tcp_subr.c revision 1.36 1 1.36 thorpej /* $NetBSD: tcp_subr.c,v 1.36 1997/12/11 22:47:25 thorpej Exp $ */
2 1.11 cgd
3 1.1 cgd /*
4 1.10 mycroft * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 1.10 mycroft * The Regents of the University of California. All rights reserved.
6 1.1 cgd *
7 1.1 cgd * Redistribution and use in source and binary forms, with or without
8 1.1 cgd * modification, are permitted provided that the following conditions
9 1.1 cgd * are met:
10 1.1 cgd * 1. Redistributions of source code must retain the above copyright
11 1.1 cgd * notice, this list of conditions and the following disclaimer.
12 1.1 cgd * 2. Redistributions in binary form must reproduce the above copyright
13 1.1 cgd * notice, this list of conditions and the following disclaimer in the
14 1.1 cgd * documentation and/or other materials provided with the distribution.
15 1.1 cgd * 3. All advertising materials mentioning features or use of this software
16 1.1 cgd * must display the following acknowledgement:
17 1.1 cgd * This product includes software developed by the University of
18 1.1 cgd * California, Berkeley and its contributors.
19 1.1 cgd * 4. Neither the name of the University nor the names of its contributors
20 1.1 cgd * may be used to endorse or promote products derived from this software
21 1.1 cgd * without specific prior written permission.
22 1.1 cgd *
23 1.1 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 1.1 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 1.1 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 1.1 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 1.1 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 1.1 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 1.1 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 1.1 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 1.1 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 1.1 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 1.1 cgd * SUCH DAMAGE.
34 1.1 cgd *
35 1.11 cgd * @(#)tcp_subr.c 8.1 (Berkeley) 6/10/93
36 1.1 cgd */
37 1.1 cgd
38 1.30 explorer #include "rnd.h"
39 1.30 explorer
40 1.5 mycroft #include <sys/param.h>
41 1.10 mycroft #include <sys/proc.h>
42 1.5 mycroft #include <sys/systm.h>
43 1.5 mycroft #include <sys/malloc.h>
44 1.5 mycroft #include <sys/mbuf.h>
45 1.5 mycroft #include <sys/socket.h>
46 1.5 mycroft #include <sys/socketvar.h>
47 1.5 mycroft #include <sys/protosw.h>
48 1.5 mycroft #include <sys/errno.h>
49 1.27 thorpej #include <sys/kernel.h>
50 1.30 explorer #if NRND > 0
51 1.29 explorer #include <sys/rnd.h>
52 1.30 explorer #endif
53 1.1 cgd
54 1.5 mycroft #include <net/route.h>
55 1.5 mycroft #include <net/if.h>
56 1.1 cgd
57 1.5 mycroft #include <netinet/in.h>
58 1.5 mycroft #include <netinet/in_systm.h>
59 1.5 mycroft #include <netinet/ip.h>
60 1.5 mycroft #include <netinet/in_pcb.h>
61 1.5 mycroft #include <netinet/ip_var.h>
62 1.5 mycroft #include <netinet/ip_icmp.h>
63 1.5 mycroft #include <netinet/tcp.h>
64 1.5 mycroft #include <netinet/tcp_fsm.h>
65 1.5 mycroft #include <netinet/tcp_seq.h>
66 1.5 mycroft #include <netinet/tcp_timer.h>
67 1.5 mycroft #include <netinet/tcp_var.h>
68 1.5 mycroft #include <netinet/tcpip.h>
69 1.1 cgd
70 1.1 cgd /* patchable/settable parameters for tcp */
71 1.1 cgd int tcp_mssdflt = TCP_MSS;
72 1.1 cgd int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
73 1.10 mycroft int tcp_do_rfc1323 = 1;
74 1.36 thorpej int tcp_init_win = 1;
75 1.1 cgd
76 1.21 mycroft #ifndef TCBHASHSIZE
77 1.21 mycroft #define TCBHASHSIZE 128
78 1.21 mycroft #endif
79 1.21 mycroft int tcbhashsize = TCBHASHSIZE;
80 1.1 cgd
81 1.35 thorpej int tcp_freeq __P((struct tcpcb *));
82 1.35 thorpej
83 1.1 cgd /*
84 1.1 cgd * Tcp initialization
85 1.1 cgd */
86 1.7 mycroft void
87 1.1 cgd tcp_init()
88 1.1 cgd {
89 1.1 cgd
90 1.24 mycroft in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
91 1.1 cgd if (max_protohdr < sizeof(struct tcpiphdr))
92 1.1 cgd max_protohdr = sizeof(struct tcpiphdr);
93 1.1 cgd if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
94 1.1 cgd panic("tcp_init");
95 1.1 cgd }
96 1.1 cgd
97 1.1 cgd /*
98 1.1 cgd * Create template to be used to send tcp packets on a connection.
99 1.1 cgd * Call after host entry created, allocates an mbuf and fills
100 1.1 cgd * in a skeletal tcp/ip header, minimizing the amount of work
101 1.1 cgd * necessary when the connection is used.
