1 /*
2 * Copyright (c) 1995, Mike Mitchell
3 * Copyright (c) 1984, 1985, 1986, 1987, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 * must display the following acknowledgement:
16 * This product includes software developed by the University of
17 * California, Berkeley and its contributors.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)spx_usrreq.h
35 *
36 * $FreeBSD: releng/5.1/sys/netipx/spx_usrreq.c 111119 2003-02-19 05:47:46Z imp $
37 */
38
39 #include <sys/param.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/protosw.h>
46 #include <sys/signalvar.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sx.h>
50 #include <sys/systm.h>
51
52 #include <net/route.h>
53 #include <netinet/tcp_fsm.h>
54
55 #include <netipx/ipx.h>
56 #include <netipx/ipx_pcb.h>
57 #include <netipx/ipx_var.h>
58 #include <netipx/spx.h>
59 #include <netipx/spx_debug.h>
60 #include <netipx/spx_timer.h>
61 #include <netipx/spx_var.h>
62
63 /*
64 * SPX protocol implementation.
65 */
66 static u_short spx_iss;
67 static u_short spx_newchecks[50];
68 static int spx_hardnosed;
69 static int spx_use_delack = 0;
70 static int traceallspxs = 0;
71 static struct spx spx_savesi;
72 static struct spx_istat spx_istat;
73
74 /* Following was struct spxstat spxstat; */
75 #ifndef spxstat
76 #define spxstat spx_istat.newstats
77 #endif
78
79 static int spx_backoff[SPX_MAXRXTSHIFT+1] =
80 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
81
82 static struct spxpcb *spx_close(struct spxpcb *cb);
83 static struct spxpcb *spx_disconnect(struct spxpcb *cb);
84 static struct spxpcb *spx_drop(struct spxpcb *cb, int errno);
85 static int spx_output(struct spxpcb *cb, struct mbuf *m0);
86 static int spx_reass(struct spxpcb *cb, struct spx *si);
87 static void spx_setpersist(struct spxpcb *cb);
88 static void spx_template(struct spxpcb *cb);
89 static struct spxpcb *spx_timers(struct spxpcb *cb, int timer);
90 static struct spxpcb *spx_usrclosed(struct spxpcb *cb);
91
92 static int spx_usr_abort(struct socket *so);
93 static int spx_accept(struct socket *so, struct sockaddr **nam);
94 static int spx_attach(struct socket *so, int proto, struct thread *td);
95 static int spx_bind(struct socket *so, struct sockaddr *nam, struct thread *td);
96 static int spx_connect(struct socket *so, struct sockaddr *nam,
97 struct thread *td);
98 static int spx_detach(struct socket *so);
99 static int spx_usr_disconnect(struct socket *so);
100 static int spx_listen(struct socket *so, struct thread *td);
101 static int spx_rcvd(struct socket *so, int flags);
102 static int spx_rcvoob(struct socket *so, struct mbuf *m, int flags);
103 static int spx_send(struct socket *so, int flags, struct mbuf *m,
104 struct sockaddr *addr, struct mbuf *control,
105 struct thread *td);
106 static int spx_shutdown(struct socket *so);
107 static int spx_sp_attach(struct socket *so, int proto, struct thread *td);
108
109 struct pr_usrreqs spx_usrreqs = {
110 spx_usr_abort, spx_accept, spx_attach, spx_bind,
111 spx_connect, pru_connect2_notsupp, ipx_control, spx_detach,
112 spx_usr_disconnect, spx_listen, ipx_peeraddr, spx_rcvd,
113 spx_rcvoob, spx_send, pru_sense_null, spx_shutdown,
114 ipx_sockaddr, sosend, soreceive, sopoll
115 };
116
117 struct pr_usrreqs spx_usrreq_sps = {
118 spx_usr_abort, spx_accept, spx_sp_attach, spx_bind,
119 spx_connect, pru_connect2_notsupp, ipx_control, spx_detach,
120 spx_usr_disconnect, spx_listen, ipx_peeraddr, spx_rcvd,
121 spx_rcvoob, spx_send, pru_sense_null, spx_shutdown,
122 ipx_sockaddr, sosend, soreceive, sopoll
123 };
124
125 void
126 spx_init()
127 {
128
129 spx_iss = 1; /* WRONG !! should fish it out of TODR */
130 }
131
132 void
133 spx_input(m, ipxp)
134 register struct mbuf *m;
135 register struct ipxpcb *ipxp;
136 {
137 register struct spxpcb *cb;
138 register struct spx *si = mtod(m, struct spx *);
139 register struct socket *so;
140 int dropsocket = 0;
141 short ostate = 0;
142
143 spxstat.spxs_rcvtotal++;
144 if (ipxp == NULL) {
145 panic("No ipxpcb in spx_input\n");
146 return;
147 }
148
149 cb = ipxtospxpcb(ipxp);
150 if (cb == NULL)
151 goto bad;
152
153 if (m->m_len < sizeof(*si)) {
154 if ((m = m_pullup(m, sizeof(*si))) == NULL) {
155 spxstat.spxs_rcvshort++;
156 return;
157 }
158 si = mtod(m, struct spx *);
159 }
160 si->si_seq = ntohs(si->si_seq);
161 si->si_ack = ntohs(si->si_ack);
162 si->si_alo = ntohs(si->si_alo);
163
164 so = ipxp->ipxp_socket;
165
166 if (so->so_options & SO_DEBUG || traceallspxs) {
167 ostate = cb->s_state;
168 spx_savesi = *si;
169 }
170 if (so->so_options & SO_ACCEPTCONN) {
171 struct spxpcb *ocb = cb;
172
173 so = sonewconn(so, 0);
174 if (so == NULL) {
175 goto drop;
176 }
177 /*
178 * This is ugly, but ....
179 *
180 * Mark socket as temporary until we're
181 * committed to keeping it. The code at
182 * ``drop'' and ``dropwithreset'' check the
183 * flag dropsocket to see if the temporary
184 * socket created here should be discarded.
185 * We mark the socket as discardable until
186 * we're committed to it below in TCPS_LISTEN.
187 */
188 dropsocket++;
189 ipxp = (struct ipxpcb *)so->so_pcb;
190 ipxp->ipxp_laddr = si->si_dna;
191 cb = ipxtospxpcb(ipxp);
192 cb->s_mtu = ocb->s_mtu; /* preserve sockopts */
193 cb->s_flags = ocb->s_flags; /* preserve sockopts */
194 cb->s_flags2 = ocb->s_flags2; /* preserve sockopts */
195 cb->s_state = TCPS_LISTEN;
196 }
197
198 /*
199 * Packet received on connection.
200 * reset idle time and keep-alive timer;
201 */
202 cb->s_idle = 0;
203 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
204
205 switch (cb->s_state) {
206
207 case TCPS_LISTEN:{
208 struct sockaddr_ipx *sipx, ssipx;
209 struct ipx_addr laddr;
210
211 /*
212 * If somebody here was carying on a conversation
213 * and went away, and his pen pal thinks he can
214 * still talk, we get the misdirected packet.
215 */
216 if (spx_hardnosed && (si->si_did != 0 || si->si_seq != 0)) {
217 spx_istat.gonawy++;
218 goto dropwithreset;
219 }
220 sipx = &ssipx;
221 bzero(sipx, sizeof *sipx);
222 sipx->sipx_len = sizeof(*sipx);
223 sipx->sipx_family = AF_IPX;
224 sipx->sipx_addr = si->si_sna;
225 laddr = ipxp->ipxp_laddr;
226 if (ipx_nullhost(laddr))
227 ipxp->ipxp_laddr = si->si_dna;
228 if (ipx_pcbconnect(ipxp, (struct sockaddr *)sipx, &thread0)) {
229 ipxp->ipxp_laddr = laddr;
230 spx_istat.noconn++;
231 goto drop;
232 }
233 spx_template(cb);
234 dropsocket = 0; /* committed to socket */
235 cb->s_did = si->si_sid;
236 cb->s_rack = si->si_ack;
237 cb->s_ralo = si->si_alo;
238 #define THREEWAYSHAKE
239 #ifdef THREEWAYSHAKE
240 cb->s_state = TCPS_SYN_RECEIVED;
241 cb->s_force = 1 + SPXT_KEEP;
242 spxstat.spxs_accepts++;
243 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
244 }
245 break;
246 /*
247 * This state means that we have heard a response
248 * to our acceptance of their connection
249 * It is probably logically unnecessary in this
250 * implementation.
