tcp_subr.c revision 1.40 1 1.40 mellon /* $NetBSD: tcp_subr.c,v 1.40 1998/01/30 08:42:11 mellon Exp $ */
2 1.11 cgd
3 1.1 cgd /*
4 1.38 thorpej * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
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 1.38 thorpej * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
36 1.1 cgd */
37 1.1 cgd
38 1.39 scottr #include "opt_tcp_compat_42.h"
39 1.30 explorer #include "rnd.h"
40 1.30 explorer
41 1.5 mycroft #include <sys/param.h>
42 1.10 mycroft #include <sys/proc.h>
43 1.5 mycroft #include <sys/systm.h>
44 1.5 mycroft #include <sys/malloc.h>
45 1.5 mycroft #include <sys/mbuf.h>
46 1.5 mycroft #include <sys/socket.h>
47 1.5 mycroft #include <sys/socketvar.h>
48 1.5 mycroft #include <sys/protosw.h>
49 1.5 mycroft #include <sys/errno.h>
50 1.27 thorpej #include <sys/kernel.h>
51 1.30 explorer #if NRND > 0
52 1.29 explorer #include <sys/rnd.h>
53 1.30 explorer #endif
54 1.1 cgd
55 1.5 mycroft #include <net/route.h>
56 1.5 mycroft #include <net/if.h>
57 1.1 cgd
58 1.5 mycroft #include <netinet/in.h>
59 1.5 mycroft #include <netinet/in_systm.h>
60 1.5 mycroft #include <netinet/ip.h>
61 1.5 mycroft #include <netinet/in_pcb.h>
62 1.5 mycroft #include <netinet/ip_var.h>
63 1.5 mycroft #include <netinet/ip_icmp.h>
64 1.5 mycroft #include <netinet/tcp.h>
65 1.5 mycroft #include <netinet/tcp_fsm.h>
66 1.5 mycroft #include <netinet/tcp_seq.h>
67 1.5 mycroft #include <netinet/tcp_timer.h>
68 1.5 mycroft #include <netinet/tcp_var.h>
69 1.5 mycroft #include <netinet/tcpip.h>
70 1.1 cgd
71 1.1 cgd /* patchable/settable parameters for tcp */
72 1.1 cgd int tcp_mssdflt = TCP_MSS;
73 1.1 cgd int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
74 1.10 mycroft int tcp_do_rfc1323 = 1;
75 1.36 thorpej int tcp_init_win = 1;
76 1.1 cgd
77 1.21 mycroft #ifndef TCBHASHSIZE
78 1.21 mycroft #define TCBHASHSIZE 128
79 1.21 mycroft #endif
80 1.21 mycroft int tcbhashsize = TCBHASHSIZE;
81 1.1 cgd
82 1.35 thorpej int tcp_freeq __P((struct tcpcb *));
83 1.35 thorpej
84 1.1 cgd /*
85 1.1 cgd * Tcp initialization
86 1.1 cgd */
87 1.7 mycroft void
88 1.1 cgd tcp_init()
89 1.1 cgd {
90 1.1 cgd
91 1.24 mycroft in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
92 1.37 thorpej LIST_INIT(&tcp_delacks);
93 1.1 cgd if (max_protohdr < sizeof(struct tcpiphdr))
94 1.1 cgd max_protohdr = sizeof(struct tcpiphdr);
95 1.1 cgd if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
96 1.1 cgd panic("tcp_init");
97 1.1 cgd }
98 1.1 cgd
99 1.1 cgd /*
100 1.1 cgd * Create template to be used to send tcp packets on a connection.
101 1.1 cgd * Call after host entry created, allocates an mbuf and fills
102 1.1 cgd * in a skeletal tcp/ip header, minimizing the amount of work
103 1.1 cgd * necessary when the connection is used.
104 1.1 cgd */
105 1.1 cgd struct tcpiphdr *
106 1.1 cgd tcp_template(tp)
107 1.1 cgd struct tcpcb *tp;
108 1.1 cgd {
109 1.1 cgd register struct inpcb *inp = tp->t_inpcb;
110 1.1 cgd register struct tcpiphdr *n;
111 1.1 cgd
112 1.1 cgd if ((n = tp->t_template) == 0) {
113 1.26 thorpej MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
114 1.26 thorpej M_MBUF, M_NOWAIT);
115 1.26 thorpej if (n == NULL)
116 1.1 cgd return (0);
117 1.1 cgd }
118 1.20 cgd bzero(n->ti_x1, sizeof n->ti_x1);
119 1.1 cgd n->ti_pr = IPPROTO_TCP;
120 1.1 cgd n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
121 1.1 cgd n->ti_src = inp->inp_laddr;
122 1.1 cgd n->ti_dst = inp->inp_faddr;
123 1.1 cgd n->ti_sport = inp->inp_lport;
124 1.1 cgd n->ti_dport = inp->inp_fport;
125 1.1 cgd n->ti_seq = 0;
126 1.1 cgd n->ti_ack = 0;
127 1.1 cgd n->ti_x2 = 0;
128 1.1 cgd n->ti_off = 5;
129 1.1 cgd n->ti_flags = 0;
130 1.1 cgd n->ti_win = 0;
131 1.1 cgd n->ti_sum = 0;
132 1.1 cgd n->ti_urp = 0;
133 1.1 cgd return (n);
134 1.1 cgd }
135 1.1 cgd
136 1.1 cgd /*
137 1.1 cgd * Send a single message to the TCP at address specified by
138 1.1 cgd * the given TCP/IP header. If m == 0, then we make a copy
139 1.1 cgd * of the tcpiphdr at ti and send directly to the addressed host.