102 1.1 cgd */
103 1.1 cgd struct tcpiphdr *
104 1.1 cgd tcp_template(tp)
105 1.1 cgd struct tcpcb *tp;
106 1.1 cgd {
107 1.1 cgd register struct inpcb *inp = tp->t_inpcb;
108 1.1 cgd register struct tcpiphdr *n;
109 1.1 cgd
110 1.1 cgd if ((n = tp->t_template) == 0) {
111 1.26 thorpej MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
112 1.26 thorpej M_MBUF, M_NOWAIT);
113 1.26 thorpej if (n == NULL)
114 1.1 cgd return (0);
115 1.1 cgd }
116 1.20 cgd bzero(n->ti_x1, sizeof n->ti_x1);
117 1.1 cgd n->ti_pr = IPPROTO_TCP;
118 1.1 cgd n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
119 1.1 cgd n->ti_src = inp->inp_laddr;
120 1.1 cgd n->ti_dst = inp->inp_faddr;
121 1.1 cgd n->ti_sport = inp->inp_lport;
122 1.1 cgd n->ti_dport = inp->inp_fport;
123 1.1 cgd n->ti_seq = 0;
124 1.1 cgd n->ti_ack = 0;
125 1.1 cgd n->ti_x2 = 0;
126 1.1 cgd n->ti_off = 5;
127 1.1 cgd n->ti_flags = 0;
128 1.1 cgd n->ti_win = 0;
129 1.1 cgd n->ti_sum = 0;
130 1.1 cgd n->ti_urp = 0;
131 1.1 cgd return (n);
132 1.1 cgd }
133 1.1 cgd
134 1.1 cgd /*
135 1.1 cgd * Send a single message to the TCP at address specified by
136 1.1 cgd * the given TCP/IP header. If m == 0, then we make a copy
137 1.1 cgd * of the tcpiphdr at ti and send directly to the addressed host.
138 1.1 cgd * This is used to force keep alive messages out using the TCP
139 1.1 cgd * template for a connection tp->t_template. If flags are given
140 1.1 cgd * then we send a message back to the TCP which originated the
141 1.1 cgd * segment ti, and discard the mbuf containing it and any other
142 1.1 cgd * attached mbufs.
143 1.1 cgd *
144 1.1 cgd * In any case the ack and sequence number of the transmitted
145 1.1 cgd * segment are as specified by the parameters.
146 1.1 cgd */
147 1.27 thorpej int
148 1.1 cgd tcp_respond(tp, ti, m, ack, seq, flags)
149 1.1 cgd struct tcpcb *tp;
150 1.1 cgd register struct tcpiphdr *ti;
151 1.1 cgd register struct mbuf *m;
152 1.1 cgd tcp_seq ack, seq;
153 1.1 cgd int flags;
154 1.1 cgd {
155 1.1 cgd register int tlen;
156 1.1 cgd int win = 0;
157 1.1 cgd struct route *ro = 0;
158 1.1 cgd
159 1.1 cgd if (tp) {
160 1.1 cgd win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
161 1.1 cgd ro = &tp->t_inpcb->inp_route;
162 1.1 cgd }
163 1.1 cgd if (m == 0) {
164 1.1 cgd m = m_gethdr(M_DONTWAIT, MT_HEADER);
165 1.1 cgd if (m == NULL)
166 1.27 thorpej return (ENOBUFS);
167 1.1 cgd #ifdef TCP_COMPAT_42
168 1.1 cgd tlen = 1;
169 1.1 cgd #else
170 1.1 cgd tlen = 0;
171 1.1 cgd #endif
172 1.1 cgd m->m_data += max_linkhdr;
173 1.1 cgd *mtod(m, struct tcpiphdr *) = *ti;
174 1.1 cgd ti = mtod(m, struct tcpiphdr *);
175 1.1 cgd flags = TH_ACK;
176 1.1 cgd } else {
177 1.1 cgd m_freem(m->m_next);
178 1.1 cgd m->m_next = 0;
179 1.1 cgd m->m_data = (caddr_t)ti;
180 1.1 cgd m->m_len = sizeof (struct tcpiphdr);
181 1.1 cgd tlen = 0;
182 1.10 mycroft #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
183 1.13 cgd xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
184 1.13 cgd xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
185 1.1 cgd #undef xchg
186 1.1 cgd }
187 1.20 cgd bzero(ti->ti_x1, sizeof ti->ti_x1);
188 1.1 cgd ti->ti_seq = htonl(seq);
189 1.1 cgd ti->ti_ack = htonl(ack);
190 1.1 cgd ti->ti_x2 = 0;
191 1.27 thorpej if ((flags & TH_SYN) == 0) {
192 1.27 thorpej if (tp)
193 1.27 thorpej ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
194 1.27 thorpej else
195 1.27 thorpej ti->ti_win = htons((u_int16_t)win);
196 1.27 thorpej ti->ti_off = sizeof (struct tcphdr) >> 2;
197 1.27 thorpej tlen += sizeof (struct tcphdr);
198 1.27 thorpej } else
199 1.27 thorpej tlen += ti->ti_off << 2;
200 1.27 thorpej ti->ti_len = htons((u_int16_t)tlen);
201 1.27 thorpej tlen += sizeof (struct ip);
202 1.27 thorpej m->m_len = tlen;
203 1.27 thorpej m->m_pkthdr.len = tlen;
204 1.27 thorpej m->m_pkthdr.rcvif = (struct ifnet *) 0;
205 1.1 cgd ti->ti_flags = flags;
206 1.1 cgd ti->ti_urp = 0;
207 1.10 mycroft ti->ti_sum = 0;
208 1.1 cgd ti->ti_sum = in_cksum(m, tlen);
209 1.1 cgd ((struct ip *)ti)->ip_len = tlen;
210 1.10 mycroft ((struct ip *)ti)->ip_ttl = ip_defttl;
211 1.27 thorpej return ip_output(m, NULL, ro, 0, NULL);
212 1.1 cgd }
213 1.1 cgd
214 1.1 cgd /*
215 1.1 cgd * Create a new TCP control block, making an
216 1.1 cgd * empty reassembly queue and hooking it to the argument
217 1.1 cgd * protocol control block.