251 */
252 case TCPS_SYN_RECEIVED: {
253 if (si->si_did != cb->s_sid) {
254 spx_istat.wrncon++;
255 goto drop;
256 }
257 #endif
258 ipxp->ipxp_fport = si->si_sport;
259 cb->s_timer[SPXT_REXMT] = 0;
260 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
261 soisconnected(so);
262 cb->s_state = TCPS_ESTABLISHED;
263 spxstat.spxs_accepts++;
264 }
265 break;
266
267 /*
268 * This state means that we have gotten a response
269 * to our attempt to establish a connection.
270 * We fill in the data from the other side,
271 * telling us which port to respond to, instead of the well-
272 * known one we might have sent to in the first place.
273 * We also require that this is a response to our
274 * connection id.
275 */
276 case TCPS_SYN_SENT:
277 if (si->si_did != cb->s_sid) {
278 spx_istat.notme++;
279 goto drop;
280 }
281 spxstat.spxs_connects++;
282 cb->s_did = si->si_sid;
283 cb->s_rack = si->si_ack;
284 cb->s_ralo = si->si_alo;
285 cb->s_dport = ipxp->ipxp_fport = si->si_sport;
286 cb->s_timer[SPXT_REXMT] = 0;
287 cb->s_flags |= SF_ACKNOW;
288 soisconnected(so);
289 cb->s_state = TCPS_ESTABLISHED;
290 /* Use roundtrip time of connection request for initial rtt */
291 if (cb->s_rtt) {
292 cb->s_srtt = cb->s_rtt << 3;
293 cb->s_rttvar = cb->s_rtt << 1;
294 SPXT_RANGESET(cb->s_rxtcur,
295 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
296 SPXTV_MIN, SPXTV_REXMTMAX);
297 cb->s_rtt = 0;
298 }
299 }
300 if (so->so_options & SO_DEBUG || traceallspxs)
301 spx_trace(SA_INPUT, (u_char)ostate, cb, &spx_savesi, 0);
302
303 m->m_len -= sizeof(struct ipx);
304 m->m_pkthdr.len -= sizeof(struct ipx);
305 m->m_data += sizeof(struct ipx);
306
307 if (spx_reass(cb, si)) {
308 m_freem(m);
309 }
310 if (cb->s_force || (cb->s_flags & (SF_ACKNOW|SF_WIN|SF_RXT)))
311 spx_output(cb, (struct mbuf *)NULL);
312 cb->s_flags &= ~(SF_WIN|SF_RXT);
313 return;
314
315 dropwithreset:
316 if (dropsocket)
317 soabort(so);
318 si->si_seq = ntohs(si->si_seq);
319 si->si_ack = ntohs(si->si_ack);
320 si->si_alo = ntohs(si->si_alo);
321 m_freem(dtom(si));
322 if (cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG || traceallspxs)
323 spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
324 return;
325
326 drop:
327 bad:
328 if (cb == 0 || cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG ||
329 traceallspxs)
330 spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
331 m_freem(m);
332 }
333
334 static int spxrexmtthresh = 3;
335
336 /*
337 * This is structurally similar to the tcp reassembly routine
338 * but its function is somewhat different: It merely queues
339 * packets up, and suppresses duplicates.
340 */
341 static int
342 spx_reass(cb, si)
343 register struct spxpcb *cb;
344 register struct spx *si;
345 {
346 register struct spx_q *q;
347 register struct mbuf *m;
348 register struct socket *so = cb->s_ipxpcb->ipxp_socket;
349 char packetp = cb->s_flags & SF_HI;
350 int incr;
351 char wakeup = 0;
352
353 if (si == SI(0))
354 goto present;
355 /*
356 * Update our news from them.
357 */
358 if (si->si_cc & SPX_SA)
359 cb->s_flags |= (spx_use_delack ? SF_DELACK : SF_ACKNOW);
360 if (SSEQ_GT(si->si_alo, cb->s_ralo))
361 cb->s_flags |= SF_WIN;
362 if (SSEQ_LEQ(si->si_ack, cb->s_rack)) {
363 if ((si->si_cc & SPX_SP) && cb->s_rack != (cb->s_smax + 1)) {
364 spxstat.spxs_rcvdupack++;
365 /*
366 * If this is a completely duplicate ack
367 * and other conditions hold, we assume
368 * a packet has been dropped and retransmit
369 * it exactly as in tcp_input().
370 */
371 if (si->si_ack != cb->s_rack ||
372 si->si_alo != cb->s_ralo)
373 cb->s_dupacks = 0;
374 else if (++cb->s_dupacks == spxrexmtthresh) {
375 u_short onxt = cb->s_snxt;
376 int cwnd = cb->s_cwnd;
377
378 cb->s_snxt = si->si_ack;
379 cb->s_cwnd = CUNIT;
380 cb->s_force = 1 + SPXT_REXMT;
381 spx_output(cb, (struct mbuf *)NULL);
382 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
383 cb->s_rtt = 0;
384 if (cwnd >= 4 * CUNIT)
385 cb->s_cwnd = cwnd / 2;
386 if (SSEQ_GT(onxt, cb->s_snxt))
387 cb->s_snxt = onxt;
388 return (1);
389 }
390 } else
391 cb->s_dupacks = 0;
392 goto update_window;
393 }
394 cb->s_dupacks = 0;
395 /*
396 * If our correspondent acknowledges data we haven't sent
397 * TCP would drop the packet after acking. We'll be a little
398 * more permissive
399 */
400 if (SSEQ_GT(si->si_ack, (cb->s_smax + 1))) {
401 spxstat.spxs_rcvacktoomuch++;
402 si->si_ack = cb->s_smax + 1;
403 }
404 spxstat.spxs_rcvackpack++;
405 /*
406 * If transmit timer is running and timed sequence
407 * number was acked, update smoothed round trip time.
408 * See discussion of algorithm in tcp_input.c
409 */
410 if (cb->s_rtt && SSEQ_GT(si->si_ack, cb->s_rtseq)) {
411 spxstat.spxs_rttupdated++;
412 if (cb->s_srtt != 0) {
413 register short delta;
414 delta = cb->s_rtt - (cb->s_srtt >> 3);
415 if ((cb->s_srtt += delta) <= 0)
416 cb->s_srtt = 1;
417 if (delta < 0)
418 delta = -delta;
419 delta -= (cb->s_rttvar >> 2);
420 if ((cb->s_rttvar += delta) <= 0)
421 cb->s_rttvar = 1;
422 } else {
423 /*
424 * No rtt measurement yet
425 */
426 cb->s_srtt = cb->s_rtt << 3;
427 cb->s_rttvar = cb->s_rtt << 1;
428 }
429 cb->s_rtt = 0;
430 cb->s_rxtshift = 0;
431 SPXT_RANGESET(cb->s_rxtcur,
432 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
433 SPXTV_MIN, SPXTV_REXMTMAX);
434 }
435 /*
436 * If all outstanding data is acked, stop retransmit
437 * timer and remember to restart (more output or persist).
438 * If there is more data to be acked, restart retransmit
439 * timer, using current (possibly backed-off) value;
440 */
441 if (si->si_ack == cb->s_smax + 1) {
442 cb->s_timer[SPXT_REXMT] = 0;
443 cb->s_flags |= SF_RXT;
444 } else if (cb->s_timer[SPXT_PERSIST] == 0)
445 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
446 /*
447 * When new data is acked, open the congestion window.
448 * If the window gives us less than ssthresh packets
449 * in flight, open exponentially (maxseg at a time).
450 * Otherwise open linearly (maxseg^2 / cwnd at a time).
451 */
452 incr = CUNIT;
453 if (cb->s_cwnd > cb->s_ssthresh)
454 incr = max(incr * incr / cb->s_cwnd, 1);
455 cb->s_cwnd = min(cb->s_cwnd + incr, cb->s_cwmx);
456 /*
457 * Trim Acked data from output queue.
