1 /*-
2 * Copyright (c) 2004-2005 Robert N. M. Watson
3 * Copyright (c) 1995, Mike Mitchell
4 * Copyright (c) 1984, 1985, 1986, 1987, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed by the University of
18 * California, Berkeley and its contributors.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * @(#)spx_usrreq.h
36 */
37
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD: src/sys/netipx/spx_usrreq.c,v 1.46.2.15 2005/04/01 15:04:19 rwatson Exp $");
40
41 #include <sys/param.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/mutex.h>
46 #include <sys/proc.h>
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sx.h>
52 #include <sys/systm.h>
53
54 #include <net/route.h>
55 #include <netinet/tcp_fsm.h>
56
57 #include <netipx/ipx.h>
58 #include <netipx/ipx_pcb.h>
59 #include <netipx/ipx_var.h>
60 #include <netipx/spx.h>
61 #include <netipx/spx_debug.h>
62 #include <netipx/spx_timer.h>
63 #include <netipx/spx_var.h>
64
65 /*
66 * SPX protocol implementation.
67 */
68 static u_short spx_iss;
69 static u_short spx_newchecks[50];
70 static int spx_hardnosed;
71 static int spx_use_delack = 0;
72 static int traceallspxs = 0;
73 static struct spx spx_savesi;
74 static struct spx_istat spx_istat;
75
76 /* Following was struct spxstat spxstat; */
77 #ifndef spxstat
78 #define spxstat spx_istat.newstats
79 #endif
80
81 static const int spx_backoff[SPX_MAXRXTSHIFT+1] =
82 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
83
84 static void spx_close(struct spxpcb *cb);
85 static void spx_disconnect(struct spxpcb *cb);
86 static void spx_drop(struct spxpcb *cb, int errno);
87 static int spx_output(struct spxpcb *cb, struct mbuf *m0);
88 static int spx_reass(struct spxpcb *cb, struct spx *si);
89 static void spx_setpersist(struct spxpcb *cb);
90 static void spx_template(struct spxpcb *cb);
91 static struct spxpcb *spx_timers(struct spxpcb *cb, int timer);
92 static void spx_usrclosed(struct spxpcb *cb);
93
94 static int spx_usr_abort(struct socket *so);
95 static int spx_accept(struct socket *so, struct sockaddr **nam);
96 static int spx_attach(struct socket *so, int proto, struct thread *td);
97 static int spx_bind(struct socket *so, struct sockaddr *nam, struct thread *td);
98 static int spx_connect(struct socket *so, struct sockaddr *nam,
99 struct thread *td);
100 static int spx_detach(struct socket *so);
101 static int spx_usr_disconnect(struct socket *so);
102 static int spx_listen(struct socket *so, struct thread *td);
103 static int spx_rcvd(struct socket *so, int flags);
104 static int spx_rcvoob(struct socket *so, struct mbuf *m, int flags);
105 static int spx_send(struct socket *so, int flags, struct mbuf *m,
106 struct sockaddr *addr, struct mbuf *control,
107 struct thread *td);
108 static int spx_shutdown(struct socket *so);
109 static int spx_sp_attach(struct socket *so, int proto, struct thread *td);
110
111 struct pr_usrreqs spx_usrreqs = {
112 spx_usr_abort, spx_accept, spx_attach, spx_bind,
113 spx_connect, pru_connect2_notsupp, ipx_control, spx_detach,
114 spx_usr_disconnect, spx_listen, ipx_peeraddr, spx_rcvd,
115 spx_rcvoob, spx_send, pru_sense_null, spx_shutdown,
116 ipx_sockaddr, sosend, soreceive, sopoll, pru_sosetlabel_null
117 };
118
119 struct pr_usrreqs spx_usrreq_sps = {
120 spx_usr_abort, spx_accept, spx_sp_attach, spx_bind,
121 spx_connect, pru_connect2_notsupp, ipx_control, spx_detach,
122 spx_usr_disconnect, spx_listen, ipx_peeraddr, spx_rcvd,
123 spx_rcvoob, spx_send, pru_sense_null, spx_shutdown,
124 ipx_sockaddr, sosend, soreceive, sopoll, pru_sosetlabel_null
125 };
126
127 void
128 spx_init()
129 {
130
131 spx_iss = 1; /* WRONG !! should fish it out of TODR */
132 }
133
134 void
135 spx_input(m, ipxp)
136 register struct mbuf *m;
137 register struct ipxpcb *ipxp;
138 {
139 register struct spxpcb *cb;
140 register struct spx *si = mtod(m, struct spx *);
141 register struct socket *so;
142 int dropsocket = 0;
143 short ostate = 0;
144
145 spxstat.spxs_rcvtotal++;
146 KASSERT(ipxp != NULL, ("spx_input: NULL ipxpcb"));
147
148 /*
149 * spx_input() assumes that the caller will hold both the pcb list
150 * lock and also the ipxp lock. spx_input() will release both before
151 * returning, and may in fact trade in the ipxp lock for another pcb
152 * lock following sonewconn().
153 */
154 IPX_LIST_LOCK_ASSERT();
155 IPX_LOCK_ASSERT(ipxp);
156
157 cb = ipxtospxpcb(ipxp);
158 if (cb == NULL)
159 goto bad;
160
161 if (m->m_len < sizeof(*si)) {
162 if ((m = m_pullup(m, sizeof(*si))) == NULL) {
163 IPX_UNLOCK(ipxp);
164 IPX_LIST_UNLOCK();
165 spxstat.spxs_rcvshort++;
166 return;
167 }
168 si = mtod(m, struct spx *);
169 }
170 si->si_seq = ntohs(si->si_seq);
171 si->si_ack = ntohs(si->si_ack);
172 si->si_alo = ntohs(si->si_alo);
173
174 so = ipxp->ipxp_socket;
175
176 if (so->so_options & SO_DEBUG || traceallspxs) {
177 ostate = cb->s_state;
178 spx_savesi = *si;
179 }
180 if (so->so_options & SO_ACCEPTCONN) {
181 struct spxpcb *ocb = cb;
182
183 so = sonewconn(so, 0);
184 if (so == NULL) {
185 goto drop;
186 }
187 /*
188 * This is ugly, but ....
189 *
190 * Mark socket as temporary until we're
191 * committed to keeping it. The code at
192 * ``drop'' and ``dropwithreset'' check the
193 * flag dropsocket to see if the temporary
194 * socket created here should be discarded.
195 * We mark the socket as discardable until
196 * we're committed to it below in TCPS_LISTEN.
197 */
198 dropsocket++;
199 IPX_UNLOCK(ipxp);
200 ipxp = (struct ipxpcb *)so->so_pcb;
201 IPX_LOCK(ipxp);
202 ipxp->ipxp_laddr = si->si_dna;
203 cb = ipxtospxpcb(ipxp);
204 cb->s_mtu = ocb->s_mtu; /* preserve sockopts */
205 cb->s_flags = ocb->s_flags; /* preserve sockopts */
206 cb->s_flags2 = ocb->s_flags2; /* preserve sockopts */
207 cb->s_state = TCPS_LISTEN;
208 }
209
210 /*
211 * Packet received on connection.
212 * reset idle time and keep-alive timer;
213 */
214 cb->s_idle = 0;
215 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
216
217 switch (cb->s_state) {
218
219 case TCPS_LISTEN:{
220 struct sockaddr_ipx *sipx, ssipx;
221 struct ipx_addr laddr;
222
223 /*
224 * If somebody here was carying on a conversation
225 * and went away, and his pen pal thinks he can
226 * still talk, we get the misdirected packet.
227 */
228 if (spx_hardnosed && (si->si_did != 0 || si->si_seq != 0)) {
229 spx_istat.gonawy++;
230 goto dropwithreset;
231 }
232 sipx = &ssipx;
233 bzero(sipx, sizeof *sipx);
234 sipx->sipx_len = sizeof(*sipx);
235 sipx->sipx_family = AF_IPX;
236 sipx->sipx_addr = si->si_sna;
237 laddr = ipxp->ipxp_laddr;
238 if (ipx_nullhost(laddr))
239 ipxp->ipxp_laddr = si->si_dna;
240 if (ipx_pcbconnect(ipxp, (struct sockaddr *)sipx, &thread0)) {
241 ipxp->ipxp_laddr = laddr;
242 spx_istat.noconn++;
243 goto drop;
244 }
245 spx_template(cb);
246 dropsocket = 0; /* committed to socket */
247 cb->s_did = si->si_sid;
248 cb->s_rack = si->si_ack;
249 cb->s_ralo = si->si_alo;
250 #define THREEWAYSHAKE
251 #ifdef THREEWAYSHAKE
252 cb->s_state = TCPS_SYN_RECEIVED;
253 cb->s_force = 1 + SPXT_KEEP;
254 spxstat.spxs_accepts++;
255 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
256 }
257 break;
258 /*
259 * This state means that we have heard a response
260 * to our acceptance of their connection
261 * It is probably logically unnecessary in this
262 * implementation.