140 1.1 cgd * This is used to force keep alive messages out using the TCP
141 1.1 cgd * template for a connection tp->t_template. If flags are given
142 1.1 cgd * then we send a message back to the TCP which originated the
143 1.1 cgd * segment ti, and discard the mbuf containing it and any other
144 1.1 cgd * attached mbufs.
145 1.1 cgd *
146 1.1 cgd * In any case the ack and sequence number of the transmitted
147 1.1 cgd * segment are as specified by the parameters.
148 1.1 cgd */
149 1.27 thorpej int
150 1.1 cgd tcp_respond(tp, ti, m, ack, seq, flags)
151 1.1 cgd struct tcpcb *tp;
152 1.1 cgd register struct tcpiphdr *ti;
153 1.1 cgd register struct mbuf *m;
154 1.1 cgd tcp_seq ack, seq;
155 1.1 cgd int flags;
156 1.1 cgd {
157 1.1 cgd register int tlen;
158 1.1 cgd int win = 0;
159 1.1 cgd struct route *ro = 0;
160 1.1 cgd
161 1.1 cgd if (tp) {
162 1.1 cgd win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
163 1.1 cgd ro = &tp->t_inpcb->inp_route;
164 1.1 cgd }
165 1.1 cgd if (m == 0) {
166 1.1 cgd m = m_gethdr(M_DONTWAIT, MT_HEADER);
167 1.1 cgd if (m == NULL)
168 1.27 thorpej return (ENOBUFS);
169 1.1 cgd #ifdef TCP_COMPAT_42
170 1.1 cgd tlen = 1;
171 1.1 cgd #else
172 1.1 cgd tlen = 0;
173 1.1 cgd #endif
174 1.1 cgd m->m_data += max_linkhdr;
175 1.1 cgd *mtod(m, struct tcpiphdr *) = *ti;
176 1.1 cgd ti = mtod(m, struct tcpiphdr *);
177 1.1 cgd flags = TH_ACK;
178 1.1 cgd } else {
179 1.1 cgd m_freem(m->m_next);
180 1.1 cgd m->m_next = 0;
181 1.1 cgd m->m_data = (caddr_t)ti;
182 1.1 cgd m->m_len = sizeof (struct tcpiphdr);
183 1.1 cgd tlen = 0;
184 1.10 mycroft #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
185 1.13 cgd xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
186 1.13 cgd xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
187 1.1 cgd #undef xchg
188 1.1 cgd }
189 1.20 cgd bzero(ti->ti_x1, sizeof ti->ti_x1);
190 1.1 cgd ti->ti_seq = htonl(seq);
191 1.1 cgd ti->ti_ack = htonl(ack);
192 1.1 cgd ti->ti_x2 = 0;
193 1.27 thorpej if ((flags & TH_SYN) == 0) {
194 1.27 thorpej if (tp)
195 1.27 thorpej ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
196 1.27 thorpej else
197 1.27 thorpej ti->ti_win = htons((u_int16_t)win);
198 1.27 thorpej ti->ti_off = sizeof (struct tcphdr) >> 2;
199 1.27 thorpej tlen += sizeof (struct tcphdr);
200 1.27 thorpej } else
201 1.27 thorpej tlen += ti->ti_off << 2;
202 1.27 thorpej ti->ti_len = htons((u_int16_t)tlen);
203 1.27 thorpej tlen += sizeof (struct ip);
204 1.27 thorpej m->m_len = tlen;
205 1.27 thorpej m->m_pkthdr.len = tlen;
206 1.27 thorpej m->m_pkthdr.rcvif = (struct ifnet *) 0;
207 1.1 cgd ti->ti_flags = flags;
208 1.1 cgd ti->ti_urp = 0;
209 1.10 mycroft ti->ti_sum = 0;
210 1.1 cgd ti->ti_sum = in_cksum(m, tlen);
211 1.1 cgd ((struct ip *)ti)->ip_len = tlen;
212 1.10 mycroft ((struct ip *)ti)->ip_ttl = ip_defttl;
213 1.27 thorpej return ip_output(m, NULL, ro, 0, NULL);
214 1.1 cgd }
215 1.1 cgd
216 1.1 cgd /*
217 1.1 cgd * Create a new TCP control block, making an
218 1.1 cgd * empty reassembly queue and hooking it to the argument
219 1.1 cgd * protocol control block.