458 */
459 while ((m = so->so_snd.sb_mb) != NULL) {
460 if (SSEQ_LT((mtod(m, struct spx *))->si_seq, si->si_ack))
461 sbdroprecord(&so->so_snd);
462 else
463 break;
464 }
465 sowwakeup(so);
466 cb->s_rack = si->si_ack;
467 update_window:
468 if (SSEQ_LT(cb->s_snxt, cb->s_rack))
469 cb->s_snxt = cb->s_rack;
470 if (SSEQ_LT(cb->s_swl1, si->si_seq) || ((cb->s_swl1 == si->si_seq &&
471 (SSEQ_LT(cb->s_swl2, si->si_ack))) ||
472 (cb->s_swl2 == si->si_ack && SSEQ_LT(cb->s_ralo, si->si_alo)))) {
473 /* keep track of pure window updates */
474 if ((si->si_cc & SPX_SP) && cb->s_swl2 == si->si_ack
475 && SSEQ_LT(cb->s_ralo, si->si_alo)) {
476 spxstat.spxs_rcvwinupd++;
477 spxstat.spxs_rcvdupack--;
478 }
479 cb->s_ralo = si->si_alo;
480 cb->s_swl1 = si->si_seq;
481 cb->s_swl2 = si->si_ack;
482 cb->s_swnd = (1 + si->si_alo - si->si_ack);
483 if (cb->s_swnd > cb->s_smxw)
484 cb->s_smxw = cb->s_swnd;
485 cb->s_flags |= SF_WIN;
486 }
487 /*
488 * If this packet number is higher than that which
489 * we have allocated refuse it, unless urgent
490 */
491 if (SSEQ_GT(si->si_seq, cb->s_alo)) {
492 if (si->si_cc & SPX_SP) {
493 spxstat.spxs_rcvwinprobe++;
494 return (1);
495 } else
496 spxstat.spxs_rcvpackafterwin++;
497 if (si->si_cc & SPX_OB) {
498 if (SSEQ_GT(si->si_seq, cb->s_alo + 60)) {
499 m_freem(dtom(si));
500 return (0);
501 } /* else queue this packet; */
502 } else {
503 /*register struct socket *so = cb->s_ipxpcb->ipxp_socket;
504 if (so->so_state && SS_NOFDREF) {
505 spx_close(cb);
506 } else
507 would crash system*/
508 spx_istat.notyet++;
509 m_freem(dtom(si));
510 return (0);
511 }
512 }
513 /*
514 * If this is a system packet, we don't need to
515 * queue it up, and won't update acknowledge #
516 */
517 if (si->si_cc & SPX_SP) {
518 return (1);
519 }
520 /*
521 * We have already seen this packet, so drop.
522 */
523 if (SSEQ_LT(si->si_seq, cb->s_ack)) {
524 spx_istat.bdreas++;
525 spxstat.spxs_rcvduppack++;
526 if (si->si_seq == cb->s_ack - 1)
527 spx_istat.lstdup++;
528 return (1);
529 }
530 /*
531 * Loop through all packets queued up to insert in
532 * appropriate sequence.
533 */
534 for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
535 if (si->si_seq == SI(q)->si_seq) {
536 spxstat.spxs_rcvduppack++;
537 return (1);
538 }
539 if (SSEQ_LT(si->si_seq, SI(q)->si_seq)) {
540 spxstat.spxs_rcvoopack++;
541 break;
542 }
543 }
544 insque(si, q->si_prev);
545 /*
546 * If this packet is urgent, inform process
547 */
548 if (si->si_cc & SPX_OB) {
549 cb->s_iobc = ((char *)si)[1 + sizeof(*si)];
550 sohasoutofband(so);
551 cb->s_oobflags |= SF_IOOB;
552 }
553 present:
554 #define SPINC sizeof(struct spxhdr)
555 /*
556 * Loop through all packets queued up to update acknowledge
557 * number, and present all acknowledged data to user;
558 * If in packet interface mode, show packet headers.
559 */
560 for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
561 if (SI(q)->si_seq == cb->s_ack) {
562 cb->s_ack++;
563 m = dtom(q);
564 if (SI(q)->si_cc & SPX_OB) {
565 cb->s_oobflags &= ~SF_IOOB;
566 if (so->so_rcv.sb_cc)
567 so->so_oobmark = so->so_rcv.sb_cc;
568 else
569 so->so_state |= SS_RCVATMARK;
570 }
571 q = q->si_prev;
572 remque(q->si_next);
573 wakeup = 1;
574 spxstat.spxs_rcvpack++;
575 #ifdef SF_NEWCALL
576 if (cb->s_flags2 & SF_NEWCALL) {
577 struct spxhdr *sp = mtod(m, struct spxhdr *);
578 u_char dt = sp->spx_dt;
579 spx_newchecks[4]++;
580 if (dt != cb->s_rhdr.spx_dt) {
581 struct mbuf *mm =
582 m_getclr(M_DONTWAIT, MT_CONTROL);
583 spx_newchecks[0]++;
584 if (mm != NULL) {
585 u_short *s =
586 mtod(mm, u_short *);
587 cb->s_rhdr.spx_dt = dt;
588 mm->m_len = 5; /*XXX*/
589 s[0] = 5;
590 s[1] = 1;
591 *(u_char *)(&s[2]) = dt;
592 sbappend(&so->so_rcv, mm);
593 }
594 }
595 if (sp->spx_cc & SPX_OB) {
596 MCHTYPE(m, MT_OOBDATA);
597 spx_newchecks[1]++;
598 so->so_oobmark = 0;
599 so->so_state &= ~SS_RCVATMARK;
600 }
601 if (packetp == 0) {
602 m->m_data += SPINC;
603 m->m_len -= SPINC;
604 m->m_pkthdr.len -= SPINC;
605 }
606 if ((sp->spx_cc & SPX_EM) || packetp) {
607 sbappendrecord(&so->so_rcv, m);
608 spx_newchecks[9]++;
609 } else
610 sbappend(&so->so_rcv, m);
611 } else
612 #endif
613 if (packetp) {
614 sbappendrecord(&so->so_rcv, m);
615 } else {
616 cb->s_rhdr = *mtod(m, struct spxhdr *);
617 m->m_data += SPINC;
618 m->m_len -= SPINC;
619 m->m_pkthdr.len -= SPINC;
620 sbappend(&so->so_rcv, m);
621 }
622 } else
623 break;
624 }
625 if (wakeup)
626 sorwakeup(so);
627 return (0);
628 }
629
630 void
631 spx_ctlinput(cmd, arg_as_sa, dummy)
632 int cmd;
633 struct sockaddr *arg_as_sa; /* XXX should be swapped with dummy */
634 void *dummy;
635 {
636 caddr_t arg = (/* XXX */ caddr_t)arg_as_sa;
637 struct ipx_addr *na;
638 struct sockaddr_ipx *sipx;
639
640 if (cmd < 0 || cmd > PRC_NCMDS)
641 return;
642
643 switch (cmd) {
644
645 case PRC_ROUTEDEAD:
646 return;
647
648 case PRC_IFDOWN:
649 case PRC_HOSTDEAD:
650 case PRC_HOSTUNREACH:
651 sipx = (struct sockaddr_ipx *)arg;
652 if (sipx->sipx_family != AF_IPX)
653 return;
654 na = &sipx->sipx_addr;
655 break;
656
657 default:
658 break;
659 }
660 }
661
662 #ifdef notdef
663 int
664 spx_fixmtu(ipxp)
665 register struct ipxpcb *ipxp;
666 {
667 register struct spxpcb *cb = (struct spxpcb *)(ipxp->ipxp_pcb);
668 register struct mbuf *m;
669 register struct spx *si;
670 struct ipx_errp *ep;
671 struct sockbuf *sb;
672 int badseq, len;
673 struct mbuf *firstbad, *m0;
674
675 if (cb != NULL) {
676 /*
677 * The notification that we have sent
678 * too much is bad news -- we will
679 * have to go through queued up so far
680 * splitting ones which are too big and
681 * reassigning sequence numbers and checksums.
682 * we should then retransmit all packets from
683 * one above the offending packet to the last one
684 * we had sent (or our allocation)
685 * then the offending one so that the any queued
686 * data at our destination will be discarded.
