Commit | Line | Data |
---|---|---|
8ae0e4b4 | 1 | /* |
4a86c86b | 2 | * Copyright (c) 1982, 1986, 1988 Regents of the University of California. |
2b6b6284 | 3 | * All rights reserved. |
8ae0e4b4 | 4 | * |
2b6b6284 | 5 | * Redistribution and use in source and binary forms are permitted |
616d42db KB |
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. | |
2b6b6284 | 16 | * |
ccc60b2b | 17 | * @(#)tcp_output.c 7.19 (Berkeley) %G% |
8ae0e4b4 | 18 | */ |
76ee76df | 19 | |
20666ad3 JB |
20 | #include "param.h" |
21 | #include "systm.h" | |
4a86c86b | 22 | #include "malloc.h" |
20666ad3 JB |
23 | #include "mbuf.h" |
24 | #include "protosw.h" | |
25 | #include "socket.h" | |
26 | #include "socketvar.h" | |
27 | #include "errno.h" | |
f4d55810 | 28 | |
c124e997 | 29 | #include "../net/route.h" |
f4d55810 | 30 | |
20666ad3 | 31 | #include "in.h" |
20666ad3 JB |
32 | #include "in_systm.h" |
33 | #include "ip.h" | |
ccc60b2b | 34 | #include "in_pcb.h" |
20666ad3 JB |
35 | #include "ip_var.h" |
36 | #include "tcp.h" | |
0974b45c | 37 | #define TCPOUTFLAGS |
20666ad3 JB |
38 | #include "tcp_fsm.h" |
39 | #include "tcp_seq.h" | |
40 | #include "tcp_timer.h" | |
41 | #include "tcp_var.h" | |
42 | #include "tcpip.h" | |
43 | #include "tcp_debug.h" | |
76ee76df | 44 | |
8b5a83bb | 45 | /* |
77a4e3ca | 46 | * Initial options. |
8b5a83bb | 47 | */ |
8b5a83bb | 48 | u_char tcp_initopt[4] = { TCPOPT_MAXSEG, 4, 0x0, 0x0, }; |
8b5a83bb | 49 | |
ea727f86 | 50 | /* |
4aed14e3 | 51 | * Tcp output routine: figure out what should be sent and send it. |
ea727f86 | 52 | */ |
a6503abf | 53 | tcp_output(tp) |
53a5409e | 54 | register struct tcpcb *tp; |
ea727f86 | 55 | { |
53a5409e | 56 | register struct socket *so = tp->t_inpcb->inp_socket; |
8129ec6e | 57 | register long len, win; |
a6503abf | 58 | struct mbuf *m0; |
acad00cc | 59 | int off, flags, error; |
a6503abf BJ |
60 | register struct mbuf *m; |
61 | register struct tcpiphdr *ti; | |
8b5a83bb BJ |
62 | u_char *opt; |
63 | unsigned optlen = 0; | |
acad00cc | 64 | int idle, sendalot; |
76ee76df | 65 | |
a6503abf | 66 | /* |
8ae6c089 | 67 | * Determine length of data that should be transmitted, |
0974b45c BJ |
68 | * and flags that will be used. |
69 | * If there is some data or critical controls (SYN, RST) | |
70 | * to send, then transmit; otherwise, investigate further. | |
a6503abf | 71 | */ |
acad00cc | 72 | idle = (tp->snd_max == tp->snd_una); |
2266a466 BJ |
73 | again: |
74 | sendalot = 0; | |
a6503abf | 75 | off = tp->snd_nxt - tp->snd_una; |
4a86c86b | 76 | win = min(tp->snd_wnd, tp->snd_cwnd); |
e4af65f3 | 77 | |
eaf69575 MK |
78 | /* |
79 | * If in persist timeout with window of 0, send 1 byte. | |
acad00cc MK |
80 | * Otherwise, if window is small but nonzero |
81 | * and timer expired, we will send what we can | |
82 | * and go to transmit state. | |
eaf69575 MK |
83 | */ |
84 | if (tp->t_force) { | |
e4af65f3 | 85 | if (win == 0) |
eaf69575 MK |
86 | win = 1; |
87 | else { | |
88 | tp->t_timer[TCPT_PERSIST] = 0; | |
89 | tp->t_rxtshift = 0; | |
90 | } | |
91 | } | |
acad00cc | 92 | |
4a86c86b | 93 | len = min(so->so_snd.sb_cc, win) - off; |
0974b45c | 94 | flags = tcp_outflags[tp->t_state]; |
e4af65f3 MK |
95 | |
96 | if (len < 0) { | |
97 | /* | |
dd0ed0b1 MK |
98 | * If FIN has been sent but not acked, |
99 | * but we haven't been called to retransmit, | |
449e69b4 MK |
100 | * len will be -1. Otherwise, window shrank |