220 1.1 cgd */
221 1.1 cgd struct tcpcb *
222 1.1 cgd tcp_newtcpcb(inp)
223 1.1 cgd struct inpcb *inp;
224 1.1 cgd {
225 1.1 cgd register struct tcpcb *tp;
226 1.1 cgd
227 1.10 mycroft tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
228 1.10 mycroft if (tp == NULL)
229 1.1 cgd return ((struct tcpcb *)0);
230 1.23 mycroft bzero((caddr_t)tp, sizeof(struct tcpcb));
231 1.20 cgd LIST_INIT(&tp->segq);
232 1.33 kml tp->t_peermss = tcp_mssdflt;
233 1.28 thorpej tp->t_ourmss = tcp_mssdflt;
234 1.33 kml tp->t_segsz = tcp_mssdflt;
235 1.1 cgd
236 1.10 mycroft tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
237 1.1 cgd tp->t_inpcb = inp;
238 1.1 cgd /*
239 1.1 cgd * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
240 1.1 cgd * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives
241 1.1 cgd * reasonable initial retransmit time.
242 1.1 cgd */
243 1.1 cgd tp->t_srtt = TCPTV_SRTTBASE;
244 1.15 mycroft tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
245 1.1 cgd tp->t_rttmin = TCPTV_MIN;
246 1.15 mycroft TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
247 1.1 cgd TCPTV_MIN, TCPTV_REXMTMAX);
248 1.10 mycroft tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
249 1.10 mycroft tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
250 1.10 mycroft inp->inp_ip.ip_ttl = ip_defttl;
251 1.1 cgd inp->inp_ppcb = (caddr_t)tp;
252 1.1 cgd return (tp);
253 1.1 cgd }
254 1.1 cgd
255 1.1 cgd /*
256 1.1 cgd * Drop a TCP connection, reporting
257 1.1 cgd * the specified error. If connection is synchronized,
258 1.1 cgd * then send a RST to peer.
259 1.1 cgd */
260 1.1 cgd struct tcpcb *
261 1.1 cgd tcp_drop(tp, errno)
262 1.1 cgd register struct tcpcb *tp;
263 1.1 cgd int errno;
264 1.1 cgd {
265 1.1 cgd struct socket *so = tp->t_inpcb->inp_socket;
266 1.1 cgd
267 1.1 cgd if (TCPS_HAVERCVDSYN(tp->t_state)) {
268 1.1 cgd tp->t_state = TCPS_CLOSED;
269 1.1 cgd (void) tcp_output(tp);
270 1.1 cgd tcpstat.tcps_drops++;
271 1.1 cgd } else
272 1.1 cgd tcpstat.tcps_conndrops++;
273 1.1 cgd if (errno == ETIMEDOUT && tp->t_softerror)
274 1.1 cgd errno = tp->t_softerror;
275 1.1 cgd so->so_error = errno;
276 1.1 cgd return (tcp_close(tp));
277 1.1 cgd }
278 1.1 cgd
279 1.1 cgd /*
280 1.1 cgd * Close a TCP control block:
281 1.1 cgd * discard all space held by the tcp
282 1.1 cgd * discard internet protocol block
283 1.1 cgd * wake up any sleepers
284 1.1 cgd */
285 1.1 cgd struct tcpcb *
286 1.1 cgd tcp_close(tp)
287 1.1 cgd register struct tcpcb *tp;
288 1.1 cgd {
289 1.1 cgd struct inpcb *inp = tp->t_inpcb;
290 1.1 cgd struct socket *so = inp->inp_socket;
291 1.1 cgd #ifdef RTV_RTT
292 1.1 cgd register struct rtentry *rt;
293 1.1 cgd
294 1.1 cgd /*
295 1.1 cgd * If we sent enough data to get some meaningful characteristics,
296 1.10 mycroft * save them in the routing entry. 'Enough' is arbitrarily
297 1.1 cgd * defined as the sendpipesize (default 4K) * 16. This would
298 1.1 cgd * give us 16 rtt samples assuming we only get one sample per
299 1.1 cgd * window (the usual case on a long haul net). 16 samples is
300 1.1 cgd * enough for the srtt filter to converge to within 5% of the correct
301 1.1 cgd * value; fewer samples and we could save a very bogus rtt.
302 1.1 cgd *
303 1.1 cgd * Don't update the default route's characteristics and don't
304 1.1 cgd * update anything that the user "locked".
