xref: /csrg-svn/sys/netinet/tcp_subr.c (revision 44488)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)tcp_subr.c	7.18 (Berkeley) 06/28/90
8  */
9 
10 #include "param.h"
11 #include "systm.h"
12 #include "malloc.h"
13 #include "mbuf.h"
14 #include "socket.h"
15 #include "socketvar.h"
16 #include "protosw.h"
17 #include "errno.h"
18 
19 #include "../net/route.h"
20 #include "../net/if.h"
21 
22 #include "in.h"
23 #include "in_systm.h"
24 #include "ip.h"
25 #include "in_pcb.h"
26 #include "ip_var.h"
27 #include "ip_icmp.h"
28 #include "tcp.h"
29 #include "tcp_fsm.h"
30 #include "tcp_seq.h"
31 #include "tcp_timer.h"
32 #include "tcp_var.h"
33 #include "tcpip.h"
34 
35 /* patchable/settable parameters for tcp */
36 int	tcp_ttl = TCP_TTL;
37 int 	tcp_mssdflt = TCP_MSS;
38 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
39 
40 extern	struct inpcb *tcp_last_inpcb;
41 
42 /*
43  * Tcp initialization
44  */
45 tcp_init()
46 {
47 
48 	tcp_iss = 1;		/* wrong */
49 	tcb.inp_next = tcb.inp_prev = &tcb;
50 	if (max_protohdr < sizeof(struct tcpiphdr))
51 		max_protohdr = sizeof(struct tcpiphdr);
52 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
53 		panic("tcp_init");
54 }
55 
56 /*
57  * Create template to be used to send tcp packets on a connection.
58  * Call after host entry created, allocates an mbuf and fills
59  * in a skeletal tcp/ip header, minimizing the amount of work
60  * necessary when the connection is used.
61  */
62 struct tcpiphdr *
63 tcp_template(tp)
64 	struct tcpcb *tp;
65 {
66 	register struct inpcb *inp = tp->t_inpcb;
67 	register struct mbuf *m;
68 	register struct tcpiphdr *n;
69 
70 	if ((n = tp->t_template) == 0) {
71 		m = m_get(M_DONTWAIT, MT_HEADER);
72 		if (m == NULL)
73 			return (0);
74 		m->m_len = sizeof (struct tcpiphdr);
75 		n = mtod(m, struct tcpiphdr *);
76 	}
77 	n->ti_next = n->ti_prev = 0;
78 	n->ti_x1 = 0;
79 	n->ti_pr = IPPROTO_TCP;
80 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
81 	n->ti_src = inp->inp_laddr;
82 	n->ti_dst = inp->inp_faddr;
83 	n->ti_sport = inp->inp_lport;
84 	n->ti_dport = inp->inp_fport;
85 	n->ti_seq = 0;
86 	n->ti_ack = 0;
87 	n->ti_x2 = 0;
88 	n->ti_off = 5;
89 	n->ti_flags = 0;
90 	n->ti_win = 0;
91 	n->ti_sum = 0;
92 	n->ti_urp = 0;
93 	return (n);
94 }
95 
96 /*
97  * Send a single message to the TCP at address specified by
98  * the given TCP/IP header.  If m == 0, then we make a copy
99  * of the tcpiphdr at ti and send directly to the addressed host.
100  * This is used to force keep alive messages out using the TCP
101  * template for a connection tp->t_template.  If flags are given
102  * then we send a message back to the TCP which originated the
103  * segment ti, and discard the mbuf containing it and any other
104  * attached mbufs.
105  *
106  * In any case the ack and sequence number of the transmitted
107  * segment are as specified by the parameters.