101 | * after we sent into it. If window shrank to 0, | |
102 | * cancel pending retransmit and pull snd_nxt | |
103 | * back to (closed) window. We will enter persist | |
104 | * state below. If the window didn't close completely, | |
dd0ed0b1 | 105 | * just wait for an ACK. |
e4af65f3 | 106 | */ |
449e69b4 MK |
107 | len = 0; |
108 | if (win == 0) { | |
faa26a98 MK |
109 | tp->t_timer[TCPT_REXMT] = 0; |
110 | tp->snd_nxt = tp->snd_una; | |
449e69b4 | 111 | } |
faa26a98 MK |
112 | } |
113 | if (len > tp->t_maxseg) { | |
114 | len = tp->t_maxseg; | |
2e5a76f2 | 115 | sendalot = 1; |
e4af65f3 | 116 | } |
baf677ce MK |
117 | if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) |
118 | flags &= ~TH_FIN; | |
e4af65f3 | 119 | win = sbspace(&so->so_rcv); |
acad00cc | 120 | |
faa26a98 MK |
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; | |
acad00cc MK |
130 | /* |
131 | * Send if we owe peer an ACK. | |
132 | */ | |
e4af65f3 MK |
133 | if (tp->t_flags & TF_ACKNOW) |
134 | goto send; | |
faa26a98 | 135 | if (flags & (TH_SYN|TH_RST)) |
acad00cc | 136 | goto send; |
8ae6c089 | 137 | if (SEQ_GT(tp->snd_up, tp->snd_una)) |
a6503abf BJ |
138 | goto send; |
139 | ||
8ae6c089 | 140 | /* |
7d304adf MK |
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, | |
8ae6c089 | 144 | * or are forced, do it; otherwise don't bother. |
18a438b6 MK |
145 | * If peer's buffer is tiny, then send |
146 | * when window is at least half open. | |
eaf69575 MK |
147 | * If retransmitting (possibly after persist timer forced us |
148 | * to send into a small window), then must resend. | |
8ae6c089 BJ |
149 | */ |
150 | if (len) { | |
3d92549d | 151 | if (len == tp->t_maxseg) |
8ae6c089 | 152 | goto send; |
acad00cc MK |
153 | if ((idle || tp->t_flags & TF_NODELAY) && |
154 | len + off >= so->so_snd.sb_cc) | |
8ae6c089 BJ |
155 | goto send; |
156 | if (tp->t_force) | |
157 | goto send; | |
18a438b6 MK |
158 | if (len >= tp->max_sndwnd / 2) |
159 | goto send; | |
eaf69575 MK |
160 | if (SEQ_LT(tp->snd_nxt, tp->snd_max)) |
161 | goto send; | |
e4af65f3 | 162 | } |
76ee76df | 163 | |
a6503abf | 164 | /* |
acad00cc MK |
165 | * Compare available window to amount of window |
166 | * known to peer (as advertised window less | |
449e69b4 | 167 | * next expected input). If the difference is at least two |
ccc60b2b | 168 | * max size segments, or at least 50% of the maximum possible |
449e69b4 | 169 | * window, then want to send a window update to peer. |
a6503abf | 170 | */ |
a6bbda13 MK |
171 | if (win > 0) { |
172 | int adv = win - (tp->rcv_adv - tp->rcv_nxt); | |
173 | ||
ccc60b2b | 174 | if (adv >= 2 * tp->t_maxseg) |
a6bbda13 | 175 | goto send; |
ccc60b2b | 176 | if (2 * adv >= so->so_rcv.sb_hiwat) |
a6bbda13 MK |
177 | goto send; |
178 | } | |
a6503abf | 179 | |
2266a466 BJ |
180 | /* |
181 | * TCP window updates are not reliable, rather a polling protocol | |
182 | * using ``persist'' packets is used to insure receipt of window | |
183 | * updates. The three ``states'' for the output side are: | |
184 | * idle not doing retransmits or persists | |
acad00cc | 185 | * persisting to move a small or zero window |
2266a466 BJ |
186 | * (re)transmitting and thereby not persisting |
187 | * | |
188 | * tp->t_timer[TCPT_PERSIST] | |
189 | * is set when we are in persist state. | |
190 | * tp->t_force | |
191 | * is set when we are called to send a persist packet. | |
192 | * tp->t_timer[TCPT_REXMT] | |
