1 /* tcp_output.c 4.47 82/10/17 */ 2 3 #include "../h/param.h" 4 #include "../h/systm.h" 5 #include "../h/mbuf.h" 6 #include "../h/protosw.h" 7 #include "../h/socket.h" 8 #include "../h/socketvar.h" 9 #include "../netinet/in.h" 10 #include "../net/route.h" 11 #include "../netinet/in_pcb.h" 12 #include "../netinet/in_systm.h" 13 #include "../netinet/ip.h" 14 #include "../netinet/ip_var.h" 15 #include "../netinet/tcp.h" 16 #define TCPOUTFLAGS 17 #include "../netinet/tcp_fsm.h" 18 #include "../netinet/tcp_seq.h" 19 #include "../netinet/tcp_timer.h" 20 #include "../netinet/tcp_var.h" 21 #include "../netinet/tcpip.h" 22 #include "../netinet/tcp_debug.h" 23 #include <errno.h> 24 25 char *tcpstates[]; /* XXX */ 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 /* 50 * Determine length of data that should be transmitted, 51 * and flags that will be used. 52 * If there is some data or critical controls (SYN, RST) 53 * to send, then transmit; otherwise, investigate further. 54 */ 55 again: 56 sendalot = 0; 57 off = tp->snd_nxt - tp->snd_una; 58 len = MIN(so->so_snd.sb_cc, tp->snd_wnd+tp->t_force) - off; 59 if (len < 0) 60 return (0); /* ??? */ /* past FIN */ 61 if (len > tp->t_maxseg) { 62 len = tp->t_maxseg; 63 sendalot = 1; 64 } 65 66 flags = tcp_outflags[tp->t_state]; 67 if (tp->snd_nxt + len < tp->snd_una + so->so_snd.sb_cc) 68 flags &= ~TH_FIN; 69 if (flags & (TH_SYN|TH_RST|TH_FIN)) 70 goto send; 71 if (SEQ_GT(tp->snd_up, tp->snd_una)) 72 goto send; 73 74 /* 75 * Sender silly window avoidance. If can send all data, 76 * a maximum segment, at least 1/4 of window do it, 77 * or are forced, do it; otherwise don't bother. 78 */ 79 if (len) { 80 if (len == tp->t_maxseg || off+len >= so->so_snd.sb_cc) 81 goto send; 82 if (len * 4 >= tp->snd_wnd) /* a lot */ 83 goto send; 84 if (tp->t_force) 85 goto send; 86 } 87 88 /* 89 * Send if we owe peer an ACK. 90 */ 91 if (tp->t_flags&TF_ACKNOW) 92 goto send; 93 94 95 /* 96 * Calculate available window in i, and also amount 97 * of window known to peer (as advertised window less 98 * next expected input.) If this is 35% or more of the 99 * maximum possible window, then want to send a segment to peer. 100 */ 101 win = sbspace(&so->so_rcv); 102 if (win > 0 && 103 ((100*(win-(tp->rcv_adv-tp->rcv_nxt))/so->so_rcv.sb_hiwat) >= 35)) 104 goto send; 105 106 /* 107 * TCP window updates are not reliable, rather a polling protocol 108 * using ``persist'' packets is used to insure receipt of window 109 * updates. The three ``states'' for the output side are: 110 * idle not doing retransmits or persists 111 * persisting to move a zero window 112 * (re)transmitting and thereby not persisting 113 * 114 * tp->t_timer[TCPT_PERSIST] 115 * is set when we are in persist state. 116 * tp->t_force 117 * is set when we are called to send a persist packet. 118 * tp->t_timer[TCPT_REXMT] 119 * is set when we are retransmitting 120 * The output side is idle when both timers are zero. 121 * 122 * If send window is closed, there is data to transmit, and no 123 * retransmit or persist is pending, then go to persist state, 124 * arranging to force out a byte to get more current window information 125 * if nothing happens soon. 126 */ 127 if (tp->snd_wnd == 0 && so->so_snd.sb_cc && 128 tp->t_timer[TCPT_REXMT] == 0 && tp->t_timer[TCPT_PERSIST] == 0) { 129 tp->t_rxtshift = 0; 130 tcp_setpersist(tp); 131 } 132 133 /* 134 * No reason to send a segment, just return. 135 */ 136 return (0); 137 138 send: 139 /* 140 * Grab a header mbuf, attaching a copy of data to 141 * be transmitted, and initialize the header from 142 * the template for sends on this connection. 143 */ 144 MGET(m, 0); 145 if (m == 0) 146 return (ENOBUFS); 147 m->m_off = MMAXOFF - sizeof (struct tcpiphdr); 148 m->m_len = sizeof (struct tcpiphdr); 149 if (len) { 150 m->m_next = m_copy(so->so_snd.sb_mb, off, len); 151 if (m->m_next == 0) 152 len = 0; 153 } 154 ti = mtod(m, struct tcpiphdr *); 155 if (tp->t_template == 0) 156 panic("tcp_output"); 157 bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); 158 159 /* 160 * Fill in fields, remembering maximum advertised 161 * window for use in delaying messages about window sizes. 