108  */
109 tcp_respond(tp, ti, m, ack, seq, flags)
110 	struct tcpcb *tp;
111 	register struct tcpiphdr *ti;
112 	register struct mbuf *m;
113 	tcp_seq ack, seq;
114 	int flags;
115 {
116 	int win = 0, tlen;
117 	struct route *ro = 0;
118 
119 	if (tp) {
120 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
121 		ro = &tp->t_inpcb->inp_route;
122 	}
123 	if (m == 0) {
124 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
125 		if (m == NULL)
126 			return;
127 #ifdef TCP_COMPAT_42
128 		tlen = 1;
129 #else
130 		tlen = 0;
131 #endif
132 		m->m_data += max_linkhdr;
133 		*mtod(m, struct tcpiphdr *) = *ti;
134 		ti = mtod(m, struct tcpiphdr *);
135 		flags = TH_ACK;
136 	} else {
137 		m_freem(m->m_next);
138 		m->m_next = 0;
139 		m->m_data = (caddr_t)ti;
140 		m->m_len = sizeof (struct tcpiphdr);
141 		tlen = 0;
142 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
143 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
144 		xchg(ti->ti_dport, ti->ti_sport, u_short);
145 #undef xchg
146 	}
147 	m->m_len = sizeof (struct tcpiphdr) + tlen;
148 	ti->ti_next = ti->ti_prev = 0;
149 	ti->ti_x1 = 0;
150 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
151 	ti->ti_seq = htonl(seq);
152 	ti->ti_ack = htonl(ack);
153 	ti->ti_x2 = 0;
154 	ti->ti_off = sizeof (struct tcphdr) >> 2;
155 	ti->ti_flags = flags;
156 	ti->ti_win = htons((u_short)win);
157 	ti->ti_urp = 0;
158 	ti->ti_sum = in_cksum(m, sizeof (struct tcpiphdr) + tlen);
159 	((struct ip *)ti)->ip_len = sizeof (struct tcpiphdr) + tlen;
160 	((struct ip *)ti)->ip_ttl = tcp_ttl;
161 	(void) ip_output(m, (struct mbuf *)0, ro, 0);
162 }
163 
164 /*
165  * Create a new TCP control block, making an
166  * empty reassembly queue and hooking it to the argument
167  * protocol control block.
168  */
169 struct tcpcb *
170 tcp_newtcpcb(inp)
171 	struct inpcb *inp;
172 {
173 	struct mbuf *m = m_getclr(M_DONTWAIT, MT_PCB);
174 	register struct tcpcb *tp;
175 
176 	if (m == NULL)
177 		return ((struct tcpcb *)0);
178 	tp = mtod(m, struct tcpcb *);
179 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
180 	tp->t_maxseg = tcp_mssdflt;
181 
182 	tp->t_flags = 0;		/* sends options! */
183 	tp->t_inpcb = inp;
184 	/*
185 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
186 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
187 	 * reasonable initial retransmit time.
188 	 */
189 	tp->t_srtt = TCPTV_SRTTBASE;
190 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
191 	tp->t_rttmin = TCPTV_MIN;
192 	TCPT_RANGESET(tp->t_rxtcur,
193 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
194 	    TCPTV_MIN, TCPTV_REXMTMAX);
195 	tp->snd_cwnd = TCP_MAXWIN;
196 	tp->snd_ssthresh = TCP_MAXWIN;
197 	inp->inp_ip.ip_ttl = tcp_ttl;
198 	inp->inp_ppcb = (caddr_t)tp;
199 	return (tp);
200 }
201 
202 /*
203  * Drop a TCP connection, reporting
204  * the specified error.  If connection is synchronized,
205  * then send a RST to peer.
206  */
207 struct tcpcb *
208 tcp_drop(tp, errno)
209 	register struct tcpcb *tp;
210 	int errno;
211 {
212 	struct socket *so = tp->t_inpcb->inp_socket;
213 
214 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
215 		tp->t_state = TCPS_CLOSED;
216 		(void) tcp_output(tp);
217 		tcpstat.tcps_drops++;
218 	} else
219 		tcpstat.tcps_conndrops++;
220 	if (errno == ETIMEDOUT && tp->t_softerror)
221 		errno = tp->t_softerror;
222 	so->so_error = errno;
223 	return (tcp_close(tp));
224 }
225 
226 /*
227  * Close a TCP control block:
228  *	discard all space held by the tcp
229  *	discard internet protocol block
230  *	wake up any sleepers
231  */
232 struct tcpcb *
233 tcp_close(tp)
234 	register struct tcpcb *tp;
235 {
236 	register struct tcpiphdr *t;
237 	struct inpcb *inp = tp->t_inpcb;
238 	struct socket *so = inp->inp_socket;
239 	register struct mbuf *m;
240 #ifdef RTV_RTT
241 	register struct rtentry *rt;
242 
243 	/*
244 	 * If we sent enough data to get some meaningful characteristics,
245 	 * save them in the routing entry.  'Enough' is arbitrarily
246 	 * defined as 4K (default tcp_sendspace) * 16.  This would
247 	 * give us 16 rtt samples assuming we only get one sample per
248 	 * window (the usual case on a long haul net).  16 samples is
249 	 * enough for the srtt filter to converge to within 5% of the correct
250 	 * value; fewer samples and we could save a very bogus rtt.