193 | * is set when we are retransmitting | |
194 | * The output side is idle when both timers are zero. | |
195 | * | |
eaf69575 MK |
196 | * If send window is too small, there is data to transmit, and no |
197 | * retransmit or persist is pending, then go to persist state. | |
198 | * If nothing happens soon, send when timer expires: | |
199 | * if window is nonzero, transmit what we can, | |
200 | * otherwise force out a byte. | |
2266a466 | 201 | */ |
eaf69575 MK |
202 | if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && |
203 | tp->t_timer[TCPT_PERSIST] == 0) { | |
2266a466 BJ |
204 | tp->t_rxtshift = 0; |
205 | tcp_setpersist(tp); | |
206 | } | |
207 | ||
a6503abf BJ |
208 | /* |
209 | * No reason to send a segment, just return. | |
210 | */ | |
f1b2fa5b | 211 | return (0); |
a6503abf BJ |
212 | |
213 | send: | |
214 | /* | |
215 | * Grab a header mbuf, attaching a copy of data to | |
216 | * be transmitted, and initialize the header from | |
217 | * the template for sends on this connection. | |
218 | */ | |
4a86c86b | 219 | MGETHDR(m, M_DONTWAIT, MT_HEADER); |
60d68e9e | 220 | if (m == NULL) |
8a2f82db | 221 | return (ENOBUFS); |
4a86c86b | 222 | m->m_data += max_linkhdr; |
53a5409e | 223 | m->m_len = sizeof (struct tcpiphdr); |
4a86c86b | 224 | m->m_pkthdr.rcvif = (struct ifnet *)0; |
8129ec6e | 225 | ti = mtod(m, struct tcpiphdr *); |
a6503abf | 226 | if (len) { |
3b52afc5 MK |
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 | } | |
4a86c86b | 236 | if (len <= MHLEN - sizeof (struct tcpiphdr) - max_linkhdr) { |
8129ec6e MK |
237 | if (m->m_next == 0) |
238 | len = 0; | |
239 | } | |
3b52afc5 MK |
240 | } else if (tp->t_flags & TF_ACKNOW) |
241 | tcpstat.tcps_sndacks++; | |
242 | else if (flags & (TH_SYN|TH_FIN|TH_RST)) | |
243 | tcpstat.tcps_sndctrl++; | |
244 | else if (SEQ_GT(tp->snd_up, tp->snd_una)) | |
245 | tcpstat.tcps_sndurg++; | |
246 | else | |
247 | tcpstat.tcps_sndwinup++; | |
248 | ||
a6503abf BJ |
249 | if (tp->t_template == 0) |
250 | panic("tcp_output"); | |
f1b2fa5b | 251 | bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr)); |
a6503abf BJ |
252 | |
253 | /* | |
254 | * Fill in fields, remembering maximum advertised | |
255 | * window for use in delaying messages about window sizes. | |
faa26a98 | 256 | * If resending a FIN, be sure not to use a new sequence number. |
a6503abf | 257 | */ |
3d92549d MK |
258 | if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && |
259 | tp->snd_nxt == tp->snd_max) | |
faa26a98 | 260 | tp->snd_nxt--; |
acad00cc MK |
261 | ti->ti_seq = htonl(tp->snd_nxt); |
262 | ti->ti_ack = htonl(tp->rcv_nxt); | |
8b5a83bb BJ |
263 | /* |
264 | * Before ESTABLISHED, force sending of initial options | |
265 | * unless TCP set to not do any options. | |
266 | */ | |
acad00cc | 267 | opt = NULL; |
1ccb6fcd | 268 | if (flags & TH_SYN && (tp->t_flags & TF_NOOPT) == 0) { |
8011f5df | 269 | u_short mss; |
99578149 | 270 | |
4a86c86b | 271 | mss = min(so->so_rcv.sb_hiwat / 2, tcp_mss(tp)); |
acad00cc MK |
272 | if (mss > IP_MSS - sizeof(struct tcpiphdr)) { |
273 | opt = tcp_initopt; | |
274 | optlen = sizeof (tcp_initopt); | |
275 | *(u_short *)(opt + 2) = htons(mss); | |
276 | } | |
8b5a83bb | 277 | } |
77a4e3ca | 278 | if (opt) { |
f1b2fa5b | 279 | m0 = m->m_next; |
cce93e4b | 280 | m->m_next = m_get(M_DONTWAIT, MT_DATA); |
0974b45c BJ |
281 | if (m->m_next == 0) { |
282 | (void) m_free(m); | |
