xref: /original-bsd/sys/netinet/tcp_output.c (revision 3708840b)
1 /*	tcp_output.c	4.54	83/05/12	*/
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 "../h/errno.h"
10 
11 #include "../net/route.h"
12 
13 #include "../netinet/in.h"
14 #include "../netinet/in_pcb.h"
15 #include "../netinet/in_systm.h"
16 #include "../netinet/ip.h"
17 #include "../netinet/ip_var.h"
18 #include "../netinet/tcp.h"
19 #define	TCPOUTFLAGS
20 #include "../netinet/tcp_fsm.h"
21 #include "../netinet/tcp_seq.h"
22 #include "../netinet/tcp_timer.h"
23 #include "../netinet/tcp_var.h"
24 #include "../netinet/tcpip.h"
25 #include "../netinet/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 		sendalot = 1;
63 	}
64 
65 	flags = tcp_outflags[tp->t_state];
66 	if (tp->snd_nxt + len < tp->snd_una + so->so_snd.sb_cc)
67 		flags &= ~TH_FIN;
68 	if (flags & (TH_SYN|TH_RST|TH_FIN))
69 		goto send;
70 	if (SEQ_GT(tp->snd_up, tp->snd_una))
71 		goto send;
72 
73 	/*
74 	 * Sender silly window avoidance.  If can send all data,
75 	 * a maximum segment, at least 1/4 of window do it,
76 	 * or are forced, do it; otherwise don't bother.
77 	 */
78 	if (len) {
79 		if (len == tp->t_maxseg || off+len >= so->so_snd.sb_cc)
80 			goto send;
81 		if (len * 4 >= tp->snd_wnd)		/* a lot */
82 			goto send;
83 		if (tp->t_force)
84 			goto send;
85 	}
86 
87 	/*
88 	 * Send if we owe peer an ACK.
89 	 */
90 	if (tp->t_flags&TF_ACKNOW)
91 		goto send;
92 
93 
94 	/*
95 	 * Calculate available window in i, and also amount
96 	 * of window known to peer (as advertised window less
97 	 * next expected input.)  If this is 35% or more of the
98 	 * maximum possible window, then want to send a segment to peer.
99 	 */
100 	win = sbspace(&so->so_rcv);
101 	if (win > 0 &&
102 	    ((100*(win-(tp->rcv_adv-tp->rcv_nxt))/so->so_rcv.sb_hiwat) >= 35))
103 		goto send;
104 
105 	/*
106 	 * TCP window updates are not reliable, rather a polling protocol
107 	 * using ``persist'' packets is used to insure receipt of window
108 	 * updates.  The three ``states'' for the output side are:
109 	 *	idle			not doing retransmits or persists
110 	 *	persisting		to move a zero window
111 	 *	(re)transmitting	and thereby not persisting
112 	 *
113 	 * tp->t_timer[TCPT_PERSIST]
114 	 *	is set when we are in persist state.
115 	 * tp->t_force
116 	 *	is set when we are called to send a persist packet.
117 	 * tp->t_timer[TCPT_REXMT]
118 	 *	is set when we are retransmitting
119 	 * The output side is idle when both timers are zero.
120 	 *
121 	 * If send window is closed, there is data to transmit, and no
122 	 * retransmit or persist is pending, then go to persist state,
123 	 * arranging to force out a byte to get more current window information
124 	 * if nothing happens soon.
125 	 */
126 	if (tp->snd_wnd == 0 && so->so_snd.sb_cc &&
127 	    tp->t_timer[TCPT_REXMT] == 0 && tp->t_timer[TCPT_PERSIST] == 0) {
128 		tp->t_rxtshift = 0;
129 		tcp_setpersist(tp);
130 	}
131 
132 	/*
133 	 * No reason to send a segment, just return.
134 	 */
135 	return (0);
136 
137 send:
138 	/*
139 	 * Grab a header mbuf, attaching a copy of data to
140 	 * be transmitted, and initialize the header from
141 	 * the template for sends on this connection.