305 1.1 cgd */
306 1.1 cgd if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
307 1.1 cgd (rt = inp->inp_route.ro_rt) &&
308 1.23 mycroft !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
309 1.22 christos register u_long i = 0;
310 1.1 cgd
311 1.1 cgd if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
312 1.1 cgd i = tp->t_srtt *
313 1.25 mycroft ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
314 1.1 cgd if (rt->rt_rmx.rmx_rtt && i)
315 1.1 cgd /*
316 1.1 cgd * filter this update to half the old & half
317 1.1 cgd * the new values, converting scale.
318 1.1 cgd * See route.h and tcp_var.h for a
319 1.1 cgd * description of the scaling constants.
320 1.1 cgd */
321 1.1 cgd rt->rt_rmx.rmx_rtt =
322 1.1 cgd (rt->rt_rmx.rmx_rtt + i) / 2;
323 1.1 cgd else
324 1.1 cgd rt->rt_rmx.rmx_rtt = i;
325 1.1 cgd }
326 1.1 cgd if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
327 1.1 cgd i = tp->t_rttvar *
328 1.25 mycroft ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
329 1.1 cgd if (rt->rt_rmx.rmx_rttvar && i)
330 1.1 cgd rt->rt_rmx.rmx_rttvar =
331 1.1 cgd (rt->rt_rmx.rmx_rttvar + i) / 2;
332 1.1 cgd else
333 1.1 cgd rt->rt_rmx.rmx_rttvar = i;
334 1.1 cgd }
335 1.1 cgd /*
336 1.1 cgd * update the pipelimit (ssthresh) if it has been updated
337 1.1 cgd * already or if a pipesize was specified & the threshhold
338 1.1 cgd * got below half the pipesize. I.e., wait for bad news
339 1.1 cgd * before we start updating, then update on both good
340 1.1 cgd * and bad news.
341 1.1 cgd */
342 1.22 christos if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
343 1.22 christos (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
344 1.1 cgd i < (rt->rt_rmx.rmx_sendpipe / 2)) {
345 1.1 cgd /*
346 1.1 cgd * convert the limit from user data bytes to
347 1.1 cgd * packets then to packet data bytes.
348 1.1 cgd */
349 1.33 kml i = (i + tp->t_segsz / 2) / tp->t_segsz;
350 1.1 cgd if (i < 2)
351 1.1 cgd i = 2;
352 1.33 kml i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
353 1.1 cgd if (rt->rt_rmx.rmx_ssthresh)
354 1.1 cgd rt->rt_rmx.rmx_ssthresh =
355 1.1 cgd (rt->rt_rmx.rmx_ssthresh + i) / 2;
356 1.1 cgd else
357 1.1 cgd rt->rt_rmx.rmx_ssthresh = i;
358 1.1 cgd }
359 1.1 cgd }
360 1.9 mycroft #endif /* RTV_RTT */
361 1.1 cgd /* free the reassembly queue, if any */
362 1.35 thorpej (void) tcp_freeq(tp);
363 1.40 mellon TCP_CLEAR_DELACK(tp);
364 1.35 thorpej
365 1.1 cgd if (tp->t_template)
366 1.26 thorpej FREE(tp->t_template, M_MBUF);
367 1.10 mycroft free(tp, M_PCB);
368 1.1 cgd inp->inp_ppcb = 0;
369 1.1 cgd soisdisconnected(so);
370 1.1 cgd in_pcbdetach(inp);
371 1.1 cgd tcpstat.tcps_closed++;
372 1.1 cgd return ((struct tcpcb *)0);
373 1.1 cgd }
374 1.1 cgd
375 1.35 thorpej int
376 1.35 thorpej tcp_freeq(tp)
377 1.35 thorpej struct tcpcb *tp;
378 1.35 thorpej {
379 1.35 thorpej register struct ipqent *qe;
380 1.35 thorpej int rv = 0;
381 1.35 thorpej
382 1.35 thorpej while ((qe = tp->segq.lh_first) != NULL) {
383 1.35 thorpej LIST_REMOVE(qe, ipqe_q);
384 1.35 thorpej m_freem(qe->ipqe_m);
385 1.35 thorpej FREE(qe, M_IPQ);
386 1.35 thorpej rv = 1;
387 1.35 thorpej }
388 1.35 thorpej return (rv);
389 1.35 thorpej }
390 1.35 thorpej
391 1.35 thorpej /*
392 1.35 thorpej * Protocol drain routine. Called when memory is in short supply.
393 1.35 thorpej */
394 1.7 mycroft void
395 1.1 cgd tcp_drain()
396 1.1 cgd {
397 1.35 thorpej register struct inpcb *inp;
398 1.35 thorpej register struct tcpcb *tp;
399 1.1 cgd
400 1.35 thorpej /*
401 1.35 thorpej * Free the sequence queue of all TCP connections.