218 1.1 cgd */
219 1.1 cgd struct tcpcb *
220 1.1 cgd tcp_newtcpcb(inp)
221 1.1 cgd struct inpcb *inp;
222 1.1 cgd {
223 1.1 cgd register struct tcpcb *tp;
224 1.1 cgd
225 1.10 mycroft tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
226 1.10 mycroft if (tp == NULL)
227 1.1 cgd return ((struct tcpcb *)0);
228 1.23 mycroft bzero((caddr_t)tp, sizeof(struct tcpcb));
229 1.20 cgd LIST_INIT(&tp->segq);
230 1.33 kml tp->t_peermss = tcp_mssdflt;
231 1.28 thorpej tp->t_ourmss = tcp_mssdflt;
232 1.33 kml tp->t_segsz = tcp_mssdflt;
233 1.1 cgd
234 1.10 mycroft tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
235 1.1 cgd tp->t_inpcb = inp;
236 1.1 cgd /*
237 1.1 cgd * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
238 1.1 cgd * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives
239 1.1 cgd * reasonable initial retransmit time.
240 1.1 cgd */
241 1.1 cgd tp->t_srtt = TCPTV_SRTTBASE;
242 1.15 mycroft tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
243 1.1 cgd tp->t_rttmin = TCPTV_MIN;
244 1.15 mycroft TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
245 1.1 cgd TCPTV_MIN, TCPTV_REXMTMAX);
246 1.10 mycroft tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
247 1.10 mycroft tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
248 1.10 mycroft inp->inp_ip.ip_ttl = ip_defttl;
249 1.1 cgd inp->inp_ppcb = (caddr_t)tp;
250 1.1 cgd return (tp);
251 1.1 cgd }
252 1.1 cgd
253 1.1 cgd /*
254 1.1 cgd * Drop a TCP connection, reporting
255 1.1 cgd * the specified error. If connection is synchronized,
256 1.1 cgd * then send a RST to peer.
257 1.1 cgd */
258 1.1 cgd struct tcpcb *
259 1.1 cgd tcp_drop(tp, errno)
260 1.1 cgd register struct tcpcb *tp;
261 1.1 cgd int errno;
262 1.1 cgd {
263 1.1 cgd struct socket *so = tp->t_inpcb->inp_socket;
264 1.1 cgd
265 1.1 cgd if (TCPS_HAVERCVDSYN(tp->t_state)) {
266 1.1 cgd tp->t_state = TCPS_CLOSED;
267 1.1 cgd (void) tcp_output(tp);
268 1.1 cgd tcpstat.tcps_drops++;
269 1.1 cgd } else
270 1.1 cgd tcpstat.tcps_conndrops++;
271 1.1 cgd if (errno == ETIMEDOUT && tp->t_softerror)
272 1.1 cgd errno = tp->t_softerror;
273 1.1 cgd so->so_error = errno;
274 1.1 cgd return (tcp_close(tp));
275 1.1 cgd }
276 1.1 cgd
277 1.1 cgd /*
278 1.1 cgd * Close a TCP control block:
279 1.1 cgd * discard all space held by the tcp
280 1.1 cgd * discard internet protocol block
281 1.1 cgd * wake up any sleepers
282 1.1 cgd */
283 1.1 cgd struct tcpcb *
284 1.1 cgd tcp_close(tp)
285 1.1 cgd register struct tcpcb *tp;
286 1.1 cgd {
287 1.1 cgd struct inpcb *inp = tp->t_inpcb;
288 1.1 cgd struct socket *so = inp->inp_socket;
289 1.1 cgd #ifdef RTV_RTT
290 1.1 cgd register struct rtentry *rt;
291 1.1 cgd
292 1.1 cgd /*
293 1.1 cgd * If we sent enough data to get some meaningful characteristics,
294 1.10 mycroft * save them in the routing entry. 'Enough' is arbitrarily
295 1.1 cgd * defined as the sendpipesize (default 4K) * 16. This would
296 1.1 cgd * give us 16 rtt samples assuming we only get one sample per
297 1.1 cgd * window (the usual case on a long haul net). 16 samples is
298 1.1 cgd * enough for the srtt filter to converge to within 5% of the correct
299 1.1 cgd * value; fewer samples and we could save a very bogus rtt.
