1 /* tcp_output.c 6.6 84/11/01 */ 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 len = MIN(so->so_snd.sb_cc, tp->snd_wnd+tp->t_force) - off; 58 if (len < 0) 59 return (0); /* ??? */ /* past FIN */ 60 if (len > tp->t_maxseg) { 61 len = tp->t_maxseg; 62 /* 63 * Don't send more than one segment if retransmitting. 64 */ 65 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) 66 sendalot = 1; 67 } 68 69 flags = tcp_outflags[tp->t_state]; 70 if (tp->snd_nxt + len < tp->snd_una + so->so_snd.sb_cc) 71 flags &= ~TH_FIN; 72 if (flags & (TH_SYN|TH_RST|TH_FIN)) 73 goto send; 74 if (SEQ_GT(tp->snd_up, tp->snd_una)) 75 goto send; 76 77 /* 78 * Sender silly window avoidance. If connection is idle 79 * and can send all data, a maximum segment, 80 * at least a maximum default-size segment do it, 81 * or are forced, do it; otherwise don't bother. 82 */ 83 if (len) { 84 if (len == tp->t_maxseg || len >= so->so_snd.sb_cc) /* off = 0*/ 85 goto send; 86 if (len >= TCP_MSS) /* a lot */ 87 goto send; 88 if (tp->t_force) 89 goto send; 90 } 91 92 /* 93 * Send if we owe peer an ACK. 94 */ 95 if (tp->t_flags&TF_ACKNOW) 96 goto send; 97 98 99 /* 100 * Calculate available window in i, and also amount 101 * of window known to peer (as advertised window less 102 * next expected input.) If this is 35% or more of the 103 * maximum possible window, then want to send a segment to peer. 104 */ 105 win = sbspace(&so->so_rcv); 106 if (win > 0 && 107 ((100*(win-(tp->rcv_adv-tp->rcv_nxt))/so->so_rcv.sb_hiwat) >= 35)) 108 goto send; 109 110 /* 111 * TCP window updates are not reliable, rather a polling protocol 112 * using ``persist'' packets is used to insure receipt of window 113 * updates. The three ``states'' for the output side are: 114 * idle not doing retransmits or persists 115 * persisting to move a zero window 116 * (re)transmitting and thereby not persisting 117 * 118 * tp->t_timer[TCPT_PERSIST] 119 * is set when we are in persist state. 120 * tp->t_force 121 * is set when we are called to send a persist packet. 122 * tp->t_timer[TCPT_REXMT] 123 * is set when we are retransmitting 124 * The output side is idle when both timers are zero. 125 * 126 * If send window is closed, there is data to transmit, and no 127 * retransmit or persist is pending, then go to persist state, 128 * arranging to force out a byte to get more current window information 129 * if nothing happens soon. 130 */ 131 if (tp->snd_wnd == 0 && so->so_snd.sb_cc && 132 tp->t_timer[TCPT_REXMT] == 0 && tp->t_timer[TCPT_PERSIST] == 0) { 133 tp->t_rxtshift = 0; 134 tcp_setpersist(tp); 135 } 136 137 /* 138 * No reason to send a segment, just return. 139 */ 140 return (0); 141 142 send: 143 /* 144 * Grab a header mbuf, attaching a copy of data to 145 * be transmitted, and initialize the header from 146 * the template for sends on this connection. 147 */ 148 MGET(m, M_DONTWAIT, MT_HEADER); 149 if (m == NULL) 150 return (ENOBUFS); 151 m->m_off = MMAXOFF - sizeof (struct tcpiphdr); 152 m->m_len = sizeof (struct tcpiphdr); 153 if (len) { 154 m->m_next = m_copy(so->so_snd.sb_mb, off, len); 155 if (m->m_next == 0) 156 len = 0; 157 } 158 ti = mtod(m, struct tcpiphdr *); 159 if (tp->t_template == 0) 160 panic("tcp_output"); 161 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 162 163 /* 164 * Fill in fields, remembering maximum advertised 165 * window for use in delaying messages about window sizes. 166 */ 167 ti->ti_seq = tp->snd_nxt; 168 ti->ti_ack = tp->rcv_nxt; 169 ti->ti_seq = htonl(ti->ti_seq); 170 ti->ti_ack = htonl(ti->ti_ack); 171 /* 172 * Before ESTABLISHED, force sending of initial options 173 * unless TCP set to not do any options. 