402 1.35 thorpej */
403 1.35 thorpej inp = tcbtable.inpt_queue.cqh_first;
404 1.35 thorpej if (inp) /* XXX */
405 1.35 thorpej for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
406 1.35 thorpej inp = inp->inp_queue.cqe_next) {
407 1.35 thorpej if ((tp = intotcpcb(inp)) != NULL) {
408 1.35 thorpej if (tcp_freeq(tp))
409 1.35 thorpej tcpstat.tcps_connsdrained++;
410 1.35 thorpej }
411 1.35 thorpej }
412 1.1 cgd }
413 1.1 cgd
414 1.1 cgd /*
415 1.1 cgd * Notify a tcp user of an asynchronous error;
416 1.1 cgd * store error as soft error, but wake up user
417 1.1 cgd * (for now, won't do anything until can select for soft error).
418 1.1 cgd */
419 1.7 mycroft void
420 1.1 cgd tcp_notify(inp, error)
421 1.10 mycroft struct inpcb *inp;
422 1.1 cgd int error;
423 1.1 cgd {
424 1.10 mycroft register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
425 1.10 mycroft register struct socket *so = inp->inp_socket;
426 1.1 cgd
427 1.10 mycroft /*
428 1.10 mycroft * Ignore some errors if we are hooked up.
429 1.10 mycroft * If connection hasn't completed, has retransmitted several times,
430 1.10 mycroft * and receives a second error, give up now. This is better
431 1.10 mycroft * than waiting a long time to establish a connection that
432 1.10 mycroft * can never complete.
433 1.10 mycroft */
434 1.10 mycroft if (tp->t_state == TCPS_ESTABLISHED &&
435 1.10 mycroft (error == EHOSTUNREACH || error == ENETUNREACH ||
436 1.10 mycroft error == EHOSTDOWN)) {
437 1.10 mycroft return;
438 1.12 mycroft } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
439 1.12 mycroft tp->t_rxtshift > 3 && tp->t_softerror)
440 1.10 mycroft so->so_error = error;
441 1.10 mycroft else
442 1.10 mycroft tp->t_softerror = error;
443 1.10 mycroft wakeup((caddr_t) &so->so_timeo);
444 1.10 mycroft sorwakeup(so);
445 1.10 mycroft sowwakeup(so);
446 1.1 cgd }
447 1.1 cgd
448 1.22 christos void *
449 1.22 christos tcp_ctlinput(cmd, sa, v)
450 1.1 cgd int cmd;
451 1.1 cgd struct sockaddr *sa;
452 1.22 christos register void *v;
453 1.1 cgd {
454 1.22 christos register struct ip *ip = v;
455 1.1 cgd register struct tcphdr *th;
456 1.19 mycroft extern int inetctlerrmap[];
457 1.7 mycroft void (*notify) __P((struct inpcb *, int)) = tcp_notify;
458 1.19 mycroft int errno;
459 1.27 thorpej int nmatch;
460 1.1 cgd
461 1.18 mycroft if ((unsigned)cmd >= PRC_NCMDS)
462 1.22 christos return NULL;
463 1.18 mycroft errno = inetctlerrmap[cmd];
464 1.17 mycroft if (cmd == PRC_QUENCH)
465 1.17 mycroft notify = tcp_quench;
466 1.17 mycroft else if (PRC_IS_REDIRECT(cmd))
467 1.17 mycroft notify = in_rtchange, ip = 0;
468 1.32 kml else if (cmd == PRC_MSGSIZE && ip_mtudisc)
469 1.31 kml notify = tcp_mtudisc, ip = 0;
470 1.16 mycroft else if (cmd == PRC_HOSTDEAD)
471 1.17 mycroft ip = 0;
472 1.18 mycroft else if (errno == 0)
473 1.22 christos return NULL;
474 1.17 mycroft if (ip) {
475 1.1 cgd th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
476 1.27 thorpej nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
477 1.27 thorpej th->th_dport, ip->ip_src, th->th_sport, errno, notify);
478 1.27 thorpej if (nmatch == 0 && syn_cache_count &&
479 1.27 thorpej (inetctlerrmap[cmd] == EHOSTUNREACH ||
480 1.27 thorpej inetctlerrmap[cmd] == ENETUNREACH ||
481 1.27 thorpej inetctlerrmap[cmd] == EHOSTDOWN))
482 1.27 thorpej syn_cache_unreach(ip, th);
483 1.17 mycroft } else
484 1.27 thorpej (void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
485 1.23 mycroft notify);
486 1.22 christos return NULL;
487 1.1 cgd }
488 1.1 cgd
489 1.1 cgd /*
490 1.1 cgd * When a source quench is received, close congestion window
491 1.1 cgd * to one segment. We will gradually open it again as we proceed.
492 1.1 cgd */
493 1.7 mycroft void
494 1.7 mycroft tcp_quench(inp, errno)
495 1.1 cgd struct inpcb *inp;
496 1.7 mycroft int errno;
497 1.1 cgd {
498 1.1 cgd struct tcpcb *tp = intotcpcb(inp);
499 1.1 cgd
500 1.1 cgd if (tp)
501 1.33 kml tp->snd_cwnd = tp->t_segsz;
502 1.28 thorpej }
503 1.31 kml
504 1.31 kml /*
505 1.31 kml * On receipt of path MTU corrections, flush old route and replace it
506 1.31 kml * with the new one. Retransmit all unacknowledged packets, to ensure
507 1.31 kml * that all packets will be received.
