1 /* 2 * Copyright (c) 1982 Regents of the University of California. 3 * All rights reserved. The Berkeley software License Agreement 4 * specifies the terms and conditions for redistribution. 5 * 6 * @(#)tcp_output.c 6.18 (Berkeley) 04/13/86 7 */ 8 9 #include "param.h" 10 #include "systm.h" 11 #include "mbuf.h" 12 #include "protosw.h" 13 #include "socket.h" 14 #include "socketvar.h" 15 #include "errno.h" 16 17 #include "../net/route.h" 18 19 #include "in.h" 20 #include "in_pcb.h" 21 #include "in_systm.h" 22 #include "ip.h" 23 #include "ip_var.h" 24 #include "tcp.h" 25 #define TCPOUTFLAGS 26 #include "tcp_fsm.h" 27 #include "tcp_seq.h" 28 #include "tcp_timer.h" 29 #include "tcp_var.h" 30 #include "tcpip.h" 31 #include "tcp_debug.h" 32 33 /* 34 * Initial options. 35 */ 36 u_char tcp_initopt[4] = { TCPOPT_MAXSEG, 4, 0x0, 0x0, }; 37 38 /* 39 * Tcp output routine: figure out what should be sent and send it. 40 */ 41 tcp_output(tp) 42 register struct tcpcb *tp; 43 { 44 register struct socket *so = tp->t_inpcb->inp_socket; 45 register int len, win; 46 struct mbuf *m0; 47 int off, flags, error; 48 register struct mbuf *m; 49 register struct tcpiphdr *ti; 50 u_char *opt; 51 unsigned optlen = 0; 52 int idle, sendalot; 53 54 /* 55 * Determine length of data that should be transmitted, 56 * and flags that will be used. 57 * If there is some data or critical controls (SYN, RST) 58 * to send, then transmit; otherwise, investigate further. 59 */ 60 idle = (tp->snd_max == tp->snd_una); 61 again: 62 sendalot = 0; 63 off = tp->snd_nxt - tp->snd_una; 64 win = MIN(tp->snd_wnd, tp->snd_cwnd); 65 66 /* 67 * If in persist timeout with window of 0, send 1 byte. 68 * Otherwise, if window is small but nonzero 69 * and timer expired, we will send what we can 70 * and go to transmit state. 71 */ 72 if (tp->t_force) { 73 if (win == 0) 74 win = 1; 75 else { 76 tp->t_timer[TCPT_PERSIST] = 0; 77 tp->t_rxtshift = 0; 78 } 79 } 80 81 len = MIN(so->so_snd.sb_cc, win) - off; 82 flags = tcp_outflags[tp->t_state]; 83 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 84 flags &= ~TH_FIN; 85 86 if (len < 0) { 87 /* 88 * If FIN has been sent but not acked, 89 * but we haven't been called to retransmit, 90 * len will be -1; transmit if acking, otherwise no need. 91 * Otherwise, window shrank after we sent into it. 92 * If window shrank to 0, cancel pending retransmit 93 * and pull snd_nxt back to (closed) window. 94 * We will enter persist state below. 95 * If the window didn't close completely, 96 * just wait for an ACK. 97 */ 98 if (flags & TH_FIN) { 99 if (tp->t_flags & TF_ACKNOW) 100 len = 0; 101 else 102 return (0); 103 } else if (win == 0) { 104 tp->t_timer[TCPT_REXMT] = 0; 105 tp->snd_nxt = tp->snd_una; 106 len = 0; 107 } else 108 return (0); 109 } 110 if (len > tp->t_maxseg) { 111 len = tp->t_maxseg; 112 /* 113 * Don't send more than one segment if retransmitting 114 * (or persisting, but then we shouldn't be here). 115 */ 116 if (tp->t_rxtshift == 0) 117 sendalot = 1; 118 } 119 win = sbspace(&so->so_rcv); 120 121 122 /* 123 * If our state indicates that FIN should be sent 124 * and we have not yet done so, or we're retransmitting the FIN, 125 * then we need to send. 126 */ 127 if (flags & TH_FIN && 128 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 129 goto send; 130 /* 131 * Send if we owe peer an ACK. 132 */ 133 if (tp->t_flags & TF_ACKNOW) 134 goto send; 135 if (flags & (TH_SYN|TH_RST)) 136 goto send; 137 if (SEQ_GT(tp->snd_up, tp->snd_una)) 138 goto send; 139 140 /* 141 * Sender silly window avoidance. If connection is idle 142 * and can send all data, a maximum segment, 143 * at least a maximum default-size segment do it, 144 * or are forced, do it; otherwise don't bother. 145 * If peer's buffer is tiny, then send 146 * when window is at least half open. 147 * If retransmitting (possibly after persist timer forced us 148 * to send into a small window), then must resend. 