8b5a83bb | 283 | m_freem(m0); |
8a2f82db | 284 | return (ENOBUFS); |
0974b45c BJ |
285 | } |
286 | m->m_next->m_next = m0; | |
8b5a83bb | 287 | m0 = m->m_next; |
8b5a83bb | 288 | m0->m_len = optlen; |
668cc26d | 289 | bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen); |
8b5a83bb | 290 | opt = (u_char *)(mtod(m0, caddr_t) + optlen); |
8b5a83bb BJ |
291 | while (m0->m_len & 0x3) { |
292 | *opt++ = TCPOPT_EOL; | |
293 | m0->m_len++; | |
294 | } | |
295 | optlen = m0->m_len; | |
296 | ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; | |
0974b45c BJ |
297 | } |
298 | ti->ti_flags = flags; | |
acad00cc MK |
299 | /* |
300 | * Calculate receive window. Don't shrink window, | |
301 | * but avoid silly window syndrome. | |
302 | */ | |
6def2330 | 303 | if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) |
acad00cc | 304 | win = 0; |
8129ec6e MK |
305 | if (win > IP_MAXPACKET) |
306 | win = IP_MAXPACKET; | |
6def2330 MK |
307 | if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) |
308 | win = (long)(tp->rcv_adv - tp->rcv_nxt); | |
dd861483 MK |
309 | if (win > IP_MAXPACKET) |
310 | win = IP_MAXPACKET; | |
acad00cc | 311 | ti->ti_win = htons((u_short)win); |
0974b45c | 312 | if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { |
8011f5df | 313 | ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); |
a6503abf BJ |
314 | ti->ti_flags |= TH_URG; |
315 | } else | |
316 | /* | |
317 | * If no urgent pointer to send, then we pull | |
318 | * the urgent pointer to the left edge of the send window | |
319 | * so that it doesn't drift into the send window on sequence | |
320 | * number wraparound. | |
321 | */ | |
0974b45c | 322 | tp->snd_up = tp->snd_una; /* drag it along */ |
02c1608b BJ |
323 | /* |
324 | * If anything to send and we can send it all, set PUSH. | |
325 | * (This will keep happy those implementations which only | |
5cdc4d65 | 326 | * give data to the user when a buffer fills or a PUSH comes in.) |
02c1608b | 327 | */ |
02c1608b BJ |
328 | if (len && off+len == so->so_snd.sb_cc) |
329 | ti->ti_flags |= TH_PUSH; | |
a6503abf BJ |
330 | |
331 | /* | |
332 | * Put TCP length in extended header, and then | |
333 | * checksum extended header and data. | |
334 | */ | |
acad00cc MK |
335 | if (len + optlen) |
336 | ti->ti_len = htons((u_short)(sizeof(struct tcphdr) + | |
337 | optlen + len)); | |
9340d736 MK |
338 | ti->ti_sum = in_cksum(m, |
339 | (int)(sizeof (struct tcpiphdr) + (int)optlen + len)); | |
0974b45c BJ |
340 | |
341 | /* | |
2266a466 | 342 | * In transmit state, time the transmission and arrange for |
eaf69575 | 343 | * the retransmit. In persist state, just set snd_max. |
0974b45c | 344 | */ |
eaf69575 | 345 | if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { |
eb5508b2 MK |
346 | tcp_seq startseq = tp->snd_nxt; |
347 | ||
2266a466 | 348 | /* |
8931cb5b | 349 | * Advance snd_nxt over sequence space of this segment. |
2266a466 | 350 | */ |
faa26a98 MK |
351 | if (flags & TH_SYN) |
352 | tp->snd_nxt++; | |
353 | if (flags & TH_FIN) { | |
2266a466 | 354 | tp->snd_nxt++; |
faa26a98 MK |
355 | tp->t_flags |= TF_SENTFIN; |
356 | } | |
2266a466 | 357 | tp->snd_nxt += len; |
eb5508b2 | 358 | if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { |
e45e6858 | 359 | tp->snd_max = tp->snd_nxt; |
eb5508b2 MK |
360 | /* |
361 | * Time this transmission if not a retransmission and | |
362 | * not currently timing anything. | |