162 */ 163 ti->ti_seq = tp->snd_nxt; 164 ti->ti_ack = tp->rcv_nxt; 165 #if vax || pdp11 || ns16032 166 ti->ti_seq = htonl(ti->ti_seq); 167 ti->ti_ack = htonl(ti->ti_ack); 168 #endif 169 /* 170 * Before ESTABLISHED, force sending of initial options 171 * unless TCP set to not do any options. 172 */ 173 if (tp->t_state < TCPS_ESTABLISHED) { 174 if (tp->t_flags&TF_NOOPT) 175 goto noopt; 176 opt = tcp_initopt; 177 optlen = sizeof (tcp_initopt); 178 *(u_short *)(opt + 2) = MIN(so->so_rcv.sb_hiwat / 2, 1024); 179 #if vax || pdp11 || ns16032 180 *(u_short *)(opt + 2) = htons(*(u_short *)(opt + 2)); 181 #endif 182 } else { 183 if (tp->t_tcpopt == 0) 184 goto noopt; 185 opt = mtod(tp->t_tcpopt, u_char *); 186 optlen = tp->t_tcpopt->m_len; 187 } 188 if (opt) { 189 m0 = m->m_next; 190 m->m_next = m_get(M_DONTWAIT); 191 if (m->m_next == 0) { 192 (void) m_free(m); 193 m_freem(m0); 194 return (ENOBUFS); 195 } 196 m->m_next->m_next = m0; 197 m0 = m->m_next; 198 m0->m_len = optlen; 199 bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen); 200 opt = (u_char *)(mtod(m0, caddr_t) + optlen); 201 while (m0->m_len & 0x3) { 202 *opt++ = TCPOPT_EOL; 203 m0->m_len++; 204 } 205 optlen = m0->m_len; 206 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 207 } 208 noopt: 209 ti->ti_flags = flags; 210 win = sbspace(&so->so_rcv); 211 if (win < so->so_rcv.sb_hiwat / 4) /* avoid silly window */ 212 win = 0; 213 if (win > 0) 214 #if vax || pdp11 || ns16032 215 ti->ti_win = htons((u_short)win); 216 #else 217 ti->ti_win = win; 218 #endif 219 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 220 ti->ti_urp = tp->snd_up - tp->snd_nxt; 221 #if vax || pdp11 || ns16032 222 ti->ti_urp = htons(ti->ti_urp); 223 #endif 224 ti->ti_flags |= TH_URG; 225 } else 226 /* 227 * If no urgent pointer to send, then we pull 228 * the urgent pointer to the left edge of the send window 229 * so that it doesn't drift into the send window on sequence 230 * number wraparound. 231 */ 232 tp->snd_up = tp->snd_una; /* drag it along */ 233 /* 234 * If anything to send and we can send it all, set PUSH. 235 * (This will keep happy those implementations which only 236 * give data to the user when a buffer fills or a PUSH comes in. 237 */ 238 if (len && off+len == so->so_snd.sb_cc) 239 ti->ti_flags |= TH_PUSH; 240 241 /* 242 * Put TCP length in extended header, and then 243 * checksum extended header and data. 244 */ 245 if (len + optlen) { 246 ti->ti_len = sizeof (struct tcphdr) + optlen + len; 247 #if vax || pdp11 || ns16032 248 ti->ti_len = htons((u_short)ti->ti_len); 249 #endif 250 } 251 ti->ti_sum = in_cksum(m, sizeof (struct tcpiphdr) + (int)optlen + len); 252 253 /* 254 * In transmit state, time the transmission and arrange for 255 * the retransmit. In persist state, reset persist time for 256 * next persist. 257 */ 258 if (tp->t_force == 0) { 259 /* 260 * Advance snd_nxt over sequence space of this segment. 261 */ 262 if (flags & (TH_SYN|TH_FIN)) 263 tp->snd_nxt++; 264 tp->snd_nxt += len; 265 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) 266 tp->snd_max = tp->snd_nxt; 267 268 /* 269 * Time this transmission if not a retransmission and 270 * not currently timing anything. 271 */ 272 if (SEQ_GT(tp->snd_nxt, tp->snd_max) && tp->t_rtt == 0) { 273 tp->t_rtt = 1; 274 tp->t_rtseq = tp->snd_nxt - len; 275 } 276 277 /* 278 * Set retransmit timer if not currently set. 279 * Initial value for retransmit timer to tcp_beta*tp->t_srtt. 280 * Initialize shift counter which is used for exponential 281 * backoff of retransmit time. 282 */ 283 if (tp->t_timer[TCPT_REXMT] == 0 && 284 tp->snd_nxt != tp->snd_una) { 285 TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 286 tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX); 287 tp->t_rtt = 0; 288 tp->t_rxtshift = 0; 289 } 290 tp->t_timer[TCPT_PERSIST] = 0; 291 } else { 292 if (SEQ_GT(tp->snd_una+1, tp->snd_max)) 293 tp->snd_max = tp->snd_una+1; 294 } 295 296 /* 297 * Trace. 298 */ 299 if (so->so_options & SO_DEBUG) 300 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 301 302 /* 303 * Fill in IP length and desired time to live and 304 * send to IP level. 305 */ 306 ((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len; 307 ((struct ip *)ti)->ip_ttl = TCP_TTL; 308 if (error = ip_output(m, tp->t_ipopt, (so->so_options & SO_DONTROUTE) ? 309 &routetoif : &tp->t_inpcb->inp_route, 0)) 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