149 */ 150 if (len) { 151 if (len == tp->t_maxseg || len >= TCP_MSS) /* a lot */ 152 goto send; 153 if ((idle || tp->t_flags & TF_NODELAY) && 154 len + off >= so->so_snd.sb_cc) 155 goto send; 156 if (tp->t_force) 157 goto send; 158 if (len >= tp->max_sndwnd / 2) 159 goto send; 160 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 161 goto send; 162 } 163 164 /* 165 * Compare available window to amount of window 166 * known to peer (as advertised window less 167 * next expected input.) If the difference is 35% or more of the 168 * maximum possible window, then want to send a window update to peer. 169 */ 170 if (win > 0 && 171 ((100*(win-(tp->rcv_adv-tp->rcv_nxt))/so->so_rcv.sb_hiwat) >= 35)) 172 goto send; 173 174 /* 175 * TCP window updates are not reliable, rather a polling protocol 176 * using ``persist'' packets is used to insure receipt of window 177 * updates. The three ``states'' for the output side are: 178 * idle not doing retransmits or persists 179 * persisting to move a small or zero window 180 * (re)transmitting and thereby not persisting 181 * 182 * tp->t_timer[TCPT_PERSIST] 183 * is set when we are in persist state. 184 * tp->t_force 185 * is set when we are called to send a persist packet. 186 * tp->t_timer[TCPT_REXMT] 187 * is set when we are retransmitting 188 * The output side is idle when both timers are zero. 189 * 190 * If send window is too small, there is data to transmit, and no 191 * retransmit or persist is pending, then go to persist state. 192 * If nothing happens soon, send when timer expires: 193 * if window is nonzero, transmit what we can, 194 * otherwise force out a byte. 195 */ 196 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && 197 tp->t_timer[TCPT_PERSIST] == 0) { 198 tp->t_rxtshift = 0; 199 tcp_setpersist(tp); 200 } 201 202 /* 203 * No reason to send a segment, just return. 204 */ 205 return (0); 206 207 send: 208 /* 209 * Grab a header mbuf, attaching a copy of data to 210 * be transmitted, and initialize the header from 211 * the template for sends on this connection. 212 */ 213 MGET(m, M_DONTWAIT, MT_HEADER); 214 if (m == NULL) 215 return (ENOBUFS); 216 m->m_off = MMAXOFF - sizeof (struct tcpiphdr); 217 m->m_len = sizeof (struct tcpiphdr); 218 if (len) { 219 m->m_next = m_copy(so->so_snd.sb_mb, off, len); 220 if (m->m_next == 0) 221 len = 0; 222 } 223 ti = mtod(m, struct tcpiphdr *); 224 if (tp->t_template == 0) 225 panic("tcp_output"); 226 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 227 228 /* 229 * Fill in fields, remembering maximum advertised 230 * window for use in delaying messages about window sizes. 231 * If resending a FIN, be sure not to use a new sequence number. 232 */ 233 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 234 tp->snd_nxt != tp->snd_una) 235 tp->snd_nxt--; 236 ti->ti_seq = htonl(tp->snd_nxt); 237 ti->ti_ack = htonl(tp->rcv_nxt); 238 /* 239 * Before ESTABLISHED, force sending of initial options 240 * unless TCP set to not do any options. 241 */ 242 opt = NULL; 243 if (tp->t_state < TCPS_ESTABLISHED && (tp->t_flags & TF_NOOPT) == 0) { 244 u_short mss; 245 246 mss = MIN(so->so_rcv.sb_hiwat / 2, tcp_mss(tp)); 247 if (mss > IP_MSS - sizeof(struct tcpiphdr)) { 248 opt = tcp_initopt; 249 optlen = sizeof (tcp_initopt); 250 *(u_short *)(opt + 2) = htons(mss); 251 } 252 } else if (tp->t_tcpopt) { 253 opt = mtod(tp->t_tcpopt, u_char *); 254 optlen = tp->t_tcpopt->m_len; 255 } 256 if (opt) { 257 m0 = m->m_next; 258 m->m_next = m_get(M_DONTWAIT, MT_DATA); 259 if (m->m_next == 0) { 260 (void) m_free(m); 261 m_freem(m0); 262 return (ENOBUFS); 263 } 264 m->m_next->m_next = m0; 265 m0 = m->m_next; 266 m0->m_len = optlen; 267 bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen); 268 opt = (u_char *)(mtod(m0, caddr_t) + optlen); 269 while (m0->m_len & 0x3) { 270 *opt++ = TCPOPT_EOL; 271 m0->m_len++; 272 } 273 optlen = m0->m_len; 274 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 275 } 276 ti->ti_flags = flags; 277 /* 278 * Calculate receive window. Don't shrink window, 279 * but avoid silly window syndrome. 