263 */
264 case TCPS_SYN_RECEIVED: {
265 if (si->si_did != cb->s_sid) {
266 spx_istat.wrncon++;
267 goto drop;
268 }
269 #endif
270 ipxp->ipxp_fport = si->si_sport;
271 cb->s_timer[SPXT_REXMT] = 0;
272 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
273 soisconnected(so);
274 cb->s_state = TCPS_ESTABLISHED;
275 spxstat.spxs_accepts++;
276 }
277 break;
278
279 /*
280 * This state means that we have gotten a response
281 * to our attempt to establish a connection.
282 * We fill in the data from the other side,
283 * telling us which port to respond to, instead of the well-
284 * known one we might have sent to in the first place.
285 * We also require that this is a response to our
286 * connection id.
287 */
288 case TCPS_SYN_SENT:
289 if (si->si_did != cb->s_sid) {
290 spx_istat.notme++;
291 goto drop;
292 }
293 spxstat.spxs_connects++;
294 cb->s_did = si->si_sid;
295 cb->s_rack = si->si_ack;
296 cb->s_ralo = si->si_alo;
297 cb->s_dport = ipxp->ipxp_fport = si->si_sport;
298 cb->s_timer[SPXT_REXMT] = 0;
299 cb->s_flags |= SF_ACKNOW;
300 soisconnected(so);
301 cb->s_state = TCPS_ESTABLISHED;
302 /* Use roundtrip time of connection request for initial rtt */
303 if (cb->s_rtt) {
304 cb->s_srtt = cb->s_rtt << 3;
305 cb->s_rttvar = cb->s_rtt << 1;
306 SPXT_RANGESET(cb->s_rxtcur,
307 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
308 SPXTV_MIN, SPXTV_REXMTMAX);
309 cb->s_rtt = 0;
310 }
311 }
312 if (so->so_options & SO_DEBUG || traceallspxs)
313 spx_trace(SA_INPUT, (u_char)ostate, cb, &spx_savesi, 0);
314
315 m->m_len -= sizeof(struct ipx);
316 m->m_pkthdr.len -= sizeof(struct ipx);
317 m->m_data += sizeof(struct ipx);
318
319 if (spx_reass(cb, si)) {
320 m_freem(m);
321 }
322 if (cb->s_force || (cb->s_flags & (SF_ACKNOW|SF_WIN|SF_RXT)))
323 spx_output(cb, NULL);
324 cb->s_flags &= ~(SF_WIN|SF_RXT);
325 IPX_UNLOCK(ipxp);
326 IPX_LIST_UNLOCK();
327 return;
328
329 dropwithreset:
330 IPX_UNLOCK(ipxp);
331 if (dropsocket) {
332 struct socket *head;
333 ACCEPT_LOCK();
334 KASSERT((so->so_qstate & SQ_INCOMP) != 0,
335 ("spx_input: nascent socket not SQ_INCOMP on soabort()"));
336 head = so->so_head;
337 TAILQ_REMOVE(&head->so_incomp, so, so_list);
338 head->so_incqlen--;
339 so->so_qstate &= ~SQ_INCOMP;
340 so->so_head = NULL;
341 ACCEPT_UNLOCK();
342 soabort(so);
343 cb = NULL;
344 }
345 IPX_LIST_UNLOCK();
346 si->si_seq = ntohs(si->si_seq);
347 si->si_ack = ntohs(si->si_ack);
348 si->si_alo = ntohs(si->si_alo);
349 m_freem(dtom(si));
350 if (cb == NULL || cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG ||
351 traceallspxs)
352 spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
353 return;
354
355 drop:
356 bad:
357 if (cb == NULL || cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG ||
358 traceallspxs)
359 spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
360 IPX_UNLOCK(ipxp);
361 IPX_LIST_UNLOCK();
362 m_freem(m);
363 }
364
365 static int spxrexmtthresh = 3;
366
367 /*
368 * This is structurally similar to the tcp reassembly routine
369 * but its function is somewhat different: It merely queues
370 * packets up, and suppresses duplicates.
371 */
372 static int
373 spx_reass(cb, si)
374 register struct spxpcb *cb;
375 register struct spx *si;
376 {
377 register struct spx_q *q;
378 register struct mbuf *m;
379 register struct socket *so = cb->s_ipxpcb->ipxp_socket;
380 char packetp = cb->s_flags & SF_HI;
381 int incr;
382 char wakeup = 0;
383
384 IPX_LOCK_ASSERT(cb->s_ipxpcb);
385
386 if (si == SI(0))
387 goto present;
388 /*
389 * Update our news from them.
390 */
391 if (si->si_cc & SPX_SA)
392 cb->s_flags |= (spx_use_delack ? SF_DELACK : SF_ACKNOW);
393 if (SSEQ_GT(si->si_alo, cb->s_ralo))
394 cb->s_flags |= SF_WIN;
395 if (SSEQ_LEQ(si->si_ack, cb->s_rack)) {
396 if ((si->si_cc & SPX_SP) && cb->s_rack != (cb->s_smax + 1)) {
397 spxstat.spxs_rcvdupack++;
398 /*
399 * If this is a completely duplicate ack
400 * and other conditions hold, we assume
401 * a packet has been dropped and retransmit
402 * it exactly as in tcp_input().
403 */
404 if (si->si_ack != cb->s_rack ||
405 si->si_alo != cb->s_ralo)
406 cb->s_dupacks = 0;
407 else if (++cb->s_dupacks == spxrexmtthresh) {
408 u_short onxt = cb->s_snxt;
409 int cwnd = cb->s_cwnd;
410
411 cb->s_snxt = si->si_ack;
412 cb->s_cwnd = CUNIT;
413 cb->s_force = 1 + SPXT_REXMT;
414 spx_output(cb, NULL);
415 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
416 cb->s_rtt = 0;
417 if (cwnd >= 4 * CUNIT)
418 cb->s_cwnd = cwnd / 2;
419 if (SSEQ_GT(onxt, cb->s_snxt))
420 cb->s_snxt = onxt;
421 return (1);
422 }
423 } else
424 cb->s_dupacks = 0;
425 goto update_window;
426 }
427 cb->s_dupacks = 0;
428 /*
429 * If our correspondent acknowledges data we haven't sent
430 * TCP would drop the packet after acking. We'll be a little
431 * more permissive
432 */
433 if (SSEQ_GT(si->si_ack, (cb->s_smax + 1))) {
434 spxstat.spxs_rcvacktoomuch++;
435 si->si_ack = cb->s_smax + 1;
436 }
437 spxstat.spxs_rcvackpack++;
438 /*
439 * If transmit timer is running and timed sequence
440 * number was acked, update smoothed round trip time.
441 * See discussion of algorithm in tcp_input.c
442 */
443 if (cb->s_rtt && SSEQ_GT(si->si_ack, cb->s_rtseq)) {
444 spxstat.spxs_rttupdated++;
445 if (cb->s_srtt != 0) {
446 register short delta;
447 delta = cb->s_rtt - (cb->s_srtt >> 3);
448 if ((cb->s_srtt += delta) <= 0)
449 cb->s_srtt = 1;
450 if (delta < 0)
451 delta = -delta;
452 delta -= (cb->s_rttvar >> 2);
453 if ((cb->s_rttvar += delta) <= 0)
454 cb->s_rttvar = 1;
455 } else {
456 /*
457 * No rtt measurement yet
458 */
459 cb->s_srtt = cb->s_rtt << 3;
460 cb->s_rttvar = cb->s_rtt << 1;
461 }
462 cb->s_rtt = 0;
463 cb->s_rxtshift = 0;
464 SPXT_RANGESET(cb->s_rxtcur,
465 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
466 SPXTV_MIN, SPXTV_REXMTMAX);
467 }
468 /*
469 * If all outstanding data is acked, stop retransmit
470 * timer and remember to restart (more output or persist).
471 * If there is more data to be acked, restart retransmit
472 * timer, using current (possibly backed-off) value;
473 */
474 if (si->si_ack == cb->s_smax + 1) {
475 cb->s_timer[SPXT_REXMT] = 0;
476 cb->s_flags |= SF_RXT;
477 } else if (cb->s_timer[SPXT_PERSIST] == 0)
478 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
479 /*
480 * When new data is acked, open the congestion window.
481 * If the window gives us less than ssthresh packets
482 * in flight, open exponentially (maxseg at a time).
