1 /* tcp_output.c 6.7 84/11/14 */ 2 3 #include "param.h" 4 #include "systm.h" 5 #include "mbuf.h" 6 #include "protosw.h" 7 #include "socket.h" 8 #include "socketvar.h" 9 #include "errno.h" 10 11 #include "../net/route.h" 12 13 #include "in.h" 14 #include "in_pcb.h" 15 #include "in_systm.h" 16 #include "ip.h" 17 #include "ip_var.h" 18 #include "tcp.h" 19 #define TCPOUTFLAGS 20 #include "tcp_fsm.h" 21 #include "tcp_seq.h" 22 #include "tcp_timer.h" 23 #include "tcp_var.h" 24 #include "tcpip.h" 25 #include "tcp_debug.h" 26 27 /* 28 * Initial options. 29 */ 30 u_char tcp_initopt[4] = { TCPOPT_MAXSEG, 4, 0x0, 0x0, }; 31 32 /* 33 * Tcp output routine: figure out what should be sent and send it. 34 */ 35 tcp_output(tp) 36 register struct tcpcb *tp; 37 { 38 register struct socket *so = tp->t_inpcb->inp_socket; 39 register int len; 40 struct mbuf *m0; 41 int off, flags, win, error; 42 register struct mbuf *m; 43 register struct tcpiphdr *ti; 44 u_char *opt; 45 unsigned optlen = 0; 46 int sendalot; 47 48 /* 49 * Determine length of data that should be transmitted, 50 * and flags that will be used. 51 * If there is some data or critical controls (SYN, RST) 52 * to send, then transmit; otherwise, investigate further. 53 */ 54 again: 55 sendalot = 0; 56 off = tp->snd_nxt - tp->snd_una; 57 win = MIN(tp->snd_wnd, tp->snd_cwnd) + tp->t_force; 58 len = MIN(so->so_snd.sb_cc, win) - off; 59 if (len < 0) 60 return (0); /* ??? */ /* past FIN */ 61 if (len > tp->t_maxseg) { 62 len = tp->t_maxseg; 63 /* 64 * Don't send more than one segment if retransmitting. 65 */ 66 if (tp->t_rxtshift == 0) 67 sendalot = 1; 68 } 69 70 flags = tcp_outflags[tp->t_state]; 71 if (tp->snd_nxt + len < tp->snd_una + so->so_snd.sb_cc) 72 flags &= ~TH_FIN; 73 if (flags & (TH_SYN|TH_RST|TH_FIN)) 74 goto send; 75 if (SEQ_GT(tp->snd_up, tp->snd_una)) 76 goto send; 77 78 /* 79 * Sender silly window avoidance. If connection is idle 80 * and can send all data, a maximum segment, 81 * at least a maximum default-size segment do it, 82 * or are forced, do it; otherwise don't bother. 83 */ 84 if (len) { 85 if (len == tp->t_maxseg || len >= so->so_snd.sb_cc) /* off = 0*/ 86 goto send; 87 if (len >= TCP_MSS) /* a lot */ 88 goto send; 89 if (tp->t_force) 90 goto send; 91 } 92 93 /* 94 * Send if we owe peer an ACK. 95 */ 96 if (tp->t_flags&TF_ACKNOW) 97 goto send; 98 99 100 /* 101 * Calculate available window, and also amount 102 * of window known to peer (as advertised window less 103 * next expected input.) If the difference is 35% or more of the 104 * maximum possible window, then want to send a window update to peer. 105 */ 106 win = sbspace(&so->so_rcv); 107 if (win > 0 && 108 ((100*(win-(tp->rcv_adv-tp->rcv_nxt))/so->so_rcv.sb_hiwat) >= 35)) 109 goto send; 110 111 /* 112 * TCP window updates are not reliable, rather a polling protocol 113 * using ``persist'' packets is used to insure receipt of window 114 * updates. The three ``states'' for the output side are: 115 * idle not doing retransmits or persists 116 * persisting to move a zero window 117 * (re)transmitting and thereby not persisting 118 * 119 * tp->t_timer[TCPT_PERSIST] 120 * is set when we are in persist state. 121 * tp->t_force 122 * is set when we are called to send a persist packet. 