363 | */ | |
364 | if (tp->t_rtt == 0) { | |
365 | tp->t_rtt = 1; | |
366 | tp->t_rtseq = startseq; | |
367 | tcpstat.tcps_segstimed++; | |
368 | } | |
369 | } | |
405c9168 | 370 | |
2266a466 | 371 | /* |
eaf69575 | 372 | * Set retransmit timer if not currently set, |
6be9a225 | 373 | * and not doing an ack or a keep-alive probe. |
7cc62c26 MK |
374 | * Initial value for retransmit timer is smoothed |
375 | * round-trip time + 2 * round-trip time variance. | |
e4af65f3 MK |
376 | * Initialize shift counter which is used for backoff |
377 | * of retransmit time. | |
2266a466 BJ |
378 | */ |
379 | if (tp->t_timer[TCPT_REXMT] == 0 && | |
380 | tp->snd_nxt != tp->snd_una) { | |
a6bbda13 MK |
381 | tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; |
382 | if (tp->t_timer[TCPT_PERSIST]) { | |
383 | tp->t_timer[TCPT_PERSIST] = 0; | |
384 | tp->t_rxtshift = 0; | |
385 | } | |
2266a466 | 386 | } |
6be9a225 | 387 | } else |
acad00cc MK |
388 | if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) |
389 | tp->snd_max = tp->snd_nxt + len; | |
a6503abf | 390 | |
f1dd32da BJ |
391 | /* |
392 | * Trace. | |
393 | */ | |
8931cb5b | 394 | if (so->so_options & SO_DEBUG) |
f1dd32da BJ |
395 | tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); |
396 | ||
a6503abf BJ |
397 | /* |
398 | * Fill in IP length and desired time to live and | |
399 | * send to IP level. | |
400 | */ | |
8b5a83bb | 401 | ((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len; |
4a86c86b MK |
402 | if (m->m_flags & M_PKTHDR) |
403 | m->m_pkthdr.len = ((struct ip *)ti)->ip_len; | |
ccc60b2b | 404 | ((struct ip *)ti)->ip_ttl = tcp_ttl; |
9d866d2f | 405 | #if BSD>=43 |
d55475b1 MK |
406 | error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, |
407 | so->so_options & SO_DONTROUTE); | |
9d866d2f MK |
408 | #else |
409 | error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route, | |
410 | so->so_options & SO_DONTROUTE); | |
411 | #endif | |
83591b9d MK |
412 | if (error) { |
413 | if (error == ENOBUFS) { | |
414 | tcp_quench(tp->t_inpcb); | |
415 | return (0); | |
416 | } | |
8a2f82db | 417 | return (error); |
83591b9d | 418 | } |
3b52afc5 | 419 | tcpstat.tcps_sndtotal++; |
a6503abf BJ |
420 | |
421 | /* | |
422 | * Data sent (as far as we can tell). | |
423 | * If this advertises a larger window than any other segment, | |
4aed14e3 | 424 | * then remember the size of the advertised window. |
acad00cc | 425 | * Any pending ACK has now been sent. |
a6503abf | 426 | */ |
be43ac7f | 427 | if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) |
a6503abf | 428 | tp->rcv_adv = tp->rcv_nxt + win; |
0974b45c | 429 | tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); |
acad00cc | 430 | if (sendalot) |
2266a466 | 431 | goto again; |
8a2f82db | 432 | return (0); |
76ee76df | 433 | } |
2266a466 BJ |
434 | |
435 | tcp_setpersist(tp) | |
436 | register struct tcpcb *tp; | |
437 | { | |
7cc62c26 | 438 | register t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; |
2266a466 BJ |
439 | |
440 | if (tp->t_timer[TCPT_REXMT]) | |
441 | panic("tcp_output REXMT"); | |
442 | /* | |
443 | * Start/restart persistance timer. | |
444 | */ | |
445 | TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], | |
7cc62c26 | 446 | t * tcp_backoff[tp->t_rxtshift], |
3d92549d MK |
447 | TCPTV_PERSMIN, TCPTV_PERSMAX); |
448 | if (tp->t_rxtshift < TCP_MAXRXTSHIFT) | |
449 | tp->t_rxtshift++; | |
2266a466 | 450 | } |