687 */
688 ep = (struct ipx_errp *)ipxp->ipxp_notify_param;
689 sb = &ipxp->ipxp_socket->so_snd;
690 cb->s_mtu = ep->ipx_err_param;
691 badseq = SI(&ep->ipx_err_ipx)->si_seq;
692 for (m = sb->sb_mb; m != NULL; m = m->m_act) {
693 si = mtod(m, struct spx *);
694 if (si->si_seq == badseq)
695 break;
696 }
697 if (m == NULL)
698 return;
699 firstbad = m;
700 /*for (;;) {*/
701 /* calculate length */
702 for (m0 = m, len = 0; m != NULL; m = m->m_next)
703 len += m->m_len;
704 if (len > cb->s_mtu) {
705 }
706 /* FINISH THIS
707 } */
708 }
709 }
710 #endif
711
712 static int
713 spx_output(cb, m0)
714 register struct spxpcb *cb;
715 struct mbuf *m0;
716 {
717 struct socket *so = cb->s_ipxpcb->ipxp_socket;
718 register struct mbuf *m;
719 register struct spx *si = (struct spx *)NULL;
720 register struct sockbuf *sb = &so->so_snd;
721 int len = 0, win, rcv_win;
722 short span, off, recordp = 0;
723 u_short alo;
724 int error = 0, sendalot;
725 #ifdef notdef
726 int idle;
727 #endif
728 struct mbuf *mprev;
729
730 if (m0 != NULL) {
731 int mtu = cb->s_mtu;
732 int datalen;
733 /*
734 * Make sure that packet isn't too big.
735 */
736 for (m = m0; m != NULL; m = m->m_next) {
737 mprev = m;
738 len += m->m_len;
739 if (m->m_flags & M_EOR)
740 recordp = 1;
741 }
742 datalen = (cb->s_flags & SF_HO) ?
743 len - sizeof(struct spxhdr) : len;
744 if (datalen > mtu) {
745 if (cb->s_flags & SF_PI) {
746 m_freem(m0);
747 return (EMSGSIZE);
748 } else {
749 int oldEM = cb->s_cc & SPX_EM;
750
751 cb->s_cc &= ~SPX_EM;
752 while (len > mtu) {
753 /*
754 * Here we are only being called
755 * from usrreq(), so it is OK to
756 * block.
757 */
758 m = m_copym(m0, 0, mtu, M_TRYWAIT);
759 if (cb->s_flags & SF_NEWCALL) {
760 struct mbuf *mm = m;
761 spx_newchecks[7]++;
762 while (mm != NULL) {
763 mm->m_flags &= ~M_EOR;
764 mm = mm->m_next;
765 }
766 }
767 error = spx_output(cb, m);
768 if (error) {
769 cb->s_cc |= oldEM;
770 m_freem(m0);
771 return (error);
772 }
773 m_adj(m0, mtu);
774 len -= mtu;
775 }
776 cb->s_cc |= oldEM;
777 }
778 }
779 /*
780 * Force length even, by adding a "garbage byte" if
781 * necessary.
782 */
783 if (len & 1) {
784 m = mprev;
785 if (M_TRAILINGSPACE(m) >= 1)
786 m->m_len++;
787 else {
788 struct mbuf *m1 = m_get(M_DONTWAIT, MT_DATA);
789
790 if (m1 == NULL) {
791 m_freem(m0);
792 return (ENOBUFS);
793 }
794 m1->m_len = 1;
795 *(mtod(m1, u_char *)) = 0;
796 m->m_next = m1;
797 }
798 }
799 m = m_gethdr(M_DONTWAIT, MT_HEADER);
800 if (m == NULL) {
801 m_freem(m0);
802 return (ENOBUFS);
803 }
804 /*
805 * Fill in mbuf with extended SP header
806 * and addresses and length put into network format.
807 */
808 MH_ALIGN(m, sizeof(struct spx));
809 m->m_len = sizeof(struct spx);
810 m->m_next = m0;
811 si = mtod(m, struct spx *);
812 si->si_i = *cb->s_ipx;
813 si->si_s = cb->s_shdr;
814 if ((cb->s_flags & SF_PI) && (cb->s_flags & SF_HO)) {
815 register struct spxhdr *sh;
816 if (m0->m_len < sizeof(*sh)) {
817 if((m0 = m_pullup(m0, sizeof(*sh))) == NULL) {
818 m_free(m);
819 m_freem(m0);
820 return (EINVAL);
821 }
822 m->m_next = m0;
823 }
824 sh = mtod(m0, struct spxhdr *);
825 si->si_dt = sh->spx_dt;
826 si->si_cc |= sh->spx_cc & SPX_EM;
827 m0->m_len -= sizeof(*sh);
828 m0->m_data += sizeof(*sh);
829 len -= sizeof(*sh);
830 }
831 len += sizeof(*si);
832 if ((cb->s_flags2 & SF_NEWCALL) && recordp) {
833 si->si_cc |= SPX_EM;
834 spx_newchecks[8]++;
835 }
836 if (cb->s_oobflags & SF_SOOB) {
837 /*
838 * Per jqj@cornell:
839 * make sure OB packets convey exactly 1 byte.
840 * If the packet is 1 byte or larger, we
841 * have already guaranted there to be at least
842 * one garbage byte for the checksum, and
843 * extra bytes shouldn't hurt!
844 */
845 if (len > sizeof(*si)) {
846 si->si_cc |= SPX_OB;
847 len = (1 + sizeof(*si));
848 }
849 }
850 si->si_len = htons((u_short)len);
851 m->m_pkthdr.len = ((len - 1) | 1) + 1;
852 /*
853 * queue stuff up for output
854 */
855 sbappendrecord(sb, m);
856 cb->s_seq++;
857 }
858 #ifdef notdef
859 idle = (cb->s_smax == (cb->s_rack - 1));
860 #endif
861 again:
862 sendalot = 0;
863 off = cb->s_snxt - cb->s_rack;
864 win = min(cb->s_swnd, (cb->s_cwnd / CUNIT));
865
866 /*
867 * If in persist timeout with window of 0, send a probe.
868 * Otherwise, if window is small but nonzero
869 * and timer expired, send what we can and go into
870 * transmit state.
871 */
872 if (cb->s_force == 1 + SPXT_PERSIST) {
873 if (win != 0) {
874 cb->s_timer[SPXT_PERSIST] = 0;
875 cb->s_rxtshift = 0;
876 }
877 }
878 span = cb->s_seq - cb->s_rack;
879 len = min(span, win) - off;
880
881 if (len < 0) {
882 /*
883 * Window shrank after we went into it.
884 * If window shrank to 0, cancel pending
885 * restransmission and pull s_snxt back
886 * to (closed) window. We will enter persist
887 * state below. If the widndow didn't close completely,
888 * just wait for an ACK.
889 */
890 len = 0;
891 if (win == 0) {
892 cb->s_timer[SPXT_REXMT] = 0;
893 cb->s_snxt = cb->s_rack;
894 }
895 }
896 if (len > 1)
897 sendalot = 1;
898 rcv_win = sbspace(&so->so_rcv);
899
900 /*
901 * Send if we owe peer an ACK.
902 */
903 if (cb->s_oobflags & SF_SOOB) {
904 /*
905 * must transmit this out of band packet
906 */
907 cb->s_oobflags &= ~ SF_SOOB;
908 sendalot = 1;
909 spxstat.spxs_sndurg++;
910 goto found;
911 }
912 if (cb->s_flags & SF_ACKNOW)
913 goto send;
914 if (cb->s_state < TCPS_ESTABLISHED)
915 goto send;
916 /*
917 * Silly window can't happen in spx.
918 * Code from tcp deleted.
919 */
920 if (len)
921 goto send;
922 /*
923 * Compare available window to amount of window
924 * known to peer (as advertised window less
925 * next expected input.) If the difference is at least two
926 * packets or at least 35% of the mximum possible window,
927 * then want to send a window update to peer.
928 */
929 if (rcv_win > 0) {
930 u_short delta = 1 + cb->s_alo - cb->s_ack;
931 int adv = rcv_win - (delta * cb->s_mtu);
932
933 if ((so->so_rcv.sb_cc == 0 && adv >= (2 * cb->s_mtu)) ||
934 (100 * adv / so->so_rcv.sb_hiwat >= 35)) {
935 spxstat.spxs_sndwinup++;
936 cb->s_flags |= SF_ACKNOW;
937 goto send;
938 }
939
940 }
941 /*
942 * Many comments from tcp_output.c are appropriate here
943 * including . . .