280 */ 281 if (win < so->so_rcv.sb_hiwat / 4 && win < tp->t_maxseg) 282 win = 0; 283 if (win < (int)(tp->rcv_adv - tp->rcv_nxt)) 284 win = (int)(tp->rcv_adv - tp->rcv_nxt); 285 ti->ti_win = htons((u_short)win); 286 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 287 ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 288 ti->ti_flags |= TH_URG; 289 } else 290 /* 291 * If no urgent pointer to send, then we pull 292 * the urgent pointer to the left edge of the send window 293 * so that it doesn't drift into the send window on sequence 294 * number wraparound. 295 */ 296 tp->snd_up = tp->snd_una; /* drag it along */ 297 /* 298 * If anything to send and we can send it all, set PUSH. 299 * (This will keep happy those implementations which only 300 * give data to the user when a buffer fills or a PUSH comes in.) 301 */ 302 if (len && off+len == so->so_snd.sb_cc) 303 ti->ti_flags |= TH_PUSH; 304 305 /* 306 * Put TCP length in extended header, and then 307 * checksum extended header and data. 308 */ 309 if (len + optlen) 310 ti->ti_len = htons((u_short)(sizeof(struct tcphdr) + 311 optlen + len)); 312 ti->ti_sum = in_cksum(m, sizeof (struct tcpiphdr) + (int)optlen + len); 313 314 /* 315 * In transmit state, time the transmission and arrange for 316 * the retransmit. In persist state, just set snd_max. 317 */ 318 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { 319 /* 320 * Advance snd_nxt over sequence space of this segment. 321 */ 322 if (flags & TH_SYN) 323 tp->snd_nxt++; 324 if (flags & TH_FIN) { 325 tp->snd_nxt++; 326 tp->t_flags |= TF_SENTFIN; 327 } 328 tp->snd_nxt += len; 329 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 330 tp->snd_max = tp->snd_nxt; 331 /* 332 * Time this transmission if not a retransmission and 333 * not currently timing anything. 334 */ 335 if (tp->t_rtt == 0) { 336 tp->t_rtt = 1; 337 tp->t_rtseq = tp->snd_nxt - len; 338 } 339 } 340 341 /* 342 * Set retransmit timer if not currently set, 343 * and not doing an ack or a keep-alive probe. 344 * Initial value for retransmit timer is tcp_beta*tp->t_srtt. 345 * Initialize shift counter which is used for backoff 346 * of retransmit time. 347 */ 348 if (tp->t_timer[TCPT_REXMT] == 0 && 349 tp->snd_nxt != tp->snd_una) { 350 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 351 tcp_beta * (tp->t_srtt ? tp->t_srtt : TCPTV_SRTTDFLT), 352 TCPTV_MIN, TCPTV_MAX); 353 tp->t_rxtshift = 0; 354 tp->t_timer[TCPT_PERSIST] = 0; 355 } 356 } else 357 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 358 tp->snd_max = tp->snd_nxt + len; 359 360 /* 361 * Trace. 362 */ 363 if (so->so_options & SO_DEBUG) 364 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 365 366 /* 367 * Fill in IP length and desired time to live and 368 * send to IP level. 369 */ 370 ((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len; 371 ((struct ip *)ti)->ip_ttl = TCP_TTL; 372 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 373 so->so_options & SO_DONTROUTE); 374 if (error) 375 return (error); 376 377 /* 378 * Data sent (as far as we can tell). 379 * If this advertises a larger window than any other segment, 380 * then remember the size of the advertised window. 381 * Any pending ACK has now been sent. 382 */ 383 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 384 tp->rcv_adv = tp->rcv_nxt + win; 385 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 386 if (sendalot) 387 goto again; 388 return (0); 389 } 390 391 tcp_setpersist(tp) 392 register struct tcpcb *tp; 393 { 394 395 if (tp->t_timer[TCPT_REXMT]) 396 panic("tcp_output REXMT"); 397 /* 398 * Start/restart persistance timer. 399 */ 400 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 401 ((int)(tcp_beta * tp->t_srtt)) << tp->t_rxtshift, 402 TCPTV_PERSMIN, TCPTV_MAX); 403 tp->t_rxtshift++; 404 if (tp->t_rxtshift >= TCP_MAXRXTSHIFT) 405 tp->t_rxtshift = 0; 406 } 407