483 * Otherwise open linearly (maxseg^2 / cwnd at a time).
484 */
485 incr = CUNIT;
486 if (cb->s_cwnd > cb->s_ssthresh)
487 incr = max(incr * incr / cb->s_cwnd, 1);
488 cb->s_cwnd = min(cb->s_cwnd + incr, cb->s_cwmx);
489 /*
490 * Trim Acked data from output queue.
491 */
492 SOCKBUF_LOCK(&so->so_snd);
493 while ((m = so->so_snd.sb_mb) != NULL) {
494 if (SSEQ_LT((mtod(m, struct spx *))->si_seq, si->si_ack))
495 sbdroprecord_locked(&so->so_snd);
496 else
497 break;
498 }
499 sowwakeup_locked(so);
500 cb->s_rack = si->si_ack;
501 update_window:
502 if (SSEQ_LT(cb->s_snxt, cb->s_rack))
503 cb->s_snxt = cb->s_rack;
504 if (SSEQ_LT(cb->s_swl1, si->si_seq) || ((cb->s_swl1 == si->si_seq &&
505 (SSEQ_LT(cb->s_swl2, si->si_ack))) ||
506 (cb->s_swl2 == si->si_ack && SSEQ_LT(cb->s_ralo, si->si_alo)))) {
507 /* keep track of pure window updates */
508 if ((si->si_cc & SPX_SP) && cb->s_swl2 == si->si_ack
509 && SSEQ_LT(cb->s_ralo, si->si_alo)) {
510 spxstat.spxs_rcvwinupd++;
511 spxstat.spxs_rcvdupack--;
512 }
513 cb->s_ralo = si->si_alo;
514 cb->s_swl1 = si->si_seq;
515 cb->s_swl2 = si->si_ack;
516 cb->s_swnd = (1 + si->si_alo - si->si_ack);
517 if (cb->s_swnd > cb->s_smxw)
518 cb->s_smxw = cb->s_swnd;
519 cb->s_flags |= SF_WIN;
520 }
521 /*
522 * If this packet number is higher than that which
523 * we have allocated refuse it, unless urgent
524 */
525 if (SSEQ_GT(si->si_seq, cb->s_alo)) {
526 if (si->si_cc & SPX_SP) {
527 spxstat.spxs_rcvwinprobe++;
528 return (1);
529 } else
530 spxstat.spxs_rcvpackafterwin++;
531 if (si->si_cc & SPX_OB) {
532 if (SSEQ_GT(si->si_seq, cb->s_alo + 60)) {
533 m_freem(dtom(si));
534 return (0);
535 } /* else queue this packet; */
536 } else {
537 #ifdef BROKEN
538 /*
539 * XXXRW: This is broken on at least one count:
540 * spx_close() will free the ipxp and related parts,
541 * which are then touched by spx_input() after the
542 * return from spx_reass().
543 */
544 /*register struct socket *so = cb->s_ipxpcb->ipxp_socket;
545 if (so->so_state && SS_NOFDREF) {
546 spx_close(cb);
547 } else
548 would crash system*/
549 #endif
550 spx_istat.notyet++;
551 m_freem(dtom(si));
552 return (0);
553 }
554 }
555 /*
556 * If this is a system packet, we don't need to
557 * queue it up, and won't update acknowledge #
558 */
559 if (si->si_cc & SPX_SP) {
560 return (1);
561 }
562 /*
563 * We have already seen this packet, so drop.
564 */
565 if (SSEQ_LT(si->si_seq, cb->s_ack)) {
566 spx_istat.bdreas++;
567 spxstat.spxs_rcvduppack++;
568 if (si->si_seq == cb->s_ack - 1)
569 spx_istat.lstdup++;
570 return (1);
571 }
572 /*
573 * Loop through all packets queued up to insert in
574 * appropriate sequence.
575 */
576 for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
577 if (si->si_seq == SI(q)->si_seq) {
578 spxstat.spxs_rcvduppack++;
579 return (1);
580 }
581 if (SSEQ_LT(si->si_seq, SI(q)->si_seq)) {
582 spxstat.spxs_rcvoopack++;
583 break;
584 }
585 }
586 insque(si, q->si_prev);
587 /*
588 * If this packet is urgent, inform process
589 */
590 if (si->si_cc & SPX_OB) {
591 cb->s_iobc = ((char *)si)[1 + sizeof(*si)];
592 sohasoutofband(so);
593 cb->s_oobflags |= SF_IOOB;
594 }
595 present:
596 #define SPINC sizeof(struct spxhdr)
597 SOCKBUF_LOCK(&so->so_rcv);
598 /*
599 * Loop through all packets queued up to update acknowledge
600 * number, and present all acknowledged data to user;
601 * If in packet interface mode, show packet headers.
602 */
603 for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
604 if (SI(q)->si_seq == cb->s_ack) {
605 cb->s_ack++;
606 m = dtom(q);
607 if (SI(q)->si_cc & SPX_OB) {
608 cb->s_oobflags &= ~SF_IOOB;
609 if (so->so_rcv.sb_cc)
610 so->so_oobmark = so->so_rcv.sb_cc;
611 else
612 so->so_rcv.sb_state |= SBS_RCVATMARK;
613 }
614 q = q->si_prev;
615 remque(q->si_next);
616 wakeup = 1;
617 spxstat.spxs_rcvpack++;
618 #ifdef SF_NEWCALL
619 if (cb->s_flags2 & SF_NEWCALL) {
620 struct spxhdr *sp = mtod(m, struct spxhdr *);
621 u_char dt = sp->spx_dt;
622 spx_newchecks[4]++;
623 if (dt != cb->s_rhdr.spx_dt) {
624 struct mbuf *mm =
625 m_getclr(M_DONTWAIT, MT_CONTROL);
626 spx_newchecks[0]++;
627 if (mm != NULL) {
628 u_short *s =
629 mtod(mm, u_short *);
630 cb->s_rhdr.spx_dt = dt;
631 mm->m_len = 5; /*XXX*/
632 s[0] = 5;
633 s[1] = 1;
634 *(u_char *)(&s[2]) = dt;
635 sbappend_locked(&so->so_rcv, mm);
636 }
637 }
638 if (sp->spx_cc & SPX_OB) {
639 MCHTYPE(m, MT_OOBDATA);
640 spx_newchecks[1]++;
641 so->so_oobmark = 0;
642 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
643 }
644 if (packetp == 0) {
645 m->m_data += SPINC;
646 m->m_len -= SPINC;
647 m->m_pkthdr.len -= SPINC;
648 }
649 if ((sp->spx_cc & SPX_EM) || packetp) {
650 sbappendrecord_locked(&so->so_rcv, m);
651 spx_newchecks[9]++;
652 } else
653 sbappend_locked(&so->so_rcv, m);
654 } else
655 #endif
656 if (packetp) {
657 sbappendrecord_locked(&so->so_rcv, m);
658 } else {
659 cb->s_rhdr = *mtod(m, struct spxhdr *);
660 m->m_data += SPINC;
661 m->m_len -= SPINC;
662 m->m_pkthdr.len -= SPINC;
663 sbappend_locked(&so->so_rcv, m);
664 }
665 } else
666 break;
667 }
668 if (wakeup)
669 sorwakeup_locked(so);
670 else
671 SOCKBUF_UNLOCK(&so->so_rcv);
672 return (0);
673 }
674
675 void
676 spx_ctlinput(cmd, arg_as_sa, dummy)
677 int cmd;
678 struct sockaddr *arg_as_sa; /* XXX should be swapped with dummy */
679 void *dummy;
680 {
681 caddr_t arg = (/* XXX */ caddr_t)arg_as_sa;
682 struct ipx_addr *na;
683 struct sockaddr_ipx *sipx;
684
685 if (cmd < 0 || cmd >= PRC_NCMDS)
686 return;
687
688 switch (cmd) {
689
690 case PRC_ROUTEDEAD:
691 return;
692
693 case PRC_IFDOWN:
694 case PRC_HOSTDEAD:
695 case PRC_HOSTUNREACH:
696 sipx = (struct sockaddr_ipx *)arg;
697 if (sipx->sipx_family != AF_IPX)
698 return;
699 na = &sipx->sipx_addr;
700 break;
701
702 default:
703 break;
704 }
705 }
706
707 static int
708 spx_output(cb, m0)
709 register struct spxpcb *cb;
710 struct mbuf *m0;
711 {
712 struct socket *so = cb->s_ipxpcb->ipxp_socket;
713 register struct mbuf *m;
714 register struct spx *si = NULL;
715 register struct sockbuf *sb = &so->so_snd;
716 int len = 0, win, rcv_win;
717 short span, off, recordp = 0;
718 u_short alo;
719 int error = 0, sendalot;
720 #ifdef notdef
721 int idle;
722 #endif
723 struct mbuf *mprev;
724
725 IPX_LOCK_ASSERT(cb->s_ipxpcb);
726
727 if (m0 != NULL) {
728 int mtu = cb->s_mtu;
729 int datalen;
730 /*
731 * Make sure that packet isn't too big.
