; 251 else 252 tcpstat.tcps_sndwinup++; 253 254 if (tp->t_template == 0) 255 panic("tcp_output"); 256 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 257 258 /* 259 * Fill in fields, remembering maximum advertised 260 * window for use in delaying messages about window sizes. 261 * If resending a FIN, be sure not to use a new sequence number. 262 */ 263 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 264 tp->snd_nxt == tp->snd_max) 265 tp->snd_nxt--; 266 ti->ti_seq = htonl(tp->snd_nxt); 267 ti->ti_ack = htonl(tp->rcv_nxt); 268 /* 269 * Before ESTABLISHED, force sending of initial options 270 * unless TCP set to not do any options. 271 */ 272 opt = NULL; 273 if (flags & TH_SYN && (tp->t_flags & TF_NOOPT) == 0) { 274 u_short mss; 275 276 mss = MIN(so->so_rcv.sb_hiwat / 2, tcp_mss(tp)); 277 if (mss > IP_MSS - sizeof(struct tcpiphdr)) { 278 opt = tcp_initopt; 279 optlen = sizeof (tcp_initopt); 280 *(u_short *)(opt + 2) = htons(mss); 281 } 282 } 283 if (opt) { 284 m0 = m->m_next; 285 m->m_next = m_get(M_DONTWAIT, MT_DATA); 286 if (m->m_next == 0) { 287 (void) m_free(m); 288 m_freem(m0); 289 return (ENOBUFS); 290 } 291 m->m_next->m_next = m0; 292 m0 = m->m_next; 293 m0->m_len = optlen; 294 bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen); 295 opt = (u_char *)(mtod(m0, caddr_t) + optlen); 296 while (m0->m_len & 0x3) { 297 *opt++ = TCPOPT_EOL; 298 m0->m_len++; 299 } 300 optlen = m0->m_len; 301 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 302 } 303 ti->ti_flags = flags; 304 /* 305 * Calculate receive window. Don't shrink window, 306 * but avoid silly window syndrome. 307 */ 308 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) 309 win = 0; 310 if (win > IP_MAXPACKET) 311 win = IP_MAXPACKET; 312 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 313 win = (long)(tp->rcv_adv - tp->rcv_nxt); 314 ti->ti_win = htons((u_short)win); 315 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 316 ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 317 ti->ti_flags |= TH_URG; 318 } else 319 /* 320 * If no urgent pointer to send, then we pull 321 * the urgent pointer to the left edge of the send window 322 * so that it doesn't drift into the send window on sequence 323 * number wraparound. 324 */ 325 tp->snd_up = tp->snd_una; /* drag it along */ 326 /* 327 * If anything to send and we can send it all, set PUSH. 328 * (This will keep happy those implementations which only 329 * give data to the user when a buffer fills or a PUSH comes in.) 330 */ 331 if (len && off+len == so->so_snd.sb_cc) 332 ti->ti_flags |= TH_PUSH; 333 334 /* 335 * Put TCP length in extended header, and then 336 * checksum extended header and data. 337 */ 338 if (len + optlen) 339 ti->ti_len = htons((u_short)(sizeof(struct tcphdr) + 340 optlen + len)); 341 ti->ti_sum = in_cksum(m, 342 (int)(sizeof (struct tcpiphdr) + (int)optlen + len)); 343 344 /* 345 * In transmit state, time the transmission and arrange for 346 * the retransmit. In persist state, just set snd_max. 347 */ 348 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { 349 tcp_seq startseq = tp->snd_nxt; 350 351 /* 352 * Advance snd_nxt over sequence space of this segment. 353 */ 354 if (flags & TH_SYN) 355 tp->snd_nxt++; 356 if (flags & TH_FIN) { 357 tp->snd_nxt++; 358 tp->t_flags |= TF_SENTFIN; 359 } 360 tp->snd_nxt += len; 361 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 362 tp->snd_max = tp->snd_nxt; 363 /* 364 * Time this transmission if not a retransmission and 365 * not currently timing anything. 366 */ 367 if (tp->t_rtt == 0) { 368 tp->t_rtt = 1; 369 tp->t_rtseq = startseq; 370 tcpstat.tcps_segstimed++; 371 } 372 } 373 374 /* 375 * Set retransmit timer if not currently set, 376 * and not doing an ack or a keep-alive probe. 377 * Initial value for retransmit timer is smoothed 378 * round-trip time + 2 * round-trip time variance. 379 * Initialize shift counter which is used for backoff 380 * of retransmit time. 381 */ 382 if (tp->t_timer[TCPT_REXMT] == 0 && 383 tp->snd_nxt != tp->snd_una) { 384 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 385 if (tp->t_timer[TCPT_PERSIST]) { 386 tp->t_timer[TCPT_PERSIST] = 0; 387 tp->t_rxtshift = 0; 388 } 389 } 390 } else 391 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 392 tp->snd_max = tp->snd_nxt + len; 393 394 /* 395 * Trace. 396 */ 397 if (so->so_options & SO_DEBUG) 398 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 399 400 /* 401 * Fill in IP length and desired time to live and 402 * send to IP level. 403 */ 404 ((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len; 405 ((struct ip *)ti)->ip_ttl = TCP_TTL; 406 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 407 so->so_options & SO_DONTROUTE); 408 if (error) { 409 if (error == ENOBUFS) { 410 tcp_quench(tp->t_inpcb); 411 return (0); 412 } 413 return (error); 414 } 415 tcpstat.tcps_sndtotal++; 416 417 /* 418 * Data sent (as far as we can tell). 419 * If this advertises a larger window than any other segment, 420 * then remember the size of the advertised window. 421 * Any pending ACK has now been sent. 422 */ 423 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 424 tp->rcv_adv = tp->rcv_nxt + win; 425 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 426 if (sendalot) 427 goto again; 428 return (0); 429 } 430 431 tcp_setpersist(tp) 432 register struct tcpcb *tp; 433 { 434 register t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 435 436 if (tp->t_timer[TCPT_REXMT]) 437 panic("tcp_output REXMT"); 438 /* 439 * Start/restart persistance timer. 440 */ 441 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 442 t * tcp_backoff[tp->t_rxtshift], 443 TCPTV_PERSMIN, TCPTV_PERSMAX); 444 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 445 tp->t_rxtshift++; 446 } 447