300 1.1 cgd *
301 1.1 cgd * Don't update the default route's characteristics and don't
302 1.1 cgd * update anything that the user "locked".
303 1.1 cgd */
304 1.1 cgd if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
305 1.1 cgd (rt = inp->inp_route.ro_rt) &&
306 1.23 mycroft !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
307 1.22 christos register u_long i = 0;
308 1.1 cgd
309 1.1 cgd if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
310 1.1 cgd i = tp->t_srtt *
311 1.25 mycroft ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
312 1.1 cgd if (rt->rt_rmx.rmx_rtt && i)
313 1.1 cgd /*
314 1.1 cgd * filter this update to half the old & half
315 1.1 cgd * the new values, converting scale.
316 1.1 cgd * See route.h and tcp_var.h for a
317 1.1 cgd * description of the scaling constants.
318 1.1 cgd */
319 1.1 cgd rt->rt_rmx.rmx_rtt =
320 1.1 cgd (rt->rt_rmx.rmx_rtt + i) / 2;
321 1.1 cgd else
322 1.1 cgd rt->rt_rmx.rmx_rtt = i;
323 1.1 cgd }
324 1.1 cgd if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
325 1.1 cgd i = tp->t_rttvar *
326 1.25 mycroft ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
327 1.1 cgd if (rt->rt_rmx.rmx_rttvar && i)
328 1.1 cgd rt->rt_rmx.rmx_rttvar =
329 1.1 cgd (rt->rt_rmx.rmx_rttvar + i) / 2;
330 1.1 cgd else
331 1.1 cgd rt->rt_rmx.rmx_rttvar = i;
332 1.1 cgd }
333 1.1 cgd /*
334 1.1 cgd * update the pipelimit (ssthresh) if it has been updated
335 1.1 cgd * already or if a pipesize was specified & the threshhold
336 1.1 cgd * got below half the pipesize. I.e., wait for bad news
337 1.1 cgd * before we start updating, then update on both good
338 1.1 cgd * and bad news.
339 1.1 cgd */
340 1.22 christos if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
341 1.22 christos (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
342 1.1 cgd i < (rt->rt_rmx.rmx_sendpipe / 2)) {
343 1.1 cgd /*
344 1.1 cgd * convert the limit from user data bytes to
345 1.1 cgd * packets then to packet data bytes.
346 1.1 cgd */
347 1.33 kml i = (i + tp->t_segsz / 2) / tp->t_segsz;
348 1.1 cgd if (i < 2)
349 1.1 cgd i = 2;
350 1.33 kml i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
351 1.1 cgd if (rt->rt_rmx.rmx_ssthresh)
352 1.1 cgd rt->rt_rmx.rmx_ssthresh =
353 1.1 cgd (rt->rt_rmx.rmx_ssthresh + i) / 2;
354 1.1 cgd else
355 1.1 cgd rt->rt_rmx.rmx_ssthresh = i;
356 1.1 cgd }
357 1.1 cgd }
358 1.9 mycroft #endif /* RTV_RTT */
359 1.1 cgd /* free the reassembly queue, if any */
360 1.35 thorpej (void) tcp_freeq(tp);
361 1.35 thorpej
362 1.1 cgd if (tp->t_template)
363 1.26 thorpej FREE(tp->t_template, M_MBUF);
364 1.10 mycroft free(tp, M_PCB);
365 1.1 cgd inp->inp_ppcb = 0;
366 1.1 cgd soisdisconnected(so);
367 1.1 cgd in_pcbdetach(inp);
368 1.1 cgd tcpstat.tcps_closed++;
369 1.1 cgd return ((struct tcpcb *)0);
370 1.1 cgd }
371 1.1 cgd
372 1.35 thorpej int
373 1.35 thorpej tcp_freeq(tp)
374 1.35 thorpej struct tcpcb *tp;
375 1.35 thorpej {
376 1.35 thorpej register struct ipqent *qe;
377 1.35 thorpej int rv = 0;
378 1.35 thorpej
379 1.35 thorpej while ((qe = tp->segq.lh_first) != NULL) {
380 1.35 thorpej LIST_REMOVE(qe, ipqe_q);
381 1.35 thorpej m_freem(qe->ipqe_m);
382 1.35 thorpej FREE(qe, M_IPQ);
383 1.35 thorpej rv = 1;
384 1.35 thorpej }
385 1.35 thorpej return (rv);
386 1.35 thorpej }
387 1.35 thorpej
388 1.35 thorpej /*
389 1.35 thorpej * Protocol drain routine. Called when memory is in short supply.