251 	 *
252 	 * Don't update the default route's characteristics and don't
253 	 * update anything that the user "locked".
254 	 */
255 	if (SEQ_LT(tp->iss+(4096*16), tp->snd_max) &&
256 	    (rt = inp->inp_route.ro_rt) &&
257 	    ((struct sockaddr_in *) rt_key(rt))->sin_addr.s_addr !=
258 	    INADDR_ANY) {
259 		register u_long i;
260 
261 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
262 			i = tp->t_srtt *
263 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
264 			if (rt->rt_rmx.rmx_rtt && i)
265 				/*
266 				 * filter this update to half the old & half
267 				 * the new values, converting scale.
268 				 * See route.h and tcp_var.h for a
269 				 * description of the scaling constants.
270 				 */
271 				rt->rt_rmx.rmx_rtt =
272 				    (rt->rt_rmx.rmx_rtt + i) / 2;
273 			else
274 				rt->rt_rmx.rmx_rtt = i;
275 		}
276 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
277 			i = tp->t_rttvar *
278 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
279 			if (rt->rt_rmx.rmx_rttvar && i)
280 				rt->rt_rmx.rmx_rttvar =
281 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
282 			else
283 				rt->rt_rmx.rmx_rttvar = i;
284 		}
285 		/*
286 		 * update the pipelimit (ssthresh) if it has been updated
287 		 * already or if a pipesize was specified & the threshhold
288 		 * got below half the pipesize.  I.e., wait for bad news
289 		 * before we start updating, then update on both good
290 		 * and bad news.
291 		 */
292 		if ((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
293 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh ||
294 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
295 			/*
296 			 * convert the limit from user data bytes to
297 			 * packets then to packet data bytes.
298 			 */
299 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
300 			if (i < 2)
301 				i = 2;
302 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
303 			if (rt->rt_rmx.rmx_ssthresh)
304 				rt->rt_rmx.rmx_ssthresh =
305 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
306 			else
307 				rt->rt_rmx.rmx_ssthresh = i;
308 		}
309 	}
310 #endif RTV_RTT
311 	/* free the reassembly queue, if any */
312 	t = tp->seg_next;
313 	while (t != (struct tcpiphdr *)tp) {
314 		t = (struct tcpiphdr *)t->ti_next;
315 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
316 		remque(t->ti_prev);
317 		m_freem(m);
318 	}
319 	if (tp->t_template)
320 		(void) m_free(dtom(tp->t_template));
321 	(void) m_free(dtom(tp));
322 	inp->inp_ppcb = 0;
323 	soisdisconnected(so);
324 	/* clobber input pcb cache if we're closing the cached connection */
325 	if (inp == tcp_last_inpcb)
326 		tcp_last_inpcb = &tcb;
327 	in_pcbdetach(inp);
328 	tcpstat.tcps_closed++;
329 	return ((struct tcpcb *)0);
330 }
331 
332 tcp_drain()
333 {
334 
335 }
336 
337 /*
338  * Notify a tcp user of an asynchronous error;
339  * store error as soft error, but wake up user
340  * (for now, won't do anything until can select for soft error).
341  */
342 tcp_notify(inp, error)
343 	register struct inpcb *inp;
344 	int error;
345 {
346 
347 	((struct tcpcb *)inp->inp_ppcb)->t_softerror = error;
348 	wakeup((caddr_t) &inp->inp_socket->so_timeo);
349 	sorwakeup(inp->inp_socket);
350 	sowwakeup(inp->inp_socket);
351 }
352 
353 tcp_ctlinput(cmd, sa, ip)
354 	int cmd;
355 	struct sockaddr *sa;
356 	register struct ip *ip;
357 {
358 	register struct tcphdr *th;
359 	extern struct in_addr zeroin_addr;
360 	extern u_char inetctlerrmap[];
361 	int (*notify)() = tcp_notify, tcp_quench();
362 
363 	if (cmd == PRC_QUENCH)
364 		notify = tcp_quench;
365 	else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
366 		return;
367 	if (ip) {
368 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
369 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
370 			cmd, notify);
371 	} else
372 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
373 }
374 
375 /*
376  * When a source quench is received, close congestion window
377  * to one segment.  We will gradually open it again as we proceed.
378  */
379 tcp_quench(inp)
380 	struct inpcb *inp;
381 {
382 	struct tcpcb *tp = intotcpcb(inp);
383 
384 	if (tp)
385 		tp->snd_cwnd = tp->t_maxseg;
386 }
387