944 * If send window is too small, there is data to transmit, and no
945 * retransmit or persist is pending, then go to persist state.
946 * If nothing happens soon, send when timer expires:
947 * if window is nonzero, transmit what we can,
948 * otherwise send a probe.
949 */
950 if (so->so_snd.sb_cc && cb->s_timer[SPXT_REXMT] == 0 &&
951 cb->s_timer[SPXT_PERSIST] == 0) {
952 cb->s_rxtshift = 0;
953 spx_setpersist(cb);
954 }
955 /*
956 * No reason to send a packet, just return.
957 */
958 cb->s_outx = 1;
959 return (0);
960
961 send:
962 /*
963 * Find requested packet.
964 */
965 si = 0;
966 if (len > 0) {
967 cb->s_want = cb->s_snxt;
968 for (m = sb->sb_mb; m != NULL; m = m->m_act) {
969 si = mtod(m, struct spx *);
970 if (SSEQ_LEQ(cb->s_snxt, si->si_seq))
971 break;
972 }
973 found:
974 if (si != NULL) {
975 if (si->si_seq == cb->s_snxt)
976 cb->s_snxt++;
977 else
978 spxstat.spxs_sndvoid++, si = 0;
979 }
980 }
981 /*
982 * update window
983 */
984 if (rcv_win < 0)
985 rcv_win = 0;
986 alo = cb->s_ack - 1 + (rcv_win / ((short)cb->s_mtu));
987 if (SSEQ_LT(alo, cb->s_alo))
988 alo = cb->s_alo;
989
990 if (si != NULL) {
991 /*
992 * must make a copy of this packet for
993 * ipx_output to monkey with
994 */
995 m = m_copy(dtom(si), 0, (int)M_COPYALL);
996 if (m == NULL) {
997 return (ENOBUFS);
998 }
999 si = mtod(m, struct spx *);
1000 if (SSEQ_LT(si->si_seq, cb->s_smax))
1001 spxstat.spxs_sndrexmitpack++;
1002 else
1003 spxstat.spxs_sndpack++;
1004 } else if (cb->s_force || cb->s_flags & SF_ACKNOW) {
1005 /*
1006 * Must send an acknowledgement or a probe
1007 */
1008 if (cb->s_force)
1009 spxstat.spxs_sndprobe++;
1010 if (cb->s_flags & SF_ACKNOW)
1011 spxstat.spxs_sndacks++;
1012 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1013 if (m == NULL)
1014 return (ENOBUFS);
1015 /*
1016 * Fill in mbuf with extended SP header
1017 * and addresses and length put into network format.
1018 */
1019 MH_ALIGN(m, sizeof(struct spx));
1020 m->m_len = sizeof(*si);
1021 m->m_pkthdr.len = sizeof(*si);
1022 si = mtod(m, struct spx *);
1023 si->si_i = *cb->s_ipx;
1024 si->si_s = cb->s_shdr;
1025 si->si_seq = cb->s_smax + 1;
1026 si->si_len = htons(sizeof(*si));
1027 si->si_cc |= SPX_SP;
1028 } else {
1029 cb->s_outx = 3;
1030 if (so->so_options & SO_DEBUG || traceallspxs)
1031 spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1032 return (0);
1033 }
1034 /*
1035 * Stuff checksum and output datagram.
1036 */
1037 if ((si->si_cc & SPX_SP) == 0) {
1038 if (cb->s_force != (1 + SPXT_PERSIST) ||
1039 cb->s_timer[SPXT_PERSIST] == 0) {
1040 /*
1041 * If this is a new packet and we are not currently
1042 * timing anything, time this one.
1043 */
1044 if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1045 cb->s_smax = si->si_seq;
1046 if (cb->s_rtt == 0) {
1047 spxstat.spxs_segstimed++;
1048 cb->s_rtseq = si->si_seq;
1049 cb->s_rtt = 1;
1050 }
1051 }
1052 /*
1053 * Set rexmt timer if not currently set,
1054 * Initial value for retransmit timer is smoothed
1055 * round-trip time + 2 * round-trip time variance.
1056 * Initialize shift counter which is used for backoff
1057 * of retransmit time.
1058 */
1059 if (cb->s_timer[SPXT_REXMT] == 0 &&
1060 cb->s_snxt != cb->s_rack) {
1061 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1062 if (cb->s_timer[SPXT_PERSIST]) {
1063 cb->s_timer[SPXT_PERSIST] = 0;
1064 cb->s_rxtshift = 0;
1065 }
1066 }
1067 } else if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1068 cb->s_smax = si->si_seq;
1069 }
1070 } else if (cb->s_state < TCPS_ESTABLISHED) {
1071 if (cb->s_rtt == 0)
1072 cb->s_rtt = 1; /* Time initial handshake */
1073 if (cb->s_timer[SPXT_REXMT] == 0)
1074 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1075 }
1076 {
1077 /*
1078 * Do not request acks when we ack their data packets or
1079 * when we do a gratuitous window update.
1080 */
1081 if (((si->si_cc & SPX_SP) == 0) || cb->s_force)
1082 si->si_cc |= SPX_SA;
1083 si->si_seq = htons(si->si_seq);
1084 si->si_alo = htons(alo);
1085 si->si_ack = htons(cb->s_ack);
1086
1087 if (ipxcksum) {
1088 si->si_sum = ipx_cksum(m, ntohs(si->si_len));
1089 } else
1090 si->si_sum = 0xffff;
1091
1092 cb->s_outx = 4;
1093 if (so->so_options & SO_DEBUG || traceallspxs)
1094 spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1095
1096 if (so->so_options & SO_DONTROUTE)
1097 error = ipx_outputfl(m, (struct route *)NULL, IPX_ROUTETOIF);
1098 else
1099 error = ipx_outputfl(m, &cb->s_ipxpcb->ipxp_route, 0);
1100 }
1101 if (error) {
1102 return (error);
1103 }
1104 spxstat.spxs_sndtotal++;
1105 /*
1106 * Data sent (as far as we can tell).
1107 * If this advertises a larger window than any other segment,
1108 * then remember the size of the advertized window.
1109 * Any pending ACK has now been sent.
1110 */
1111 cb->s_force = 0;
1112 cb->s_flags &= ~(SF_ACKNOW|SF_DELACK);
1113 if (SSEQ_GT(alo, cb->s_alo))
1114 cb->s_alo = alo;
1115 if (sendalot)
1116 goto again;
1117 cb->s_outx = 5;
1118 return (0);
1119 }
1120
1121 static int spx_do_persist_panics = 0;
1122
1123 static void
1124 spx_setpersist(cb)
1125 register struct spxpcb *cb;
1126 {
1127 register int t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1128
1129 if (cb->s_timer[SPXT_REXMT] && spx_do_persist_panics)
1130 panic("spx_output REXMT");
1131 /*
1132 * Start/restart persistance timer.
1133 */
1134 SPXT_RANGESET(cb->s_timer[SPXT_PERSIST],
1135 t*spx_backoff[cb->s_rxtshift],
1136 SPXTV_PERSMIN, SPXTV_PERSMAX);
1137 if (cb->s_rxtshift < SPX_MAXRXTSHIFT)
1138 cb->s_rxtshift++;
1139 }
1140
1141 int
1142 spx_ctloutput(so, sopt)
1143 struct socket *so;
1144 struct sockopt *sopt;
1145 {
1146 struct ipxpcb *ipxp = sotoipxpcb(so);
1147 register struct spxpcb *cb;
1148 int mask, error;
1149 short soptval;
1150 u_short usoptval;
1151 int optval;
1152
1153 error = 0;
1154
1155 if (sopt->sopt_level != IPXPROTO_SPX) {
1156 /* This will have to be changed when we do more general
1157 stacking of protocols */
1158 return (ipx_ctloutput(so, sopt));
1159 }
1160 if (ipxp == NULL)
1161 return (EINVAL);
1162 else
1163 cb = ipxtospxpcb(ipxp);
1164
1165 switch (sopt->sopt_dir) {
1166 case SOPT_GET:
1167 switch (sopt->sopt_name) {
1168 case SO_HEADERS_ON_INPUT:
1169 mask = SF_HI;
1170 goto get_flags;
1171
1172 case SO_HEADERS_ON_OUTPUT:
1173 mask = SF_HO;
1174 get_flags:
1175 soptval = cb->s_flags & mask;
1176 error = sooptcopyout(sopt, &soptval, sizeof soptval);
1177 break;
1178
1179 case SO_MTU:
1180 usoptval = cb->s_mtu;
1181 error = sooptcopyout(sopt, &usoptval, sizeof usoptval);
1182 break;
1183
1184 case SO_LAST_HEADER:
1185 error = sooptcopyout(sopt, &cb->s_rhdr,
1186 sizeof cb->s_rhdr);
1187 break;
1188
1189 case SO_DEFAULT_HEADERS:
1190 error = sooptcopyout(sopt, &cb->s_shdr,
1191 sizeof cb->s_shdr);
1192 break;
1193
1194 default:
1195 error = ENOPROTOOPT;
1196 }
1197 break;
1198
1199 case SOPT_SET:
1200 switch (sopt->sopt_name) {
1201 /* XXX why are these shorts on get and ints on set?