390 1.35 thorpej */
391 1.7 mycroft void
392 1.1 cgd tcp_drain()
393 1.1 cgd {
394 1.35 thorpej register struct inpcb *inp;
395 1.35 thorpej register struct tcpcb *tp;
396 1.1 cgd
397 1.35 thorpej /*
398 1.35 thorpej * Free the sequence queue of all TCP connections.
399 1.35 thorpej */
400 1.35 thorpej inp = tcbtable.inpt_queue.cqh_first;
401 1.35 thorpej if (inp) /* XXX */
402 1.35 thorpej for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
403 1.35 thorpej inp = inp->inp_queue.cqe_next) {
404 1.35 thorpej if ((tp = intotcpcb(inp)) != NULL) {
405 1.35 thorpej if (tcp_freeq(tp))
406 1.35 thorpej tcpstat.tcps_connsdrained++;
407 1.35 thorpej }
408 1.35 thorpej }
409 1.1 cgd }
410 1.1 cgd
411 1.1 cgd /*
412 1.1 cgd * Notify a tcp user of an asynchronous error;
413 1.1 cgd * store error as soft error, but wake up user
414 1.1 cgd * (for now, won't do anything until can select for soft error).
415 1.1 cgd */
416 1.7 mycroft void
417 1.1 cgd tcp_notify(inp, error)
418 1.10 mycroft struct inpcb *inp;
419 1.1 cgd int error;
420 1.1 cgd {
421 1.10 mycroft register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
422 1.10 mycroft register struct socket *so = inp->inp_socket;
423 1.1 cgd
424 1.10 mycroft /*
425 1.10 mycroft * Ignore some errors if we are hooked up.
426 1.10 mycroft * If connection hasn't completed, has retransmitted several times,
427 1.10 mycroft * and receives a second error, give up now. This is better
428 1.10 mycroft * than waiting a long time to establish a connection that
429 1.10 mycroft * can never complete.
430 1.10 mycroft */
431 1.10 mycroft if (tp->t_state == TCPS_ESTABLISHED &&
432 1.10 mycroft (error == EHOSTUNREACH || error == ENETUNREACH ||
433 1.10 mycroft error == EHOSTDOWN)) {
434 1.10 mycroft return;
435 1.12 mycroft } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
436 1.12 mycroft tp->t_rxtshift > 3 && tp->t_softerror)
437 1.10 mycroft so->so_error = error;
438 1.10 mycroft else
439 1.10 mycroft tp->t_softerror = error;
440 1.10 mycroft wakeup((caddr_t) &so->so_timeo);
441 1.10 mycroft sorwakeup(so);
442 1.10 mycroft sowwakeup(so);
443 1.1 cgd }
444 1.1 cgd
445 1.22 christos void *
446 1.22 christos tcp_ctlinput(cmd, sa, v)
447 1.1 cgd int cmd;
448 1.1 cgd struct sockaddr *sa;
449 1.22 christos register void *v;
450 1.1 cgd {
451 1.22 christos register struct ip *ip = v;
452 1.1 cgd register struct tcphdr *th;
453 1.19 mycroft extern int inetctlerrmap[];
454 1.7 mycroft void (*notify) __P((struct inpcb *, int)) = tcp_notify;
455 1.19 mycroft int errno;
456 1.27 thorpej int nmatch;
457 1.1 cgd
458 1.18 mycroft if ((unsigned)cmd >= PRC_NCMDS)
459 1.22 christos return NULL;
460 1.18 mycroft errno = inetctlerrmap[cmd];
461 1.17 mycroft if (cmd == PRC_QUENCH)
462 1.17 mycroft notify = tcp_quench;
463 1.17 mycroft else if (PRC_IS_REDIRECT(cmd))
464 1.17 mycroft notify = in_rtchange, ip = 0;
465 1.32 kml else if (cmd == PRC_MSGSIZE && ip_mtudisc)
466 1.31 kml notify = tcp_mtudisc, ip = 0;
467 1.16 mycroft else if (cmd == PRC_HOSTDEAD)
468 1.17 mycroft ip = 0;
469 1.18 mycroft else if (errno == 0)
470 1.22 christos return NULL;
471 1.17 mycroft if (ip) {
472 1.1 cgd th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
473 1.27 thorpej nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
474 1.27 thorpej th->th_dport, ip->ip_src, th->th_sport, errno, notify);
475 1.27 thorpej if (nmatch == 0 && syn_cache_count &&
476 1.27 thorpej (inetctlerrmap[cmd] == EHOSTUNREACH ||
477 1.27 thorpej inetctlerrmap[cmd] == ENETUNREACH ||
478 1.27 thorpej inetctlerrmap[cmd] == EHOSTDOWN))
479 1.27 thorpej syn_cache_unreach(ip, th);
480 1.17 mycroft } else
481 1.27 thorpej (void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
482 1.23 mycroft notify);
483 1.22 christos return NULL;
484 1.1 cgd }
485 1.1 cgd
486 1.1 cgd /*
487 1.1 cgd * When a source quench is received, close congestion window
488 1.1 cgd * to one segment. We will gradually open it again as we proceed.