732 */
733 for (m = m0; m != NULL; m = m->m_next) {
734 mprev = m;
735 len += m->m_len;
736 if (m->m_flags & M_EOR)
737 recordp = 1;
738 }
739 datalen = (cb->s_flags & SF_HO) ?
740 len - sizeof(struct spxhdr) : len;
741 if (datalen > mtu) {
742 if (cb->s_flags & SF_PI) {
743 m_freem(m0);
744 return (EMSGSIZE);
745 } else {
746 int oldEM = cb->s_cc & SPX_EM;
747
748 cb->s_cc &= ~SPX_EM;
749 while (len > mtu) {
750 /*
751 * Here we are only being called
752 * from usrreq(), so it is OK to
753 * block.
754 */
755 m = m_copym(m0, 0, mtu, M_TRYWAIT);
756 if (cb->s_flags & SF_NEWCALL) {
757 struct mbuf *mm = m;
758 spx_newchecks[7]++;
759 while (mm != NULL) {
760 mm->m_flags &= ~M_EOR;
761 mm = mm->m_next;
762 }
763 }
764 error = spx_output(cb, m);
765 if (error) {
766 cb->s_cc |= oldEM;
767 m_freem(m0);
768 return (error);
769 }
770 m_adj(m0, mtu);
771 len -= mtu;
772 }
773 cb->s_cc |= oldEM;
774 }
775 }
776 /*
777 * Force length even, by adding a "garbage byte" if
778 * necessary.
779 */
780 if (len & 1) {
781 m = mprev;
782 if (M_TRAILINGSPACE(m) >= 1)
783 m->m_len++;
784 else {
785 struct mbuf *m1 = m_get(M_DONTWAIT, MT_DATA);
786
787 if (m1 == NULL) {
788 m_freem(m0);
789 return (ENOBUFS);
790 }
791 m1->m_len = 1;
792 *(mtod(m1, u_char *)) = 0;
793 m->m_next = m1;
794 }
795 }
796 m = m_gethdr(M_DONTWAIT, MT_HEADER);
797 if (m == NULL) {
798 m_freem(m0);
799 return (ENOBUFS);
800 }
801 /*
802 * Fill in mbuf with extended SP header
803 * and addresses and length put into network format.
804 */
805 MH_ALIGN(m, sizeof(struct spx));
806 m->m_len = sizeof(struct spx);
807 m->m_next = m0;
808 si = mtod(m, struct spx *);
809 si->si_i = *cb->s_ipx;
810 si->si_s = cb->s_shdr;
811 if ((cb->s_flags & SF_PI) && (cb->s_flags & SF_HO)) {
812 register struct spxhdr *sh;
813 if (m0->m_len < sizeof(*sh)) {
814 if((m0 = m_pullup(m0, sizeof(*sh))) == NULL) {
815 m_free(m);
816 m_freem(m0);
817 return (EINVAL);
818 }
819 m->m_next = m0;
820 }
821 sh = mtod(m0, struct spxhdr *);
822 si->si_dt = sh->spx_dt;
823 si->si_cc |= sh->spx_cc & SPX_EM;
824 m0->m_len -= sizeof(*sh);
825 m0->m_data += sizeof(*sh);
826 len -= sizeof(*sh);
827 }
828 len += sizeof(*si);
829 if ((cb->s_flags2 & SF_NEWCALL) && recordp) {
830 si->si_cc |= SPX_EM;
831 spx_newchecks[8]++;
832 }
833 if (cb->s_oobflags & SF_SOOB) {
834 /*
835 * Per jqj@cornell:
836 * make sure OB packets convey exactly 1 byte.
837 * If the packet is 1 byte or larger, we
838 * have already guaranted there to be at least
839 * one garbage byte for the checksum, and
840 * extra bytes shouldn't hurt!
841 */
842 if (len > sizeof(*si)) {
843 si->si_cc |= SPX_OB;
844 len = (1 + sizeof(*si));
845 }
846 }
847 si->si_len = htons((u_short)len);
848 m->m_pkthdr.len = ((len - 1) | 1) + 1;
849 /*
850 * queue stuff up for output
851 */
852 sbappendrecord(sb, m);
853 cb->s_seq++;
854 }
855 #ifdef notdef
856 idle = (cb->s_smax == (cb->s_rack - 1));
857 #endif
858 again:
859 sendalot = 0;
860 off = cb->s_snxt - cb->s_rack;
861 win = min(cb->s_swnd, (cb->s_cwnd / CUNIT));
862
863 /*
864 * If in persist timeout with window of 0, send a probe.
865 * Otherwise, if window is small but nonzero
866 * and timer expired, send what we can and go into
867 * transmit state.
868 */
869 if (cb->s_force == 1 + SPXT_PERSIST) {
870 if (win != 0) {
871 cb->s_timer[SPXT_PERSIST] = 0;
872 cb->s_rxtshift = 0;
873 }
874 }
875 span = cb->s_seq - cb->s_rack;
876 len = min(span, win) - off;
877
878 if (len < 0) {
879 /*
880 * Window shrank after we went into it.
881 * If window shrank to 0, cancel pending
882 * restransmission and pull s_snxt back
883 * to (closed) window. We will enter persist
884 * state below. If the widndow didn't close completely,
885 * just wait for an ACK.
886 */
887 len = 0;
888 if (win == 0) {
889 cb->s_timer[SPXT_REXMT] = 0;
890 cb->s_snxt = cb->s_rack;
891 }
892 }
893 if (len > 1)
894 sendalot = 1;
895 rcv_win = sbspace(&so->so_rcv);
896
897 /*
898 * Send if we owe peer an ACK.
899 */
900 if (cb->s_oobflags & SF_SOOB) {
901 /*
902 * must transmit this out of band packet
903 */
904 cb->s_oobflags &= ~ SF_SOOB;
905 sendalot = 1;
906 spxstat.spxs_sndurg++;
907 goto found;
908 }
909 if (cb->s_flags & SF_ACKNOW)
910 goto send;
911 if (cb->s_state < TCPS_ESTABLISHED)
912 goto send;
913 /*
914 * Silly window can't happen in spx.
915 * Code from tcp deleted.
916 */
917 if (len)
918 goto send;
919 /*
920 * Compare available window to amount of window
921 * known to peer (as advertised window less
922 * next expected input.) If the difference is at least two
923 * packets or at least 35% of the mximum possible window,
924 * then want to send a window update to peer.
925 */
926 if (rcv_win > 0) {
927 u_short delta = 1 + cb->s_alo - cb->s_ack;
928 int adv = rcv_win - (delta * cb->s_mtu);
929
930 if ((so->so_rcv.sb_cc == 0 && adv >= (2 * cb->s_mtu)) ||
931 (100 * adv / so->so_rcv.sb_hiwat >= 35)) {
932 spxstat.spxs_sndwinup++;
933 cb->s_flags |= SF_ACKNOW;
934 goto send;
935 }
936
937 }
938 /*
939 * Many comments from tcp_output.c are appropriate here
940 * including . . .
941 * If send window is too small, there is data to transmit, and no
942 * retransmit or persist is pending, then go to persist state.
943 * If nothing happens soon, send when timer expires:
944 * if window is nonzero, transmit what we can,
945 * otherwise send a probe.
946 */
947 if (so->so_snd.sb_cc && cb->s_timer[SPXT_REXMT] == 0 &&
948 cb->s_timer[SPXT_PERSIST] == 0) {
949 cb->s_rxtshift = 0;
950 spx_setpersist(cb);
951 }
952 /*
953 * No reason to send a packet, just return.
954 */
955 cb->s_outx = 1;
956 return (0);
957
958 send:
959 /*
960 * Find requested packet.