1202 that doesn't make any sense... */
1203 case SO_HEADERS_ON_INPUT:
1204 mask = SF_HI;
1205 goto set_head;
1206
1207 case SO_HEADERS_ON_OUTPUT:
1208 mask = SF_HO;
1209 set_head:
1210 error = sooptcopyin(sopt, &optval, sizeof optval,
1211 sizeof optval);
1212 if (error)
1213 break;
1214
1215 if (cb->s_flags & SF_PI) {
1216 if (optval)
1217 cb->s_flags |= mask;
1218 else
1219 cb->s_flags &= ~mask;
1220 } else error = EINVAL;
1221 break;
1222
1223 case SO_MTU:
1224 error = sooptcopyin(sopt, &usoptval, sizeof usoptval,
1225 sizeof usoptval);
1226 if (error)
1227 break;
1228 cb->s_mtu = usoptval;
1229 break;
1230
1231 #ifdef SF_NEWCALL
1232 case SO_NEWCALL:
1233 error = sooptcopyin(sopt, &optval, sizeof optval,
1234 sizeof optval);
1235 if (error)
1236 break;
1237 if (optval) {
1238 cb->s_flags2 |= SF_NEWCALL;
1239 spx_newchecks[5]++;
1240 } else {
1241 cb->s_flags2 &= ~SF_NEWCALL;
1242 spx_newchecks[6]++;
1243 }
1244 break;
1245 #endif
1246
1247 case SO_DEFAULT_HEADERS:
1248 {
1249 struct spxhdr sp;
1250
1251 error = sooptcopyin(sopt, &sp, sizeof sp,
1252 sizeof sp);
1253 if (error)
1254 break;
1255 cb->s_dt = sp.spx_dt;
1256 cb->s_cc = sp.spx_cc & SPX_EM;
1257 }
1258 break;
1259
1260 default:
1261 error = ENOPROTOOPT;
1262 }
1263 break;
1264 }
1265 return (error);
1266 }
1267
1268 static int
1269 spx_usr_abort(so)
1270 struct socket *so;
1271 {
1272 int s;
1273 struct ipxpcb *ipxp;
1274 struct spxpcb *cb;
1275
1276 ipxp = sotoipxpcb(so);
1277 cb = ipxtospxpcb(ipxp);
1278
1279 s = splnet();
1280 spx_drop(cb, ECONNABORTED);
1281 splx(s);
1282 return (0);
1283 }
1284
1285 /*
1286 * Accept a connection. Essentially all the work is
1287 * done at higher levels; just return the address
1288 * of the peer, storing through addr.
1289 */
1290 static int
1291 spx_accept(so, nam)
1292 struct socket *so;
1293 struct sockaddr **nam;
1294 {
1295 struct ipxpcb *ipxp;
1296 struct sockaddr_ipx *sipx, ssipx;
1297
1298 ipxp = sotoipxpcb(so);
1299 sipx = &ssipx;
1300 bzero(sipx, sizeof *sipx);
1301 sipx->sipx_len = sizeof *sipx;
1302 sipx->sipx_family = AF_IPX;
1303 sipx->sipx_addr = ipxp->ipxp_faddr;
1304 *nam = dup_sockaddr((struct sockaddr *)sipx, 0);
1305 return (0);
1306 }
1307
1308 static int
1309 spx_attach(so, proto, td)
1310 struct socket *so;
1311 int proto;
1312 struct thread *td;
1313 {
1314 int error;
1315 int s;
1316 struct ipxpcb *ipxp;
1317 struct spxpcb *cb;
1318 struct mbuf *mm;
1319 struct sockbuf *sb;
1320
1321 ipxp = sotoipxpcb(so);
1322 cb = ipxtospxpcb(ipxp);
1323
1324 if (ipxp != NULL)
1325 return (EISCONN);
1326 s = splnet();
1327 error = ipx_pcballoc(so, &ipxpcb, td);
1328 if (error)
1329 goto spx_attach_end;
1330 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1331 error = soreserve(so, (u_long) 3072, (u_long) 3072);
1332 if (error)
1333 goto spx_attach_end;
1334 }
1335 ipxp = sotoipxpcb(so);
1336
1337 MALLOC(cb, struct spxpcb *, sizeof *cb, M_PCB, M_NOWAIT | M_ZERO);
1338
1339 if (cb == NULL) {
1340 error = ENOBUFS;
1341 goto spx_attach_end;
1342 }
1343 sb = &so->so_snd;
1344
1345 mm = m_getclr(M_DONTWAIT, MT_HEADER);
1346 if (mm == NULL) {
1347 FREE(cb, M_PCB);
1348 error = ENOBUFS;
1349 goto spx_attach_end;
1350 }
1351 cb->s_ipx = mtod(mm, struct ipx *);
1352 cb->s_state = TCPS_LISTEN;
1353 cb->s_smax = -1;
1354 cb->s_swl1 = -1;
1355 cb->s_q.si_next = cb->s_q.si_prev = &cb->s_q;
1356 cb->s_ipxpcb = ipxp;
1357 cb->s_mtu = 576 - sizeof(struct spx);
1358 cb->s_cwnd = sbspace(sb) * CUNIT / cb->s_mtu;
1359 cb->s_ssthresh = cb->s_cwnd;
1360 cb->s_cwmx = sbspace(sb) * CUNIT / (2 * sizeof(struct spx));
1361 /* Above is recomputed when connecting to account
1362 for changed buffering or mtu's */
1363 cb->s_rtt = SPXTV_SRTTBASE;
1364 cb->s_rttvar = SPXTV_SRTTDFLT << 2;
1365 SPXT_RANGESET(cb->s_rxtcur,
1366 ((SPXTV_SRTTBASE >> 2) + (SPXTV_SRTTDFLT << 2)) >> 1,
1367 SPXTV_MIN, SPXTV_REXMTMAX);
1368 ipxp->ipxp_pcb = (caddr_t)cb;
1369 spx_attach_end:
1370 splx(s);
1371 return (error);
1372 }
1373
1374 static int
1375 spx_bind(so, nam, td)
1376 struct socket *so;
1377 struct sockaddr *nam;
1378 struct thread *td;
1379 {
1380 struct ipxpcb *ipxp;
1381
1382 ipxp = sotoipxpcb(so);
1383
1384 return (ipx_pcbbind(ipxp, nam, td));
1385 }
1386
1387 /*
1388 * Initiate connection to peer.
1389 * Enter SYN_SENT state, and mark socket as connecting.
1390 * Start keep-alive timer, setup prototype header,
1391 * Send initial system packet requesting connection.
1392 */
1393 static int
1394 spx_connect(so, nam, td)
1395 struct socket *so;
1396 struct sockaddr *nam;
1397 struct thread *td;
1398 {
1399 int error;
1400 int s;
1401 struct ipxpcb *ipxp;
1402 struct spxpcb *cb;
1403
1404 ipxp = sotoipxpcb(so);
1405 cb = ipxtospxpcb(ipxp);
1406
1407 s = splnet();
1408 if (ipxp->ipxp_lport == 0) {
1409 error = ipx_pcbbind(ipxp, (struct sockaddr *)NULL, td);
1410 if (error)
1411 goto spx_connect_end;
1412 }
1413 error = ipx_pcbconnect(ipxp, nam, td);
1414 if (error)
1415 goto spx_connect_end;
1416 soisconnecting(so);
1417 spxstat.spxs_connattempt++;
1418 cb->s_state = TCPS_SYN_SENT;
1419 cb->s_did = 0;
1420 spx_template(cb);
1421 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1422 cb->s_force = 1 + SPXTV_KEEP;
1423 /*
1424 * Other party is required to respond to
1425 * the port I send from, but he is not
1426 * required to answer from where I am sending to,
1427 * so allow wildcarding.