489 1.1 cgd */
490 1.7 mycroft void
491 1.7 mycroft tcp_quench(inp, errno)
492 1.1 cgd struct inpcb *inp;
493 1.7 mycroft int errno;
494 1.1 cgd {
495 1.1 cgd struct tcpcb *tp = intotcpcb(inp);
496 1.1 cgd
497 1.1 cgd if (tp)
498 1.33 kml tp->snd_cwnd = tp->t_segsz;
499 1.28 thorpej }
500 1.31 kml
501 1.31 kml /*
502 1.31 kml * On receipt of path MTU corrections, flush old route and replace it
503 1.31 kml * with the new one. Retransmit all unacknowledged packets, to ensure
504 1.31 kml * that all packets will be received.
505 1.31 kml */
506 1.31 kml void
507 1.31 kml tcp_mtudisc(inp, errno)
508 1.31 kml struct inpcb *inp;
509 1.31 kml int errno;
510 1.31 kml {
511 1.31 kml struct tcpcb *tp = intotcpcb(inp);
512 1.31 kml struct rtentry *rt = in_pcbrtentry(inp);
513 1.31 kml
514 1.31 kml if (tp != 0) {
515 1.31 kml if (rt != 0) {
516 1.36 thorpej /*
517 1.36 thorpej * If this was not a host route, remove and realloc.
518 1.36 thorpej */
519 1.31 kml if ((rt->rt_flags & RTF_HOST) == 0) {
520 1.31 kml in_rtchange(inp, errno);
521 1.31 kml if ((rt = in_pcbrtentry(inp)) == 0)
522 1.31 kml return;
523 1.31 kml }
524 1.36 thorpej
525 1.36 thorpej /*
526 1.36 thorpej * Slow start out of the error condition. We
527 1.36 thorpej * use the MTU because we know it's smaller
528 1.36 thorpej * than the previously transmitted segment.
529 1.36 thorpej */
530 1.33 kml if (rt->rt_rmx.rmx_mtu != 0)
531 1.36 thorpej tp->snd_cwnd =
532 1.36 thorpej TCP_INITIAL_WINDOW(rt->rt_rmx.rmx_mtu);
533 1.31 kml }
534 1.31 kml
535 1.36 thorpej /*
536 1.36 thorpej * Resend unacknowledged packets.
537 1.36 thorpej */
538 1.31 kml tp->snd_nxt = tp->snd_una;
539 1.31 kml tcp_output(tp);
540 1.31 kml }
541 1.31 kml }
542 1.31 kml
543 1.28 thorpej
544 1.28 thorpej /*
545 1.28 thorpej * Compute the MSS to advertise to the peer. Called only during
546 1.28 thorpej * the 3-way handshake. If we are the server (peer initiated
547 1.28 thorpej * connection), we are called with the TCPCB for the listen
548 1.28 thorpej * socket. If we are the client (we initiated connection), we
549 1.28 thorpej * are called witht he TCPCB for the actual connection.
550 1.28 thorpej */
551 1.28 thorpej int
552 1.28 thorpej tcp_mss_to_advertise(tp)
553 1.28 thorpej const struct tcpcb *tp;
554 1.28 thorpej {
555 1.28 thorpej extern u_long in_maxmtu;
556 1.28 thorpej struct inpcb *inp;
557 1.28 thorpej struct socket *so;
558 1.28 thorpej int mss;
559 1.28 thorpej
560 1.28 thorpej inp = tp->t_inpcb;
561 1.28 thorpej so = inp->inp_socket;
562 1.28 thorpej
563 1.28 thorpej /*
564 1.28 thorpej * In order to avoid defeating path MTU discovery on the peer,
565 1.28 thorpej * we advertise the max MTU of all attached networks as our MSS,
566 1.28 thorpej * per RFC 1191, section 3.1.
567 1.28 thorpej *
568 1.28 thorpej * XXX Should we allow room for the timestamp option if
569 1.28 thorpej * XXX rfc1323 is enabled?
570 1.28 thorpej */
571 1.28 thorpej mss = in_maxmtu - sizeof(struct tcpiphdr);
572 1.28 thorpej
573 1.28 thorpej return (mss);
574 1.28 thorpej }
575 1.28 thorpej
576 1.28 thorpej /*
577 1.28 thorpej * Set connection variables based on the peer's advertised MSS.
578 1.28 thorpej * We are passed the TCPCB for the actual connection. If we
579 1.28 thorpej * are the server, we are called by the compressed state engine
580 1.28 thorpej * when the 3-way handshake is complete. If we are the client,
581 1.28 thorpej * we are called when we recieve the SYN,ACK from the server.
582 1.28 thorpej *
583 1.28 thorpej * NOTE: Our advertised MSS value must be initialized in the TCPCB
584 1.28 thorpej * before this routine is called!