961 */
962 si = 0;
963 if (len > 0) {
964 cb->s_want = cb->s_snxt;
965 for (m = sb->sb_mb; m != NULL; m = m->m_act) {
966 si = mtod(m, struct spx *);
967 if (SSEQ_LEQ(cb->s_snxt, si->si_seq))
968 break;
969 }
970 found:
971 if (si != NULL) {
972 if (si->si_seq == cb->s_snxt)
973 cb->s_snxt++;
974 else
975 spxstat.spxs_sndvoid++, si = 0;
976 }
977 }
978 /*
979 * update window
980 */
981 if (rcv_win < 0)
982 rcv_win = 0;
983 alo = cb->s_ack - 1 + (rcv_win / ((short)cb->s_mtu));
984 if (SSEQ_LT(alo, cb->s_alo))
985 alo = cb->s_alo;
986
987 if (si != NULL) {
988 /*
989 * must make a copy of this packet for
990 * ipx_output to monkey with
991 */
992 m = m_copy(dtom(si), 0, (int)M_COPYALL);
993 if (m == NULL) {
994 return (ENOBUFS);
995 }
996 si = mtod(m, struct spx *);
997 if (SSEQ_LT(si->si_seq, cb->s_smax))
998 spxstat.spxs_sndrexmitpack++;
999 else
1000 spxstat.spxs_sndpack++;
1001 } else if (cb->s_force || cb->s_flags & SF_ACKNOW) {
1002 /*
1003 * Must send an acknowledgement or a probe
1004 */
1005 if (cb->s_force)
1006 spxstat.spxs_sndprobe++;
1007 if (cb->s_flags & SF_ACKNOW)
1008 spxstat.spxs_sndacks++;
1009 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1010 if (m == NULL)
1011 return (ENOBUFS);
1012 /*
1013 * Fill in mbuf with extended SP header
1014 * and addresses and length put into network format.
1015 */
1016 MH_ALIGN(m, sizeof(struct spx));
1017 m->m_len = sizeof(*si);
1018 m->m_pkthdr.len = sizeof(*si);
1019 si = mtod(m, struct spx *);
1020 si->si_i = *cb->s_ipx;
1021 si->si_s = cb->s_shdr;
1022 si->si_seq = cb->s_smax + 1;
1023 si->si_len = htons(sizeof(*si));
1024 si->si_cc |= SPX_SP;
1025 } else {
1026 cb->s_outx = 3;
1027 if (so->so_options & SO_DEBUG || traceallspxs)
1028 spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1029 return (0);
1030 }
1031 /*
1032 * Stuff checksum and output datagram.
1033 */
1034 if ((si->si_cc & SPX_SP) == 0) {
1035 if (cb->s_force != (1 + SPXT_PERSIST) ||
1036 cb->s_timer[SPXT_PERSIST] == 0) {
1037 /*
1038 * If this is a new packet and we are not currently
1039 * timing anything, time this one.
1040 */
1041 if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1042 cb->s_smax = si->si_seq;
1043 if (cb->s_rtt == 0) {
1044 spxstat.spxs_segstimed++;
1045 cb->s_rtseq = si->si_seq;
1046 cb->s_rtt = 1;
1047 }
1048 }
1049 /*
1050 * Set rexmt timer if not currently set,
1051 * Initial value for retransmit timer is smoothed
1052 * round-trip time + 2 * round-trip time variance.
1053 * Initialize shift counter which is used for backoff
1054 * of retransmit time.
1055 */
1056 if (cb->s_timer[SPXT_REXMT] == 0 &&
1057 cb->s_snxt != cb->s_rack) {
1058 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1059 if (cb->s_timer[SPXT_PERSIST]) {
1060 cb->s_timer[SPXT_PERSIST] = 0;
1061 cb->s_rxtshift = 0;
1062 }
1063 }
1064 } else if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1065 cb->s_smax = si->si_seq;
1066 }
1067 } else if (cb->s_state < TCPS_ESTABLISHED) {
1068 if (cb->s_rtt == 0)
1069 cb->s_rtt = 1; /* Time initial handshake */
1070 if (cb->s_timer[SPXT_REXMT] == 0)
1071 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1072 }
1073 {
1074 /*
1075 * Do not request acks when we ack their data packets or
1076 * when we do a gratuitous window update.
1077 */
1078 if (((si->si_cc & SPX_SP) == 0) || cb->s_force)
1079 si->si_cc |= SPX_SA;
1080 si->si_seq = htons(si->si_seq);
1081 si->si_alo = htons(alo);
1082 si->si_ack = htons(cb->s_ack);
1083
1084 if (ipxcksum) {
1085 si->si_sum = ipx_cksum(m, ntohs(si->si_len));
1086 } else
1087 si->si_sum = 0xffff;
1088
1089 cb->s_outx = 4;
1090 if (so->so_options & SO_DEBUG || traceallspxs)
1091 spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1092
1093 if (so->so_options & SO_DONTROUTE)
1094 error = ipx_outputfl(m, NULL, IPX_ROUTETOIF);
1095 else
1096 error = ipx_outputfl(m, &cb->s_ipxpcb->ipxp_route, 0);
1097 }
1098 if (error) {
1099 return (error);
1100 }
1101 spxstat.spxs_sndtotal++;
1102 /*
1103 * Data sent (as far as we can tell).
1104 * If this advertises a larger window than any other segment,
1105 * then remember the size of the advertized window.
1106 * Any pending ACK has now been sent.
1107 */
1108 cb->s_force = 0;
1109 cb->s_flags &= ~(SF_ACKNOW|SF_DELACK);
1110 if (SSEQ_GT(alo, cb->s_alo))
1111 cb->s_alo = alo;
1112 if (sendalot)
1113 goto again;
1114 cb->s_outx = 5;
1115 return (0);
1116 }
1117
1118 static int spx_do_persist_panics = 0;
1119
1120 static void
1121 spx_setpersist(cb)
1122 register struct spxpcb *cb;
1123 {
1124 register int t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1125
1126 IPX_LOCK_ASSERT(cb->s_ipxpcb);
1127
1128 if (cb->s_timer[SPXT_REXMT] && spx_do_persist_panics)
1129 panic("spx_output REXMT");
1130 /*
1131 * Start/restart persistance timer.
1132 */
1133 SPXT_RANGESET(cb->s_timer[SPXT_PERSIST],
1134 t*spx_backoff[cb->s_rxtshift],
1135 SPXTV_PERSMIN, SPXTV_PERSMAX);
1136 if (cb->s_rxtshift < SPX_MAXRXTSHIFT)
1137 cb->s_rxtshift++;
1138 }
1139
1140 int
1141 spx_ctloutput(so, sopt)
1142 struct socket *so;
1143 struct sockopt *sopt;
1144 {
1145 struct ipxpcb *ipxp = sotoipxpcb(so);
1146 register struct spxpcb *cb;
1147 int mask, error;
1148 short soptval;
1149 u_short usoptval;
1150 int optval;
1151
1152 error = 0;
1153
1154 if (sopt->sopt_level != IPXPROTO_SPX) {
1155 /* This will have to be changed when we do more general
1156 stacking of protocols */
1157 return (ipx_ctloutput(so, sopt));
1158 }
1159 if (ipxp == NULL)
1160 return (EINVAL);
1161 else
1162 cb = ipxtospxpcb(ipxp);
1163
1164 switch (sopt->sopt_dir) {
1165 case SOPT_GET:
1166 switch (sopt->sopt_name) {
1167 case SO_HEADERS_ON_INPUT:
1168 mask = SF_HI;
1169 goto get_flags;
1170
1171 case SO_HEADERS_ON_OUTPUT:
1172 mask = SF_HO;
1173 get_flags:
1174 /* Unlocked read. */
1175 soptval = cb->s_flags & mask;
1176 error = sooptcopyout(sopt, &soptval, sizeof soptval);
1177 break;
1178
1179 case SO_MTU:
1180 /* Unlocked read. */
1181 usoptval = cb->s_mtu;
1182 error = sooptcopyout(sopt, &usoptval, sizeof usoptval);
1183 break;
1184
1185 case SO_LAST_HEADER:
1186 /* Unlocked read. */
1187 error = sooptcopyout(sopt, &cb->s_rhdr,
1188 sizeof cb->s_rhdr);
1189 break;
1190
1191 case SO_DEFAULT_HEADERS:
1192 /* Unlocked read. */
1193 error = sooptcopyout(sopt, &cb->s_shdr,
1194 sizeof cb->s_shdr);
1195 break;
1196
1197 default:
1198 error = ENOPROTOOPT;
1199 }
1200 break;
1201
1202 case SOPT_SET:
1203 switch (sopt->sopt_name) {
1204 /* XXX why are these shorts on get and ints on set?