142 	 */
143 	MGET(m, M_DONTWAIT, MT_HEADER);
144 	if (m == NULL)
145 		return (ENOBUFS);
146 	m->m_off = MMAXOFF - sizeof (struct tcpiphdr);
147 	m->m_len = sizeof (struct tcpiphdr);
148 	if (len) {
149 		m->m_next = m_copy(so->so_snd.sb_mb, off, len);
150 		if (m->m_next == 0)
151 			len = 0;
152 	}
153 	ti = mtod(m, struct tcpiphdr *);
154 	if (tp->t_template == 0)
155 		panic("tcp_output");
156 	bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr));
157 
158 	/*
159 	 * Fill in fields, remembering maximum advertised
160 	 * window for use in delaying messages about window sizes.
161 	 */
162 	ti->ti_seq = tp->snd_nxt;
163 	ti->ti_ack = tp->rcv_nxt;
164 	ti->ti_seq = htonl(ti->ti_seq);
165 	ti->ti_ack = htonl(ti->ti_ack);
166 	/*
167 	 * Before ESTABLISHED, force sending of initial options
168 	 * unless TCP set to not do any options.
169 	 */
170 	if (tp->t_state < TCPS_ESTABLISHED) {
171 		if (tp->t_flags&TF_NOOPT)
172 			goto noopt;
173 		opt = tcp_initopt;
174 		optlen = sizeof (tcp_initopt);
175 		*(u_short *)(opt + 2) = MIN(so->so_rcv.sb_hiwat / 2, 1024);
176 		*(u_short *)(opt + 2) = htons(*(u_short *)(opt + 2));
177 	} else {
178 		if (tp->t_tcpopt == 0)
179 			goto noopt;
180 		opt = mtod(tp->t_tcpopt, u_char *);
181 		optlen = tp->t_tcpopt->m_len;
182 	}
183 	if (opt) {
184 		m0 = m->m_next;
185 		m->m_next = m_get(M_DONTWAIT, MT_DATA);
186 		if (m->m_next == 0) {
187 			(void) m_free(m);
188 			m_freem(m0);
189 			return (ENOBUFS);
190 		}
191 		m->m_next->m_next = m0;
192 		m0 = m->m_next;
193 		m0->m_len = optlen;
194 		bcopy((caddr_t)opt, mtod(m0, caddr_t), optlen);
195 		opt = (u_char *)(mtod(m0, caddr_t) + optlen);
196 		while (m0->m_len & 0x3) {
197 			*opt++ = TCPOPT_EOL;
198 			m0->m_len++;
199 		}
200 		optlen = m0->m_len;
201 		ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
202 	}
203 noopt:
204 	ti->ti_flags = flags;
205 	win = sbspace(&so->so_rcv);
206 	if (win < so->so_rcv.sb_hiwat / 4)	/* avoid silly window */
207 		win = 0;
208 	if (win > 0)
209 		ti->ti_win = htons((u_short)win);
210 	if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
211 		ti->ti_urp = tp->snd_up - tp->snd_nxt;
212 		ti->ti_urp = htons(ti->ti_urp);
213 		ti->ti_flags |= TH_URG;
214 	} else
215 		/*
216 		 * If no urgent pointer to send, then we pull
217 		 * the urgent pointer to the left edge of the send window
218 		 * so that it doesn't drift into the send window on sequence
219 		 * number wraparound.
220 		 */
221 		tp->snd_up = tp->snd_una;		/* drag it along */
222 	/*
223 	 * If anything to send and we can send it all, set PUSH.
224 	 * (This will keep happy those implementations which only
225 	 * give data to the user when a buffer fills or a PUSH comes in.)
226 	 */
227 	if (len && off+len == so->so_snd.sb_cc)
228 		ti->ti_flags |= TH_PUSH;
229 
230 	/*
231 	 * Put TCP length in extended header, and then
232 	 * checksum extended header and data.