508 1.31 kml */
509 1.31 kml void
510 1.31 kml tcp_mtudisc(inp, errno)
511 1.31 kml struct inpcb *inp;
512 1.31 kml int errno;
513 1.31 kml {
514 1.31 kml struct tcpcb *tp = intotcpcb(inp);
515 1.31 kml struct rtentry *rt = in_pcbrtentry(inp);
516 1.31 kml
517 1.31 kml if (tp != 0) {
518 1.31 kml if (rt != 0) {
519 1.36 thorpej /*
520 1.36 thorpej * If this was not a host route, remove and realloc.
521 1.36 thorpej */
522 1.31 kml if ((rt->rt_flags & RTF_HOST) == 0) {
523 1.31 kml in_rtchange(inp, errno);
524 1.31 kml if ((rt = in_pcbrtentry(inp)) == 0)
525 1.31 kml return;
526 1.31 kml }
527 1.36 thorpej
528 1.36 thorpej /*
529 1.36 thorpej * Slow start out of the error condition. We
530 1.36 thorpej * use the MTU because we know it's smaller
531 1.36 thorpej * than the previously transmitted segment.
532 1.36 thorpej */
533 1.33 kml if (rt->rt_rmx.rmx_mtu != 0)
534 1.36 thorpej tp->snd_cwnd =
535 1.36 thorpej TCP_INITIAL_WINDOW(rt->rt_rmx.rmx_mtu);
536 1.31 kml }
537 1.31 kml
538 1.36 thorpej /*
539 1.36 thorpej * Resend unacknowledged packets.
540 1.36 thorpej */
541 1.31 kml tp->snd_nxt = tp->snd_una;
542 1.31 kml tcp_output(tp);
543 1.31 kml }
544 1.31 kml }
545 1.31 kml
546 1.28 thorpej
547 1.28 thorpej /*
548 1.28 thorpej * Compute the MSS to advertise to the peer. Called only during
549 1.28 thorpej * the 3-way handshake. If we are the server (peer initiated
550 1.28 thorpej * connection), we are called with the TCPCB for the listen
551 1.28 thorpej * socket. If we are the client (we initiated connection), we
552 1.28 thorpej * are called witht he TCPCB for the actual connection.
553 1.28 thorpej */
554 1.28 thorpej int
555 1.28 thorpej tcp_mss_to_advertise(tp)
556 1.28 thorpej const struct tcpcb *tp;
557 1.28 thorpej {
558 1.28 thorpej extern u_long in_maxmtu;
559 1.28 thorpej struct inpcb *inp;
560 1.28 thorpej struct socket *so;
561 1.28 thorpej int mss;
562 1.28 thorpej
563 1.28 thorpej inp = tp->t_inpcb;
564 1.28 thorpej so = inp->inp_socket;
565 1.28 thorpej
566 1.28 thorpej /*
567 1.28 thorpej * In order to avoid defeating path MTU discovery on the peer,
568 1.28 thorpej * we advertise the max MTU of all attached networks as our MSS,
569 1.28 thorpej * per RFC 1191, section 3.1.
570 1.28 thorpej *
571 1.28 thorpej * XXX Should we allow room for the timestamp option if
572 1.28 thorpej * XXX rfc1323 is enabled?
573 1.28 thorpej */
574 1.28 thorpej mss = in_maxmtu - sizeof(struct tcpiphdr);
575 1.28 thorpej
576 1.28 thorpej return (mss);
577 1.28 thorpej }
578 1.28 thorpej
579 1.28 thorpej /*
580 1.28 thorpej * Set connection variables based on the peer's advertised MSS.
581 1.28 thorpej * We are passed the TCPCB for the actual connection. If we
582 1.28 thorpej * are the server, we are called by the compressed state engine
583 1.28 thorpej * when the 3-way handshake is complete. If we are the client,
584 1.28 thorpej * we are called when we recieve the SYN,ACK from the server.
585 1.28 thorpej *
586 1.28 thorpej * NOTE: Our advertised MSS value must be initialized in the TCPCB
587 1.28 thorpej * before this routine is called!
588 1.28 thorpej */
589 1.28 thorpej void
590 1.28 thorpej tcp_mss_from_peer(tp, offer)
591 1.28 thorpej struct tcpcb *tp;
592 1.28 thorpej int offer;
593 1.28 thorpej {
594 1.28 thorpej struct inpcb *inp = tp->t_inpcb;
595 1.28 thorpej struct socket *so = inp->inp_socket;
596 1.28 thorpej #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
597 1.28 thorpej struct rtentry *rt = in_pcbrtentry(inp);
598 1.28 thorpej #endif
599 1.28 thorpej u_long bufsize;
600 1.28 thorpej int mss;
601 1.28 thorpej
602 1.28 thorpej /*
603 1.28 thorpej * Assume our MSS is the MSS of the peer, unless they sent us
604 1.28 thorpej * an offer. Do not accept offers less than 32 bytes.