1205 that doesn't make any sense... */
1206 case SO_HEADERS_ON_INPUT:
1207 mask = SF_HI;
1208 goto set_head;
1209
1210 case SO_HEADERS_ON_OUTPUT:
1211 mask = SF_HO;
1212 set_head:
1213 error = sooptcopyin(sopt, &optval, sizeof optval,
1214 sizeof optval);
1215 if (error)
1216 break;
1217
1218 IPX_LOCK(ipxp);
1219 if (cb->s_flags & SF_PI) {
1220 if (optval)
1221 cb->s_flags |= mask;
1222 else
1223 cb->s_flags &= ~mask;
1224 } else error = EINVAL;
1225 IPX_UNLOCK(ipxp);
1226 break;
1227
1228 case SO_MTU:
1229 error = sooptcopyin(sopt, &usoptval, sizeof usoptval,
1230 sizeof usoptval);
1231 if (error)
1232 break;
1233 /* Unlocked write. */
1234 cb->s_mtu = usoptval;
1235 break;
1236
1237 #ifdef SF_NEWCALL
1238 case SO_NEWCALL:
1239 error = sooptcopyin(sopt, &optval, sizeof optval,
1240 sizeof optval);
1241 if (error)
1242 break;
1243 IPX_LOCK(ipxp);
1244 if (optval) {
1245 cb->s_flags2 |= SF_NEWCALL;
1246 spx_newchecks[5]++;
1247 } else {
1248 cb->s_flags2 &= ~SF_NEWCALL;
1249 spx_newchecks[6]++;
1250 }
1251 IPX_UNLOCK(ipxp);
1252 break;
1253 #endif
1254
1255 case SO_DEFAULT_HEADERS:
1256 {
1257 struct spxhdr sp;
1258
1259 error = sooptcopyin(sopt, &sp, sizeof sp,
1260 sizeof sp);
1261 if (error)
1262 break;
1263 IPX_LOCK(ipxp);
1264 cb->s_dt = sp.spx_dt;
1265 cb->s_cc = sp.spx_cc & SPX_EM;
1266 IPX_UNLOCK(ipxp);
1267 }
1268 break;
1269
1270 default:
1271 error = ENOPROTOOPT;
1272 }
1273 break;
1274 }
1275 return (error);
1276 }
1277
1278 static int
1279 spx_usr_abort(so)
1280 struct socket *so;
1281 {
1282 struct ipxpcb *ipxp;
1283 struct spxpcb *cb;
1284
1285 ipxp = sotoipxpcb(so);
1286 cb = ipxtospxpcb(ipxp);
1287
1288 IPX_LIST_LOCK();
1289 IPX_LOCK(ipxp);
1290 spx_drop(cb, ECONNABORTED);
1291 IPX_LIST_UNLOCK();
1292 return (0);
1293 }
1294
1295 /*
1296 * Accept a connection. Essentially all the work is
1297 * done at higher levels; just return the address
1298 * of the peer, storing through addr.
1299 */
1300 static int
1301 spx_accept(so, nam)
1302 struct socket *so;
1303 struct sockaddr **nam;
1304 {
1305 struct ipxpcb *ipxp;
1306 struct sockaddr_ipx *sipx, ssipx;
1307
1308 ipxp = sotoipxpcb(so);
1309 sipx = &ssipx;
1310 bzero(sipx, sizeof *sipx);
1311 sipx->sipx_len = sizeof *sipx;
1312 sipx->sipx_family = AF_IPX;
1313 IPX_LOCK(ipxp);
1314 sipx->sipx_addr = ipxp->ipxp_faddr;
1315 IPX_UNLOCK(ipxp);
1316 *nam = sodupsockaddr((struct sockaddr *)sipx, M_WAITOK);
1317 return (0);
1318 }
1319
1320 static int
1321 spx_attach(so, proto, td)
1322 struct socket *so;
1323 int proto;
1324 struct thread *td;
1325 {
1326 struct ipxpcb *ipxp;
1327 struct spxpcb *cb;
1328 struct mbuf *mm;
1329 struct sockbuf *sb;
1330 int error;
1331
1332 ipxp = sotoipxpcb(so);
1333 cb = ipxtospxpcb(ipxp);
1334
1335 if (ipxp != NULL)
1336 return (EISCONN);
1337 IPX_LIST_LOCK();
1338 error = ipx_pcballoc(so, &ipxpcb_list, td);
1339 if (error)
1340 goto spx_attach_end;
1341 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1342 error = soreserve(so, (u_long) 3072, (u_long) 3072);
1343 if (error)
1344 goto spx_attach_end;
1345 }
1346 ipxp = sotoipxpcb(so);
1347
1348 MALLOC(cb, struct spxpcb *, sizeof *cb, M_PCB, M_NOWAIT | M_ZERO);
1349
1350 if (cb == NULL) {
1351 error = ENOBUFS;
1352 goto spx_attach_end;
1353 }
1354 sb = &so->so_snd;
1355
1356 mm = m_getclr(M_DONTWAIT, MT_HEADER);
1357 if (mm == NULL) {
1358 FREE(cb, M_PCB);
1359 error = ENOBUFS;
1360 goto spx_attach_end;
1361 }
1362 cb->s_ipx = mtod(mm, struct ipx *);
1363 cb->s_state = TCPS_LISTEN;
1364 cb->s_smax = -1;
1365 cb->s_swl1 = -1;
1366 cb->s_q.si_next = cb->s_q.si_prev = &cb->s_q;
1367 cb->s_ipxpcb = ipxp;
1368 cb->s_mtu = 576 - sizeof(struct spx);
1369 cb->s_cwnd = sbspace(sb) * CUNIT / cb->s_mtu;
1370 cb->s_ssthresh = cb->s_cwnd;
1371 cb->s_cwmx = sbspace(sb) * CUNIT / (2 * sizeof(struct spx));
1372 /* Above is recomputed when connecting to account
1373 for changed buffering or mtu's */
1374 cb->s_rtt = SPXTV_SRTTBASE;
1375 cb->s_rttvar = SPXTV_SRTTDFLT << 2;
1376 SPXT_RANGESET(cb->s_rxtcur,
1377 ((SPXTV_SRTTBASE >> 2) + (SPXTV_SRTTDFLT << 2)) >> 1,
1378 SPXTV_MIN, SPXTV_REXMTMAX);
1379 ipxp->ipxp_pcb = (caddr_t)cb;
1380 spx_attach_end:
1381 IPX_LIST_UNLOCK();
1382 return (error);
1383 }
1384
1385 static int
1386 spx_bind(so, nam, td)
1387 struct socket *so;
1388 struct sockaddr *nam;
1389 struct thread *td;
1390 {
1391 struct ipxpcb *ipxp;
1392 int error;
1393
1394 ipxp = sotoipxpcb(so);
1395
1396 IPX_LIST_LOCK();
1397 IPX_LOCK(ipxp);
1398 error = ipx_pcbbind(ipxp, nam, td);
1399 IPX_UNLOCK(ipxp);
1400 IPX_LIST_UNLOCK();
1401 return (error);
1402 }
1403
1404 /*
1405 * Initiate connection to peer.
1406 * Enter SYN_SENT state, and mark socket as connecting.
1407 * Start keep-alive timer, setup prototype header,
1408 * Send initial system packet requesting connection.
1409 */
1410 static int
1411 spx_connect(so, nam, td)
1412 struct socket *so;
1413 struct sockaddr *nam;
1414 struct thread *td;
1415 {
1416 struct ipxpcb *ipxp;
1417 struct spxpcb *cb;
1418 int error;
1419
1420 ipxp = sotoipxpcb(so);
1421 cb = ipxtospxpcb(ipxp);
1422
1423 IPX_LIST_LOCK();
1424 IPX_LOCK(ipxp);
1425 if (ipxp->ipxp_lport == 0) {
1426 error = ipx_pcbbind(ipxp, NULL, td);
1427 if (error)
1428 goto spx_connect_end;
1429 }
1430 error = ipx_pcbconnect(ipxp, nam, td);
1431 if (error)
1432 goto spx_connect_end;
1433 soisconnecting(so);
1434 spxstat.spxs_connattempt++;
1435 cb->s_state = TCPS_SYN_SENT;
1436 cb->s_did = 0;
1437 spx_template(cb);
1438 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1439 cb->s_force = 1 + SPXTV_KEEP;
1440 /*
1441 * Other party is required to respond to
1442 * the port I send from, but he is not
1443 * required to answer from where I am sending to,
1444 * so allow wildcarding.
1445 * original port I am sending to is still saved in
1446 * cb->s_dport.
1447 */
1448 ipxp->ipxp_fport = 0;
1449 error = spx_output(cb, NULL);
1450 spx_connect_end:
1451 IPX_UNLOCK(ipxp);
1452 IPX_LIST_UNLOCK();
1453 return (error);
1454 }
1455
1456 static int
1457 spx_detach(so)
1458 struct socket *so;
1459 {
1460 struct ipxpcb *ipxp;
1461 struct spxpcb *cb;
1462
1463 ipxp = sotoipxpcb(so);
1464 cb = ipxtospxpcb(ipxp);
1465
1466 if (ipxp == NULL)
1467 return (ENOTCONN);
1468 IPX_LIST_LOCK();
1469 IPX_LOCK(ipxp);
1470 if (cb->s_state > TCPS_LISTEN)
1471 spx_disconnect(cb);
1472 else
1473 spx_close(cb);
1474 IPX_LIST_UNLOCK();
1475 return (0);
1476 }
1477
1478 /*
1479 * We may decide later to implement connection closing
1480 * handshaking at the spx level optionally.