1428 * original port I am sending to is still saved in
1429 * cb->s_dport.
1430 */
1431 ipxp->ipxp_fport = 0;
1432 error = spx_output(cb, (struct mbuf *)NULL);
1433 spx_connect_end:
1434 splx(s);
1435 return (error);
1436 }
1437
1438 static int
1439 spx_detach(so)
1440 struct socket *so;
1441 {
1442 int s;
1443 struct ipxpcb *ipxp;
1444 struct spxpcb *cb;
1445
1446 ipxp = sotoipxpcb(so);
1447 cb = ipxtospxpcb(ipxp);
1448
1449 if (ipxp == NULL)
1450 return (ENOTCONN);
1451 s = splnet();
1452 if (cb->s_state > TCPS_LISTEN)
1453 spx_disconnect(cb);
1454 else
1455 spx_close(cb);
1456 splx(s);
1457 return (0);
1458 }
1459
1460 /*
1461 * We may decide later to implement connection closing
1462 * handshaking at the spx level optionally.
1463 * here is the hook to do it:
1464 */
1465 static int
1466 spx_usr_disconnect(so)
1467 struct socket *so;
1468 {
1469 int s;
1470 struct ipxpcb *ipxp;
1471 struct spxpcb *cb;
1472
1473 ipxp = sotoipxpcb(so);
1474 cb = ipxtospxpcb(ipxp);
1475
1476 s = splnet();
1477 spx_disconnect(cb);
1478 splx(s);
1479 return (0);
1480 }
1481
1482 static int
1483 spx_listen(so, td)
1484 struct socket *so;
1485 struct thread *td;
1486 {
1487 int error;
1488 struct ipxpcb *ipxp;
1489 struct spxpcb *cb;
1490
1491 error = 0;
1492 ipxp = sotoipxpcb(so);
1493 cb = ipxtospxpcb(ipxp);
1494
1495 if (ipxp->ipxp_lport == 0)
1496 error = ipx_pcbbind(ipxp, (struct sockaddr *)NULL, td);
1497 if (error == 0)
1498 cb->s_state = TCPS_LISTEN;
1499 return (error);
1500 }
1501
1502 /*
1503 * After a receive, possibly send acknowledgment
1504 * updating allocation.
1505 */
1506 static int
1507 spx_rcvd(so, flags)
1508 struct socket *so;
1509 int flags;
1510 {
1511 int s;
1512 struct ipxpcb *ipxp;
1513 struct spxpcb *cb;
1514
1515 ipxp = sotoipxpcb(so);
1516 cb = ipxtospxpcb(ipxp);
1517
1518 s = splnet();
1519 cb->s_flags |= SF_RVD;
1520 spx_output(cb, (struct mbuf *)NULL);
1521 cb->s_flags &= ~SF_RVD;
1522 splx(s);
1523 return (0);
1524 }
1525
1526 static int
1527 spx_rcvoob(so, m, flags)
1528 struct socket *so;
1529 struct mbuf *m;
1530 int flags;
1531 {
1532 struct ipxpcb *ipxp;
1533 struct spxpcb *cb;
1534
1535 ipxp = sotoipxpcb(so);
1536 cb = ipxtospxpcb(ipxp);
1537
1538 if ((cb->s_oobflags & SF_IOOB) || so->so_oobmark ||
1539 (so->so_state & SS_RCVATMARK)) {
1540 m->m_len = 1;
1541 *mtod(m, caddr_t) = cb->s_iobc;
1542 return (0);
1543 }
1544 return (EINVAL);
1545 }
1546
1547 static int
1548 spx_send(so, flags, m, addr, controlp, td)
1549 struct socket *so;
1550 int flags;
1551 struct mbuf *m;
1552 struct sockaddr *addr;
1553 struct mbuf *controlp;
1554 struct thread *td;
1555 {
1556 int error;
1557 int s;
1558 struct ipxpcb *ipxp;
1559 struct spxpcb *cb;
1560
1561 error = 0;
1562 ipxp = sotoipxpcb(so);
1563 cb = ipxtospxpcb(ipxp);
1564
1565 s = splnet();
1566 if (flags & PRUS_OOB) {
1567 if (sbspace(&so->so_snd) < -512) {
1568 error = ENOBUFS;
1569 goto spx_send_end;
1570 }
1571 cb->s_oobflags |= SF_SOOB;
1572 }
1573 if (controlp != NULL) {
1574 u_short *p = mtod(controlp, u_short *);
1575 spx_newchecks[2]++;
1576 if ((p[0] == 5) && (p[1] == 1)) { /* XXXX, for testing */
1577 cb->s_shdr.spx_dt = *(u_char *)(&p[2]);
1578 spx_newchecks[3]++;
1579 }
1580 m_freem(controlp);
1581 }
1582 controlp = NULL;
1583 error = spx_output(cb, m);
1584 m = NULL;
1585 spx_send_end:
1586 if (controlp != NULL)
1587 m_freem(controlp);
1588 if (m != NULL)
1589 m_freem(m);
1590 splx(s);
1591 return (error);
1592 }
1593
1594 static int
1595 spx_shutdown(so)
1596 struct socket *so;
1597 {
1598 int error;
1599 int s;
1600 struct ipxpcb *ipxp;
1601 struct spxpcb *cb;
1602
1603 error = 0;
1604 ipxp = sotoipxpcb(so);
1605 cb = ipxtospxpcb(ipxp);
1606
1607 s = splnet();
1608 socantsendmore(so);
1609 cb = spx_usrclosed(cb);
1610 if (cb != NULL)
1611 error = spx_output(cb, (struct mbuf *)NULL);
1612 splx(s);
1613 return (error);
1614 }
1615
1616 static int
1617 spx_sp_attach(so, proto, td)
1618 struct socket *so;
1619 int proto;
1620 struct thread *td;
1621 {
1622 int error;
1623 struct ipxpcb *ipxp;
1624
1625 error = spx_attach(so, proto, td);
1626 if (error == 0) {
1627 ipxp = sotoipxpcb(so);
1628 ((struct spxpcb *)ipxp->ipxp_pcb)->s_flags |=
1629 (SF_HI | SF_HO | SF_PI);
1630 }
1631 return (error);
1632 }
1633
1634 /*
1635 * Create template to be used to send spx packets on a connection.
1636 * Called after host entry created, fills
1637 * in a skeletal spx header (choosing connection id),
1638 * minimizing the amount of work necessary when the connection is used.
1639 */
1640 static void
1641 spx_template(cb)
1642 register struct spxpcb *cb;
1643 {
1644 register struct ipxpcb *ipxp = cb->s_ipxpcb;
1645 register struct ipx *ipx = cb->s_ipx;
1646 register struct sockbuf *sb = &(ipxp->ipxp_socket->so_snd);
1647
1648 ipx->ipx_pt = IPXPROTO_SPX;
1649 ipx->ipx_sna = ipxp->ipxp_laddr;
1650 ipx->ipx_dna = ipxp->ipxp_faddr;
1651 cb->s_sid = htons(spx_iss);
1652 spx_iss += SPX_ISSINCR/2;
1653 cb->s_alo = 1;
1654 cb->s_cwnd = (sbspace(sb) * CUNIT) / cb->s_mtu;
1655 cb->s_ssthresh = cb->s_cwnd; /* Try to expand fast to full complement
1656 of large packets */
1657 cb->s_cwmx = (sbspace(sb) * CUNIT) / (2 * sizeof(struct spx));
1658 cb->s_cwmx = max(cb->s_cwmx, cb->s_cwnd);
1659 /* But allow for lots of little packets as well */
1660 }
1661
1662 /*
1663 * Close a SPIP control block:
1664 * discard spx control block itself
1665 * discard ipx protocol control block
1666 * wake up any sleepers
1667 */
1668 static struct spxpcb *
1669 spx_close(cb)
1670 register struct spxpcb *cb;
1671 {
1672 register struct spx_q *s;
1673 struct ipxpcb *ipxp = cb->s_ipxpcb;
1674 struct socket *so = ipxp->ipxp_socket;
1675 register struct mbuf *m;
1676
1677 s = cb->s_q.si_next;
1678 while (s != &(cb->s_q)) {
1679 s = s->si_next;
1680 m = dtom(s->si_prev);
1681 remque(s->si_prev);
1682 m_freem(m);
1683 }
1684 m_free(dtom(cb->s_ipx));
1685 FREE(cb, M_PCB);
1686 ipxp->ipxp_pcb = 0;
1687 soisdisconnected(so);
1688 ipx_pcbdetach(ipxp);
1689 spxstat.spxs_closed++;
1690 return ((struct spxpcb *)NULL);
1691 }
1692
1693 /*
1694 * Someday we may do level 3 handshaking
1695 * to close a connection or send a xerox style error.