605 1.28 thorpej */
606 1.28 thorpej mss = tp->t_ourmss;
607 1.28 thorpej if (offer)
608 1.28 thorpej mss = offer;
609 1.28 thorpej mss = max(mss, 32); /* sanity */
610 1.28 thorpej
611 1.28 thorpej /*
612 1.28 thorpej * If there's a pipesize, change the socket buffer to that size.
613 1.28 thorpej * Make the socket buffer an integral number of MSS units. If
614 1.28 thorpej * the MSS is larger than the socket buffer, artificially decrease
615 1.28 thorpej * the MSS.
616 1.28 thorpej */
617 1.28 thorpej #ifdef RTV_SPIPE
618 1.28 thorpej if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
619 1.28 thorpej bufsize = rt->rt_rmx.rmx_sendpipe;
620 1.28 thorpej else
621 1.28 thorpej #endif
622 1.28 thorpej bufsize = so->so_snd.sb_hiwat;
623 1.28 thorpej if (bufsize < mss)
624 1.28 thorpej mss = bufsize;
625 1.28 thorpej else {
626 1.28 thorpej bufsize = roundup(bufsize, mss);
627 1.28 thorpej if (bufsize > sb_max)
628 1.28 thorpej bufsize = sb_max;
629 1.28 thorpej (void) sbreserve(&so->so_snd, bufsize);
630 1.28 thorpej }
631 1.33 kml tp->t_peermss = mss;
632 1.33 kml tp->t_segsz = mss;
633 1.28 thorpej
634 1.28 thorpej /* Initialize the initial congestion window. */
635 1.36 thorpej tp->snd_cwnd = TCP_INITIAL_WINDOW(mss);
636 1.28 thorpej
637 1.28 thorpej #ifdef RTV_SSTHRESH
638 1.28 thorpej if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
639 1.28 thorpej /*
640 1.28 thorpej * There's some sort of gateway or interface buffer
641 1.28 thorpej * limit on the path. Use this to set the slow
642 1.28 thorpej * start threshold, but set the threshold to no less
643 1.28 thorpej * than 2 * MSS.
644 1.28 thorpej */
645 1.28 thorpej tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
646 1.28 thorpej }
647 1.28 thorpej #endif
648 1.28 thorpej }
649 1.28 thorpej
650 1.28 thorpej /*
651 1.28 thorpej * Processing necessary when a TCP connection is established.
652 1.28 thorpej */
653 1.28 thorpej void
654 1.28 thorpej tcp_established(tp)
655 1.28 thorpej struct tcpcb *tp;
656 1.28 thorpej {
657 1.28 thorpej struct inpcb *inp = tp->t_inpcb;
658 1.28 thorpej struct socket *so = inp->inp_socket;
659 1.28 thorpej #ifdef RTV_RPIPE
660 1.28 thorpej struct rtentry *rt = in_pcbrtentry(inp);
661 1.28 thorpej #endif
662 1.28 thorpej u_long bufsize;
663 1.28 thorpej
664 1.28 thorpej tp->t_state = TCPS_ESTABLISHED;
665 1.28 thorpej tp->t_timer[TCPT_KEEP] = tcp_keepidle;
666 1.28 thorpej
667 1.28 thorpej #ifdef RTV_RPIPE
668 1.28 thorpej if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
669 1.28 thorpej bufsize = rt->rt_rmx.rmx_recvpipe;
670 1.28 thorpej else
671 1.28 thorpej #endif
672 1.28 thorpej bufsize = so->so_rcv.sb_hiwat;
673 1.28 thorpej if (bufsize > tp->t_ourmss) {
674 1.28 thorpej bufsize = roundup(bufsize, tp->t_ourmss);
675 1.28 thorpej if (bufsize > sb_max)
676 1.28 thorpej bufsize = sb_max;
677 1.28 thorpej (void) sbreserve(&so->so_rcv, bufsize);
678 1.28 thorpej }
679 1.28 thorpej }
680 1.28 thorpej
681 1.28 thorpej /*
682 1.28 thorpej * Check if there's an initial rtt or rttvar. Convert from the
683 1.28 thorpej * route-table units to scaled multiples of the slow timeout timer.
684 1.28 thorpej * Called only during the 3-way handshake.