233 	 */
234 	if (len + optlen) {
235 		ti->ti_len = sizeof (struct tcphdr) + optlen + len;
236 		ti->ti_len = htons((u_short)ti->ti_len);
237 	}
238 	ti->ti_sum = in_cksum(m, sizeof (struct tcpiphdr) + (int)optlen + len);
239 
240 	/*
241 	 * In transmit state, time the transmission and arrange for
242 	 * the retransmit.  In persist state, reset persist time for
243 	 * next persist.
244 	 */
245 	if (tp->t_force == 0) {
246 		/*
247 		 * Advance snd_nxt over sequence space of this segment.
248 		 */
249 		if (flags & (TH_SYN|TH_FIN))
250 			tp->snd_nxt++;
251 		tp->snd_nxt += len;
252 		if (SEQ_GT(tp->snd_nxt, tp->snd_max))
253 			tp->snd_max = tp->snd_nxt;
254 
255 		/*
256 		 * Time this transmission if not a retransmission and
257 		 * not currently timing anything.
258 		 */
259 		if (SEQ_GT(tp->snd_nxt, tp->snd_max) && tp->t_rtt == 0) {
260 			tp->t_rtt = 1;
261 			tp->t_rtseq = tp->snd_nxt - len;
262 		}
263 
264 		/*
265 		 * Set retransmit timer if not currently set.
266 		 * Initial value for retransmit timer to tcp_beta*tp->t_srtt.
267 		 * Initialize shift counter which is used for exponential
268 		 * backoff of retransmit time.
269 		 */
270 		if (tp->t_timer[TCPT_REXMT] == 0 &&
271 		    tp->snd_nxt != tp->snd_una) {
272 			TCPT_RANGESET(tp->t_timer[TCPT_REXMT],
273 			    tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX);
274 			tp->t_rtt = 0;
275 			tp->t_rxtshift = 0;
276 		}
277 		tp->t_timer[TCPT_PERSIST] = 0;
278 	} else {
279 		if (SEQ_GT(tp->snd_una+1, tp->snd_max))
280 			tp->snd_max = tp->snd_una+1;
281 	}
282 
283 	/*
284 	 * Trace.
285 	 */
286 	if (so->so_options & SO_DEBUG)
287 		tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0);
288 
289 	/*
290 	 * Fill in IP length and desired time to live and
291 	 * send to IP level.
292 	 */
293 	((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + optlen + len;
294 	((struct ip *)ti)->ip_ttl = TCP_TTL;
295 	if (so->so_options & SO_DONTROUTE)
296 		error = ip_output(m, tp->t_ipopt, 0, IP_ROUTETOIF);
297 	else
298 		error = ip_output(m, tp->t_ipopt, &tp->t_inpcb->inp_route, 0);
299 	if (error)
300 		return (error);
301 
302 	/*
303 	 * Data sent (as far as we can tell).
304 	 * If this advertises a larger window than any other segment,
305 	 * then remember the size of the advertised window.
306 	 * Drop send for purpose of ACK requirements.
307 	 */
308 	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
309 		tp->rcv_adv = tp->rcv_nxt + win;
310 	tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
311 	if (sendalot && tp->t_force == 0)
312 		goto again;
313 	return (0);
314 }
315 
316 tcp_setpersist(tp)
317 	register struct tcpcb *tp;
318 {
319 
320 	if (tp->t_timer[TCPT_REXMT])
321 		panic("tcp_output REXMT");
322 	/*
323 	 * Start/restart persistance timer.
324 	 */
325 	TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
326 	    ((int)(tcp_beta * tp->t_srtt)) << tp->t_rxtshift,
327 	    TCPTV_PERSMIN, TCPTV_MAX);
328 	tp->t_rxtshift++;
329 	if (tp->t_rxtshift >= TCP_MAXRXTSHIFT)
330 		tp->t_rxtshift = 0;
331 }
332