1696 * For now, just close.
1697 */
1698 static struct spxpcb *
1699 spx_usrclosed(cb)
1700 register struct spxpcb *cb;
1701 {
1702 return (spx_close(cb));
1703 }
1704
1705 static struct spxpcb *
1706 spx_disconnect(cb)
1707 register struct spxpcb *cb;
1708 {
1709 return (spx_close(cb));
1710 }
1711
1712 /*
1713 * Drop connection, reporting
1714 * the specified error.
1715 */
1716 static struct spxpcb *
1717 spx_drop(cb, errno)
1718 register struct spxpcb *cb;
1719 int errno;
1720 {
1721 struct socket *so = cb->s_ipxpcb->ipxp_socket;
1722
1723 /*
1724 * someday, in the xerox world
1725 * we will generate error protocol packets
1726 * announcing that the socket has gone away.
1727 */
1728 if (TCPS_HAVERCVDSYN(cb->s_state)) {
1729 spxstat.spxs_drops++;
1730 cb->s_state = TCPS_CLOSED;
1731 /*tcp_output(cb);*/
1732 } else
1733 spxstat.spxs_conndrops++;
1734 so->so_error = errno;
1735 return (spx_close(cb));
1736 }
1737
1738 /*
1739 * Fast timeout routine for processing delayed acks
1740 */
1741 void
1742 spx_fasttimo()
1743 {
1744 register struct ipxpcb *ipxp;
1745 register struct spxpcb *cb;
1746 int s = splnet();
1747
1748 ipxp = ipxpcb.ipxp_next;
1749 if (ipxp != NULL)
1750 for (; ipxp != &ipxpcb; ipxp = ipxp->ipxp_next)
1751 if ((cb = (struct spxpcb *)ipxp->ipxp_pcb) != NULL &&
1752 (cb->s_flags & SF_DELACK)) {
1753 cb->s_flags &= ~SF_DELACK;
1754 cb->s_flags |= SF_ACKNOW;
1755 spxstat.spxs_delack++;
1756 spx_output(cb, (struct mbuf *)NULL);
1757 }
1758 splx(s);
1759 }
1760
1761 /*
1762 * spx protocol timeout routine called every 500 ms.
1763 * Updates the timers in all active pcb's and
1764 * causes finite state machine actions if timers expire.
1765 */
1766 void
1767 spx_slowtimo()
1768 {
1769 register struct ipxpcb *ip, *ipnxt;
1770 register struct spxpcb *cb;
1771 int s = splnet();
1772 register int i;
1773
1774 /*
1775 * Search through tcb's and update active timers.
1776 */
1777 ip = ipxpcb.ipxp_next;
1778 if (ip == NULL) {
1779 splx(s);
1780 return;
1781 }
1782 while (ip != &ipxpcb) {
1783 cb = ipxtospxpcb(ip);
1784 ipnxt = ip->ipxp_next;
1785 if (cb == NULL)
1786 goto tpgone;
1787 for (i = 0; i < SPXT_NTIMERS; i++) {
1788 if (cb->s_timer[i] && --cb->s_timer[i] == 0) {
1789 spx_timers(cb, i);
1790 if (ipnxt->ipxp_prev != ip)
1791 goto tpgone;
1792 }
1793 }
1794 cb->s_idle++;
1795 if (cb->s_rtt)
1796 cb->s_rtt++;
1797 tpgone:
1798 ip = ipnxt;
1799 }
1800 spx_iss += SPX_ISSINCR/PR_SLOWHZ; /* increment iss */
1801 splx(s);
1802 }
1803
1804 /*
1805 * SPX timer processing.
1806 */
1807 static struct spxpcb *
1808 spx_timers(cb, timer)
1809 register struct spxpcb *cb;
1810 int timer;
1811 {
1812 long rexmt;
1813 int win;
1814
1815 cb->s_force = 1 + timer;
1816 switch (timer) {
1817
1818 /*
1819 * 2 MSL timeout in shutdown went off. TCP deletes connection
1820 * control block.
1821 */
1822 case SPXT_2MSL:
1823 printf("spx: SPXT_2MSL went off for no reason\n");
1824 cb->s_timer[timer] = 0;
1825 break;
1826
1827 /*
1828 * Retransmission timer went off. Message has not
1829 * been acked within retransmit interval. Back off
1830 * to a longer retransmit interval and retransmit one packet.
1831 */
1832 case SPXT_REXMT:
1833 if (++cb->s_rxtshift > SPX_MAXRXTSHIFT) {
1834 cb->s_rxtshift = SPX_MAXRXTSHIFT;
1835 spxstat.spxs_timeoutdrop++;
1836 cb = spx_drop(cb, ETIMEDOUT);
1837 break;
1838 }
1839 spxstat.spxs_rexmttimeo++;
1840 rexmt = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1841 rexmt *= spx_backoff[cb->s_rxtshift];
1842 SPXT_RANGESET(cb->s_rxtcur, rexmt, SPXTV_MIN, SPXTV_REXMTMAX);
1843 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1844 /*
1845 * If we have backed off fairly far, our srtt
1846 * estimate is probably bogus. Clobber it
1847 * so we'll take the next rtt measurement as our srtt;
1848 * move the current srtt into rttvar to keep the current
1849 * retransmit times until then.
1850 */
1851 if (cb->s_rxtshift > SPX_MAXRXTSHIFT / 4 ) {
1852 cb->s_rttvar += (cb->s_srtt >> 2);
1853 cb->s_srtt = 0;
1854 }
1855 cb->s_snxt = cb->s_rack;
1856 /*
1857 * If timing a packet, stop the timer.
1858 */
1859 cb->s_rtt = 0;
1860 /*
1861 * See very long discussion in tcp_timer.c about congestion
1862 * window and sstrhesh
1863 */
1864 win = min(cb->s_swnd, (cb->s_cwnd/CUNIT)) / 2;
1865 if (win < 2)
1866 win = 2;
1867 cb->s_cwnd = CUNIT;
1868 cb->s_ssthresh = win * CUNIT;
1869 spx_output(cb, (struct mbuf *)NULL);
1870 break;
1871
1872 /*
1873 * Persistance timer into zero window.
1874 * Force a probe to be sent.
1875 */
1876 case SPXT_PERSIST:
1877 spxstat.spxs_persisttimeo++;
1878 spx_setpersist(cb);
1879 spx_output(cb, (struct mbuf *)NULL);
1880 break;
1881
1882 /*
1883 * Keep-alive timer went off; send something
1884 * or drop connection if idle for too long.
1885 */
1886 case SPXT_KEEP:
1887 spxstat.spxs_keeptimeo++;
1888 if (cb->s_state < TCPS_ESTABLISHED)
1889 goto dropit;
1890 if (cb->s_ipxpcb->ipxp_socket->so_options & SO_KEEPALIVE) {
1891 if (cb->s_idle >= SPXTV_MAXIDLE)
1892 goto dropit;
1893 spxstat.spxs_keepprobe++;
1894 spx_output(cb, (struct mbuf *)NULL);
1895 } else
1896 cb->s_idle = 0;
1897 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1898 break;
1899 dropit:
1900 spxstat.spxs_keepdrops++;
1901 cb = spx_drop(cb, ETIMEDOUT);
1902 break;
1903 }
1904 return (cb);
1905 }
Cache object: 2e8ab24e6a97cf0dcfa2600f6d56c98e
|