585 1.28 thorpej */
586 1.28 thorpej void
587 1.28 thorpej tcp_mss_from_peer(tp, offer)
588 1.28 thorpej struct tcpcb *tp;
589 1.28 thorpej int offer;
590 1.28 thorpej {
591 1.28 thorpej struct inpcb *inp = tp->t_inpcb;
592 1.28 thorpej struct socket *so = inp->inp_socket;
593 1.28 thorpej #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
594 1.28 thorpej struct rtentry *rt = in_pcbrtentry(inp);
595 1.28 thorpej #endif
596 1.28 thorpej u_long bufsize;
597 1.28 thorpej int mss;
598 1.28 thorpej
599 1.28 thorpej /*
600 1.28 thorpej * Assume our MSS is the MSS of the peer, unless they sent us
601 1.28 thorpej * an offer. Do not accept offers less than 32 bytes.
602 1.28 thorpej */
603 1.28 thorpej mss = tp->t_ourmss;
604 1.28 thorpej if (offer)
605 1.28 thorpej mss = offer;
606 1.28 thorpej mss = max(mss, 32); /* sanity */
607 1.28 thorpej
608 1.28 thorpej /*
609 1.28 thorpej * If there's a pipesize, change the socket buffer to that size.
610 1.28 thorpej * Make the socket buffer an integral number of MSS units. If
611 1.28 thorpej * the MSS is larger than the socket buffer, artificially decrease
612 1.28 thorpej * the MSS.
613 1.28 thorpej */
614 1.28 thorpej #ifdef RTV_SPIPE
615 1.28 thorpej if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
616 1.28 thorpej bufsize = rt->rt_rmx.rmx_sendpipe;
617 1.28 thorpej else
618 1.28 thorpej #endif
619 1.28 thorpej bufsize = so->so_snd.sb_hiwat;
620 1.28 thorpej if (bufsize < mss)
621 1.28 thorpej mss = bufsize;
622 1.28 thorpej else {
623 1.28 thorpej bufsize = roundup(bufsize, mss);
624 1.28 thorpej if (bufsize > sb_max)
625 1.28 thorpej bufsize = sb_max;
626 1.28 thorpej (void) sbreserve(&so->so_snd, bufsize);
627 1.28 thorpej }
628 1.33 kml tp->t_peermss = mss;
629 1.33 kml tp->t_segsz = mss;
630 1.28 thorpej
631 1.28 thorpej /* Initialize the initial congestion window. */
632 1.36 thorpej tp->snd_cwnd = TCP_INITIAL_WINDOW(mss);
633 1.28 thorpej
634 1.28 thorpej #ifdef RTV_SSTHRESH
635 1.28 thorpej if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
636 1.28 thorpej /*
637 1.28 thorpej * There's some sort of gateway or interface buffer
638 1.28 thorpej * limit on the path. Use this to set the slow
639 1.28 thorpej * start threshold, but set the threshold to no less
640 1.28 thorpej * than 2 * MSS.
641 1.28 thorpej */
642 1.28 thorpej tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
643 1.28 thorpej }
644 1.28 thorpej #endif
645 1.28 thorpej }
646 1.28 thorpej
647 1.28 thorpej /*
648 1.28 thorpej * Processing necessary when a TCP connection is established.
649 1.28 thorpej */
650 1.28 thorpej void
651 1.28 thorpej tcp_established(tp)
652 1.28 thorpej struct tcpcb *tp;
653 1.28 thorpej {
654 1.28 thorpej struct inpcb *inp = tp->t_inpcb;
655 1.28 thorpej struct socket *so = inp->inp_socket;
656 1.28 thorpej #ifdef RTV_RPIPE
657 1.28 thorpej struct rtentry *rt = in_pcbrtentry(inp);
658 1.28 thorpej #endif
659 1.28 thorpej u_long bufsize;
660 1.28 thorpej
661 1.28 thorpej tp->t_state = TCPS_ESTABLISHED;
662 1.28 thorpej tp->t_timer[TCPT_KEEP] = tcp_keepidle;
663 1.28 thorpej
664 1.28 thorpej #ifdef RTV_RPIPE
665 1.28 thorpej if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
666 1.28 thorpej bufsize = rt->rt_rmx.rmx_recvpipe;
667 1.28 thorpej else
668 1.28 thorpej #endif
669 1.28 thorpej bufsize = so->so_rcv.sb_hiwat;
670 1.28 thorpej if (bufsize > tp->t_ourmss) {
671 1.28 thorpej bufsize = roundup(bufsize, tp->t_ourmss);
672 1.28 thorpej if (bufsize > sb_max)
673 1.28 thorpej bufsize = sb_max;
674 1.28 thorpej (void) sbreserve(&so->so_rcv, bufsize);
675 1.28 thorpej }
676 1.28 thorpej }
677 1.28 thorpej
678 1.28 thorpej /*
679 1.28 thorpej * Check if there's an initial rtt or rttvar. Convert from the
680 1.28 thorpej * route-table units to scaled multiples of the slow timeout timer.
681 1.28 thorpej * Called only during the 3-way handshake.