123 * tp->t_timer[TCPT_REXMT] 124 * is set when we are retransmitting 125 * The output side is idle when both timers are zero. 126 * 127 * If send window is closed, there is data to transmit, and no 128 * retransmit or persist is pending, then go to persist state, 129 * arranging to force out a byte to get more current window information 130 * if nothing happens soon. 131 */ 132 if (tp->snd_wnd == 0 && so->so_snd.sb_cc && 133 tp->t_timer[TCPT_REXMT] == 0 && tp->t_timer[TCPT_PERSIST] == 0) { 134 tp->t_rxtshift = 0; 135 tcp_setpersist(tp); 136 } 137 138 /* 139 * No reason to send a segment, just return. 140 */ 141 return (0); 142 143 send: 144 /* 145 * Grab a header mbuf, attaching a copy of data to 146 * be transmitted, and initialize the header from 147 * the template for sends on this connection. 148 */ 149 MGET(m, M_DONTWAIT, MT_HEADER); 150 if (m == NULL) 151 return (ENOBUFS); 152 m->m_off = MMAXOFF - sizeof (struct tcpiphdr); 153 m->m_len = sizeof (struct tcpiphdr); 154 if (len) { 155 m->m_next = m_copy(so->so_snd.sb_mb, off, len); 156 if (m->m_next == 0) 157 len = 0; 158 } 159 ti = mtod(m, struct tcpiphdr *); 160 if (tp->t_template == 0) 161 panic("tcp_output"); 162 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 163 164 /* 165 * Fill in fields, remembering maximum advertised 166 * window for use in delaying messages about window sizes. 167 */ 168 ti->ti_seq = tp->snd_nxt; 169 ti->ti_ack = tp->rcv_nxt; 170 ti->ti_seq = htonl(ti->ti_seq); 171 ti->ti_ack = htonl(ti->ti_ack); 172 /* 173 * Before ESTABLISHED, force sending of initial options 174 * unless TCP set to not do any options. 175 */ 176 if (tp->t_state < TCPS_ESTABLISHED) { 177 int mss; 178 179 if (tp->t_flags&TF_NOOPT) 180 goto noopt; 181 mss = MIN(so->so_rcv.sb_hiwat / 2, tcp_mss(tp)); 182 if (mss <= IP_MSS - sizeof(struct tcpiphdr)) 183 goto noopt; 184 opt = tcp_initopt; 185 optlen = sizeof (tcp_initopt); 186 *(u_short *)(opt + 2) = htons(mss); 187 } else { 188 if (tp->t_tcpopt == 0) 189 goto noopt; 190 opt = mtod(tp->t_tcpopt, u_char *); 191 optlen = tp->t_tcpopt->m_len; 192 } 193 if (opt) { 194 m0 = m->m_next; 195 m->m_next = m_get(M_DONTWAIT, MT_DATA); 196 if (m->m_next == 0) { 197 (void) m_free(m); 198 m_freem(m0); 199 return (ENOBUFS); 200 } 201 m->m_next->m_next = m0; 202 m0 = m->m_next; 203 m0->m_len = optlen; 204 bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen); 205 opt = (u_char *)(mtod(m0, caddr_t) + optlen); 206 while (m0->m_len & 0x3) { 207 *opt++ = TCPOPT_EOL; 208 m0->m_len++; 209 } 210 optlen = m0->m_len; 211 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 212 } 213 noopt: 214 ti->ti_flags = flags; 215 win = sbspace(&so->so_rcv); 216 if (win < so->so_rcv.sb_hiwat / 4) /* avoid silly window */ 217 win = 0; 218 if (win > 0) 219 ti->ti_win = htons((u_short)win); 220 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 221 ti->ti_urp = tp->snd_up - tp->snd_nxt; 222 ti->ti_urp = htons(ti->ti_urp); 223 ti->ti_flags |= TH_URG; 224 } else 225 /* 226 * If no urgent pointer to send, then we pull 227 * the urgent pointer to the left edge of the send window 228 * so that it doesn't drift into the send window on sequence 229 * number wraparound. 