174 */ 175 if (tp->t_state < TCPS_ESTABLISHED) { 176 int mss; 177 178 if (tp->t_flags&TF_NOOPT) 179 goto noopt; 180 mss = MIN(so->so_rcv.sb_hiwat / 2, tcp_mss(tp)); 181 if (mss <= IP_MSS - sizeof(struct tcpiphdr)) 182 goto noopt; 183 opt = tcp_initopt; 184 optlen = sizeof (tcp_initopt); 185 *(u_short *)(opt + 2) = htons(mss); 186 } else { 187 if (tp->t_tcpopt == 0) 188 goto noopt; 189 opt = mtod(tp->t_tcpopt, u_char *); 190 optlen = tp->t_tcpopt->m_len; 191 } 192 if (opt) { 193 m0 = m->m_next; 194 m->m_next = m_get(M_DONTWAIT, MT_DATA); 195 if (m->m_next == 0) { 196 (void) m_free(m); 197 m_freem(m0); 198 return (ENOBUFS); 199 } 200 m->m_next->m_next = m0; 201 m0 = m->m_next; 202 m0->m_len = optlen; 203 bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen); 204 opt = (u_char *)(mtod(m0, caddr_t) + optlen); 205 while (m0->m_len & 0x3) { 206 *opt++ = TCPOPT_EOL; 207 m0->m_len++; 208 } 209 optlen = m0->m_len; 210 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 211 } 212 noopt: 213 ti->ti_flags = flags; 214 win = sbspace(&so->so_rcv); 215 if (win < so->so_rcv.sb_hiwat / 4) /* avoid silly window */ 216 win = 0; 217 if (win > 0) 218 ti->ti_win = htons((u_short)win); 219 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 220 ti->ti_urp = tp->snd_up - tp->snd_nxt; 221 ti->ti_urp = htons(ti->ti_urp); 222 ti->ti_flags |= TH_URG; 223 } else 224 /* 225 * If no urgent pointer to send, then we pull 226 * the urgent pointer to the left edge of the send window 227 * so that it doesn't drift into the send window on sequence 228 * number wraparound. 229 */ 230 tp->snd_up = tp->snd_una; /* drag it along */ 231 /* 232 * If anything to send and we can send it all, set PUSH. 233 * (This will keep happy those implementations which only 234 * give data to the user when a buffer fills or a PUSH comes in.) 235 */ 236 if (len && off+len == so->so_snd.sb_cc) 237 ti->ti_flags |= TH_PUSH; 238 239 /* 240 * Put TCP length in extended header, and then 241 * checksum extended header and data. 242 */ 243 if (len + optlen) { 244 ti->ti_len = sizeof (struct tcphdr) + optlen + len; 245 ti->ti_len = htons((u_short)ti->ti_len); 246 } 247 ti->ti_sum = in_cksum(m, sizeof (struct tcpiphdr) + (int)optlen + len); 248 249 /* 250 * In transmit state, time the transmission and arrange for 251 * the retransmit. In persist state, reset persist time for 252 * next persist. 253 */ 254 if (tp->t_force == 0) { 255 /* 256 * Advance snd_nxt over sequence space of this segment. 257 */ 258 if (flags & (TH_SYN|TH_FIN)) 259 tp->snd_nxt++; 260 tp->snd_nxt += len; 261 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 262 tp->snd_max = tp->snd_nxt; 263 /* 264 * Time this transmission if not a retransmission and 265 * not currently timing anything. 266 */ 267 if (tp->t_rtt == 0) { 268 tp->t_rtt = 1; 269 tp->t_rtseq = tp->snd_nxt - len; 270 } 271 } 272 273 /* 274 * Set retransmit timer if not currently set. 275 * Initial value for retransmit timer to tcp_beta*tp->t_srtt. 276 * Initialize shift counter which is used for exponential 277 * backoff of retransmit time. 278 */ 279 if (tp->t_timer[TCPT_REXMT] == 0 && 280 tp->snd_nxt != tp->snd_una) { 281 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 282 tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX); 283 tp->t_rxtshift = 0; 284 } 285 tp->t_timer[TCPT_PERSIST] = 0; 286 } else { 287 if (SEQ_GT(tp->snd_una+1, tp->snd_max)) 288 tp->snd_max = tp->snd_una+1; 289 } 290 291 /* 292 * Trace. 293 */ 294 if (so->so_options & SO_DEBUG) 295 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 296 297 /* 298 * Fill in IP length and desired time to live and 299 * send to IP level. 300 */ 301 ((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len; 302 ((struct ip *)ti)->ip_ttl = TCP_TTL; 303 if (so->so_options & SO_DONTROUTE) 304 error = 305 ip_output(m, tp->t_ipopt, (struct route *)0, IP_ROUTETOIF); 306 else 307 error = ip_output(m, tp->t_ipopt, &tp->t_inpcb->inp_route, 0); 308 if (error) 309 return (error); 310 311 /* 312 * Data sent (as far as we can tell). 313 * If this advertises a larger window than any other segment, 314 * then remember the size of the advertised window. 315 * Drop send for purpose of ACK requirements. 316 */ 317 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 318 tp->rcv_adv = tp->rcv_nxt + win; 319 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 320 if (sendalot && tp->t_force == 0) 321 goto again; 322 return (0); 323 } 324 325 tcp_setpersist(tp) 326 register struct tcpcb *tp; 327 { 328 329 if (tp->t_timer[TCPT_REXMT]) 330 panic("tcp_output REXMT"); 331 /* 332 * Start/restart persistance timer. 333 */ 334 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 335 ((int)(tcp_beta * tp->t_srtt)) << tp->t_rxtshift, 336 TCPTV_PERSMIN, TCPTV_MAX); 337 tp->t_rxtshift++; 338 if (tp->t_rxtshift >= TCP_MAXRXTSHIFT) 339 tp->t_rxtshift = 0; 340 } 341