1481 * here is the hook to do it:
1482 */
1483 static int
1484 spx_usr_disconnect(so)
1485 struct socket *so;
1486 {
1487 struct ipxpcb *ipxp;
1488 struct spxpcb *cb;
1489
1490 ipxp = sotoipxpcb(so);
1491 cb = ipxtospxpcb(ipxp);
1492
1493 IPX_LIST_LOCK();
1494 IPX_LOCK(ipxp);
1495 spx_disconnect(cb);
1496 IPX_LIST_UNLOCK();
1497 return (0);
1498 }
1499
1500 static int
1501 spx_listen(so, td)
1502 struct socket *so;
1503 struct thread *td;
1504 {
1505 int error;
1506 struct ipxpcb *ipxp;
1507 struct spxpcb *cb;
1508
1509 error = 0;
1510 ipxp = sotoipxpcb(so);
1511 cb = ipxtospxpcb(ipxp);
1512
1513 IPX_LIST_LOCK();
1514 IPX_LOCK(ipxp);
1515 SOCK_LOCK(so);
1516 error = solisten_proto_check(so);
1517 if (error == 0 && ipxp->ipxp_lport == 0)
1518 error = ipx_pcbbind(ipxp, NULL, td);
1519 if (error == 0) {
1520 cb->s_state = TCPS_LISTEN;
1521 solisten_proto(so);
1522 }
1523 SOCK_UNLOCK(so);
1524 IPX_UNLOCK(ipxp);
1525 IPX_LIST_UNLOCK();
1526 return (error);
1527 }
1528
1529 /*
1530 * After a receive, possibly send acknowledgment
1531 * updating allocation.
1532 */
1533 static int
1534 spx_rcvd(so, flags)
1535 struct socket *so;
1536 int flags;
1537 {
1538 struct ipxpcb *ipxp;
1539 struct spxpcb *cb;
1540
1541 ipxp = sotoipxpcb(so);
1542 cb = ipxtospxpcb(ipxp);
1543
1544 IPX_LOCK(ipxp);
1545 cb->s_flags |= SF_RVD;
1546 spx_output(cb, NULL);
1547 cb->s_flags &= ~SF_RVD;
1548 IPX_UNLOCK(ipxp);
1549 return (0);
1550 }
1551
1552 static int
1553 spx_rcvoob(so, m, flags)
1554 struct socket *so;
1555 struct mbuf *m;
1556 int flags;
1557 {
1558 struct ipxpcb *ipxp;
1559 struct spxpcb *cb;
1560
1561 ipxp = sotoipxpcb(so);
1562 cb = ipxtospxpcb(ipxp);
1563
1564 SOCKBUF_LOCK(&so->so_rcv);
1565 if ((cb->s_oobflags & SF_IOOB) || so->so_oobmark ||
1566 (so->so_rcv.sb_state & SBS_RCVATMARK)) {
1567 SOCKBUF_UNLOCK(&so->so_rcv);
1568 m->m_len = 1;
1569 /* Unlocked read. */
1570 *mtod(m, caddr_t) = cb->s_iobc;
1571 return (0);
1572 }
1573 SOCKBUF_UNLOCK(&so->so_rcv);
1574 return (EINVAL);
1575 }
1576
1577 static int
1578 spx_send(so, flags, m, addr, controlp, td)
1579 struct socket *so;
1580 int flags;
1581 struct mbuf *m;
1582 struct sockaddr *addr;
1583 struct mbuf *controlp;
1584 struct thread *td;
1585 {
1586 int error;
1587 struct ipxpcb *ipxp;
1588 struct spxpcb *cb;
1589
1590 error = 0;
1591 ipxp = sotoipxpcb(so);
1592 cb = ipxtospxpcb(ipxp);
1593
1594 IPX_LOCK(ipxp);
1595 if (flags & PRUS_OOB) {
1596 if (sbspace(&so->so_snd) < -512) {
1597 error = ENOBUFS;
1598 goto spx_send_end;
1599 }
1600 cb->s_oobflags |= SF_SOOB;
1601 }
1602 if (controlp != NULL) {
1603 u_short *p = mtod(controlp, u_short *);
1604 spx_newchecks[2]++;
1605 if ((p[0] == 5) && (p[1] == 1)) { /* XXXX, for testing */
1606 cb->s_shdr.spx_dt = *(u_char *)(&p[2]);
1607 spx_newchecks[3]++;
1608 }
1609 m_freem(controlp);
1610 }
1611 controlp = NULL;
1612 error = spx_output(cb, m);
1613 m = NULL;
1614 spx_send_end:
1615 IPX_UNLOCK(ipxp);
1616 if (controlp != NULL)
1617 m_freem(controlp);
1618 if (m != NULL)
1619 m_freem(m);
1620 return (error);
1621 }
1622
1623 static int
1624 spx_shutdown(so)
1625 struct socket *so;
1626 {
1627 struct ipxpcb *ipxp;
1628 struct spxpcb *cb;
1629
1630 ipxp = sotoipxpcb(so);
1631 cb = ipxtospxpcb(ipxp);
1632
1633 socantsendmore(so);
1634 IPX_LIST_LOCK();
1635 IPX_LOCK(ipxp);
1636 spx_usrclosed(cb);
1637 IPX_LIST_UNLOCK();
1638 return (0);
1639 }
1640
1641 static int
1642 spx_sp_attach(so, proto, td)
1643 struct socket *so;
1644 int proto;
1645 struct thread *td;
1646 {
1647 int error;
1648 struct ipxpcb *ipxp;
1649
1650 error = spx_attach(so, proto, td);
1651 if (error == 0) {
1652 ipxp = sotoipxpcb(so);
1653 ((struct spxpcb *)ipxp->ipxp_pcb)->s_flags |=
1654 (SF_HI | SF_HO | SF_PI);
1655 }
1656 return (error);
1657 }
1658
1659 /*
1660 * Create template to be used to send spx packets on a connection.
1661 * Called after host entry created, fills
1662 * in a skeletal spx header (choosing connection id),
1663 * minimizing the amount of work necessary when the connection is used.
1664 */
1665 static void
1666 spx_template(cb)
1667 register struct spxpcb *cb;
1668 {
1669 register struct ipxpcb *ipxp = cb->s_ipxpcb;
1670 register struct ipx *ipx = cb->s_ipx;
1671 register struct sockbuf *sb = &(ipxp->ipxp_socket->so_snd);
1672
1673 IPX_LOCK_ASSERT(ipxp);
1674
1675 ipx->ipx_pt = IPXPROTO_SPX;
1676 ipx->ipx_sna = ipxp->ipxp_laddr;
1677 ipx->ipx_dna = ipxp->ipxp_faddr;
1678 cb->s_sid = htons(spx_iss);
1679 spx_iss += SPX_ISSINCR/2;
1680 cb->s_alo = 1;
1681 cb->s_cwnd = (sbspace(sb) * CUNIT) / cb->s_mtu;
1682 cb->s_ssthresh = cb->s_cwnd; /* Try to expand fast to full complement
1683 of large packets */
1684 cb->s_cwmx = (sbspace(sb) * CUNIT) / (2 * sizeof(struct spx));
1685 cb->s_cwmx = max(cb->s_cwmx, cb->s_cwnd);
1686 /* But allow for lots of little packets as well */
1687 }
1688
1689 /*
1690 * Close a SPIP control block:
1691 * discard spx control block itself
1692 * discard ipx protocol control block
1693 * wake up any sleepers
1694 * cb will always be invalid after this call.
1695 */
1696 void
1697 spx_close(cb)
1698 register struct spxpcb *cb;
1699 {
1700 register struct spx_q *s;
1701 struct ipxpcb *ipxp = cb->s_ipxpcb;
1702 struct socket *so = ipxp->ipxp_socket;
1703 register struct mbuf *m;
1704
1705 IPX_LIST_LOCK_ASSERT();
1706 IPX_LOCK_ASSERT(ipxp);
1707
1708 s = cb->s_q.si_next;
1709 while (s != &(cb->s_q)) {
1710 s = s->si_next;
1711 m = dtom(s->si_prev);
1712 remque(s->si_prev);
1713 m_freem(m);
1714 }
1715 m_free(dtom(cb->s_ipx));
1716 FREE(cb, M_PCB);
1717 ipxp->ipxp_pcb = NULL;
1718 soisdisconnected(so);
1719 ipx_pcbdetach(ipxp);
1720 spxstat.spxs_closed++;
1721 }
1722
1723 /*
1724 * Someday we may do level 3 handshaking
1725 * to close a connection or send a xerox style error.