685 1.28 thorpej */
686 1.28 thorpej void
687 1.28 thorpej tcp_rmx_rtt(tp)
688 1.28 thorpej struct tcpcb *tp;
689 1.28 thorpej {
690 1.28 thorpej #ifdef RTV_RTT
691 1.28 thorpej struct rtentry *rt;
692 1.28 thorpej int rtt;
693 1.28 thorpej
694 1.28 thorpej if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
695 1.28 thorpej return;
696 1.28 thorpej
697 1.28 thorpej if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
698 1.28 thorpej /*
699 1.28 thorpej * XXX The lock bit for MTU indicates that the value
700 1.28 thorpej * is also a minimum value; this is subject to time.
701 1.28 thorpej */
702 1.28 thorpej if (rt->rt_rmx.rmx_locks & RTV_RTT)
703 1.28 thorpej tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
704 1.28 thorpej tp->t_srtt = rtt /
705 1.28 thorpej ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
706 1.28 thorpej if (rt->rt_rmx.rmx_rttvar) {
707 1.28 thorpej tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
708 1.28 thorpej ((RTM_RTTUNIT / PR_SLOWHZ) >>
709 1.28 thorpej (TCP_RTTVAR_SHIFT + 2));
710 1.28 thorpej } else {
711 1.28 thorpej /* Default variation is +- 1 rtt */
712 1.28 thorpej tp->t_rttvar =
713 1.28 thorpej tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
714 1.28 thorpej }
715 1.28 thorpej TCPT_RANGESET(tp->t_rxtcur,
716 1.28 thorpej ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
717 1.28 thorpej tp->t_rttmin, TCPTV_REXMTMAX);
718 1.28 thorpej }
719 1.28 thorpej #endif
720 1.29 explorer }
721 1.29 explorer
722 1.30 explorer tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */
723 1.30 explorer
724 1.29 explorer /*
725 1.29 explorer * Get a new sequence value given a tcp control block
726 1.29 explorer */
727 1.29 explorer tcp_seq
728 1.29 explorer tcp_new_iss(tp, len, addin)
729 1.29 explorer void *tp;
730 1.29 explorer u_long len;
731 1.29 explorer tcp_seq addin;
732 1.29 explorer {
733 1.29 explorer tcp_seq tcp_iss;
734 1.29 explorer
735 1.29 explorer /*
736 1.29 explorer * add randomness about this connection, but do not estimate
737 1.29 explorer * entropy from the timing, since the physical device driver would
738 1.29 explorer * have done that for us.
739 1.29 explorer */
740 1.30 explorer #if NRND > 0
741 1.29 explorer if (tp != NULL)
742 1.29 explorer rnd_add_data(NULL, tp, len, 0);
743 1.30 explorer #endif
744 1.29 explorer
745 1.29 explorer /*
746 1.29 explorer * randomize.
747 1.29 explorer */
748 1.30 explorer #if NRND > 0
749 1.29 explorer rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
750 1.30 explorer #else
751 1.30 explorer tcp_iss = random();
752 1.30 explorer #endif
753 1.29 explorer
754 1.29 explorer /*
755 1.29 explorer * If we were asked to add some amount to a known value,
756 1.29 explorer * we will take a random value obtained above, mask off the upper
757 1.29 explorer * bits, and add in the known value. We also add in a constant to
758 1.29 explorer * ensure that we are at least a certain distance from the original
759 1.29 explorer * value.
760 1.29 explorer *
761 1.29 explorer * This is used when an old connection is in timed wait
762 1.29 explorer * and we have a new one coming in, for instance.
763 1.29 explorer */
764 1.29 explorer if (addin != 0) {
765 1.29 explorer #ifdef TCPISS_DEBUG
766 1.29 explorer printf("Random %08x, ", tcp_iss);
767 1.29 explorer #endif
768 1.29 explorer tcp_iss &= TCP_ISS_RANDOM_MASK;
769 1.30 explorer tcp_iss = tcp_iss + addin + TCP_ISSINCR;
770 1.30 explorer tcp_iss_seq += TCP_ISSINCR;
771 1.30 explorer tcp_iss += tcp_iss_seq;
772 1.29 explorer #ifdef TCPISS_DEBUG
773 1.29 explorer printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
774 1.29 explorer #endif
775 1.29 explorer } else {
776 1.29 explorer tcp_iss &= TCP_ISS_RANDOM_MASK;
777 1.30 explorer tcp_iss_seq += TCP_ISSINCR;
778 1.29 explorer tcp_iss += tcp_iss_seq;
779 1.29 explorer #ifdef TCPISS_DEBUG
780 1.29 explorer printf("ISS %08x\n", tcp_iss);
781 1.29 explorer #endif
782 1.29 explorer }
783 1.29 explorer
784 1.29 explorer #ifdef TCP_COMPAT_42
785 1.29 explorer /*
786 1.29 explorer * limit it to the positive range for really old TCP implementations
787 1.29 explorer */
788 1.29 explorer if ((int)tcp_iss < 0)
789 1.29 explorer tcp_iss &= 0x7fffffff; /* XXX */
790 1.29 explorer #endif
791 1.29 explorer
792 1.29 explorer return tcp_iss;
793 1.1 cgd }
794