230 */ 231 tp->snd_up = tp->snd_una; /* drag it along */ 232 /* 233 * If anything to send and we can send it all, set PUSH. 234 * (This will keep happy those implementations which only 235 * give data to the user when a buffer fills or a PUSH comes in.) 236 */ 237 if (len && off+len == so->so_snd.sb_cc) 238 ti->ti_flags |= TH_PUSH; 239 240 /* 241 * Put TCP length in extended header, and then 242 * checksum extended header and data. 243 */ 244 if (len + optlen) { 245 ti->ti_len = sizeof (struct tcphdr) + optlen + len; 246 ti->ti_len = htons((u_short)ti->ti_len); 247 } 248 ti->ti_sum = in_cksum(m, sizeof (struct tcpiphdr) + (int)optlen + len); 249 250 /* 251 * In transmit state, time the transmission and arrange for 252 * the retransmit. In persist state, reset persist time for 253 * next persist. 254 */ 255 if (tp->t_force == 0) { 256 /* 257 * Advance snd_nxt over sequence space of this segment. 258 */ 259 if (flags & (TH_SYN|TH_FIN)) 260 tp->snd_nxt++; 261 tp->snd_nxt += len; 262 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 263 tp->snd_max = tp->snd_nxt; 264 /* 265 * Time this transmission if not a retransmission and 266 * not currently timing anything. 267 */ 268 if (tp->t_rtt == 0) { 269 tp->t_rtt = 1; 270 tp->t_rtseq = tp->snd_nxt - len; 271 } 272 } 273 274 /* 275 * Set retransmit timer if not currently set. 276 * Initial value for retransmit timer to tcp_beta*tp->t_srtt. 277 * Initialize shift counter which is used for exponential 278 * backoff of retransmit time. 279 */ 280 if (tp->t_timer[TCPT_REXMT] == 0 && 281 tp->snd_nxt != tp->snd_una) { 282 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 283 tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX); 284 tp->t_rxtshift = 0; 285 } 286 tp->t_timer[TCPT_PERSIST] = 0; 287 } else { 288 if (SEQ_GT(tp->snd_una+1, tp->snd_max)) 289 tp->snd_max = tp->snd_una+1; 290 } 291 292 /* 293 * Trace. 294 */ 295 if (so->so_options & SO_DEBUG) 296 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 297 298 /* 299 * Fill in IP length and desired time to live and 300 * send to IP level. 301 */ 302 ((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len; 303 ((struct ip *)ti)->ip_ttl = TCP_TTL; 304 if (so->so_options & SO_DONTROUTE) 305 error = 306 ip_output(m, tp->t_ipopt, (struct route *)0, IP_ROUTETOIF); 307 else 308 error = ip_output(m, tp->t_ipopt, &tp->t_inpcb->inp_route, 0); 309 if (error) 310 return (error); 311 312 /* 313 * Data sent (as far as we can tell). 314 * If this advertises a larger window than any other segment, 315 * then remember the size of the advertised window. 316 * Drop send for purpose of ACK requirements. 317 */ 318 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 319 tp->rcv_adv = tp->rcv_nxt + win; 320 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 321 if (sendalot && tp->t_force == 0) 322 goto again; 323 return (0); 324 } 325 326 tcp_setpersist(tp) 327 register struct tcpcb *tp; 328 { 329 330 if (tp->t_timer[TCPT_REXMT]) 331 panic("tcp_output REXMT"); 332 /* 333 * Start/restart persistance timer. 334 */ 335 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 336 ((int)(tcp_beta * tp->t_srtt)) << tp->t_rxtshift, 337 TCPTV_PERSMIN, TCPTV_MAX); 338 tp->t_rxtshift++; 339 if (tp->t_rxtshift >= TCP_MAXRXTSHIFT) 340 tp->t_rxtshift = 0; 341 } 342