1726 * For now, just close.
1727 * cb will always be invalid after this call.
1728 */
1729 static void
1730 spx_usrclosed(cb)
1731 register struct spxpcb *cb;
1732 {
1733
1734 IPX_LIST_LOCK_ASSERT();
1735 IPX_LOCK_ASSERT(cb->s_ipxpcb);
1736
1737 spx_close(cb);
1738 }
1739
1740 /*
1741 * cb will always be invalid after this call.
1742 */
1743 static void
1744 spx_disconnect(cb)
1745 register struct spxpcb *cb;
1746 {
1747
1748 IPX_LIST_LOCK_ASSERT();
1749 IPX_LOCK_ASSERT(cb->s_ipxpcb);
1750
1751 spx_close(cb);
1752 }
1753
1754 /*
1755 * Drop connection, reporting
1756 * the specified error.
1757 * cb will always be invalid after this call.
1758 */
1759 static void
1760 spx_drop(cb, errno)
1761 register struct spxpcb *cb;
1762 int errno;
1763 {
1764 struct socket *so = cb->s_ipxpcb->ipxp_socket;
1765
1766 IPX_LIST_LOCK_ASSERT();
1767 IPX_LOCK_ASSERT(cb->s_ipxpcb);
1768
1769 /*
1770 * someday, in the xerox world
1771 * we will generate error protocol packets
1772 * announcing that the socket has gone away.
1773 */
1774 if (TCPS_HAVERCVDSYN(cb->s_state)) {
1775 spxstat.spxs_drops++;
1776 cb->s_state = TCPS_CLOSED;
1777 /*tcp_output(cb);*/
1778 } else
1779 spxstat.spxs_conndrops++;
1780 so->so_error = errno;
1781 spx_close(cb);
1782 }
1783
1784 /*
1785 * Fast timeout routine for processing delayed acks
1786 */
1787 void
1788 spx_fasttimo()
1789 {
1790 struct ipxpcb *ipxp;
1791 struct spxpcb *cb;
1792
1793 IPX_LIST_LOCK();
1794 LIST_FOREACH(ipxp, &ipxpcb_list, ipxp_list) {
1795 IPX_LOCK(ipxp);
1796 if ((cb = (struct spxpcb *)ipxp->ipxp_pcb) != NULL &&
1797 (cb->s_flags & SF_DELACK)) {
1798 cb->s_flags &= ~SF_DELACK;
1799 cb->s_flags |= SF_ACKNOW;
1800 spxstat.spxs_delack++;
1801 spx_output(cb, NULL);
1802 }
1803 IPX_UNLOCK(ipxp);
1804 }
1805 IPX_LIST_UNLOCK();
1806 }
1807
1808 /*
1809 * spx protocol timeout routine called every 500 ms.
1810 * Updates the timers in all active pcb's and
1811 * causes finite state machine actions if timers expire.
1812 */
1813 void
1814 spx_slowtimo()
1815 {
1816 struct ipxpcb *ip, *ip_temp;
1817 struct spxpcb *cb;
1818 int i;
1819
1820 /*
1821 * Search through tcb's and update active timers. Note that timers
1822 * may free the ipxpcb, so be sure to handle that case.
1823 *
1824 * spx_timers() may remove an ipxpcb entry, so we have to be ready to
1825 * continue despite that. The logic here is a bit obfuscated.
1826 */
1827 IPX_LIST_LOCK();
1828 LIST_FOREACH_SAFE(ip, &ipxpcb_list, ipxp_list, ip_temp) {
1829 cb = ipxtospxpcb(ip);
1830 if (cb == NULL)
1831 continue;
1832 IPX_LOCK(cb->s_ipxpcb);
1833 for (i = 0; i < SPXT_NTIMERS; i++) {
1834 if (cb->s_timer[i] && --cb->s_timer[i] == 0) {
1835 /*
1836 * spx_timers() returns (NULL) if it free'd
1837 * the pcb.
1838 */
1839 cb = spx_timers(cb, i);
1840 if (cb == NULL)
1841 break;
1842 }
1843 }
1844 if (cb != NULL) {
1845 cb->s_idle++;
1846 if (cb->s_rtt)
1847 cb->s_rtt++;
1848 IPX_UNLOCK(cb->s_ipxpcb);
1849 }
1850 }
1851 spx_iss += SPX_ISSINCR/PR_SLOWHZ; /* increment iss */
1852 IPX_LIST_UNLOCK();
1853 }
1854
1855 /*
1856 * SPX timer processing.
1857 */
1858 static struct spxpcb *
1859 spx_timers(cb, timer)
1860 register struct spxpcb *cb;
1861 int timer;
1862 {
1863 long rexmt;
1864 int win;
1865
1866 IPX_LIST_LOCK_ASSERT();
1867 IPX_LOCK_ASSERT(cb->s_ipxpcb);
1868
1869 cb->s_force = 1 + timer;
1870 switch (timer) {
1871
1872 /*
1873 * 2 MSL timeout in shutdown went off. TCP deletes connection
1874 * control block.
1875 */
1876 case SPXT_2MSL:
1877 printf("spx: SPXT_2MSL went off for no reason\n");
1878 cb->s_timer[timer] = 0;
1879 break;
1880
1881 /*
1882 * Retransmission timer went off. Message has not
1883 * been acked within retransmit interval. Back off
1884 * to a longer retransmit interval and retransmit one packet.
1885 */
1886 case SPXT_REXMT:
1887 if (++cb->s_rxtshift > SPX_MAXRXTSHIFT) {
1888 cb->s_rxtshift = SPX_MAXRXTSHIFT;
1889 spxstat.spxs_timeoutdrop++;
1890 spx_drop(cb, ETIMEDOUT);
1891 cb = NULL;
1892 break;
1893 }
1894 spxstat.spxs_rexmttimeo++;
1895 rexmt = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1896 rexmt *= spx_backoff[cb->s_rxtshift];
1897 SPXT_RANGESET(cb->s_rxtcur, rexmt, SPXTV_MIN, SPXTV_REXMTMAX);
1898 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1899 /*
1900 * If we have backed off fairly far, our srtt
1901 * estimate is probably bogus. Clobber it
1902 * so we'll take the next rtt measurement as our srtt;
1903 * move the current srtt into rttvar to keep the current
1904 * retransmit times until then.
1905 */
1906 if (cb->s_rxtshift > SPX_MAXRXTSHIFT / 4 ) {
1907 cb->s_rttvar += (cb->s_srtt >> 2);
1908 cb->s_srtt = 0;
1909 }
1910 cb->s_snxt = cb->s_rack;
1911 /*
1912 * If timing a packet, stop the timer.
1913 */
1914 cb->s_rtt = 0;
1915 /*
1916 * See very long discussion in tcp_timer.c about congestion
1917 * window and sstrhesh
1918 */
1919 win = min(cb->s_swnd, (cb->s_cwnd/CUNIT)) / 2;
1920 if (win < 2)
1921 win = 2;
1922 cb->s_cwnd = CUNIT;
1923 cb->s_ssthresh = win * CUNIT;
1924 spx_output(cb, NULL);
1925 break;
1926
1927 /*
1928 * Persistance timer into zero window.
1929 * Force a probe to be sent.
1930 */
1931 case SPXT_PERSIST:
1932 spxstat.spxs_persisttimeo++;
1933 spx_setpersist(cb);
1934 spx_output(cb, NULL);
1935 break;
1936
1937 /*
1938 * Keep-alive timer went off; send something
1939 * or drop connection if idle for too long.
1940 */
1941 case SPXT_KEEP:
1942 spxstat.spxs_keeptimeo++;
1943 if (cb->s_state < TCPS_ESTABLISHED)
1944 goto dropit;
1945 if (cb->s_ipxpcb->ipxp_socket->so_options & SO_KEEPALIVE) {
1946 if (cb->s_idle >= SPXTV_MAXIDLE)
1947 goto dropit;
1948 spxstat.spxs_keepprobe++;
1949 spx_output(cb, NULL);
1950 } else
1951 cb->s_idle = 0;
1952 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1953 break;
1954 dropit:
1955 spxstat.spxs_keepdrops++;
1956 spx_drop(cb, ETIMEDOUT);
1957 cb = NULL;
1958 break;
1959 }
1960 return (cb);
1961 }
Cache object: 25f6e7d2928237eb472807641707a4cf
|