682 1.28 thorpej */
683 1.28 thorpej void
684 1.28 thorpej tcp_rmx_rtt(tp)
685 1.28 thorpej struct tcpcb *tp;
686 1.28 thorpej {
687 1.28 thorpej #ifdef RTV_RTT
688 1.28 thorpej struct rtentry *rt;
689 1.28 thorpej int rtt;
690 1.28 thorpej
691 1.28 thorpej if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
692 1.28 thorpej return;
693 1.28 thorpej
694 1.28 thorpej if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
695 1.28 thorpej /*
696 1.28 thorpej * XXX The lock bit for MTU indicates that the value
697 1.28 thorpej * is also a minimum value; this is subject to time.
698 1.28 thorpej */
699 1.28 thorpej if (rt->rt_rmx.rmx_locks & RTV_RTT)
700 1.28 thorpej tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
701 1.28 thorpej tp->t_srtt = rtt /
702 1.28 thorpej ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
703 1.28 thorpej if (rt->rt_rmx.rmx_rttvar) {
704 1.28 thorpej tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
705 1.28 thorpej ((RTM_RTTUNIT / PR_SLOWHZ) >>
706 1.28 thorpej (TCP_RTTVAR_SHIFT + 2));
707 1.28 thorpej } else {
708 1.28 thorpej /* Default variation is +- 1 rtt */
709 1.28 thorpej tp->t_rttvar =
710 1.28 thorpej tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
711 1.28 thorpej }
712 1.28 thorpej TCPT_RANGESET(tp->t_rxtcur,
713 1.28 thorpej ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
714 1.28 thorpej tp->t_rttmin, TCPTV_REXMTMAX);
715 1.28 thorpej }
716 1.28 thorpej #endif
717 1.29 explorer }
718 1.29 explorer
719 1.30 explorer tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */
720 1.30 explorer
721 1.29 explorer /*
722 1.29 explorer * Get a new sequence value given a tcp control block
723 1.29 explorer */
724 1.29 explorer tcp_seq
725 1.29 explorer tcp_new_iss(tp, len, addin)
726 1.29 explorer void *tp;
727 1.29 explorer u_long len;
728 1.29 explorer tcp_seq addin;
729 1.29 explorer {
730 1.29 explorer tcp_seq tcp_iss;
731 1.29 explorer
732 1.29 explorer /*
733 1.29 explorer * add randomness about this connection, but do not estimate
734 1.29 explorer * entropy from the timing, since the physical device driver would
735 1.29 explorer * have done that for us.
736 1.29 explorer */
737 1.30 explorer #if NRND > 0
738 1.29 explorer if (tp != NULL)
739 1.29 explorer rnd_add_data(NULL, tp, len, 0);
740 1.30 explorer #endif
741 1.29 explorer
742 1.29 explorer /*
743 1.29 explorer * randomize.
744 1.29 explorer */
745 1.30 explorer #if NRND > 0
746 1.29 explorer rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
747 1.30 explorer #else
748 1.30 explorer tcp_iss = random();
749 1.30 explorer #endif
750 1.29 explorer
751 1.29 explorer /*
752 1.29 explorer * If we were asked to add some amount to a known value,
753 1.29 explorer * we will take a random value obtained above, mask off the upper
754 1.29 explorer * bits, and add in the known value. We also add in a constant to
755 1.29 explorer * ensure that we are at least a certain distance from the original
756 1.29 explorer * value.
757 1.29 explorer *
758 1.29 explorer * This is used when an old connection is in timed wait
759 1.29 explorer * and we have a new one coming in, for instance.
760 1.29 explorer */
761 1.29 explorer if (addin != 0) {
762 1.29 explorer #ifdef TCPISS_DEBUG
763 1.29 explorer printf("Random %08x, ", tcp_iss);
764 1.29 explorer #endif
765 1.29 explorer tcp_iss &= TCP_ISS_RANDOM_MASK;
766 1.30 explorer tcp_iss = tcp_iss + addin + TCP_ISSINCR;
767 1.30 explorer tcp_iss_seq += TCP_ISSINCR;
768 1.30 explorer tcp_iss += tcp_iss_seq;
769 1.29 explorer #ifdef TCPISS_DEBUG
770 1.29 explorer printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
771 1.29 explorer #endif
772 1.29 explorer } else {
773 1.29 explorer tcp_iss &= TCP_ISS_RANDOM_MASK;
774 1.30 explorer tcp_iss_seq += TCP_ISSINCR;
775 1.29 explorer tcp_iss += tcp_iss_seq;
776 1.29 explorer #ifdef TCPISS_DEBUG
777 1.29 explorer printf("ISS %08x\n", tcp_iss);
778 1.29 explorer #endif
779 1.29 explorer }
780 1.29 explorer
781 1.29 explorer #ifdef TCP_COMPAT_42
782 1.29 explorer /*
783 1.29 explorer * limit it to the positive range for really old TCP implementations
784 1.29 explorer */
785 1.29 explorer if ((int)tcp_iss < 0)
786 1.29 explorer tcp_iss &= 0x7fffffff; /* XXX */
787 1.29 explorer #endif
788 1.29 explorer
789 1.29 explorer return tcp_iss;
790 1.1 cgd }
791